1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/sched/mm.h> 8 #include <linux/bio.h> 9 #include <linux/slab.h> 10 #include <linux/blkdev.h> 11 #include <linux/ratelimit.h> 12 #include <linux/kthread.h> 13 #include <linux/raid/pq.h> 14 #include <linux/semaphore.h> 15 #include <linux/uuid.h> 16 #include <linux/list_sort.h> 17 #include "misc.h" 18 #include "ctree.h" 19 #include "extent_map.h" 20 #include "disk-io.h" 21 #include "transaction.h" 22 #include "print-tree.h" 23 #include "volumes.h" 24 #include "raid56.h" 25 #include "async-thread.h" 26 #include "check-integrity.h" 27 #include "rcu-string.h" 28 #include "dev-replace.h" 29 #include "sysfs.h" 30 #include "tree-checker.h" 31 #include "space-info.h" 32 #include "block-group.h" 33 #include "discard.h" 34 #include "zoned.h" 35 36 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { 37 [BTRFS_RAID_RAID10] = { 38 .sub_stripes = 2, 39 .dev_stripes = 1, 40 .devs_max = 0, /* 0 == as many as possible */ 41 .devs_min = 2, 42 .tolerated_failures = 1, 43 .devs_increment = 2, 44 .ncopies = 2, 45 .nparity = 0, 46 .raid_name = "raid10", 47 .bg_flag = BTRFS_BLOCK_GROUP_RAID10, 48 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, 49 }, 50 [BTRFS_RAID_RAID1] = { 51 .sub_stripes = 1, 52 .dev_stripes = 1, 53 .devs_max = 2, 54 .devs_min = 2, 55 .tolerated_failures = 1, 56 .devs_increment = 2, 57 .ncopies = 2, 58 .nparity = 0, 59 .raid_name = "raid1", 60 .bg_flag = BTRFS_BLOCK_GROUP_RAID1, 61 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, 62 }, 63 [BTRFS_RAID_RAID1C3] = { 64 .sub_stripes = 1, 65 .dev_stripes = 1, 66 .devs_max = 3, 67 .devs_min = 3, 68 .tolerated_failures = 2, 69 .devs_increment = 3, 70 .ncopies = 3, 71 .nparity = 0, 72 .raid_name = "raid1c3", 73 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, 74 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, 75 }, 76 [BTRFS_RAID_RAID1C4] = { 77 .sub_stripes = 1, 78 .dev_stripes = 1, 79 .devs_max = 4, 80 .devs_min = 4, 81 .tolerated_failures = 3, 82 .devs_increment = 4, 83 .ncopies = 4, 84 .nparity = 0, 85 .raid_name = "raid1c4", 86 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, 87 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, 88 }, 89 [BTRFS_RAID_DUP] = { 90 .sub_stripes = 1, 91 .dev_stripes = 2, 92 .devs_max = 1, 93 .devs_min = 1, 94 .tolerated_failures = 0, 95 .devs_increment = 1, 96 .ncopies = 2, 97 .nparity = 0, 98 .raid_name = "dup", 99 .bg_flag = BTRFS_BLOCK_GROUP_DUP, 100 .mindev_error = 0, 101 }, 102 [BTRFS_RAID_RAID0] = { 103 .sub_stripes = 1, 104 .dev_stripes = 1, 105 .devs_max = 0, 106 .devs_min = 1, 107 .tolerated_failures = 0, 108 .devs_increment = 1, 109 .ncopies = 1, 110 .nparity = 0, 111 .raid_name = "raid0", 112 .bg_flag = BTRFS_BLOCK_GROUP_RAID0, 113 .mindev_error = 0, 114 }, 115 [BTRFS_RAID_SINGLE] = { 116 .sub_stripes = 1, 117 .dev_stripes = 1, 118 .devs_max = 1, 119 .devs_min = 1, 120 .tolerated_failures = 0, 121 .devs_increment = 1, 122 .ncopies = 1, 123 .nparity = 0, 124 .raid_name = "single", 125 .bg_flag = 0, 126 .mindev_error = 0, 127 }, 128 [BTRFS_RAID_RAID5] = { 129 .sub_stripes = 1, 130 .dev_stripes = 1, 131 .devs_max = 0, 132 .devs_min = 2, 133 .tolerated_failures = 1, 134 .devs_increment = 1, 135 .ncopies = 1, 136 .nparity = 1, 137 .raid_name = "raid5", 138 .bg_flag = BTRFS_BLOCK_GROUP_RAID5, 139 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, 140 }, 141 [BTRFS_RAID_RAID6] = { 142 .sub_stripes = 1, 143 .dev_stripes = 1, 144 .devs_max = 0, 145 .devs_min = 3, 146 .tolerated_failures = 2, 147 .devs_increment = 1, 148 .ncopies = 1, 149 .nparity = 2, 150 .raid_name = "raid6", 151 .bg_flag = BTRFS_BLOCK_GROUP_RAID6, 152 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, 153 }, 154 }; 155 156 /* 157 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 158 * can be used as index to access btrfs_raid_array[]. 159 */ 160 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) 161 { 162 if (flags & BTRFS_BLOCK_GROUP_RAID10) 163 return BTRFS_RAID_RAID10; 164 else if (flags & BTRFS_BLOCK_GROUP_RAID1) 165 return BTRFS_RAID_RAID1; 166 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3) 167 return BTRFS_RAID_RAID1C3; 168 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4) 169 return BTRFS_RAID_RAID1C4; 170 else if (flags & BTRFS_BLOCK_GROUP_DUP) 171 return BTRFS_RAID_DUP; 172 else if (flags & BTRFS_BLOCK_GROUP_RAID0) 173 return BTRFS_RAID_RAID0; 174 else if (flags & BTRFS_BLOCK_GROUP_RAID5) 175 return BTRFS_RAID_RAID5; 176 else if (flags & BTRFS_BLOCK_GROUP_RAID6) 177 return BTRFS_RAID_RAID6; 178 179 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ 180 } 181 182 const char *btrfs_bg_type_to_raid_name(u64 flags) 183 { 184 const int index = btrfs_bg_flags_to_raid_index(flags); 185 186 if (index >= BTRFS_NR_RAID_TYPES) 187 return NULL; 188 189 return btrfs_raid_array[index].raid_name; 190 } 191 192 /* 193 * Fill @buf with textual description of @bg_flags, no more than @size_buf 194 * bytes including terminating null byte. 195 */ 196 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) 197 { 198 int i; 199 int ret; 200 char *bp = buf; 201 u64 flags = bg_flags; 202 u32 size_bp = size_buf; 203 204 if (!flags) { 205 strcpy(bp, "NONE"); 206 return; 207 } 208 209 #define DESCRIBE_FLAG(flag, desc) \ 210 do { \ 211 if (flags & (flag)) { \ 212 ret = snprintf(bp, size_bp, "%s|", (desc)); \ 213 if (ret < 0 || ret >= size_bp) \ 214 goto out_overflow; \ 215 size_bp -= ret; \ 216 bp += ret; \ 217 flags &= ~(flag); \ 218 } \ 219 } while (0) 220 221 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); 222 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); 223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); 224 225 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); 226 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) 227 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, 228 btrfs_raid_array[i].raid_name); 229 #undef DESCRIBE_FLAG 230 231 if (flags) { 232 ret = snprintf(bp, size_bp, "0x%llx|", flags); 233 size_bp -= ret; 234 } 235 236 if (size_bp < size_buf) 237 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ 238 239 /* 240 * The text is trimmed, it's up to the caller to provide sufficiently 241 * large buffer 242 */ 243 out_overflow:; 244 } 245 246 static int init_first_rw_device(struct btrfs_trans_handle *trans); 247 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); 248 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); 249 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); 250 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 251 enum btrfs_map_op op, 252 u64 logical, u64 *length, 253 struct btrfs_io_context **bioc_ret, 254 int mirror_num, int need_raid_map); 255 256 /* 257 * Device locking 258 * ============== 259 * 260 * There are several mutexes that protect manipulation of devices and low-level 261 * structures like chunks but not block groups, extents or files 262 * 263 * uuid_mutex (global