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, const char *device_path, 2080 u64 devid, struct block_device **bdev, fmode_t *mode) 2081 { 2082 struct btrfs_device *device; 2083 struct btrfs_fs_devices *cur_devices; 2084 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2085 u64 num_devices; 2086 int ret = 0; 2087 2088 /* 2089 * The device list in fs_devices is accessed without locks (neither 2090 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2091 * filesystem and another device rm cannot run. 2092 */ 2093 num_devices = btrfs_num_devices(fs_info); 2094 2095 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2096 if (ret) 2097 goto out; 2098 2099 device = btrfs_find_device_by_devspec(fs_info, devid, device_path); 2100 2101 if (IS_ERR(device)) { 2102 if (PTR_ERR(device) == -ENOENT && 2103 device_path && strcmp(device_path, "missing") == 0) 2104 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2105 else 2106 ret = PTR_ERR(device); 2107 goto out; 2108 } 2109 2110 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2111 btrfs_warn_in_rcu(fs_info, 2112 "cannot remove device %s (devid %llu) due to active swapfile", 2113 rcu_str_deref(device->name), device->devid); 2114 ret = -ETXTBSY; 2115 goto out; 2116 } 2117 2118 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2119 ret = BTRFS_ERROR_DEV_TGT_REPLACE; 2120 goto out; 2121 } 2122 2123 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2124 fs_info->fs_devices->rw_devices == 1) { 2125 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; 2126 goto out; 2127 } 2128 2129 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2130 mutex_lock(&fs_info->chunk_mutex); 2131 list_del_init(&device->dev_alloc_list); 2132 device->fs_devices->rw_devices--; 2133 mutex_unlock(&fs_info->chunk_mutex); 2134 } 2135 2136 ret = btrfs_shrink_device(device, 0); 2137 if (!ret) 2138 btrfs_reada_remove_dev(device); 2139 if (ret) 2140 goto error_undo; 2141 2142 /* 2143 * TODO: the superblock still includes this device in its num_devices 2144 * counter although write_all_supers() is not locked out. This 2145 * could give a filesystem state which requires a degraded mount. 2146 */ 2147 ret = btrfs_rm_dev_item(device); 2148 if (ret) 2149 goto error_undo; 2150 2151 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2152 btrfs_scrub_cancel_dev(device); 2153 2154 /* 2155 * the device list mutex makes sure that we don't change 2156 * the device list while someone else is writing out all 2157 * the device supers. Whoever is writing all supers, should 2158 * lock the device list mutex before getting the number of 2159 * devices in the super block (super_copy). Conversely, 2160 * whoever updates the number of devices in the super block 2161 * (super_copy) should hold the device list mutex. 2162 */ 2163 2164 /* 2165 * In normal cases the cur_devices == fs_devices. But in case 2166 * of deleting a seed device, the cur_devices should point to 2167 * its own fs_devices listed under the fs_devices->seed_list. 2168 */ 2169 cur_devices = device->fs_devices; 2170 mutex_lock(&fs_devices->device_list_mutex); 2171 list_del_rcu(&device->dev_list); 2172 2173 cur_devices->num_devices--; 2174 cur_devices->total_devices--; 2175 /* Update total_devices of the parent fs_devices if it's seed */ 2176 if (cur_devices != fs_devices) 2177 fs_devices->total_devices--; 2178 2179 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2180 cur_devices->missing_devices--; 2181 2182 btrfs_assign_next_active_device(device, NULL); 2183 2184 if (device->bdev) { 2185 cur_devices->open_devices--; 2186 /* remove sysfs entry */ 2187 btrfs_sysfs_remove_device(device); 2188 } 2189 2190 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2191 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2192 mutex_unlock(&fs_devices->device_list_mutex); 2193 2194 /* 2195 * At this point, the device is zero sized and detached from the 2196 * devices list. All that's left is to zero out the old supers and 2197 * free the device. 2198 * 2199 * We cannot call btrfs_close_bdev() here because we're holding the sb 2200 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2201 * block device and it's dependencies. Instead just flush the device 2202 * and let the caller do the final blkdev_put. 2203 */ 2204 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2205 btrfs_scratch_superblocks(fs_info, device->bdev, 2206 device->name->str); 2207 if (device->bdev) { 2208 sync_blockdev(device->bdev); 2209 invalidate_bdev(device->bdev); 2210 } 2211 } 2212 2213 *bdev = device->bdev; 2214 *mode = device->mode; 2215 synchronize_rcu(); 2216 btrfs_free_device(device); 2217 2218 /* 2219 * This can happen if cur_devices is the private seed devices list. We 2220 * cannot call close_fs_devices() here because it expects the uuid_mutex 2221 * to be held, but in fact we don't need that for the private 2222 * seed_devices, we can simply decrement cur_devices->opened and then 2223 * remove it from our list and free the fs_devices. 2224 */ 2225 if (cur_devices->num_devices == 0) { 2226 list_del_init(&cur_devices->seed_list); 2227 ASSERT(cur_devices->opened == 1); 2228 cur_devices->opened--; 2229 free_fs_devices(cur_devices); 2230 } 2231 2232 out: 2233 return ret; 2234 2235 error_undo: 2236 btrfs_reada_undo_remove_dev(device); 2237 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2238 mutex_lock(&fs_info->chunk_mutex); 2239 list_add(&device->dev_alloc_list, 2240 &fs_devices->alloc_list); 2241 device->fs_devices->rw_devices++; 2242 mutex_unlock(&fs_info->chunk_mutex); 2243 } 2244 goto out; 2245 } 2246 2247 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2248 { 2249 struct btrfs_fs_devices *fs_devices; 2250 2251 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2252 2253 /* 2254 * in case of fs with no seed, srcdev->fs_devices will point 2255 * to fs_devices of fs_info. However when the dev being replaced is 2256 * a seed dev it will point to the seed's local fs_devices. In short 2257 * srcdev will have its correct fs_devices in both the cases. 2258 */ 2259 fs_devices = srcdev->fs_devices; 2260 2261 list_del_rcu(&srcdev->dev_list); 2262 list_del(&srcdev->dev_alloc_list); 2263 fs_devices->num_devices--; 2264 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2265 fs_devices->missing_devices--; 2266 2267 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2268 fs_devices->rw_devices--; 2269 2270 if (srcdev->bdev) 2271 fs_devices->open_devices--; 2272 } 2273 2274 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2275 { 2276 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2277 2278 mutex_lock(&uuid_mutex); 2279 2280 btrfs_close_bdev(srcdev); 2281 synchronize_rcu(); 2282 btrfs_free_device(srcdev); 2283 2284 /* if this is no devs we rather delete the fs_devices */ 2285 if (!fs_devices->num_devices) { 2286 /* 2287 * On a mounted FS, num_devices can't be zero unless it's a 2288 * seed. In case of a seed device being replaced, the replace 2289 * target added to the sprout FS, so there will be no more 2290 * device left under the seed FS. 2291 */ 2292 ASSERT(fs_devices->seeding); 2293 2294 list_del_init(&fs_devices->seed_list); 2295 close_fs_devices(fs_devices); 2296 free_fs_devices(fs_devices); 2297 } 2298 mutex_unlock(&uuid_mutex); 2299 } 2300 2301 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2302 { 2303 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2304 2305 mutex_lock(&fs_devices->device_list_mutex); 2306 2307 btrfs_sysfs_remove_device(tgtdev); 2308 2309 if (tgtdev->bdev) 2310 fs_devices->open_devices--; 2311 2312 fs_devices->num_devices--; 2313 2314 btrfs_assign_next_active_device(tgtdev, NULL); 2315 2316 list_del_rcu(&tgtdev->dev_list); 2317 2318 mutex_unlock(&fs_devices->device_list_mutex); 2319 2320 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2321 tgtdev->name->str); 2322 2323 btrfs_close_bdev(tgtdev); 2324 synchronize_rcu(); 2325 btrfs_free_device(tgtdev); 2326 } 2327 2328 static struct btrfs_device *btrfs_find_device_by_path( 2329 struct btrfs_fs_info *fs_info, const char *device_path) 2330 { 2331 BTRFS_DEV_LOOKUP_ARGS(args); 2332 int ret = 0; 2333 struct btrfs_super_block *disk_super; 2334 struct block_device *bdev; 2335 struct btrfs_device *device; 2336 2337 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, 2338 fs_info->bdev_holder, 0, &bdev, &disk_super); 2339 if (ret) 2340 return ERR_PTR(ret); 2341 2342 args.devid = btrfs_stack_device_id(&disk_super->dev_item); 2343 args.uuid = disk_super->dev_item.uuid; 2344 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2345 args.fsid = disk_super->metadata_uuid; 2346 else 2347 args.fsid = disk_super->fsid; 2348 2349 device = btrfs_find_device(fs_info->fs_devices, &args); 2350 2351 btrfs_release_disk_super(disk_super); 2352 if (!device) 2353 device = ERR_PTR(-ENOENT); 2354 blkdev_put(bdev, FMODE_READ); 2355 return device; 2356 } 2357 2358 /* 2359 * Lookup a device given by device id, or the path if the id is 0. 2360 */ 2361 struct btrfs_device *btrfs_find_device_by_devspec( 2362 struct btrfs_fs_info *fs_info, u64 devid, 2363 const char *device_path) 2364 { 2365 BTRFS_DEV_LOOKUP_ARGS(args); 2366 struct btrfs_device *device; 2367 2368 if (devid) { 2369 args.devid = devid; 2370 device = btrfs_find_device(fs_info->fs_devices, &args); 2371 if (!device) 2372 return ERR_PTR(-ENOENT); 2373 return device; 2374 } 2375 2376 if (!device_path || !device_path[0]) 2377 return ERR_PTR(-EINVAL); 2378 2379 if (strcmp(device_path, "missing") == 0) { 2380 args.missing = true; 2381 device = btrfs_find_device(fs_info->fs_devices, &args); 2382 if (!device) 2383 return ERR_PTR(-ENOENT); 2384 return device; 2385 } 2386 2387 return btrfs_find_device_by_path(fs_info, device_path); 2388 } 2389 2390 /* 2391 * does all the dirty work required for changing file system's UUID. 2392 */ 2393 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) 2394 { 2395 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2396 struct btrfs_fs_devices *old_devices; 2397 struct btrfs_fs_devices *seed_devices; 2398 struct btrfs_super_block *disk_super = fs_info->super_copy; 2399 struct btrfs_device *device; 2400 u64 super_flags; 2401 2402 lockdep_assert_held(&uuid_mutex); 2403 if (!fs_devices->seeding) 2404 return -EINVAL; 2405 2406 /* 2407 * Private copy of the seed devices, anchored at 2408 * fs_info->fs_devices->seed_list 2409 */ 2410 seed_devices = alloc_fs_devices(NULL, NULL); 2411 if (IS_ERR(seed_devices)) 2412 return PTR_ERR(seed_devices); 2413 2414 /* 2415 * It's necessary to retain a copy of the original seed fs_devices in 2416 * fs_uuids so that filesystems which have been seeded can successfully 2417 * reference the seed device from open_seed_devices. This also supports 2418 * multiple fs seed. 2419 */ 2420 old_devices = clone_fs_devices(fs_devices); 2421 if (IS_ERR(old_devices)) { 2422 kfree(seed_devices); 2423 return PTR_ERR(old_devices); 2424 } 2425 2426 list_add(&old_devices->fs_list, &fs_uuids); 2427 2428 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2429 seed_devices->opened = 1; 2430 INIT_LIST_HEAD(&seed_devices->devices); 2431 INIT_LIST_HEAD(&seed_devices->alloc_list); 2432 mutex_init(&seed_devices->device_list_mutex); 2433 2434 mutex_lock(&fs_devices->device_list_mutex); 2435 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2436 synchronize_rcu); 2437 list_for_each_entry(device, &seed_devices->devices, dev_list) 2438 device->fs_devices = seed_devices; 2439 2440 fs_devices->seeding = false; 2441 fs_devices->num_devices = 0; 2442 fs_devices->open_devices = 0; 2443 fs_devices->missing_devices = 0; 2444 fs_devices->rotating = false; 2445 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2446 2447 generate_random_uuid(fs_devices->fsid); 2448 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2449 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2450 mutex_unlock(&fs_devices->device_list_mutex); 2451 2452 super_flags = btrfs_super_flags(disk_super) & 2453 ~BTRFS_SUPER_FLAG_SEEDING; 2454 btrfs_set_super_flags(disk_super, super_flags); 2455 2456 return 0; 2457 } 2458 2459 /* 2460 * Store the expected generation for seed devices in device items. 2461 */ 2462 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2463 { 2464 BTRFS_DEV_LOOKUP_ARGS(args); 2465 struct btrfs_fs_info *fs_info = trans->fs_info; 2466 struct btrfs_root *root = fs_info->chunk_root; 2467 struct btrfs_path *path; 2468 struct extent_buffer *leaf; 2469 struct btrfs_dev_item *dev_item; 2470 struct btrfs_device *device; 2471 struct btrfs_key key; 2472 u8 fs_uuid[BTRFS_FSID_SIZE]; 2473 u8 dev_uuid[BTRFS_UUID_SIZE]; 2474 int ret; 2475 2476 path = btrfs_alloc_path(); 2477 if (!path) 2478 return -ENOMEM; 2479 2480 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2481 key.offset = 0; 2482 key.type = BTRFS_DEV_ITEM_KEY; 2483 2484 while (1) { 2485 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2486 if (ret < 0) 2487 goto error; 2488 2489 leaf = path->nodes[0]; 2490 next_slot: 2491 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2492 ret = btrfs_next_leaf(root, path); 2493 if (ret > 0) 2494 break; 2495 if (ret < 0) 2496 goto error; 2497 leaf = path->nodes[0]; 2498 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2499 btrfs_release_path(path); 2500 continue; 2501 } 2502 2503 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2504 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2505 key.type != BTRFS_DEV_ITEM_KEY) 2506 break; 2507 2508 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2509 struct btrfs_dev_item); 2510 args.devid = btrfs_device_id(leaf, dev_item); 2511 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2512 BTRFS_UUID_SIZE); 2513 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2514 BTRFS_FSID_SIZE); 2515 args.uuid = dev_uuid; 2516 args.fsid = fs_uuid; 2517 device = btrfs_find_device(fs_info->fs_devices, &args); 2518 BUG_ON(!device); /* Logic error */ 2519 2520 if (device->fs_devices->seeding) { 2521 btrfs_set_device_generation(leaf, dev_item, 2522 device->generation); 2523 btrfs_mark_buffer_dirty(leaf); 2524 } 2525 2526 path->slots[0]++; 2527 goto next_slot; 2528 } 2529 ret = 0; 2530 error: 2531 btrfs_free_path(path); 2532 return ret; 2533 } 2534 2535 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2536 { 2537 struct btrfs_root *root = fs_info->dev_root; 2538 struct request_queue *q; 2539 struct btrfs_trans_handle *trans; 2540 struct btrfs_device *device; 2541 struct block_device *bdev; 2542 struct super_block *sb = fs_info->sb; 2543 struct rcu_string *name; 2544 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2545 u64 orig_super_total_bytes; 2546 u64 orig_super_num_devices; 2547 int seeding_dev = 0; 2548 int ret = 0; 2549 bool locked = false; 2550 2551 if (sb_rdonly(sb) && !fs_devices->seeding) 2552 return -EROFS; 2553 2554 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2555 fs_info->bdev_holder); 2556 if (IS_ERR(bdev)) 2557 return PTR_ERR(bdev); 2558 2559 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2560 ret = -EINVAL; 2561 goto error; 2562 } 2563 2564 if (fs_devices->seeding) { 2565 seeding_dev = 1; 2566 down_write(&sb->s_umount); 2567 mutex_lock(&uuid_mutex); 2568 locked = true; 2569 } 2570 2571 sync_blockdev(bdev); 2572 2573 rcu_read_lock(); 2574 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2575 if (device->bdev == bdev) { 2576 ret = -EEXIST; 2577 rcu_read_unlock(); 2578 goto error; 2579 } 2580 } 2581 rcu_read_unlock(); 2582 2583 device = btrfs_alloc_device(fs_info, NULL, NULL); 2584 if (IS_ERR(device)) { 2585 /* we can safely leave the fs_devices entry around */ 2586 ret = PTR_ERR(device); 2587 goto error; 2588 } 2589 2590 name = rcu_string_strdup(device_path, GFP_KERNEL); 2591 if (!name) { 2592 ret = -ENOMEM; 2593 goto error_free_device; 2594 } 2595 rcu_assign_pointer(device->name, name); 2596 2597 device->fs_info = fs_info; 2598 device->bdev = bdev; 2599 2600 ret = btrfs_get_dev_zone_info(device); 2601 if (ret) 2602 goto error_free_device; 2603 2604 trans = btrfs_start_transaction(root, 0); 2605 if (IS_ERR(trans)) { 2606 ret = PTR_ERR(trans); 2607 goto error_free_zone; 2608 } 2609 2610 q = bdev_get_queue(bdev); 2611 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2612 device->generation = trans->transid; 2613 device->io_width = fs_info->sectorsize; 2614 device->io_align = fs_info->sectorsize; 2615 device->sector_size = fs_info->sectorsize; 2616 device->total_bytes = round_down(i_size_read(bdev->bd_inode), 2617 fs_info->sectorsize); 2618 device->disk_total_bytes = device->total_bytes; 2619 device->commit_total_bytes = device->total_bytes; 2620 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2621 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2622 device->mode = FMODE_EXCL; 2623 device->dev_stats_valid = 1; 2624 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2625 2626 if (seeding_dev) { 2627 btrfs_clear_sb_rdonly(sb); 2628 ret = btrfs_prepare_sprout(fs_info); 2629 if (ret) { 2630 btrfs_abort_transaction(trans, ret); 2631 goto error_trans; 2632 } 2633 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2634 device); 2635 } 2636 2637 device->fs_devices = fs_devices; 2638 2639 mutex_lock(&fs_devices->device_list_mutex); 2640 mutex_lock(&fs_info->chunk_mutex); 2641 list_add_rcu(&device->dev_list, &fs_devices->devices); 2642 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2643 fs_devices->num_devices++; 2644 fs_devices->open_devices++; 2645 fs_devices->rw_devices++; 2646 fs_devices->total_devices++; 2647 fs_devices->total_rw_bytes += device->total_bytes; 2648 2649 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2650 2651 if (!blk_queue_nonrot(q)) 2652 fs_devices->rotating = true; 2653 2654 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2655 btrfs_set_super_total_bytes(fs_info->super_copy, 2656 round_down(orig_super_total_bytes + device->total_bytes, 2657 fs_info->sectorsize)); 2658 2659 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2660 btrfs_set_super_num_devices(fs_info->super_copy, 2661 orig_super_num_devices + 1); 2662 2663 /* 2664 * we've got more storage, clear any full flags on the space 2665 * infos 2666 */ 2667 btrfs_clear_space_info_full(fs_info); 2668 2669 mutex_unlock(&fs_info->chunk_mutex); 2670 2671 /* Add sysfs device entry */ 2672 btrfs_sysfs_add_device(device); 2673 2674 mutex_unlock(&fs_devices->device_list_mutex); 2675 2676 if (seeding_dev) { 2677 mutex_lock(&fs_info->chunk_mutex); 2678 ret = init_first_rw_device(trans); 2679 mutex_unlock(&fs_info->chunk_mutex); 2680 if (ret) { 2681 btrfs_abort_transaction(trans, ret); 2682 goto error_sysfs; 2683 } 2684 } 2685 2686 ret = btrfs_add_dev_item(trans, device); 2687 if (ret) { 2688 btrfs_abort_transaction(trans, ret); 2689 goto error_sysfs; 2690 } 2691 2692 if (seeding_dev) { 2693 ret = btrfs_finish_sprout(trans); 2694 if (ret) { 2695 btrfs_abort_transaction(trans, ret); 2696 goto error_sysfs; 2697 } 2698 2699 /* 2700 * fs_devices now represents the newly sprouted filesystem and 2701 * its fsid has been changed by btrfs_prepare_sprout 2702 */ 2703 btrfs_sysfs_update_sprout_fsid(fs_devices); 2704 } 2705 2706 ret = btrfs_commit_transaction(trans); 2707 2708 if (seeding_dev) { 2709 mutex_unlock(&uuid_mutex); 2710 up_write(&sb->s_umount); 2711 locked = false; 2712 2713 if (ret) /* transaction commit */ 2714 return ret; 2715 2716 ret = btrfs_relocate_sys_chunks(fs_info); 2717 if (ret < 0) 2718 btrfs_handle_fs_error(fs_info, ret, 2719 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2720 trans = btrfs_attach_transaction(root); 2721 if (IS_ERR(trans)) { 2722 if (PTR_ERR(trans) == -ENOENT) 2723 return 0; 2724 ret = PTR_ERR(trans); 2725 trans = NULL; 2726 goto error_sysfs; 2727 } 2728 ret = btrfs_commit_transaction(trans); 2729 } 2730 2731 /* 2732 * Now that we have written a new super block to this device, check all 2733 * other fs_devices list if device_path alienates any other scanned 2734 * device. 2735 * We can ignore the return value as it typically returns -EINVAL and 2736 * only succeeds if the device was an alien. 2737 */ 2738 btrfs_forget_devices(device_path); 2739 2740 /* Update ctime/mtime for blkid or udev */ 2741 update_dev_time(bdev); 2742 2743 return ret; 2744 2745 error_sysfs: 2746 btrfs_sysfs_remove_device(device); 2747 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2748 mutex_lock(&fs_info->chunk_mutex); 2749 list_del_rcu(&device->dev_list); 2750 list_del(&device->dev_alloc_list); 2751 fs_info->fs_devices->num_devices--; 2752 fs_info->fs_devices->open_devices--; 2753 fs_info->fs_devices->rw_devices--; 2754 fs_info->fs_devices->total_devices--; 2755 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2756 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2757 btrfs_set_super_total_bytes(fs_info->super_copy, 2758 orig_super_total_bytes); 2759 btrfs_set_super_num_devices(fs_info->super_copy, 2760 orig_super_num_devices); 2761 mutex_unlock(&fs_info->chunk_mutex); 2762 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2763 error_trans: 2764 if (seeding_dev) 2765 btrfs_set_sb_rdonly(sb); 2766 if (trans) 2767 btrfs_end_transaction(trans); 2768 error_free_zone: 2769 btrfs_destroy_dev_zone_info(device); 2770 error_free_device: 2771 btrfs_free_device(device); 2772 error: 2773 blkdev_put(bdev, FMODE_EXCL); 2774 if (locked) { 2775 mutex_unlock(&uuid_mutex); 2776 up_write(&sb->s_umount); 2777 } 2778 return ret; 2779 } 2780 2781 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2782 struct btrfs_device *device) 2783 { 2784 int ret; 2785 struct btrfs_path *path; 2786 struct btrfs_root *root = device->fs_info->chunk_root; 2787 struct btrfs_dev_item *dev_item; 2788 struct extent_buffer *leaf; 2789 struct btrfs_key key; 2790 2791 path = btrfs_alloc_path(); 2792 if (!path) 2793 return -ENOMEM; 2794 2795 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2796 key.type = BTRFS_DEV_ITEM_KEY; 2797 key.offset = device->devid; 2798 2799 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2800 if (ret < 0) 2801 goto out; 2802 2803 if (ret > 0) { 2804 ret = -ENOENT; 2805 goto out; 2806 } 2807 2808 leaf = path->nodes[0]; 2809 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2810 2811 btrfs_set_device_id(leaf, dev_item, device->devid); 2812 btrfs_set_device_type(leaf, dev_item, device->type); 2813 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2814 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2815 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2816 btrfs_set_device_total_bytes(leaf, dev_item, 2817 btrfs_device_get_disk_total_bytes(device)); 2818 btrfs_set_device_bytes_used(leaf, dev_item, 2819 btrfs_device_get_bytes_used(device)); 2820 btrfs_mark_buffer_dirty(leaf); 2821 2822 out: 2823 btrfs_free_path(path); 2824 return ret; 2825 } 2826 2827 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2828 struct btrfs_device *device, u64 new_size) 2829 { 2830 struct btrfs_fs_info *fs_info = device->fs_info; 2831 struct btrfs_super_block *super_copy = fs_info->super_copy; 2832 u64 old_total; 2833 u64 diff; 2834 2835 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2836 return -EACCES; 2837 2838 new_size = round_down(new_size, fs_info->sectorsize); 2839 2840 mutex_lock(&fs_info->chunk_mutex); 2841 old_total = btrfs_super_total_bytes(super_copy); 2842 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2843 2844 if (new_size <= device->total_bytes || 2845 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2846 mutex_unlock(&fs_info->chunk_mutex); 2847 return -EINVAL; 2848 } 2849 2850 btrfs_set_super_total_bytes(super_copy, 2851 round_down(old_total + diff, fs_info->sectorsize)); 2852 device->fs_devices->total_rw_bytes += diff; 2853 2854 btrfs_device_set_total_bytes(device, new_size); 2855 btrfs_device_set_disk_total_bytes(device, new_size); 2856 btrfs_clear_space_info_full(device->fs_info); 2857 if (list_empty(&device->post_commit_list)) 2858 list_add_tail(&device->post_commit_list, 2859 &trans->transaction->dev_update_list); 2860 mutex_unlock(&fs_info->chunk_mutex); 2861 2862 return btrfs_update_device(trans, device); 2863 } 2864 2865 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2866 { 2867 struct btrfs_fs_info *fs_info = trans->fs_info; 2868 struct btrfs_root *root = fs_info->chunk_root; 2869 int ret; 2870 struct btrfs_path *path; 2871 struct btrfs_key key; 2872 2873 path = btrfs_alloc_path(); 2874 if (!path) 2875 return -ENOMEM; 2876 2877 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2878 key.offset = chunk_offset; 2879 key.type = BTRFS_CHUNK_ITEM_KEY; 2880 2881 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2882 if (ret < 0) 2883 goto out; 2884 else if (ret > 0) { /* Logic error or corruption */ 2885 btrfs_handle_fs_error(fs_info, -ENOENT, 2886 "Failed lookup while freeing chunk."); 2887 ret = -ENOENT; 2888 goto out; 2889 } 2890 2891 ret = btrfs_del_item(trans, root, path); 2892 if (ret < 0) 2893 btrfs_handle_fs_error(fs_info, ret, 2894 "Failed to delete chunk item."); 2895 out: 2896 btrfs_free_path(path); 2897 return ret; 2898 } 2899 2900 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2901 { 2902 struct btrfs_super_block *super_copy = fs_info->super_copy; 2903 struct btrfs_disk_key *disk_key; 2904 struct btrfs_chunk *chunk; 2905 u8 *ptr; 2906 int ret = 0; 2907 u32 num_stripes; 2908 u32 array_size; 2909 u32 len = 0; 2910 u32 cur; 2911 struct btrfs_key key; 2912 2913 lockdep_assert_held(&fs_info->chunk_mutex); 2914 array_size = btrfs_super_sys_array_size(super_copy); 2915 2916 ptr = super_copy->sys_chunk_array; 2917 cur = 0; 2918 2919 while (cur < array_size) { 2920 disk_key = (struct btrfs_disk_key *)ptr; 2921 btrfs_disk_key_to_cpu(&key, disk_key); 2922 2923 len = sizeof(*disk_key); 2924 2925 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 2926 chunk = (struct btrfs_chunk *)(ptr + len); 2927 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 2928 len += btrfs_chunk_item_size(num_stripes); 2929 } else { 2930 ret = -EIO; 2931 break; 2932 } 2933 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 2934 key.offset == chunk_offset) { 2935 memmove(ptr, ptr + len, array_size - (cur + len)); 2936 array_size -= len; 2937 btrfs_set_super_sys_array_size(super_copy, array_size); 2938 } else { 2939 ptr += len; 2940 cur += len; 2941 } 2942 } 2943 return ret; 2944 } 2945 2946 /* 2947 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 2948 * @logical: Logical block offset in bytes. 