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 mutex_lock(&fs_devices->device_list_mutex); 816 device = btrfs_find_device(fs_devices, devid, 817 disk_super->dev_item.uuid, NULL); 818 819 /* 820 * If this disk has been pulled into an fs devices created by 821 * a device which had the CHANGING_FSID_V2 flag then replace the 822 * metadata_uuid/fsid values of the fs_devices. 823 */ 824 if (fs_devices->fsid_change && 825 found_transid > fs_devices->latest_generation) { 826 memcpy(fs_devices->fsid, disk_super->fsid, 827 BTRFS_FSID_SIZE); 828 829 if (has_metadata_uuid) 830 memcpy(fs_devices->metadata_uuid, 831 disk_super->metadata_uuid, 832 BTRFS_FSID_SIZE); 833 else 834 memcpy(fs_devices->metadata_uuid, 835 disk_super->fsid, BTRFS_FSID_SIZE); 836 837 fs_devices->fsid_change = false; 838 } 839 } 840 841 if (!device) { 842 if (fs_devices->opened) { 843 mutex_unlock(&fs_devices->device_list_mutex); 844 return ERR_PTR(-EBUSY); 845 } 846 847 device = btrfs_alloc_device(NULL, &devid, 848 disk_super->dev_item.uuid); 849 if (IS_ERR(device)) { 850 mutex_unlock(&fs_devices->device_list_mutex); 851 /* we can safely leave the fs_devices entry around */ 852 return device; 853 } 854 855 name = rcu_string_strdup(path, GFP_NOFS); 856 if (!name) { 857 btrfs_free_device(device); 858 mutex_unlock(&fs_devices->device_list_mutex); 859 return ERR_PTR(-ENOMEM); 860 } 861 rcu_assign_pointer(device->name, name); 862 863 list_add_rcu(&device->dev_list, &fs_devices->devices); 864 fs_devices->num_devices++; 865 866 device->fs_devices = fs_devices; 867 *new_device_added = true; 868 869 if (disk_super->label[0]) 870 pr_info( 871 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 872 disk_super->label, devid, found_transid, path, 873 current->comm, task_pid_nr(current)); 874 else 875 pr_info( 876 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 877 disk_super->fsid, devid, found_transid, path, 878 current->comm, task_pid_nr(current)); 879 880 } else if (!device->name || strcmp(device->name->str, path)) { 881 /* 882 * When FS is already mounted. 883 * 1. If you are here and if the device->name is NULL that 884 * means this device was missing at time of FS mount. 885 * 2. If you are here and if the device->name is different 886 * from 'path' that means either 887 * a. The same device disappeared and reappeared with 888 * different name. or 889 * b. The missing-disk-which-was-replaced, has 890 * reappeared now. 891 * 892 * We must allow 1 and 2a above. But 2b would be a spurious 893 * and unintentional. 894 * 895 * Further in case of 1 and 2a above, the disk at 'path' 896 * would have missed some transaction when it was away and 897 * in case of 2a the stale bdev has to be updated as well. 898 * 2b must not be allowed at all time. 899 */ 900 901 /* 902 * For now, we do allow update to btrfs_fs_device through the 903 * btrfs dev scan cli after FS has been mounted. We're still 904 * tracking a problem where systems fail mount by subvolume id 905 * when we reject replacement on a mounted FS. 906 */ 907 if (!fs_devices->opened && found_transid < device->generation) { 908 /* 909 * That is if the FS is _not_ mounted and if you 910 * are here, that means there is more than one 911 * disk with same uuid and devid.We keep the one 912 * with larger generation number or the last-in if 913 * generation are equal. 914 */ 915 mutex_unlock(&fs_devices->device_list_mutex); 916 return ERR_PTR(-EEXIST); 917 } 918 919 /* 920 * We are going to replace the device path for a given devid, 921 * make sure it's the same device if the device is mounted 922 */ 923 if (device->bdev) { 924 int error; 925 dev_t path_dev; 926 927 error = lookup_bdev(path, &path_dev); 928 if (error) { 929 mutex_unlock(&fs_devices->device_list_mutex); 930 return ERR_PTR(error); 931 } 932 933 if (device->bdev->bd_dev != path_dev) { 934 mutex_unlock(&fs_devices->device_list_mutex); 935 /* 936 * device->fs_info may not be reliable here, so 937 * pass in a NULL instead. This avoids a 938 * possible use-after-free when the fs_info and 939 * fs_info->sb are already torn down. 940 */ 941 btrfs_warn_in_rcu(NULL, 942 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 943 path, devid, found_transid, 944 current->comm, 945 task_pid_nr(current)); 946 return ERR_PTR(-EEXIST); 947 } 948 btrfs_info_in_rcu(device->fs_info, 949 "devid %llu device path %s changed to %s scanned by %s (%d)", 950 devid, rcu_str_deref(device->name), 951 path, current->comm, 952 task_pid_nr(current)); 953 } 954 955 name = rcu_string_strdup(path, GFP_NOFS); 956 if (!name) { 957 mutex_unlock(&fs_devices->device_list_mutex); 958 return ERR_PTR(-ENOMEM); 959 } 960 rcu_string_free(device->name); 961 rcu_assign_pointer(device->name, name); 962 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 963 fs_devices->missing_devices--; 964 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 965 } 966 } 967 968 /* 969 * Unmount does not free the btrfs_device struct but would zero 970 * generation along with most of the other members. So just update 971 * it back. We need it to pick the disk with largest generation 972 * (as above). 973 */ 974 if (!fs_devices->opened) { 975 device->generation = found_transid; 976 fs_devices->latest_generation = max_t(u64, found_transid, 977 fs_devices->latest_generation); 978 } 979 980 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 981 982 mutex_unlock(&fs_devices->device_list_mutex); 983 return device; 984 } 985 986 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 987 { 988 struct btrfs_fs_devices *fs_devices; 989 struct btrfs_device *device; 990 struct btrfs_device *orig_dev; 991 int ret = 0; 992 993 lockdep_assert_held(&uuid_mutex); 994 995 fs_devices = alloc_fs_devices(orig->fsid, NULL); 996 if (IS_ERR(fs_devices)) 997 return fs_devices; 998 999 fs_devices->total_devices = orig->total_devices; 1000 1001 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 1002 struct rcu_string *name; 1003 1004 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1005 orig_dev->uuid); 1006 if (IS_ERR(device)) { 1007 ret = PTR_ERR(device); 1008 goto error; 1009 } 1010 1011 /* 1012 * This is ok to do without rcu read locked because we hold the 1013 * uuid mutex so nothing we touch in here is going to disappear. 1014 */ 1015 if (orig_dev->name) { 1016 name = rcu_string_strdup(orig_dev->name->str, 1017 GFP_KERNEL); 1018 if (!name) { 1019 btrfs_free_device(device); 1020 ret = -ENOMEM; 1021 goto error; 1022 } 1023 rcu_assign_pointer(device->name, name); 1024 } 1025 1026 list_add(&device->dev_list, &fs_devices->devices); 1027 device->fs_devices = fs_devices; 1028 fs_devices->num_devices++; 1029 } 1030 return fs_devices; 1031 error: 1032 free_fs_devices(fs_devices); 1033 return ERR_PTR(ret); 1034 } 1035 1036 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1037 struct btrfs_device **latest_dev) 1038 { 1039 struct btrfs_device *device, *next; 1040 1041 /* This is the initialized path, it is safe to release the devices. */ 1042 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1043 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1044 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1045 &device->dev_state) && 1046 !test_bit(BTRFS_DEV_STATE_MISSING, 1047 &device->dev_state) && 1048 (!*latest_dev || 1049 device->generation > (*latest_dev)->generation)) { 1050 *latest_dev = device; 1051 } 1052 continue; 1053 } 1054 1055 /* 1056 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1057 * in btrfs_init_dev_replace() so just continue. 1058 */ 1059 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1060 continue; 1061 1062 if (device->bdev) { 1063 blkdev_put(device->bdev, device->mode); 1064 device->bdev = NULL; 1065 fs_devices->open_devices--; 1066 } 1067 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1068 list_del_init(&device->dev_alloc_list); 1069 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1070 fs_devices->rw_devices--; 1071 } 1072 list_del_init(&device->dev_list); 1073 fs_devices->num_devices--; 1074 btrfs_free_device(device); 1075 } 1076 1077 } 1078 1079 /* 1080 * After we have read the system tree and know devids belonging to this 1081 * filesystem, remove the device which does not belong there. 1082 */ 1083 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1084 { 1085 struct btrfs_device *latest_dev = NULL; 1086 struct btrfs_fs_devices *seed_dev; 1087 1088 mutex_lock(&uuid_mutex); 1089 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1090 1091 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1092 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1093 1094 fs_devices->latest_dev = latest_dev; 1095 1096 mutex_unlock(&uuid_mutex); 1097 } 1098 1099 static void btrfs_close_bdev(struct btrfs_device *device) 1100 { 1101 if (!device->bdev) 1102 return; 1103 1104 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1105 sync_blockdev(device->bdev); 1106 invalidate_bdev(device->bdev); 1107 } 1108 1109 blkdev_put(device->bdev, device->mode); 1110 } 1111 1112 static void btrfs_close_one_device(struct btrfs_device *device) 1113 { 1114 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1115 1116 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1117 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1118 list_del_init(&device->dev_alloc_list); 1119 fs_devices->rw_devices--; 1120 } 1121 1122 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1123 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1124 1125 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 1126 fs_devices->missing_devices--; 1127 1128 btrfs_close_bdev(device); 1129 if (device->bdev) { 1130 fs_devices->open_devices--; 1131 device->bdev = NULL; 1132 } 1133 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1134 btrfs_destroy_dev_zone_info(device); 1135 1136 device->fs_info = NULL; 1137 atomic_set(&device->dev_stats_ccnt, 0); 1138 extent_io_tree_release(&device->alloc_state); 1139 1140 /* 1141 * Reset the flush error record. We might have a transient flush error 1142 * in this mount, and if so we aborted the current transaction and set 1143 * the fs to an error state, guaranteeing no super blocks can be further 1144 * committed. However that error might be transient and if we unmount the 1145 * filesystem and mount it again, we should allow the mount to succeed 1146 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1147 * filesystem again we still get flush errors, then we will again abort 1148 * any transaction and set the error state, guaranteeing no commits of 1149 * unsafe super blocks. 1150 */ 1151 device->last_flush_error = 0; 1152 1153 /* Verify the device is back in a pristine state */ 1154 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1155 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1156 ASSERT(list_empty(&device->dev_alloc_list)); 1157 ASSERT(list_empty(&device->post_commit_list)); 1158 ASSERT(atomic_read(&device->reada_in_flight) == 0); 1159 } 1160 1161 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1162 { 1163 struct btrfs_device *device, *tmp; 1164 1165 lockdep_assert_held(&uuid_mutex); 1166 1167 if (--fs_devices->opened > 0) 1168 return; 1169 1170 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1171 btrfs_close_one_device(device); 1172 1173 WARN_ON(fs_devices->open_devices); 1174 WARN_ON(fs_devices->rw_devices); 1175 fs_devices->opened = 0; 1176 fs_devices->seeding = false; 1177 fs_devices->fs_info = NULL; 1178 } 1179 1180 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1181 { 1182 LIST_HEAD(list); 1183 struct btrfs_fs_devices *tmp; 1184 1185 mutex_lock(&uuid_mutex); 1186 close_fs_devices(fs_devices); 1187 if (!fs_devices->opened) 1188 list_splice_init(&fs_devices->seed_list, &list); 1189 1190 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1191 close_fs_devices(fs_devices); 1192 list_del(&fs_devices->seed_list); 1193 free_fs_devices(fs_devices); 1194 } 1195 mutex_unlock(&uuid_mutex); 1196 } 1197 1198 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1199 fmode_t flags, void *holder) 1200 { 1201 struct btrfs_device *device; 1202 struct btrfs_device *latest_dev = NULL; 1203 struct btrfs_device *tmp_device; 1204 1205 flags |= FMODE_EXCL; 1206 1207 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1208 dev_list) { 1209 int ret; 1210 1211 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1212 if (ret == 0 && 1213 (!latest_dev || device->generation > latest_dev->generation)) { 1214 latest_dev = device; 1215 } else if (ret == -ENODATA) { 1216 fs_devices->num_devices--; 1217 list_del(&device->dev_list); 1218 btrfs_free_device(device); 1219 } 1220 } 1221 if (fs_devices->open_devices == 0) 1222 return -EINVAL; 1223 1224 fs_devices->opened = 1; 1225 fs_devices->latest_dev = latest_dev; 1226 fs_devices->total_rw_bytes = 0; 1227 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1228 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1229 1230 return 0; 1231 } 1232 1233 static int devid_cmp(void *priv, const struct list_head *a, 1234 const struct list_head *b) 1235 { 1236 const struct btrfs_device *dev1, *dev2; 1237 1238 dev1 = list_entry(a, struct btrfs_device, dev_list); 1239 dev2 = list_entry(b, struct btrfs_device, dev_list); 1240 1241 if (dev1->devid < dev2->devid) 1242 return -1; 1243 else if (dev1->devid > dev2->devid) 1244 return 1; 1245 return 0; 1246 } 1247 1248 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1249 fmode_t flags, void *holder) 1250 { 1251 int ret; 1252 1253 lockdep_assert_held(&uuid_mutex); 1254 /* 1255 * The device_list_mutex cannot be taken here in case opening the 1256 * underlying device takes further locks like open_mutex. 1257 * 1258 * We also don't need the lock here as this is called during mount and 1259 * exclusion is provided by uuid_mutex 1260 */ 1261 1262 if (fs_devices->opened) { 1263 fs_devices->opened++; 1264 ret = 0; 1265 } else { 1266 list_sort(NULL, &fs_devices->devices, devid_cmp); 1267 ret = open_fs_devices(fs_devices, flags, holder); 1268 } 1269 1270 return ret; 1271 } 1272 1273 void btrfs_release_disk_super(struct btrfs_super_block *super) 1274 { 1275 struct page *page = virt_to_page(super); 1276 1277 put_page(page); 1278 } 1279 1280 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1281 u64 bytenr, u64 bytenr_orig) 1282 { 1283 struct btrfs_super_block *disk_super; 1284 struct page *page; 1285 void *p; 1286 pgoff_t index; 1287 1288 /* make sure our super fits in the device */ 1289 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) 1290 return ERR_PTR(-EINVAL); 1291 1292 /* make sure our super fits in the page */ 1293 if (sizeof(*disk_super) > PAGE_SIZE) 1294 return ERR_PTR(-EINVAL); 1295 1296 /* make sure our super doesn't straddle pages on disk */ 1297 index = bytenr >> PAGE_SHIFT; 1298 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1299 return ERR_PTR(-EINVAL); 1300 1301 /* pull in the page with our super */ 1302 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1303 1304 if (IS_ERR(page)) 1305 return ERR_CAST(page); 1306 1307 p = page_address(page); 1308 1309 /* align our pointer to the offset of the super block */ 1310 disk_super = p + offset_in_page(bytenr); 1311 1312 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1313 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1314 btrfs_release_disk_super(p); 1315 return ERR_PTR(-EINVAL); 1316 } 1317 1318 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1319 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1320 1321 return disk_super; 1322 } 1323 1324 int btrfs_forget_devices(const char *path) 1325 { 1326 int ret; 1327 1328 mutex_lock(&uuid_mutex); 1329 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); 1330 mutex_unlock(&uuid_mutex); 1331 1332 return ret; 1333 } 1334 1335 /* 1336 * Look for a btrfs signature on a device. This may be called out of the mount path 1337 * and we are not allowed to call set_blocksize during the scan. The superblock 1338 * is read via pagecache 1339 */ 1340 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1341 void *holder) 1342 { 1343 struct btrfs_super_block *disk_super; 1344 bool new_device_added = false; 1345 struct btrfs_device *device = NULL; 1346 struct block_device *bdev; 1347 u64 bytenr, bytenr_orig; 1348 int ret; 1349 1350 lockdep_assert_held(&uuid_mutex); 1351 1352 /* 1353 * we would like to check all the supers, but that would make 1354 * a btrfs mount succeed after a mkfs from a different FS. 1355 * So, we need to add a special mount option to scan for 1356 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1357 */ 1358 flags |= FMODE_EXCL; 1359 1360 bdev = blkdev_get_by_path(path, flags, holder); 1361 if (IS_ERR(bdev)) 1362 return ERR_CAST(bdev); 1363 1364 bytenr_orig = btrfs_sb_offset(0); 1365 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1366 if (ret) 1367 return ERR_PTR(ret); 1368 1369 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1370 if (IS_ERR(disk_super)) { 1371 device = ERR_CAST(disk_super); 1372 goto error_bdev_put; 1373 } 1374 1375 device = device_list_add(path, disk_super, &new_device_added); 1376 if (!IS_ERR(device)) { 1377 if (new_device_added) 1378 btrfs_free_stale_devices(path, device); 1379 } 1380 1381 btrfs_release_disk_super(disk_super); 1382 1383 error_bdev_put: 1384 blkdev_put(bdev, flags); 1385 1386 return device; 1387 } 1388 1389 /* 1390 * Try to find a chunk that intersects [start, start + len] range and when one 1391 * such is found, record the end of it in *start 1392 */ 1393 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1394 u64 len) 1395 { 1396 u64 physical_start, physical_end; 1397 1398 lockdep_assert_held(&device->fs_info->chunk_mutex); 1399 1400 if (!find_first_extent_bit(&device->alloc_state, *start, 1401 &physical_start, &physical_end, 1402 CHUNK_ALLOCATED, NULL)) { 1403 1404 if (in_range(physical_start, *start, len) || 1405 in_range(*start, physical_start, 1406 physical_end - physical_start)) { 1407 *start = physical_end + 1; 1408 return true; 1409 } 1410 } 1411 return false; 1412 } 1413 1414 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1415 { 1416 switch (device->fs_devices->chunk_alloc_policy) { 1417 case BTRFS_CHUNK_ALLOC_REGULAR: 1418 /* 1419 * We don't want to overwrite the superblock on the drive nor 1420 * any area used by the boot loader (grub for example), so we 1421 * make sure to start at an offset of at least 1MB. 1422 */ 1423 return max_t(u64, start, SZ_1M); 1424 case BTRFS_CHUNK_ALLOC_ZONED: 1425 /* 1426 * We don't care about the starting region like regular 1427 * allocator, because we anyway use/reserve the first two zones 1428 * for superblock logging. 1429 */ 1430 return ALIGN(start, device->zone_info->zone_size); 1431 default: 1432 BUG(); 1433 } 1434 } 1435 1436 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1437 u64 *hole_start, u64 *hole_size, 1438 u64 num_bytes) 1439 { 1440 u64 zone_size = device->zone_info->zone_size; 1441 u64 pos; 1442 int ret; 1443 bool changed = false; 1444 1445 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1446 1447 while (*hole_size > 0) { 1448 pos = btrfs_find_allocatable_zones(device, *hole_start, 1449 *hole_start + *hole_size, 1450 num_bytes); 1451 if (pos != *hole_start) { 1452 *hole_size = *hole_start + *hole_size - pos; 1453 *hole_start = pos; 1454 changed = true; 1455 if (*hole_size < num_bytes) 1456 break; 1457 } 1458 1459 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1460 1461 /* Range is ensured to be empty */ 1462 if (!ret) 1463 return changed; 1464 1465 /* Given hole range was invalid (outside of device) */ 1466 if (ret == -ERANGE) { 1467 *hole_start += *hole_size; 1468 *hole_size = 0; 1469 return true; 1470 } 1471 1472 *hole_start += zone_size; 1473 *hole_size -= zone_size; 1474 changed = true; 1475 } 1476 1477 return changed; 1478 } 1479 1480 /** 1481 * dev_extent_hole_check - check if specified hole is suitable for allocation 1482 * @device: the device which we have the hole 1483 * @hole_start: starting position of the hole 1484 * @hole_size: the size of the hole 1485 * @num_bytes: the size of the free space that we need 1486 * 1487 * This function may modify @hole_start and @hole_size to reflect the suitable 1488 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1489 */ 1490 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1491 u64 *hole_size, u64 num_bytes) 1492 { 1493 bool changed = false; 1494 u64 hole_end = *hole_start + *hole_size; 1495 1496 for (;;) { 1497 /* 1498 * Check before we set max_hole_start, otherwise we could end up 1499 * sending back this offset anyway. 1500 */ 1501 if (contains_pending_extent(device, hole_start, *hole_size)) { 1502 if (hole_end >= *hole_start) 1503 *hole_size = hole_end - *hole_start; 1504 else 1505 *hole_size = 0; 1506 changed = true; 1507 } 1508 1509 switch (device->fs_devices->chunk_alloc_policy) { 1510 case BTRFS_CHUNK_ALLOC_REGULAR: 1511 /* No extra check */ 1512 break; 1513 case BTRFS_CHUNK_ALLOC_ZONED: 1514 if (dev_extent_hole_check_zoned(device, hole_start, 1515 hole_size, num_bytes)) { 1516 changed = true; 1517 /* 1518 * The changed hole can contain pending extent. 1519 * Loop again to check that. 1520 */ 1521 continue; 1522 } 1523 break; 1524 default: 1525 BUG(); 1526 } 1527 1528 break; 1529 } 1530 1531 return changed; 1532 } 1533 1534 /* 1535 * find_free_dev_extent_start - find free space in the specified device 1536 * @device: the device which we search the free space in 1537 * @num_bytes: the size of the free space that we need 1538 * @search_start: the position from which to begin the search 1539 * @start: store the start of the free space. 1540 * @len: the size of the free space. that we find, or the size 1541 * of the max free space if we don't find suitable free space 1542 * 1543 * this uses a pretty simple search, the expectation is that it is 1544 * called very infrequently and that a given device has a small number 1545 * of extents 1546 * 1547 * @start is used to store the start of the free space if we find. But if we 1548 * don't find suitable free space, it will be used to store the start position 1549 * of the max free space. 1550 * 1551 * @len is used to store the size of the free space that we find. 1552 * But if we don't find suitable free space, it is used to store the size of 1553 * the max free space. 1554 * 1555 * NOTE: This function will search *commit* root of device tree, and does extra 1556 * check to ensure dev extents are not double allocated. 1557 * This makes the function safe to allocate dev extents but may not report 1558 * correct usable device space, as device extent freed in current transaction 1559 * is not reported as available. 1560 */ 1561 static int find_free_dev_extent_start(struct btrfs_device *device, 1562 u64 num_bytes, u64 search_start, u64 *start, 1563 u64 *len) 1564 { 1565 struct btrfs_fs_info *fs_info = device->fs_info; 1566 struct btrfs_root *root = fs_info->dev_root; 1567 struct btrfs_key key; 1568 struct btrfs_dev_extent *dev_extent; 1569 struct btrfs_path *path; 1570 u64 hole_size; 1571 u64 max_hole_start; 1572 u64 max_hole_size; 1573 u64 extent_end; 1574 u64 search_end = device->total_bytes; 1575 int ret; 1576 int slot; 1577 struct extent_buffer *l; 1578 1579 search_start = dev_extent_search_start(device, search_start); 1580 1581 WARN_ON(device->zone_info && 1582 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1583 1584 path = btrfs_alloc_path(); 1585 if (!path) 1586 return -ENOMEM; 1587 1588 max_hole_start = search_start; 1589 max_hole_size = 0; 1590 1591 again: 1592 if (search_start >= search_end || 1593 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1594 ret = -ENOSPC; 1595 goto out; 1596 } 1597 1598 path->reada = READA_FORWARD; 1599 path->search_commit_root = 1; 1600 path->skip_locking = 1; 1601 1602 key.objectid = device->devid; 1603 key.offset = search_start; 1604 key.type = BTRFS_DEV_EXTENT_KEY; 1605 1606 ret = btrfs_search_backwards(root, &key, path); 1607 if (ret < 0) 1608 goto out; 1609 1610 while (1) { 1611 l = path->nodes[0]; 1612 slot = path->slots[0]; 1613 if (slot >= btrfs_header_nritems(l)) { 1614 ret = btrfs_next_leaf(root, path); 1615 if (ret == 0) 1616 continue; 1617 if (ret < 0) 1618 goto out; 1619 1620 break; 1621 } 1622 btrfs_item_key_to_cpu(l, &key, slot); 1623 1624 if (key.objectid < device->devid) 1625 goto next; 1626 1627 if (key.objectid > device->devid) 1628 break; 1629 1630 if (key.type != BTRFS_DEV_EXTENT_KEY) 1631 goto next; 1632 1633 if (key.offset > search_start) { 1634 hole_size = key.offset - search_start; 1635 dev_extent_hole_check(device, &search_start, &hole_size, 1636 num_bytes); 1637 1638 if (hole_size > max_hole_size) { 1639 max_hole_start = search_start; 1640 max_hole_size = hole_size; 1641 } 1642 1643 /* 1644 * If this free space is greater than which we need, 1645 * it must be the max free space that we have found 1646 * until now, so max_hole_start must point to the start 1647 * of this free space and the length of this free space 1648 * is stored in max_hole_size. Thus, we return 1649 * max_hole_start and max_hole_size and go back to the 1650 * caller. 1651 */ 1652 if (hole_size >= num_bytes) { 1653 ret = 0; 1654 goto out; 1655 } 1656 } 1657 1658 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1659 extent_end = key.offset + btrfs_dev_extent_length(l, 1660 dev_extent); 1661 if (extent_end > search_start) 1662 search_start = extent_end; 1663 next: 1664 path->slots[0]++; 1665 cond_resched(); 1666 } 1667 1668 /* 1669 * At this point, search_start should be the end of 1670 * allocated dev extents, and when shrinking the device, 1671 * search_end may be smaller than search_start. 