lock) 264 * ------------------------ 265 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from 266 * the SCAN_DEV ioctl registration or from mount either implicitly (the first 267 * device) or requested by the device= mount option 268 * 269 * the mutex can be very coarse and can cover long-running operations 270 * 271 * protects: updates to fs_devices counters like missing devices, rw devices, 272 * seeding, structure cloning, opening/closing devices at mount/umount time 273 * 274 * global::fs_devs - add, remove, updates to the global list 275 * 276 * does not protect: manipulation of the fs_devices::devices list in general 277 * but in mount context it could be used to exclude list modifications by eg. 278 * scan ioctl 279 * 280 * btrfs_device::name - renames (write side), read is RCU 281 * 282 * fs_devices::device_list_mutex (per-fs, with RCU) 283 * ------------------------------------------------ 284 * protects updates to fs_devices::devices, ie. adding and deleting 285 * 286 * simple list traversal with read-only actions can be done with RCU protection 287 * 288 * may be used to exclude some operations from running concurrently without any 289 * modifications to the list (see write_all_supers) 290 * 291 * Is not required at mount and close times, because our device list is 292 * protected by the uuid_mutex at that point. 293 * 294 * balance_mutex 295 * ------------- 296 * protects balance structures (status, state) and context accessed from 297 * several places (internally, ioctl) 298 * 299 * chunk_mutex 300 * ----------- 301 * protects chunks, adding or removing during allocation, trim or when a new 302 * device is added/removed. Additionally it also protects post_commit_list of 303 * individual devices, since they can be added to the transaction's 304 * post_commit_list only with chunk_mutex held. 305 * 306 * cleaner_mutex 307 * ------------- 308 * a big lock that is held by the cleaner thread and prevents running subvolume 309 * cleaning together with relocation or delayed iputs 310 * 311 * 312 * Lock nesting 313 * ============ 314 * 315 * uuid_mutex 316 * device_list_mutex 317 * chunk_mutex 318 * balance_mutex 319 * 320 * 321 * Exclusive operations 322 * ==================== 323 * 324 * Maintains the exclusivity of the following operations that apply to the 325 * whole filesystem and cannot run in parallel. 326 * 327 * - Balance (*) 328 * - Device add 329 * - Device remove 330 * - Device replace (*) 331 * - Resize 332 * 333 * The device operations (as above) can be in one of the following states: 334 * 335 * - Running state 336 * - Paused state 337 * - Completed state 338 * 339 * Only device operations marked with (*) can go into the Paused state for the 340 * following reasons: 341 * 342 * - ioctl (only Balance can be Paused through ioctl) 343 * - filesystem remounted as read-only 344 * - filesystem unmounted and mounted as read-only 345 * - system power-cycle and filesystem mounted as read-only 346 * - filesystem or device errors leading to forced read-only 347 * 348 * The status of exclusive operation is set and cleared atomically. 349 * During the course of Paused state, fs_info::exclusive_operation remains set. 350 * A device operation in Paused or Running state can be canceled or resumed 351 * either by ioctl (Balance only) or when remounted as read-write. 352 * The exclusive status is cleared when the device operation is canceled or 353 * completed. 354 */ 355 356 DEFINE_MUTEX(uuid_mutex); 357 static LIST_HEAD(fs_uuids); 358 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) 359 { 360 return &fs_uuids; 361 } 362 363 /* 364 * alloc_fs_devices - allocate struct btrfs_fs_devices 365 * @fsid: if not NULL, copy the UUID to fs_devices::fsid 366 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid 367 * 368 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). 369 * The returned struct is not linked onto any lists and can be destroyed with 370 * kfree() right away. 371 */ 372 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, 373 const u8 *metadata_fsid) 374 { 375 struct btrfs_fs_devices *fs_devs; 376 377 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); 378 if (!fs_devs) 379 return ERR_PTR(-ENOMEM); 380 381 mutex_init(&fs_devs->device_list_mutex); 382 383 INIT_LIST_HEAD(&fs_devs->devices); 384 INIT_LIST_HEAD(&fs_devs->alloc_list); 385 INIT_LIST_HEAD(&fs_devs->fs_list); 386 INIT_LIST_HEAD(&fs_devs->seed_list); 387 if (fsid) 388 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); 389 390 if (metadata_fsid) 391 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); 392 else if (fsid) 393 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); 394 395 return fs_devs; 396 } 397 398 void btrfs_free_device(struct btrfs_device *device) 399 { 400 WARN_ON(!list_empty(&device->post_commit_list)); 401 rcu_string_free(device->name); 402 extent_io_tree_release(&device->alloc_state); 403 bio_put(device->flush_bio); 404 btrfs_destroy_dev_zone_info(device); 405 kfree(device); 406 } 407 408 static void free_fs_devices(struct btrfs_fs_devices *fs_devices) 409 { 410 struct btrfs_device *device; 411 WARN_ON(fs_devices->opened); 412 while (!list_empty(&fs_devices->devices)) { 413 device = list_entry(fs_devices->devices.next, 414 struct btrfs_device, dev_list); 415 list_del(&device->dev_list); 416 btrfs_free_device(device); 417 } 418 kfree(fs_devices); 419 } 420 421 void __exit btrfs_cleanup_fs_uuids(void) 422 { 423 struct btrfs_fs_devices *fs_devices; 424 425 while (!list_empty(&fs_uuids)) { 426 fs_devices = list_entry(fs_uuids.next, 427 struct btrfs_fs_devices, fs_list); 428 list_del(&fs_devices->fs_list); 429 free_fs_devices(fs_devices); 430 } 431 } 432 433 static noinline struct btrfs_fs_devices *find_fsid( 434 const u8 *fsid, const u8 *metadata_fsid) 435 { 436 struct btrfs_fs_devices *fs_devices; 437 438 ASSERT(fsid); 439 440 /* Handle non-split brain cases */ 441 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 442 if (metadata_fsid) { 443 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 444 && memcmp(metadata_fsid, fs_devices->metadata_uuid, 445 BTRFS_FSID_SIZE) == 0) 446 return fs_devices; 447 } else { 448 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) 449 return fs_devices; 450 } 451 } 452 return NULL; 453 } 454 455 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( 456 struct btrfs_super_block *disk_super) 457 { 458 459 struct btrfs_fs_devices *fs_devices; 460 461 /* 462 * Handle scanned device having completed its fsid change but 463 * belonging to a fs_devices that was created by first scanning 464 * a device which didn't have its fsid/metadata_uuid changed 465 * at all and the CHANGING_FSID_V2 flag set. 466 */ 467 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 468 if (fs_devices->fsid_change && 469 memcmp(disk_super->metadata_uuid, fs_devices->fsid, 470 BTRFS_FSID_SIZE) == 0 && 471 memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 472 