2949 * @length: Length of extent in bytes. 2950 * 2951 * Return: Chunk mapping or ERR_PTR. 2952 */ 2953 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 2954 u64 logical, u64 length) 2955 { 2956 struct extent_map_tree *em_tree; 2957 struct extent_map *em; 2958 2959 em_tree = &fs_info->mapping_tree; 2960 read_lock(&em_tree->lock); 2961 em = lookup_extent_mapping(em_tree, logical, length); 2962 read_unlock(&em_tree->lock); 2963 2964 if (!em) { 2965 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 2966 logical, length); 2967 return ERR_PTR(-EINVAL); 2968 } 2969 2970 if (em->start > logical || em->start + em->len < logical) { 2971 btrfs_crit(fs_info, 2972 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 2973 logical, length, em->start, em->start + em->len); 2974 free_extent_map(em); 2975 return ERR_PTR(-EINVAL); 2976 } 2977 2978 /* callers are responsible for dropping em's ref. */ 2979 return em; 2980 } 2981 2982 static int remove_chunk_item(struct btrfs_trans_handle *trans, 2983 struct map_lookup *map, u64 chunk_offset) 2984 { 2985 int i; 2986 2987 /* 2988 * Removing chunk items and updating the device items in the chunks btree 2989 * requires holding the chunk_mutex. 2990 * See the comment at btrfs_chunk_alloc() for the details. 2991 */ 2992 lockdep_assert_held(&trans->fs_info->chunk_mutex); 2993 2994 for (i = 0; i < map->num_stripes; i++) { 2995 int ret; 2996 2997 ret = btrfs_update_device(trans, map->stripes[i].dev); 2998 if (ret) 2999 return ret; 3000 } 3001 3002 return btrfs_free_chunk(trans, chunk_offset); 3003 } 3004 3005 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3006 { 3007 struct btrfs_fs_info *fs_info = trans->fs_info; 3008 struct extent_map *em; 3009 struct map_lookup *map; 3010 u64 dev_extent_len = 0; 3011 int i, ret = 0; 3012 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3013 3014 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3015 if (IS_ERR(em)) { 3016 /* 3017 * This is a logic error, but we don't want to just rely on the 3018 * user having built with ASSERT enabled, so if ASSERT doesn't 3019 * do anything we still error out. 3020 */ 3021 ASSERT(0); 3022 return PTR_ERR(em); 3023 } 3024 map = em->map_lookup; 3025 3026 /* 3027 * First delete the device extent items from the devices btree. 3028 * We take the device_list_mutex to avoid racing with the finishing phase 3029 * of a device replace operation. See the comment below before acquiring 3030 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3031 * because that can result in a deadlock when deleting the device extent 3032 * items from the devices btree - COWing an extent buffer from the btree 3033 * may result in allocating a new metadata chunk, which would attempt to 3034 * lock again fs_info->chunk_mutex. 3035 */ 3036 mutex_lock(&fs_devices->device_list_mutex); 3037 for (i = 0; i < map->num_stripes; i++) { 3038 struct btrfs_device *device = map->stripes[i].dev; 3039 ret = btrfs_free_dev_extent(trans, device, 3040 map->stripes[i].physical, 3041 &dev_extent_len); 3042 if (ret) { 3043 mutex_unlock(&fs_devices->device_list_mutex); 3044 btrfs_abort_transaction(trans, ret); 3045 goto out; 3046 } 3047 3048 if (device->bytes_used > 0) { 3049 mutex_lock(&fs_info->chunk_mutex); 3050 btrfs_device_set_bytes_used(device, 3051 device->bytes_used - dev_extent_len); 3052 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3053 btrfs_clear_space_info_full(fs_info); 3054 mutex_unlock(&fs_info->chunk_mutex); 3055 } 3056 } 3057 mutex_unlock(&fs_devices->device_list_mutex); 3058 3059 /* 3060 * We acquire fs_info->chunk_mutex for 2 reasons: 3061 * 3062 * 1) Just like with the first phase of the chunk allocation, we must 3063 * reserve system space, do all chunk btree updates and deletions, and 3064 * update the system chunk array in the superblock while holding this 3065 * mutex. This is for similar reasons as explained on the comment at 3066 * the top of btrfs_chunk_alloc(); 3067 * 3068 * 2) Prevent races with the final phase of a device replace operation 3069 * that replaces the device object associated with the map's stripes, 3070 * because the device object's id can change at any time during that 3071 * final phase of the device replace operation 3072 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3073 * replaced device and then see it with an ID of 3074 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3075 * the device item, which does not exists on the chunk btree. 3076 * The finishing phase of device replace acquires both the 3077 * device_list_mutex and the chunk_mutex, in that order, so we are 3078 * safe by just acquiring the chunk_mutex. 3079 */ 3080 trans->removing_chunk = true; 3081 mutex_lock(&fs_info->chunk_mutex); 3082 3083 check_system_chunk(trans, map->type); 3084 3085 ret = remove_chunk_item(trans, map, chunk_offset); 3086 /* 3087 * Normally we should not get -ENOSPC since we reserved space before 3088 * through the call to check_system_chunk(). 3089 * 3090 * Despite our system space_info having enough free space, we may not 3091 * be able to allocate extents from its block groups, because all have 3092 * an incompatible profile, which will force us to allocate a new system 3093 * block group with the right profile, or right after we called 3094 * check_system_space() above, a scrub turned the only system block group 3095 * with enough free space into RO mode. 3096 * This is explained with more detail at do_chunk_alloc(). 3097 * 3098 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3099 */ 3100 if (ret == -ENOSPC) { 3101 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3102 struct btrfs_block_group *sys_bg; 3103 3104 sys_bg = btrfs_create_chunk(trans, sys_flags); 3105 if (IS_ERR(sys_bg)) { 3106 ret = PTR_ERR(sys_bg); 3107 btrfs_abort_transaction(trans, ret); 3108 goto out; 3109 } 3110 3111 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3112 if (ret) { 3113 btrfs_abort_transaction(trans, ret); 3114 goto out; 3115 } 3116 3117 ret = remove_chunk_item(trans, map, chunk_offset); 3118 if (ret) { 3119 btrfs_abort_transaction(trans, ret); 3120 goto out; 3121 } 3122 } else if (ret) { 3123 btrfs_abort_transaction(trans, ret); 3124 goto out; 3125 } 3126 3127 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3128 3129 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3130 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3131 if (ret) { 3132 btrfs_abort_transaction(trans, ret); 3133 goto out; 3134 } 3135 } 3136 3137 mutex_unlock(&fs_info->chunk_mutex); 3138 trans->removing_chunk = false; 3139 3140 /* 3141 * We are done with chunk btree updates and deletions, so release the 3142 * system space we previously reserved (with check_system_chunk()). 3143 */ 3144 btrfs_trans_release_chunk_metadata(trans); 3145 3146 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3147 if (ret) { 3148 btrfs_abort_transaction(trans, ret); 3149 goto out; 3150 } 3151 3152 out: 3153 if (trans->removing_chunk) { 3154 mutex_unlock(&fs_info->chunk_mutex); 3155 trans->removing_chunk = false; 3156 } 3157 /* once for us */ 3158 free_extent_map(em); 3159 return ret; 3160 } 3161 3162 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3163 { 3164 struct btrfs_root *root = fs_info->chunk_root; 3165 struct btrfs_trans_handle *trans; 3166 struct btrfs_block_group *block_group; 3167 u64 length; 3168 int ret; 3169 3170 /* 3171 * Prevent races with automatic removal of unused block groups. 3172 * After we relocate and before we remove the chunk with offset 3173 * chunk_offset, automatic removal of the block group can kick in, 3174 * resulting in a failure when calling btrfs_remove_chunk() below. 3175 * 3176 * Make sure to acquire this mutex before doing a tree search (dev 3177 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3178 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3179 * we release the path used to search the chunk/dev tree and before 3180 * the current task acquires this mutex and calls us. 3181 */ 3182 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3183 3184 /* step one, relocate all the extents inside this chunk */ 3185 btrfs_scrub_pause(fs_info); 3186 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3187 btrfs_scrub_continue(fs_info); 3188 if (ret) 3189 return ret; 3190 3191 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3192 if (!block_group) 3193 return -ENOENT; 3194 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3195 length = block_group->length; 3196 btrfs_put_block_group(block_group); 3197 3198 /* 3199 * On a zoned file system, discard the whole block group, this will 3200 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3201 * resetting the zone fails, don't treat it as a fatal problem from the 3202 * filesystem's point of view. 3203 */ 3204 if (btrfs_is_zoned(fs_info)) { 3205 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3206 if (ret) 3207 btrfs_info(fs_info, 3208 "failed to reset zone %llu after relocation", 3209 chunk_offset); 3210 } 3211 3212 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3213 chunk_offset); 3214 if (IS_ERR(trans)) { 3215 ret = PTR_ERR(trans); 3216 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3217 return ret; 3218 } 3219 3220 /* 3221 * step two, delete the device extents and the 3222 * chunk tree entries 3223 */ 3224 ret = btrfs_remove_chunk(trans, chunk_offset); 3225 btrfs_end_transaction(trans); 3226 return ret; 3227 } 3228 3229 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3230 { 3231 struct btrfs_root *chunk_root = fs_info->chunk_root; 3232 struct btrfs_path *path; 3233 struct extent_buffer *leaf; 3234 struct btrfs_chunk *chunk; 3235 struct btrfs_key key; 3236 struct btrfs_key found_key; 3237 u64 chunk_type; 3238 bool retried = false; 3239 int failed = 0; 3240 int ret; 3241 3242 path = btrfs_alloc_path(); 3243 if (!path) 3244 return -ENOMEM; 3245 3246 again: 3247 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3248 key.offset = (u64)-1; 3249 key.type = BTRFS_CHUNK_ITEM_KEY; 3250 3251 while (1) { 3252 mutex_lock(&fs_info->reclaim_bgs_lock); 3253 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3254 if (ret < 0) { 3255 mutex_unlock(&fs_info->reclaim_bgs_lock); 3256 goto error; 3257 } 3258 BUG_ON(ret == 0); /* Corruption */ 3259 3260 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3261 key.type); 3262 if (ret) 3263 mutex_unlock(&fs_info->reclaim_bgs_lock); 3264 if (ret < 0) 3265 goto error; 3266 if (ret > 0) 3267 break; 3268 3269 leaf = path->nodes[0]; 3270 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3271 3272 chunk = btrfs_item_ptr(leaf, path->slots[0], 3273 struct btrfs_chunk); 3274 chunk_type = btrfs_chunk_type(leaf, chunk); 3275 btrfs_release_path(path); 3276 3277 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3278 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3279 if (ret == -ENOSPC) 3280 failed++; 3281 else 3282 BUG_ON(ret); 3283 } 3284 mutex_unlock(&fs_info->reclaim_bgs_lock); 3285 3286 if (found_key.offset == 0) 3287 break; 3288 key.offset = found_key.offset - 1; 3289 } 3290 ret = 0; 3291 if (failed && !retried) { 3292 failed = 0; 3293 retried = true; 3294 goto again; 3295 } else if (WARN_ON(failed && retried)) { 3296 ret = -ENOSPC; 3297 } 3298 error: 3299 btrfs_free_path(path); 3300 return ret; 3301 } 3302 3303 /* 3304 * return 1 : allocate a data chunk successfully, 3305 * return <0: errors during allocating a data chunk, 3306 * return 0 : no need to allocate a data chunk. 3307 */ 3308 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3309 u64 chunk_offset) 3310 { 3311 struct btrfs_block_group *cache; 3312 u64 bytes_used; 3313 u64 chunk_type; 3314 3315 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3316 ASSERT(cache); 3317 chunk_type = cache->flags; 3318 btrfs_put_block_group(cache); 3319 3320 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3321 return 0; 3322 3323 spin_lock(&fs_info->data_sinfo->lock); 3324 bytes_used = fs_info->data_sinfo->bytes_used; 3325 spin_unlock(&fs_info->data_sinfo->lock); 3326 3327 if (!bytes_used) { 3328 struct btrfs_trans_handle *trans; 3329 int ret; 3330 3331 trans = btrfs_join_transaction(fs_info->tree_root); 3332 if (IS_ERR(trans)) 3333 return PTR_ERR(trans); 3334 3335 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3336 btrfs_end_transaction(trans); 3337 if (ret < 0) 3338 return ret; 3339 return 1; 3340 } 3341 3342 return 0; 3343 } 3344 3345 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3346 struct btrfs_balance_control *bctl) 3347 { 3348 struct btrfs_root *root = fs_info->tree_root; 3349 struct btrfs_trans_handle *trans; 3350 struct btrfs_balance_item *item; 3351 struct btrfs_disk_balance_args disk_bargs; 3352 struct btrfs_path *path; 3353 struct extent_buffer *leaf; 3354 struct btrfs_key key; 3355 int ret, err; 3356 3357 path = btrfs_alloc_path(); 3358 if (!path) 3359 return -ENOMEM; 3360 3361 trans = btrfs_start_transaction(root, 0); 3362 if (IS_ERR(trans)) { 3363 btrfs_free_path(path); 3364 return PTR_ERR(trans); 3365 } 3366 3367 key.objectid = BTRFS_BALANCE_OBJECTID; 3368 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3369 key.offset = 0; 3370 3371 ret = btrfs_insert_empty_item(trans, root, path, &key, 3372 sizeof(*item)); 3373 if (ret) 3374 goto out; 3375 3376 leaf = path->nodes[0]; 3377 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3378 3379 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3380 3381 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3382 btrfs_set_balance_data(leaf, item, &disk_bargs); 3383 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3384 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3385 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3386 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3387 3388 btrfs_set_balance_flags(leaf, item, bctl->flags); 3389 3390 btrfs_mark_buffer_dirty(leaf); 3391 out: 3392 btrfs_free_path(path); 3393 err = btrfs_commit_transaction(trans); 3394 if (err && !ret) 3395 ret = err; 3396 return ret; 3397 } 3398 3399 static int del_balance_item(struct btrfs_fs_info *fs_info) 3400 { 3401 struct btrfs_root *root = fs_info->tree_root; 3402 struct btrfs_trans_handle *trans; 3403 struct btrfs_path *path; 3404 struct btrfs_key key; 3405 int ret, err; 3406 3407 path = btrfs_alloc_path(); 3408 if (!path) 3409 return -ENOMEM; 3410 3411 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3412 if (IS_ERR(trans)) { 3413 btrfs_free_path(path); 3414 return PTR_ERR(trans); 3415 } 3416 3417 key.objectid = BTRFS_BALANCE_OBJECTID; 3418 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3419 key.offset = 0; 3420 3421 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3422 if (ret < 0) 3423 goto out; 3424 if (ret > 0) { 3425 ret = -ENOENT; 3426 goto out; 3427 } 3428 3429 ret = btrfs_del_item(trans, root, path); 3430 out: 3431 btrfs_free_path(path); 3432 err = btrfs_commit_transaction(trans); 3433 if (err && !ret) 3434 ret = err; 3435 return ret; 3436 } 3437 3438 /* 3439 * This is a heuristic used to reduce the number of chunks balanced on 3440 * resume after balance was interrupted. 3441 */ 3442 static void update_balance_args(struct btrfs_balance_control *bctl) 3443 { 3444 /* 3445 * Turn on soft mode for chunk types that were being converted. 3446 */ 3447 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3448 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3449 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3450 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3451 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3452 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3453 3454 /* 3455 * Turn on usage filter if is not already used. The idea is 3456 * that chunks that we have already balanced should be 3457 * reasonably full. Don't do it for chunks that are being 3458 * converted - that will keep us from relocating unconverted 3459 * (albeit full) chunks. 3460 */ 3461 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3462 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3463 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3464 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3465 bctl->data.usage = 90; 3466 } 3467 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3468 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3469 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3470 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3471 bctl->sys.usage = 90; 3472 } 3473 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3474 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3475 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3476 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3477 bctl->meta.usage = 90; 3478 } 3479 } 3480 3481 /* 3482 * Clear the balance status in fs_info and delete the balance item from disk. 3483 */ 3484 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3485 { 3486 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3487 int ret; 3488 3489 BUG_ON(!fs_info->balance_ctl); 3490 3491 spin_lock(&fs_info->balance_lock); 3492 fs_info->balance_ctl = NULL; 3493 spin_unlock(&fs_info->balance_lock); 3494 3495 kfree(bctl); 3496 ret = del_balance_item(fs_info); 3497 if (ret) 3498 btrfs_handle_fs_error(fs_info, ret, NULL); 3499 } 3500 3501 /* 3502 * Balance filters. Return 1 if chunk should be filtered out 3503 * (should not be balanced). 