1672 */ 1673 if (search_end > search_start) { 1674 hole_size = search_end - search_start; 1675 if (dev_extent_hole_check(device, &search_start, &hole_size, 1676 num_bytes)) { 1677 btrfs_release_path(path); 1678 goto again; 1679 } 1680 1681 if (hole_size > max_hole_size) { 1682 max_hole_start = search_start; 1683 max_hole_size = hole_size; 1684 } 1685 } 1686 1687 /* See above. */ 1688 if (max_hole_size < num_bytes) 1689 ret = -ENOSPC; 1690 else 1691 ret = 0; 1692 1693 out: 1694 btrfs_free_path(path); 1695 *start = max_hole_start; 1696 if (len) 1697 *len = max_hole_size; 1698 return ret; 1699 } 1700 1701 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1702 u64 *start, u64 *len) 1703 { 1704 /* FIXME use last free of some kind */ 1705 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1706 } 1707 1708 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1709 struct btrfs_device *device, 1710 u64 start, u64 *dev_extent_len) 1711 { 1712 struct btrfs_fs_info *fs_info = device->fs_info; 1713 struct btrfs_root *root = fs_info->dev_root; 1714 int ret; 1715 struct btrfs_path *path; 1716 struct btrfs_key key; 1717 struct btrfs_key found_key; 1718 struct extent_buffer *leaf = NULL; 1719 struct btrfs_dev_extent *extent = NULL; 1720 1721 path = btrfs_alloc_path(); 1722 if (!path) 1723 return -ENOMEM; 1724 1725 key.objectid = device->devid; 1726 key.offset = start; 1727 key.type = BTRFS_DEV_EXTENT_KEY; 1728 again: 1729 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1730 if (ret > 0) { 1731 ret = btrfs_previous_item(root, path, key.objectid, 1732 BTRFS_DEV_EXTENT_KEY); 1733 if (ret) 1734 goto out; 1735 leaf = path->nodes[0]; 1736 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1737 extent = btrfs_item_ptr(leaf, path->slots[0], 1738 struct btrfs_dev_extent); 1739 BUG_ON(found_key.offset > start || found_key.offset + 1740 btrfs_dev_extent_length(leaf, extent) < start); 1741 key = found_key; 1742 btrfs_release_path(path); 1743 goto again; 1744 } else if (ret == 0) { 1745 leaf = path->nodes[0]; 1746 extent = btrfs_item_ptr(leaf, path->slots[0], 1747 struct btrfs_dev_extent); 1748 } else { 1749 goto out; 1750 } 1751 1752 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1753 1754 ret = btrfs_del_item(trans, root, path); 1755 if (ret == 0) 1756 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1757 out: 1758 btrfs_free_path(path); 1759 return ret; 1760 } 1761 1762 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1763 { 1764 struct extent_map_tree *em_tree; 1765 struct extent_map *em; 1766 struct rb_node *n; 1767 u64 ret = 0; 1768 1769 em_tree = &fs_info->mapping_tree; 1770 read_lock(&em_tree->lock); 1771 n = rb_last(&em_tree->map.rb_root); 1772 if (n) { 1773 em = rb_entry(n, struct extent_map, rb_node); 1774 ret = em->start + em->len; 1775 } 1776 read_unlock(&em_tree->lock); 1777 1778 return ret; 1779 } 1780 1781 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1782 u64 *devid_ret) 1783 { 1784 int ret; 1785 struct btrfs_key key; 1786 struct btrfs_key found_key; 1787 struct btrfs_path *path; 1788 1789 path = btrfs_alloc_path(); 1790 if (!path) 1791 return -ENOMEM; 1792 1793 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1794 key.type = BTRFS_DEV_ITEM_KEY; 1795 key.offset = (u64)-1; 1796 1797 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1798 if (ret < 0) 1799 goto error; 1800 1801 if (ret == 0) { 1802 /* Corruption */ 1803 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1804 ret = -EUCLEAN; 1805 goto error; 1806 } 1807 1808 ret = btrfs_previous_item(fs_info->chunk_root, path, 1809 BTRFS_DEV_ITEMS_OBJECTID, 1810 BTRFS_DEV_ITEM_KEY); 1811 if (ret) { 1812 *devid_ret = 1; 1813 } else { 1814 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1815 path->slots[0]); 1816 *devid_ret = found_key.offset + 1; 1817 } 1818 ret = 0; 1819 error: 1820 btrfs_free_path(path); 1821 return ret; 1822 } 1823 1824 /* 1825 * the device information is stored in the chunk root 1826 * the btrfs_device struct should be fully filled in 1827 */ 1828 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1829 struct btrfs_device *device) 1830 { 1831 int ret; 1832 struct btrfs_path *path; 1833 struct btrfs_dev_item *dev_item; 1834 struct extent_buffer *leaf; 1835 struct btrfs_key key; 1836 unsigned long ptr; 1837 1838 path = btrfs_alloc_path(); 1839 if (!path) 1840 return -ENOMEM; 1841 1842 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1843 key.type = BTRFS_DEV_ITEM_KEY; 1844 key.offset = device->devid; 1845 1846 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1847 &key, sizeof(*dev_item)); 1848 if (ret) 1849 goto out; 1850 1851 leaf = path->nodes[0]; 1852 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1853 1854 btrfs_set_device_id(leaf, dev_item, device->devid); 1855 btrfs_set_device_generation(leaf, dev_item, 0); 1856 btrfs_set_device_type(leaf, dev_item, device->type); 1857 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1858 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1859 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1860 btrfs_set_device_total_bytes(leaf, dev_item, 1861 btrfs_device_get_disk_total_bytes(device)); 1862 btrfs_set_device_bytes_used(leaf, dev_item, 1863 btrfs_device_get_bytes_used(device)); 1864 btrfs_set_device_group(leaf, dev_item, 0); 1865 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1866 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1867 btrfs_set_device_start_offset(leaf, dev_item, 0); 1868 1869 ptr = btrfs_device_uuid(dev_item); 1870 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1871 ptr = btrfs_device_fsid(dev_item); 1872 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1873 ptr, BTRFS_FSID_SIZE); 1874 btrfs_mark_buffer_dirty(leaf); 1875 1876 ret = 0; 1877 out: 1878 btrfs_free_path(path); 1879 return ret; 1880 } 1881 1882 /* 1883 * Function to update ctime/mtime for a given device path. 1884 * Mainly used for ctime/mtime based probe like libblkid. 1885 */ 1886 static void update_dev_time(struct block_device *bdev) 1887 { 1888 struct inode *inode = bdev->bd_inode; 1889 struct timespec64 now; 1890 1891 /* Shouldn't happen but just in case. */ 1892 if (!inode) 1893 return; 1894 1895 now = current_time(inode); 1896 generic_update_time(inode, &now, S_MTIME | S_CTIME); 1897 } 1898 1899 static int btrfs_rm_dev_item(struct btrfs_device *device) 1900 { 1901 struct btrfs_root *root = device->fs_info->chunk_root; 1902 int ret; 1903 struct btrfs_path *path; 1904 struct btrfs_key key; 1905 struct btrfs_trans_handle *trans; 1906 1907 path = btrfs_alloc_path(); 1908 if (!path) 1909 return -ENOMEM; 1910 1911 trans = btrfs_start_transaction(root, 0); 1912 if (IS_ERR(trans)) { 1913 btrfs_free_path(path); 1914 return PTR_ERR(trans); 1915 } 1916 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1917 key.type = BTRFS_DEV_ITEM_KEY; 1918 key.offset = device->devid; 1919 1920 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1921 if (ret) { 1922 if (ret > 0) 1923 ret = -ENOENT; 1924 btrfs_abort_transaction(trans, ret); 1925 btrfs_end_transaction(trans); 1926 goto out; 1927 } 1928 1929 ret = btrfs_del_item(trans, root, path); 1930 if (ret) { 1931 btrfs_abort_transaction(trans, ret); 1932 btrfs_end_transaction(trans); 1933 } 1934 1935 out: 1936 btrfs_free_path(path); 1937 if (!ret) 1938 ret = btrfs_commit_transaction(trans); 1939 return ret; 1940 } 1941 1942 /* 1943 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1944 * filesystem. It's up to the caller to adjust that number regarding eg. device 1945 * replace. 1946 */ 1947 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1948 u64 num_devices) 1949 { 1950 u64 all_avail; 1951 unsigned seq; 1952 int i; 1953 1954 do { 1955 seq = read_seqbegin(&fs_info->profiles_lock); 1956 1957 all_avail = fs_info->avail_data_alloc_bits | 1958 fs_info->avail_system_alloc_bits | 1959 fs_info->avail_metadata_alloc_bits; 1960 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1961 1962 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1963 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1964 continue; 1965 1966 if (num_devices < btrfs_raid_array[i].devs_min) 1967 return btrfs_raid_array[i].mindev_error; 1968 } 1969 1970 return 0; 1971 } 1972 1973 static struct btrfs_device * btrfs_find_next_active_device( 1974 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1975 { 1976 struct btrfs_device *next_device; 1977 1978 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1979 if (next_device != device && 1980 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1981 && next_device->bdev) 1982 return next_device; 1983 } 1984 1985 return NULL; 1986 } 1987 1988 /* 1989 * Helper function to check if the given device is part of s_bdev / latest_dev 1990 * and replace it with the provided or the next active device, in the context 1991 * where this function called, there should be always be another device (or 1992 * this_dev) which is active. 1993 */ 1994 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1995 struct btrfs_device *next_device) 1996 { 1997 struct btrfs_fs_info *fs_info = device->fs_info; 1998 1999 if (!next_device) 2000 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2001 device); 2002 ASSERT(next_device); 2003 2004 if (fs_info->sb->s_bdev && 2005 (fs_info->sb->s_bdev == device->bdev)) 2006 fs_info->sb->s_bdev = next_device->bdev; 2007 2008 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2009 fs_info->fs_devices->latest_dev = next_device; 2010 } 2011 2012 /* 2013 * Return btrfs_fs_devices::num_devices excluding the device that's being 2014 * currently replaced. 2015 */ 2016 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2017 { 2018 u64 num_devices = fs_info->fs_devices->num_devices; 2019 2020 down_read(&fs_info->dev_replace.rwsem); 2021 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2022 ASSERT(num_devices > 1); 2023 num_devices--; 2024 } 2025 up_read(&fs_info->dev_replace.rwsem); 2026 2027 return num_devices; 2028 } 2029 2030 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2031 struct block_device *bdev, 2032 const char *device_path) 2033 { 2034 struct btrfs_super_block *disk_super; 2035 int copy_num; 2036 2037 if (!bdev) 2038 return; 2039 2040 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2041 struct page *page; 2042 int ret; 2043 2044 disk_super = btrfs_read_dev_one_super(bdev, copy_num); 2045 if (IS_ERR(disk_super)) 2046 continue; 2047 2048 if (bdev_is_zoned(bdev)) { 2049 btrfs_reset_sb_log_zones(bdev, copy_num); 2050 continue; 2051 } 2052 2053 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2054 2055 page = virt_to_page(disk_super); 2056 set_page_dirty(page); 2057 lock_page(page); 2058 /* write_on_page() unlocks the page */ 2059 ret = write_one_page(page); 2060 if (ret) 2061 btrfs_warn(fs_info, 2062 "error clearing superblock number %d (%d)", 2063 copy_num, ret); 2064 btrfs_release_disk_super(disk_super); 2065 2066 } 2067 2068 /* Notify udev that device has changed */ 2069 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2070 2071 /* Update ctime/mtime for device path for libblkid */ 2072 update_dev_time(bdev); 2073 } 2074 2075 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, 2076 u64 devid, struct block_device **bdev, fmode_t *mode) 2077 { 2078 struct btrfs_device *device; 2079 struct btrfs_fs_devices *cur_devices; 2080 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2081 u64 num_devices; 2082 int ret = 0; 2083 2084 /* 2085 * The device list in fs_devices is accessed without locks (neither 2086 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2087 * filesystem and another device rm cannot run. 2088 */ 2089 num_devices = btrfs_num_devices(fs_info); 2090 2091 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2092 if (ret) 2093 goto out; 2094 2095 device = btrfs_find_device_by_devspec(fs_info, devid, device_path); 2096 2097 if (IS_ERR(device)) { 2098 if (PTR_ERR(device) == -ENOENT && 2099 device_path && strcmp(device_path, "missing") == 0) 2100 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2101 else 2102 ret = PTR_ERR(device); 2103 goto out; 2104 } 2105 2106 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2107 btrfs_warn_in_rcu(fs_info, 2108 "cannot remove device %s (devid %llu) due to active swapfile", 2109 rcu_str_deref(device->name), device->devid); 2110 ret = -ETXTBSY; 2111 goto out; 2112 } 2113 2114 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2115 ret = BTRFS_ERROR_DEV_TGT_REPLACE; 2116 goto out; 2117 } 2118 2119 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2120 fs_info->fs_devices->rw_devices == 1) { 2121 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; 2122 goto out; 2123 } 2124 2125 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2126 mutex_lock(&fs_info->chunk_mutex); 2127 list_del_init(&device->dev_alloc_list); 2128 device->fs_devices->rw_devices--; 2129 mutex_unlock(&fs_info->chunk_mutex); 2130 } 2131 2132 ret = btrfs_shrink_device(device, 0); 2133 if (!ret) 2134 btrfs_reada_remove_dev(device); 2135 if (ret) 2136 goto error_undo; 2137 2138 /* 2139 * TODO: the superblock still includes this device in its num_devices 2140 * counter although write_all_supers() is not locked out. This 2141 * could give a filesystem state which requires a degraded mount. 2142 */ 2143 ret = btrfs_rm_dev_item(device); 2144 if (ret) 2145 goto error_undo; 2146 2147 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2148 btrfs_scrub_cancel_dev(device); 2149 2150 /* 2151 * the device list mutex makes sure that we don't change 2152 * the device list while someone else is writing out all 2153 * the device supers. Whoever is writing all supers, should 2154 * lock the device list mutex before getting the number of 2155 * devices in the super block (super_copy). Conversely, 2156 * whoever updates the number of devices in the super block 2157 * (super_copy) should hold the device list mutex. 2158 */ 2159 2160 /* 2161 * In normal cases the cur_devices == fs_devices. But in case 2162 * of deleting a seed device, the cur_devices should point to 2163 * its own fs_devices listed under the fs_devices->seed_list. 2164 */ 2165 cur_devices = device->fs_devices; 2166 mutex_lock(&fs_devices->device_list_mutex); 2167 list_del_rcu(&device->dev_list); 2168 2169 cur_devices->num_devices--; 2170 cur_devices->total_devices--; 2171 /* Update total_devices of the parent fs_devices if it's seed */ 2172 if (cur_devices != fs_devices) 2173 fs_devices->total_devices--; 2174 2175 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2176 cur_devices->missing_devices--; 2177 2178 btrfs_assign_next_active_device(device, NULL); 2179 2180 if (device->bdev) { 2181 cur_devices->open_devices--; 2182 /* remove sysfs entry */ 2183 btrfs_sysfs_remove_device(device); 2184 } 2185 2186 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2187 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2188 mutex_unlock(&fs_devices->device_list_mutex); 2189 2190 /* 2191 * At this point, the device is zero sized and detached from the 2192 * devices list. All that's left is to zero out the old supers and 2193 * free the device. 2194 * 2195 * We cannot call btrfs_close_bdev() here because we're holding the sb 2196 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2197 * block device and it's dependencies. Instead just flush the device 2198 * and let the caller do the final blkdev_put. 2199 */ 2200 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2201 btrfs_scratch_superblocks(fs_info, device->bdev, 2202 device->name->str); 2203 if (device->bdev) { 2204 sync_blockdev(device->bdev); 2205 invalidate_bdev(device->bdev); 2206 } 2207 } 2208 2209 *bdev = device->bdev; 2210 *mode = device->mode; 2211 synchronize_rcu(); 2212 btrfs_free_device(device); 2213 2214 /* 2215 * This can happen if cur_devices is the private seed devices list. We 2216 * cannot call close_fs_devices() here because it expects the uuid_mutex 2217 * to be held, but in fact we don't need that for the private 2218 * seed_devices, we can simply decrement cur_devices->opened and then 2219 * remove it from our list and free the fs_devices. 2220 */ 2221 if (cur_devices->num_devices == 0) { 2222 list_del_init(&cur_devices->seed_list); 2223 ASSERT(cur_devices->opened == 1); 2224 cur_devices->opened--; 2225 free_fs_devices(cur_devices); 2226 } 2227 2228 out: 2229 return ret; 2230 2231 error_undo: 2232 btrfs_reada_undo_remove_dev(device); 2233 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2234 mutex_lock(&fs_info->chunk_mutex); 2235 list_add(&device->dev_alloc_list, 2236 &fs_devices->alloc_list); 2237 device->fs_devices->rw_devices++; 2238 mutex_unlock(&fs_info->chunk_mutex); 2239 } 2240 goto out; 2241 } 2242 2243 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2244 { 2245 struct btrfs_fs_devices *fs_devices; 2246 2247 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2248 2249 /* 2250 * in case of fs with no seed, srcdev->fs_devices will point 2251 * to fs_devices of fs_info. However when the dev being replaced is 2252 * a seed dev it will point to the seed's local fs_devices. In short 2253 * srcdev will have its correct fs_devices in both the cases. 2254 */ 2255 fs_devices = srcdev->fs_devices; 2256 2257 list_del_rcu(&srcdev->dev_list); 2258 list_del(&srcdev->dev_alloc_list); 2259 fs_devices->num_devices--; 2260 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2261 fs_devices->missing_devices--; 2262 2263 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2264 fs_devices->rw_devices--; 2265 2266 if (srcdev->bdev) 2267 fs_devices->open_devices--; 2268 } 2269 2270 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2271 { 2272 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2273 2274 mutex_lock(&uuid_mutex); 2275 2276 btrfs_close_bdev(srcdev); 2277 synchronize_rcu(); 2278 btrfs_free_device(srcdev); 2279 2280 /* if this is no devs we rather delete the fs_devices */ 2281 if (!fs_devices->num_devices) { 2282 /* 2283 * On a mounted FS, num_devices can't be zero unless it's a 2284 * seed. In case of a seed device being replaced, the replace 2285 * target added to the sprout FS, so there will be no more 2286 * device left under the seed FS. 2287 */ 2288 ASSERT(fs_devices->seeding); 2289 2290 list_del_init(&fs_devices->seed_list); 2291 close_fs_devices(fs_devices); 2292 free_fs_devices(fs_devices); 2293 } 2294 mutex_unlock(&uuid_mutex); 2295 } 2296 2297 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2298 { 2299 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2300 2301 mutex_lock(&fs_devices->device_list_mutex); 2302 2303 btrfs_sysfs_remove_device(tgtdev); 2304 2305 if (tgtdev->bdev) 2306 fs_devices->open_devices--; 2307 2308 fs_devices->num_devices--; 2309 2310 btrfs_assign_next_active_device(tgtdev, NULL); 2311 2312 list_del_rcu(&tgtdev->dev_list); 2313 2314 mutex_unlock(&fs_devices->device_list_mutex); 2315 2316 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2317 tgtdev->name->str); 2318 2319 btrfs_close_bdev(tgtdev); 2320 synchronize_rcu(); 2321 btrfs_free_device(tgtdev); 2322 } 2323 2324 static struct btrfs_device *btrfs_find_device_by_path( 2325 struct btrfs_fs_info *fs_info, const char *device_path) 2326 { 2327 int ret = 0; 2328 struct btrfs_super_block *disk_super; 2329 u64 devid; 2330 u8 *dev_uuid; 2331 struct block_device *bdev; 2332 struct btrfs_device *device; 2333 2334 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, 2335 fs_info->bdev_holder, 0, &bdev, &disk_super); 2336 if (ret) 2337 return ERR_PTR(ret); 2338 2339 devid = btrfs_stack_device_id(&disk_super->dev_item); 2340 dev_uuid = disk_super->dev_item.uuid; 2341 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2342 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, 2343 disk_super->metadata_uuid); 2344 else 2345 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, 2346 disk_super->fsid); 2347 2348 btrfs_release_disk_super(disk_super); 2349 if (!device) 2350 device = ERR_PTR(-ENOENT); 2351 blkdev_put(bdev, FMODE_READ); 2352 return device; 2353 } 2354 2355 /* 2356 * Lookup a device given by device id, or the path if the id is 0. 2357 */ 2358 struct btrfs_device *btrfs_find_device_by_devspec( 2359 struct btrfs_fs_info *fs_info, u64 devid, 2360 const char *device_path) 2361 { 2362 struct btrfs_device *device; 2363 2364 if (devid) { 2365 device = btrfs_find_device(fs_info->fs_devices, devid, NULL, 2366 NULL); 2367 if (!device) 2368 return ERR_PTR(-ENOENT); 2369 return device; 2370 } 2371 2372 if (!device_path || !device_path[0]) 2373 return ERR_PTR(-EINVAL); 2374 2375 if (strcmp(device_path, "missing") == 0) { 2376 /* Find first missing device */ 2377 list_for_each_entry(device, &fs_info->fs_devices->devices, 2378 dev_list) { 2379 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 2380 &device->dev_state) && !device->bdev) 2381 return device; 2382 } 2383 return ERR_PTR(-ENOENT); 2384 } 2385 2386 return btrfs_find_device_by_path(fs_info, device_path); 2387 } 2388 2389 /* 2390 * does all the dirty work required for changing file system's UUID. 2391 */ 2392 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) 2393 { 2394 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2395 struct btrfs_fs_devices *old_devices; 2396 struct btrfs_fs_devices *seed_devices; 2397 struct btrfs_super_block *disk_super = fs_info->super_copy; 2398 struct btrfs_device *device; 2399 u64 super_flags; 2400 2401 lockdep_assert_held(&uuid_mutex); 2402 if (!fs_devices->seeding) 2403 return -EINVAL; 2404 2405 /* 2406 * Private copy of the seed devices, anchored at 2407 * fs_info->fs_devices->seed_list 2408 */ 2409 seed_devices = alloc_fs_devices(NULL, NULL); 2410 if (IS_ERR(seed_devices)) 2411 return PTR_ERR(seed_devices); 2412 2413 /* 2414 * It's necessary to retain a copy of the original seed fs_devices in 2415 * fs_uuids so that filesystems which have been seeded can successfully 2416 * reference the seed device from open_seed_devices. This also supports 2417 * multiple fs seed. 2418 */ 2419 old_devices = clone_fs_devices(fs_devices); 2420 if (IS_ERR(old_devices)) { 2421 kfree(seed_devices); 2422 return PTR_ERR(old_devices); 2423 } 2424 2425 list_add(&old_devices->fs_list, &fs_uuids); 2426 2427 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2428 seed_devices->opened = 1; 2429 INIT_LIST_HEAD(&seed_devices->devices); 2430 INIT_LIST_HEAD(&seed_devices->alloc_list); 2431 mutex_init(&seed_devices->device_list_mutex); 2432 2433 mutex_lock(&fs_devices->device_list_mutex); 2434 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2435 synchronize_rcu); 2436 list_for_each_entry(device, &seed_devices->devices, dev_list) 2437 device->fs_devices = seed_devices; 2438 2439 fs_devices->seeding = false; 2440 fs_devices->num_devices = 0; 2441 fs_devices->open_devices = 0; 2442 fs_devices->missing_devices = 0; 2443 fs_devices->rotating = false; 2444 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2445 2446 generate_random_uuid(fs_devices->fsid); 2447 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2448 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2449 mutex_unlock(&fs_devices->device_list_mutex); 2450 2451 super_flags = btrfs_super_flags(disk_super) & 2452 ~BTRFS_SUPER_FLAG_SEEDING; 2453 btrfs_set_super_flags(disk_super, super_flags); 2454 2455 return 0; 2456 } 2457 2458 /* 2459 * Store the expected generation for seed devices in device items. 2460 */ 2461 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2462 { 2463 struct btrfs_fs_info *fs_info = trans->fs_info; 2464 struct btrfs_root *root = fs_info->chunk_root; 2465 struct btrfs_path *path; 2466 struct extent_buffer *leaf; 2467 struct btrfs_dev_item *dev_item; 2468 struct btrfs_device *device; 2469 struct btrfs_key key; 2470 u8 fs_uuid[BTRFS_FSID_SIZE]; 2471 u8 dev_uuid[BTRFS_UUID_SIZE]; 2472 u64 devid; 2473 int ret; 2474 2475 path = btrfs_alloc_path(); 2476 if (!path) 2477 return -ENOMEM; 2478 2479 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2480 key.offset = 0; 2481 key.type = BTRFS_DEV_ITEM_KEY; 2482 2483 while (1) { 2484 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2485 if (ret < 0) 2486 goto error; 2487 2488 leaf = path->nodes[0]; 2489 next_slot: 2490 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2491 ret = btrfs_next_leaf(root, path); 2492 if (ret > 0) 2493 break; 2494 if (ret < 0) 2495 goto error; 2496 leaf = path->nodes[0]; 2497 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2498 btrfs_release_path(path); 2499 continue; 2500 } 2501 2502 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2503 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2504 key.type != BTRFS_DEV_ITEM_KEY) 2505 break; 2506 2507 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2508 struct btrfs_dev_item); 2509 devid = btrfs_device_id(leaf, dev_item); 2510 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2511 BTRFS_UUID_SIZE); 2512 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2513 BTRFS_FSID_SIZE); 2514 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, 2515 fs_uuid); 2516 BUG_ON(!device); /* Logic error */ 2517 2518 if (device->fs_devices->seeding) { 2519 btrfs_set_device_generation(leaf, dev_item, 2520 device->generation); 2521 btrfs_mark_buffer_dirty(leaf); 2522 } 2523 2524 path->slots[0]++; 2525 goto next_slot; 2526 } 2527 ret = 0; 2528 error: 2529 btrfs_free_path(path); 2530 return ret; 2531 } 2532 2533 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2534 { 2535 struct btrfs_root *root = fs_info->dev_root; 2536 struct request_queue *q; 2537 struct btrfs_trans_handle *trans; 2538 struct btrfs_device *device; 2539 struct block_device *bdev; 2540 struct super_block *sb = fs_info->sb; 2541 struct rcu_string *name; 2542 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2543 u64 orig_super_total_bytes; 2544 u64 orig_super_num_devices; 2545 int seeding_dev = 0; 2546 int ret = 0; 2547 bool locked = false; 2548 2549 if (sb_rdonly(sb) && !fs_devices->seeding) 2550 return -EROFS; 2551 2552 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2553 fs_info->bdev_holder); 2554 if (IS_ERR(bdev)) 2555 return PTR_ERR(bdev); 2556 2557 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2558 ret = -EINVAL; 2559 goto error; 2560 } 2561 2562 if (fs_devices->seeding) { 2563 seeding_dev = 1; 2564 down_write(&sb->s_umount); 2565 mutex_lock(&uuid_mutex); 2566 locked = true; 2567 } 2568 2569 sync_blockdev(bdev); 2570 2571 rcu_read_lock(); 2572 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2573 if (device->bdev == bdev) { 2574 ret = -EEXIST; 2575 rcu_read_unlock(); 2576 goto error; 2577 } 2578 } 2579 rcu_read_unlock(); 2580 2581 device = btrfs_alloc_device(fs_info, NULL, NULL); 2582 if (IS_ERR(device)) { 2583 /* we can safely leave the fs_devices entry around */ 2584 ret = PTR_ERR(device); 2585 goto error; 2586 } 2587 2588 name = rcu_string_strdup(device_path, GFP_KERNEL); 2589 if (!name) { 2590 ret = -ENOMEM; 2591 goto error_free_device; 2592 } 2593 rcu_assign_pointer(device->name, name); 2594 2595 device->fs_info = fs_info; 2596 device->bdev = bdev; 2597 2598 ret = btrfs_get_dev_zone_info(device); 2599 if (ret) 2600 goto error_free_device; 2601 2602 trans = btrfs_start_transaction(root, 0); 2603 if (IS_ERR(trans)) { 2604 ret = PTR_ERR(trans); 2605 goto error_free_zone; 2606 } 2607 2608 q = bdev_get_queue(bdev); 2609 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2610 device->generation = trans->transid; 2611 device->io_width = fs_info->sectorsize; 2612 device->io_align = fs_info->sectorsize; 2613 device->sector_size = fs_info->sectorsize; 2614 device->total_bytes = round_down(i_size_read(bdev->bd_inode), 2615 fs_info->sectorsize); 2616 device->disk_total_bytes = device->total_bytes; 2617 device->commit_total_bytes = device->total_bytes; 2618 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2619 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2620 device->mode = FMODE_EXCL; 2621 device->dev_stats_valid = 1; 2622 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2623 2624 if (seeding_dev) { 2625 btrfs_clear_sb_rdonly(sb); 2626 ret = btrfs_prepare_sprout(fs_info); 2627 if (ret) { 2628 btrfs_abort_transaction(trans, ret); 2629 goto error_trans; 2630 } 2631 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2632 device); 2633 } 2634 2635 device->fs_devices = fs_devices; 2636 2637 mutex_lock(&fs_devices->device_list_mutex); 2638 mutex_lock(&fs_info->chunk_mutex); 2639 list_add_rcu(&device->dev_list, &fs_devices->devices); 2640 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2641 fs_devices->num_devices++; 2642 fs_devices->open_devices++; 2643 fs_devices->rw_devices++; 2644 fs_devices->total_devices++; 2645 fs_devices->total_rw_bytes += device->total_bytes; 2646 2647 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2648 2649 if (!blk_queue_nonrot(q)) 2650 fs_devices->rotating = true; 2651 2652 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2653 btrfs_set_super_total_bytes(fs_info->super_copy, 2654 round_down(orig_super_total_bytes + device->total_bytes, 2655 fs_info->sectorsize)); 2656 2657 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2658 btrfs_set_super_num_devices(fs_info->super_copy, 2659 orig_super_num_devices + 1); 2660 2661 /* 2662 * we've got more storage, clear any full flags on the space 2663 * infos 2664 */ 2665 btrfs_clear_space_info_full(fs_info); 2666 2667 mutex_unlock(&fs_info->chunk_mutex); 2668 2669 /* Add sysfs device entry */ 2670 btrfs_sysfs_add_device(device); 2671 2672 mutex_unlock(&fs_devices->device_list_mutex); 2673 2674 if (seeding_dev) { 2675 mutex_lock(&fs_info->chunk_mutex); 2676 ret = init_first_rw_device(trans); 2677 mutex_unlock(&fs_info->chunk_mutex); 2678 if (ret) { 2679 btrfs_abort_transaction(trans, ret); 2680 goto error_sysfs; 2681 } 2682 } 2683 2684 ret = btrfs_add_dev_item(trans, device); 2685 if (ret) { 2686 btrfs_abort_transaction(trans, ret); 2687 goto error_sysfs; 2688 } 2689 2690 if (seeding_dev) { 2691 ret = btrfs_finish_sprout(trans); 2692 if (ret) { 2693 btrfs_abort_transaction(trans, ret); 2694 goto error_sysfs; 2695 } 2696 2697 /* 2698 * fs_devices now represents the newly sprouted filesystem and 2699 * its fsid has been changed by btrfs_prepare_sprout 2700 */ 2701 btrfs_sysfs_update_sprout_fsid(fs_devices); 2702 } 2703 2704 ret = btrfs_commit_transaction(trans); 2705 2706 if (seeding_dev) { 2707 mutex_unlock(&uuid_mutex); 2708 up_write(&sb->s_umount); 2709 locked = false; 2710 2711 if (ret) /* transaction commit */ 2712 return ret; 2713 2714 ret = btrfs_relocate_sys_chunks(fs_info); 2715 if (ret < 0) 2716 btrfs_handle_fs_error(fs_info, ret, 2717 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2718 trans = btrfs_attach_transaction(root); 2719 if (IS_ERR(trans)) { 2720 if (PTR_ERR(trans) == -ENOENT) 2721 return 0; 2722 ret = PTR_ERR(trans); 2723 trans = NULL; 2724 goto error_sysfs; 2725 } 2726 ret = btrfs_commit_transaction(trans); 2727 } 2728 2729 /* 2730 * Now that we have written a new super block to this device, check all 2731 * other fs_devices list if device_path alienates any other scanned 2732 * device. 