BTRFS_FSID_SIZE) == 0) { 473 return fs_devices; 474 } 475 } 476 /* 477 * Handle scanned device having completed its fsid change but 478 * belonging to a fs_devices that was created by a device that 479 * has an outdated pair of fsid/metadata_uuid and 480 * CHANGING_FSID_V2 flag set. 481 */ 482 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 483 if (fs_devices->fsid_change && 484 memcmp(fs_devices->metadata_uuid, 485 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && 486 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, 487 BTRFS_FSID_SIZE) == 0) { 488 return fs_devices; 489 } 490 } 491 492 return find_fsid(disk_super->fsid, disk_super->metadata_uuid); 493 } 494 495 496 static int 497 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, 498 int flush, struct block_device **bdev, 499 struct btrfs_super_block **disk_super) 500 { 501 int ret; 502 503 *bdev = blkdev_get_by_path(device_path, flags, holder); 504 505 if (IS_ERR(*bdev)) { 506 ret = PTR_ERR(*bdev); 507 goto error; 508 } 509 510 if (flush) 511 filemap_write_and_wait((*bdev)->bd_inode->i_mapping); 512 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); 513 if (ret) { 514 blkdev_put(*bdev, flags); 515 goto error; 516 } 517 invalidate_bdev(*bdev); 518 *disk_super = btrfs_read_dev_super(*bdev); 519 if (IS_ERR(*disk_super)) { 520 ret = PTR_ERR(*disk_super); 521 blkdev_put(*bdev, flags); 522 goto error; 523 } 524 525 return 0; 526 527 error: 528 *bdev = NULL; 529 return ret; 530 } 531 532 static bool device_path_matched(const char *path, struct btrfs_device *device) 533 { 534 int found; 535 536 rcu_read_lock(); 537 found = strcmp(rcu_str_deref(device->name), path); 538 rcu_read_unlock(); 539 540 return found == 0; 541 } 542 543 /* 544 * Search and remove all stale (devices which are not mounted) devices. 545 * When both inputs are NULL, it will search and release all stale devices. 546 * path: Optional. When provided will it release all unmounted devices 547 * matching this path only. 548 * skip_dev: Optional. Will skip this device when searching for the stale 549 * devices. 550 * Return: 0 for success or if @path is NULL. 551 * -EBUSY if @path is a mounted device. 552 * -ENOENT if @path does not match any device in the list. 553 */ 554 static int btrfs_free_stale_devices(const char *path, 555 struct btrfs_device *skip_device) 556 { 557 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; 558 struct btrfs_device *device, *tmp_device; 559 int ret = 0; 560 561 lockdep_assert_held(&uuid_mutex); 562 563 if (path) 564 ret = -ENOENT; 565 566 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { 567 568 mutex_lock(&fs_devices->device_list_mutex); 569 list_for_each_entry_safe(device, tmp_device, 570 &fs_devices->devices, dev_list) { 571 if (skip_device && skip_device == device) 572 continue; 573 if (path && !device->name) 574 continue; 575 if (path && !device_path_matched(path, device)) 576 continue; 577 if (fs_devices->opened) { 578 /* for an already deleted device return 0 */ 579 if (path && ret != 0) 580 ret = -EBUSY; 581 break; 582 } 583 584 /* delete the stale device */ 585 fs_devices->num_devices--; 586 list_del(&device->dev_list); 587 btrfs_free_device(device); 588 589 ret = 0; 590 } 591 mutex_unlock(&fs_devices->device_list_mutex); 592 593 if (fs_devices->num_devices == 0) { 594 btrfs_sysfs_remove_fsid(fs_devices); 595 list_del(&fs_devices->fs_list); 596 free_fs_devices(fs_devices); 597 } 598 } 599 600 return ret; 601 } 602 603 /* 604 * This is only used on mount, and we are protected from competing things 605 * messing with our fs_devices by the uuid_mutex, thus we do not need the 606 * fs_devices->device_list_mutex here. 607 */ 608 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, 609 struct btrfs_device *device, fmode_t flags, 610 void *holder) 611 { 612 struct request_queue *q; 613 struct block_device *bdev; 614 struct btrfs_super_block *disk_super; 615 u64 devid; 616 int ret; 617 618 if (device->bdev) 619 return -EINVAL; 620 if (!device->name) 621 return -EINVAL; 622 623 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, 624 &bdev, &disk_super); 625 if (ret) 626 return ret; 627 628 devid = btrfs_stack_device_id(&disk_super->dev_item); 629 if (devid != device->devid) 630 goto error_free_page; 631 632 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) 633 goto error_free_page; 634 635 device->generation = btrfs_super_generation(disk_super); 636 637 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { 638 if (btrfs_super_incompat_flags(disk_super) & 639 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { 640 pr_err( 641 "BTRFS: Invalid seeding and uuid-changed device detected\n"); 642 goto error_free_page; 643 } 644 645 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 646 fs_devices->seeding = true; 647 } else { 648 if (bdev_read_only(bdev)) 649 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 650 else 651 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 652 } 653 654 q = bdev_get_queue(bdev); 655 if (!blk_queue_nonrot(q)) 656 fs_devices->rotating = true; 657 658 device->bdev = bdev; 659 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 660 device->mode = flags; 661 662 fs_devices->open_devices++; 663 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 664 device->devid != BTRFS_DEV_REPLACE_DEVID) { 665 fs_devices->rw_devices++; 666 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); 667 } 668 btrfs_release_disk_super(disk_super); 669 670 return 0; 671 672 error_free_page: 673 btrfs_release_disk_super(disk_super); 674 blkdev_put(bdev, flags); 675 676 return -EINVAL; 677 } 678 679 /* 680 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices 681 * being created with a disk that has already completed its fsid change. Such 682 * disk can belong to an fs which has its FSID changed or to one which doesn't. 683 * Handle both cases here. 684 */ 685 static struct btrfs_fs_devices *find_fsid_inprogress( 686 struct btrfs_super_block *disk_super) 687 { 688 struct btrfs_fs_devices *fs_devices; 689 690 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 691 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 692 BTRFS_FSID_SIZE) != 0 && 693 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 694 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { 695 return fs_devices; 696 } 697 } 698 699 return find_fsid(disk_super->fsid, NULL); 700 } 701 702 703 static struct btrfs_fs_devices *find_fsid_changed( 704 struct btrfs_super_block *disk_super) 705 { 706 struct btrfs_fs_devices *fs_devices; 707 708 /* 709 * Handles the case where scanned device is part of an fs that had 710 * multiple successful changes of FSID but currently device didn't 711 * observe it. Meaning our fsid will be different than theirs. We need 712 * to handle two subcases : 713 * 1 - The fs still continues to have different METADATA/FSID uuids. 714 * 2 - The fs is switched back to its original FSID (METADATA/FSID 715 * are equal). 