3504 */ 3505 static int chunk_profiles_filter(u64 chunk_type, 3506 struct btrfs_balance_args *bargs) 3507 { 3508 chunk_type = chunk_to_extended(chunk_type) & 3509 BTRFS_EXTENDED_PROFILE_MASK; 3510 3511 if (bargs->profiles & chunk_type) 3512 return 0; 3513 3514 return 1; 3515 } 3516 3517 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3518 struct btrfs_balance_args *bargs) 3519 { 3520 struct btrfs_block_group *cache; 3521 u64 chunk_used; 3522 u64 user_thresh_min; 3523 u64 user_thresh_max; 3524 int ret = 1; 3525 3526 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3527 chunk_used = cache->used; 3528 3529 if (bargs->usage_min == 0) 3530 user_thresh_min = 0; 3531 else 3532 user_thresh_min = div_factor_fine(cache->length, 3533 bargs->usage_min); 3534 3535 if (bargs->usage_max == 0) 3536 user_thresh_max = 1; 3537 else if (bargs->usage_max > 100) 3538 user_thresh_max = cache->length; 3539 else 3540 user_thresh_max = div_factor_fine(cache->length, 3541 bargs->usage_max); 3542 3543 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3544 ret = 0; 3545 3546 btrfs_put_block_group(cache); 3547 return ret; 3548 } 3549 3550 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3551 u64 chunk_offset, struct btrfs_balance_args *bargs) 3552 { 3553 struct btrfs_block_group *cache; 3554 u64 chunk_used, user_thresh; 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 = 1; 3562 else if (bargs->usage > 100) 3563 user_thresh = cache->length; 3564 else 3565 user_thresh = div_factor_fine(cache->length, bargs->usage); 3566 3567 if (chunk_used < user_thresh) 3568 ret = 0; 3569 3570 btrfs_put_block_group(cache); 3571 return ret; 3572 } 3573 3574 static int chunk_devid_filter(struct extent_buffer *leaf, 3575 struct btrfs_chunk *chunk, 3576 struct btrfs_balance_args *bargs) 3577 { 3578 struct btrfs_stripe *stripe; 3579 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3580 int i; 3581 3582 for (i = 0; i < num_stripes; i++) { 3583 stripe = btrfs_stripe_nr(chunk, i); 3584 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3585 return 0; 3586 } 3587 3588 return 1; 3589 } 3590 3591 static u64 calc_data_stripes(u64 type, int num_stripes) 3592 { 3593 const int index = btrfs_bg_flags_to_raid_index(type); 3594 const int ncopies = btrfs_raid_array[index].ncopies; 3595 const int nparity = btrfs_raid_array[index].nparity; 3596 3597 return (num_stripes - nparity) / ncopies; 3598 } 3599 3600 /* [pstart, pend) */ 3601 static int chunk_drange_filter(struct extent_buffer *leaf, 3602 struct btrfs_chunk *chunk, 3603 struct btrfs_balance_args *bargs) 3604 { 3605 struct btrfs_stripe *stripe; 3606 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3607 u64 stripe_offset; 3608 u64 stripe_length; 3609 u64 type; 3610 int factor; 3611 int i; 3612 3613 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3614 return 0; 3615 3616 type = btrfs_chunk_type(leaf, chunk); 3617 factor = calc_data_stripes(type, num_stripes); 3618 3619 for (i = 0; i < num_stripes; i++) { 3620 stripe = btrfs_stripe_nr(chunk, i); 3621 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3622 continue; 3623 3624 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3625 stripe_length = btrfs_chunk_length(leaf, chunk); 3626 stripe_length = div_u64(stripe_length, factor); 3627 3628 if (stripe_offset < bargs->pend && 3629 stripe_offset + stripe_length > bargs->pstart) 3630 return 0; 3631 } 3632 3633 return 1; 3634 } 3635 3636 /* [vstart, vend) */ 3637 static int chunk_vrange_filter(struct extent_buffer *leaf, 3638 struct btrfs_chunk *chunk, 3639 u64 chunk_offset, 3640 struct btrfs_balance_args *bargs) 3641 { 3642 if (chunk_offset < bargs->vend && 3643 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3644 /* at least part of the chunk is inside this vrange */ 3645 return 0; 3646 3647 return 1; 3648 } 3649 3650 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3651 struct btrfs_chunk *chunk, 3652 struct btrfs_balance_args *bargs) 3653 { 3654 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3655 3656 if (bargs->stripes_min <= num_stripes 3657 && num_stripes <= bargs->stripes_max) 3658 return 0; 3659 3660 return 1; 3661 } 3662 3663 static int chunk_soft_convert_filter(u64 chunk_type, 3664 struct btrfs_balance_args *bargs) 3665 { 3666 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3667 return 0; 3668 3669 chunk_type = chunk_to_extended(chunk_type) & 3670 BTRFS_EXTENDED_PROFILE_MASK; 3671 3672 if (bargs->target == chunk_type) 3673 return 1; 3674 3675 return 0; 3676 } 3677 3678 static int should_balance_chunk(struct extent_buffer *leaf, 3679 struct btrfs_chunk *chunk, u64 chunk_offset) 3680 { 3681 struct btrfs_fs_info *fs_info = leaf->fs_info; 3682 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3683 struct btrfs_balance_args *bargs = NULL; 3684 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3685 3686 /* type filter */ 3687 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3688 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3689 return 0; 3690 } 3691 3692 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3693 bargs = &bctl->data; 3694 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3695 bargs = &bctl->sys; 3696 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3697 bargs = &bctl->meta; 3698 3699 /* profiles filter */ 3700 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3701 chunk_profiles_filter(chunk_type, bargs)) { 3702 return 0; 3703 } 3704 3705 /* usage filter */ 3706 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3707 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3708 return 0; 3709 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3710 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3711 return 0; 3712 } 3713 3714 /* devid filter */ 3715 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3716 chunk_devid_filter(leaf, chunk, bargs)) { 3717 return 0; 3718 } 3719 3720 /* drange filter, makes sense only with devid filter */ 3721 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3722 chunk_drange_filter(leaf, chunk, bargs)) { 3723 return 0; 3724 } 3725 3726 /* vrange filter */ 3727 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3728 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3729 return 0; 3730 } 3731 3732 /* stripes filter */ 3733 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3734 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3735 return 0; 3736 } 3737 3738 /* soft profile changing mode */ 3739 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3740 chunk_soft_convert_filter(chunk_type, bargs)) { 3741 return 0; 3742 } 3743 3744 /* 3745 * limited by count, must be the last filter 3746 */ 3747 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3748 if (bargs->limit == 0) 3749 return 0; 3750 else 3751 bargs->limit--; 3752 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3753 /* 3754 * Same logic as the 'limit' filter; the minimum cannot be 3755 * determined here because we do not have the global information 3756 * about the count of all chunks that satisfy the filters. 3757 */ 3758 if (bargs->limit_max == 0) 3759 return 0; 3760 else 3761 bargs->limit_max--; 3762 } 3763 3764 return 1; 3765 } 3766 3767 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3768 { 3769 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3770 struct btrfs_root *chunk_root = fs_info->chunk_root; 3771 u64 chunk_type; 3772 struct btrfs_chunk *chunk; 3773 struct btrfs_path *path = NULL; 3774 struct btrfs_key key; 3775 struct btrfs_key found_key; 3776 struct extent_buffer *leaf; 3777 int slot; 3778 int ret; 3779 int enospc_errors = 0; 3780 bool counting = true; 3781 /* The single value limit and min/max limits use the same bytes in the */ 3782 u64 limit_data = bctl->data.limit; 3783 u64 limit_meta = bctl->meta.limit; 3784 u64 limit_sys = bctl->sys.limit; 3785 u32 count_data = 0; 3786 u32 count_meta = 0; 3787 u32 count_sys = 0; 3788 int chunk_reserved = 0; 3789 3790 path = btrfs_alloc_path(); 3791 if (!path) { 3792 ret = -ENOMEM; 3793 goto error; 3794 } 3795 3796 /* zero out stat counters */ 3797 spin_lock(&fs_info->balance_lock); 3798 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3799 spin_unlock(&fs_info->balance_lock); 3800 again: 3801 if (!counting) { 3802 /* 3803 * The single value limit and min/max limits use the same bytes 3804 * in the 3805 */ 3806 bctl->data.limit = limit_data; 3807 bctl->meta.limit = limit_meta; 3808 bctl->sys.limit = limit_sys; 3809 } 3810 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3811 key.offset = (u64)-1; 3812 key.type = BTRFS_CHUNK_ITEM_KEY; 3813 3814 while (1) { 3815 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3816 atomic_read(&fs_info->balance_cancel_req)) { 3817 ret = -ECANCELED; 3818 goto error; 3819 } 3820 3821 mutex_lock(&fs_info->reclaim_bgs_lock); 3822 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3823 if (ret < 0) { 3824 mutex_unlock(&fs_info->reclaim_bgs_lock); 3825 goto error; 3826 } 3827 3828 /* 3829 * this shouldn't happen, it means the last relocate 3830 * failed 3831 */ 3832 if (ret == 0) 3833 BUG(); /* FIXME break ? */ 3834 3835 ret = btrfs_previous_item(chunk_root, path, 0, 3836 BTRFS_CHUNK_ITEM_KEY); 3837 if (ret) { 3838 mutex_unlock(&fs_info->reclaim_bgs_lock); 3839 ret = 0; 3840 break; 3841 } 3842 3843 leaf = path->nodes[0]; 3844 slot = path->slots[0]; 3845 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3846 3847 if (found_key.objectid != key.objectid) { 3848 mutex_unlock(&fs_info->reclaim_bgs_lock); 3849 break; 3850 } 3851 3852 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3853 chunk_type = btrfs_chunk_type(leaf, chunk); 3854 3855 if (!counting) { 3856 spin_lock(&fs_info->balance_lock); 3857 bctl->stat.considered++; 3858 spin_unlock(&fs_info->balance_lock); 3859 } 3860 3861 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3862 3863 btrfs_release_path(path); 3864 if (!ret) { 3865 mutex_unlock(&fs_info->reclaim_bgs_lock); 3866 goto loop; 3867 } 3868 3869 if (counting) { 3870 mutex_unlock(&fs_info->reclaim_bgs_lock); 3871 spin_lock(&fs_info->balance_lock); 3872 bctl->stat.expected++; 3873 spin_unlock(&fs_info->balance_lock); 3874 3875 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3876 count_data++; 3877 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3878 count_sys++; 3879 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3880 count_meta++; 3881 3882 goto loop; 3883 } 3884 3885 /* 3886 * Apply limit_min filter, no need to check if the LIMITS 3887 * filter is used, limit_min is 0 by default 3888 */ 3889 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3890 count_data < bctl->data.limit_min) 3891 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3892 count_meta < bctl->meta.limit_min) 3893 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3894 count_sys < bctl->sys.limit_min)) { 3895 mutex_unlock(&fs_info->reclaim_bgs_lock); 3896 goto loop; 3897 } 3898 3899 if (!chunk_reserved) { 3900 /* 3901 * We may be relocating the only data chunk we have, 3902 * which could potentially end up with losing data's 3903 * raid profile, so lets allocate an empty one in 3904 * advance. 3905 */ 3906 ret = btrfs_may_alloc_data_chunk(fs_info, 3907 found_key.offset); 3908 if (ret < 0) { 3909 mutex_unlock(&fs_info->reclaim_bgs_lock); 3910 goto error; 3911 } else if (ret == 1) { 3912 chunk_reserved = 1; 3913 } 3914 } 3915 3916 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3917 mutex_unlock(&fs_info->reclaim_bgs_lock); 3918 if (ret == -ENOSPC) { 3919 enospc_errors++; 3920 } else if (ret == -ETXTBSY) { 3921 btrfs_info(fs_info, 3922 "skipping relocation of block group %llu due to active swapfile", 3923 found_key.offset); 3924 ret = 0; 3925 } else if (ret) { 3926 goto error; 3927 } else { 3928 spin_lock(&fs_info->balance_lock); 3929 bctl->stat.completed++; 3930 spin_unlock(&fs_info->balance_lock); 3931 } 3932 loop: 3933 if (found_key.offset == 0) 3934 break; 3935 key.offset = found_key.offset - 1; 3936 } 3937 3938 if (counting) { 3939 btrfs_release_path(path); 3940 counting = false; 3941 goto again; 3942 } 3943 error: 3944 btrfs_free_path(path); 3945 if (enospc_errors) { 3946 btrfs_info(fs_info, "%d enospc errors during balance", 3947 enospc_errors); 3948 if (!ret) 3949 ret = -ENOSPC; 3950 } 3951 3952 return ret; 3953 } 3954 3955 /** 3956 * alloc_profile_is_valid - see if a given profile is valid and reduced 3957 * @flags: profile to validate 3958 * @extended: if true @flags is treated as an extended profile 3959 */ 3960 static int alloc_profile_is_valid(u64 flags, int extended) 3961 { 3962 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 3963 BTRFS_BLOCK_GROUP_PROFILE_MASK); 3964 3965 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 3966 3967 /* 1) check that all other bits are zeroed */ 3968 if (flags & ~mask) 3969 return 0; 3970 3971 /* 2) see if profile is reduced */ 3972 if (flags == 0) 3973 return !extended; /* "0" is valid for usual profiles */ 3974 3975 return has_single_bit_set(flags); 3976 } 3977 3978 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 3979 { 3980 /* cancel requested || normal exit path */ 3981 return atomic_read(&fs_info->balance_cancel_req) || 3982 (atomic_read(&fs_info->balance_pause_req) == 0 && 3983 atomic_read(&fs_info->balance_cancel_req) == 0); 3984 } 3985 3986 /* 3987 * Validate target profile against allowed profiles and return true if it's OK. 3988 * Otherwise print the error message and return false. 3989 */ 3990 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 3991 const struct btrfs_balance_args *bargs, 3992 u64 allowed, const char *type) 3993 { 3994 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3995 return true; 3996 3997 if (fs_info->sectorsize < PAGE_SIZE && 3998 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) { 3999 btrfs_err(fs_info, 4000 "RAID56 is not yet supported for sectorsize %u with page size %lu", 4001 fs_info->sectorsize, PAGE_SIZE); 4002 return false; 4003 } 4004 /* Profile is valid and does not have bits outside of the allowed set */ 4005 if (alloc_profile_is_valid(bargs->target, 1) && 4006 (bargs->target & ~allowed) == 0) 4007 return true; 4008 4009 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4010 type, btrfs_bg_type_to_raid_name(bargs->target)); 4011 return false; 4012 } 4013 4014 /* 4015 * Fill @buf with textual description of balance filter flags @bargs, up to 4016 * @size_buf including the terminating null. The output may be trimmed if it 4017 * does not fit into the provided buffer. 4018 */ 4019 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4020 u32 size_buf) 4021 { 4022 int ret; 4023 u32 size_bp = size_buf; 4024 char *bp = buf; 4025 u64 flags = bargs->flags; 4026 char tmp_buf[128] = {'\0'}; 4027 4028 if (!flags) 4029 return; 4030 4031 #define CHECK_APPEND_NOARG(a) \ 4032 do { \ 4033 ret = snprintf(bp, size_bp, (a)); \ 4034 if (ret < 0 || ret >= size_bp) \ 4035 goto out_overflow; \ 4036 size_bp -= ret; \ 4037 bp += ret; \ 4038 } while (0) 4039 4040 #define CHECK_APPEND_1ARG(a, v1) \ 4041 do { \ 4042 ret = snprintf(bp, size_bp, (a), (v1)); \ 4043 if (ret < 0 || ret >= size_bp) \ 4044 goto out_overflow; \ 4045 size_bp -= ret; \ 4046 bp += ret; \ 4047 } while (0) 4048 4049 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4050 do { \ 4051 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4052 if (ret < 0 || ret >= size_bp) \ 4053 goto out_overflow; \ 4054 size_bp -= ret; \ 4055 bp += ret; \ 4056 } while (0) 4057 4058 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4059 CHECK_APPEND_1ARG("convert=%s,", 4060 btrfs_bg_type_to_raid_name(bargs->target)); 4061 4062 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4063 CHECK_APPEND_NOARG("soft,"); 4064 4065 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4066 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4067 sizeof(tmp_buf)); 4068 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4069 } 4070 4071 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4072 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4073 4074 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4075 CHECK_APPEND_2ARG("usage=%u..%u,", 4076 bargs->usage_min, bargs->usage_max); 4077 4078 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4079 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4080 4081 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4082 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4083 bargs->pstart, bargs->pend); 4084 4085 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4086 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4087 bargs->vstart, bargs->vend); 4088 4089 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4090 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4091 4092 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4093 CHECK_APPEND_2ARG("limit=%u..%u,", 4094 bargs->limit_min, bargs->limit_max); 4095 4096 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4097 CHECK_APPEND_2ARG("stripes=%u..%u,", 4098 bargs->stripes_min, bargs->stripes_max); 4099 4100 #undef CHECK_APPEND_2ARG 4101 #undef CHECK_APPEND_1ARG 4102 #undef CHECK_APPEND_NOARG 4103 4104 out_overflow: 4105 4106 if (size_bp < size_buf) 4107 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4108 else 4109 buf[0] = '\0'; 4110 } 4111 4112 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4113 { 4114 u32 size_buf = 1024; 4115 char tmp_buf[192] = {'\0'}; 4116 char *buf; 4117 char *bp; 4118 u32 size_bp = size_buf; 4119 int ret; 4120 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4121 4122 buf = kzalloc(size_buf, GFP_KERNEL); 4123 if (!buf) 4124 return; 4125 4126 bp = buf; 4127 4128 #define CHECK_APPEND_1ARG(a, v1) \ 4129 do { \ 4130 ret = snprintf(bp, size_bp, (a), (v1)); \ 4131 if (ret < 0 || ret >= size_bp) \ 4132 goto out_overflow; \ 4133 size_bp -= ret; \ 4134 bp += ret; \ 4135 } while (0) 4136 4137 if (bctl->flags & BTRFS_BALANCE_FORCE) 4138 CHECK_APPEND_1ARG("%s", "-f "); 4139 4140 if (bctl->flags & BTRFS_BALANCE_DATA) { 4141 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4142 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4143 } 4144 4145 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4146 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4147 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4148 } 4149 4150 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4151 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4152 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4153 } 4154 4155 #undef CHECK_APPEND_1ARG 4156 4157 out_overflow: 4158 4159 if (size_bp < size_buf) 4160 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4161 btrfs_info(fs_info, "balance: %s %s", 4162 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4163 "resume" : "start", buf); 4164 4165 kfree(buf); 4166 } 4167 4168 /* 4169 * Should be called with balance mutexe held 4170 */ 4171 int btrfs_balance(struct btrfs_fs_info *fs_info, 4172 struct btrfs_balance_control *bctl, 4173 struct btrfs_ioctl_balance_args *bargs) 4174 { 4175 u64 meta_target, data_target; 4176 u64 allowed; 4177 int mixed = 0; 4178 int ret; 4179 u64 num_devices; 4180 unsigned seq; 4181 bool reducing_redundancy; 4182 int i; 4183 4184 if (btrfs_fs_closing(fs_info) || 4185 atomic_read(&fs_info->balance_pause_req) || 4186 btrfs_should_cancel_balance(fs_info)) { 4187 ret = -EINVAL; 4188 goto out; 4189 } 4190 4191 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4192 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4193 mixed = 1; 4194 4195 /* 4196 * In case of mixed groups both data and meta should be picked, 4197 * and identical options should be given for both of them. 4198 */ 4199 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4200 if (mixed && (bctl->flags & allowed)) { 4201 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4202 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4203 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4204 btrfs_err(fs_info, 4205 "balance: mixed groups data and metadata options must be the same"); 4206 ret = -EINVAL; 4207 goto out; 4208 } 4209 } 4210 4211 /* 4212 * rw_devices will not change at the moment, device add/delete/replace 4213 * are exclusive 4214 */ 4215 num_devices = fs_info->fs_devices->rw_devices; 4216 4217 /* 4218 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4219 * special bit for it, to make it easier to distinguish. Thus we need 4220 * to set it manually, or balance would refuse the profile. 4221 */ 4222 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4223 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4224 if (num_devices >= btrfs_raid_array[i].devs_min) 4225 allowed |= btrfs_raid_array[i].bg_flag; 4226 4227 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4228 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4229 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4230 ret = -EINVAL; 4231 goto out; 4232 } 4233 4234 /* 4235 * Allow to reduce metadata or system integrity only if force set for 4236 * profiles with redundancy (copies, parity) 4237 */ 4238 allowed = 0; 4239 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4240 if (btrfs_raid_array[i].ncopies >= 2 || 4241 btrfs_raid_array[i].tolerated_failures >= 1) 4242 allowed |= btrfs_raid_array[i].bg_flag; 4243 } 4244 do { 4245 seq = read_seqbegin(&fs_info->profiles_lock); 4246 4247 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4248 (fs_info->avail_system_alloc_bits & allowed) && 4249 !(bctl->sys.target & allowed)) || 4250 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4251 (fs_info->avail_metadata_alloc_bits & allowed) && 4252 !(bctl->meta.target & allowed))) 4253 reducing_redundancy = true; 4254 else 4255 reducing_redundancy = false; 4256 4257 /* if we're not converting, the target field is uninitialized */ 4258 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4259 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4260 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4261 bctl->data.target : fs_info->avail_data_alloc_bits; 4262 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4263 4264 if (reducing_redundancy) { 4265 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4266 btrfs_info(fs_info, 4267 "balance: force reducing metadata redundancy"); 4268 } else { 4269 btrfs_err(fs_info, 4270 "balance: reduces metadata redundancy, use --force if you want this"); 4271 ret = -EINVAL; 4272 goto out; 4273 } 4274 } 4275 4276 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4277 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4278 btrfs_warn(fs_info, 4279 "balance: metadata profile %s has lower redundancy than data profile %s", 4280 btrfs_bg_type_to_raid_name(meta_target), 4281 btrfs_bg_type_to_raid_name(data_target)); 4282 } 4283 4284 ret = insert_balance_item(fs_info, bctl); 4285 if (ret && ret != -EEXIST) 4286 goto out; 4287 4288 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4289 BUG_ON(ret == -EEXIST); 4290 BUG_ON(fs_info->balance_ctl); 4291 spin_lock(&fs_info->balance_lock); 4292 fs_info->balance_ctl = bctl; 4293 spin_unlock(&fs_info->balance_lock); 4294 } else { 4295 BUG_ON(ret != -EEXIST); 4296 spin_lock(&fs_info->balance_lock); 4297 update_balance_args(bctl); 4298 spin_unlock(&fs_info->balance_lock); 4299 } 4300 4301 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4302 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4303 describe_balance_start_or_resume(fs_info); 4304 mutex_unlock(&fs_info->balance_mutex); 4305 4306 ret = __btrfs_balance(fs_info); 4307 4308 mutex_lock(&fs_info->balance_mutex); 4309 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) 4310 btrfs_info(fs_info, "balance: paused"); 4311 /* 4312 * Balance can be canceled by: 4313 * 4314 * - Regular cancel request 4315 * Then ret == -ECANCELED and balance_cancel_req > 0 4316 * 4317 * - Fatal signal to "btrfs" process 4318 * Either the signal caught by wait_reserve_ticket() and callers 4319 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4320 * got -ECANCELED. 4321 * Either way, in this case balance_cancel_req = 0, and 4322 * ret == -EINTR or ret == -ECANCELED. 4323 * 4324 * So here we only check the return value to catch canceled balance. 4325 */ 4326 else if (ret == -ECANCELED || ret == -EINTR) 4327 btrfs_info(fs_info, "balance: canceled"); 4328 else 4329 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4330 4331 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4332 4333 if (bargs) { 4334 memset(bargs, 0, sizeof(*bargs)); 4335 btrfs_update_ioctl_balance_args(fs_info, bargs); 4336 } 4337 4338 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4339 balance_need_close(fs_info)) { 4340 reset_balance_state(fs_info); 4341 btrfs_exclop_finish(fs_info); 4342 } 4343 4344 wake_up(&fs_info->balance_wait_q); 4345 4346 return ret; 4347 out: 4348 if (bctl->flags & BTRFS_BALANCE_RESUME) 4349 reset_balance_state(fs_info); 4350 else 4351 kfree(bctl); 4352 btrfs_exclop_finish(fs_info); 4353 4354 return ret; 4355 } 4356 4357 static int balance_kthread(void *data) 4358 { 4359 struct btrfs_fs_info *fs_info = data; 4360 int ret = 0; 4361 4362 mutex_lock(&fs_info->balance_mutex); 4363 if (fs_info->balance_ctl) 4364 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4365 mutex_unlock(&fs_info->balance_mutex); 4366 4367 return ret; 4368 } 4369 4370 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4371 { 4372 struct task_struct *tsk; 4373 4374 mutex_lock(&fs_info->balance_mutex); 4375 if (!fs_info->balance_ctl) { 4376 mutex_unlock(&fs_info->balance_mutex); 4377 return 0; 4378 } 4379 mutex_unlock(&fs_info->balance_mutex); 4380 4381 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4382 btrfs_info(fs_info, "balance: resume skipped"); 4383 return 0; 4384 } 4385 4386 /* 4387 * A ro->rw remount sequence should continue with the paused balance 4388 * regardless of who pauses it, system or the user as of now, so set 4389 * the resume flag. 4390 */ 4391 spin_lock(&fs_info->balance_lock); 4392 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4393 spin_unlock(&fs_info->balance_lock); 4394 4395 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4396 return PTR_ERR_OR_ZERO(tsk); 4397 } 4398 4399 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4400 { 4401 struct btrfs_balance_control *bctl; 4402 struct btrfs_balance_item *item; 4403 struct btrfs_disk_balance_args disk_bargs; 4404 struct btrfs_path *path; 4405 struct extent_buffer *leaf; 4406 struct btrfs_key key; 4407 int ret; 4408 4409 path = btrfs_alloc_path(); 4410 if (!path) 4411 return -ENOMEM; 4412 4413 key.objectid = BTRFS_BALANCE_OBJECTID; 4414 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4415 key.offset = 0; 4416 4417 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4418 if (ret < 0) 4419 goto out; 4420 if (ret > 0) { /* ret = -ENOENT; */ 4421 ret = 0; 4422 goto out; 4423 } 4424 4425 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4426 if (!bctl) { 4427 ret = -ENOMEM; 4428 goto out; 4429 } 4430 4431 leaf = path->nodes[0]; 4432 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4433 4434 bctl->flags = btrfs_balance_flags(leaf, item); 4435 bctl->flags |= BTRFS_BALANCE_RESUME; 4436 4437 btrfs_balance_data(leaf, item, &disk_bargs); 4438 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4439 btrfs_balance_meta(leaf, item, &disk_bargs); 4440 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4441 btrfs_balance_sys(leaf, item, &disk_bargs); 4442 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4443 4444 /* 4445 * This should never happen, as the paused balance state is recovered 4446 * during mount without any chance of other exclusive ops to collide. 