2733 * We can ignore the return value as it typically returns -EINVAL and 2734 * only succeeds if the device was an alien. 2735 */ 2736 btrfs_forget_devices(device_path); 2737 2738 /* Update ctime/mtime for blkid or udev */ 2739 update_dev_time(bdev); 2740 2741 return ret; 2742 2743 error_sysfs: 2744 btrfs_sysfs_remove_device(device); 2745 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2746 mutex_lock(&fs_info->chunk_mutex); 2747 list_del_rcu(&device->dev_list); 2748 list_del(&device->dev_alloc_list); 2749 fs_info->fs_devices->num_devices--; 2750 fs_info->fs_devices->open_devices--; 2751 fs_info->fs_devices->rw_devices--; 2752 fs_info->fs_devices->total_devices--; 2753 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2754 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2755 btrfs_set_super_total_bytes(fs_info->super_copy, 2756 orig_super_total_bytes); 2757 btrfs_set_super_num_devices(fs_info->super_copy, 2758 orig_super_num_devices); 2759 mutex_unlock(&fs_info->chunk_mutex); 2760 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2761 error_trans: 2762 if (seeding_dev) 2763 btrfs_set_sb_rdonly(sb); 2764 if (trans) 2765 btrfs_end_transaction(trans); 2766 error_free_zone: 2767 btrfs_destroy_dev_zone_info(device); 2768 error_free_device: 2769 btrfs_free_device(device); 2770 error: 2771 blkdev_put(bdev, FMODE_EXCL); 2772 if (locked) { 2773 mutex_unlock(&uuid_mutex); 2774 up_write(&sb->s_umount); 2775 } 2776 return ret; 2777 } 2778 2779 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2780 struct btrfs_device *device) 2781 { 2782 int ret; 2783 struct btrfs_path *path; 2784 struct btrfs_root *root = device->fs_info->chunk_root; 2785 struct btrfs_dev_item *dev_item; 2786 struct extent_buffer *leaf; 2787 struct btrfs_key key; 2788 2789 path = btrfs_alloc_path(); 2790 if (!path) 2791 return -ENOMEM; 2792 2793 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2794 key.type = BTRFS_DEV_ITEM_KEY; 2795 key.offset = device->devid; 2796 2797 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2798 if (ret < 0) 2799 goto out; 2800 2801 if (ret > 0) { 2802 ret = -ENOENT; 2803 goto out; 2804 } 2805 2806 leaf = path->nodes[0]; 2807 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2808 2809 btrfs_set_device_id(leaf, dev_item, device->devid); 2810 btrfs_set_device_type(leaf, dev_item, device->type); 2811 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2812 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2813 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2814 btrfs_set_device_total_bytes(leaf, dev_item, 2815 btrfs_device_get_disk_total_bytes(device)); 2816 btrfs_set_device_bytes_used(leaf, dev_item, 2817 btrfs_device_get_bytes_used(device)); 2818 btrfs_mark_buffer_dirty(leaf); 2819 2820 out: 2821 btrfs_free_path(path); 2822 return ret; 2823 } 2824 2825 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2826 struct btrfs_device *device, u64 new_size) 2827 { 2828 struct btrfs_fs_info *fs_info = device->fs_info; 2829 struct btrfs_super_block *super_copy = fs_info->super_copy; 2830 u64 old_total; 2831 u64 diff; 2832 2833 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2834 return -EACCES; 2835 2836 new_size = round_down(new_size, fs_info->sectorsize); 2837 2838 mutex_lock(&fs_info->chunk_mutex); 2839 old_total = btrfs_super_total_bytes(super_copy); 2840 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2841 2842 if (new_size <= device->total_bytes || 2843 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2844 mutex_unlock(&fs_info->chunk_mutex); 2845 return -EINVAL; 2846 } 2847 2848 btrfs_set_super_total_bytes(super_copy, 2849 round_down(old_total + diff, fs_info->sectorsize)); 2850 device->fs_devices->total_rw_bytes += diff; 2851 2852 btrfs_device_set_total_bytes(device, new_size); 2853 btrfs_device_set_disk_total_bytes(device, new_size); 2854 btrfs_clear_space_info_full(device->fs_info); 2855 if (list_empty(&device->post_commit_list)) 2856 list_add_tail(&device->post_commit_list, 2857 &trans->transaction->dev_update_list); 2858 mutex_unlock(&fs_info->chunk_mutex); 2859 2860 return btrfs_update_device(trans, device); 2861 } 2862 2863 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2864 { 2865 struct btrfs_fs_info *fs_info = trans->fs_info; 2866 struct btrfs_root *root = fs_info->chunk_root; 2867 int ret; 2868 struct btrfs_path *path; 2869 struct btrfs_key key; 2870 2871 path = btrfs_alloc_path(); 2872 if (!path) 2873 return -ENOMEM; 2874 2875 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2876 key.offset = chunk_offset; 2877 key.type = BTRFS_CHUNK_ITEM_KEY; 2878 2879 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2880 if (ret < 0) 2881 goto out; 2882 else if (ret > 0) { /* Logic error or corruption */ 2883 btrfs_handle_fs_error(fs_info, -ENOENT, 2884 "Failed lookup while freeing chunk."); 2885 ret = -ENOENT; 2886 goto out; 2887 } 2888 2889 ret = btrfs_del_item(trans, root, path); 2890 if (ret < 0) 2891 btrfs_handle_fs_error(fs_info, ret, 2892 "Failed to delete chunk item."); 2893 out: 2894 btrfs_free_path(path); 2895 return ret; 2896 } 2897 2898 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2899 { 2900 struct btrfs_super_block *super_copy = fs_info->super_copy; 2901 struct btrfs_disk_key *disk_key; 2902 struct btrfs_chunk *chunk; 2903 u8 *ptr; 2904 int ret = 0; 2905 u32 num_stripes; 2906 u32 array_size; 2907 u32 len = 0; 2908 u32 cur; 2909 struct btrfs_key key; 2910 2911 lockdep_assert_held(&fs_info->chunk_mutex); 2912 array_size = btrfs_super_sys_array_size(super_copy); 2913 2914 ptr = super_copy->sys_chunk_array; 2915 cur = 0; 2916 2917 while (cur < array_size) { 2918 disk_key = (struct btrfs_disk_key *)ptr; 2919 btrfs_disk_key_to_cpu(&key, disk_key); 2920 2921 len = sizeof(*disk_key); 2922 2923 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 2924 chunk = (struct btrfs_chunk *)(ptr + len); 2925 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 2926 len += btrfs_chunk_item_size(num_stripes); 2927 } else { 2928 ret = -EIO; 2929 break; 2930 } 2931 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 2932 key.offset == chunk_offset) { 2933 memmove(ptr, ptr + len, array_size - (cur + len)); 2934 array_size -= len; 2935 btrfs_set_super_sys_array_size(super_copy, array_size); 2936 } else { 2937 ptr += len; 2938 cur += len; 2939 } 2940 } 2941 return ret; 2942 } 2943 2944 /* 2945 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 2946 * @logical: Logical block offset in bytes. 2947 * @length: Length of extent in bytes. 2948 * 2949 * Return: Chunk mapping or ERR_PTR. 2950 */ 2951 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 2952 u64 logical, u64 length) 2953 { 2954 struct extent_map_tree *em_tree; 2955 struct extent_map *em; 2956 2957 em_tree = &fs_info->mapping_tree; 2958 read_lock(&em_tree->lock); 2959 em = lookup_extent_mapping(em_tree, logical, length); 2960 read_unlock(&em_tree->lock); 2961 2962 if (!em) { 2963 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 2964 logical, length); 2965 return ERR_PTR(-EINVAL); 2966 } 2967 2968 if (em->start > logical || em->start + em->len < logical) { 2969 btrfs_crit(fs_info, 2970 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 2971 logical, length, em->start, em->start + em->len); 2972 free_extent_map(em); 2973 return ERR_PTR(-EINVAL); 2974 } 2975 2976 /* callers are responsible for dropping em's ref. */ 2977 return em; 2978 } 2979 2980 static int remove_chunk_item(struct btrfs_trans_handle *trans, 2981 struct map_lookup *map, u64 chunk_offset) 2982 { 2983 int i; 2984 2985 /* 2986 * Removing chunk items and updating the device items in the chunks btree 2987 * requires holding the chunk_mutex. 2988 * See the comment at btrfs_chunk_alloc() for the details. 2989 */ 2990 lockdep_assert_held(&trans->fs_info->chunk_mutex); 2991 2992 for (i = 0; i < map->num_stripes; i++) { 2993 int ret; 2994 2995 ret = btrfs_update_device(trans, map->stripes[i].dev); 2996 if (ret) 2997 return ret; 2998 } 2999 3000 return btrfs_free_chunk(trans, chunk_offset); 3001 } 3002 3003 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3004 { 3005 struct btrfs_fs_info *fs_info = trans->fs_info; 3006 struct extent_map *em; 3007 struct map_lookup *map; 3008 u64 dev_extent_len = 0; 3009 int i, ret = 0; 3010 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3011 3012 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3013 if (IS_ERR(em)) { 3014 /* 3015 * This is a logic error, but we don't want to just rely on the 3016 * user having built with ASSERT enabled, so if ASSERT doesn't 3017 * do anything we still error out. 3018 */ 3019 ASSERT(0); 3020 return PTR_ERR(em); 3021 } 3022 map = em->map_lookup; 3023 3024 /* 3025 * First delete the device extent items from the devices btree. 3026 * We take the device_list_mutex to avoid racing with the finishing phase 3027 * of a device replace operation. See the comment below before acquiring 3028 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3029 * because that can result in a deadlock when deleting the device extent 3030 * items from the devices btree - COWing an extent buffer from the btree 3031 * may result in allocating a new metadata chunk, which would attempt to 3032 * lock again fs_info->chunk_mutex. 3033 */ 3034 mutex_lock(&fs_devices->device_list_mutex); 3035 for (i = 0; i < map->num_stripes; i++) { 3036 struct btrfs_device *device = map->stripes[i].dev; 3037 ret = btrfs_free_dev_extent(trans, device, 3038 map->stripes[i].physical, 3039 &dev_extent_len); 3040 if (ret) { 3041 mutex_unlock(&fs_devices->device_list_mutex); 3042 btrfs_abort_transaction(trans, ret); 3043 goto out; 3044 } 3045 3046 if (device->bytes_used > 0) { 3047 mutex_lock(&fs_info->chunk_mutex); 3048 btrfs_device_set_bytes_used(device, 3049 device->bytes_used - dev_extent_len); 3050 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3051 btrfs_clear_space_info_full(fs_info); 3052 mutex_unlock(&fs_info->chunk_mutex); 3053 } 3054 } 3055 mutex_unlock(&fs_devices->device_list_mutex); 3056 3057 /* 3058 * We acquire fs_info->chunk_mutex for 2 reasons: 3059 * 3060 * 1) Just like with the first phase of the chunk allocation, we must 3061 * reserve system space, do all chunk btree updates and deletions, and 3062 * update the system chunk array in the superblock while holding this 3063 * mutex. This is for similar reasons as explained on the comment at 3064 * the top of btrfs_chunk_alloc(); 3065 * 3066 * 2) Prevent races with the final phase of a device replace operation 3067 * that replaces the device object associated with the map's stripes, 3068 * because the device object's id can change at any time during that 3069 * final phase of the device replace operation 3070 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3071 * replaced device and then see it with an ID of 3072 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3073 * the device item, which does not exists on the chunk btree. 3074 * The finishing phase of device replace acquires both the 3075 * device_list_mutex and the chunk_mutex, in that order, so we are 3076 * safe by just acquiring the chunk_mutex. 3077 */ 3078 trans->removing_chunk = true; 3079 mutex_lock(&fs_info->chunk_mutex); 3080 3081 check_system_chunk(trans, map->type); 3082 3083 ret = remove_chunk_item(trans, map, chunk_offset); 3084 /* 3085 * Normally we should not get -ENOSPC since we reserved space before 3086 * through the call to check_system_chunk(). 3087 * 3088 * Despite our system space_info having enough free space, we may not 3089 * be able to allocate extents from its block groups, because all have 3090 * an incompatible profile, which will force us to allocate a new system 3091 * block group with the right profile, or right after we called 3092 * check_system_space() above, a scrub turned the only system block group 3093 * with enough free space into RO mode. 3094 * This is explained with more detail at do_chunk_alloc(). 3095 * 3096 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3097 */ 3098 if (ret == -ENOSPC) { 3099 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3100 struct btrfs_block_group *sys_bg; 3101 3102 sys_bg = btrfs_create_chunk(trans, sys_flags); 3103 if (IS_ERR(sys_bg)) { 3104 ret = PTR_ERR(sys_bg); 3105 btrfs_abort_transaction(trans, ret); 3106 goto out; 3107 } 3108 3109 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3110 if (ret) { 3111 btrfs_abort_transaction(trans, ret); 3112 goto out; 3113 } 3114 3115 ret = remove_chunk_item(trans, map, chunk_offset); 3116 if (ret) { 3117 btrfs_abort_transaction(trans, ret); 3118 goto out; 3119 } 3120 } else if (ret) { 3121 btrfs_abort_transaction(trans, ret); 3122 goto out; 3123 } 3124 3125 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3126 3127 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3128 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3129 if (ret) { 3130 btrfs_abort_transaction(trans, ret); 3131 goto out; 3132 } 3133 } 3134 3135 mutex_unlock(&fs_info->chunk_mutex); 3136 trans->removing_chunk = false; 3137 3138 /* 3139 * We are done with chunk btree updates and deletions, so release the 3140 * system space we previously reserved (with check_system_chunk()). 3141 */ 3142 btrfs_trans_release_chunk_metadata(trans); 3143 3144 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3145 if (ret) { 3146 btrfs_abort_transaction(trans, ret); 3147 goto out; 3148 } 3149 3150 out: 3151 if (trans->removing_chunk) { 3152 mutex_unlock(&fs_info->chunk_mutex); 3153 trans->removing_chunk = false; 3154 } 3155 /* once for us */ 3156 free_extent_map(em); 3157 return ret; 3158 } 3159 3160 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3161 { 3162 struct btrfs_root *root = fs_info->chunk_root; 3163 struct btrfs_trans_handle *trans; 3164 struct btrfs_block_group *block_group; 3165 u64 length; 3166 int ret; 3167 3168 /* 3169 * Prevent races with automatic removal of unused block groups. 3170 * After we relocate and before we remove the chunk with offset 3171 * chunk_offset, automatic removal of the block group can kick in, 3172 * resulting in a failure when calling btrfs_remove_chunk() below. 3173 * 3174 * Make sure to acquire this mutex before doing a tree search (dev 3175 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3176 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3177 * we release the path used to search the chunk/dev tree and before 3178 * the current task acquires this mutex and calls us. 3179 */ 3180 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3181 3182 /* step one, relocate all the extents inside this chunk */ 3183 btrfs_scrub_pause(fs_info); 3184 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3185 btrfs_scrub_continue(fs_info); 3186 if (ret) 3187 return ret; 3188 3189 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3190 if (!block_group) 3191 return -ENOENT; 3192 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3193 length = block_group->length; 3194 btrfs_put_block_group(block_group); 3195 3196 /* 3197 * On a zoned file system, discard the whole block group, this will 3198 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3199 * resetting the zone fails, don't treat it as a fatal problem from the 3200 * filesystem's point of view. 3201 */ 3202 if (btrfs_is_zoned(fs_info)) { 3203 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3204 if (ret) 3205 btrfs_info(fs_info, 3206 "failed to reset zone %llu after relocation", 3207 chunk_offset); 3208 } 3209 3210 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3211 chunk_offset); 3212 if (IS_ERR(trans)) { 3213 ret = PTR_ERR(trans); 3214 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3215 return ret; 3216 } 3217 3218 /* 3219 * step two, delete the device extents and the 3220 * chunk tree entries 3221 */ 3222 ret = btrfs_remove_chunk(trans, chunk_offset); 3223 btrfs_end_transaction(trans); 3224 return ret; 3225 } 3226 3227 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3228 { 3229 struct btrfs_root *chunk_root = fs_info->chunk_root; 3230 struct btrfs_path *path; 3231 struct extent_buffer *leaf; 3232 struct btrfs_chunk *chunk; 3233 struct btrfs_key key; 3234 struct btrfs_key found_key; 3235 u64 chunk_type; 3236 bool retried = false; 3237 int failed = 0; 3238 int ret; 3239 3240 path = btrfs_alloc_path(); 3241 if (!path) 3242 return -ENOMEM; 3243 3244 again: 3245 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3246 key.offset = (u64)-1; 3247 key.type = BTRFS_CHUNK_ITEM_KEY; 3248 3249 while (1) { 3250 mutex_lock(&fs_info->reclaim_bgs_lock); 3251 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3252 if (ret < 0) { 3253 mutex_unlock(&fs_info->reclaim_bgs_lock); 3254 goto error; 3255 } 3256 BUG_ON(ret == 0); /* Corruption */ 3257 3258 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3259 key.type); 3260 if (ret) 3261 mutex_unlock(&fs_info->reclaim_bgs_lock); 3262 if (ret < 0) 3263 goto error; 3264 if (ret > 0) 3265 break; 3266 3267 leaf = path->nodes[0]; 3268 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3269 3270 chunk = btrfs_item_ptr(leaf, path->slots[0], 3271 struct btrfs_chunk); 3272 chunk_type = btrfs_chunk_type(leaf, chunk); 3273 btrfs_release_path(path); 3274 3275 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3276 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3277 if (ret == -ENOSPC) 3278 failed++; 3279 else 3280 BUG_ON(ret); 3281 } 3282 mutex_unlock(&fs_info->reclaim_bgs_lock); 3283 3284 if (found_key.offset == 0) 3285 break; 3286 key.offset = found_key.offset - 1; 3287 } 3288 ret = 0; 3289 if (failed && !retried) { 3290 failed = 0; 3291 retried = true; 3292 goto again; 3293 } else if (WARN_ON(failed && retried)) { 3294 ret = -ENOSPC; 3295 } 3296 error: 3297 btrfs_free_path(path); 3298 return ret; 3299 } 3300 3301 /* 3302 * return 1 : allocate a data chunk successfully, 3303 * return <0: errors during allocating a data chunk, 3304 * return 0 : no need to allocate a data chunk. 3305 */ 3306 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3307 u64 chunk_offset) 3308 { 3309 struct btrfs_block_group *cache; 3310 u64 bytes_used; 3311 u64 chunk_type; 3312 3313 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3314 ASSERT(cache); 3315 chunk_type = cache->flags; 3316 btrfs_put_block_group(cache); 3317 3318 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3319 return 0; 3320 3321 spin_lock(&fs_info->data_sinfo->lock); 3322 bytes_used = fs_info->data_sinfo->bytes_used; 3323 spin_unlock(&fs_info->data_sinfo->lock); 3324 3325 if (!bytes_used) { 3326 struct btrfs_trans_handle *trans; 3327 int ret; 3328 3329 trans = btrfs_join_transaction(fs_info->tree_root); 3330 if (IS_ERR(trans)) 3331 return PTR_ERR(trans); 3332 3333 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3334 btrfs_end_transaction(trans); 3335 if (ret < 0) 3336 return ret; 3337 return 1; 3338 } 3339 3340 return 0; 3341 } 3342 3343 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3344 struct btrfs_balance_control *bctl) 3345 { 3346 struct btrfs_root *root = fs_info->tree_root; 3347 struct btrfs_trans_handle *trans; 3348 struct btrfs_balance_item *item; 3349 struct btrfs_disk_balance_args disk_bargs; 3350 struct btrfs_path *path; 3351 struct extent_buffer *leaf; 3352 struct btrfs_key key; 3353 int ret, err; 3354 3355 path = btrfs_alloc_path(); 3356 if (!path) 3357 return -ENOMEM; 3358 3359 trans = btrfs_start_transaction(root, 0); 3360 if (IS_ERR(trans)) { 3361 btrfs_free_path(path); 3362 return PTR_ERR(trans); 3363 } 3364 3365 key.objectid = BTRFS_BALANCE_OBJECTID; 3366 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3367 key.offset = 0; 3368 3369 ret = btrfs_insert_empty_item(trans, root, path, &key, 3370 sizeof(*item)); 3371 if (ret) 3372 goto out; 3373 3374 leaf = path->nodes[0]; 3375 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3376 3377 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3378 3379 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3380 btrfs_set_balance_data(leaf, item, &disk_bargs); 3381 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3382 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3383 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3384 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3385 3386 btrfs_set_balance_flags(leaf, item, bctl->flags); 3387 3388 btrfs_mark_buffer_dirty(leaf); 3389 out: 3390 btrfs_free_path(path); 3391 err = btrfs_commit_transaction(trans); 3392 if (err && !ret) 3393 ret = err; 3394 return ret; 3395 } 3396 3397 static int del_balance_item(struct btrfs_fs_info *fs_info) 3398 { 3399 struct btrfs_root *root = fs_info->tree_root; 3400 struct btrfs_trans_handle *trans; 3401 struct btrfs_path *path; 3402 struct btrfs_key key; 3403 int ret, err; 3404 3405 path = btrfs_alloc_path(); 3406 if (!path) 3407 return -ENOMEM; 3408 3409 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3410 if (IS_ERR(trans)) { 3411 btrfs_free_path(path); 3412 return PTR_ERR(trans); 3413 } 3414 3415 key.objectid = BTRFS_BALANCE_OBJECTID; 3416 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3417 key.offset = 0; 3418 3419 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3420 if (ret < 0) 3421 goto out; 3422 if (ret > 0) { 3423 ret = -ENOENT; 3424 goto out; 3425 } 3426 3427 ret = btrfs_del_item(trans, root, path); 3428 out: 3429 btrfs_free_path(path); 3430 err = btrfs_commit_transaction(trans); 3431 if (err && !ret) 3432 ret = err; 3433 return ret; 3434 } 3435 3436 /* 3437 * This is a heuristic used to reduce the number of chunks balanced on 3438 * resume after balance was interrupted. 3439 */ 3440 static void update_balance_args(struct btrfs_balance_control *bctl) 3441 { 3442 /* 3443 * Turn on soft mode for chunk types that were being converted. 3444 */ 3445 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3446 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3447 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3448 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3449 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3450 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3451 3452 /* 3453 * Turn on usage filter if is not already used. The idea is 3454 * that chunks that we have already balanced should be 3455 * reasonably full. Don't do it for chunks that are being 3456 * converted - that will keep us from relocating unconverted 3457 * (albeit full) chunks. 3458 */ 3459 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3460 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3461 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3462 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3463 bctl->data.usage = 90; 3464 } 3465 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3466 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3467 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3468 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3469 bctl->sys.usage = 90; 3470 } 3471 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3472 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3473 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3474 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3475 bctl->meta.usage = 90; 3476 } 3477 } 3478 3479 /* 3480 * Clear the balance status in fs_info and delete the balance item from disk. 3481 */ 3482 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3483 { 3484 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3485 int ret; 3486 3487 BUG_ON(!fs_info->balance_ctl); 3488 3489 spin_lock(&fs_info->balance_lock); 3490 fs_info->balance_ctl = NULL; 3491 spin_unlock(&fs_info->balance_lock); 3492 3493 kfree(bctl); 3494 ret = del_balance_item(fs_info); 3495 if (ret) 3496 btrfs_handle_fs_error(fs_info, ret, NULL); 3497 } 3498 3499 /* 3500 * Balance filters. Return 1 if chunk should be filtered out 3501 * (should not be balanced). 