716 */ 717 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 718 /* Changed UUIDs */ 719 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 720 BTRFS_FSID_SIZE) != 0 && 721 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, 722 BTRFS_FSID_SIZE) == 0 && 723 memcmp(fs_devices->fsid, disk_super->fsid, 724 BTRFS_FSID_SIZE) != 0) 725 return fs_devices; 726 727 /* Unchanged UUIDs */ 728 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, 729 BTRFS_FSID_SIZE) == 0 && 730 memcmp(fs_devices->fsid, disk_super->metadata_uuid, 731 BTRFS_FSID_SIZE) == 0) 732 return fs_devices; 733 } 734 735 return NULL; 736 } 737 738 static struct btrfs_fs_devices *find_fsid_reverted_metadata( 739 struct btrfs_super_block *disk_super) 740 { 741 struct btrfs_fs_devices *fs_devices; 742 743 /* 744 * Handle the case where the scanned device is part of an fs whose last 745 * metadata UUID change reverted it to the original FSID. At the same 746 * time * fs_devices was first created by another constitutent device 747 * which didn't fully observe the operation. This results in an 748 * btrfs_fs_devices created with metadata/fsid different AND 749 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the 750 * fs_devices equal to the FSID of the disk. 751 */ 752 list_for_each_entry(fs_devices, &fs_uuids, fs_list) { 753 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, 754 BTRFS_FSID_SIZE) != 0 && 755 memcmp(fs_devices->metadata_uuid, disk_super->fsid, 756 BTRFS_FSID_SIZE) == 0 && 757 fs_devices->fsid_change) 758 return fs_devices; 759 } 760 761 return NULL; 762 } 763 /* 764 * Add new device to list of registered devices 765 * 766 * Returns: 767 * device pointer which was just added or updated when successful 768 * error pointer when failed 769 */ 770 static noinline struct btrfs_device *device_list_add(const char *path, 771 struct btrfs_super_block *disk_super, 772 bool *new_device_added) 773 { 774 struct btrfs_device *device; 775 struct btrfs_fs_devices *fs_devices = NULL; 776 struct rcu_string *name; 777 u64 found_transid = btrfs_super_generation(disk_super); 778 u64 devid = btrfs_stack_device_id(&disk_super->dev_item); 779 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & 780 BTRFS_FEATURE_INCOMPAT_METADATA_UUID); 781 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & 782 BTRFS_SUPER_FLAG_CHANGING_FSID_V2); 783 784 if (fsid_change_in_progress) { 785 if (!has_metadata_uuid) 786 fs_devices = find_fsid_inprogress(disk_super); 787 else 788 fs_devices = find_fsid_changed(disk_super); 789 } else if (has_metadata_uuid) { 790 fs_devices = find_fsid_with_metadata_uuid(disk_super); 791 } else { 792 fs_devices = find_fsid_reverted_metadata(disk_super); 793 if (!fs_devices) 794 fs_devices = find_fsid(disk_super->fsid, NULL); 795 } 796 797 798 if (!fs_devices) { 799 if (has_metadata_uuid) 800 fs_devices = alloc_fs_devices(disk_super->fsid, 801 disk_super->metadata_uuid); 802 else 803 fs_devices = alloc_fs_devices(disk_super->fsid, NULL); 804 805 if (IS_ERR(fs_devices)) 806 return ERR_CAST(fs_devices); 807 808 fs_devices->fsid_change = fsid_change_in_progress; 809 810 mutex_lock(&fs_devices->device_list_mutex); 811 list_add(&fs_devices->fs_list, &fs_uuids); 812 813 device = NULL; 814 } else { 815 struct btrfs_dev_lookup_args args = { 816 .devid = devid, 817 .uuid = disk_super->dev_item.uuid, 818 }; 819 820 mutex_lock(&fs_devices->device_list_mutex); 821 device = btrfs_find_device(fs_devices, &args); 822 823 /* 824 * If this disk has been pulled into an fs devices created by 825 * a device which had the CHANGING_FSID_V2 flag then replace the 826 * metadata_uuid/fsid values of the fs_devices. 827 */ 828 if (fs_devices->fsid_change && 829 found_transid > fs_devices->latest_generation) { 830 memcpy(fs_devices->fsid, disk_super->fsid, 831 BTRFS_FSID_SIZE); 832 833 if (has_metadata_uuid) 834 memcpy(fs_devices->metadata_uuid, 835 disk_super->metadata_uuid, 836 BTRFS_FSID_SIZE); 837 else 838 memcpy(fs_devices->metadata_uuid, 839 disk_super->fsid, BTRFS_FSID_SIZE); 840 841 fs_devices->fsid_change = false; 842 } 843 } 844 845 if (!device) { 846 if (fs_devices->opened) { 847 mutex_unlock(&fs_devices->device_list_mutex); 848 return ERR_PTR(-EBUSY); 849 } 850 851 device = btrfs_alloc_device(NULL, &devid, 852 disk_super->dev_item.uuid); 853 if (IS_ERR(device)) { 854 mutex_unlock(&fs_devices->device_list_mutex); 855 /* we can safely leave the fs_devices entry around */ 856 return device; 857 } 858 859 name = rcu_string_strdup(path, GFP_NOFS); 860 if (!name) { 861 btrfs_free_device(device); 862 mutex_unlock(&fs_devices->device_list_mutex); 863 return ERR_PTR(-ENOMEM); 864 } 865 rcu_assign_pointer(device->name, name); 866 867 list_add_rcu(&device->dev_list, &fs_devices->devices); 868 fs_devices->num_devices++; 869 870 device->fs_devices = fs_devices; 871 *new_device_added = true; 872 873 if (disk_super->label[0]) 874 pr_info( 875 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", 876 disk_super->label, devid, found_transid, path, 877 current->comm, task_pid_nr(current)); 878 else 879 pr_info( 880 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", 881 disk_super->fsid, devid, found_transid, path, 882 current->comm, task_pid_nr(current)); 883 884 } else if (!device->name || strcmp(device->name->str, path)) { 885 /* 886 * When FS is already mounted. 887 * 1. If you are here and if the device->name is NULL that 888 * means this device was missing at time of FS mount. 889 * 2. If you are here and if the device->name is different 890 * from 'path' that means either 891 * a. The same device disappeared and reappeared with 892 * different name. or 893 * b. The missing-disk-which-was-replaced, has 894 * reappeared now. 895 * 896 * We must allow 1 and 2a above. But 2b would be a spurious 897 * and unintentional. 898 * 899 * Further in case of 1 and 2a above, the disk at 'path' 900 * would have missed some transaction when it was away and 901 * in case of 2a the stale bdev has to be updated as well. 902 * 2b must not be allowed at all time. 903 */ 904 905 /* 906 * For now, we do allow update to btrfs_fs_device through the 907 * btrfs dev scan cli after FS has been mounted. We're still 908 * tracking a problem where systems fail mount by subvolume id 909 * when we reject replacement on a mounted FS. 910 */ 911 if (!fs_devices->opened && found_transid < device->generation) { 912 /* 913 * That is if the FS is _not_ mounted and if you 914 * are here, that means there is more than one 915 * disk with same uuid and devid.We keep the one 916 * with larger generation number or the last-in if 917 * generation are equal. 