4447 * 4448 * This gives the exclusive op status to balance and keeps in paused 4449 * state until user intervention (cancel or umount). If the ownership 4450 * cannot be assigned, show a message but do not fail. The balance 4451 * is in a paused state and must have fs_info::balance_ctl properly 4452 * set up. 4453 */ 4454 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) 4455 btrfs_warn(fs_info, 4456 "balance: cannot set exclusive op status, resume manually"); 4457 4458 btrfs_release_path(path); 4459 4460 mutex_lock(&fs_info->balance_mutex); 4461 BUG_ON(fs_info->balance_ctl); 4462 spin_lock(&fs_info->balance_lock); 4463 fs_info->balance_ctl = bctl; 4464 spin_unlock(&fs_info->balance_lock); 4465 mutex_unlock(&fs_info->balance_mutex); 4466 out: 4467 btrfs_free_path(path); 4468 return ret; 4469 } 4470 4471 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4472 { 4473 int ret = 0; 4474 4475 mutex_lock(&fs_info->balance_mutex); 4476 if (!fs_info->balance_ctl) { 4477 mutex_unlock(&fs_info->balance_mutex); 4478 return -ENOTCONN; 4479 } 4480 4481 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4482 atomic_inc(&fs_info->balance_pause_req); 4483 mutex_unlock(&fs_info->balance_mutex); 4484 4485 wait_event(fs_info->balance_wait_q, 4486 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4487 4488 mutex_lock(&fs_info->balance_mutex); 4489 /* we are good with balance_ctl ripped off from under us */ 4490 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4491 atomic_dec(&fs_info->balance_pause_req); 4492 } else { 4493 ret = -ENOTCONN; 4494 } 4495 4496 mutex_unlock(&fs_info->balance_mutex); 4497 return ret; 4498 } 4499 4500 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4501 { 4502 mutex_lock(&fs_info->balance_mutex); 4503 if (!fs_info->balance_ctl) { 4504 mutex_unlock(&fs_info->balance_mutex); 4505 return -ENOTCONN; 4506 } 4507 4508 /* 4509 * A paused balance with the item stored on disk can be resumed at 4510 * mount time if the mount is read-write. Otherwise it's still paused 4511 * and we must not allow cancelling as it deletes the item. 4512 */ 4513 if (sb_rdonly(fs_info->sb)) { 4514 mutex_unlock(&fs_info->balance_mutex); 4515 return -EROFS; 4516 } 4517 4518 atomic_inc(&fs_info->balance_cancel_req); 4519 /* 4520 * if we are running just wait and return, balance item is 4521 * deleted in btrfs_balance in this case 4522 */ 4523 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4524 mutex_unlock(&fs_info->balance_mutex); 4525 wait_event(fs_info->balance_wait_q, 4526 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4527 mutex_lock(&fs_info->balance_mutex); 4528 } else { 4529 mutex_unlock(&fs_info->balance_mutex); 4530 /* 4531 * Lock released to allow other waiters to continue, we'll 4532 * reexamine the status again. 4533 */ 4534 mutex_lock(&fs_info->balance_mutex); 4535 4536 if (fs_info->balance_ctl) { 4537 reset_balance_state(fs_info); 4538 btrfs_exclop_finish(fs_info); 4539 btrfs_info(fs_info, "balance: canceled"); 4540 } 4541 } 4542 4543 BUG_ON(fs_info->balance_ctl || 4544 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4545 atomic_dec(&fs_info->balance_cancel_req); 4546 mutex_unlock(&fs_info->balance_mutex); 4547 return 0; 4548 } 4549 4550 int btrfs_uuid_scan_kthread(void *data) 4551 { 4552 struct btrfs_fs_info *fs_info = data; 4553 struct btrfs_root *root = fs_info->tree_root; 4554 struct btrfs_key key; 4555 struct btrfs_path *path = NULL; 4556 int ret = 0; 4557 struct extent_buffer *eb; 4558 int slot; 4559 struct btrfs_root_item root_item; 4560 u32 item_size; 4561 struct btrfs_trans_handle *trans = NULL; 4562 bool closing = false; 4563 4564 path = btrfs_alloc_path(); 4565 if (!path) { 4566 ret = -ENOMEM; 4567 goto out; 4568 } 4569 4570 key.objectid = 0; 4571 key.type = BTRFS_ROOT_ITEM_KEY; 4572 key.offset = 0; 4573 4574 while (1) { 4575 if (btrfs_fs_closing(fs_info)) { 4576 closing = true; 4577 break; 4578 } 4579 ret = btrfs_search_forward(root, &key, path, 4580 BTRFS_OLDEST_GENERATION); 4581 if (ret) { 4582 if (ret > 0) 4583 ret = 0; 4584 break; 4585 } 4586 4587 if (key.type != BTRFS_ROOT_ITEM_KEY || 4588 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4589 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4590 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4591 goto skip; 4592 4593 eb = path->nodes[0]; 4594 slot = path->slots[0]; 4595 item_size = btrfs_item_size_nr(eb, slot); 4596 if (item_size < sizeof(root_item)) 4597 goto skip; 4598 4599 read_extent_buffer(eb, &root_item, 4600 btrfs_item_ptr_offset(eb, slot), 4601 (int)sizeof(root_item)); 4602 if (btrfs_root_refs(&root_item) == 0) 4603 goto skip; 4604 4605 if (!btrfs_is_empty_uuid(root_item.uuid) || 4606 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4607 if (trans) 4608 goto update_tree; 4609 4610 btrfs_release_path(path); 4611 /* 4612 * 1 - subvol uuid item 4613 * 1 - received_subvol uuid item 4614 */ 4615 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4616 if (IS_ERR(trans)) { 4617 ret = PTR_ERR(trans); 4618 break; 4619 } 4620 continue; 4621 } else { 4622 goto skip; 4623 } 4624 update_tree: 4625 btrfs_release_path(path); 4626 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4627 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4628 BTRFS_UUID_KEY_SUBVOL, 4629 key.objectid); 4630 if (ret < 0) { 4631 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4632 ret); 4633 break; 4634 } 4635 } 4636 4637 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4638 ret = btrfs_uuid_tree_add(trans, 4639 root_item.received_uuid, 4640 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4641 key.objectid); 4642 if (ret < 0) { 4643 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4644 ret); 4645 break; 4646 } 4647 } 4648 4649 skip: 4650 btrfs_release_path(path); 4651 if (trans) { 4652 ret = btrfs_end_transaction(trans); 4653 trans = NULL; 4654 if (ret) 4655 break; 4656 } 4657 4658 if (key.offset < (u64)-1) { 4659 key.offset++; 4660 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4661 key.offset = 0; 4662 key.type = BTRFS_ROOT_ITEM_KEY; 4663 } else if (key.objectid < (u64)-1) { 4664 key.offset = 0; 4665 key.type = BTRFS_ROOT_ITEM_KEY; 4666 key.objectid++; 4667 } else { 4668 break; 4669 } 4670 cond_resched(); 4671 } 4672 4673 out: 4674 btrfs_free_path(path); 4675 if (trans && !IS_ERR(trans)) 4676 btrfs_end_transaction(trans); 4677 if (ret) 4678 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4679 else if (!closing) 4680 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4681 up(&fs_info->uuid_tree_rescan_sem); 4682 return 0; 4683 } 4684 4685 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4686 { 4687 struct btrfs_trans_handle *trans; 4688 struct btrfs_root *tree_root = fs_info->tree_root; 4689 struct btrfs_root *uuid_root; 4690 struct task_struct *task; 4691 int ret; 4692 4693 /* 4694 * 1 - root node 4695 * 1 - root item 4696 */ 4697 trans = btrfs_start_transaction(tree_root, 2); 4698 if (IS_ERR(trans)) 4699 return PTR_ERR(trans); 4700 4701 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4702 if (IS_ERR(uuid_root)) { 4703 ret = PTR_ERR(uuid_root); 4704 btrfs_abort_transaction(trans, ret); 4705 btrfs_end_transaction(trans); 4706 return ret; 4707 } 4708 4709 fs_info->uuid_root = uuid_root; 4710 4711 ret = btrfs_commit_transaction(trans); 4712 if (ret) 4713 return ret; 4714 4715 down(&fs_info->uuid_tree_rescan_sem); 4716 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4717 if (IS_ERR(task)) { 4718 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4719 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4720 up(&fs_info->uuid_tree_rescan_sem); 4721 return PTR_ERR(task); 4722 } 4723 4724 return 0; 4725 } 4726 4727 /* 4728 * shrinking a device means finding all of the device extents past 4729 * the new size, and then following the back refs to the chunks. 4730 * The chunk relocation code actually frees the device extent 4731 */ 4732 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4733 { 4734 struct btrfs_fs_info *fs_info = device->fs_info; 4735 struct btrfs_root *root = fs_info->dev_root; 4736 struct btrfs_trans_handle *trans; 4737 struct btrfs_dev_extent *dev_extent = NULL; 4738 struct btrfs_path *path; 4739 u64 length; 4740 u64 chunk_offset; 4741 int ret; 4742 int slot; 4743 int failed = 0; 4744 bool retried = false; 4745 struct extent_buffer *l; 4746 struct btrfs_key key; 4747 struct btrfs_super_block *super_copy = fs_info->super_copy; 4748 u64 old_total = btrfs_super_total_bytes(super_copy); 4749 u64 old_size = btrfs_device_get_total_bytes(device); 4750 u64 diff; 4751 u64 start; 4752 4753 new_size = round_down(new_size, fs_info->sectorsize); 4754 start = new_size; 4755 diff = round_down(old_size - new_size, fs_info->sectorsize); 4756 4757 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4758 return -EINVAL; 4759 4760 path = btrfs_alloc_path(); 4761 if (!path) 4762 return -ENOMEM; 4763 4764 path->reada = READA_BACK; 4765 4766 trans = btrfs_start_transaction(root, 0); 4767 if (IS_ERR(trans)) { 4768 btrfs_free_path(path); 4769 return PTR_ERR(trans); 4770 } 4771 4772 mutex_lock(&fs_info->chunk_mutex); 4773 4774 btrfs_device_set_total_bytes(device, new_size); 4775 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4776 device->fs_devices->total_rw_bytes -= diff; 4777 atomic64_sub(diff, &fs_info->free_chunk_space); 4778 } 4779 4780 /* 4781 * Once the device's size has been set to the new size, ensure all 4782 * in-memory chunks are synced to disk so that the loop below sees them 4783 * and relocates them accordingly. 4784 */ 4785 if (contains_pending_extent(device, &start, diff)) { 4786 mutex_unlock(&fs_info->chunk_mutex); 4787 ret = btrfs_commit_transaction(trans); 4788 if (ret) 4789 goto done; 4790 } else { 4791 mutex_unlock(&fs_info->chunk_mutex); 4792 btrfs_end_transaction(trans); 4793 } 4794 4795 again: 4796 key.objectid = device->devid; 4797 key.offset = (u64)-1; 4798 key.type = BTRFS_DEV_EXTENT_KEY; 4799 4800 do { 4801 mutex_lock(&fs_info->reclaim_bgs_lock); 4802 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4803 if (ret < 0) { 4804 mutex_unlock(&fs_info->reclaim_bgs_lock); 4805 goto done; 4806 } 4807 4808 ret = btrfs_previous_item(root, path, 0, key.type); 4809 if (ret) { 4810 mutex_unlock(&fs_info->reclaim_bgs_lock); 4811 if (ret < 0) 4812 goto done; 4813 ret = 0; 4814 btrfs_release_path(path); 4815 break; 4816 } 4817 4818 l = path->nodes[0]; 4819 slot = path->slots[0]; 4820 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4821 4822 if (key.objectid != device->devid) { 4823 mutex_unlock(&fs_info->reclaim_bgs_lock); 4824 btrfs_release_path(path); 4825 break; 4826 } 4827 4828 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4829 length = btrfs_dev_extent_length(l, dev_extent); 4830 4831 if (key.offset + length <= new_size) { 4832 mutex_unlock(&fs_info->reclaim_bgs_lock); 4833 btrfs_release_path(path); 4834 break; 4835 } 4836 4837 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4838 btrfs_release_path(path); 4839 4840 /* 4841 * We may be relocating the only data chunk we have, 4842 * which could potentially end up with losing data's 4843 * raid profile, so lets allocate an empty one in 4844 * advance. 4845 */ 4846 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4847 if (ret < 0) { 4848 mutex_unlock(&fs_info->reclaim_bgs_lock); 4849 goto done; 4850 } 4851 4852 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4853 mutex_unlock(&fs_info->reclaim_bgs_lock); 4854 if (ret == -ENOSPC) { 4855 failed++; 4856 } else if (ret) { 4857 if (ret == -ETXTBSY) { 4858 btrfs_warn(fs_info, 4859 "could not shrink block group %llu due to active swapfile", 4860 chunk_offset); 4861 } 4862 goto done; 4863 } 4864 } while (key.offset-- > 0); 4865 4866 if (failed && !retried) { 4867 failed = 0; 4868 retried = true; 4869 goto again; 4870 } else if (failed && retried) { 4871 ret = -ENOSPC; 4872 goto done; 4873 } 4874 4875 /* Shrinking succeeded, else we would be at "done". */ 4876 trans = btrfs_start_transaction(root, 0); 4877 if (IS_ERR(trans)) { 4878 ret = PTR_ERR(trans); 4879 goto done; 4880 } 4881 4882 mutex_lock(&fs_info->chunk_mutex); 4883 /* Clear all state bits beyond the shrunk device size */ 4884 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4885 CHUNK_STATE_MASK); 4886 4887 btrfs_device_set_disk_total_bytes(device, new_size); 4888 if (list_empty(&device->post_commit_list)) 4889 list_add_tail(&device->post_commit_list, 4890 &trans->transaction->dev_update_list); 4891 4892 WARN_ON(diff > old_total); 4893 btrfs_set_super_total_bytes(super_copy, 4894 round_down(old_total - diff, fs_info->sectorsize)); 4895 mutex_unlock(&fs_info->chunk_mutex); 4896 4897 /* Now btrfs_update_device() will change the on-disk size. */ 4898 ret = btrfs_update_device(trans, device); 4899 if (ret < 0) { 4900 btrfs_abort_transaction(trans, ret); 4901 btrfs_end_transaction(trans); 4902 } else { 4903 ret = btrfs_commit_transaction(trans); 4904 } 4905 done: 4906 btrfs_free_path(path); 4907 if (ret) { 4908 mutex_lock(&fs_info->chunk_mutex); 4909 btrfs_device_set_total_bytes(device, old_size); 4910 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4911 device->fs_devices->total_rw_bytes += diff; 4912 atomic64_add(diff, &fs_info->free_chunk_space); 4913 mutex_unlock(&fs_info->chunk_mutex); 4914 } 4915 return ret; 4916 } 4917 4918 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 4919 struct btrfs_key *key, 4920 struct btrfs_chunk *chunk, int item_size) 4921 { 4922 struct btrfs_super_block *super_copy = fs_info->super_copy; 4923 struct btrfs_disk_key disk_key; 4924 u32 array_size; 4925 u8 *ptr; 4926 4927 lockdep_assert_held(&fs_info->chunk_mutex); 4928 4929 array_size = btrfs_super_sys_array_size(super_copy); 4930 if (array_size + item_size + sizeof(disk_key) 4931 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 4932 return -EFBIG; 4933 4934 ptr = super_copy->sys_chunk_array + array_size; 4935 btrfs_cpu_key_to_disk(&disk_key, key); 4936 memcpy(ptr, &disk_key, sizeof(disk_key)); 4937 ptr += sizeof(disk_key); 4938 memcpy(ptr, chunk, item_size); 4939 item_size += sizeof(disk_key); 4940 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 4941 4942 return 0; 4943 } 4944 4945 /* 4946 * sort the devices in descending order by max_avail, total_avail 4947 */ 4948 static int btrfs_cmp_device_info(const void *a, const void *b) 4949 { 4950 const struct btrfs_device_info *di_a = a; 4951 const struct btrfs_device_info *di_b = b; 4952 4953 if (di_a->max_avail > di_b->max_avail) 4954 return -1; 4955 if (di_a->max_avail < di_b->max_avail) 4956 return 1; 4957 if (di_a->total_avail > di_b->total_avail) 4958 return -1; 4959 if (di_a->total_avail < di_b->total_avail) 4960 return 1; 4961 return 0; 4962 } 4963 4964 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 4965 { 4966 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 4967 return; 4968 4969 btrfs_set_fs_incompat(info, RAID56); 4970 } 4971 4972 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 4973 { 4974 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 4975 return; 4976 4977 btrfs_set_fs_incompat(info, RAID1C34); 4978 } 4979 4980 /* 4981 * Structure used internally for btrfs_create_chunk() function. 4982 * Wraps needed parameters. 4983 */ 4984 struct alloc_chunk_ctl { 4985 u64 start; 4986 u64 type; 4987 /* Total number of stripes to allocate */ 4988 int num_stripes; 4989 /* sub_stripes info for map */ 4990 int sub_stripes; 4991 /* Stripes per device */ 4992 int dev_stripes; 4993 /* Maximum number of devices to use */ 4994 int devs_max; 4995 /* Minimum number of devices to use */ 4996 int devs_min; 4997 /* ndevs has to be a multiple of this */ 4998 int devs_increment; 4999 /* Number of copies */ 5000 int ncopies; 5001 /* Number of stripes worth of bytes to store parity information */ 5002 int nparity; 5003 u64 max_stripe_size; 5004 u64 max_chunk_size; 5005 u64 dev_extent_min; 5006 u64 stripe_size; 5007 u64 chunk_size; 5008 int ndevs; 5009 }; 5010 5011 static void init_alloc_chunk_ctl_policy_regular( 5012 struct btrfs_fs_devices *fs_devices, 5013 struct alloc_chunk_ctl *ctl) 5014 { 5015 u64 type = ctl->type; 5016 5017 if (type & BTRFS_BLOCK_GROUP_DATA) { 5018 ctl->max_stripe_size = SZ_1G; 5019 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; 5020 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5021 /* For larger filesystems, use larger metadata chunks */ 5022 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) 5023 ctl->max_stripe_size = SZ_1G; 5024 else 5025 ctl->max_stripe_size = SZ_256M; 5026 ctl->max_chunk_size = ctl->max_stripe_size; 5027 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5028 ctl->max_stripe_size = SZ_32M; 5029 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5030 ctl->devs_max = min_t(int, ctl->devs_max, 5031 BTRFS_MAX_DEVS_SYS_CHUNK); 5032 } else { 5033 BUG(); 5034 } 5035 5036 /* We don't want a chunk larger than 10% of writable space */ 5037 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5038 ctl->max_chunk_size); 5039 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5040 } 5041 5042 static void init_alloc_chunk_ctl_policy_zoned( 5043 struct btrfs_fs_devices *fs_devices, 5044 struct alloc_chunk_ctl *ctl) 5045 { 5046 u64 zone_size = fs_devices->fs_info->zone_size; 5047 u64 limit; 5048 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5049 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5050 u64 min_chunk_size = min_data_stripes * zone_size; 5051 u64 type = ctl->type; 5052 5053 ctl->max_stripe_size = zone_size; 5054 if (type & BTRFS_BLOCK_GROUP_DATA) { 5055 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5056 zone_size); 5057 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5058 ctl->max_chunk_size = ctl->max_stripe_size; 5059 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 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 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5069 zone_size), 5070 min_chunk_size); 5071 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5072 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5073 } 5074 5075 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5076 struct alloc_chunk_ctl *ctl) 5077 { 5078 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5079 5080 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5081 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5082 ctl->devs_max = btrfs_raid_array[index].devs_max; 5083 if (!ctl->devs_max) 5084 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5085 ctl->devs_min = btrfs_raid_array[index].devs_min; 5086 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5087 ctl->ncopies = btrfs_raid_array[index].ncopies; 5088 ctl->nparity = btrfs_raid_array[index].nparity; 5089 ctl->ndevs = 0; 5090 5091 switch (fs_devices->chunk_alloc_policy) { 5092 case BTRFS_CHUNK_ALLOC_REGULAR: 5093 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5094 break; 5095 case BTRFS_CHUNK_ALLOC_ZONED: 5096 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5097 break; 5098 default: 5099 BUG(); 5100 } 5101 } 5102 5103 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5104 struct alloc_chunk_ctl *ctl, 5105 struct btrfs_device_info *devices_info) 5106 { 5107 struct btrfs_fs_info *info = fs_devices->fs_info; 5108 struct btrfs_device *device; 5109 u64 total_avail; 5110 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5111 int ret; 5112 int ndevs = 0; 5113 u64 max_avail; 5114 u64 dev_offset; 5115 5116 /* 5117 * in the first pass through the devices list, we gather information 5118 * about the available holes on each device. 5119 */ 5120 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5121 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5122 WARN(1, KERN_ERR 5123 "BTRFS: read-only device in alloc_list\n"); 5124 continue; 5125 } 5126 5127 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5128 &device->dev_state) || 5129 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5130 continue; 5131 5132 if (device->total_bytes > device->bytes_used) 5133 total_avail = device->total_bytes - device->bytes_used; 5134 else 5135 total_avail = 0; 5136 5137 /* If there is no space on this device, skip it. */ 5138 if (total_avail < ctl->dev_extent_min) 5139 continue; 5140 5141 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5142 &max_avail); 5143 if (ret && ret != -ENOSPC) 5144 return ret; 5145 5146 if (ret == 0) 5147 max_avail = dev_extent_want; 5148 5149 if (max_avail < ctl->dev_extent_min) { 5150 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5151 btrfs_debug(info, 5152 "%s: devid %llu has no free space, have=%llu want=%llu", 5153 __func__, device->devid, max_avail, 5154 ctl->dev_extent_min); 5155 continue; 5156 } 5157 5158 if (ndevs == fs_devices->rw_devices) { 5159 WARN(1, "%s: found more than %llu devices\n", 5160 __func__, fs_devices->rw_devices); 5161 break; 5162 } 5163 devices_info[ndevs].dev_offset = dev_offset; 5164 devices_info[ndevs].max_avail = max_avail; 5165 devices_info[ndevs].total_avail = total_avail; 5166 devices_info[ndevs].dev = device; 5167 ++ndevs; 5168 } 5169 ctl->ndevs = ndevs; 5170 5171 /* 5172 * now sort the devices by hole size / available space 5173 */ 5174 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5175 btrfs_cmp_device_info, NULL); 5176 5177 return 0; 5178 } 5179 5180 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5181 struct btrfs_device_info *devices_info) 5182 { 5183 /* Number of stripes that count for block group size */ 5184 int data_stripes; 5185 5186 /* 5187 * The primary goal is to maximize the number of stripes, so use as 5188 * many devices as possible, even if the stripes are not maximum sized. 