3502 */ 3503 static int chunk_profiles_filter(u64 chunk_type, 3504 struct btrfs_balance_args *bargs) 3505 { 3506 chunk_type = chunk_to_extended(chunk_type) & 3507 BTRFS_EXTENDED_PROFILE_MASK; 3508 3509 if (bargs->profiles & chunk_type) 3510 return 0; 3511 3512 return 1; 3513 } 3514 3515 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3516 struct btrfs_balance_args *bargs) 3517 { 3518 struct btrfs_block_group *cache; 3519 u64 chunk_used; 3520 u64 user_thresh_min; 3521 u64 user_thresh_max; 3522 int ret = 1; 3523 3524 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3525 chunk_used = cache->used; 3526 3527 if (bargs->usage_min == 0) 3528 user_thresh_min = 0; 3529 else 3530 user_thresh_min = div_factor_fine(cache->length, 3531 bargs->usage_min); 3532 3533 if (bargs->usage_max == 0) 3534 user_thresh_max = 1; 3535 else if (bargs->usage_max > 100) 3536 user_thresh_max = cache->length; 3537 else 3538 user_thresh_max = div_factor_fine(cache->length, 3539 bargs->usage_max); 3540 3541 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3542 ret = 0; 3543 3544 btrfs_put_block_group(cache); 3545 return ret; 3546 } 3547 3548 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3549 u64 chunk_offset, struct btrfs_balance_args *bargs) 3550 { 3551 struct btrfs_block_group *cache; 3552 u64 chunk_used, user_thresh; 3553 int ret = 1; 3554 3555 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3556 chunk_used = cache->used; 3557 3558 if (bargs->usage_min == 0) 3559 user_thresh = 1; 3560 else if (bargs->usage > 100) 3561 user_thresh = cache->length; 3562 else 3563 user_thresh = div_factor_fine(cache->length, bargs->usage); 3564 3565 if (chunk_used < user_thresh) 3566 ret = 0; 3567 3568 btrfs_put_block_group(cache); 3569 return ret; 3570 } 3571 3572 static int chunk_devid_filter(struct extent_buffer *leaf, 3573 struct btrfs_chunk *chunk, 3574 struct btrfs_balance_args *bargs) 3575 { 3576 struct btrfs_stripe *stripe; 3577 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3578 int i; 3579 3580 for (i = 0; i < num_stripes; i++) { 3581 stripe = btrfs_stripe_nr(chunk, i); 3582 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3583 return 0; 3584 } 3585 3586 return 1; 3587 } 3588 3589 static u64 calc_data_stripes(u64 type, int num_stripes) 3590 { 3591 const int index = btrfs_bg_flags_to_raid_index(type); 3592 const int ncopies = btrfs_raid_array[index].ncopies; 3593 const int nparity = btrfs_raid_array[index].nparity; 3594 3595 return (num_stripes - nparity) / ncopies; 3596 } 3597 3598 /* [pstart, pend) */ 3599 static int chunk_drange_filter(struct extent_buffer *leaf, 3600 struct btrfs_chunk *chunk, 3601 struct btrfs_balance_args *bargs) 3602 { 3603 struct btrfs_stripe *stripe; 3604 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3605 u64 stripe_offset; 3606 u64 stripe_length; 3607 u64 type; 3608 int factor; 3609 int i; 3610 3611 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3612 return 0; 3613 3614 type = btrfs_chunk_type(leaf, chunk); 3615 factor = calc_data_stripes(type, num_stripes); 3616 3617 for (i = 0; i < num_stripes; i++) { 3618 stripe = btrfs_stripe_nr(chunk, i); 3619 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3620 continue; 3621 3622 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3623 stripe_length = btrfs_chunk_length(leaf, chunk); 3624 stripe_length = div_u64(stripe_length, factor); 3625 3626 if (stripe_offset < bargs->pend && 3627 stripe_offset + stripe_length > bargs->pstart) 3628 return 0; 3629 } 3630 3631 return 1; 3632 } 3633 3634 /* [vstart, vend) */ 3635 static int chunk_vrange_filter(struct extent_buffer *leaf, 3636 struct btrfs_chunk *chunk, 3637 u64 chunk_offset, 3638 struct btrfs_balance_args *bargs) 3639 { 3640 if (chunk_offset < bargs->vend && 3641 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3642 /* at least part of the chunk is inside this vrange */ 3643 return 0; 3644 3645 return 1; 3646 } 3647 3648 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3649 struct btrfs_chunk *chunk, 3650 struct btrfs_balance_args *bargs) 3651 { 3652 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3653 3654 if (bargs->stripes_min <= num_stripes 3655 && num_stripes <= bargs->stripes_max) 3656 return 0; 3657 3658 return 1; 3659 } 3660 3661 static int chunk_soft_convert_filter(u64 chunk_type, 3662 struct btrfs_balance_args *bargs) 3663 { 3664 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3665 return 0; 3666 3667 chunk_type = chunk_to_extended(chunk_type) & 3668 BTRFS_EXTENDED_PROFILE_MASK; 3669 3670 if (bargs->target == chunk_type) 3671 return 1; 3672 3673 return 0; 3674 } 3675 3676 static int should_balance_chunk(struct extent_buffer *leaf, 3677 struct btrfs_chunk *chunk, u64 chunk_offset) 3678 { 3679 struct btrfs_fs_info *fs_info = leaf->fs_info; 3680 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3681 struct btrfs_balance_args *bargs = NULL; 3682 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3683 3684 /* type filter */ 3685 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3686 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3687 return 0; 3688 } 3689 3690 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3691 bargs = &bctl->data; 3692 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3693 bargs = &bctl->sys; 3694 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3695 bargs = &bctl->meta; 3696 3697 /* profiles filter */ 3698 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3699 chunk_profiles_filter(chunk_type, bargs)) { 3700 return 0; 3701 } 3702 3703 /* usage filter */ 3704 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3705 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3706 return 0; 3707 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3708 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3709 return 0; 3710 } 3711 3712 /* devid filter */ 3713 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3714 chunk_devid_filter(leaf, chunk, bargs)) { 3715 return 0; 3716 } 3717 3718 /* drange filter, makes sense only with devid filter */ 3719 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3720 chunk_drange_filter(leaf, chunk, bargs)) { 3721 return 0; 3722 } 3723 3724 /* vrange filter */ 3725 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3726 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3727 return 0; 3728 } 3729 3730 /* stripes filter */ 3731 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3732 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3733 return 0; 3734 } 3735 3736 /* soft profile changing mode */ 3737 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3738 chunk_soft_convert_filter(chunk_type, bargs)) { 3739 return 0; 3740 } 3741 3742 /* 3743 * limited by count, must be the last filter 3744 */ 3745 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3746 if (bargs->limit == 0) 3747 return 0; 3748 else 3749 bargs->limit--; 3750 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3751 /* 3752 * Same logic as the 'limit' filter; the minimum cannot be 3753 * determined here because we do not have the global information 3754 * about the count of all chunks that satisfy the filters. 3755 */ 3756 if (bargs->limit_max == 0) 3757 return 0; 3758 else 3759 bargs->limit_max--; 3760 } 3761 3762 return 1; 3763 } 3764 3765 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3766 { 3767 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3768 struct btrfs_root *chunk_root = fs_info->chunk_root; 3769 u64 chunk_type; 3770 struct btrfs_chunk *chunk; 3771 struct btrfs_path *path = NULL; 3772 struct btrfs_key key; 3773 struct btrfs_key found_key; 3774 struct extent_buffer *leaf; 3775 int slot; 3776 int ret; 3777 int enospc_errors = 0; 3778 bool counting = true; 3779 /* The single value limit and min/max limits use the same bytes in the */ 3780 u64 limit_data = bctl->data.limit; 3781 u64 limit_meta = bctl->meta.limit; 3782 u64 limit_sys = bctl->sys.limit; 3783 u32 count_data = 0; 3784 u32 count_meta = 0; 3785 u32 count_sys = 0; 3786 int chunk_reserved = 0; 3787 3788 path = btrfs_alloc_path(); 3789 if (!path) { 3790 ret = -ENOMEM; 3791 goto error; 3792 } 3793 3794 /* zero out stat counters */ 3795 spin_lock(&fs_info->balance_lock); 3796 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3797 spin_unlock(&fs_info->balance_lock); 3798 again: 3799 if (!counting) { 3800 /* 3801 * The single value limit and min/max limits use the same bytes 3802 * in the 3803 */ 3804 bctl->data.limit = limit_data; 3805 bctl->meta.limit = limit_meta; 3806 bctl->sys.limit = limit_sys; 3807 } 3808 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3809 key.offset = (u64)-1; 3810 key.type = BTRFS_CHUNK_ITEM_KEY; 3811 3812 while (1) { 3813 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3814 atomic_read(&fs_info->balance_cancel_req)) { 3815 ret = -ECANCELED; 3816 goto error; 3817 } 3818 3819 mutex_lock(&fs_info->reclaim_bgs_lock); 3820 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3821 if (ret < 0) { 3822 mutex_unlock(&fs_info->reclaim_bgs_lock); 3823 goto error; 3824 } 3825 3826 /* 3827 * this shouldn't happen, it means the last relocate 3828 * failed 3829 */ 3830 if (ret == 0) 3831 BUG(); /* FIXME break ? */ 3832 3833 ret = btrfs_previous_item(chunk_root, path, 0, 3834 BTRFS_CHUNK_ITEM_KEY); 3835 if (ret) { 3836 mutex_unlock(&fs_info->reclaim_bgs_lock); 3837 ret = 0; 3838 break; 3839 } 3840 3841 leaf = path->nodes[0]; 3842 slot = path->slots[0]; 3843 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3844 3845 if (found_key.objectid != key.objectid) { 3846 mutex_unlock(&fs_info->reclaim_bgs_lock); 3847 break; 3848 } 3849 3850 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3851 chunk_type = btrfs_chunk_type(leaf, chunk); 3852 3853 if (!counting) { 3854 spin_lock(&fs_info->balance_lock); 3855 bctl->stat.considered++; 3856 spin_unlock(&fs_info->balance_lock); 3857 } 3858 3859 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3860 3861 btrfs_release_path(path); 3862 if (!ret) { 3863 mutex_unlock(&fs_info->reclaim_bgs_lock); 3864 goto loop; 3865 } 3866 3867 if (counting) { 3868 mutex_unlock(&fs_info->reclaim_bgs_lock); 3869 spin_lock(&fs_info->balance_lock); 3870 bctl->stat.expected++; 3871 spin_unlock(&fs_info->balance_lock); 3872 3873 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3874 count_data++; 3875 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3876 count_sys++; 3877 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3878 count_meta++; 3879 3880 goto loop; 3881 } 3882 3883 /* 3884 * Apply limit_min filter, no need to check if the LIMITS 3885 * filter is used, limit_min is 0 by default 3886 */ 3887 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3888 count_data < bctl->data.limit_min) 3889 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3890 count_meta < bctl->meta.limit_min) 3891 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3892 count_sys < bctl->sys.limit_min)) { 3893 mutex_unlock(&fs_info->reclaim_bgs_lock); 3894 goto loop; 3895 } 3896 3897 if (!chunk_reserved) { 3898 /* 3899 * We may be relocating the only data chunk we have, 3900 * which could potentially end up with losing data's 3901 * raid profile, so lets allocate an empty one in 3902 * advance. 3903 */ 3904 ret = btrfs_may_alloc_data_chunk(fs_info, 3905 found_key.offset); 3906 if (ret < 0) { 3907 mutex_unlock(&fs_info->reclaim_bgs_lock); 3908 goto error; 3909 } else if (ret == 1) { 3910 chunk_reserved = 1; 3911 } 3912 } 3913 3914 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3915 mutex_unlock(&fs_info->reclaim_bgs_lock); 3916 if (ret == -ENOSPC) { 3917 enospc_errors++; 3918 } else if (ret == -ETXTBSY) { 3919 btrfs_info(fs_info, 3920 "skipping relocation of block group %llu due to active swapfile", 3921 found_key.offset); 3922 ret = 0; 3923 } else if (ret) { 3924 goto error; 3925 } else { 3926 spin_lock(&fs_info->balance_lock); 3927 bctl->stat.completed++; 3928 spin_unlock(&fs_info->balance_lock); 3929 } 3930 loop: 3931 if (found_key.offset == 0) 3932 break; 3933 key.offset = found_key.offset - 1; 3934 } 3935 3936 if (counting) { 3937 btrfs_release_path(path); 3938 counting = false; 3939 goto again; 3940 } 3941 error: 3942 btrfs_free_path(path); 3943 if (enospc_errors) { 3944 btrfs_info(fs_info, "%d enospc errors during balance", 3945 enospc_errors); 3946 if (!ret) 3947 ret = -ENOSPC; 3948 } 3949 3950 return ret; 3951 } 3952 3953 /** 3954 * alloc_profile_is_valid - see if a given profile is valid and reduced 3955 * @flags: profile to validate 3956 * @extended: if true @flags is treated as an extended profile 3957 */ 3958 static int alloc_profile_is_valid(u64 flags, int extended) 3959 { 3960 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 3961 BTRFS_BLOCK_GROUP_PROFILE_MASK); 3962 3963 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 3964 3965 /* 1) check that all other bits are zeroed */ 3966 if (flags & ~mask) 3967 return 0; 3968 3969 /* 2) see if profile is reduced */ 3970 if (flags == 0) 3971 return !extended; /* "0" is valid for usual profiles */ 3972 3973 return has_single_bit_set(flags); 3974 } 3975 3976 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 3977 { 3978 /* cancel requested || normal exit path */ 3979 return atomic_read(&fs_info->balance_cancel_req) || 3980 (atomic_read(&fs_info->balance_pause_req) == 0 && 3981 atomic_read(&fs_info->balance_cancel_req) == 0); 3982 } 3983 3984 /* 3985 * Validate target profile against allowed profiles and return true if it's OK. 3986 * Otherwise print the error message and return false. 3987 */ 3988 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 3989 const struct btrfs_balance_args *bargs, 3990 u64 allowed, const char *type) 3991 { 3992 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3993 return true; 3994 3995 if (fs_info->sectorsize < PAGE_SIZE && 3996 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) { 3997 btrfs_err(fs_info, 3998 "RAID56 is not yet supported for sectorsize %u with page size %lu", 3999 fs_info->sectorsize, PAGE_SIZE); 4000 return false; 4001 } 4002 /* Profile is valid and does not have bits outside of the allowed set */ 4003 if (alloc_profile_is_valid(bargs->target, 1) && 4004 (bargs->target & ~allowed) == 0) 4005 return true; 4006 4007 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4008 type, btrfs_bg_type_to_raid_name(bargs->target)); 4009 return false; 4010 } 4011 4012 /* 4013 * Fill @buf with textual description of balance filter flags @bargs, up to 4014 * @size_buf including the terminating null. The output may be trimmed if it 4015 * does not fit into the provided buffer. 4016 */ 4017 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4018 u32 size_buf) 4019 { 4020 int ret; 4021 u32 size_bp = size_buf; 4022 char *bp = buf; 4023 u64 flags = bargs->flags; 4024 char tmp_buf[128] = {'\0'}; 4025 4026 if (!flags) 4027 return; 4028 4029 #define CHECK_APPEND_NOARG(a) \ 4030 do { \ 4031 ret = snprintf(bp, size_bp, (a)); \ 4032 if (ret < 0 || ret >= size_bp) \ 4033 goto out_overflow; \ 4034 size_bp -= ret; \ 4035 bp += ret; \ 4036 } while (0) 4037 4038 #define CHECK_APPEND_1ARG(a, v1) \ 4039 do { \ 4040 ret = snprintf(bp, size_bp, (a), (v1)); \ 4041 if (ret < 0 || ret >= size_bp) \ 4042 goto out_overflow; \ 4043 size_bp -= ret; \ 4044 bp += ret; \ 4045 } while (0) 4046 4047 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4048 do { \ 4049 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4050 if (ret < 0 || ret >= size_bp) \ 4051 goto out_overflow; \ 4052 size_bp -= ret; \ 4053 bp += ret; \ 4054 } while (0) 4055 4056 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4057 CHECK_APPEND_1ARG("convert=%s,", 4058 btrfs_bg_type_to_raid_name(bargs->target)); 4059 4060 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4061 CHECK_APPEND_NOARG("soft,"); 4062 4063 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4064 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4065 sizeof(tmp_buf)); 4066 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4067 } 4068 4069 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4070 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4071 4072 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4073 CHECK_APPEND_2ARG("usage=%u..%u,", 4074 bargs->usage_min, bargs->usage_max); 4075 4076 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4077 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4078 4079 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4080 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4081 bargs->pstart, bargs->pend); 4082 4083 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4084 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4085 bargs->vstart, bargs->vend); 4086 4087 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4088 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4089 4090 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4091 CHECK_APPEND_2ARG("limit=%u..%u,", 4092 bargs->limit_min, bargs->limit_max); 4093 4094 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4095 CHECK_APPEND_2ARG("stripes=%u..%u,", 4096 bargs->stripes_min, bargs->stripes_max); 4097 4098 #undef CHECK_APPEND_2ARG 4099 #undef CHECK_APPEND_1ARG 4100 #undef CHECK_APPEND_NOARG 4101 4102 out_overflow: 4103 4104 if (size_bp < size_buf) 4105 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4106 else 4107 buf[0] = '\0'; 4108 } 4109 4110 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4111 { 4112 u32 size_buf = 1024; 4113 char tmp_buf[192] = {'\0'}; 4114 char *buf; 4115 char *bp; 4116 u32 size_bp = size_buf; 4117 int ret; 4118 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4119 4120 buf = kzalloc(size_buf, GFP_KERNEL); 4121 if (!buf) 4122 return; 4123 4124 bp = buf; 4125 4126 #define CHECK_APPEND_1ARG(a, v1) \ 4127 do { \ 4128 ret = snprintf(bp, size_bp, (a), (v1)); \ 4129 if (ret < 0 || ret >= size_bp) \ 4130 goto out_overflow; \ 4131 size_bp -= ret; \ 4132 bp += ret; \ 4133 } while (0) 4134 4135 if (bctl->flags & BTRFS_BALANCE_FORCE) 4136 CHECK_APPEND_1ARG("%s", "-f "); 4137 4138 if (bctl->flags & BTRFS_BALANCE_DATA) { 4139 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4140 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4141 } 4142 4143 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4144 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4145 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4146 } 4147 4148 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4149 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4150 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4151 } 4152 4153 #undef CHECK_APPEND_1ARG 4154 4155 out_overflow: 4156 4157 if (size_bp < size_buf) 4158 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4159 btrfs_info(fs_info, "balance: %s %s", 4160 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4161 "resume" : "start", buf); 4162 4163 kfree(buf); 4164 } 4165 4166 /* 4167 * Should be called with balance mutexe held 4168 */ 4169 int btrfs_balance(struct btrfs_fs_info *fs_info, 4170 struct btrfs_balance_control *bctl, 4171 struct btrfs_ioctl_balance_args *bargs) 4172 { 4173 u64 meta_target, data_target; 4174 u64 allowed; 4175 int mixed = 0; 4176 int ret; 4177 u64 num_devices; 4178 unsigned seq; 4179 bool reducing_redundancy; 4180 int i; 4181 4182 if (btrfs_fs_closing(fs_info) || 4183 atomic_read(&fs_info->balance_pause_req) || 4184 btrfs_should_cancel_balance(fs_info)) { 4185 ret = -EINVAL; 4186 goto out; 4187 } 4188 4189 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4190 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4191 mixed = 1; 4192 4193 /* 4194 * In case of mixed groups both data and meta should be picked, 4195 * and identical options should be given for both of them. 4196 */ 4197 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4198 if (mixed && (bctl->flags & allowed)) { 4199 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4200 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4201 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4202 btrfs_err(fs_info, 4203 "balance: mixed groups data and metadata options must be the same"); 4204 ret = -EINVAL; 4205 goto out; 4206 } 4207 } 4208 4209 /* 4210 * rw_devices will not change at the moment, device add/delete/replace 4211 * are exclusive 4212 */ 4213 num_devices = fs_info->fs_devices->rw_devices; 4214 4215 /* 4216 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4217 * special bit for it, to make it easier to distinguish. Thus we need 4218 * to set it manually, or balance would refuse the profile. 4219 */ 4220 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4221 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4222 if (num_devices >= btrfs_raid_array[i].devs_min) 4223 allowed |= btrfs_raid_array[i].bg_flag; 4224 4225 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4226 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4227 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4228 ret = -EINVAL; 4229 goto out; 4230 } 4231 4232 /* 4233 * Allow to reduce metadata or system integrity only if force set for 4234 * profiles with redundancy (copies, parity) 4235 */ 4236 allowed = 0; 4237 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4238 if (btrfs_raid_array[i].ncopies >= 2 || 4239 btrfs_raid_array[i].tolerated_failures >= 1) 4240 allowed |= btrfs_raid_array[i].bg_flag; 4241 } 4242 do { 4243 seq = read_seqbegin(&fs_info->profiles_lock); 4244 4245 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4246 (fs_info->avail_system_alloc_bits & allowed) && 4247 !(bctl->sys.target & allowed)) || 4248 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4249 (fs_info->avail_metadata_alloc_bits & allowed) && 4250 !(bctl->meta.target & allowed))) 4251 reducing_redundancy = true; 4252 else 4253 reducing_redundancy = false; 4254 4255 /* if we're not converting, the target field is uninitialized */ 4256 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4257 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4258 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4259 bctl->data.target : fs_info->avail_data_alloc_bits; 4260 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4261 4262 if (reducing_redundancy) { 4263 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4264 btrfs_info(fs_info, 4265 "balance: force reducing metadata redundancy"); 4266 } else { 4267 btrfs_err(fs_info, 4268 "balance: reduces metadata redundancy, use --force if you want this"); 4269 ret = -EINVAL; 4270 goto out; 4271 } 4272 } 4273 4274 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4275 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4276 btrfs_warn(fs_info, 4277 "balance: metadata profile %s has lower redundancy than data profile %s", 4278 btrfs_bg_type_to_raid_name(meta_target), 4279 btrfs_bg_type_to_raid_name(data_target)); 4280 } 4281 4282 ret = insert_balance_item(fs_info, bctl); 4283 if (ret && ret != -EEXIST) 4284 goto out; 4285 4286 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4287 BUG_ON(ret == -EEXIST); 4288 BUG_ON(fs_info->balance_ctl); 4289 spin_lock(&fs_info->balance_lock); 4290 fs_info->balance_ctl = bctl; 4291 spin_unlock(&fs_info->balance_lock); 4292 } else { 4293 BUG_ON(ret != -EEXIST); 4294 spin_lock(&fs_info->balance_lock); 4295 update_balance_args(bctl); 4296 spin_unlock(&fs_info->balance_lock); 4297 } 4298 4299 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4300 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4301 describe_balance_start_or_resume(fs_info); 4302 mutex_unlock(&fs_info->balance_mutex); 4303 4304 ret = __btrfs_balance(fs_info); 4305 4306 mutex_lock(&fs_info->balance_mutex); 4307 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) 4308 btrfs_info(fs_info, "balance: paused"); 4309 /* 4310 * Balance can be canceled by: 4311 * 4312 * - Regular cancel request 4313 * Then ret == -ECANCELED and balance_cancel_req > 0 4314 * 4315 * - Fatal signal to "btrfs" process 4316 * Either the signal caught by wait_reserve_ticket() and callers 4317 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4318 * got -ECANCELED. 4319 * Either way, in this case balance_cancel_req = 0, and 4320 * ret == -EINTR or ret == -ECANCELED. 4321 * 4322 * So here we only check the return value to catch canceled balance. 4323 */ 4324 else if (ret == -ECANCELED || ret == -EINTR) 4325 btrfs_info(fs_info, "balance: canceled"); 4326 else 4327 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4328 4329 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4330 4331 if (bargs) { 4332 memset(bargs, 0, sizeof(*bargs)); 4333 btrfs_update_ioctl_balance_args(fs_info, bargs); 4334 } 4335 4336 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4337 balance_need_close(fs_info)) { 4338 reset_balance_state(fs_info); 4339 btrfs_exclop_finish(fs_info); 4340 } 4341 4342 wake_up(&fs_info->balance_wait_q); 4343 4344 return ret; 4345 out: 4346 if (bctl->flags & BTRFS_BALANCE_RESUME) 4347 reset_balance_state(fs_info); 4348 else 4349 kfree(bctl); 4350 btrfs_exclop_finish(fs_info); 4351 4352 return ret; 4353 } 4354 4355 static int balance_kthread(void *data) 4356 { 4357 struct btrfs_fs_info *fs_info = data; 4358 int ret = 0; 4359 4360 mutex_lock(&fs_info->balance_mutex); 4361 if (fs_info->balance_ctl) 4362 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4363 mutex_unlock(&fs_info->balance_mutex); 4364 4365 return ret; 4366 } 4367 4368 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4369 { 4370 struct task_struct *tsk; 4371 4372 mutex_lock(&fs_info->balance_mutex); 4373 if (!fs_info->balance_ctl) { 4374 mutex_unlock(&fs_info->balance_mutex); 4375 return 0; 4376 } 4377 mutex_unlock(&fs_info->balance_mutex); 4378 4379 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4380 btrfs_info(fs_info, "balance: resume skipped"); 4381 return 0; 4382 } 4383 4384 /* 4385 * A ro->rw remount sequence should continue with the paused balance 4386 * regardless of who pauses it, system or the user as of now, so set 4387 * the resume flag. 4388 */ 4389 spin_lock(&fs_info->balance_lock); 4390 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4391 spin_unlock(&fs_info->balance_lock); 4392 4393 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4394 return PTR_ERR_OR_ZERO(tsk); 4395 } 4396 4397 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4398 { 4399 struct btrfs_balance_control *bctl; 4400 struct btrfs_balance_item *item; 4401 struct btrfs_disk_balance_args disk_bargs; 4402 struct btrfs_path *path; 4403 struct extent_buffer *leaf; 4404 struct btrfs_key key; 4405 int ret; 4406 4407 path = btrfs_alloc_path(); 4408 if (!path) 4409 return -ENOMEM; 4410 4411 key.objectid = BTRFS_BALANCE_OBJECTID; 4412 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4413 key.offset = 0; 4414 4415 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4416 if (ret < 0) 4417 goto out; 4418 if (ret > 0) { /* ret = -ENOENT; */ 4419 ret = 0; 4420 goto out; 4421 } 4422 4423 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4424 if (!bctl) { 4425 ret = -ENOMEM; 4426 goto out; 4427 } 4428 4429 leaf = path->nodes[0]; 4430 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4431 4432 bctl->flags = btrfs_balance_flags(leaf, item); 4433 bctl->flags |= BTRFS_BALANCE_RESUME; 4434 4435 btrfs_balance_data(leaf, item, &disk_bargs); 4436 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4437 btrfs_balance_meta(leaf, item, &disk_bargs); 4438 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4439 btrfs_balance_sys(leaf, item, &disk_bargs); 4440 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4441 4442 /* 4443 * This should never happen, as the paused balance state is recovered 4444 * during mount without any chance of other exclusive ops to collide. 4445 * 4446 * This gives the exclusive op status to balance and keeps in paused 4447 * state until user intervention (cancel or umount). If the ownership 4448 * cannot be assigned, show a message but do not fail. The balance 4449 * is in a paused state and must have fs_info::balance_ctl properly 4450 * set up. 4451 */ 4452 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) 4453 btrfs_warn(fs_info, 4454 "balance: cannot set exclusive op status, resume manually"); 4455 4456 btrfs_release_path(path); 4457 4458 mutex_lock(&fs_info->balance_mutex); 4459 BUG_ON(fs_info->balance_ctl); 4460 spin_lock(&fs_info->balance_lock); 4461 fs_info->balance_ctl = bctl; 4462 spin_unlock(&fs_info->balance_lock); 4463 mutex_unlock(&fs_info->balance_mutex); 4464 out: 4465 btrfs_free_path(path); 4466 return ret; 4467 } 4468 4469 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4470 { 4471 int ret = 0; 4472 4473 mutex_lock(&fs_info->balance_mutex); 4474 if (!fs_info->balance_ctl) { 4475 mutex_unlock(&fs_info->balance_mutex); 4476 return -ENOTCONN; 4477 } 4478 4479 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4480 atomic_inc(&fs_info->balance_pause_req); 4481 mutex_unlock(&fs_info->balance_mutex); 4482 4483 wait_event(fs_info->balance_wait_q, 4484 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4485 4486 mutex_lock(&fs_info->balance_mutex); 4487 /* we are good with balance_ctl ripped off from under us */ 4488 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4489 atomic_dec(&fs_info->balance_pause_req); 4490 } else { 4491 ret = -ENOTCONN; 4492 } 4493 4494 mutex_unlock(&fs_info->balance_mutex); 4495 return ret; 4496 } 4497 4498 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4499 { 4500 mutex_lock(&fs_info->balance_mutex); 4501 if (!fs_info->balance_ctl) { 4502 mutex_unlock(&fs_info->balance_mutex); 4503 return -ENOTCONN; 4504 } 4505 4506 /* 4507 * A paused balance with the item stored on disk can be resumed at 4508 * mount time if the mount is read-write. Otherwise it's still paused 4509 * and we must not allow cancelling as it deletes the item. 4510 */ 4511 if (sb_rdonly(fs_info->sb)) { 4512 mutex_unlock(&fs_info->balance_mutex); 4513 return -EROFS; 4514 } 4515 4516 atomic_inc(&fs_info->balance_cancel_req); 4517 /* 4518 * if we are running just wait and return, balance item is 4519 * deleted in btrfs_balance in this case 4520 */ 4521 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4522 mutex_unlock(&fs_info->balance_mutex); 4523 wait_event(fs_info->balance_wait_q, 4524 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4525 mutex_lock(&fs_info->balance_mutex); 4526 } else { 4527 mutex_unlock(&fs_info->balance_mutex); 4528 /* 4529 * Lock released to allow other waiters to continue, we'll 4530 * reexamine the status again. 