918 */ 919 mutex_unlock(&fs_devices->device_list_mutex); 920 return ERR_PTR(-EEXIST); 921 } 922 923 /* 924 * We are going to replace the device path for a given devid, 925 * make sure it's the same device if the device is mounted 926 */ 927 if (device->bdev) { 928 int error; 929 dev_t path_dev; 930 931 error = lookup_bdev(path, &path_dev); 932 if (error) { 933 mutex_unlock(&fs_devices->device_list_mutex); 934 return ERR_PTR(error); 935 } 936 937 if (device->bdev->bd_dev != path_dev) { 938 mutex_unlock(&fs_devices->device_list_mutex); 939 /* 940 * device->fs_info may not be reliable here, so 941 * pass in a NULL instead. This avoids a 942 * possible use-after-free when the fs_info and 943 * fs_info->sb are already torn down. 944 */ 945 btrfs_warn_in_rcu(NULL, 946 "duplicate device %s devid %llu generation %llu scanned by %s (%d)", 947 path, devid, found_transid, 948 current->comm, 949 task_pid_nr(current)); 950 return ERR_PTR(-EEXIST); 951 } 952 btrfs_info_in_rcu(device->fs_info, 953 "devid %llu device path %s changed to %s scanned by %s (%d)", 954 devid, rcu_str_deref(device->name), 955 path, current->comm, 956 task_pid_nr(current)); 957 } 958 959 name = rcu_string_strdup(path, GFP_NOFS); 960 if (!name) { 961 mutex_unlock(&fs_devices->device_list_mutex); 962 return ERR_PTR(-ENOMEM); 963 } 964 rcu_string_free(device->name); 965 rcu_assign_pointer(device->name, name); 966 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 967 fs_devices->missing_devices--; 968 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 969 } 970 } 971 972 /* 973 * Unmount does not free the btrfs_device struct but would zero 974 * generation along with most of the other members. So just update 975 * it back. We need it to pick the disk with largest generation 976 * (as above). 977 */ 978 if (!fs_devices->opened) { 979 device->generation = found_transid; 980 fs_devices->latest_generation = max_t(u64, found_transid, 981 fs_devices->latest_generation); 982 } 983 984 fs_devices->total_devices = btrfs_super_num_devices(disk_super); 985 986 mutex_unlock(&fs_devices->device_list_mutex); 987 return device; 988 } 989 990 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) 991 { 992 struct btrfs_fs_devices *fs_devices; 993 struct btrfs_device *device; 994 struct btrfs_device *orig_dev; 995 int ret = 0; 996 997 lockdep_assert_held(&uuid_mutex); 998 999 fs_devices = alloc_fs_devices(orig->fsid, NULL); 1000 if (IS_ERR(fs_devices)) 1001 return fs_devices; 1002 1003 fs_devices->total_devices = orig->total_devices; 1004 1005 list_for_each_entry(orig_dev, &orig->devices, dev_list) { 1006 struct rcu_string *name; 1007 1008 device = btrfs_alloc_device(NULL, &orig_dev->devid, 1009 orig_dev->uuid); 1010 if (IS_ERR(device)) { 1011 ret = PTR_ERR(device); 1012 goto error; 1013 } 1014 1015 /* 1016 * This is ok to do without rcu read locked because we hold the 1017 * uuid mutex so nothing we touch in here is going to disappear. 1018 */ 1019 if (orig_dev->name) { 1020 name = rcu_string_strdup(orig_dev->name->str, 1021 GFP_KERNEL); 1022 if (!name) { 1023 btrfs_free_device(device); 1024 ret = -ENOMEM; 1025 goto error; 1026 } 1027 rcu_assign_pointer(device->name, name); 1028 } 1029 1030 list_add(&device->dev_list, &fs_devices->devices); 1031 device->fs_devices = fs_devices; 1032 fs_devices->num_devices++; 1033 } 1034 return fs_devices; 1035 error: 1036 free_fs_devices(fs_devices); 1037 return ERR_PTR(ret); 1038 } 1039 1040 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, 1041 struct btrfs_device **latest_dev) 1042 { 1043 struct btrfs_device *device, *next; 1044 1045 /* This is the initialized path, it is safe to release the devices. */ 1046 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 1047 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { 1048 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, 1049 &device->dev_state) && 1050 !test_bit(BTRFS_DEV_STATE_MISSING, 1051 &device->dev_state) && 1052 (!*latest_dev || 1053 device->generation > (*latest_dev)->generation)) { 1054 *latest_dev = device; 1055 } 1056 continue; 1057 } 1058 1059 /* 1060 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, 1061 * in btrfs_init_dev_replace() so just continue. 1062 */ 1063 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1064 continue; 1065 1066 if (device->bdev) { 1067 blkdev_put(device->bdev, device->mode); 1068 device->bdev = NULL; 1069 fs_devices->open_devices--; 1070 } 1071 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1072 list_del_init(&device->dev_alloc_list); 1073 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1074 fs_devices->rw_devices--; 1075 } 1076 list_del_init(&device->dev_list); 1077 fs_devices->num_devices--; 1078 btrfs_free_device(device); 1079 } 1080 1081 } 1082 1083 /* 1084 * After we have read the system tree and know devids belonging to this 1085 * filesystem, remove the device which does not belong there. 1086 */ 1087 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) 1088 { 1089 struct btrfs_device *latest_dev = NULL; 1090 struct btrfs_fs_devices *seed_dev; 1091 1092 mutex_lock(&uuid_mutex); 1093 __btrfs_free_extra_devids(fs_devices, &latest_dev); 1094 1095 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) 1096 __btrfs_free_extra_devids(seed_dev, &latest_dev); 1097 1098 fs_devices->latest_dev = latest_dev; 1099 1100 mutex_unlock(&uuid_mutex); 1101 } 1102 1103 static void btrfs_close_bdev(struct btrfs_device *device) 1104 { 1105 if (!device->bdev) 1106 return; 1107 1108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 1109 sync_blockdev(device->bdev); 1110 invalidate_bdev(device->bdev); 1111 } 1112 1113 blkdev_put(device->bdev, device->mode); 1114 } 1115 1116 static void btrfs_close_one_device(struct btrfs_device *device) 1117 { 1118 struct btrfs_fs_devices *fs_devices = device->fs_devices; 1119 1120 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 1121 device->devid != BTRFS_DEV_REPLACE_DEVID) { 1122 list_del_init(&device->dev_alloc_list); 1123 fs_devices->rw_devices--; 1124 } 1125 1126 if (device->devid == BTRFS_DEV_REPLACE_DEVID) 1127 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 1128 1129 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 1130 fs_devices->missing_devices--; 1131 1132 btrfs_close_bdev(device); 1133 if (device->bdev) { 1134 fs_devices->open_devices--; 1135 device->bdev = NULL; 1136 } 1137 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 1138 btrfs_destroy_dev_zone_info(device); 1139 1140 device->fs_info = NULL; 1141 atomic_set(&device->dev_stats_ccnt, 0); 1142 extent_io_tree_release(&device->alloc_state); 1143 1144 /* 1145 * Reset the flush error record. We might have a transient flush error 1146 * in this mount, and if so we aborted the current transaction and set 1147 * the fs to an error state, guaranteeing no super blocks can be further 1148 * committed. However that error might be transient and if we unmount the 1149 * filesystem and mount it again, we should allow the mount to succeed 1150 * (btrfs_check_rw_degradable() should not fail) - if after mounting the 1151 * filesystem again we still get flush errors, then we will again abort 1152 * any transaction and set the error state, guaranteeing no commits of 1153 * unsafe super blocks. 