5189 * 5190 * The DUP profile stores more than one stripe per device, the 5191 * max_avail is the total size so we have to adjust. 5192 */ 5193 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5194 ctl->dev_stripes); 5195 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5196 5197 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5198 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5199 5200 /* 5201 * Use the number of data stripes to figure out how big this chunk is 5202 * really going to be in terms of logical address space, and compare 5203 * that answer with the max chunk size. If it's higher, we try to 5204 * reduce stripe_size. 5205 */ 5206 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5207 /* 5208 * Reduce stripe_size, round it up to a 16MB boundary again and 5209 * then use it, unless it ends up being even bigger than the 5210 * previous value we had already. 5211 */ 5212 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5213 data_stripes), SZ_16M), 5214 ctl->stripe_size); 5215 } 5216 5217 /* Align to BTRFS_STRIPE_LEN */ 5218 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5219 ctl->chunk_size = ctl->stripe_size * data_stripes; 5220 5221 return 0; 5222 } 5223 5224 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5225 struct btrfs_device_info *devices_info) 5226 { 5227 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5228 /* Number of stripes that count for block group size */ 5229 int data_stripes; 5230 5231 /* 5232 * It should hold because: 5233 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5234 */ 5235 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5236 5237 ctl->stripe_size = zone_size; 5238 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5239 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5240 5241 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5242 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5243 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5244 ctl->stripe_size) + ctl->nparity, 5245 ctl->dev_stripes); 5246 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5247 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5248 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5249 } 5250 5251 ctl->chunk_size = ctl->stripe_size * data_stripes; 5252 5253 return 0; 5254 } 5255 5256 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5257 struct alloc_chunk_ctl *ctl, 5258 struct btrfs_device_info *devices_info) 5259 { 5260 struct btrfs_fs_info *info = fs_devices->fs_info; 5261 5262 /* 5263 * Round down to number of usable stripes, devs_increment can be any 5264 * number so we can't use round_down() that requires power of 2, while 5265 * rounddown is safe. 5266 */ 5267 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5268 5269 if (ctl->ndevs < ctl->devs_min) { 5270 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5271 btrfs_debug(info, 5272 "%s: not enough devices with free space: have=%d minimum required=%d", 5273 __func__, ctl->ndevs, ctl->devs_min); 5274 } 5275 return -ENOSPC; 5276 } 5277 5278 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5279 5280 switch (fs_devices->chunk_alloc_policy) { 5281 case BTRFS_CHUNK_ALLOC_REGULAR: 5282 return decide_stripe_size_regular(ctl, devices_info); 5283 case BTRFS_CHUNK_ALLOC_ZONED: 5284 return decide_stripe_size_zoned(ctl, devices_info); 5285 default: 5286 BUG(); 5287 } 5288 } 5289 5290 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5291 struct alloc_chunk_ctl *ctl, 5292 struct btrfs_device_info *devices_info) 5293 { 5294 struct btrfs_fs_info *info = trans->fs_info; 5295 struct map_lookup *map = NULL; 5296 struct extent_map_tree *em_tree; 5297 struct btrfs_block_group *block_group; 5298 struct extent_map *em; 5299 u64 start = ctl->start; 5300 u64 type = ctl->type; 5301 int ret; 5302 int i; 5303 int j; 5304 5305 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5306 if (!map) 5307 return ERR_PTR(-ENOMEM); 5308 map->num_stripes = ctl->num_stripes; 5309 5310 for (i = 0; i < ctl->ndevs; ++i) { 5311 for (j = 0; j < ctl->dev_stripes; ++j) { 5312 int s = i * ctl->dev_stripes + j; 5313 map->stripes[s].dev = devices_info[i].dev; 5314 map->stripes[s].physical = devices_info[i].dev_offset + 5315 j * ctl->stripe_size; 5316 } 5317 } 5318 map->stripe_len = BTRFS_STRIPE_LEN; 5319 map->io_align = BTRFS_STRIPE_LEN; 5320 map->io_width = BTRFS_STRIPE_LEN; 5321 map->type = type; 5322 map->sub_stripes = ctl->sub_stripes; 5323 5324 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5325 5326 em = alloc_extent_map(); 5327 if (!em) { 5328 kfree(map); 5329 return ERR_PTR(-ENOMEM); 5330 } 5331 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5332 em->map_lookup = map; 5333 em->start = start; 5334 em->len = ctl->chunk_size; 5335 em->block_start = 0; 5336 em->block_len = em->len; 5337 em->orig_block_len = ctl->stripe_size; 5338 5339 em_tree = &info->mapping_tree; 5340 write_lock(&em_tree->lock); 5341 ret = add_extent_mapping(em_tree, em, 0); 5342 if (ret) { 5343 write_unlock(&em_tree->lock); 5344 free_extent_map(em); 5345 return ERR_PTR(ret); 5346 } 5347 write_unlock(&em_tree->lock); 5348 5349 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5350 if (IS_ERR(block_group)) 5351 goto error_del_extent; 5352 5353 for (i = 0; i < map->num_stripes; i++) { 5354 struct btrfs_device *dev = map->stripes[i].dev; 5355 5356 btrfs_device_set_bytes_used(dev, 5357 dev->bytes_used + ctl->stripe_size); 5358 if (list_empty(&dev->post_commit_list)) 5359 list_add_tail(&dev->post_commit_list, 5360 &trans->transaction->dev_update_list); 5361 } 5362 5363 atomic64_sub(ctl->stripe_size * map->num_stripes, 5364 &info->free_chunk_space); 5365 5366 free_extent_map(em); 5367 check_raid56_incompat_flag(info, type); 5368 check_raid1c34_incompat_flag(info, type); 5369 5370 return block_group; 5371 5372 error_del_extent: 5373 write_lock(&em_tree->lock); 5374 remove_extent_mapping(em_tree, em); 5375 write_unlock(&em_tree->lock); 5376 5377 /* One for our allocation */ 5378 free_extent_map(em); 5379 /* One for the tree reference */ 5380 free_extent_map(em); 5381 5382 return block_group; 5383 } 5384 5385 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5386 u64 type) 5387 { 5388 struct btrfs_fs_info *info = trans->fs_info; 5389 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5390 struct btrfs_device_info *devices_info = NULL; 5391 struct alloc_chunk_ctl ctl; 5392 struct btrfs_block_group *block_group; 5393 int ret; 5394 5395 lockdep_assert_held(&info->chunk_mutex); 5396 5397 if (!alloc_profile_is_valid(type, 0)) { 5398 ASSERT(0); 5399 return ERR_PTR(-EINVAL); 5400 } 5401 5402 if (list_empty(&fs_devices->alloc_list)) { 5403 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5404 btrfs_debug(info, "%s: no writable device", __func__); 5405 return ERR_PTR(-ENOSPC); 5406 } 5407 5408 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5409 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5410 ASSERT(0); 5411 return ERR_PTR(-EINVAL); 5412 } 5413 5414 ctl.start = find_next_chunk(info); 5415 ctl.type = type; 5416 init_alloc_chunk_ctl(fs_devices, &ctl); 5417 5418 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5419 GFP_NOFS); 5420 if (!devices_info) 5421 return ERR_PTR(-ENOMEM); 5422 5423 ret = gather_device_info(fs_devices, &ctl, devices_info); 5424 if (ret < 0) { 5425 block_group = ERR_PTR(ret); 5426 goto out; 5427 } 5428 5429 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5430 if (ret < 0) { 5431 block_group = ERR_PTR(ret); 5432 goto out; 5433 } 5434 5435 block_group = create_chunk(trans, &ctl, devices_info); 5436 5437 out: 5438 kfree(devices_info); 5439 return block_group; 5440 } 5441 5442 /* 5443 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5444 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5445 * chunks. 5446 * 5447 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5448 * phases. 5449 */ 5450 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5451 struct btrfs_block_group *bg) 5452 { 5453 struct btrfs_fs_info *fs_info = trans->fs_info; 5454 struct btrfs_root *extent_root = fs_info->extent_root; 5455 struct btrfs_root *chunk_root = fs_info->chunk_root; 5456 struct btrfs_key key; 5457 struct btrfs_chunk *chunk; 5458 struct btrfs_stripe *stripe; 5459 struct extent_map *em; 5460 struct map_lookup *map; 5461 size_t item_size; 5462 int i; 5463 int ret; 5464 5465 /* 5466 * We take the chunk_mutex for 2 reasons: 5467 * 5468 * 1) Updates and insertions in the chunk btree must be done while holding 5469 * the chunk_mutex, as well as updating the system chunk array in the 5470 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5471 * details; 5472 * 5473 * 2) To prevent races with the final phase of a device replace operation 5474 * that replaces the device object associated with the map's stripes, 5475 * because the device object's id can change at any time during that 5476 * final phase of the device replace operation 5477 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5478 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5479 * which would cause a failure when updating the device item, which does 5480 * not exists, or persisting a stripe of the chunk item with such ID. 5481 * Here we can't use the device_list_mutex because our caller already 5482 * has locked the chunk_mutex, and the final phase of device replace 5483 * acquires both mutexes - first the device_list_mutex and then the 5484 * chunk_mutex. Using any of those two mutexes protects us from a 5485 * concurrent device replace. 5486 */ 5487 lockdep_assert_held(&fs_info->chunk_mutex); 5488 5489 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5490 if (IS_ERR(em)) { 5491 ret = PTR_ERR(em); 5492 btrfs_abort_transaction(trans, ret); 5493 return ret; 5494 } 5495 5496 map = em->map_lookup; 5497 item_size = btrfs_chunk_item_size(map->num_stripes); 5498 5499 chunk = kzalloc(item_size, GFP_NOFS); 5500 if (!chunk) { 5501 ret = -ENOMEM; 5502 btrfs_abort_transaction(trans, ret); 5503 goto out; 5504 } 5505 5506 for (i = 0; i < map->num_stripes; i++) { 5507 struct btrfs_device *device = map->stripes[i].dev; 5508 5509 ret = btrfs_update_device(trans, device); 5510 if (ret) 5511 goto out; 5512 } 5513 5514 stripe = &chunk->stripe; 5515 for (i = 0; i < map->num_stripes; i++) { 5516 struct btrfs_device *device = map->stripes[i].dev; 5517 const u64 dev_offset = map->stripes[i].physical; 5518 5519 btrfs_set_stack_stripe_devid(stripe, device->devid); 5520 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5521 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5522 stripe++; 5523 } 5524 5525 btrfs_set_stack_chunk_length(chunk, bg->length); 5526 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); 5527 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5528 btrfs_set_stack_chunk_type(chunk, map->type); 5529 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5530 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5531 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5532 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5533 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5534 5535 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5536 key.type = BTRFS_CHUNK_ITEM_KEY; 5537 key.offset = bg->start; 5538 5539 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5540 if (ret) 5541 goto out; 5542 5543 bg->chunk_item_inserted = 1; 5544 5545 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5546 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5547 if (ret) 5548 goto out; 5549 } 5550 5551 out: 5552 kfree(chunk); 5553 free_extent_map(em); 5554 return ret; 5555 } 5556 5557 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5558 { 5559 struct btrfs_fs_info *fs_info = trans->fs_info; 5560 u64 alloc_profile; 5561 struct btrfs_block_group *meta_bg; 5562 struct btrfs_block_group *sys_bg; 5563 5564 /* 5565 * When adding a new device for sprouting, the seed device is read-only 5566 * so we must first allocate a metadata and a system chunk. But before 5567 * adding the block group items to the extent, device and chunk btrees, 5568 * we must first: 5569 * 5570 * 1) Create both chunks without doing any changes to the btrees, as 5571 * otherwise we would get -ENOSPC since the block groups from the 5572 * seed device are read-only; 5573 * 5574 * 2) Add the device item for the new sprout device - finishing the setup 5575 * of a new block group requires updating the device item in the chunk 5576 * btree, so it must exist when we attempt to do it. The previous step 5577 * ensures this does not fail with -ENOSPC. 5578 * 5579 * After that we can add the block group items to their btrees: 5580 * update existing device item in the chunk btree, add a new block group 5581 * item to the extent btree, add a new chunk item to the chunk btree and 5582 * finally add the new device extent items to the devices btree. 5583 */ 5584 5585 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5586 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5587 if (IS_ERR(meta_bg)) 5588 return PTR_ERR(meta_bg); 5589 5590 alloc_profile = btrfs_system_alloc_profile(fs_info); 5591 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5592 if (IS_ERR(sys_bg)) 5593 return PTR_ERR(sys_bg); 5594 5595 return 0; 5596 } 5597 5598 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5599 { 5600 const int index = btrfs_bg_flags_to_raid_index(map->type); 5601 5602 return btrfs_raid_array[index].tolerated_failures; 5603 } 5604 5605 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5606 { 5607 struct extent_map *em; 5608 struct map_lookup *map; 5609 int miss_ndevs = 0; 5610 int i; 5611 bool ret = true; 5612 5613 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5614 if (IS_ERR(em)) 5615 return false; 5616 5617 map = em->map_lookup; 5618 for (i = 0; i < map->num_stripes; i++) { 5619 if (test_bit(BTRFS_DEV_STATE_MISSING, 5620 &map->stripes[i].dev->dev_state)) { 5621 miss_ndevs++; 5622 continue; 5623 } 5624 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5625 &map->stripes[i].dev->dev_state)) { 5626 ret = false; 5627 goto end; 5628 } 5629 } 5630 5631 /* 5632 * If the number of missing devices is larger than max errors, we can 5633 * not write the data into that chunk successfully. 5634 */ 5635 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5636 ret = false; 5637 end: 5638 free_extent_map(em); 5639 return ret; 5640 } 5641 5642 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5643 { 5644 struct extent_map *em; 5645 5646 while (1) { 5647 write_lock(&tree->lock); 5648 em = lookup_extent_mapping(tree, 0, (u64)-1); 5649 if (em) 5650 remove_extent_mapping(tree, em); 5651 write_unlock(&tree->lock); 5652 if (!em) 5653 break; 5654 /* once for us */ 5655 free_extent_map(em); 5656 /* once for the tree */ 5657 free_extent_map(em); 5658 } 5659 } 5660 5661 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5662 { 5663 struct extent_map *em; 5664 struct map_lookup *map; 5665 int ret; 5666 5667 em = btrfs_get_chunk_map(fs_info, logical, len); 5668 if (IS_ERR(em)) 5669 /* 5670 * We could return errors for these cases, but that could get 5671 * ugly and we'd probably do the same thing which is just not do 5672 * anything else and exit, so return 1 so the callers don't try 5673 * to use other copies. 5674 */ 5675 return 1; 5676 5677 map = em->map_lookup; 5678 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) 5679 ret = map->num_stripes; 5680 else if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5681 ret = map->sub_stripes; 5682 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5683 ret = 2; 5684 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5685 /* 5686 * There could be two corrupted data stripes, we need 5687 * to loop retry in order to rebuild the correct data. 5688 * 5689 * Fail a stripe at a time on every retry except the 5690 * stripe under reconstruction. 5691 */ 5692 ret = map->num_stripes; 5693 else 5694 ret = 1; 5695 free_extent_map(em); 5696 5697 down_read(&fs_info->dev_replace.rwsem); 5698 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5699 fs_info->dev_replace.tgtdev) 5700 ret++; 5701 up_read(&fs_info->dev_replace.rwsem); 5702 5703 return ret; 5704 } 5705 5706 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5707 u64 logical) 5708 { 5709 struct extent_map *em; 5710 struct map_lookup *map; 5711 unsigned long len = fs_info->sectorsize; 5712 5713 em = btrfs_get_chunk_map(fs_info, logical, len); 5714 5715 if (!WARN_ON(IS_ERR(em))) { 5716 map = em->map_lookup; 5717 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5718 len = map->stripe_len * nr_data_stripes(map); 5719 free_extent_map(em); 5720 } 5721 return len; 5722 } 5723 5724 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5725 { 5726 struct extent_map *em; 5727 struct map_lookup *map; 5728 int ret = 0; 5729 5730 em = btrfs_get_chunk_map(fs_info, logical, len); 5731 5732 if(!WARN_ON(IS_ERR(em))) { 5733 map = em->map_lookup; 5734 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5735 ret = 1; 5736 free_extent_map(em); 5737 } 5738 return ret; 5739 } 5740 5741 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5742 struct map_lookup *map, int first, 5743 int dev_replace_is_ongoing) 5744 { 5745 int i; 5746 int num_stripes; 5747 int preferred_mirror; 5748 int tolerance; 5749 struct btrfs_device *srcdev; 5750 5751 ASSERT((map->type & 5752 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5753 5754 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5755 num_stripes = map->sub_stripes; 5756 else 5757 num_stripes = map->num_stripes; 5758 5759 switch (fs_info->fs_devices->read_policy) { 5760 default: 5761 /* Shouldn't happen, just warn and use pid instead of failing */ 5762 btrfs_warn_rl(fs_info, 5763 "unknown read_policy type %u, reset to pid", 5764 fs_info->fs_devices->read_policy); 5765 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5766 fallthrough; 5767 case BTRFS_READ_POLICY_PID: 5768 preferred_mirror = first + (current->pid % num_stripes); 5769 break; 5770 } 5771 5772 if (dev_replace_is_ongoing && 5773 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5774 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5775 srcdev = fs_info->dev_replace.srcdev; 5776 else 5777 srcdev = NULL; 5778 5779 /* 5780 * try to avoid the drive that is the source drive for a 5781 * dev-replace procedure, only choose it if no other non-missing 5782 * mirror is available 5783 */ 5784 for (tolerance = 0; tolerance < 2; tolerance++) { 5785 if (map->stripes[preferred_mirror].dev->bdev && 5786 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5787 return preferred_mirror; 5788 for (i = first; i < first + num_stripes; i++) { 5789 if (map->stripes[i].dev->bdev && 5790 (tolerance || map->stripes[i].dev != srcdev)) 5791 return i; 5792 } 5793 } 5794 5795 /* we couldn't find one that doesn't fail. Just return something 5796 * and the io error handling code will clean up eventually 5797 */ 5798 return preferred_mirror; 5799 } 5800 5801 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5802 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5803 { 5804 int i; 5805 int again = 1; 5806 5807 while (again) { 5808 again = 0; 5809 for (i = 0; i < num_stripes - 1; i++) { 5810 /* Swap if parity is on a smaller index */ 5811 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5812 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5813 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5814 again = 1; 5815 } 5816 } 5817 } 5818 } 5819 5820 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5821 int total_stripes, 5822 int real_stripes) 5823 { 5824 struct btrfs_io_context *bioc = kzalloc( 5825 /* The size of btrfs_io_context */ 5826 sizeof(struct btrfs_io_context) + 5827 /* Plus the variable array for the stripes */ 5828 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5829 /* Plus the variable array for the tgt dev */ 5830 sizeof(int) * (real_stripes) + 5831 /* 5832 * Plus the raid_map, which includes both the tgt dev 5833 * and the stripes. 5834 */ 5835 sizeof(u64) * (total_stripes), 5836 GFP_NOFS|__GFP_NOFAIL); 5837 5838 atomic_set(&bioc->error, 0); 5839 refcount_set(&bioc->refs, 1); 5840 5841 bioc->fs_info = fs_info; 5842 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5843 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5844 5845 return bioc; 5846 } 5847 5848 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5849 { 5850 WARN_ON(!refcount_read(&bioc->refs)); 5851 refcount_inc(&bioc->refs); 5852 } 5853 5854 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5855 { 5856 if (!bioc) 5857 return; 5858 if (refcount_dec_and_test(&bioc->refs)) 5859 kfree(bioc); 5860 } 5861 5862 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ 5863 /* 5864 * Please note that, discard won't be sent to target device of device 5865 * replace. 5866 */ 5867 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, 5868 u64 logical, u64 *length_ret, 5869 struct btrfs_io_context **bioc_ret) 5870 { 5871 struct extent_map *em; 5872 struct map_lookup *map; 5873 struct btrfs_io_context *bioc; 5874 u64 length = *length_ret; 5875 u64 offset; 5876 u64 stripe_nr; 5877 u64 stripe_nr_end; 5878 u64 stripe_end_offset; 5879 u64 stripe_cnt; 5880 u64 stripe_len; 5881 u64 stripe_offset; 5882 u64 num_stripes; 5883 u32 stripe_index; 5884 u32 factor = 0; 5885 u32 sub_stripes = 0; 5886 u64 stripes_per_dev = 0; 5887 u32 remaining_stripes = 0; 5888 u32 last_stripe = 0; 5889 int ret = 0; 5890 int i; 5891 5892 /* Discard always returns a bioc. */ 5893 ASSERT(bioc_ret); 5894 5895 em = btrfs_get_chunk_map(fs_info, logical, length); 5896 if (IS_ERR(em)) 5897 return PTR_ERR(em); 5898 5899 map = em->map_lookup; 5900 /* we don't discard raid56 yet */ 5901 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5902 ret = -EOPNOTSUPP; 5903 goto out; 5904 } 5905 5906 offset = logical - em->start; 5907 length = min_t(u64, em->start + em->len - logical, length); 5908 *length_ret = length; 5909 5910 stripe_len = map->stripe_len; 5911 /* 5912 * stripe_nr counts the total number of stripes we have to stride 5913 * to get to this block 5914 */ 5915 stripe_nr = div64_u64(offset, stripe_len); 5916 5917 /* stripe_offset is the offset of this block in its stripe */ 5918 stripe_offset = offset - stripe_nr * stripe_len; 5919 5920 stripe_nr_end = round_up(offset + length, map->stripe_len); 5921 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 5922 stripe_cnt = stripe_nr_end - stripe_nr; 5923 stripe_end_offset = stripe_nr_end * map->stripe_len - 5924 (offset + length); 5925 /* 5926 * after this, stripe_nr is the number of stripes on this 5927 * device we have to walk to find the data, and stripe_index is 5928 * the number of our device in the stripe array 5929 */ 5930 num_stripes = 1; 5931 stripe_index = 0; 5932 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5933 BTRFS_BLOCK_GROUP_RAID10)) { 5934 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 5935 sub_stripes = 1; 5936 else 5937 sub_stripes = map->sub_stripes; 5938 5939 factor = map->num_stripes / sub_stripes; 5940 num_stripes = min_t(u64, map->num_stripes, 5941 sub_stripes * stripe_cnt); 5942 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 5943 stripe_index *= sub_stripes; 5944 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 5945 &remaining_stripes); 5946 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 5947 last_stripe *= sub_stripes; 5948 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 5949 BTRFS_BLOCK_GROUP_DUP)) { 5950 num_stripes = map->num_stripes; 5951 } else { 5952 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 5953 &stripe_index); 5954 } 5955 5956 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0); 5957 if (!bioc) { 5958 ret = -ENOMEM; 5959 goto out; 5960 } 5961 5962 for (i = 0; i < num_stripes; i++) { 5963 bioc->stripes[i].physical = 5964 map->stripes[stripe_index].physical + 5965 stripe_offset + stripe_nr * map->stripe_len; 5966 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 5967 5968 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5969 BTRFS_BLOCK_GROUP_RAID10)) { 5970 bioc->stripes[i].length = stripes_per_dev * 5971 map->stripe_len; 5972 5973 if (i / sub_stripes < remaining_stripes) 5974 bioc->stripes[i].length += map->stripe_len; 5975 5976 /* 5977 * Special for the first stripe and 5978 * the last stripe: 5979 * 5980 * |-------|...