4531 */ 4532 mutex_lock(&fs_info->balance_mutex); 4533 4534 if (fs_info->balance_ctl) { 4535 reset_balance_state(fs_info); 4536 btrfs_exclop_finish(fs_info); 4537 btrfs_info(fs_info, "balance: canceled"); 4538 } 4539 } 4540 4541 BUG_ON(fs_info->balance_ctl || 4542 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4543 atomic_dec(&fs_info->balance_cancel_req); 4544 mutex_unlock(&fs_info->balance_mutex); 4545 return 0; 4546 } 4547 4548 int btrfs_uuid_scan_kthread(void *data) 4549 { 4550 struct btrfs_fs_info *fs_info = data; 4551 struct btrfs_root *root = fs_info->tree_root; 4552 struct btrfs_key key; 4553 struct btrfs_path *path = NULL; 4554 int ret = 0; 4555 struct extent_buffer *eb; 4556 int slot; 4557 struct btrfs_root_item root_item; 4558 u32 item_size; 4559 struct btrfs_trans_handle *trans = NULL; 4560 bool closing = false; 4561 4562 path = btrfs_alloc_path(); 4563 if (!path) { 4564 ret = -ENOMEM; 4565 goto out; 4566 } 4567 4568 key.objectid = 0; 4569 key.type = BTRFS_ROOT_ITEM_KEY; 4570 key.offset = 0; 4571 4572 while (1) { 4573 if (btrfs_fs_closing(fs_info)) { 4574 closing = true; 4575 break; 4576 } 4577 ret = btrfs_search_forward(root, &key, path, 4578 BTRFS_OLDEST_GENERATION); 4579 if (ret) { 4580 if (ret > 0) 4581 ret = 0; 4582 break; 4583 } 4584 4585 if (key.type != BTRFS_ROOT_ITEM_KEY || 4586 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4587 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4588 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4589 goto skip; 4590 4591 eb = path->nodes[0]; 4592 slot = path->slots[0]; 4593 item_size = btrfs_item_size_nr(eb, slot); 4594 if (item_size < sizeof(root_item)) 4595 goto skip; 4596 4597 read_extent_buffer(eb, &root_item, 4598 btrfs_item_ptr_offset(eb, slot), 4599 (int)sizeof(root_item)); 4600 if (btrfs_root_refs(&root_item) == 0) 4601 goto skip; 4602 4603 if (!btrfs_is_empty_uuid(root_item.uuid) || 4604 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4605 if (trans) 4606 goto update_tree; 4607 4608 btrfs_release_path(path); 4609 /* 4610 * 1 - subvol uuid item 4611 * 1 - received_subvol uuid item 4612 */ 4613 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4614 if (IS_ERR(trans)) { 4615 ret = PTR_ERR(trans); 4616 break; 4617 } 4618 continue; 4619 } else { 4620 goto skip; 4621 } 4622 update_tree: 4623 btrfs_release_path(path); 4624 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4625 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4626 BTRFS_UUID_KEY_SUBVOL, 4627 key.objectid); 4628 if (ret < 0) { 4629 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4630 ret); 4631 break; 4632 } 4633 } 4634 4635 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4636 ret = btrfs_uuid_tree_add(trans, 4637 root_item.received_uuid, 4638 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4639 key.objectid); 4640 if (ret < 0) { 4641 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4642 ret); 4643 break; 4644 } 4645 } 4646 4647 skip: 4648 btrfs_release_path(path); 4649 if (trans) { 4650 ret = btrfs_end_transaction(trans); 4651 trans = NULL; 4652 if (ret) 4653 break; 4654 } 4655 4656 if (key.offset < (u64)-1) { 4657 key.offset++; 4658 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4659 key.offset = 0; 4660 key.type = BTRFS_ROOT_ITEM_KEY; 4661 } else if (key.objectid < (u64)-1) { 4662 key.offset = 0; 4663 key.type = BTRFS_ROOT_ITEM_KEY; 4664 key.objectid++; 4665 } else { 4666 break; 4667 } 4668 cond_resched(); 4669 } 4670 4671 out: 4672 btrfs_free_path(path); 4673 if (trans && !IS_ERR(trans)) 4674 btrfs_end_transaction(trans); 4675 if (ret) 4676 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4677 else if (!closing) 4678 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4679 up(&fs_info->uuid_tree_rescan_sem); 4680 return 0; 4681 } 4682 4683 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4684 { 4685 struct btrfs_trans_handle *trans; 4686 struct btrfs_root *tree_root = fs_info->tree_root; 4687 struct btrfs_root *uuid_root; 4688 struct task_struct *task; 4689 int ret; 4690 4691 /* 4692 * 1 - root node 4693 * 1 - root item 4694 */ 4695 trans = btrfs_start_transaction(tree_root, 2); 4696 if (IS_ERR(trans)) 4697 return PTR_ERR(trans); 4698 4699 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4700 if (IS_ERR(uuid_root)) { 4701 ret = PTR_ERR(uuid_root); 4702 btrfs_abort_transaction(trans, ret); 4703 btrfs_end_transaction(trans); 4704 return ret; 4705 } 4706 4707 fs_info->uuid_root = uuid_root; 4708 4709 ret = btrfs_commit_transaction(trans); 4710 if (ret) 4711 return ret; 4712 4713 down(&fs_info->uuid_tree_rescan_sem); 4714 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4715 if (IS_ERR(task)) { 4716 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4717 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4718 up(&fs_info->uuid_tree_rescan_sem); 4719 return PTR_ERR(task); 4720 } 4721 4722 return 0; 4723 } 4724 4725 /* 4726 * shrinking a device means finding all of the device extents past 4727 * the new size, and then following the back refs to the chunks. 4728 * The chunk relocation code actually frees the device extent 4729 */ 4730 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4731 { 4732 struct btrfs_fs_info *fs_info = device->fs_info; 4733 struct btrfs_root *root = fs_info->dev_root; 4734 struct btrfs_trans_handle *trans; 4735 struct btrfs_dev_extent *dev_extent = NULL; 4736 struct btrfs_path *path; 4737 u64 length; 4738 u64 chunk_offset; 4739 int ret; 4740 int slot; 4741 int failed = 0; 4742 bool retried = false; 4743 struct extent_buffer *l; 4744 struct btrfs_key key; 4745 struct btrfs_super_block *super_copy = fs_info->super_copy; 4746 u64 old_total = btrfs_super_total_bytes(super_copy); 4747 u64 old_size = btrfs_device_get_total_bytes(device); 4748 u64 diff; 4749 u64 start; 4750 4751 new_size = round_down(new_size, fs_info->sectorsize); 4752 start = new_size; 4753 diff = round_down(old_size - new_size, fs_info->sectorsize); 4754 4755 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4756 return -EINVAL; 4757 4758 path = btrfs_alloc_path(); 4759 if (!path) 4760 return -ENOMEM; 4761 4762 path->reada = READA_BACK; 4763 4764 trans = btrfs_start_transaction(root, 0); 4765 if (IS_ERR(trans)) { 4766 btrfs_free_path(path); 4767 return PTR_ERR(trans); 4768 } 4769 4770 mutex_lock(&fs_info->chunk_mutex); 4771 4772 btrfs_device_set_total_bytes(device, new_size); 4773 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4774 device->fs_devices->total_rw_bytes -= diff; 4775 atomic64_sub(diff, &fs_info->free_chunk_space); 4776 } 4777 4778 /* 4779 * Once the device's size has been set to the new size, ensure all 4780 * in-memory chunks are synced to disk so that the loop below sees them 4781 * and relocates them accordingly. 4782 */ 4783 if (contains_pending_extent(device, &start, diff)) { 4784 mutex_unlock(&fs_info->chunk_mutex); 4785 ret = btrfs_commit_transaction(trans); 4786 if (ret) 4787 goto done; 4788 } else { 4789 mutex_unlock(&fs_info->chunk_mutex); 4790 btrfs_end_transaction(trans); 4791 } 4792 4793 again: 4794 key.objectid = device->devid; 4795 key.offset = (u64)-1; 4796 key.type = BTRFS_DEV_EXTENT_KEY; 4797 4798 do { 4799 mutex_lock(&fs_info->reclaim_bgs_lock); 4800 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4801 if (ret < 0) { 4802 mutex_unlock(&fs_info->reclaim_bgs_lock); 4803 goto done; 4804 } 4805 4806 ret = btrfs_previous_item(root, path, 0, key.type); 4807 if (ret) { 4808 mutex_unlock(&fs_info->reclaim_bgs_lock); 4809 if (ret < 0) 4810 goto done; 4811 ret = 0; 4812 btrfs_release_path(path); 4813 break; 4814 } 4815 4816 l = path->nodes[0]; 4817 slot = path->slots[0]; 4818 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4819 4820 if (key.objectid != device->devid) { 4821 mutex_unlock(&fs_info->reclaim_bgs_lock); 4822 btrfs_release_path(path); 4823 break; 4824 } 4825 4826 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4827 length = btrfs_dev_extent_length(l, dev_extent); 4828 4829 if (key.offset + length <= new_size) { 4830 mutex_unlock(&fs_info->reclaim_bgs_lock); 4831 btrfs_release_path(path); 4832 break; 4833 } 4834 4835 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4836 btrfs_release_path(path); 4837 4838 /* 4839 * We may be relocating the only data chunk we have, 4840 * which could potentially end up with losing data's 4841 * raid profile, so lets allocate an empty one in 4842 * advance. 4843 */ 4844 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4845 if (ret < 0) { 4846 mutex_unlock(&fs_info->reclaim_bgs_lock); 4847 goto done; 4848 } 4849 4850 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4851 mutex_unlock(&fs_info->reclaim_bgs_lock); 4852 if (ret == -ENOSPC) { 4853 failed++; 4854 } else if (ret) { 4855 if (ret == -ETXTBSY) { 4856 btrfs_warn(fs_info, 4857 "could not shrink block group %llu due to active swapfile", 4858 chunk_offset); 4859 } 4860 goto done; 4861 } 4862 } while (key.offset-- > 0); 4863 4864 if (failed && !retried) { 4865 failed = 0; 4866 retried = true; 4867 goto again; 4868 } else if (failed && retried) { 4869 ret = -ENOSPC; 4870 goto done; 4871 } 4872 4873 /* Shrinking succeeded, else we would be at "done". */ 4874 trans = btrfs_start_transaction(root, 0); 4875 if (IS_ERR(trans)) { 4876 ret = PTR_ERR(trans); 4877 goto done; 4878 } 4879 4880 mutex_lock(&fs_info->chunk_mutex); 4881 /* Clear all state bits beyond the shrunk device size */ 4882 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4883 CHUNK_STATE_MASK); 4884 4885 btrfs_device_set_disk_total_bytes(device, new_size); 4886 if (list_empty(&device->post_commit_list)) 4887 list_add_tail(&device->post_commit_list, 4888 &trans->transaction->dev_update_list); 4889 4890 WARN_ON(diff > old_total); 4891 btrfs_set_super_total_bytes(super_copy, 4892 round_down(old_total - diff, fs_info->sectorsize)); 4893 mutex_unlock(&fs_info->chunk_mutex); 4894 4895 /* Now btrfs_update_device() will change the on-disk size. */ 4896 ret = btrfs_update_device(trans, device); 4897 if (ret < 0) { 4898 btrfs_abort_transaction(trans, ret); 4899 btrfs_end_transaction(trans); 4900 } else { 4901 ret = btrfs_commit_transaction(trans); 4902 } 4903 done: 4904 btrfs_free_path(path); 4905 if (ret) { 4906 mutex_lock(&fs_info->chunk_mutex); 4907 btrfs_device_set_total_bytes(device, old_size); 4908 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4909 device->fs_devices->total_rw_bytes += diff; 4910 atomic64_add(diff, &fs_info->free_chunk_space); 4911 mutex_unlock(&fs_info->chunk_mutex); 4912 } 4913 return ret; 4914 } 4915 4916 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 4917 struct btrfs_key *key, 4918 struct btrfs_chunk *chunk, int item_size) 4919 { 4920 struct btrfs_super_block *super_copy = fs_info->super_copy; 4921 struct btrfs_disk_key disk_key; 4922 u32 array_size; 4923 u8 *ptr; 4924 4925 lockdep_assert_held(&fs_info->chunk_mutex); 4926 4927 array_size = btrfs_super_sys_array_size(super_copy); 4928 if (array_size + item_size + sizeof(disk_key) 4929 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 4930 return -EFBIG; 4931 4932 ptr = super_copy->sys_chunk_array + array_size; 4933 btrfs_cpu_key_to_disk(&disk_key, key); 4934 memcpy(ptr, &disk_key, sizeof(disk_key)); 4935 ptr += sizeof(disk_key); 4936 memcpy(ptr, chunk, item_size); 4937 item_size += sizeof(disk_key); 4938 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 4939 4940 return 0; 4941 } 4942 4943 /* 4944 * sort the devices in descending order by max_avail, total_avail 4945 */ 4946 static int btrfs_cmp_device_info(const void *a, const void *b) 4947 { 4948 const struct btrfs_device_info *di_a = a; 4949 const struct btrfs_device_info *di_b = b; 4950 4951 if (di_a->max_avail > di_b->max_avail) 4952 return -1; 4953 if (di_a->max_avail < di_b->max_avail) 4954 return 1; 4955 if (di_a->total_avail > di_b->total_avail) 4956 return -1; 4957 if (di_a->total_avail < di_b->total_avail) 4958 return 1; 4959 return 0; 4960 } 4961 4962 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 4963 { 4964 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 4965 return; 4966 4967 btrfs_set_fs_incompat(info, RAID56); 4968 } 4969 4970 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 4971 { 4972 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 4973 return; 4974 4975 btrfs_set_fs_incompat(info, RAID1C34); 4976 } 4977 4978 /* 4979 * Structure used internally for btrfs_create_chunk() function. 4980 * Wraps needed parameters. 4981 */ 4982 struct alloc_chunk_ctl { 4983 u64 start; 4984 u64 type; 4985 /* Total number of stripes to allocate */ 4986 int num_stripes; 4987 /* sub_stripes info for map */ 4988 int sub_stripes; 4989 /* Stripes per device */ 4990 int dev_stripes; 4991 /* Maximum number of devices to use */ 4992 int devs_max; 4993 /* Minimum number of devices to use */ 4994 int devs_min; 4995 /* ndevs has to be a multiple of this */ 4996 int devs_increment; 4997 /* Number of copies */ 4998 int ncopies; 4999 /* Number of stripes worth of bytes to store parity information */ 5000 int nparity; 5001 u64 max_stripe_size; 5002 u64 max_chunk_size; 5003 u64 dev_extent_min; 5004 u64 stripe_size; 5005 u64 chunk_size; 5006 int ndevs; 5007 }; 5008 5009 static void init_alloc_chunk_ctl_policy_regular( 5010 struct btrfs_fs_devices *fs_devices, 5011 struct alloc_chunk_ctl *ctl) 5012 { 5013 u64 type = ctl->type; 5014 5015 if (type & BTRFS_BLOCK_GROUP_DATA) { 5016 ctl->max_stripe_size = SZ_1G; 5017 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; 5018 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5019 /* For larger filesystems, use larger metadata chunks */ 5020 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) 5021 ctl->max_stripe_size = SZ_1G; 5022 else 5023 ctl->max_stripe_size = SZ_256M; 5024 ctl->max_chunk_size = ctl->max_stripe_size; 5025 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5026 ctl->max_stripe_size = SZ_32M; 5027 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5028 ctl->devs_max = min_t(int, ctl->devs_max, 5029 BTRFS_MAX_DEVS_SYS_CHUNK); 5030 } else { 5031 BUG(); 5032 } 5033 5034 /* We don't want a chunk larger than 10% of writable space */ 5035 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5036 ctl->max_chunk_size); 5037 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5038 } 5039 5040 static void init_alloc_chunk_ctl_policy_zoned( 5041 struct btrfs_fs_devices *fs_devices, 5042 struct alloc_chunk_ctl *ctl) 5043 { 5044 u64 zone_size = fs_devices->fs_info->zone_size; 5045 u64 limit; 5046 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5047 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5048 u64 min_chunk_size = min_data_stripes * zone_size; 5049 u64 type = ctl->type; 5050 5051 ctl->max_stripe_size = zone_size; 5052 if (type & BTRFS_BLOCK_GROUP_DATA) { 5053 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5054 zone_size); 5055 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5056 ctl->max_chunk_size = ctl->max_stripe_size; 5057 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5058 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5059 ctl->devs_max = min_t(int, ctl->devs_max, 5060 BTRFS_MAX_DEVS_SYS_CHUNK); 5061 } else { 5062 BUG(); 5063 } 5064 5065 /* We don't want a chunk larger than 10% of writable space */ 5066 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5067 zone_size), 5068 min_chunk_size); 5069 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5070 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5071 } 5072 5073 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5074 struct alloc_chunk_ctl *ctl) 5075 { 5076 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5077 5078 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5079 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5080 ctl->devs_max = btrfs_raid_array[index].devs_max; 5081 if (!ctl->devs_max) 5082 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5083 ctl->devs_min = btrfs_raid_array[index].devs_min; 5084 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5085 ctl->ncopies = btrfs_raid_array[index].ncopies; 5086 ctl->nparity = btrfs_raid_array[index].nparity; 5087 ctl->ndevs = 0; 5088 5089 switch (fs_devices->chunk_alloc_policy) { 5090 case BTRFS_CHUNK_ALLOC_REGULAR: 5091 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5092 break; 5093 case BTRFS_CHUNK_ALLOC_ZONED: 5094 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5095 break; 5096 default: 5097 BUG(); 5098 } 5099 } 5100 5101 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5102 struct alloc_chunk_ctl *ctl, 5103 struct btrfs_device_info *devices_info) 5104 { 5105 struct btrfs_fs_info *info = fs_devices->fs_info; 5106 struct btrfs_device *device; 5107 u64 total_avail; 5108 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5109 int ret; 5110 int ndevs = 0; 5111 u64 max_avail; 5112 u64 dev_offset; 5113 5114 /* 5115 * in the first pass through the devices list, we gather information 5116 * about the available holes on each device. 5117 */ 5118 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5119 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5120 WARN(1, KERN_ERR 5121 "BTRFS: read-only device in alloc_list\n"); 5122 continue; 5123 } 5124 5125 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5126 &device->dev_state) || 5127 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5128 continue; 5129 5130 if (device->total_bytes > device->bytes_used) 5131 total_avail = device->total_bytes - device->bytes_used; 5132 else 5133 total_avail = 0; 5134 5135 /* If there is no space on this device, skip it. */ 5136 if (total_avail < ctl->dev_extent_min) 5137 continue; 5138 5139 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5140 &max_avail); 5141 if (ret && ret != -ENOSPC) 5142 return ret; 5143 5144 if (ret == 0) 5145 max_avail = dev_extent_want; 5146 5147 if (max_avail < ctl->dev_extent_min) { 5148 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5149 btrfs_debug(info, 5150 "%s: devid %llu has no free space, have=%llu want=%llu", 5151 __func__, device->devid, max_avail, 5152 ctl->dev_extent_min); 5153 continue; 5154 } 5155 5156 if (ndevs == fs_devices->rw_devices) { 5157 WARN(1, "%s: found more than %llu devices\n", 5158 __func__, fs_devices->rw_devices); 5159 break; 5160 } 5161 devices_info[ndevs].dev_offset = dev_offset; 5162 devices_info[ndevs].max_avail = max_avail; 5163 devices_info[ndevs].total_avail = total_avail; 5164 devices_info[ndevs].dev = device; 5165 ++ndevs; 5166 } 5167 ctl->ndevs = ndevs; 5168 5169 /* 5170 * now sort the devices by hole size / available space 5171 */ 5172 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5173 btrfs_cmp_device_info, NULL); 5174 5175 return 0; 5176 } 5177 5178 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5179 struct btrfs_device_info *devices_info) 5180 { 5181 /* Number of stripes that count for block group size */ 5182 int data_stripes; 5183 5184 /* 5185 * The primary goal is to maximize the number of stripes, so use as 5186 * many devices as possible, even if the stripes are not maximum sized. 5187 * 5188 * The DUP profile stores more than one stripe per device, the 5189 * max_avail is the total size so we have to adjust. 5190 */ 5191 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5192 ctl->dev_stripes); 5193 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5194 5195 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5196 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5197 5198 /* 5199 * Use the number of data stripes to figure out how big this chunk is 5200 * really going to be in terms of logical address space, and compare 5201 * that answer with the max chunk size. If it's higher, we try to 5202 * reduce stripe_size. 5203 */ 5204 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5205 /* 5206 * Reduce stripe_size, round it up to a 16MB boundary again and 5207 * then use it, unless it ends up being even bigger than the 5208 * previous value we had already. 5209 */ 5210 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5211 data_stripes), SZ_16M), 5212 ctl->stripe_size); 5213 } 5214 5215 /* Align to BTRFS_STRIPE_LEN */ 5216 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5217 ctl->chunk_size = ctl->stripe_size * data_stripes; 5218 5219 return 0; 5220 } 5221 5222 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5223 struct btrfs_device_info *devices_info) 5224 { 5225 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5226 /* Number of stripes that count for block group size */ 5227 int data_stripes; 5228 5229 /* 5230 * It should hold because: 5231 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5232 */ 5233 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5234 5235 ctl->stripe_size = zone_size; 5236 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5237 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5238 5239 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5240 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5241 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5242 ctl->stripe_size) + ctl->nparity, 5243 ctl->dev_stripes); 5244 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5245 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5246 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5247 } 5248 5249 ctl->chunk_size = ctl->stripe_size * data_stripes; 5250 5251 return 0; 5252 } 5253 5254 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5255 struct alloc_chunk_ctl *ctl, 5256 struct btrfs_device_info *devices_info) 5257 { 5258 struct btrfs_fs_info *info = fs_devices->fs_info; 5259 5260 /* 5261 * Round down to number of usable stripes, devs_increment can be any 5262 * number so we can't use round_down() that requires power of 2, while 5263 * rounddown is safe. 5264 */ 5265 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5266 5267 if (ctl->ndevs < ctl->devs_min) { 5268 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5269 btrfs_debug(info, 5270 "%s: not enough devices with free space: have=%d minimum required=%d", 5271 __func__, ctl->ndevs, ctl->devs_min); 5272 } 5273 return -ENOSPC; 5274 } 5275 5276 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5277 5278 switch (fs_devices->chunk_alloc_policy) { 5279 case BTRFS_CHUNK_ALLOC_REGULAR: 5280 return decide_stripe_size_regular(ctl, devices_info); 5281 case BTRFS_CHUNK_ALLOC_ZONED: 5282 return decide_stripe_size_zoned(ctl, devices_info); 5283 default: 5284 BUG(); 5285 } 5286 } 5287 5288 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5289 struct alloc_chunk_ctl *ctl, 5290 struct btrfs_device_info *devices_info) 5291 { 5292 struct btrfs_fs_info *info = trans->fs_info; 5293 struct map_lookup *map = NULL; 5294 struct extent_map_tree *em_tree; 5295 struct btrfs_block_group *block_group; 5296 struct extent_map *em; 5297 u64 start = ctl->start; 5298 u64 type = ctl->type; 5299 int ret; 5300 int i; 5301 int j; 5302 5303 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5304 if (!map) 5305 return ERR_PTR(-ENOMEM); 5306 map->num_stripes = ctl->num_stripes; 5307 5308 for (i = 0; i < ctl->ndevs; ++i) { 5309 for (j = 0; j < ctl->dev_stripes; ++j) { 5310 int s = i * ctl->dev_stripes + j; 5311 map->stripes[s].dev = devices_info[i].dev; 5312 map->stripes[s].physical = devices_info[i].dev_offset + 5313 j * ctl->stripe_size; 5314 } 5315 } 5316 map->stripe_len = BTRFS_STRIPE_LEN; 5317 map->io_align = BTRFS_STRIPE_LEN; 5318 map->io_width = BTRFS_STRIPE_LEN; 5319 map->type = type; 5320 map->sub_stripes = ctl->sub_stripes; 5321 5322 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5323 5324 em = alloc_extent_map(); 5325 if (!em) { 5326 kfree(map); 5327 return ERR_PTR(-ENOMEM); 5328 } 5329 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5330 em->map_lookup = map; 5331 em->start = start; 5332 em->len = ctl->chunk_size; 5333 em->block_start = 0; 5334 em->block_len = em->len; 5335 em->orig_block_len = ctl->stripe_size; 5336 5337 em_tree = &info->mapping_tree; 5338 write_lock(&em_tree->lock); 5339 ret = add_extent_mapping(em_tree, em, 0); 5340 if (ret) { 5341 write_unlock(&em_tree->lock); 5342 free_extent_map(em); 5343 return ERR_PTR(ret); 5344 } 5345 write_unlock(&em_tree->lock); 5346 5347 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5348 if (IS_ERR(block_group)) 5349 goto error_del_extent; 5350 5351 for (i = 0; i < map->num_stripes; i++) { 5352 struct btrfs_device *dev = map->stripes[i].dev; 5353 5354 btrfs_device_set_bytes_used(dev, 5355 dev->bytes_used + ctl->stripe_size); 5356 if (list_empty(&dev->post_commit_list)) 5357 list_add_tail(&dev->post_commit_list, 5358 &trans->transaction->dev_update_list); 5359 } 5360 5361 atomic64_sub(ctl->stripe_size * map->num_stripes, 5362 &info->free_chunk_space); 5363 5364 free_extent_map(em); 5365 check_raid56_incompat_flag(info, type); 5366 check_raid1c34_incompat_flag(info, type); 5367 5368 return block_group; 5369 5370 error_del_extent: 5371 write_lock(&em_tree->lock); 5372 remove_extent_mapping(em_tree, em); 5373 write_unlock(&em_tree->lock); 5374 5375 /* One for our allocation */ 5376 free_extent_map(em); 5377 /* One for the tree reference */ 5378 free_extent_map(em); 5379 5380 return block_group; 5381 } 5382 5383 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5384 u64 type) 5385 { 5386 struct btrfs_fs_info *info = trans->fs_info; 5387 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5388 struct btrfs_device_info *devices_info = NULL; 5389 struct alloc_chunk_ctl ctl; 5390 struct btrfs_block_group *block_group; 5391 int ret; 5392 5393 lockdep_assert_held(&info->chunk_mutex); 5394 5395 if (!alloc_profile_is_valid(type, 0)) { 5396 ASSERT(0); 5397 return ERR_PTR(-EINVAL); 5398 } 5399 5400 if (list_empty(&fs_devices->alloc_list)) { 5401 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5402 btrfs_debug(info, "%s: no writable device", __func__); 5403 return ERR_PTR(-ENOSPC); 5404 } 5405 5406 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5407 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5408 ASSERT(0); 5409 return ERR_PTR(-EINVAL); 5410 } 5411 5412 ctl.start = find_next_chunk(info); 5413 ctl.type = type; 5414 init_alloc_chunk_ctl(fs_devices, &ctl); 5415 5416 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5417 GFP_NOFS); 5418 if (!devices_info) 5419 return ERR_PTR(-ENOMEM); 5420 5421 ret = gather_device_info(fs_devices, &ctl, devices_info); 5422 if (ret < 0) { 5423 block_group = ERR_PTR(ret); 5424 goto out; 5425 } 5426 5427 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5428 if (ret < 0) { 5429 block_group = ERR_PTR(ret); 5430 goto out; 5431 } 5432 5433 block_group = create_chunk(trans, &ctl, devices_info); 5434 5435 out: 5436 kfree(devices_info); 5437 return block_group; 5438 } 5439 5440 /* 5441 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5442 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5443 * chunks. 5444 * 5445 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5446 * phases. 5447 */ 5448 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5449 struct btrfs_block_group *bg) 5450 { 5451 struct btrfs_fs_info *fs_info = trans->fs_info; 5452 struct btrfs_root *extent_root = fs_info->extent_root; 5453 struct btrfs_root *chunk_root = fs_info->chunk_root; 5454 struct btrfs_key key; 5455 struct btrfs_chunk *chunk; 5456 struct btrfs_stripe *stripe; 5457 struct extent_map *em; 5458 struct map_lookup *map; 5459 size_t item_size; 5460 int i; 5461 int ret; 5462 5463 /* 5464 * We take the chunk_mutex for 2 reasons: 5465 * 5466 * 1) Updates and insertions in the chunk btree must be done while holding 5467 * the chunk_mutex, as well as updating the system chunk array in the 5468 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5469 * details; 5470 * 5471 * 2) To prevent races with the final phase of a device replace operation 5472 * that replaces the device object associated with the map's stripes, 5473 * because the device object's id can change at any time during that 5474 * final phase of the device replace operation 5475 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5476 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5477 * which would cause a failure when updating the device item, which does 5478 * not exists, or persisting a stripe of the chunk item with such ID. 5479 * Here we can't use the device_list_mutex because our caller already 5480 * has locked the chunk_mutex, and the final phase of device replace 5481 * acquires both mutexes - first the device_list_mutex and then the 5482 * chunk_mutex. Using any of those two mutexes protects us from a 5483 * concurrent device replace. 