1154 */ 1155 device->last_flush_error = 0; 1156 1157 /* Verify the device is back in a pristine state */ 1158 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); 1159 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); 1160 ASSERT(list_empty(&device->dev_alloc_list)); 1161 ASSERT(list_empty(&device->post_commit_list)); 1162 ASSERT(atomic_read(&device->reada_in_flight) == 0); 1163 } 1164 1165 static void close_fs_devices(struct btrfs_fs_devices *fs_devices) 1166 { 1167 struct btrfs_device *device, *tmp; 1168 1169 lockdep_assert_held(&uuid_mutex); 1170 1171 if (--fs_devices->opened > 0) 1172 return; 1173 1174 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) 1175 btrfs_close_one_device(device); 1176 1177 WARN_ON(fs_devices->open_devices); 1178 WARN_ON(fs_devices->rw_devices); 1179 fs_devices->opened = 0; 1180 fs_devices->seeding = false; 1181 fs_devices->fs_info = NULL; 1182 } 1183 1184 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) 1185 { 1186 LIST_HEAD(list); 1187 struct btrfs_fs_devices *tmp; 1188 1189 mutex_lock(&uuid_mutex); 1190 close_fs_devices(fs_devices); 1191 if (!fs_devices->opened) 1192 list_splice_init(&fs_devices->seed_list, &list); 1193 1194 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { 1195 close_fs_devices(fs_devices); 1196 list_del(&fs_devices->seed_list); 1197 free_fs_devices(fs_devices); 1198 } 1199 mutex_unlock(&uuid_mutex); 1200 } 1201 1202 static int open_fs_devices(struct btrfs_fs_devices *fs_devices, 1203 fmode_t flags, void *holder) 1204 { 1205 struct btrfs_device *device; 1206 struct btrfs_device *latest_dev = NULL; 1207 struct btrfs_device *tmp_device; 1208 1209 flags |= FMODE_EXCL; 1210 1211 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, 1212 dev_list) { 1213 int ret; 1214 1215 ret = btrfs_open_one_device(fs_devices, device, flags, holder); 1216 if (ret == 0 && 1217 (!latest_dev || device->generation > latest_dev->generation)) { 1218 latest_dev = device; 1219 } else if (ret == -ENODATA) { 1220 fs_devices->num_devices--; 1221 list_del(&device->dev_list); 1222 btrfs_free_device(device); 1223 } 1224 } 1225 if (fs_devices->open_devices == 0) 1226 return -EINVAL; 1227 1228 fs_devices->opened = 1; 1229 fs_devices->latest_dev = latest_dev; 1230 fs_devices->total_rw_bytes = 0; 1231 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; 1232 fs_devices->read_policy = BTRFS_READ_POLICY_PID; 1233 1234 return 0; 1235 } 1236 1237 static int devid_cmp(void *priv, const struct list_head *a, 1238 const struct list_head *b) 1239 { 1240 const struct btrfs_device *dev1, *dev2; 1241 1242 dev1 = list_entry(a, struct btrfs_device, dev_list); 1243 dev2 = list_entry(b, struct btrfs_device, dev_list); 1244 1245 if (dev1->devid < dev2->devid) 1246 return -1; 1247 else if (dev1->devid > dev2->devid) 1248 return 1; 1249 return 0; 1250 } 1251 1252 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 1253 fmode_t flags, void *holder) 1254 { 1255 int ret; 1256 1257 lockdep_assert_held(&uuid_mutex); 1258 /* 1259 * The device_list_mutex cannot be taken here in case opening the 1260 * underlying device takes further locks like open_mutex. 1261 * 1262 * We also don't need the lock here as this is called during mount and 1263 * exclusion is provided by uuid_mutex 1264 */ 1265 1266 if (fs_devices->opened) { 1267 fs_devices->opened++; 1268 ret = 0; 1269 } else { 1270 list_sort(NULL, &fs_devices->devices, devid_cmp); 1271 ret = open_fs_devices(fs_devices, flags, holder); 1272 } 1273 1274 return ret; 1275 } 1276 1277 void btrfs_release_disk_super(struct btrfs_super_block *super) 1278 { 1279 struct page *page = virt_to_page(super); 1280 1281 put_page(page); 1282 } 1283 1284 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, 1285 u64 bytenr, u64 bytenr_orig) 1286 { 1287 struct btrfs_super_block *disk_super; 1288 struct page *page; 1289 void *p; 1290 pgoff_t index; 1291 1292 /* make sure our super fits in the device */ 1293 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) 1294 return ERR_PTR(-EINVAL); 1295 1296 /* make sure our super fits in the page */ 1297 if (sizeof(*disk_super) > PAGE_SIZE) 1298 return ERR_PTR(-EINVAL); 1299 1300 /* make sure our super doesn't straddle pages on disk */ 1301 index = bytenr >> PAGE_SHIFT; 1302 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) 1303 return ERR_PTR(-EINVAL); 1304 1305 /* pull in the page with our super */ 1306 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); 1307 1308 if (IS_ERR(page)) 1309 return ERR_CAST(page); 1310 1311 p = page_address(page); 1312 1313 /* align our pointer to the offset of the super block */ 1314 disk_super = p + offset_in_page(bytenr); 1315 1316 if (btrfs_super_bytenr(disk_super) != bytenr_orig || 1317 btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 1318 btrfs_release_disk_super(p); 1319 return ERR_PTR(-EINVAL); 1320 } 1321 1322 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) 1323 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; 1324 1325 return disk_super; 1326 } 1327 1328 int btrfs_forget_devices(const char *path) 1329 { 1330 int ret; 1331 1332 mutex_lock(&uuid_mutex); 1333 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); 1334 mutex_unlock(&uuid_mutex); 1335 1336 return ret; 1337 } 1338 1339 /* 1340 * Look for a btrfs signature on a device. This may be called out of the mount path 1341 * and we are not allowed to call set_blocksize during the scan. The superblock 1342 * is read via pagecache 1343 */ 1344 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, 1345 void *holder) 1346 { 1347 struct btrfs_super_block *disk_super; 1348 bool new_device_added = false; 1349 struct btrfs_device *device = NULL; 1350 struct block_device *bdev; 1351 u64 bytenr, bytenr_orig; 1352 int ret; 1353 1354 lockdep_assert_held(&uuid_mutex); 1355 1356 /* 1357 * we would like to check all the supers, but that would make 1358 * a btrfs mount succeed after a mkfs from a different FS. 1359 * So, we need to add a special mount option to scan for 1360 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 1361 */ 1362 flags |= FMODE_EXCL; 1363 1364 bdev = blkdev_get_by_path(path, flags, holder); 1365 if (IS_ERR(bdev)) 1366 return ERR_CAST(bdev); 1367 1368 bytenr_orig = btrfs_sb_offset(0); 1369 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); 1370 if (ret) 1371 return ERR_PTR(ret); 1372 1373 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); 1374 if (IS_ERR(disk_super)) { 1375 device = ERR_CAST(disk_super); 1376 goto error_bdev_put; 1377 } 1378 1379 device = device_list_add(path, disk_super, &new_device_added); 1380 if (!IS_ERR(device)) { 1381 if (new_device_added) 1382 btrfs_free_stale_devices(path, device); 1383 } 1384 1385 btrfs_release_disk_super(disk_super); 1386 1387 error_bdev_put: 1388 blkdev_put(bdev, flags); 1389 1390 return device; 1391 } 1392 1393 /* 1394 * Try to find a chunk that intersects [start, start + len] range and when one 1395 * such is found, record the end of it in *start 1396 */ 1397 static bool contains_pending_extent(struct