|-------| 5981 * |----------| 5982 * off end_off 5983 */ 5984 if (i < sub_stripes) 5985 bioc->stripes[i].length -= stripe_offset; 5986 5987 if (stripe_index >= last_stripe && 5988 stripe_index <= (last_stripe + 5989 sub_stripes - 1)) 5990 bioc->stripes[i].length -= stripe_end_offset; 5991 5992 if (i == sub_stripes - 1) 5993 stripe_offset = 0; 5994 } else { 5995 bioc->stripes[i].length = length; 5996 } 5997 5998 stripe_index++; 5999 if (stripe_index == map->num_stripes) { 6000 stripe_index = 0; 6001 stripe_nr++; 6002 } 6003 } 6004 6005 *bioc_ret = bioc; 6006 bioc->map_type = map->type; 6007 bioc->num_stripes = num_stripes; 6008 out: 6009 free_extent_map(em); 6010 return ret; 6011 } 6012 6013 /* 6014 * In dev-replace case, for repair case (that's the only case where the mirror 6015 * is selected explicitly when calling btrfs_map_block), blocks left of the 6016 * left cursor can also be read from the target drive. 6017 * 6018 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6019 * array of stripes. 6020 * For READ, it also needs to be supported using the same mirror number. 6021 * 6022 * If the requested block is not left of the left cursor, EIO is returned. This 6023 * can happen because btrfs_num_copies() returns one more in the dev-replace 6024 * case. 6025 */ 6026 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6027 u64 logical, u64 length, 6028 u64 srcdev_devid, int *mirror_num, 6029 u64 *physical) 6030 { 6031 struct btrfs_io_context *bioc = NULL; 6032 int num_stripes; 6033 int index_srcdev = 0; 6034 int found = 0; 6035 u64 physical_of_found = 0; 6036 int i; 6037 int ret = 0; 6038 6039 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6040 logical, &length, &bioc, 0, 0); 6041 if (ret) { 6042 ASSERT(bioc == NULL); 6043 return ret; 6044 } 6045 6046 num_stripes = bioc->num_stripes; 6047 if (*mirror_num > num_stripes) { 6048 /* 6049 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6050 * that means that the requested area is not left of the left 6051 * cursor 6052 */ 6053 btrfs_put_bioc(bioc); 6054 return -EIO; 6055 } 6056 6057 /* 6058 * process the rest of the function using the mirror_num of the source 6059 * drive. Therefore look it up first. At the end, patch the device 6060 * pointer to the one of the target drive. 6061 */ 6062 for (i = 0; i < num_stripes; i++) { 6063 if (bioc->stripes[i].dev->devid != srcdev_devid) 6064 continue; 6065 6066 /* 6067 * In case of DUP, in order to keep it simple, only add the 6068 * mirror with the lowest physical address 6069 */ 6070 if (found && 6071 physical_of_found <= bioc->stripes[i].physical) 6072 continue; 6073 6074 index_srcdev = i; 6075 found = 1; 6076 physical_of_found = bioc->stripes[i].physical; 6077 } 6078 6079 btrfs_put_bioc(bioc); 6080 6081 ASSERT(found); 6082 if (!found) 6083 return -EIO; 6084 6085 *mirror_num = index_srcdev + 1; 6086 *physical = physical_of_found; 6087 return ret; 6088 } 6089 6090 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6091 { 6092 struct btrfs_block_group *cache; 6093 bool ret; 6094 6095 /* Non zoned filesystem does not use "to_copy" flag */ 6096 if (!btrfs_is_zoned(fs_info)) 6097 return false; 6098 6099 cache = btrfs_lookup_block_group(fs_info, logical); 6100 6101 spin_lock(&cache->lock); 6102 ret = cache->to_copy; 6103 spin_unlock(&cache->lock); 6104 6105 btrfs_put_block_group(cache); 6106 return ret; 6107 } 6108 6109 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6110 struct btrfs_io_context **bioc_ret, 6111 struct btrfs_dev_replace *dev_replace, 6112 u64 logical, 6113 int *num_stripes_ret, int *max_errors_ret) 6114 { 6115 struct btrfs_io_context *bioc = *bioc_ret; 6116 u64 srcdev_devid = dev_replace->srcdev->devid; 6117 int tgtdev_indexes = 0; 6118 int num_stripes = *num_stripes_ret; 6119 int max_errors = *max_errors_ret; 6120 int i; 6121 6122 if (op == BTRFS_MAP_WRITE) { 6123 int index_where_to_add; 6124 6125 /* 6126 * A block group which have "to_copy" set will eventually 6127 * copied by dev-replace process. We can avoid cloning IO here. 6128 */ 6129 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6130 return; 6131 6132 /* 6133 * duplicate the write operations while the dev replace 6134 * procedure is running. Since the copying of the old disk to 6135 * the new disk takes place at run time while the filesystem is 6136 * mounted writable, the regular write operations to the old 6137 * disk have to be duplicated to go to the new disk as well. 6138 * 6139 * Note that device->missing is handled by the caller, and that 6140 * the write to the old disk is already set up in the stripes 6141 * array. 6142 */ 6143 index_where_to_add = num_stripes; 6144 for (i = 0; i < num_stripes; i++) { 6145 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6146 /* write to new disk, too */ 6147 struct btrfs_io_stripe *new = 6148 bioc->stripes + index_where_to_add; 6149 struct btrfs_io_stripe *old = 6150 bioc->stripes + i; 6151 6152 new->physical = old->physical; 6153 new->length = old->length; 6154 new->dev = dev_replace->tgtdev; 6155 bioc->tgtdev_map[i] = index_where_to_add; 6156 index_where_to_add++; 6157 max_errors++; 6158 tgtdev_indexes++; 6159 } 6160 } 6161 num_stripes = index_where_to_add; 6162 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6163 int index_srcdev = 0; 6164 int found = 0; 6165 u64 physical_of_found = 0; 6166 6167 /* 6168 * During the dev-replace procedure, the target drive can also 6169 * be used to read data in case it is needed to repair a corrupt 6170 * block elsewhere. This is possible if the requested area is 6171 * left of the left cursor. In this area, the target drive is a 6172 * full copy of the source drive. 6173 */ 6174 for (i = 0; i < num_stripes; i++) { 6175 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6176 /* 6177 * In case of DUP, in order to keep it simple, 6178 * only add the mirror with the lowest physical 6179 * address 6180 */ 6181 if (found && 6182 physical_of_found <= bioc->stripes[i].physical) 6183 continue; 6184 index_srcdev = i; 6185 found = 1; 6186 physical_of_found = bioc->stripes[i].physical; 6187 } 6188 } 6189 if (found) { 6190 struct btrfs_io_stripe *tgtdev_stripe = 6191 bioc->stripes + num_stripes; 6192 6193 tgtdev_stripe->physical = physical_of_found; 6194 tgtdev_stripe->length = 6195 bioc->stripes[index_srcdev].length; 6196 tgtdev_stripe->dev = dev_replace->tgtdev; 6197 bioc->tgtdev_map[index_srcdev] = num_stripes; 6198 6199 tgtdev_indexes++; 6200 num_stripes++; 6201 } 6202 } 6203 6204 *num_stripes_ret = num_stripes; 6205 *max_errors_ret = max_errors; 6206 bioc->num_tgtdevs = tgtdev_indexes; 6207 *bioc_ret = bioc; 6208 } 6209 6210 static bool need_full_stripe(enum btrfs_map_op op) 6211 { 6212 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6213 } 6214 6215 /* 6216 * Calculate the geometry of a particular (address, len) tuple. This 6217 * information is used to calculate how big a particular bio can get before it 6218 * straddles a stripe. 6219 * 6220 * @fs_info: the filesystem 6221 * @em: mapping containing the logical extent 6222 * @op: type of operation - write or read 6223 * @logical: address that we want to figure out the geometry of 6224 * @io_geom: pointer used to return values 6225 * 6226 * Returns < 0 in case a chunk for the given logical address cannot be found, 6227 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6228 */ 6229 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6230 enum btrfs_map_op op, u64 logical, 6231 struct btrfs_io_geometry *io_geom) 6232 { 6233 struct map_lookup *map; 6234 u64 len; 6235 u64 offset; 6236 u64 stripe_offset; 6237 u64 stripe_nr; 6238 u64 stripe_len; 6239 u64 raid56_full_stripe_start = (u64)-1; 6240 int data_stripes; 6241 6242 ASSERT(op != BTRFS_MAP_DISCARD); 6243 6244 map = em->map_lookup; 6245 /* Offset of this logical address in the chunk */ 6246 offset = logical - em->start; 6247 /* Len of a stripe in a chunk */ 6248 stripe_len = map->stripe_len; 6249 /* Stripe where this block falls in */ 6250 stripe_nr = div64_u64(offset, stripe_len); 6251 /* Offset of stripe in the chunk */ 6252 stripe_offset = stripe_nr * stripe_len; 6253 if (offset < stripe_offset) { 6254 btrfs_crit(fs_info, 6255 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", 6256 stripe_offset, offset, em->start, logical, stripe_len); 6257 return -EINVAL; 6258 } 6259 6260 /* stripe_offset is the offset of this block in its stripe */ 6261 stripe_offset = offset - stripe_offset; 6262 data_stripes = nr_data_stripes(map); 6263 6264 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { 6265 u64 max_len = stripe_len - stripe_offset; 6266 6267 /* 6268 * In case of raid56, we need to know the stripe aligned start 6269 */ 6270 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6271 unsigned long full_stripe_len = stripe_len * data_stripes; 6272 raid56_full_stripe_start = offset; 6273 6274 /* 6275 * Allow a write of a full stripe, but make sure we 6276 * don't allow straddling of stripes 6277 */ 6278 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6279 full_stripe_len); 6280 raid56_full_stripe_start *= full_stripe_len; 6281 6282 /* 6283 * For writes to RAID[56], allow a full stripeset across 6284 * all disks. For other RAID types and for RAID[56] 6285 * reads, just allow a single stripe (on a single disk). 6286 */ 6287 if (op == BTRFS_MAP_WRITE) { 6288 max_len = stripe_len * data_stripes - 6289 (offset - raid56_full_stripe_start); 6290 } 6291 } 6292 len = min_t(u64, em->len - offset, max_len); 6293 } else { 6294 len = em->len - offset; 6295 } 6296 6297 io_geom->len = len; 6298 io_geom->offset = offset; 6299 io_geom->stripe_len = stripe_len; 6300 io_geom->stripe_nr = stripe_nr; 6301 io_geom->stripe_offset = stripe_offset; 6302 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6303 6304 return 0; 6305 } 6306 6307 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6308 enum btrfs_map_op op, 6309 u64 logical, u64 *length, 6310 struct btrfs_io_context **bioc_ret, 6311 int mirror_num, int need_raid_map) 6312 { 6313 struct extent_map *em; 6314 struct map_lookup *map; 6315 u64 stripe_offset; 6316 u64 stripe_nr; 6317 u64 stripe_len; 6318 u32 stripe_index; 6319 int data_stripes; 6320 int i; 6321 int ret = 0; 6322 int num_stripes; 6323 int max_errors = 0; 6324 int tgtdev_indexes = 0; 6325 struct btrfs_io_context *bioc = NULL; 6326 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6327 int dev_replace_is_ongoing = 0; 6328 int num_alloc_stripes; 6329 int patch_the_first_stripe_for_dev_replace = 0; 6330 u64 physical_to_patch_in_first_stripe = 0; 6331 u64 raid56_full_stripe_start = (u64)-1; 6332 struct btrfs_io_geometry geom; 6333 6334 ASSERT(bioc_ret); 6335 ASSERT(op != BTRFS_MAP_DISCARD); 6336 6337 em = btrfs_get_chunk_map(fs_info, logical, *length); 6338 ASSERT(!IS_ERR(em)); 6339 6340 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6341 if (ret < 0) 6342 return ret; 6343 6344 map = em->map_lookup; 6345 6346 *length = geom.len; 6347 stripe_len = geom.stripe_len; 6348 stripe_nr = geom.stripe_nr; 6349 stripe_offset = geom.stripe_offset; 6350 raid56_full_stripe_start = geom.raid56_stripe_offset; 6351 data_stripes = nr_data_stripes(map); 6352 6353 down_read(&dev_replace->rwsem); 6354 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6355 /* 6356 * Hold the semaphore for read during the whole operation, write is 6357 * requested at commit time but must wait. 6358 */ 6359 if (!dev_replace_is_ongoing) 6360 up_read(&dev_replace->rwsem); 6361 6362 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6363 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6364 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6365 dev_replace->srcdev->devid, 6366 &mirror_num, 6367 &physical_to_patch_in_first_stripe); 6368 if (ret) 6369 goto out; 6370 else 6371 patch_the_first_stripe_for_dev_replace = 1; 6372 } else if (mirror_num > map->num_stripes) { 6373 mirror_num = 0; 6374 } 6375 6376 num_stripes = 1; 6377 stripe_index = 0; 6378 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6379 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6380 &stripe_index); 6381 if (!need_full_stripe(op)) 6382 mirror_num = 1; 6383 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6384 if (need_full_stripe(op)) 6385 num_stripes = map->num_stripes; 6386 else if (mirror_num) 6387 stripe_index = mirror_num - 1; 6388 else { 6389 stripe_index = find_live_mirror(fs_info, map, 0, 6390 dev_replace_is_ongoing); 6391 mirror_num = stripe_index + 1; 6392 } 6393 6394 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6395 if (need_full_stripe(op)) { 6396 num_stripes = map->num_stripes; 6397 } else if (mirror_num) { 6398 stripe_index = mirror_num - 1; 6399 } else { 6400 mirror_num = 1; 6401 } 6402 6403 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6404 u32 factor = map->num_stripes / map->sub_stripes; 6405 6406 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6407 stripe_index *= map->sub_stripes; 6408 6409 if (need_full_stripe(op)) 6410 num_stripes = map->sub_stripes; 6411 else if (mirror_num) 6412 stripe_index += mirror_num - 1; 6413 else { 6414 int old_stripe_index = stripe_index; 6415 stripe_index = find_live_mirror(fs_info, map, 6416 stripe_index, 6417 dev_replace_is_ongoing); 6418 mirror_num = stripe_index - old_stripe_index + 1; 6419 } 6420 6421 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6422 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6423 /* push stripe_nr back to the start of the full stripe */ 6424 stripe_nr = div64_u64(raid56_full_stripe_start, 6425 stripe_len * data_stripes); 6426 6427 /* RAID[56] write or recovery. Return all stripes */ 6428 num_stripes = map->num_stripes; 6429 max_errors = nr_parity_stripes(map); 6430 6431 *length = map->stripe_len; 6432 stripe_index = 0; 6433 stripe_offset = 0; 6434 } else { 6435 /* 6436 * Mirror #0 or #1 means the original data block. 6437 * Mirror #2 is RAID5 parity block. 6438 * Mirror #3 is RAID6 Q block. 6439 */ 6440 stripe_nr = div_u64_rem(stripe_nr, 6441 data_stripes, &stripe_index); 6442 if (mirror_num > 1) 6443 stripe_index = data_stripes + mirror_num - 2; 6444 6445 /* We distribute the parity blocks across stripes */ 6446 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6447 &stripe_index); 6448 if (!need_full_stripe(op) && mirror_num <= 1) 6449 mirror_num = 1; 6450 } 6451 } else { 6452 /* 6453 * after this, stripe_nr is the number of stripes on this 6454 * device we have to walk to find the data, and stripe_index is 6455 * the number of our device in the stripe array 6456 */ 6457 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6458 &stripe_index); 6459 mirror_num = stripe_index + 1; 6460 } 6461 if (stripe_index >= map->num_stripes) { 6462 btrfs_crit(fs_info, 6463 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6464 stripe_index, map->num_stripes); 6465 ret = -EINVAL; 6466 goto out; 6467 } 6468 6469 num_alloc_stripes = num_stripes; 6470 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6471 if (op == BTRFS_MAP_WRITE) 6472 num_alloc_stripes <<= 1; 6473 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6474 num_alloc_stripes++; 6475 tgtdev_indexes = num_stripes; 6476 } 6477 6478 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6479 if (!bioc) { 6480 ret = -ENOMEM; 6481 goto out; 6482 } 6483 6484 for (i = 0; i < num_stripes; i++) { 6485 bioc->stripes[i].physical = map->stripes[stripe_index].physical + 6486 stripe_offset + stripe_nr * map->stripe_len; 6487 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 6488 stripe_index++; 6489 } 6490 6491 /* Build raid_map */ 6492 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6493 (need_full_stripe(op) || mirror_num > 1)) { 6494 u64 tmp; 6495 unsigned rot; 6496 6497 /* Work out the disk rotation on this stripe-set */ 6498 div_u64_rem(stripe_nr, num_stripes, &rot); 6499 6500 /* Fill in the logical address of each stripe */ 6501 tmp = stripe_nr * data_stripes; 6502 for (i = 0; i < data_stripes; i++) 6503 bioc->raid_map[(i + rot) % num_stripes] = 6504 em->start + (tmp + i) * map->stripe_len; 6505 6506 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6507 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6508 bioc->raid_map[(i + rot + 1) % num_stripes] = 6509 RAID6_Q_STRIPE; 6510 6511 sort_parity_stripes(bioc, num_stripes); 6512 } 6513 6514 if (need_full_stripe(op)) 6515 max_errors = btrfs_chunk_max_errors(map); 6516 6517 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6518 need_full_stripe(op)) { 6519 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6520 &num_stripes, &max_errors); 6521 } 6522 6523 *bioc_ret = bioc; 6524 bioc->map_type = map->type; 6525 bioc->num_stripes = num_stripes; 6526 bioc->max_errors = max_errors; 6527 bioc->mirror_num = mirror_num; 6528 6529 /* 6530 * this is the case that REQ_READ && dev_replace_is_ongoing && 6531 * mirror_num == num_stripes + 1 && dev_replace target drive is 6532 * available as a mirror 6533 */ 6534 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6535 WARN_ON(num_stripes > 1); 6536 bioc->stripes[0].dev = dev_replace->tgtdev; 6537 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6538 bioc->mirror_num = map->num_stripes + 1; 6539 } 6540 out: 6541 if (dev_replace_is_ongoing) { 6542 lockdep_assert_held(&dev_replace->rwsem); 6543 /* Unlock and let waiting writers proceed */ 6544 up_read(&dev_replace->rwsem); 6545 } 6546 free_extent_map(em); 6547 return ret; 6548 } 6549 6550 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6551 u64 logical, u64 *length, 6552 struct btrfs_io_context **bioc_ret, int mirror_num) 6553 { 6554 if (op == BTRFS_MAP_DISCARD) 6555 return __btrfs_map_block_for_discard(fs_info, logical, 6556 length, bioc_ret); 6557 6558 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6559 mirror_num, 0); 6560 } 6561 6562 /* For Scrub/replace */ 6563 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6564 u64 logical, u64 *length, 6565 struct btrfs_io_context **bioc_ret) 6566 { 6567 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1); 6568 } 6569 6570 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio) 6571 { 6572 bio->bi_private = bioc->private; 6573 bio->bi_end_io = bioc->end_io; 6574 bio_endio(bio); 6575 6576 btrfs_put_bioc(bioc); 6577 } 6578 6579 static void btrfs_end_bio(struct bio *bio) 6580 { 6581 struct btrfs_io_context *bioc = bio->bi_private; 6582 int is_orig_bio = 0; 6583 6584 if (bio->bi_status) { 6585 atomic_inc(&bioc->error); 6586 if (bio->bi_status == BLK_STS_IOERR || 6587 bio->bi_status == BLK_STS_TARGET) { 6588 struct btrfs_device *dev = btrfs_bio(bio)->device; 6589 6590 ASSERT(dev->bdev); 6591 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6592 btrfs_dev_stat_inc_and_print(dev, 6593 BTRFS_DEV_STAT_WRITE_ERRS); 6594 else if (!