5484 */ 5485 lockdep_assert_held(&fs_info->chunk_mutex); 5486 5487 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5488 if (IS_ERR(em)) { 5489 ret = PTR_ERR(em); 5490 btrfs_abort_transaction(trans, ret); 5491 return ret; 5492 } 5493 5494 map = em->map_lookup; 5495 item_size = btrfs_chunk_item_size(map->num_stripes); 5496 5497 chunk = kzalloc(item_size, GFP_NOFS); 5498 if (!chunk) { 5499 ret = -ENOMEM; 5500 btrfs_abort_transaction(trans, ret); 5501 goto out; 5502 } 5503 5504 for (i = 0; i < map->num_stripes; i++) { 5505 struct btrfs_device *device = map->stripes[i].dev; 5506 5507 ret = btrfs_update_device(trans, device); 5508 if (ret) 5509 goto out; 5510 } 5511 5512 stripe = &chunk->stripe; 5513 for (i = 0; i < map->num_stripes; i++) { 5514 struct btrfs_device *device = map->stripes[i].dev; 5515 const u64 dev_offset = map->stripes[i].physical; 5516 5517 btrfs_set_stack_stripe_devid(stripe, device->devid); 5518 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5519 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5520 stripe++; 5521 } 5522 5523 btrfs_set_stack_chunk_length(chunk, bg->length); 5524 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); 5525 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5526 btrfs_set_stack_chunk_type(chunk, map->type); 5527 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5528 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5529 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5530 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5531 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5532 5533 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5534 key.type = BTRFS_CHUNK_ITEM_KEY; 5535 key.offset = bg->start; 5536 5537 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5538 if (ret) 5539 goto out; 5540 5541 bg->chunk_item_inserted = 1; 5542 5543 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5544 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5545 if (ret) 5546 goto out; 5547 } 5548 5549 out: 5550 kfree(chunk); 5551 free_extent_map(em); 5552 return ret; 5553 } 5554 5555 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5556 { 5557 struct btrfs_fs_info *fs_info = trans->fs_info; 5558 u64 alloc_profile; 5559 struct btrfs_block_group *meta_bg; 5560 struct btrfs_block_group *sys_bg; 5561 5562 /* 5563 * When adding a new device for sprouting, the seed device is read-only 5564 * so we must first allocate a metadata and a system chunk. But before 5565 * adding the block group items to the extent, device and chunk btrees, 5566 * we must first: 5567 * 5568 * 1) Create both chunks without doing any changes to the btrees, as 5569 * otherwise we would get -ENOSPC since the block groups from the 5570 * seed device are read-only; 5571 * 5572 * 2) Add the device item for the new sprout device - finishing the setup 5573 * of a new block group requires updating the device item in the chunk 5574 * btree, so it must exist when we attempt to do it. The previous step 5575 * ensures this does not fail with -ENOSPC. 5576 * 5577 * After that we can add the block group items to their btrees: 5578 * update existing device item in the chunk btree, add a new block group 5579 * item to the extent btree, add a new chunk item to the chunk btree and 5580 * finally add the new device extent items to the devices btree. 5581 */ 5582 5583 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5584 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5585 if (IS_ERR(meta_bg)) 5586 return PTR_ERR(meta_bg); 5587 5588 alloc_profile = btrfs_system_alloc_profile(fs_info); 5589 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5590 if (IS_ERR(sys_bg)) 5591 return PTR_ERR(sys_bg); 5592 5593 return 0; 5594 } 5595 5596 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5597 { 5598 const int index = btrfs_bg_flags_to_raid_index(map->type); 5599 5600 return btrfs_raid_array[index].tolerated_failures; 5601 } 5602 5603 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5604 { 5605 struct extent_map *em; 5606 struct map_lookup *map; 5607 int miss_ndevs = 0; 5608 int i; 5609 bool ret = true; 5610 5611 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5612 if (IS_ERR(em)) 5613 return false; 5614 5615 map = em->map_lookup; 5616 for (i = 0; i < map->num_stripes; i++) { 5617 if (test_bit(BTRFS_DEV_STATE_MISSING, 5618 &map->stripes[i].dev->dev_state)) { 5619 miss_ndevs++; 5620 continue; 5621 } 5622 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5623 &map->stripes[i].dev->dev_state)) { 5624 ret = false; 5625 goto end; 5626 } 5627 } 5628 5629 /* 5630 * If the number of missing devices is larger than max errors, we can 5631 * not write the data into that chunk successfully. 5632 */ 5633 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5634 ret = false; 5635 end: 5636 free_extent_map(em); 5637 return ret; 5638 } 5639 5640 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5641 { 5642 struct extent_map *em; 5643 5644 while (1) { 5645 write_lock(&tree->lock); 5646 em = lookup_extent_mapping(tree, 0, (u64)-1); 5647 if (em) 5648 remove_extent_mapping(tree, em); 5649 write_unlock(&tree->lock); 5650 if (!em) 5651 break; 5652 /* once for us */ 5653 free_extent_map(em); 5654 /* once for the tree */ 5655 free_extent_map(em); 5656 } 5657 } 5658 5659 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5660 { 5661 struct extent_map *em; 5662 struct map_lookup *map; 5663 int ret; 5664 5665 em = btrfs_get_chunk_map(fs_info, logical, len); 5666 if (IS_ERR(em)) 5667 /* 5668 * We could return errors for these cases, but that could get 5669 * ugly and we'd probably do the same thing which is just not do 5670 * anything else and exit, so return 1 so the callers don't try 5671 * to use other copies. 5672 */ 5673 return 1; 5674 5675 map = em->map_lookup; 5676 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) 5677 ret = map->num_stripes; 5678 else if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5679 ret = map->sub_stripes; 5680 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5681 ret = 2; 5682 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5683 /* 5684 * There could be two corrupted data stripes, we need 5685 * to loop retry in order to rebuild the correct data. 5686 * 5687 * Fail a stripe at a time on every retry except the 5688 * stripe under reconstruction. 5689 */ 5690 ret = map->num_stripes; 5691 else 5692 ret = 1; 5693 free_extent_map(em); 5694 5695 down_read(&fs_info->dev_replace.rwsem); 5696 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5697 fs_info->dev_replace.tgtdev) 5698 ret++; 5699 up_read(&fs_info->dev_replace.rwsem); 5700 5701 return ret; 5702 } 5703 5704 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5705 u64 logical) 5706 { 5707 struct extent_map *em; 5708 struct map_lookup *map; 5709 unsigned long len = fs_info->sectorsize; 5710 5711 em = btrfs_get_chunk_map(fs_info, logical, len); 5712 5713 if (!WARN_ON(IS_ERR(em))) { 5714 map = em->map_lookup; 5715 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5716 len = map->stripe_len * nr_data_stripes(map); 5717 free_extent_map(em); 5718 } 5719 return len; 5720 } 5721 5722 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5723 { 5724 struct extent_map *em; 5725 struct map_lookup *map; 5726 int ret = 0; 5727 5728 em = btrfs_get_chunk_map(fs_info, logical, len); 5729 5730 if(!WARN_ON(IS_ERR(em))) { 5731 map = em->map_lookup; 5732 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5733 ret = 1; 5734 free_extent_map(em); 5735 } 5736 return ret; 5737 } 5738 5739 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5740 struct map_lookup *map, int first, 5741 int dev_replace_is_ongoing) 5742 { 5743 int i; 5744 int num_stripes; 5745 int preferred_mirror; 5746 int tolerance; 5747 struct btrfs_device *srcdev; 5748 5749 ASSERT((map->type & 5750 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5751 5752 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5753 num_stripes = map->sub_stripes; 5754 else 5755 num_stripes = map->num_stripes; 5756 5757 switch (fs_info->fs_devices->read_policy) { 5758 default: 5759 /* Shouldn't happen, just warn and use pid instead of failing */ 5760 btrfs_warn_rl(fs_info, 5761 "unknown read_policy type %u, reset to pid", 5762 fs_info->fs_devices->read_policy); 5763 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5764 fallthrough; 5765 case BTRFS_READ_POLICY_PID: 5766 preferred_mirror = first + (current->pid % num_stripes); 5767 break; 5768 } 5769 5770 if (dev_replace_is_ongoing && 5771 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5772 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5773 srcdev = fs_info->dev_replace.srcdev; 5774 else 5775 srcdev = NULL; 5776 5777 /* 5778 * try to avoid the drive that is the source drive for a 5779 * dev-replace procedure, only choose it if no other non-missing 5780 * mirror is available 5781 */ 5782 for (tolerance = 0; tolerance < 2; tolerance++) { 5783 if (map->stripes[preferred_mirror].dev->bdev && 5784 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5785 return preferred_mirror; 5786 for (i = first; i < first + num_stripes; i++) { 5787 if (map->stripes[i].dev->bdev && 5788 (tolerance || map->stripes[i].dev != srcdev)) 5789 return i; 5790 } 5791 } 5792 5793 /* we couldn't find one that doesn't fail. Just return something 5794 * and the io error handling code will clean up eventually 5795 */ 5796 return preferred_mirror; 5797 } 5798 5799 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5800 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5801 { 5802 int i; 5803 int again = 1; 5804 5805 while (again) { 5806 again = 0; 5807 for (i = 0; i < num_stripes - 1; i++) { 5808 /* Swap if parity is on a smaller index */ 5809 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5810 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5811 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5812 again = 1; 5813 } 5814 } 5815 } 5816 } 5817 5818 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5819 int total_stripes, 5820 int real_stripes) 5821 { 5822 struct btrfs_io_context *bioc = kzalloc( 5823 /* The size of btrfs_io_context */ 5824 sizeof(struct btrfs_io_context) + 5825 /* Plus the variable array for the stripes */ 5826 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5827 /* Plus the variable array for the tgt dev */ 5828 sizeof(int) * (real_stripes) + 5829 /* 5830 * Plus the raid_map, which includes both the tgt dev 5831 * and the stripes. 5832 */ 5833 sizeof(u64) * (total_stripes), 5834 GFP_NOFS|__GFP_NOFAIL); 5835 5836 atomic_set(&bioc->error, 0); 5837 refcount_set(&bioc->refs, 1); 5838 5839 bioc->fs_info = fs_info; 5840 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5841 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5842 5843 return bioc; 5844 } 5845 5846 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5847 { 5848 WARN_ON(!refcount_read(&bioc->refs)); 5849 refcount_inc(&bioc->refs); 5850 } 5851 5852 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5853 { 5854 if (!bioc) 5855 return; 5856 if (refcount_dec_and_test(&bioc->refs)) 5857 kfree(bioc); 5858 } 5859 5860 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ 5861 /* 5862 * Please note that, discard won't be sent to target device of device 5863 * replace. 5864 */ 5865 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, 5866 u64 logical, u64 *length_ret, 5867 struct btrfs_io_context **bioc_ret) 5868 { 5869 struct extent_map *em; 5870 struct map_lookup *map; 5871 struct btrfs_io_context *bioc; 5872 u64 length = *length_ret; 5873 u64 offset; 5874 u64 stripe_nr; 5875 u64 stripe_nr_end; 5876 u64 stripe_end_offset; 5877 u64 stripe_cnt; 5878 u64 stripe_len; 5879 u64 stripe_offset; 5880 u64 num_stripes; 5881 u32 stripe_index; 5882 u32 factor = 0; 5883 u32 sub_stripes = 0; 5884 u64 stripes_per_dev = 0; 5885 u32 remaining_stripes = 0; 5886 u32 last_stripe = 0; 5887 int ret = 0; 5888 int i; 5889 5890 /* Discard always returns a bioc. */ 5891 ASSERT(bioc_ret); 5892 5893 em = btrfs_get_chunk_map(fs_info, logical, length); 5894 if (IS_ERR(em)) 5895 return PTR_ERR(em); 5896 5897 map = em->map_lookup; 5898 /* we don't discard raid56 yet */ 5899 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5900 ret = -EOPNOTSUPP; 5901 goto out; 5902 } 5903 5904 offset = logical - em->start; 5905 length = min_t(u64, em->start + em->len - logical, length); 5906 *length_ret = length; 5907 5908 stripe_len = map->stripe_len; 5909 /* 5910 * stripe_nr counts the total number of stripes we have to stride 5911 * to get to this block 5912 */ 5913 stripe_nr = div64_u64(offset, stripe_len); 5914 5915 /* stripe_offset is the offset of this block in its stripe */ 5916 stripe_offset = offset - stripe_nr * stripe_len; 5917 5918 stripe_nr_end = round_up(offset + length, map->stripe_len); 5919 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 5920 stripe_cnt = stripe_nr_end - stripe_nr; 5921 stripe_end_offset = stripe_nr_end * map->stripe_len - 5922 (offset + length); 5923 /* 5924 * after this, stripe_nr is the number of stripes on this 5925 * device we have to walk to find the data, and stripe_index is 5926 * the number of our device in the stripe array 5927 */ 5928 num_stripes = 1; 5929 stripe_index = 0; 5930 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5931 BTRFS_BLOCK_GROUP_RAID10)) { 5932 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 5933 sub_stripes = 1; 5934 else 5935 sub_stripes = map->sub_stripes; 5936 5937 factor = map->num_stripes / sub_stripes; 5938 num_stripes = min_t(u64, map->num_stripes, 5939 sub_stripes * stripe_cnt); 5940 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 5941 stripe_index *= sub_stripes; 5942 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 5943 &remaining_stripes); 5944 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 5945 last_stripe *= sub_stripes; 5946 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 5947 BTRFS_BLOCK_GROUP_DUP)) { 5948 num_stripes = map->num_stripes; 5949 } else { 5950 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 5951 &stripe_index); 5952 } 5953 5954 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0); 5955 if (!bioc) { 5956 ret = -ENOMEM; 5957 goto out; 5958 } 5959 5960 for (i = 0; i < num_stripes; i++) { 5961 bioc->stripes[i].physical = 5962 map->stripes[stripe_index].physical + 5963 stripe_offset + stripe_nr * map->stripe_len; 5964 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 5965 5966 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5967 BTRFS_BLOCK_GROUP_RAID10)) { 5968 bioc->stripes[i].length = stripes_per_dev * 5969 map->stripe_len; 5970 5971 if (i / sub_stripes < remaining_stripes) 5972 bioc->stripes[i].length += map->stripe_len; 5973 5974 /* 5975 * Special for the first stripe and 5976 * the last stripe: 5977 * 5978 * |-------|...|-------| 5979 * |----------| 5980 * off end_off 5981 */ 5982 if (i < sub_stripes) 5983 bioc->stripes[i].length -= stripe_offset; 5984 5985 if (stripe_index >= last_stripe && 5986 stripe_index <= (last_stripe + 5987 sub_stripes - 1)) 5988 bioc->stripes[i].length -= stripe_end_offset; 5989 5990 if (i == sub_stripes - 1) 5991 stripe_offset = 0; 5992 } else { 5993 bioc->stripes[i].length = length; 5994 } 5995 5996 stripe_index++; 5997 if (stripe_index == map->num_stripes) { 5998 stripe_index = 0; 5999 stripe_nr++; 6000 } 6001 } 6002 6003 *bioc_ret = bioc; 6004 bioc->map_type = map->type; 6005 bioc->num_stripes = num_stripes; 6006 out: 6007 free_extent_map(em); 6008 return ret; 6009 } 6010 6011 /* 6012 * In dev-replace case, for repair case (that's the only case where the mirror 6013 * is selected explicitly when calling btrfs_map_block), blocks left of the 6014 * left cursor can also be read from the target drive. 6015 * 6016 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6017 * array of stripes. 6018 * For READ, it also needs to be supported using the same mirror number. 6019 * 6020 * If the requested block is not left of the left cursor, EIO is returned. This 6021 * can happen because btrfs_num_copies() returns one more in the dev-replace 6022 * case. 6023 */ 6024 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6025 u64 logical, u64 length, 6026 u64 srcdev_devid, int *mirror_num, 6027 u64 *physical) 6028 { 6029 struct btrfs_io_context *bioc = NULL; 6030 int num_stripes; 6031 int index_srcdev = 0; 6032 int found = 0; 6033 u64 physical_of_found = 0; 6034 int i; 6035 int ret = 0; 6036 6037 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6038 logical, &length, &bioc, 0, 0); 6039 if (ret) { 6040 ASSERT(bioc == NULL); 6041 return ret; 6042 } 6043 6044 num_stripes = bioc->num_stripes; 6045 if (*mirror_num > num_stripes) { 6046 /* 6047 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6048 * that means that the requested area is not left of the left 6049 * cursor 6050 */ 6051 btrfs_put_bioc(bioc); 6052 return -EIO; 6053 } 6054 6055 /* 6056 * process the rest of the function using the mirror_num of the source 6057 * drive. Therefore look it up first. At the end, patch the device 6058 * pointer to the one of the target drive. 6059 */ 6060 for (i = 0; i < num_stripes; i++) { 6061 if (bioc->stripes[i].dev->devid != srcdev_devid) 6062 continue; 6063 6064 /* 6065 * In case of DUP, in order to keep it simple, only add the 6066 * mirror with the lowest physical address 6067 */ 6068 if (found && 6069 physical_of_found <= bioc->stripes[i].physical) 6070 continue; 6071 6072 index_srcdev = i; 6073 found = 1; 6074 physical_of_found = bioc->stripes[i].physical; 6075 } 6076 6077 btrfs_put_bioc(bioc); 6078 6079 ASSERT(found); 6080 if (!found) 6081 return -EIO; 6082 6083 *mirror_num = index_srcdev + 1; 6084 *physical = physical_of_found; 6085 return ret; 6086 } 6087 6088 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6089 { 6090 struct btrfs_block_group *cache; 6091 bool ret; 6092 6093 /* Non zoned filesystem does not use "to_copy" flag */ 6094 if (!btrfs_is_zoned(fs_info)) 6095 return false; 6096 6097 cache = btrfs_lookup_block_group(fs_info, logical); 6098 6099 spin_lock(&cache->lock); 6100 ret = cache->to_copy; 6101 spin_unlock(&cache->lock); 6102 6103 btrfs_put_block_group(cache); 6104 return ret; 6105 } 6106 6107 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6108 struct btrfs_io_context **bioc_ret, 6109 struct btrfs_dev_replace *dev_replace, 6110 u64 logical, 6111 int *num_stripes_ret, int *max_errors_ret) 6112 { 6113 struct btrfs_io_context *bioc = *bioc_ret; 6114 u64 srcdev_devid = dev_replace->srcdev->devid; 6115 int tgtdev_indexes = 0; 6116 int num_stripes = *num_stripes_ret; 6117 int max_errors = *max_errors_ret; 6118 int i; 6119 6120 if (op == BTRFS_MAP_WRITE) { 6121 int index_where_to_add; 6122 6123 /* 6124 * A block group which have "to_copy" set will eventually 6125 * copied by dev-replace process. We can avoid cloning IO here. 6126 */ 6127 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6128 return; 6129 6130 /* 6131 * duplicate the write operations while the dev replace 6132 * procedure is running. Since the copying of the old disk to 6133 * the new disk takes place at run time while the filesystem is 6134 * mounted writable, the regular write operations to the old 6135 * disk have to be duplicated to go to the new disk as well. 6136 * 6137 * Note that device->missing is handled by the caller, and that 6138 * the write to the old disk is already set up in the stripes 6139 * array. 6140 */ 6141 index_where_to_add = num_stripes; 6142 for (i = 0; i < num_stripes; i++) { 6143 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6144 /* write to new disk, too */ 6145 struct btrfs_io_stripe *new = 6146 bioc->stripes + index_where_to_add; 6147 struct btrfs_io_stripe *old = 6148 bioc->stripes + i; 6149 6150 new->physical = old->physical; 6151 new->length = old->length; 6152 new->dev = dev_replace->tgtdev; 6153 bioc->tgtdev_map[i] = index_where_to_add; 6154 index_where_to_add++; 6155 max_errors++; 6156 tgtdev_indexes++; 6157 } 6158 } 6159 num_stripes = index_where_to_add; 6160 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6161 int index_srcdev = 0; 6162 int found = 0; 6163 u64 physical_of_found = 0; 6164 6165 /* 6166 * During the dev-replace procedure, the target drive can also 6167 * be used to read data in case it is needed to repair a corrupt 6168 * block elsewhere. This is possible if the requested area is 6169 * left of the left cursor. In this area, the target drive is a 6170 * full copy of the source drive. 6171 */ 6172 for (i = 0; i < num_stripes; i++) { 6173 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6174 /* 6175 * In case of DUP, in order to keep it simple, 6176 * only add the mirror with the lowest physical 6177 * address 6178 */ 6179 if (found && 6180 physical_of_found <= bioc->stripes[i].physical) 6181 continue; 6182 index_srcdev = i; 6183 found = 1; 6184 physical_of_found = bioc->stripes[i].physical; 6185 } 6186 } 6187 if (found) { 6188 struct btrfs_io_stripe *tgtdev_stripe = 6189 bioc->stripes + num_stripes; 6190 6191 tgtdev_stripe->physical = physical_of_found; 6192 tgtdev_stripe->length = 6193 bioc->stripes[index_srcdev].length; 6194 tgtdev_stripe->dev = dev_replace->tgtdev; 6195 bioc->tgtdev_map[index_srcdev] = num_stripes; 6196 6197 tgtdev_indexes++; 6198 num_stripes++; 6199 } 6200 } 6201 6202 *num_stripes_ret = num_stripes; 6203 *max_errors_ret = max_errors; 6204 bioc->num_tgtdevs = tgtdev_indexes; 6205 *bioc_ret = bioc; 6206 } 6207 6208 static bool need_full_stripe(enum btrfs_map_op op) 6209 { 6210 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6211 } 6212 6213 /* 6214 * Calculate the geometry of a particular (address, len) tuple. This 6215 * information is used to calculate how big a particular bio can get before it 6216 * straddles a stripe. 6217 * 6218 * @fs_info: the filesystem 6219 * @em: mapping containing the logical extent 6220 * @op: type of operation - write or read 6221 * @logical: address that we want to figure out the geometry of 6222 * @io_geom: pointer used to return values 6223 * 6224 * Returns < 0 in case a chunk for the given logical address cannot be found, 6225 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6226 */ 6227 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6228 enum btrfs_map_op op, u64 logical, 6229 struct btrfs_io_geometry *io_geom) 6230 { 6231 struct map_lookup *map; 6232 u64 len; 6233 u64 offset; 6234 u64 stripe_offset; 6235 u64 stripe_nr; 6236 u64 stripe_len; 6237 u64 raid56_full_stripe_start = (u64)-1; 6238 int data_stripes; 6239 6240 ASSERT(op != BTRFS_MAP_DISCARD); 6241 6242 map = em->map_lookup; 6243 /* Offset of this logical address in the chunk */ 6244 offset = logical - em->start; 6245 /* Len of a stripe in a chunk */ 6246 stripe_len = map->stripe_len; 6247 /* Stripe where this block falls in */ 6248 stripe_nr = div64_u64(offset, stripe_len); 6249 /* Offset of stripe in the chunk */ 6250 stripe_offset = stripe_nr * stripe_len; 6251 if (offset < stripe_offset) { 6252 btrfs_crit(fs_info, 6253 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", 6254 stripe_offset, offset, em->start, logical, stripe_len); 6255 return -EINVAL; 6256 } 6257 6258 /* stripe_offset is the offset of this block in its stripe */ 6259 stripe_offset = offset - stripe_offset; 6260 data_stripes = nr_data_stripes(map); 6261 6262 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { 6263 u64 max_len = stripe_len - stripe_offset; 6264 6265 /* 6266 * In case of raid56, we need to know the stripe aligned start 6267 */ 6268 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6269 unsigned long full_stripe_len = stripe_len * data_stripes; 6270 raid56_full_stripe_start = offset; 6271 6272 /* 6273 * Allow a write of a full stripe, but make sure we 6274 * don't allow straddling of stripes 6275 */ 6276 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6277 full_stripe_len); 6278 raid56_full_stripe_start *= full_stripe_len; 6279 6280 /* 6281 * For writes to RAID[56], allow a full stripeset across 6282 * all disks. For other RAID types and for RAID[56] 6283 * reads, just allow a single stripe (on a single disk). 6284 */ 6285 if (op == BTRFS_MAP_WRITE) { 6286 max_len = stripe_len * data_stripes - 6287 (offset - raid56_full_stripe_start); 6288 } 6289 } 6290 len = min_t(u64, em->len - offset, max_len); 6291 } else { 6292 len = em->len - offset; 6293 } 6294 6295 io_geom->len = len; 6296 io_geom->offset = offset; 6297 io_geom->stripe_len = stripe_len; 6298 io_geom->stripe_nr = stripe_nr; 6299 io_geom->stripe_offset = stripe_offset; 6300 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6301 6302 return 0; 6303 } 6304 6305 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6306 enum btrfs_map_op op, 6307 u64 logical, u64 *length, 6308 struct btrfs_io_context **bioc_ret, 6309 int mirror_num, int need_raid_map) 6310 { 6311 struct extent_map *em; 6312 struct map_lookup *map; 6313 u64 stripe_offset; 6314 u64 stripe_nr; 6315 u64 stripe_len; 6316 u32 stripe_index; 6317 int data_stripes; 6318 int i; 6319 int ret = 0; 6320 int num_stripes; 6321 int max_errors = 0; 6322 int tgtdev_indexes = 0; 6323 struct btrfs_io_context *bioc = NULL; 6324 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6325 int dev_replace_is_ongoing = 0; 6326 int num_alloc_stripes; 6327 int patch_the_first_stripe_for_dev_replace = 0; 6328 u64 physical_to_patch_in_first_stripe = 0; 6329 u64 raid56_full_stripe_start = (u64)-1; 6330 struct btrfs_io_geometry geom; 6331 6332 ASSERT(bioc_ret); 6333 ASSERT(op != BTRFS_MAP_DISCARD); 6334 6335 em = btrfs_get_chunk_map(fs_info, logical, *length); 6336 ASSERT(!IS_ERR(em)); 6337 6338 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6339 if (ret < 0) 6340 return ret; 6341 6342 map = em->map_lookup; 6343 6344 *length = geom.len; 6345 stripe_len = geom.stripe_len; 6346 stripe_nr = geom.stripe_nr; 6347 stripe_offset = geom.stripe_offset; 6348 raid56_full_stripe_start = geom.raid56_stripe_offset; 6349 data_stripes = nr_data_stripes(map); 6350 6351 down_read(&dev_replace->rwsem); 6352 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6353 /* 6354 * Hold the semaphore for read during the whole operation, write is 6355 * requested at commit time but must wait. 6356 */ 6357 if (!dev_replace_is_ongoing) 6358 up_read(&dev_replace->rwsem); 6359 6360 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6361 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6362 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6363 dev_replace->srcdev->devid, 6364 &mirror_num, 6365 &physical_to_patch_in_first_stripe); 6366 if (ret) 6367 goto out; 6368 else 6369 patch_the_first_stripe_for_dev_replace = 1; 6370 } else if (mirror_num > map->num_stripes) { 6371 mirror_num = 0; 6372 } 6373 6374 num_stripes = 1; 6375 stripe_index = 0; 6376 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6377 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6378 &stripe_index); 6379 if (!need_full_stripe(op)) 6380 mirror_num = 1; 6381 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6382 if (need_full_stripe(op)) 6383 num_stripes = map->num_stripes; 6384 else if (mirror_num) 6385 stripe_index = mirror_num - 1; 6386 else { 6387 stripe_index = find_live_mirror(fs_info, map, 0, 6388 dev_replace_is_ongoing); 6389 mirror_num = stripe_index + 1; 6390 } 6391 6392 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6393 if (need_full_stripe(op)) { 6394 num_stripes = map->num_stripes; 6395 } else if (mirror_num) { 6396 stripe_index = mirror_num - 1; 6397 } else { 6398 mirror_num = 1; 6399 } 6400 6401 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6402 u32 factor = map->num_stripes / map->sub_stripes; 6403 6404 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6405 stripe_index *= map->sub_stripes; 6406 6407 if (need_full_stripe(op)) 6408 num_stripes = map->sub_stripes; 6409 else if (mirror_num) 6410 stripe_index += mirror_num - 1; 6411 else { 6412 int old_stripe_index = stripe_index; 6413 stripe_index = find_live_mirror(fs_info, map, 6414 stripe_index, 6415 dev_replace_is_ongoing); 6416 mirror_num = stripe_index - old_stripe_index + 1; 6417 } 6418 6419 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6420 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6421 /* push stripe_nr back to the start of the full stripe */ 6422 stripe_nr = div64_u64(raid56_full_stripe_start, 6423 stripe_len * data_stripes); 6424 6425 /* RAID[56] write or recovery. Return all stripes */ 6426 num_stripes = map->num_stripes; 6427 max_errors = nr_parity_stripes(map); 6428 6429 *length = map->stripe_len; 6430 stripe_index = 0; 6431 stripe_offset = 0; 6432 } else { 6433 /* 6434 * Mirror #0 or #1 means the original data block. 