btrfs_device *device, u64 *start, 1398 u64 len) 1399 { 1400 u64 physical_start, physical_end; 1401 1402 lockdep_assert_held(&device->fs_info->chunk_mutex); 1403 1404 if (!find_first_extent_bit(&device->alloc_state, *start, 1405 &physical_start, &physical_end, 1406 CHUNK_ALLOCATED, NULL)) { 1407 1408 if (in_range(physical_start, *start, len) || 1409 in_range(*start, physical_start, 1410 physical_end - physical_start)) { 1411 *start = physical_end + 1; 1412 return true; 1413 } 1414 } 1415 return false; 1416 } 1417 1418 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) 1419 { 1420 switch (device->fs_devices->chunk_alloc_policy) { 1421 case BTRFS_CHUNK_ALLOC_REGULAR: 1422 /* 1423 * We don't want to overwrite the superblock on the drive nor 1424 * any area used by the boot loader (grub for example), so we 1425 * make sure to start at an offset of at least 1MB. 1426 */ 1427 return max_t(u64, start, SZ_1M); 1428 case BTRFS_CHUNK_ALLOC_ZONED: 1429 /* 1430 * We don't care about the starting region like regular 1431 * allocator, because we anyway use/reserve the first two zones 1432 * for superblock logging. 1433 */ 1434 return ALIGN(start, device->zone_info->zone_size); 1435 default: 1436 BUG(); 1437 } 1438 } 1439 1440 static bool dev_extent_hole_check_zoned(struct btrfs_device *device, 1441 u64 *hole_start, u64 *hole_size, 1442 u64 num_bytes) 1443 { 1444 u64 zone_size = device->zone_info->zone_size; 1445 u64 pos; 1446 int ret; 1447 bool changed = false; 1448 1449 ASSERT(IS_ALIGNED(*hole_start, zone_size)); 1450 1451 while (*hole_size > 0) { 1452 pos = btrfs_find_allocatable_zones(device, *hole_start, 1453 *hole_start + *hole_size, 1454 num_bytes); 1455 if (pos != *hole_start) { 1456 *hole_size = *hole_start + *hole_size - pos; 1457 *hole_start = pos; 1458 changed = true; 1459 if (*hole_size < num_bytes) 1460 break; 1461 } 1462 1463 ret = btrfs_ensure_empty_zones(device, pos, num_bytes); 1464 1465 /* Range is ensured to be empty */ 1466 if (!ret) 1467 return changed; 1468 1469 /* Given hole range was invalid (outside of device) */ 1470 if (ret == -ERANGE) { 1471 *hole_start += *hole_size; 1472 *hole_size = 0; 1473 return true; 1474 } 1475 1476 *hole_start += zone_size; 1477 *hole_size -= zone_size; 1478 changed = true; 1479 } 1480 1481 return changed; 1482 } 1483 1484 /** 1485 * dev_extent_hole_check - check if specified hole is suitable for allocation 1486 * @device: the device which we have the hole 1487 * @hole_start: starting position of the hole 1488 * @hole_size: the size of the hole 1489 * @num_bytes: the size of the free space that we need 1490 * 1491 * This function may modify @hole_start and @hole_size to reflect the suitable 1492 * position for allocation. Returns 1 if hole position is updated, 0 otherwise. 1493 */ 1494 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, 1495 u64 *hole_size, u64 num_bytes) 1496 { 1497 bool changed = false; 1498 u64 hole_end = *hole_start + *hole_size; 1499 1500 for (;;) { 1501 /* 1502 * Check before we set max_hole_start, otherwise we could end up 1503 * sending back this offset anyway. 1504 */ 1505 if (contains_pending_extent(device, hole_start, *hole_size)) { 1506 if (hole_end >= *hole_start) 1507 *hole_size = hole_end - *hole_start; 1508 else 1509 *hole_size = 0; 1510 changed = true; 1511 } 1512 1513 switch (device->fs_devices->chunk_alloc_policy) { 1514 case BTRFS_CHUNK_ALLOC_REGULAR: 1515 /* No extra check */ 1516 break; 1517 case BTRFS_CHUNK_ALLOC_ZONED: 1518 if (dev_extent_hole_check_zoned(device, hole_start, 1519 hole_size, num_bytes)) { 1520 changed = true; 1521 /* 1522 * The changed hole can contain pending extent. 1523 * Loop again to check that. 1524 */ 1525 continue; 1526 } 1527 break; 1528 default: 1529 BUG(); 1530 } 1531 1532 break; 1533 } 1534 1535 return changed; 1536 } 1537 1538 /* 1539 * find_free_dev_extent_start - find free space in the specified device 1540 * @device: the device which we search the free space in 1541 * @num_bytes: the size of the free space that we need 1542 * @search_start: the position from which to begin the search 1543 * @start: store the start of the free space. 1544 * @len: the size of the free space. that we find, or the size 1545 * of the max free space if we don't find suitable free space 1546 * 1547 * this uses a pretty simple search, the expectation is that it is 1548 * called very infrequently and that a given device has a small number 1549 * of extents 1550 * 1551 * @start is used to store the start of the free space if we find. But if we 1552 * don't find suitable free space, it will be used to store the start position 1553 * of the max free space. 1554 * 1555 * @len is used to store the size of the free space that we find. 1556 * But if we don't find suitable free space, it is used to store the size of 1557 * the max free space. 1558 * 1559 * NOTE: This function will search *commit* root of device tree, and does extra 1560 * check to ensure dev extents are not double allocated. 1561 * This makes the function safe to allocate dev extents but may not report 1562 * correct usable device space, as device extent freed in current transaction 1563 * is not reported as available. 1564 */ 1565 static int find_free_dev_extent_start(struct btrfs_device *device, 1566 u64 num_bytes, u64 search_start, u64 *start, 1567 u64 *len) 1568 { 1569 struct btrfs_fs_info *fs_info = device->fs_info; 1570 struct btrfs_root *root = fs_info->dev_root; 1571 struct btrfs_key key; 1572 struct btrfs_dev_extent *dev_extent; 1573 struct btrfs_path *path; 1574 u64 hole_size; 1575 u64 max_hole_start; 1576 u64 max_hole_size; 1577 u64 extent_end; 1578 u64 search_end = device->total_bytes; 1579 int ret; 1580 int slot; 1581 struct extent_buffer *l; 1582 1583 search_start = dev_extent_search_start(device, search_start); 1584 1585 WARN_ON(device->zone_info && 1586 !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); 1587 1588 path = btrfs_alloc_path(); 1589 if (!path) 1590 return -ENOMEM; 1591 1592 max_hole_start = search_start; 1593 max_hole_size = 0; 1594 1595 again: 1596 if (search_start >= search_end || 1597 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1598 ret = -ENOSPC; 1599 goto out; 1600 } 1601 1602 path->reada = READA_FORWARD; 1603 path->search_commit_root = 1; 1604 path->skip_locking = 1; 1605 1606 key.objectid = device->devid; 1607 key.offset = search_start; 1608 key.type = BTRFS_DEV_EXTENT_KEY; 1609 1610 ret = btrfs_search_backwards(root, &key, path); 1611 if (ret < 0) 1612 goto out; 1613 1614 while (1) { 1615 l = path->nodes[0]; 1616 slot = path->slots[0]; 1617 if (slot >= btrfs_header_nritems(l)) { 1618 ret = btrfs_next_leaf(root, path); 1619 if (ret == 0) 1620 continue; 1621 if (ret < 0) 1622 goto out; 1623 1624 break; 1625 } 1626 btrfs_item_key_to_cpu(l, &key, slot); 1627 1628 if (key.objectid < device->devid) 1629 goto next; 1630 1631 if (key.objectid > device->devid) 1632 break; 1633 1634 if (key.type != BTRFS_DEV_EXTENT_KEY) 1635 goto next; 1636 1637 if (key.offset > search_start) { 1638 hole_size = key.offset - search_start; 1639 dev_extent_hole_check(device, &search_start, &hole_size, 1640 num_bytes); 1641 1642 if (hole_size > max_hole_size) { 1643 max_hole_start = search_start; 1644 max_hole_size = hole_size; 1645 } 1646 1647 /* 1648 * If this free space is greater than which we need, 1649 * it must be the max free space that we have found 1650 * until now, so max_hole_start must point to the start 1651 * of this free space and the length of this free space 1652 * is stored in max_hole_size. Thus, we return 1653 * max_hole_start and max_hole_size and go back to the 1654 * caller. 