(bio->bi_opf & REQ_RAHEAD)) 6595 btrfs_dev_stat_inc_and_print(dev, 6596 BTRFS_DEV_STAT_READ_ERRS); 6597 if (bio->bi_opf & REQ_PREFLUSH) 6598 btrfs_dev_stat_inc_and_print(dev, 6599 BTRFS_DEV_STAT_FLUSH_ERRS); 6600 } 6601 } 6602 6603 if (bio == bioc->orig_bio) 6604 is_orig_bio = 1; 6605 6606 btrfs_bio_counter_dec(bioc->fs_info); 6607 6608 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6609 if (!is_orig_bio) { 6610 bio_put(bio); 6611 bio = bioc->orig_bio; 6612 } 6613 6614 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6615 /* only send an error to the higher layers if it is 6616 * beyond the tolerance of the btrfs bio 6617 */ 6618 if (atomic_read(&bioc->error) > bioc->max_errors) { 6619 bio->bi_status = BLK_STS_IOERR; 6620 } else { 6621 /* 6622 * this bio is actually up to date, we didn't 6623 * go over the max number of errors 6624 */ 6625 bio->bi_status = BLK_STS_OK; 6626 } 6627 6628 btrfs_end_bioc(bioc, bio); 6629 } else if (!is_orig_bio) { 6630 bio_put(bio); 6631 } 6632 } 6633 6634 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio, 6635 u64 physical, struct btrfs_device *dev) 6636 { 6637 struct btrfs_fs_info *fs_info = bioc->fs_info; 6638 6639 bio->bi_private = bioc; 6640 btrfs_bio(bio)->device = dev; 6641 bio->bi_end_io = btrfs_end_bio; 6642 bio->bi_iter.bi_sector = physical >> 9; 6643 /* 6644 * For zone append writing, bi_sector must point the beginning of the 6645 * zone 6646 */ 6647 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6648 if (btrfs_dev_is_sequential(dev, physical)) { 6649 u64 zone_start = round_down(physical, fs_info->zone_size); 6650 6651 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6652 } else { 6653 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6654 bio->bi_opf |= REQ_OP_WRITE; 6655 } 6656 } 6657 btrfs_debug_in_rcu(fs_info, 6658 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6659 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6660 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6661 dev->devid, bio->bi_iter.bi_size); 6662 bio_set_dev(bio, dev->bdev); 6663 6664 btrfs_bio_counter_inc_noblocked(fs_info); 6665 6666 btrfsic_submit_bio(bio); 6667 } 6668 6669 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical) 6670 { 6671 atomic_inc(&bioc->error); 6672 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6673 /* Should be the original bio. */ 6674 WARN_ON(bio != bioc->orig_bio); 6675 6676 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6677 bio->bi_iter.bi_sector = logical >> 9; 6678 if (atomic_read(&bioc->error) > bioc->max_errors) 6679 bio->bi_status = BLK_STS_IOERR; 6680 else 6681 bio->bi_status = BLK_STS_OK; 6682 btrfs_end_bioc(bioc, bio); 6683 } 6684 } 6685 6686 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 6687 int mirror_num) 6688 { 6689 struct btrfs_device *dev; 6690 struct bio *first_bio = bio; 6691 u64 logical = bio->bi_iter.bi_sector << 9; 6692 u64 length = 0; 6693 u64 map_length; 6694 int ret; 6695 int dev_nr; 6696 int total_devs; 6697 struct btrfs_io_context *bioc = NULL; 6698 6699 length = bio->bi_iter.bi_size; 6700 map_length = length; 6701 6702 btrfs_bio_counter_inc_blocked(fs_info); 6703 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, 6704 &map_length, &bioc, mirror_num, 1); 6705 if (ret) { 6706 btrfs_bio_counter_dec(fs_info); 6707 return errno_to_blk_status(ret); 6708 } 6709 6710 total_devs = bioc->num_stripes; 6711 bioc->orig_bio = first_bio; 6712 bioc->private = first_bio->bi_private; 6713 bioc->end_io = first_bio->bi_end_io; 6714 atomic_set(&bioc->stripes_pending, bioc->num_stripes); 6715 6716 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && 6717 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { 6718 /* In this case, map_length has been set to the length of 6719 a single stripe; not the whole write */ 6720 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 6721 ret = raid56_parity_write(bio, bioc, map_length); 6722 } else { 6723 ret = raid56_parity_recover(bio, bioc, map_length, 6724 mirror_num, 1); 6725 } 6726 6727 btrfs_bio_counter_dec(fs_info); 6728 return errno_to_blk_status(ret); 6729 } 6730 6731 if (map_length < length) { 6732 btrfs_crit(fs_info, 6733 "mapping failed logical %llu bio len %llu len %llu", 6734 logical, length, map_length); 6735 BUG(); 6736 } 6737 6738 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { 6739 dev = bioc->stripes[dev_nr].dev; 6740 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, 6741 &dev->dev_state) || 6742 (btrfs_op(first_bio) == BTRFS_MAP_WRITE && 6743 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6744 bioc_error(bioc, first_bio, logical); 6745 continue; 6746 } 6747 6748 if (dev_nr < total_devs - 1) 6749 bio = btrfs_bio_clone(first_bio); 6750 else 6751 bio = first_bio; 6752 6753 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev); 6754 } 6755 btrfs_bio_counter_dec(fs_info); 6756 return BLK_STS_OK; 6757 } 6758 6759 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6760 const struct btrfs_fs_devices *fs_devices) 6761 { 6762 if (args->fsid == NULL) 6763 return true; 6764 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6765 return true; 6766 return false; 6767 } 6768 6769 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6770 const struct btrfs_device *device) 6771 { 6772 ASSERT((args->devid != (u64)-1) || args->missing); 6773 6774 if ((args->devid != (u64)-1) && device->devid != args->devid) 6775 return false; 6776 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6777 return false; 6778 if (!args->missing) 6779 return true; 6780 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6781 !device->bdev) 6782 return true; 6783 return false; 6784 } 6785 6786 /* 6787 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6788 * return NULL. 6789 * 6790 * If devid and uuid are both specified, the match must be exact, otherwise 6791 * only devid is used. 6792 */ 6793 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6794 const struct btrfs_dev_lookup_args *args) 6795 { 6796 struct btrfs_device *device; 6797 struct btrfs_fs_devices *seed_devs; 6798 6799 if (dev_args_match_fs_devices(args, fs_devices)) { 6800 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6801 if (dev_args_match_device(args, device)) 6802 return device; 6803 } 6804 } 6805 6806 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6807 if (!dev_args_match_fs_devices(args, seed_devs)) 6808 continue; 6809 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6810 if (dev_args_match_device(args, device)) 6811 return device; 6812 } 6813 } 6814 6815 return NULL; 6816 } 6817 6818 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6819 u64 devid, u8 *dev_uuid) 6820 { 6821 struct btrfs_device *device; 6822 unsigned int nofs_flag; 6823 6824 /* 6825 * We call this under the chunk_mutex, so we want to use NOFS for this 6826 * allocation, however we don't want to change btrfs_alloc_device() to 6827 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6828 * places. 6829 */ 6830 nofs_flag = memalloc_nofs_save(); 6831 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 6832 memalloc_nofs_restore(nofs_flag); 6833 if (IS_ERR(device)) 6834 return device; 6835 6836 list_add(&device->dev_list, &fs_devices->devices); 6837 device->fs_devices = fs_devices; 6838 fs_devices->num_devices++; 6839 6840 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6841 fs_devices->missing_devices++; 6842 6843 return device; 6844 } 6845 6846 /** 6847 * btrfs_alloc_device - allocate struct btrfs_device 6848 * @fs_info: used only for generating a new devid, can be NULL if 6849 * devid is provided (i.e. @devid != NULL). 6850 * @devid: a pointer to devid for this device. If NULL a new devid 6851 * is generated. 6852 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6853 * is generated. 6854 * 6855 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6856 * on error. Returned struct is not linked onto any lists and must be 6857 * destroyed with btrfs_free_device. 6858 */ 6859 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6860 const u64 *devid, 6861 const u8 *uuid) 6862 { 6863 struct btrfs_device *dev; 6864 u64 tmp; 6865 6866 if (WARN_ON(!devid && !fs_info)) 6867 return ERR_PTR(-EINVAL); 6868 6869 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6870 if (!dev) 6871 return ERR_PTR(-ENOMEM); 6872 6873 /* 6874 * Preallocate a bio that's always going to be used for flushing device 6875 * barriers and matches the device lifespan 6876 */ 6877 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0); 6878 if (!dev->flush_bio) { 6879 kfree(dev); 6880 return ERR_PTR(-ENOMEM); 6881 } 6882 6883 INIT_LIST_HEAD(&dev->dev_list); 6884 INIT_LIST_HEAD(&dev->dev_alloc_list); 6885 INIT_LIST_HEAD(&dev->post_commit_list); 6886 6887 atomic_set(&dev->reada_in_flight, 0); 6888 atomic_set(&dev->dev_stats_ccnt, 0); 6889 btrfs_device_data_ordered_init(dev); 6890 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6891 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6892 extent_io_tree_init(fs_info, &dev->alloc_state, 6893 IO_TREE_DEVICE_ALLOC_STATE, NULL); 6894 6895 if (devid) 6896 tmp = *devid; 6897 else { 6898 int ret; 6899 6900 ret = find_next_devid(fs_info, &tmp); 6901 if (ret) { 6902 btrfs_free_device(dev); 6903 return ERR_PTR(ret); 6904 } 6905 } 6906 dev->devid = tmp; 6907 6908 if (uuid) 6909 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6910 else 6911 generate_random_uuid(dev->uuid); 6912 6913 return dev; 6914 } 6915 6916 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6917 u64 devid, u8 *uuid, bool error) 6918 { 6919 if (error) 6920 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6921 devid, uuid); 6922 else 6923 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6924 devid, uuid); 6925 } 6926 6927 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) 6928 { 6929 const int data_stripes = calc_data_stripes(type, num_stripes); 6930 6931 return div_u64(chunk_len, data_stripes); 6932 } 6933 6934 #if BITS_PER_LONG == 32 6935 /* 6936 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6937 * can't be accessed on 32bit systems. 6938 * 6939 * This function do mount time check to reject the fs if it already has 6940 * metadata chunk beyond that limit. 6941 */ 6942 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6943 u64 logical, u64 length, u64 type) 6944 { 6945 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6946 return 0; 6947 6948 if (logical + length < MAX_LFS_FILESIZE) 6949 return 0; 6950 6951 btrfs_err_32bit_limit(fs_info); 6952 return -EOVERFLOW; 6953 } 6954 6955 /* 6956 * This is to give early warning for any metadata chunk reaching 6957 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6958 * Although we can still access the metadata, it's not going to be possible 6959 * once the limit is reached. 6960 */ 6961 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6962 u64 logical, u64 length, u64 type) 6963 { 6964 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6965 return; 6966 6967 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6968 return; 6969 6970 btrfs_warn_32bit_limit(fs_info); 6971 } 6972 #endif 6973 6974 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 6975 struct btrfs_chunk *chunk) 6976 { 6977 BTRFS_DEV_LOOKUP_ARGS(args); 6978 struct btrfs_fs_info *fs_info = leaf->fs_info; 6979 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 6980 struct map_lookup *map; 6981 struct extent_map *em; 6982 u64 logical; 6983 u64 length; 6984 u64 devid; 6985 u64 type; 6986 u8 uuid[BTRFS_UUID_SIZE]; 6987 int num_stripes; 6988 int ret; 6989 int i; 6990 6991 logical = key->offset; 6992 length = btrfs_chunk_length(leaf, chunk); 6993 type = btrfs_chunk_type(leaf, chunk); 6994 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 6995 6996 #if BITS_PER_LONG == 32 6997 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 6998 if (ret < 0) 6999 return ret; 7000 warn_32bit_meta_chunk(fs_info, logical, length, type); 7001 #endif 7002 7003 /* 7004 * Only need to verify chunk item if we're reading from sys chunk array, 7005 * as chunk item in tree block is already verified by tree-checker. 7006 */ 7007 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7008 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7009 if (ret) 7010 return ret; 7011 } 7012 7013 read_lock(&map_tree->lock); 7014 em = lookup_extent_mapping(map_tree, logical, 1); 7015 read_unlock(&map_tree->lock); 7016 7017 /* already mapped? */ 7018 if (em && em->start <= logical && em->start + em->len > logical) { 7019 free_extent_map(em); 7020 return 0; 7021 } else if (em) { 7022 free_extent_map(em); 7023 } 7024 7025 em = alloc_extent_map(); 7026 if (!em) 7027 return -ENOMEM; 7028 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7029 if (!map) { 7030 free_extent_map(em); 7031 return -ENOMEM; 7032 } 7033 7034 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7035 em->map_lookup = map; 7036 em->start = logical; 7037 em->len = length; 7038 em->orig_start = 0; 7039 em->block_start = 0; 7040 em->block_len = em->len; 7041 7042 map->num_stripes = num_stripes; 7043 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7044 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7045 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7046 map->type = type; 7047 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 7048 map->verified_stripes = 0; 7049 em->orig_block_len = calc_stripe_length(type, em->len, 7050 map->num_stripes); 7051 for (i = 0; i < num_stripes; i++) { 7052 map->stripes[i].physical = 7053 btrfs_stripe_offset_nr(leaf, chunk, i); 7054 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7055 args.devid = devid; 7056 read_extent_buffer(leaf, uuid, (unsigned long) 7057 btrfs_stripe_dev_uuid_nr(chunk, i), 7058 BTRFS_UUID_SIZE); 7059 args.uuid = uuid; 7060 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7061 if (!map->stripes[i].dev && 7062 !btrfs_test_opt(fs_info, DEGRADED)) { 7063 free_extent_map(em); 7064 btrfs_report_missing_device(fs_info, devid, uuid, true); 7065 return -ENOENT; 7066 } 7067 if (!map->stripes[i].dev) { 7068 map->stripes[i].dev = 7069 add_missing_dev(fs_info->fs_devices, devid, 7070 uuid); 7071 if (IS_ERR(map->stripes[i].dev)) { 7072 free_extent_map(em); 7073 btrfs_err(fs_info, 7074 "failed to init missing dev %llu: %ld", 7075 devid, PTR_ERR(map->stripes[i].dev)); 7076 return PTR_ERR(map->stripes[i].dev); 7077 } 7078 btrfs_report_missing_device(fs_info, devid, uuid, false); 7079 } 7080 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7081 &(map->stripes[i].dev->dev_state)); 7082 7083 } 7084 7085 write_lock(&map_tree->lock); 7086 ret = add_extent_mapping(map_tree, em, 0); 7087 write_unlock(&map_tree->lock); 7088 if (ret < 0) { 7089 btrfs_err(fs_info, 7090 "failed to add chunk map, start=%llu len=%llu: %d", 7091 em->start, em->len, ret); 7092 } 7093 free_extent_map(em); 7094 7095 return ret; 7096 } 7097 7098 static void fill_device_from_item(struct extent_buffer *leaf, 7099 struct btrfs_dev_item *dev_item, 7100 struct btrfs_device *device) 7101 { 7102 unsigned long ptr; 7103 7104 device->devid = btrfs_device_id(leaf, dev_item); 7105 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7106 device->total_bytes = device->disk_total_bytes; 7107 device->commit_total_bytes = device->disk_total_bytes; 7108 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7109 device->commit_bytes_used = device->bytes_used; 7110 device->type = btrfs_device_type(leaf, dev_item); 7111 device->io_align = btrfs_device_io_align(leaf, dev_item); 7112 device->io_width = btrfs_device_io_width(leaf, dev_item); 7113 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7114 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7115 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7116 7117 ptr = btrfs_device_uuid(dev_item); 7118 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7119 } 7120 7121 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7122 u8 *fsid) 7123 { 7124 struct btrfs_fs_devices *fs_devices; 7125 int ret; 7126 7127 lockdep_assert_held(&uuid_mutex); 7128 ASSERT(fsid); 7129 7130 /* This will match only for multi-device seed fs */ 7131 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7132 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7133 return fs_devices; 7134 7135 7136 fs_devices = find_fsid(fsid, NULL); 7137 if (!fs_devices) { 7138 if (!btrfs_test_opt(fs_info, DEGRADED)) 7139 return ERR_PTR(-ENOENT); 7140 7141 fs_devices = alloc_fs_devices(fsid, NULL); 7142 if (IS_ERR(fs_devices)) 7143 return fs_devices; 7144 7145 fs_devices->seeding = true; 7146 fs_devices->opened = 1; 7147 return fs_devices; 7148 } 7149 7150 /* 7151 * Upon first call for a seed fs fsid, just create a private copy of the 7152 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7153 */ 7154 fs_devices = clone_fs_devices(fs_devices); 7155 if (IS_ERR(fs_devices)) 7156 return fs_devices; 7157 7158 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7159 if (ret) { 7160 free_fs_devices(fs_devices); 7161 return ERR_PTR(ret); 7162 } 7163 7164 if (!fs_devices->seeding) { 7165 close_fs_devices(fs_devices); 7166 free_fs_devices(fs_devices); 7167 return ERR_PTR(-EINVAL); 7168 } 7169 7170 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7171 7172 return fs_devices; 7173 } 7174 7175 static int read_one_dev(struct extent_buffer *leaf, 7176 struct btrfs_dev_item *dev_item) 7177 { 7178 BTRFS_DEV_LOOKUP_ARGS(args); 7179 struct btrfs_fs_info *fs_info = leaf->fs_info; 7180 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7181 struct btrfs_device *device; 7182 u64 devid; 7183 int ret; 7184 u8 fs_uuid[BTRFS_FSID_SIZE]; 7185 u8 dev_uuid[BTRFS_UUID_SIZE]; 7186 7187 devid = args.devid = btrfs_device_id(leaf, dev_item); 7188 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7189 BTRFS_UUID_SIZE); 7190 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7191 BTRFS_FSID_SIZE); 7192 args.uuid = dev_uuid; 7193 args.fsid = fs_uuid; 7194 7195 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7196 fs_devices = open_seed_devices(fs_info, fs_uuid); 7197 if (IS_ERR(fs_devices)) 7198 return PTR_ERR(fs_devices); 7199 } 7200 7201 device = btrfs_find_device(fs_info->fs_devices, &args); 7202 if (!device) { 7203 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7204 btrfs_report_missing_device(fs_info, devid, 7205 dev_uuid, true); 7206 return -ENOENT; 7207 } 7208 7209 device = add_missing_dev(fs_devices, devid, dev_uuid); 7210 if (IS_ERR(device)) { 7211 btrfs_err(fs_info, 7212 "failed to add missing dev %llu: %ld", 7213 devid, PTR_ERR(device)); 7214 return PTR_ERR(device); 7215 } 7216 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7217 } else { 7218 if (!device->bdev) { 7219 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7220 btrfs_report_missing_device(fs_info, 7221 devid, dev_uuid, true); 7222 return -ENOENT; 7223 } 7224 btrfs_report_missing_device(fs_info, devid, 7225 dev_uuid, false); 7226 } 7227 7228 if (!device->bdev && 7229 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7230 /* 7231 * this happens when a device that was properly setup 7232 * in the device info lists suddenly goes bad. 7233 * device->bdev is NULL, and so we have to set 7234 * device->missing to one here 7235 */ 7236 device->fs_devices->missing_devices++; 7237 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7238 } 7239 7240 /* Move the device to its own fs_devices */ 7241 if (device->fs_devices != fs_devices) { 7242 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7243 &device->dev_state)); 7244 7245 list_move(&device->dev_list, &fs_devices->devices); 7246 device->fs_devices->num_devices--; 7247 fs_devices->num_devices++; 7248 7249 device->fs_devices->missing_devices--; 7250 fs_devices->missing_devices++; 7251 7252 device->fs_devices = fs_devices; 7253 } 7254 } 7255 7256 if (device->fs_devices != fs_info->fs_devices) { 7257 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7258 if (device->generation != 7259 btrfs_device_generation(leaf, dev_item)) 7260 return -EINVAL; 7261 } 7262 7263 fill_device_from_item(leaf, dev_item, device); 7264 if (device->bdev) { 7265 u64 max_total_bytes = i_size_read(device->bdev->bd_inode); 7266 7267 if (device->total_bytes > max_total_bytes) { 7268 btrfs_err(fs_info, 7269 "device total_bytes should be at most %llu but found %llu", 7270 max_total_bytes, device->total_bytes); 7271 return -EINVAL; 7272 } 7273 } 7274 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7275 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7276 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7277 device->fs_devices->total_rw_bytes += device->total_bytes; 7278 atomic64_add(device->total_bytes - device->bytes_used, 7279 &fs_info->free_chunk_space); 7280 } 7281 ret = 0; 7282 return ret; 7283 } 7284 7285 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7286 { 7287 struct btrfs_root *root = fs_info->tree_root; 7288 struct btrfs_super_block *super_copy = fs_info->super_copy; 7289 struct extent_buffer *sb; 7290 struct btrfs_disk_key *disk_key; 7291 struct btrfs_chunk *chunk; 7292 u8 *array_ptr; 7293 unsigned long sb_array_offset; 7294 int ret = 0; 7295 u32 num_stripes; 7296 u32 array_size; 7297 u32 len = 0; 7298 u32 cur_offset; 7299 u64 type; 7300 struct btrfs_key key; 7301 7302 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7303 /* 7304 * This will create extent buffer of nodesize, superblock size is 7305 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will 7306 * overallocate but we can keep it as-is, only the first page is used. 7307 */ 7308 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, 7309 root->root_key.objectid, 0); 7310 if (IS_ERR(sb)) 7311 return PTR_ERR(sb); 7312 set_extent_buffer_uptodate(sb); 7313 /* 7314 * The sb extent buffer is artificial and just used to read the system array. 7315 * set_extent_buffer_uptodate() call does not properly mark all it's 7316 * pages up-to-date when the page is larger: extent does not cover the 7317 * whole page and consequently check_page_uptodate does not find all 7318 * the page's extents up-to-date (the hole beyond sb), 7319 * write_extent_buffer then triggers a WARN_ON. 7320 * 7321 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, 7322 * but sb spans only this function. Add an explicit SetPageUptodate call 7323 * to silence the warning eg. on PowerPC 64. 7324 */ 7325 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) 7326 SetPageUptodate(sb->pages[0]); 7327 7328 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7329 array_size = btrfs_super_sys_array_size(super_copy); 7330 7331 array_ptr = super_copy->sys_chunk_array; 7332 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7333 cur_offset = 0; 7334 7335 while (cur_offset < array_size) { 7336 disk_key = (struct btrfs_disk_key *)array_ptr; 7337 len = sizeof(*disk_key); 7338 if (cur_offset + len > array_size) 7339 goto out_short_read; 7340 7341 btrfs_disk_key_to_cpu(&key, disk_key); 7342 7343 array_ptr += len; 7344 sb_array_offset += len; 7345 cur_offset += len; 7346 7347 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7348 btrfs_err(fs_info, 7349 "unexpected item type %u in sys_array at offset %u", 7350 (u32)key.type, cur_offset); 7351 ret = -EIO; 7352 break; 7353 } 7354 7355 chunk = (struct btrfs_chunk *)sb_array_offset; 7356 /* 7357 * At least one btrfs_chunk with one stripe must be present, 7358 * exact stripe count check comes afterwards 7359 */ 7360 len = btrfs_chunk_item_size(1); 7361 if (cur_offset + len > array_size) 7362 goto out_short_read; 7363 7364 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7365 if (!num_stripes) { 7366 btrfs_err(fs_info, 7367 "invalid number of stripes %u in sys_array at offset %u", 7368 num_stripes, cur_offset); 7369 ret = -EIO; 7370 break; 7371 } 7372 7373 type = btrfs_chunk_type(sb, chunk); 7374 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7375 btrfs_err(fs_info, 7376 "invalid chunk type %llu in sys_array at offset %u", 7377 type, cur_offset); 7378 ret = -EIO; 7379 break; 7380 } 7381 7382 len = btrfs_chunk_item_size(num_stripes); 7383 if (cur_offset + len > array_size) 7384 goto out_short_read; 7385 7386 ret = read_one_chunk(&key, sb, chunk); 7387 if (ret) 7388 break; 7389 7390 array_ptr += len; 7391 sb_array_offset += len; 7392 cur_offset += len; 7393 } 7394 clear_extent_buffer_uptodate(sb); 7395 free_extent_buffer_stale(sb); 7396 return ret; 7397 7398 out_short_read: 7399 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7400 len, cur_offset); 7401 clear_extent_buffer_uptodate(sb); 7402 free_extent_buffer_stale(sb); 7403 return -EIO; 7404 } 7405 7406 /* 7407 * Check if all chunks in the fs are OK for read-write degraded mount 7408 * 7409 * If the @failing_dev is specified, it's accounted as missing. 7410 * 7411 * Return true if all chunks meet the minimal RW mount requirements. 