6435 * Mirror #2 is RAID5 parity block. 6436 * Mirror #3 is RAID6 Q block. 6437 */ 6438 stripe_nr = div_u64_rem(stripe_nr, 6439 data_stripes, &stripe_index); 6440 if (mirror_num > 1) 6441 stripe_index = data_stripes + mirror_num - 2; 6442 6443 /* We distribute the parity blocks across stripes */ 6444 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6445 &stripe_index); 6446 if (!need_full_stripe(op) && mirror_num <= 1) 6447 mirror_num = 1; 6448 } 6449 } else { 6450 /* 6451 * after this, stripe_nr is the number of stripes on this 6452 * device we have to walk to find the data, and stripe_index is 6453 * the number of our device in the stripe array 6454 */ 6455 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6456 &stripe_index); 6457 mirror_num = stripe_index + 1; 6458 } 6459 if (stripe_index >= map->num_stripes) { 6460 btrfs_crit(fs_info, 6461 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6462 stripe_index, map->num_stripes); 6463 ret = -EINVAL; 6464 goto out; 6465 } 6466 6467 num_alloc_stripes = num_stripes; 6468 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6469 if (op == BTRFS_MAP_WRITE) 6470 num_alloc_stripes <<= 1; 6471 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6472 num_alloc_stripes++; 6473 tgtdev_indexes = num_stripes; 6474 } 6475 6476 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6477 if (!bioc) { 6478 ret = -ENOMEM; 6479 goto out; 6480 } 6481 6482 for (i = 0; i < num_stripes; i++) { 6483 bioc->stripes[i].physical = map->stripes[stripe_index].physical + 6484 stripe_offset + stripe_nr * map->stripe_len; 6485 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 6486 stripe_index++; 6487 } 6488 6489 /* Build raid_map */ 6490 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6491 (need_full_stripe(op) || mirror_num > 1)) { 6492 u64 tmp; 6493 unsigned rot; 6494 6495 /* Work out the disk rotation on this stripe-set */ 6496 div_u64_rem(stripe_nr, num_stripes, &rot); 6497 6498 /* Fill in the logical address of each stripe */ 6499 tmp = stripe_nr * data_stripes; 6500 for (i = 0; i < data_stripes; i++) 6501 bioc->raid_map[(i + rot) % num_stripes] = 6502 em->start + (tmp + i) * map->stripe_len; 6503 6504 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6505 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6506 bioc->raid_map[(i + rot + 1) % num_stripes] = 6507 RAID6_Q_STRIPE; 6508 6509 sort_parity_stripes(bioc, num_stripes); 6510 } 6511 6512 if (need_full_stripe(op)) 6513 max_errors = btrfs_chunk_max_errors(map); 6514 6515 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6516 need_full_stripe(op)) { 6517 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6518 &num_stripes, &max_errors); 6519 } 6520 6521 *bioc_ret = bioc; 6522 bioc->map_type = map->type; 6523 bioc->num_stripes = num_stripes; 6524 bioc->max_errors = max_errors; 6525 bioc->mirror_num = mirror_num; 6526 6527 /* 6528 * this is the case that REQ_READ && dev_replace_is_ongoing && 6529 * mirror_num == num_stripes + 1 && dev_replace target drive is 6530 * available as a mirror 6531 */ 6532 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6533 WARN_ON(num_stripes > 1); 6534 bioc->stripes[0].dev = dev_replace->tgtdev; 6535 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6536 bioc->mirror_num = map->num_stripes + 1; 6537 } 6538 out: 6539 if (dev_replace_is_ongoing) { 6540 lockdep_assert_held(&dev_replace->rwsem); 6541 /* Unlock and let waiting writers proceed */ 6542 up_read(&dev_replace->rwsem); 6543 } 6544 free_extent_map(em); 6545 return ret; 6546 } 6547 6548 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6549 u64 logical, u64 *length, 6550 struct btrfs_io_context **bioc_ret, int mirror_num) 6551 { 6552 if (op == BTRFS_MAP_DISCARD) 6553 return __btrfs_map_block_for_discard(fs_info, logical, 6554 length, bioc_ret); 6555 6556 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6557 mirror_num, 0); 6558 } 6559 6560 /* For Scrub/replace */ 6561 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6562 u64 logical, u64 *length, 6563 struct btrfs_io_context **bioc_ret) 6564 { 6565 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1); 6566 } 6567 6568 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio) 6569 { 6570 bio->bi_private = bioc->private; 6571 bio->bi_end_io = bioc->end_io; 6572 bio_endio(bio); 6573 6574 btrfs_put_bioc(bioc); 6575 } 6576 6577 static void btrfs_end_bio(struct bio *bio) 6578 { 6579 struct btrfs_io_context *bioc = bio->bi_private; 6580 int is_orig_bio = 0; 6581 6582 if (bio->bi_status) { 6583 atomic_inc(&bioc->error); 6584 if (bio->bi_status == BLK_STS_IOERR || 6585 bio->bi_status == BLK_STS_TARGET) { 6586 struct btrfs_device *dev = btrfs_bio(bio)->device; 6587 6588 ASSERT(dev->bdev); 6589 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6590 btrfs_dev_stat_inc_and_print(dev, 6591 BTRFS_DEV_STAT_WRITE_ERRS); 6592 else if (!(bio->bi_opf & REQ_RAHEAD)) 6593 btrfs_dev_stat_inc_and_print(dev, 6594 BTRFS_DEV_STAT_READ_ERRS); 6595 if (bio->bi_opf & REQ_PREFLUSH) 6596 btrfs_dev_stat_inc_and_print(dev, 6597 BTRFS_DEV_STAT_FLUSH_ERRS); 6598 } 6599 } 6600 6601 if (bio == bioc->orig_bio) 6602 is_orig_bio = 1; 6603 6604 btrfs_bio_counter_dec(bioc->fs_info); 6605 6606 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6607 if (!is_orig_bio) { 6608 bio_put(bio); 6609 bio = bioc->orig_bio; 6610 } 6611 6612 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6613 /* only send an error to the higher layers if it is 6614 * beyond the tolerance of the btrfs bio 6615 */ 6616 if (atomic_read(&bioc->error) > bioc->max_errors) { 6617 bio->bi_status = BLK_STS_IOERR; 6618 } else { 6619 /* 6620 * this bio is actually up to date, we didn't 6621 * go over the max number of errors 6622 */ 6623 bio->bi_status = BLK_STS_OK; 6624 } 6625 6626 btrfs_end_bioc(bioc, bio); 6627 } else if (!is_orig_bio) { 6628 bio_put(bio); 6629 } 6630 } 6631 6632 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio, 6633 u64 physical, struct btrfs_device *dev) 6634 { 6635 struct btrfs_fs_info *fs_info = bioc->fs_info; 6636 6637 bio->bi_private = bioc; 6638 btrfs_bio(bio)->device = dev; 6639 bio->bi_end_io = btrfs_end_bio; 6640 bio->bi_iter.bi_sector = physical >> 9; 6641 /* 6642 * For zone append writing, bi_sector must point the beginning of the 6643 * zone 6644 */ 6645 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6646 if (btrfs_dev_is_sequential(dev, physical)) { 6647 u64 zone_start = round_down(physical, fs_info->zone_size); 6648 6649 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6650 } else { 6651 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6652 bio->bi_opf |= REQ_OP_WRITE; 6653 } 6654 } 6655 btrfs_debug_in_rcu(fs_info, 6656 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6657 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6658 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6659 dev->devid, bio->bi_iter.bi_size); 6660 bio_set_dev(bio, dev->bdev); 6661 6662 btrfs_bio_counter_inc_noblocked(fs_info); 6663 6664 btrfsic_submit_bio(bio); 6665 } 6666 6667 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical) 6668 { 6669 atomic_inc(&bioc->error); 6670 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6671 /* Should be the original bio. */ 6672 WARN_ON(bio != bioc->orig_bio); 6673 6674 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6675 bio->bi_iter.bi_sector = logical >> 9; 6676 if (atomic_read(&bioc->error) > bioc->max_errors) 6677 bio->bi_status = BLK_STS_IOERR; 6678 else 6679 bio->bi_status = BLK_STS_OK; 6680 btrfs_end_bioc(bioc, bio); 6681 } 6682 } 6683 6684 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 6685 int mirror_num) 6686 { 6687 struct btrfs_device *dev; 6688 struct bio *first_bio = bio; 6689 u64 logical = bio->bi_iter.bi_sector << 9; 6690 u64 length = 0; 6691 u64 map_length; 6692 int ret; 6693 int dev_nr; 6694 int total_devs; 6695 struct btrfs_io_context *bioc = NULL; 6696 6697 length = bio->bi_iter.bi_size; 6698 map_length = length; 6699 6700 btrfs_bio_counter_inc_blocked(fs_info); 6701 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, 6702 &map_length, &bioc, mirror_num, 1); 6703 if (ret) { 6704 btrfs_bio_counter_dec(fs_info); 6705 return errno_to_blk_status(ret); 6706 } 6707 6708 total_devs = bioc->num_stripes; 6709 bioc->orig_bio = first_bio; 6710 bioc->private = first_bio->bi_private; 6711 bioc->end_io = first_bio->bi_end_io; 6712 atomic_set(&bioc->stripes_pending, bioc->num_stripes); 6713 6714 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && 6715 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { 6716 /* In this case, map_length has been set to the length of 6717 a single stripe; not the whole write */ 6718 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 6719 ret = raid56_parity_write(bio, bioc, map_length); 6720 } else { 6721 ret = raid56_parity_recover(bio, bioc, map_length, 6722 mirror_num, 1); 6723 } 6724 6725 btrfs_bio_counter_dec(fs_info); 6726 return errno_to_blk_status(ret); 6727 } 6728 6729 if (map_length < length) { 6730 btrfs_crit(fs_info, 6731 "mapping failed logical %llu bio len %llu len %llu", 6732 logical, length, map_length); 6733 BUG(); 6734 } 6735 6736 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { 6737 dev = bioc->stripes[dev_nr].dev; 6738 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, 6739 &dev->dev_state) || 6740 (btrfs_op(first_bio) == BTRFS_MAP_WRITE && 6741 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6742 bioc_error(bioc, first_bio, logical); 6743 continue; 6744 } 6745 6746 if (dev_nr < total_devs - 1) 6747 bio = btrfs_bio_clone(first_bio); 6748 else 6749 bio = first_bio; 6750 6751 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev); 6752 } 6753 btrfs_bio_counter_dec(fs_info); 6754 return BLK_STS_OK; 6755 } 6756 6757 /* 6758 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6759 * return NULL. 6760 * 6761 * If devid and uuid are both specified, the match must be exact, otherwise 6762 * only devid is used. 6763 */ 6764 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, 6765 u64 devid, u8 *uuid, u8 *fsid) 6766 { 6767 struct btrfs_device *device; 6768 struct btrfs_fs_devices *seed_devs; 6769 6770 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { 6771 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6772 if (device->devid == devid && 6773 (!uuid || memcmp(device->uuid, uuid, 6774 BTRFS_UUID_SIZE) == 0)) 6775 return device; 6776 } 6777 } 6778 6779 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6780 if (!fsid || 6781 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { 6782 list_for_each_entry(device, &seed_devs->devices, 6783 dev_list) { 6784 if (device->devid == devid && 6785 (!uuid || memcmp(device->uuid, uuid, 6786 BTRFS_UUID_SIZE) == 0)) 6787 return device; 6788 } 6789 } 6790 } 6791 6792 return NULL; 6793 } 6794 6795 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6796 u64 devid, u8 *dev_uuid) 6797 { 6798 struct btrfs_device *device; 6799 unsigned int nofs_flag; 6800 6801 /* 6802 * We call this under the chunk_mutex, so we want to use NOFS for this 6803 * allocation, however we don't want to change btrfs_alloc_device() to 6804 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6805 * places. 6806 */ 6807 nofs_flag = memalloc_nofs_save(); 6808 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 6809 memalloc_nofs_restore(nofs_flag); 6810 if (IS_ERR(device)) 6811 return device; 6812 6813 list_add(&device->dev_list, &fs_devices->devices); 6814 device->fs_devices = fs_devices; 6815 fs_devices->num_devices++; 6816 6817 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6818 fs_devices->missing_devices++; 6819 6820 return device; 6821 } 6822 6823 /** 6824 * btrfs_alloc_device - allocate struct btrfs_device 6825 * @fs_info: used only for generating a new devid, can be NULL if 6826 * devid is provided (i.e. @devid != NULL). 6827 * @devid: a pointer to devid for this device. If NULL a new devid 6828 * is generated. 6829 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6830 * is generated. 6831 * 6832 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6833 * on error. Returned struct is not linked onto any lists and must be 6834 * destroyed with btrfs_free_device. 6835 */ 6836 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6837 const u64 *devid, 6838 const u8 *uuid) 6839 { 6840 struct btrfs_device *dev; 6841 u64 tmp; 6842 6843 if (WARN_ON(!devid && !fs_info)) 6844 return ERR_PTR(-EINVAL); 6845 6846 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6847 if (!dev) 6848 return ERR_PTR(-ENOMEM); 6849 6850 /* 6851 * Preallocate a bio that's always going to be used for flushing device 6852 * barriers and matches the device lifespan 6853 */ 6854 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0); 6855 if (!dev->flush_bio) { 6856 kfree(dev); 6857 return ERR_PTR(-ENOMEM); 6858 } 6859 6860 INIT_LIST_HEAD(&dev->dev_list); 6861 INIT_LIST_HEAD(&dev->dev_alloc_list); 6862 INIT_LIST_HEAD(&dev->post_commit_list); 6863 6864 atomic_set(&dev->reada_in_flight, 0); 6865 atomic_set(&dev->dev_stats_ccnt, 0); 6866 btrfs_device_data_ordered_init(dev); 6867 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6868 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6869 extent_io_tree_init(fs_info, &dev->alloc_state, 6870 IO_TREE_DEVICE_ALLOC_STATE, NULL); 6871 6872 if (devid) 6873 tmp = *devid; 6874 else { 6875 int ret; 6876 6877 ret = find_next_devid(fs_info, &tmp); 6878 if (ret) { 6879 btrfs_free_device(dev); 6880 return ERR_PTR(ret); 6881 } 6882 } 6883 dev->devid = tmp; 6884 6885 if (uuid) 6886 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6887 else 6888 generate_random_uuid(dev->uuid); 6889 6890 return dev; 6891 } 6892 6893 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6894 u64 devid, u8 *uuid, bool error) 6895 { 6896 if (error) 6897 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6898 devid, uuid); 6899 else 6900 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6901 devid, uuid); 6902 } 6903 6904 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) 6905 { 6906 const int data_stripes = calc_data_stripes(type, num_stripes); 6907 6908 return div_u64(chunk_len, data_stripes); 6909 } 6910 6911 #if BITS_PER_LONG == 32 6912 /* 6913 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6914 * can't be accessed on 32bit systems. 6915 * 6916 * This function do mount time check to reject the fs if it already has 6917 * metadata chunk beyond that limit. 6918 */ 6919 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6920 u64 logical, u64 length, u64 type) 6921 { 6922 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6923 return 0; 6924 6925 if (logical + length < MAX_LFS_FILESIZE) 6926 return 0; 6927 6928 btrfs_err_32bit_limit(fs_info); 6929 return -EOVERFLOW; 6930 } 6931 6932 /* 6933 * This is to give early warning for any metadata chunk reaching 6934 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6935 * Although we can still access the metadata, it's not going to be possible 6936 * once the limit is reached. 6937 */ 6938 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6939 u64 logical, u64 length, u64 type) 6940 { 6941 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6942 return; 6943 6944 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 6945 return; 6946 6947 btrfs_warn_32bit_limit(fs_info); 6948 } 6949 #endif 6950 6951 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 6952 struct btrfs_chunk *chunk) 6953 { 6954 struct btrfs_fs_info *fs_info = leaf->fs_info; 6955 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 6956 struct map_lookup *map; 6957 struct extent_map *em; 6958 u64 logical; 6959 u64 length; 6960 u64 devid; 6961 u64 type; 6962 u8 uuid[BTRFS_UUID_SIZE]; 6963 int num_stripes; 6964 int ret; 6965 int i; 6966 6967 logical = key->offset; 6968 length = btrfs_chunk_length(leaf, chunk); 6969 type = btrfs_chunk_type(leaf, chunk); 6970 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 6971 6972 #if BITS_PER_LONG == 32 6973 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 6974 if (ret < 0) 6975 return ret; 6976 warn_32bit_meta_chunk(fs_info, logical, length, type); 6977 #endif 6978 6979 /* 6980 * Only need to verify chunk item if we're reading from sys chunk array, 6981 * as chunk item in tree block is already verified by tree-checker. 6982 */ 6983 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 6984 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 6985 if (ret) 6986 return ret; 6987 } 6988 6989 read_lock(&map_tree->lock); 6990 em = lookup_extent_mapping(map_tree, logical, 1); 6991 read_unlock(&map_tree->lock); 6992 6993 /* already mapped? */ 6994 if (em && em->start <= logical && em->start + em->len > logical) { 6995 free_extent_map(em); 6996 return 0; 6997 } else if (em) { 6998 free_extent_map(em); 6999 } 7000 7001 em = alloc_extent_map(); 7002 if (!em) 7003 return -ENOMEM; 7004 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7005 if (!map) { 7006 free_extent_map(em); 7007 return -ENOMEM; 7008 } 7009 7010 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7011 em->map_lookup = map; 7012 em->start = logical; 7013 em->len = length; 7014 em->orig_start = 0; 7015 em->block_start = 0; 7016 em->block_len = em->len; 7017 7018 map->num_stripes = num_stripes; 7019 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7020 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7021 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7022 map->type = type; 7023 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 7024 map->verified_stripes = 0; 7025 em->orig_block_len = calc_stripe_length(type, em->len, 7026 map->num_stripes); 7027 for (i = 0; i < num_stripes; i++) { 7028 map->stripes[i].physical = 7029 btrfs_stripe_offset_nr(leaf, chunk, i); 7030 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7031 read_extent_buffer(leaf, uuid, (unsigned long) 7032 btrfs_stripe_dev_uuid_nr(chunk, i), 7033 BTRFS_UUID_SIZE); 7034 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, 7035 devid, uuid, NULL); 7036 if (!map->stripes[i].dev && 7037 !btrfs_test_opt(fs_info, DEGRADED)) { 7038 free_extent_map(em); 7039 btrfs_report_missing_device(fs_info, devid, uuid, true); 7040 return -ENOENT; 7041 } 7042 if (!map->stripes[i].dev) { 7043 map->stripes[i].dev = 7044 add_missing_dev(fs_info->fs_devices, devid, 7045 uuid); 7046 if (IS_ERR(map->stripes[i].dev)) { 7047 free_extent_map(em); 7048 btrfs_err(fs_info, 7049 "failed to init missing dev %llu: %ld", 7050 devid, PTR_ERR(map->stripes[i].dev)); 7051 return PTR_ERR(map->stripes[i].dev); 7052 } 7053 btrfs_report_missing_device(fs_info, devid, uuid, false); 7054 } 7055 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7056 &(map->stripes[i].dev->dev_state)); 7057 7058 } 7059 7060 write_lock(&map_tree->lock); 7061 ret = add_extent_mapping(map_tree, em, 0); 7062 write_unlock(&map_tree->lock); 7063 if (ret < 0) { 7064 btrfs_err(fs_info, 7065 "failed to add chunk map, start=%llu len=%llu: %d", 7066 em->start, em->len, ret); 7067 } 7068 free_extent_map(em); 7069 7070 return ret; 7071 } 7072 7073 static void fill_device_from_item(struct extent_buffer *leaf, 7074 struct btrfs_dev_item *dev_item, 7075 struct btrfs_device *device) 7076 { 7077 unsigned long ptr; 7078 7079 device->devid = btrfs_device_id(leaf, dev_item); 7080 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7081 device->total_bytes = device->disk_total_bytes; 7082 device->commit_total_bytes = device->disk_total_bytes; 7083 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7084 device->commit_bytes_used = device->bytes_used; 7085 device->type = btrfs_device_type(leaf, dev_item); 7086 device->io_align = btrfs_device_io_align(leaf, dev_item); 7087 device->io_width = btrfs_device_io_width(leaf, dev_item); 7088 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7089 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7090 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7091 7092 ptr = btrfs_device_uuid(dev_item); 7093 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7094 } 7095 7096 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7097 u8 *fsid) 7098 { 7099 struct btrfs_fs_devices *fs_devices; 7100 int ret; 7101 7102 lockdep_assert_held(&uuid_mutex); 7103 ASSERT(fsid); 7104 7105 /* This will match only for multi-device seed fs */ 7106 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7107 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7108 return fs_devices; 7109 7110 7111 fs_devices = find_fsid(fsid, NULL); 7112 if (!fs_devices) { 7113 if (!btrfs_test_opt(fs_info, DEGRADED)) 7114 return ERR_PTR(-ENOENT); 7115 7116 fs_devices = alloc_fs_devices(fsid, NULL); 7117 if (IS_ERR(fs_devices)) 7118 return fs_devices; 7119 7120 fs_devices->seeding = true; 7121 fs_devices->opened = 1; 7122 return fs_devices; 7123 } 7124 7125 /* 7126 * Upon first call for a seed fs fsid, just create a private copy of the 7127 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7128 */ 7129 fs_devices = clone_fs_devices(fs_devices); 7130 if (IS_ERR(fs_devices)) 7131 return fs_devices; 7132 7133 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7134 if (ret) { 7135 free_fs_devices(fs_devices); 7136 return ERR_PTR(ret); 7137 } 7138 7139 if (!fs_devices->seeding) { 7140 close_fs_devices(fs_devices); 7141 free_fs_devices(fs_devices); 7142 return ERR_PTR(-EINVAL); 7143 } 7144 7145 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7146 7147 return fs_devices; 7148 } 7149 7150 static int read_one_dev(struct extent_buffer *leaf, 7151 struct btrfs_dev_item *dev_item) 7152 { 7153 struct btrfs_fs_info *fs_info = leaf->fs_info; 7154 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7155 struct btrfs_device *device; 7156 u64 devid; 7157 int ret; 7158 u8 fs_uuid[BTRFS_FSID_SIZE]; 7159 u8 dev_uuid[BTRFS_UUID_SIZE]; 7160 7161 devid = btrfs_device_id(leaf, dev_item); 7162 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7163 BTRFS_UUID_SIZE); 7164 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7165 BTRFS_FSID_SIZE); 7166 7167 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7168 fs_devices = open_seed_devices(fs_info, fs_uuid); 7169 if (IS_ERR(fs_devices)) 7170 return PTR_ERR(fs_devices); 7171 } 7172 7173 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, 7174 fs_uuid); 7175 if (!device) { 7176 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7177 btrfs_report_missing_device(fs_info, devid, 7178 dev_uuid, true); 7179 return -ENOENT; 7180 } 7181 7182 device = add_missing_dev(fs_devices, devid, dev_uuid); 7183 if (IS_ERR(device)) { 7184 btrfs_err(fs_info, 7185 "failed to add missing dev %llu: %ld", 7186 devid, PTR_ERR(device)); 7187 return PTR_ERR(device); 7188 } 7189 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7190 } else { 7191 if (!device->bdev) { 7192 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7193 btrfs_report_missing_device(fs_info, 7194 devid, dev_uuid, true); 7195 return -ENOENT; 7196 } 7197 btrfs_report_missing_device(fs_info, devid, 7198 dev_uuid, false); 7199 } 7200 7201 if (!device->bdev && 7202 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7203 /* 7204 * this happens when a device that was properly setup 7205 * in the device info lists suddenly goes bad. 7206 * device->bdev is NULL, and so we have to set 7207 * device->missing to one here 7208 */ 7209 device->fs_devices->missing_devices++; 7210 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7211 } 7212 7213 /* Move the device to its own fs_devices */ 7214 if (device->fs_devices != fs_devices) { 7215 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7216 &device->dev_state)); 7217 7218 list_move(&device->dev_list, &fs_devices->devices); 7219 device->fs_devices->num_devices--; 7220 fs_devices->num_devices++; 7221 7222 device->fs_devices->missing_devices--; 7223 fs_devices->missing_devices++; 7224 7225 device->fs_devices = fs_devices; 7226 } 7227 } 7228 7229 if (device->fs_devices != fs_info->fs_devices) { 7230 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7231 if (device->generation != 7232 btrfs_device_generation(leaf, dev_item)) 7233 return -EINVAL; 7234 } 7235 7236 fill_device_from_item(leaf, dev_item, device); 7237 if (device->bdev) { 7238 u64 max_total_bytes = i_size_read(device->bdev->bd_inode); 7239 7240 if (device->total_bytes > max_total_bytes) { 7241 btrfs_err(fs_info, 7242 "device total_bytes should be at most %llu but found %llu", 7243 max_total_bytes, device->total_bytes); 7244 return -EINVAL; 7245 } 7246 } 7247 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7248 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7249 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7250 device->fs_devices->total_rw_bytes += device->total_bytes; 7251 atomic64_add(device->total_bytes - device->bytes_used, 7252 &fs_info->free_chunk_space); 7253 } 7254 ret = 0; 7255 return ret; 7256 } 7257 7258 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7259 { 7260 struct btrfs_root *root = fs_info->tree_root; 7261 struct btrfs_super_block *super_copy = fs_info->super_copy; 7262 struct extent_buffer *sb; 7263 struct btrfs_disk_key *disk_key; 7264 struct btrfs_chunk *chunk; 7265 u8 *array_ptr; 7266 unsigned long sb_array_offset; 7267 int ret = 0; 7268 u32 num_stripes; 7269 u32 array_size; 7270 u32 len = 0; 7271 u32 cur_offset; 7272 u64 type; 7273 struct btrfs_key key; 7274 7275 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7276 /* 7277 * This will create extent buffer of nodesize, superblock size is 7278 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will 7279 * overallocate but we can keep it as-is, only the first page is used. 7280 */ 7281 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, 7282 root->root_key.objectid, 0); 7283 if (IS_ERR(sb)) 7284 return PTR_ERR(sb); 7285 set_extent_buffer_uptodate(sb); 7286 /* 7287 * The sb extent buffer is artificial and just used to read the system array. 