7412 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7413 */ 7414 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7415 struct btrfs_device *failing_dev) 7416 { 7417 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7418 struct extent_map *em; 7419 u64 next_start = 0; 7420 bool ret = true; 7421 7422 read_lock(&map_tree->lock); 7423 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7424 read_unlock(&map_tree->lock); 7425 /* No chunk at all? Return false anyway */ 7426 if (!em) { 7427 ret = false; 7428 goto out; 7429 } 7430 while (em) { 7431 struct map_lookup *map; 7432 int missing = 0; 7433 int max_tolerated; 7434 int i; 7435 7436 map = em->map_lookup; 7437 max_tolerated = 7438 btrfs_get_num_tolerated_disk_barrier_failures( 7439 map->type); 7440 for (i = 0; i < map->num_stripes; i++) { 7441 struct btrfs_device *dev = map->stripes[i].dev; 7442 7443 if (!dev || !dev->bdev || 7444 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7445 dev->last_flush_error) 7446 missing++; 7447 else if (failing_dev && failing_dev == dev) 7448 missing++; 7449 } 7450 if (missing > max_tolerated) { 7451 if (!failing_dev) 7452 btrfs_warn(fs_info, 7453 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7454 em->start, missing, max_tolerated); 7455 free_extent_map(em); 7456 ret = false; 7457 goto out; 7458 } 7459 next_start = extent_map_end(em); 7460 free_extent_map(em); 7461 7462 read_lock(&map_tree->lock); 7463 em = lookup_extent_mapping(map_tree, next_start, 7464 (u64)(-1) - next_start); 7465 read_unlock(&map_tree->lock); 7466 } 7467 out: 7468 return ret; 7469 } 7470 7471 static void readahead_tree_node_children(struct extent_buffer *node) 7472 { 7473 int i; 7474 const int nr_items = btrfs_header_nritems(node); 7475 7476 for (i = 0; i < nr_items; i++) 7477 btrfs_readahead_node_child(node, i); 7478 } 7479 7480 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7481 { 7482 struct btrfs_root *root = fs_info->chunk_root; 7483 struct btrfs_path *path; 7484 struct extent_buffer *leaf; 7485 struct btrfs_key key; 7486 struct btrfs_key found_key; 7487 int ret; 7488 int slot; 7489 u64 total_dev = 0; 7490 u64 last_ra_node = 0; 7491 7492 path = btrfs_alloc_path(); 7493 if (!path) 7494 return -ENOMEM; 7495 7496 /* 7497 * uuid_mutex is needed only if we are mounting a sprout FS 7498 * otherwise we don't need it. 7499 */ 7500 mutex_lock(&uuid_mutex); 7501 7502 /* 7503 * It is possible for mount and umount to race in such a way that 7504 * we execute this code path, but open_fs_devices failed to clear 7505 * total_rw_bytes. We certainly want it cleared before reading the 7506 * device items, so clear it here. 7507 */ 7508 fs_info->fs_devices->total_rw_bytes = 0; 7509 7510 /* 7511 * Read all device items, and then all the chunk items. All 7512 * device items are found before any chunk item (their object id 7513 * is smaller than the lowest possible object id for a chunk 7514 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7515 */ 7516 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7517 key.offset = 0; 7518 key.type = 0; 7519 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7520 if (ret < 0) 7521 goto error; 7522 while (1) { 7523 struct extent_buffer *node; 7524 7525 leaf = path->nodes[0]; 7526 slot = path->slots[0]; 7527 if (slot >= btrfs_header_nritems(leaf)) { 7528 ret = btrfs_next_leaf(root, path); 7529 if (ret == 0) 7530 continue; 7531 if (ret < 0) 7532 goto error; 7533 break; 7534 } 7535 /* 7536 * The nodes on level 1 are not locked but we don't need to do 7537 * that during mount time as nothing else can access the tree 7538 */ 7539 node = path->nodes[1]; 7540 if (node) { 7541 if (last_ra_node != node->start) { 7542 readahead_tree_node_children(node); 7543 last_ra_node = node->start; 7544 } 7545 } 7546 btrfs_item_key_to_cpu(leaf, &found_key, slot); 7547 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7548 struct btrfs_dev_item *dev_item; 7549 dev_item = btrfs_item_ptr(leaf, slot, 7550 struct btrfs_dev_item); 7551 ret = read_one_dev(leaf, dev_item); 7552 if (ret) 7553 goto error; 7554 total_dev++; 7555 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7556 struct btrfs_chunk *chunk; 7557 7558 /* 7559 * We are only called at mount time, so no need to take 7560 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7561 * we always lock first fs_info->chunk_mutex before 7562 * acquiring any locks on the chunk tree. This is a 7563 * requirement for chunk allocation, see the comment on 7564 * top of btrfs_chunk_alloc() for details. 7565 */ 7566 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7567 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7568 ret = read_one_chunk(&found_key, leaf, chunk); 7569 if (ret) 7570 goto error; 7571 } 7572 path->slots[0]++; 7573 } 7574 7575 /* 7576 * After loading chunk tree, we've got all device information, 7577 * do another round of validation checks. 7578 */ 7579 if (total_dev != fs_info->fs_devices->total_devices) { 7580 btrfs_err(fs_info, 7581 "super_num_devices %llu mismatch with num_devices %llu found here", 7582 btrfs_super_num_devices(fs_info->super_copy), 7583 total_dev); 7584 ret = -EINVAL; 7585 goto error; 7586 } 7587 if (btrfs_super_total_bytes(fs_info->super_copy) < 7588 fs_info->fs_devices->total_rw_bytes) { 7589 btrfs_err(fs_info, 7590 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7591 btrfs_super_total_bytes(fs_info->super_copy), 7592 fs_info->fs_devices->total_rw_bytes); 7593 ret = -EINVAL; 7594 goto error; 7595 } 7596 ret = 0; 7597 error: 7598 mutex_unlock(&uuid_mutex); 7599 7600 btrfs_free_path(path); 7601 return ret; 7602 } 7603 7604 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7605 { 7606 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7607 struct btrfs_device *device; 7608 7609 fs_devices->fs_info = fs_info; 7610 7611 mutex_lock(&fs_devices->device_list_mutex); 7612 list_for_each_entry(device, &fs_devices->devices, dev_list) 7613 device->fs_info = fs_info; 7614 7615 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7616 list_for_each_entry(device, &seed_devs->devices, dev_list) 7617 device->fs_info = fs_info; 7618 7619 seed_devs->fs_info = fs_info; 7620 } 7621 mutex_unlock(&fs_devices->device_list_mutex); 7622 } 7623 7624 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7625 const struct btrfs_dev_stats_item *ptr, 7626 int index) 7627 { 7628 u64 val; 7629 7630 read_extent_buffer(eb, &val, 7631 offsetof(struct btrfs_dev_stats_item, values) + 7632 ((unsigned long)ptr) + (index * sizeof(u64)), 7633 sizeof(val)); 7634 return val; 7635 } 7636 7637 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7638 struct btrfs_dev_stats_item *ptr, 7639 int index, u64 val) 7640 { 7641 write_extent_buffer(eb, &val, 7642 offsetof(struct btrfs_dev_stats_item, values) + 7643 ((unsigned long)ptr) + (index * sizeof(u64)), 7644 sizeof(val)); 7645 } 7646 7647 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7648 struct btrfs_path *path) 7649 { 7650 struct btrfs_dev_stats_item *ptr; 7651 struct extent_buffer *eb; 7652 struct btrfs_key key; 7653 int item_size; 7654 int i, ret, slot; 7655 7656 if (!device->fs_info->dev_root) 7657 return 0; 7658 7659 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7660 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7661 key.offset = device->devid; 7662 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7663 if (ret) { 7664 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7665 btrfs_dev_stat_set(device, i, 0); 7666 device->dev_stats_valid = 1; 7667 btrfs_release_path(path); 7668 return ret < 0 ? ret : 0; 7669 } 7670 slot = path->slots[0]; 7671 eb = path->nodes[0]; 7672 item_size = btrfs_item_size_nr(eb, slot); 7673 7674 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7675 7676 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7677 if (item_size >= (1 + i) * sizeof(__le64)) 7678 btrfs_dev_stat_set(device, i, 7679 btrfs_dev_stats_value(eb, ptr, i)); 7680 else 7681 btrfs_dev_stat_set(device, i, 0); 7682 } 7683 7684 device->dev_stats_valid = 1; 7685 btrfs_dev_stat_print_on_load(device); 7686 btrfs_release_path(path); 7687 7688 return 0; 7689 } 7690 7691 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7692 { 7693 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7694 struct btrfs_device *device; 7695 struct btrfs_path *path = NULL; 7696 int ret = 0; 7697 7698 path = btrfs_alloc_path(); 7699 if (!path) 7700 return -ENOMEM; 7701 7702 mutex_lock(&fs_devices->device_list_mutex); 7703 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7704 ret = btrfs_device_init_dev_stats(device, path); 7705 if (ret) 7706 goto out; 7707 } 7708 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7709 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7710 ret = btrfs_device_init_dev_stats(device, path); 7711 if (ret) 7712 goto out; 7713 } 7714 } 7715 out: 7716 mutex_unlock(&fs_devices->device_list_mutex); 7717 7718 btrfs_free_path(path); 7719 return ret; 7720 } 7721 7722 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7723 struct btrfs_device *device) 7724 { 7725 struct btrfs_fs_info *fs_info = trans->fs_info; 7726 struct btrfs_root *dev_root = fs_info->dev_root; 7727 struct btrfs_path *path; 7728 struct btrfs_key key; 7729 struct extent_buffer *eb; 7730 struct btrfs_dev_stats_item *ptr; 7731 int ret; 7732 int i; 7733 7734 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7735 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7736 key.offset = device->devid; 7737 7738 path = btrfs_alloc_path(); 7739 if (!path) 7740 return -ENOMEM; 7741 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7742 if (ret < 0) { 7743 btrfs_warn_in_rcu(fs_info, 7744 "error %d while searching for dev_stats item for device %s", 7745 ret, rcu_str_deref(device->name)); 7746 goto out; 7747 } 7748 7749 if (ret == 0 && 7750 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7751 /* need to delete old one and insert a new one */ 7752 ret = btrfs_del_item(trans, dev_root, path); 7753 if (ret != 0) { 7754 btrfs_warn_in_rcu(fs_info, 7755 "delete too small dev_stats item for device %s failed %d", 7756 rcu_str_deref(device->name), ret); 7757 goto out; 7758 } 7759 ret = 1; 7760 } 7761 7762 if (ret == 1) { 7763 /* need to insert a new item */ 7764 btrfs_release_path(path); 7765 ret = btrfs_insert_empty_item(trans, dev_root, path, 7766 &key, sizeof(*ptr)); 7767 if (ret < 0) { 7768 btrfs_warn_in_rcu(fs_info, 7769 "insert dev_stats item for device %s failed %d", 7770 rcu_str_deref(device->name), ret); 7771 goto out; 7772 } 7773 } 7774 7775 eb = path->nodes[0]; 7776 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7777 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7778 btrfs_set_dev_stats_value(eb, ptr, i, 7779 btrfs_dev_stat_read(device, i)); 7780 btrfs_mark_buffer_dirty(eb); 7781 7782 out: 7783 btrfs_free_path(path); 7784 return ret; 7785 } 7786 7787 /* 7788 * called from commit_transaction. Writes all changed device stats to disk. 7789 */ 7790 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7791 { 7792 struct btrfs_fs_info *fs_info = trans->fs_info; 7793 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7794 struct btrfs_device *device; 7795 int stats_cnt; 7796 int ret = 0; 7797 7798 mutex_lock(&fs_devices->device_list_mutex); 7799 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7800 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7801 if (!device->dev_stats_valid || stats_cnt == 0) 7802 continue; 7803 7804 7805 /* 7806 * There is a LOAD-LOAD control dependency between the value of 7807 * dev_stats_ccnt and updating the on-disk values which requires 7808 * reading the in-memory counters. Such control dependencies 7809 * require explicit read memory barriers. 7810 * 7811 * This memory barriers pairs with smp_mb__before_atomic in 7812 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7813 * barrier implied by atomic_xchg in 7814 * btrfs_dev_stats_read_and_reset 7815 */ 7816 smp_rmb(); 7817 7818 ret = update_dev_stat_item(trans, device); 7819 if (!ret) 7820 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7821 } 7822 mutex_unlock(&fs_devices->device_list_mutex); 7823 7824 return ret; 7825 } 7826 7827 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7828 { 7829 btrfs_dev_stat_inc(dev, index); 7830 btrfs_dev_stat_print_on_error(dev); 7831 } 7832 7833 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) 7834 { 7835 if (!dev->dev_stats_valid) 7836 return; 7837 btrfs_err_rl_in_rcu(dev->fs_info, 7838 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7839 rcu_str_deref(dev->name), 7840 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7841 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7842 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7843 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7844 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7845 } 7846 7847 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7848 { 7849 int i; 7850 7851 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7852 if (btrfs_dev_stat_read(dev, i) != 0) 7853 break; 7854 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7855 return; /* all values == 0, suppress message */ 7856 7857 btrfs_info_in_rcu(dev->fs_info, 7858 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7859 rcu_str_deref(dev->name), 7860 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7861 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7862 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7863 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7864 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7865 } 7866 7867 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7868 struct btrfs_ioctl_get_dev_stats *stats) 7869 { 7870 BTRFS_DEV_LOOKUP_ARGS(args); 7871 struct btrfs_device *dev; 7872 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7873 int i; 7874 7875 mutex_lock(&fs_devices->device_list_mutex); 7876 args.devid = stats->devid; 7877 dev = btrfs_find_device(fs_info->fs_devices, &args); 7878 mutex_unlock(&fs_devices->device_list_mutex); 7879 7880 if (!dev) { 7881 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7882 return -ENODEV; 7883 } else if (!dev->dev_stats_valid) { 7884 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7885 return -ENODEV; 7886 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7887 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7888 if (stats->nr_items > i) 7889 stats->values[i] = 7890 btrfs_dev_stat_read_and_reset(dev, i); 7891 else 7892 btrfs_dev_stat_set(dev, i, 0); 7893 } 7894 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7895 current->comm, task_pid_nr(current)); 7896 } else { 7897 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7898 if (stats->nr_items > i) 7899 stats->values[i] = btrfs_dev_stat_read(dev, i); 7900 } 7901 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7902 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7903 return 0; 7904 } 7905 7906 /* 7907 * Update the size and bytes used for each device where it changed. This is 7908 * delayed since we would otherwise get errors while writing out the 7909 * superblocks. 7910 * 7911 * Must be invoked during transaction commit. 7912 */ 7913 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7914 { 7915 struct btrfs_device *curr, *next; 7916 7917 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7918 7919 if (list_empty(&trans->dev_update_list)) 7920 return; 7921 7922 /* 7923 * We don't need the device_list_mutex here. This list is owned by the 7924 * transaction and the transaction must complete before the device is 7925 * released. 7926 */ 7927 mutex_lock(&trans->fs_info->chunk_mutex); 7928 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7929 post_commit_list) { 7930 list_del_init(&curr->post_commit_list); 7931 curr->commit_total_bytes = curr->disk_total_bytes; 7932 curr->commit_bytes_used = curr->bytes_used; 7933 } 7934 mutex_unlock(&trans->fs_info->chunk_mutex); 7935 } 7936 7937 /* 7938 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7939 */ 7940 int btrfs_bg_type_to_factor(u64 flags) 7941 { 7942 const int index = btrfs_bg_flags_to_raid_index(flags); 7943 7944 return btrfs_raid_array[index].ncopies; 7945 } 7946 7947 7948 7949 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7950 u64 chunk_offset, u64 devid, 7951 u64 physical_offset, u64 physical_len) 7952 { 7953 struct btrfs_dev_lookup_args args = { .devid = devid }; 7954 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7955 struct extent_map *em; 7956 struct map_lookup *map; 7957 struct btrfs_device *dev; 7958 u64 stripe_len; 7959 bool found = false; 7960 int ret = 0; 7961 int i; 7962 7963 read_lock(&em_tree->lock); 7964 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7965 read_unlock(&em_tree->lock); 7966 7967 if (!em) { 7968 btrfs_err(fs_info, 7969 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 7970 physical_offset, devid); 7971 ret = -EUCLEAN; 7972 goto out; 7973 } 7974 7975 map = em->map_lookup; 7976 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); 7977 if (physical_len != stripe_len) { 7978 btrfs_err(fs_info, 7979 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 7980 physical_offset, devid, em->start, physical_len, 7981 stripe_len); 7982 ret = -EUCLEAN; 7983 goto out; 7984 } 7985 7986 for (i = 0; i < map->num_stripes; i++) { 7987 if (map->stripes[i].dev->devid == devid && 7988 map->stripes[i].physical == physical_offset) { 7989 found = true; 7990 if (map->verified_stripes >= map->num_stripes) { 7991 btrfs_err(fs_info, 7992 "too many dev extents for chunk %llu found", 7993 em->start); 7994 ret = -EUCLEAN; 7995 goto out; 7996 } 7997 map->verified_stripes++; 7998 break; 7999 } 8000 } 8001 if (!found) { 8002 btrfs_err(fs_info, 8003 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8004 physical_offset, devid); 8005 ret = -EUCLEAN; 8006 } 8007 8008 /* Make sure no dev extent is beyond device boundary */ 8009 dev = btrfs_find_device(fs_info->fs_devices, &args); 8010 if (!dev) { 8011 btrfs_err(fs_info, "failed to find devid %llu", devid); 8012 ret = -EUCLEAN; 8013 goto out; 8014 } 8015 8016 if (physical_offset + physical_len > dev->disk_total_bytes) { 8017 btrfs_err(fs_info, 8018 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8019 devid, physical_offset, physical_len, 8020 dev->disk_total_bytes); 8021 ret = -EUCLEAN; 8022 goto out; 8023 } 8024 8025 if (dev->zone_info) { 8026 u64 zone_size = dev->zone_info->zone_size; 8027 8028 if (!IS_ALIGNED(physical_offset, zone_size) || 8029 !IS_ALIGNED(physical_len, zone_size)) { 8030 btrfs_err(fs_info, 8031 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8032 devid, physical_offset, physical_len); 8033 ret = -EUCLEAN; 8034 goto out; 8035 } 8036 } 8037 8038 out: 8039 free_extent_map(em); 8040 return ret; 8041 } 8042 8043 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8044 { 8045 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8046 struct extent_map *em; 8047 struct rb_node *node; 8048 int ret = 0; 8049 8050 read_lock(&em_tree->lock); 8051 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8052 em = rb_entry(node, struct extent_map, rb_node); 8053 if (em->map_lookup->num_stripes != 8054 em->map_lookup->verified_stripes) { 8055 btrfs_err(fs_info, 8056 "chunk %llu has missing dev extent, have %d expect %d", 8057 em->start, em->map_lookup->verified_stripes, 8058 em->map_lookup->num_stripes); 8059 ret = -EUCLEAN; 8060 goto out; 8061 } 8062 } 8063 out: 8064 read_unlock(&em_tree->lock); 8065 return ret; 8066 } 8067 8068 /* 8069 * Ensure that all dev extents are mapped to correct chunk, otherwise 8070 * later chunk allocation/free would cause unexpected behavior. 8071 * 8072 * NOTE: This will iterate through the whole device tree, which should be of 8073 * the same size level as the chunk tree. This slightly increases mount time. 8074 */ 8075 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8076 { 8077 struct btrfs_path *path; 8078 struct btrfs_root *root = fs_info->dev_root; 8079 struct btrfs_key key; 8080 u64 prev_devid = 0; 8081 u64 prev_dev_ext_end = 0; 8082 int ret = 0; 8083 8084 /* 8085 * We don't have a dev_root because we mounted with ignorebadroots and 8086 * failed to load the root, so we want to skip the verification in this 8087 * case for sure. 8088 * 8089 * However if the dev root is fine, but the tree itself is corrupted 8090 * we'd still fail to mount. This verification is only to make sure 8091 * writes can happen safely, so instead just bypass this check 8092 * completely in the case of IGNOREBADROOTS. 8093 */ 8094 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8095 return 0; 8096 8097 key.objectid = 1; 8098 key.type = BTRFS_DEV_EXTENT_KEY; 8099 key.offset = 0; 8100 8101 path = btrfs_alloc_path(); 8102 if (!path) 8103 return -ENOMEM; 8104 8105 path->reada = READA_FORWARD; 8106 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8107 if (ret < 0) 8108 goto out; 8109 8110 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8111 ret = btrfs_next_leaf(root, path); 8112 if (ret < 0) 8113 goto out; 8114 /* No dev extents at all? Not good */ 8115 if (ret > 0) { 8116 ret = -EUCLEAN; 8117 goto out; 8118 } 8119 } 8120 while (1) { 8121 struct extent_buffer *leaf = path->nodes[0]; 8122 struct btrfs_dev_extent *dext; 8123 int slot = path->slots[0]; 8124 u64 chunk_offset; 8125 u64 physical_offset; 8126 u64 physical_len; 8127 u64 devid; 8128 8129 btrfs_item_key_to_cpu(leaf, &key, slot); 8130 if (key.type != BTRFS_DEV_EXTENT_KEY) 8131 break; 8132 devid = key.objectid; 8133 physical_offset = key.offset; 8134 8135 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8136 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8137 physical_len = btrfs_dev_extent_length(leaf, dext); 8138 8139 /* Check if this dev extent overlaps with the previous one */ 8140 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8141 btrfs_err(fs_info, 8142 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8143 devid, physical_offset, prev_dev_ext_end); 8144 ret = -EUCLEAN; 8145 goto out; 8146 } 8147 8148 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8149 physical_offset, physical_len); 8150 if (ret < 0) 8151 goto out; 8152 prev_devid = devid; 8153 prev_dev_ext_end = physical_offset + physical_len; 8154 8155 ret = btrfs_next_item(root, path); 8156 if (ret < 0) 8157 goto out; 8158 if (ret > 0) { 8159 ret = 0; 8160 break; 8161 } 8162 } 8163 8164 /* Ensure all chunks have corresponding dev extents */ 8165 ret = verify_chunk_dev_extent_mapping(fs_info); 8166 out: 8167 btrfs_free_path(path); 8168 return ret; 8169 } 8170 8171 /* 8172 * Check whether the given block group or device is pinned by any inode being 8173 * used as a swapfile. 8174 */ 8175 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8176 { 8177 struct btrfs_swapfile_pin *sp; 8178 struct rb_node *node; 8179 8180 spin_lock(&fs_info->swapfile_pins_lock); 8181 node = fs_info->swapfile_pins.rb_node; 8182 while (node) { 8183 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8184 if (ptr < sp->ptr) 8185 node = node->rb_left; 8186 else if (ptr > sp->ptr) 8187 node = node->rb_right; 8188 else 8189 break; 8190 } 8191 spin_unlock(&fs_info->swapfile_pins_lock); 8192 return node != NULL; 8193 } 8194 8195 static int relocating_repair_kthread(void *data) 8196 { 8197 struct btrfs_block_group *cache = (struct btrfs_block_group *)data; 8198 struct btrfs_fs_info *fs_info = cache->fs_info; 8199 u64 target; 8200 int ret = 0; 8201 8202 target = cache->start; 8203 btrfs_put_block_group(cache); 8204 8205 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8206 btrfs_info(fs_info, 8207 "zoned: skip relocating block group %llu to repair: EBUSY", 8208 target); 8209 return -EBUSY; 8210 } 8211 8212 mutex_lock(&fs_info->reclaim_bgs_lock); 8213 8214 /* Ensure block group still exists */ 8215 cache = btrfs_lookup_block_group(fs_info, target); 8216 if (!cache) 8217 goto out; 8218 8219 if (!cache->relocating_repair) 8220 goto out; 8221 8222 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8223 if (ret < 0) 8224 goto out; 8225 8226 btrfs_info(fs_info, 8227 "zoned: relocating block group %llu to repair IO failure", 8228 target); 8229 ret = btrfs_relocate_chunk(fs_info, target); 8230 8231 out: 8232 if (cache) 8233 btrfs_put_block_group(cache); 8234 mutex_unlock(&fs_info->reclaim_bgs_lock); 8235 btrfs_exclop_finish(fs_info); 8236 8237 return ret; 8238 } 8239 8240 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8241 { 8242 struct btrfs_block_group *cache; 8243 8244 /* Do not attempt to repair in degraded state */ 8245 if (btrfs_test_opt(fs_info, DEGRADED)) 8246 return 0; 8247 8248 cache = btrfs_lookup_block_group(fs_info, logical); 8249 if (!cache) 8250 return 0; 8251 8252 spin_lock(&cache->lock); 8253 if (cache->relocating_repair) { 8254 spin_unlock(&cache->lock); 8255 btrfs_put_block_group(cache); 8256 return 0; 8257 } 8258 cache->relocating_repair = 1; 8259 spin_unlock(&cache->lock); 8260 8261 kthread_run(relocating_repair_kthread, cache, 8262 "btrfs-relocating-repair"); 8263 8264 return 0; 8265 } 8266