7288 * set_extent_buffer_uptodate() call does not properly mark all it's 7289 * pages up-to-date when the page is larger: extent does not cover the 7290 * whole page and consequently check_page_uptodate does not find all 7291 * the page's extents up-to-date (the hole beyond sb), 7292 * write_extent_buffer then triggers a WARN_ON. 7293 * 7294 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, 7295 * but sb spans only this function. Add an explicit SetPageUptodate call 7296 * to silence the warning eg. on PowerPC 64. 7297 */ 7298 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) 7299 SetPageUptodate(sb->pages[0]); 7300 7301 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7302 array_size = btrfs_super_sys_array_size(super_copy); 7303 7304 array_ptr = super_copy->sys_chunk_array; 7305 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7306 cur_offset = 0; 7307 7308 while (cur_offset < array_size) { 7309 disk_key = (struct btrfs_disk_key *)array_ptr; 7310 len = sizeof(*disk_key); 7311 if (cur_offset + len > array_size) 7312 goto out_short_read; 7313 7314 btrfs_disk_key_to_cpu(&key, disk_key); 7315 7316 array_ptr += len; 7317 sb_array_offset += len; 7318 cur_offset += len; 7319 7320 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7321 btrfs_err(fs_info, 7322 "unexpected item type %u in sys_array at offset %u", 7323 (u32)key.type, cur_offset); 7324 ret = -EIO; 7325 break; 7326 } 7327 7328 chunk = (struct btrfs_chunk *)sb_array_offset; 7329 /* 7330 * At least one btrfs_chunk with one stripe must be present, 7331 * exact stripe count check comes afterwards 7332 */ 7333 len = btrfs_chunk_item_size(1); 7334 if (cur_offset + len > array_size) 7335 goto out_short_read; 7336 7337 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7338 if (!num_stripes) { 7339 btrfs_err(fs_info, 7340 "invalid number of stripes %u in sys_array at offset %u", 7341 num_stripes, cur_offset); 7342 ret = -EIO; 7343 break; 7344 } 7345 7346 type = btrfs_chunk_type(sb, chunk); 7347 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7348 btrfs_err(fs_info, 7349 "invalid chunk type %llu in sys_array at offset %u", 7350 type, cur_offset); 7351 ret = -EIO; 7352 break; 7353 } 7354 7355 len = btrfs_chunk_item_size(num_stripes); 7356 if (cur_offset + len > array_size) 7357 goto out_short_read; 7358 7359 ret = read_one_chunk(&key, sb, chunk); 7360 if (ret) 7361 break; 7362 7363 array_ptr += len; 7364 sb_array_offset += len; 7365 cur_offset += len; 7366 } 7367 clear_extent_buffer_uptodate(sb); 7368 free_extent_buffer_stale(sb); 7369 return ret; 7370 7371 out_short_read: 7372 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7373 len, cur_offset); 7374 clear_extent_buffer_uptodate(sb); 7375 free_extent_buffer_stale(sb); 7376 return -EIO; 7377 } 7378 7379 /* 7380 * Check if all chunks in the fs are OK for read-write degraded mount 7381 * 7382 * If the @failing_dev is specified, it's accounted as missing. 7383 * 7384 * Return true if all chunks meet the minimal RW mount requirements. 7385 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7386 */ 7387 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7388 struct btrfs_device *failing_dev) 7389 { 7390 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7391 struct extent_map *em; 7392 u64 next_start = 0; 7393 bool ret = true; 7394 7395 read_lock(&map_tree->lock); 7396 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7397 read_unlock(&map_tree->lock); 7398 /* No chunk at all? Return false anyway */ 7399 if (!em) { 7400 ret = false; 7401 goto out; 7402 } 7403 while (em) { 7404 struct map_lookup *map; 7405 int missing = 0; 7406 int max_tolerated; 7407 int i; 7408 7409 map = em->map_lookup; 7410 max_tolerated = 7411 btrfs_get_num_tolerated_disk_barrier_failures( 7412 map->type); 7413 for (i = 0; i < map->num_stripes; i++) { 7414 struct btrfs_device *dev = map->stripes[i].dev; 7415 7416 if (!dev || !dev->bdev || 7417 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7418 dev->last_flush_error) 7419 missing++; 7420 else if (failing_dev && failing_dev == dev) 7421 missing++; 7422 } 7423 if (missing > max_tolerated) { 7424 if (!failing_dev) 7425 btrfs_warn(fs_info, 7426 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7427 em->start, missing, max_tolerated); 7428 free_extent_map(em); 7429 ret = false; 7430 goto out; 7431 } 7432 next_start = extent_map_end(em); 7433 free_extent_map(em); 7434 7435 read_lock(&map_tree->lock); 7436 em = lookup_extent_mapping(map_tree, next_start, 7437 (u64)(-1) - next_start); 7438 read_unlock(&map_tree->lock); 7439 } 7440 out: 7441 return ret; 7442 } 7443 7444 static void readahead_tree_node_children(struct extent_buffer *node) 7445 { 7446 int i; 7447 const int nr_items = btrfs_header_nritems(node); 7448 7449 for (i = 0; i < nr_items; i++) 7450 btrfs_readahead_node_child(node, i); 7451 } 7452 7453 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7454 { 7455 struct btrfs_root *root = fs_info->chunk_root; 7456 struct btrfs_path *path; 7457 struct extent_buffer *leaf; 7458 struct btrfs_key key; 7459 struct btrfs_key found_key; 7460 int ret; 7461 int slot; 7462 u64 total_dev = 0; 7463 u64 last_ra_node = 0; 7464 7465 path = btrfs_alloc_path(); 7466 if (!path) 7467 return -ENOMEM; 7468 7469 /* 7470 * uuid_mutex is needed only if we are mounting a sprout FS 7471 * otherwise we don't need it. 7472 */ 7473 mutex_lock(&uuid_mutex); 7474 7475 /* 7476 * It is possible for mount and umount to race in such a way that 7477 * we execute this code path, but open_fs_devices failed to clear 7478 * total_rw_bytes. We certainly want it cleared before reading the 7479 * device items, so clear it here. 7480 */ 7481 fs_info->fs_devices->total_rw_bytes = 0; 7482 7483 /* 7484 * Read all device items, and then all the chunk items. All 7485 * device items are found before any chunk item (their object id 7486 * is smaller than the lowest possible object id for a chunk 7487 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7488 */ 7489 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7490 key.offset = 0; 7491 key.type = 0; 7492 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7493 if (ret < 0) 7494 goto error; 7495 while (1) { 7496 struct extent_buffer *node; 7497 7498 leaf = path->nodes[0]; 7499 slot = path->slots[0]; 7500 if (slot >= btrfs_header_nritems(leaf)) { 7501 ret = btrfs_next_leaf(root, path); 7502 if (ret == 0) 7503 continue; 7504 if (ret < 0) 7505 goto error; 7506 break; 7507 } 7508 /* 7509 * The nodes on level 1 are not locked but we don't need to do 7510 * that during mount time as nothing else can access the tree 7511 */ 7512 node = path->nodes[1]; 7513 if (node) { 7514 if (last_ra_node != node->start) { 7515 readahead_tree_node_children(node); 7516 last_ra_node = node->start; 7517 } 7518 } 7519 btrfs_item_key_to_cpu(leaf, &found_key, slot); 7520 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7521 struct btrfs_dev_item *dev_item; 7522 dev_item = btrfs_item_ptr(leaf, slot, 7523 struct btrfs_dev_item); 7524 ret = read_one_dev(leaf, dev_item); 7525 if (ret) 7526 goto error; 7527 total_dev++; 7528 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7529 struct btrfs_chunk *chunk; 7530 7531 /* 7532 * We are only called at mount time, so no need to take 7533 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7534 * we always lock first fs_info->chunk_mutex before 7535 * acquiring any locks on the chunk tree. This is a 7536 * requirement for chunk allocation, see the comment on 7537 * top of btrfs_chunk_alloc() for details. 7538 */ 7539 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7540 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7541 ret = read_one_chunk(&found_key, leaf, chunk); 7542 if (ret) 7543 goto error; 7544 } 7545 path->slots[0]++; 7546 } 7547 7548 /* 7549 * After loading chunk tree, we've got all device information, 7550 * do another round of validation checks. 7551 */ 7552 if (total_dev != fs_info->fs_devices->total_devices) { 7553 btrfs_err(fs_info, 7554 "super_num_devices %llu mismatch with num_devices %llu found here", 7555 btrfs_super_num_devices(fs_info->super_copy), 7556 total_dev); 7557 ret = -EINVAL; 7558 goto error; 7559 } 7560 if (btrfs_super_total_bytes(fs_info->super_copy) < 7561 fs_info->fs_devices->total_rw_bytes) { 7562 btrfs_err(fs_info, 7563 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7564 btrfs_super_total_bytes(fs_info->super_copy), 7565 fs_info->fs_devices->total_rw_bytes); 7566 ret = -EINVAL; 7567 goto error; 7568 } 7569 ret = 0; 7570 error: 7571 mutex_unlock(&uuid_mutex); 7572 7573 btrfs_free_path(path); 7574 return ret; 7575 } 7576 7577 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7578 { 7579 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7580 struct btrfs_device *device; 7581 7582 fs_devices->fs_info = fs_info; 7583 7584 mutex_lock(&fs_devices->device_list_mutex); 7585 list_for_each_entry(device, &fs_devices->devices, dev_list) 7586 device->fs_info = fs_info; 7587 7588 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7589 list_for_each_entry(device, &seed_devs->devices, dev_list) 7590 device->fs_info = fs_info; 7591 7592 seed_devs->fs_info = fs_info; 7593 } 7594 mutex_unlock(&fs_devices->device_list_mutex); 7595 } 7596 7597 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7598 const struct btrfs_dev_stats_item *ptr, 7599 int index) 7600 { 7601 u64 val; 7602 7603 read_extent_buffer(eb, &val, 7604 offsetof(struct btrfs_dev_stats_item, values) + 7605 ((unsigned long)ptr) + (index * sizeof(u64)), 7606 sizeof(val)); 7607 return val; 7608 } 7609 7610 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7611 struct btrfs_dev_stats_item *ptr, 7612 int index, u64 val) 7613 { 7614 write_extent_buffer(eb, &val, 7615 offsetof(struct btrfs_dev_stats_item, values) + 7616 ((unsigned long)ptr) + (index * sizeof(u64)), 7617 sizeof(val)); 7618 } 7619 7620 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7621 struct btrfs_path *path) 7622 { 7623 struct btrfs_dev_stats_item *ptr; 7624 struct extent_buffer *eb; 7625 struct btrfs_key key; 7626 int item_size; 7627 int i, ret, slot; 7628 7629 if (!device->fs_info->dev_root) 7630 return 0; 7631 7632 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7633 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7634 key.offset = device->devid; 7635 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7636 if (ret) { 7637 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7638 btrfs_dev_stat_set(device, i, 0); 7639 device->dev_stats_valid = 1; 7640 btrfs_release_path(path); 7641 return ret < 0 ? ret : 0; 7642 } 7643 slot = path->slots[0]; 7644 eb = path->nodes[0]; 7645 item_size = btrfs_item_size_nr(eb, slot); 7646 7647 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7648 7649 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7650 if (item_size >= (1 + i) * sizeof(__le64)) 7651 btrfs_dev_stat_set(device, i, 7652 btrfs_dev_stats_value(eb, ptr, i)); 7653 else 7654 btrfs_dev_stat_set(device, i, 0); 7655 } 7656 7657 device->dev_stats_valid = 1; 7658 btrfs_dev_stat_print_on_load(device); 7659 btrfs_release_path(path); 7660 7661 return 0; 7662 } 7663 7664 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7665 { 7666 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7667 struct btrfs_device *device; 7668 struct btrfs_path *path = NULL; 7669 int ret = 0; 7670 7671 path = btrfs_alloc_path(); 7672 if (!path) 7673 return -ENOMEM; 7674 7675 mutex_lock(&fs_devices->device_list_mutex); 7676 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7677 ret = btrfs_device_init_dev_stats(device, path); 7678 if (ret) 7679 goto out; 7680 } 7681 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7682 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7683 ret = btrfs_device_init_dev_stats(device, path); 7684 if (ret) 7685 goto out; 7686 } 7687 } 7688 out: 7689 mutex_unlock(&fs_devices->device_list_mutex); 7690 7691 btrfs_free_path(path); 7692 return ret; 7693 } 7694 7695 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7696 struct btrfs_device *device) 7697 { 7698 struct btrfs_fs_info *fs_info = trans->fs_info; 7699 struct btrfs_root *dev_root = fs_info->dev_root; 7700 struct btrfs_path *path; 7701 struct btrfs_key key; 7702 struct extent_buffer *eb; 7703 struct btrfs_dev_stats_item *ptr; 7704 int ret; 7705 int i; 7706 7707 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7708 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7709 key.offset = device->devid; 7710 7711 path = btrfs_alloc_path(); 7712 if (!path) 7713 return -ENOMEM; 7714 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7715 if (ret < 0) { 7716 btrfs_warn_in_rcu(fs_info, 7717 "error %d while searching for dev_stats item for device %s", 7718 ret, rcu_str_deref(device->name)); 7719 goto out; 7720 } 7721 7722 if (ret == 0 && 7723 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7724 /* need to delete old one and insert a new one */ 7725 ret = btrfs_del_item(trans, dev_root, path); 7726 if (ret != 0) { 7727 btrfs_warn_in_rcu(fs_info, 7728 "delete too small dev_stats item for device %s failed %d", 7729 rcu_str_deref(device->name), ret); 7730 goto out; 7731 } 7732 ret = 1; 7733 } 7734 7735 if (ret == 1) { 7736 /* need to insert a new item */ 7737 btrfs_release_path(path); 7738 ret = btrfs_insert_empty_item(trans, dev_root, path, 7739 &key, sizeof(*ptr)); 7740 if (ret < 0) { 7741 btrfs_warn_in_rcu(fs_info, 7742 "insert dev_stats item for device %s failed %d", 7743 rcu_str_deref(device->name), ret); 7744 goto out; 7745 } 7746 } 7747 7748 eb = path->nodes[0]; 7749 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7750 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7751 btrfs_set_dev_stats_value(eb, ptr, i, 7752 btrfs_dev_stat_read(device, i)); 7753 btrfs_mark_buffer_dirty(eb); 7754 7755 out: 7756 btrfs_free_path(path); 7757 return ret; 7758 } 7759 7760 /* 7761 * called from commit_transaction. Writes all changed device stats to disk. 7762 */ 7763 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7764 { 7765 struct btrfs_fs_info *fs_info = trans->fs_info; 7766 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7767 struct btrfs_device *device; 7768 int stats_cnt; 7769 int ret = 0; 7770 7771 mutex_lock(&fs_devices->device_list_mutex); 7772 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7773 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7774 if (!device->dev_stats_valid || stats_cnt == 0) 7775 continue; 7776 7777 7778 /* 7779 * There is a LOAD-LOAD control dependency between the value of 7780 * dev_stats_ccnt and updating the on-disk values which requires 7781 * reading the in-memory counters. Such control dependencies 7782 * require explicit read memory barriers. 7783 * 7784 * This memory barriers pairs with smp_mb__before_atomic in 7785 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7786 * barrier implied by atomic_xchg in 7787 * btrfs_dev_stats_read_and_reset 7788 */ 7789 smp_rmb(); 7790 7791 ret = update_dev_stat_item(trans, device); 7792 if (!ret) 7793 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7794 } 7795 mutex_unlock(&fs_devices->device_list_mutex); 7796 7797 return ret; 7798 } 7799 7800 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7801 { 7802 btrfs_dev_stat_inc(dev, index); 7803 btrfs_dev_stat_print_on_error(dev); 7804 } 7805 7806 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) 7807 { 7808 if (!dev->dev_stats_valid) 7809 return; 7810 btrfs_err_rl_in_rcu(dev->fs_info, 7811 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7812 rcu_str_deref(dev->name), 7813 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7814 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7815 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7816 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7817 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7818 } 7819 7820 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7821 { 7822 int i; 7823 7824 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7825 if (btrfs_dev_stat_read(dev, i) != 0) 7826 break; 7827 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7828 return; /* all values == 0, suppress message */ 7829 7830 btrfs_info_in_rcu(dev->fs_info, 7831 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7832 rcu_str_deref(dev->name), 7833 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7834 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7835 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7836 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7837 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7838 } 7839 7840 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7841 struct btrfs_ioctl_get_dev_stats *stats) 7842 { 7843 struct btrfs_device *dev; 7844 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7845 int i; 7846 7847 mutex_lock(&fs_devices->device_list_mutex); 7848 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL); 7849 mutex_unlock(&fs_devices->device_list_mutex); 7850 7851 if (!dev) { 7852 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7853 return -ENODEV; 7854 } else if (!dev->dev_stats_valid) { 7855 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7856 return -ENODEV; 7857 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7858 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7859 if (stats->nr_items > i) 7860 stats->values[i] = 7861 btrfs_dev_stat_read_and_reset(dev, i); 7862 else 7863 btrfs_dev_stat_set(dev, i, 0); 7864 } 7865 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7866 current->comm, task_pid_nr(current)); 7867 } else { 7868 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7869 if (stats->nr_items > i) 7870 stats->values[i] = btrfs_dev_stat_read(dev, i); 7871 } 7872 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7873 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7874 return 0; 7875 } 7876 7877 /* 7878 * Update the size and bytes used for each device where it changed. This is 7879 * delayed since we would otherwise get errors while writing out the 7880 * superblocks. 7881 * 7882 * Must be invoked during transaction commit. 7883 */ 7884 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7885 { 7886 struct btrfs_device *curr, *next; 7887 7888 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7889 7890 if (list_empty(&trans->dev_update_list)) 7891 return; 7892 7893 /* 7894 * We don't need the device_list_mutex here. This list is owned by the 7895 * transaction and the transaction must complete before the device is 7896 * released. 7897 */ 7898 mutex_lock(&trans->fs_info->chunk_mutex); 7899 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7900 post_commit_list) { 7901 list_del_init(&curr->post_commit_list); 7902 curr->commit_total_bytes = curr->disk_total_bytes; 7903 curr->commit_bytes_used = curr->bytes_used; 7904 } 7905 mutex_unlock(&trans->fs_info->chunk_mutex); 7906 } 7907 7908 /* 7909 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7910 */ 7911 int btrfs_bg_type_to_factor(u64 flags) 7912 { 7913 const int index = btrfs_bg_flags_to_raid_index(flags); 7914 7915 return btrfs_raid_array[index].ncopies; 7916 } 7917 7918 7919 7920 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7921 u64 chunk_offset, u64 devid, 7922 u64 physical_offset, u64 physical_len) 7923 { 7924 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7925 struct extent_map *em; 7926 struct map_lookup *map; 7927 struct btrfs_device *dev; 7928 u64 stripe_len; 7929 bool found = false; 7930 int ret = 0; 7931 int i; 7932 7933 read_lock(&em_tree->lock); 7934 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7935 read_unlock(&em_tree->lock); 7936 7937 if (!em) { 7938 btrfs_err(fs_info, 7939 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 7940 physical_offset, devid); 7941 ret = -EUCLEAN; 7942 goto out; 7943 } 7944 7945 map = em->map_lookup; 7946 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); 7947 if (physical_len != stripe_len) { 7948 btrfs_err(fs_info, 7949 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 7950 physical_offset, devid, em->start, physical_len, 7951 stripe_len); 7952 ret = -EUCLEAN; 7953 goto out; 7954 } 7955 7956 for (i = 0; i < map->num_stripes; i++) { 7957 if (map->stripes[i].dev->devid == devid && 7958 map->stripes[i].physical == physical_offset) { 7959 found = true; 7960 if (map->verified_stripes >= map->num_stripes) { 7961 btrfs_err(fs_info, 7962 "too many dev extents for chunk %llu found", 7963 em->start); 7964 ret = -EUCLEAN; 7965 goto out; 7966 } 7967 map->verified_stripes++; 7968 break; 7969 } 7970 } 7971 if (!found) { 7972 btrfs_err(fs_info, 7973 "dev extent physical offset %llu devid %llu has no corresponding chunk", 7974 physical_offset, devid); 7975 ret = -EUCLEAN; 7976 } 7977 7978 /* Make sure no dev extent is beyond device boundary */ 7979 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL); 7980 if (!dev) { 7981 btrfs_err(fs_info, "failed to find devid %llu", devid); 7982 ret = -EUCLEAN; 7983 goto out; 7984 } 7985 7986 if (physical_offset + physical_len > dev->disk_total_bytes) { 7987 btrfs_err(fs_info, 7988 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 7989 devid, physical_offset, physical_len, 7990 dev->disk_total_bytes); 7991 ret = -EUCLEAN; 7992 goto out; 7993 } 7994 7995 if (dev->zone_info) { 7996 u64 zone_size = dev->zone_info->zone_size; 7997 7998 if (!IS_ALIGNED(physical_offset, zone_size) || 7999 !IS_ALIGNED(physical_len, zone_size)) { 8000 btrfs_err(fs_info, 8001 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8002 devid, physical_offset, physical_len); 8003 ret = -EUCLEAN; 8004 goto out; 8005 } 8006 } 8007 8008 out: 8009 free_extent_map(em); 8010 return ret; 8011 } 8012 8013 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8014 { 8015 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8016 struct extent_map *em; 8017 struct rb_node *node; 8018 int ret = 0; 8019 8020 read_lock(&em_tree->lock); 8021 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8022 em = rb_entry(node, struct extent_map, rb_node); 8023 if (em->map_lookup->num_stripes != 8024 em->map_lookup->verified_stripes) { 8025 btrfs_err(fs_info, 8026 "chunk %llu has missing dev extent, have %d expect %d", 8027 em->start, em->map_lookup->verified_stripes, 8028 em->map_lookup->num_stripes); 8029 ret = -EUCLEAN; 8030 goto out; 8031 } 8032 } 8033 out: 8034 read_unlock(&em_tree->lock); 8035 return ret; 8036 } 8037 8038 /* 8039 * Ensure that all dev extents are mapped to correct chunk, otherwise 8040 * later chunk allocation/free would cause unexpected behavior. 8041 * 8042 * NOTE: This will iterate through the whole device tree, which should be of 8043 * the same size level as the chunk tree. This slightly increases mount time. 8044 */ 8045 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8046 { 8047 struct btrfs_path *path; 8048 struct btrfs_root *root = fs_info->dev_root; 8049 struct btrfs_key key; 8050 u64 prev_devid = 0; 8051 u64 prev_dev_ext_end = 0; 8052 int ret = 0; 8053 8054 /* 8055 * We don't have a dev_root because we mounted with ignorebadroots and 8056 * failed to load the root, so we want to skip the verification in this 8057 * case for sure. 8058 * 8059 * However if the dev root is fine, but the tree itself is corrupted 8060 * we'd still fail to mount. This verification is only to make sure 8061 * writes can happen safely, so instead just bypass this check 8062 * completely in the case of IGNOREBADROOTS. 8063 */ 8064 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8065 return 0; 8066 8067 key.objectid = 1; 8068 key.type = BTRFS_DEV_EXTENT_KEY; 8069 key.offset = 0; 8070 8071 path = btrfs_alloc_path(); 8072 if (!path) 8073 return -ENOMEM; 8074 8075 path->reada = READA_FORWARD; 8076 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8077 if (ret < 0) 8078 goto out; 8079 8080 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8081 ret = btrfs_next_leaf(root, path); 8082 if (ret < 0) 8083 goto out; 8084 /* No dev extents at all? Not good */ 8085 if (ret > 0) { 8086 ret = -EUCLEAN; 8087 goto out; 8088 } 8089 } 8090 while (1) { 8091 struct extent_buffer *leaf = path->nodes[0]; 8092 struct btrfs_dev_extent *dext; 8093 int slot = path->slots[0]; 8094 u64 chunk_offset; 8095 u64 physical_offset; 8096 u64 physical_len; 8097 u64 devid; 8098 8099 btrfs_item_key_to_cpu(leaf, &key, slot); 8100 if (key.type != BTRFS_DEV_EXTENT_KEY) 8101 break; 8102 devid = key.objectid; 8103 physical_offset = key.offset; 8104 8105 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8106 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8107 physical_len = btrfs_dev_extent_length(leaf, dext); 8108 8109 /* Check if this dev extent overlaps with the previous one */ 8110 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8111 btrfs_err(fs_info, 8112 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8113 devid, physical_offset, prev_dev_ext_end); 8114 ret = -EUCLEAN; 8115 goto out; 8116 } 8117 8118 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8119 physical_offset, physical_len); 8120 if (ret < 0) 8121 goto out; 8122 prev_devid = devid; 8123 prev_dev_ext_end = physical_offset + physical_len; 8124 8125 ret = btrfs_next_item(root, path); 8126 if (ret < 0) 8127 goto out; 8128 if (ret > 0) { 8129 ret = 0; 8130 break; 8131 } 8132 } 8133 8134 /* Ensure all chunks have corresponding dev extents */ 8135 ret = verify_chunk_dev_extent_mapping(fs_info); 8136 out: 8137 btrfs_free_path(path); 8138 return ret; 8139 } 8140 8141 /* 8142 * Check whether the given block group or device is pinned by any inode being 8143 * used as a swapfile. 8144 */ 8145 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8146 { 8147 struct btrfs_swapfile_pin *sp; 8148 struct rb_node *node; 8149 8150 spin_lock(&fs_info->swapfile_pins_lock); 8151 node = fs_info->swapfile_pins.rb_node; 8152 while (node) { 8153 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8154 if (ptr < sp->ptr) 8155 node = node->rb_left; 8156 else if (ptr > sp->ptr) 8157 node = node->rb_right; 8158 else 8159 break; 8160 } 8161 spin_unlock(&fs_info->swapfile_pins_lock); 8162 return node != NULL; 8163 } 8164 8165 static int relocating_repair_kthread(void *data) 8166 { 8167 struct btrfs_block_group *cache = (struct btrfs_block_group *)data; 8168 struct btrfs_fs_info *fs_info = cache->fs_info; 8169 u64 target; 8170 int ret = 0; 8171 8172 target = cache->start; 8173 btrfs_put_block_group(cache); 8174 8175 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8176 btrfs_info(fs_info, 8177 "zoned: skip relocating block group %llu to repair: EBUSY", 8178 target); 8179 return -EBUSY; 8180 } 8181 8182 mutex_lock(&fs_info->reclaim_bgs_lock); 8183 8184 /* Ensure block group still exists */ 8185 cache = btrfs_lookup_block_group(fs_info, target); 8186 if (!cache) 8187 goto out; 8188 8189 if (!cache->relocating_repair) 8190 goto out; 8191 8192 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8193 if (ret < 0) 8194 goto out; 8195 8196 btrfs_info(fs_info, 8197 "zoned: relocating block group %llu to repair IO failure", 8198 target); 8199 ret = btrfs_relocate_chunk(fs_info, target); 8200 8201 out: 8202 if (cache) 8203 btrfs_put_block_group(cache); 8204 mutex_unlock(&fs_info->reclaim_bgs_lock); 8205 btrfs_exclop_finish(fs_info); 8206 8207 return ret; 8208 } 8209 8210 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8211 { 8212 struct btrfs_block_group *cache; 8213 8214 /* Do not attempt to repair in degraded state */ 8215 if (btrfs_test_opt(fs_info, DEGRADED)) 8216 return 0; 8217 8218 cache = btrfs_lookup_block_group(fs_info, logical); 8219 if (!cache) 8220 return 0; 8221 8222 spin_lock(&cache->lock); 8223 if (cache->relocating_repair) { 8224 spin_unlock(&cache->lock); 8225 btrfs_put_block_group(cache); 8226 return 0; 8227 } 8228 cache->relocating_repair = 1; 8229 spin_unlock(&cache->lock); 8230 8231 kthread_run(relocating_repair_kthread, cache, 8232 "btrfs-relocating-repair"); 8233 8234 return 0; 8235 } 8236