1655 */ 1656 if (hole_size >= num_bytes) { 1657 ret = 0; 1658 goto out; 1659 } 1660 } 1661 1662 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 1663 extent_end = key.offset + btrfs_dev_extent_length(l, 1664 dev_extent); 1665 if (extent_end > search_start) 1666 search_start = extent_end; 1667 next: 1668 path->slots[0]++; 1669 cond_resched(); 1670 } 1671 1672 /* 1673 * At this point, search_start should be the end of 1674 * allocated dev extents, and when shrinking the device, 1675 * search_end may be smaller than search_start. 1676 */ 1677 if (search_end > search_start) { 1678 hole_size = search_end - search_start; 1679 if (dev_extent_hole_check(device, &search_start, &hole_size, 1680 num_bytes)) { 1681 btrfs_release_path(path); 1682 goto again; 1683 } 1684 1685 if (hole_size > max_hole_size) { 1686 max_hole_start = search_start; 1687 max_hole_size = hole_size; 1688 } 1689 } 1690 1691 /* See above. */ 1692 if (max_hole_size < num_bytes) 1693 ret = -ENOSPC; 1694 else 1695 ret = 0; 1696 1697 out: 1698 btrfs_free_path(path); 1699 *start = max_hole_start; 1700 if (len) 1701 *len = max_hole_size; 1702 return ret; 1703 } 1704 1705 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 1706 u64 *start, u64 *len) 1707 { 1708 /* FIXME use last free of some kind */ 1709 return find_free_dev_extent_start(device, num_bytes, 0, start, len); 1710 } 1711 1712 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, 1713 struct btrfs_device *device, 1714 u64 start, u64 *dev_extent_len) 1715 { 1716 struct btrfs_fs_info *fs_info = device->fs_info; 1717 struct btrfs_root *root = fs_info->dev_root; 1718 int ret; 1719 struct btrfs_path *path; 1720 struct btrfs_key key; 1721 struct btrfs_key found_key; 1722 struct extent_buffer *leaf = NULL; 1723 struct btrfs_dev_extent *extent = NULL; 1724 1725 path = btrfs_alloc_path(); 1726 if (!path) 1727 return -ENOMEM; 1728 1729 key.objectid = device->devid; 1730 key.offset = start; 1731 key.type = BTRFS_DEV_EXTENT_KEY; 1732 again: 1733 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1734 if (ret > 0) { 1735 ret = btrfs_previous_item(root, path, key.objectid, 1736 BTRFS_DEV_EXTENT_KEY); 1737 if (ret) 1738 goto out; 1739 leaf = path->nodes[0]; 1740 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 1741 extent = btrfs_item_ptr(leaf, path->slots[0], 1742 struct btrfs_dev_extent); 1743 BUG_ON(found_key.offset > start || found_key.offset + 1744 btrfs_dev_extent_length(leaf, extent) < start); 1745 key = found_key; 1746 btrfs_release_path(path); 1747 goto again; 1748 } else if (ret == 0) { 1749 leaf = path->nodes[0]; 1750 extent = btrfs_item_ptr(leaf, path->slots[0], 1751 struct btrfs_dev_extent); 1752 } else { 1753 goto out; 1754 } 1755 1756 *dev_extent_len = btrfs_dev_extent_length(leaf, extent); 1757 1758 ret = btrfs_del_item(trans, root, path); 1759 if (ret == 0) 1760 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); 1761 out: 1762 btrfs_free_path(path); 1763 return ret; 1764 } 1765 1766 static u64 find_next_chunk(struct btrfs_fs_info *fs_info) 1767 { 1768 struct extent_map_tree *em_tree; 1769 struct extent_map *em; 1770 struct rb_node *n; 1771 u64 ret = 0; 1772 1773 em_tree = &fs_info->mapping_tree; 1774 read_lock(&em_tree->lock); 1775 n = rb_last(&em_tree->map.rb_root); 1776 if (n) { 1777 em = rb_entry(n, struct extent_map, rb_node); 1778 ret = em->start + em->len; 1779 } 1780 read_unlock(&em_tree->lock); 1781 1782 return ret; 1783 } 1784 1785 static noinline int find_next_devid(struct btrfs_fs_info *fs_info, 1786 u64 *devid_ret) 1787 { 1788 int ret; 1789 struct btrfs_key key; 1790 struct btrfs_key found_key; 1791 struct btrfs_path *path; 1792 1793 path = btrfs_alloc_path(); 1794 if (!path) 1795 return -ENOMEM; 1796 1797 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1798 key.type = BTRFS_DEV_ITEM_KEY; 1799 key.offset = (u64)-1; 1800 1801 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); 1802 if (ret < 0) 1803 goto error; 1804 1805 if (ret == 0) { 1806 /* Corruption */ 1807 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); 1808 ret = -EUCLEAN; 1809 goto error; 1810 } 1811 1812 ret = btrfs_previous_item(fs_info->chunk_root, path, 1813 BTRFS_DEV_ITEMS_OBJECTID, 1814 BTRFS_DEV_ITEM_KEY); 1815 if (ret) { 1816 *devid_ret = 1; 1817 } else { 1818 btrfs_item_key_to_cpu(path->nodes[0], &found_key, 1819 path->slots[0]); 1820 *devid_ret = found_key.offset + 1; 1821 } 1822 ret = 0; 1823 error: 1824 btrfs_free_path(path); 1825 return ret; 1826 } 1827 1828 /* 1829 * the device information is stored in the chunk root 1830 * the btrfs_device struct should be fully filled in 1831 */ 1832 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, 1833 struct btrfs_device *device) 1834 { 1835 int ret; 1836 struct btrfs_path *path; 1837 struct btrfs_dev_item *dev_item; 1838 struct extent_buffer *leaf; 1839 struct btrfs_key key; 1840 unsigned long ptr; 1841 1842 path = btrfs_alloc_path(); 1843 if (!path) 1844 return -ENOMEM; 1845 1846 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1847 key.type = BTRFS_DEV_ITEM_KEY; 1848 key.offset = device->devid; 1849 1850 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, 1851 &key, sizeof(*dev_item)); 1852 if (ret) 1853 goto out; 1854 1855 leaf = path->nodes[0]; 1856 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 1857 1858 btrfs_set_device_id(leaf, dev_item, device->devid); 1859 btrfs_set_device_generation(leaf, dev_item, 0); 1860 btrfs_set_device_type(leaf, dev_item, device->type); 1861 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 1862 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 1863 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 1864 btrfs_set_device_total_bytes(leaf, dev_item, 1865 btrfs_device_get_disk_total_bytes(device)); 1866 btrfs_set_device_bytes_used(leaf, dev_item, 1867 btrfs_device_get_bytes_used(device)); 1868 btrfs_set_device_group(leaf, dev_item, 0); 1869 btrfs_set_device_seek_speed(leaf, dev_item, 0); 1870 btrfs_set_device_bandwidth(leaf, dev_item, 0); 1871 btrfs_set_device_start_offset(leaf, dev_item, 0); 1872 1873 ptr = btrfs_device_uuid(dev_item); 1874 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 1875 ptr = btrfs_device_fsid(dev_item); 1876 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, 1877 ptr, BTRFS_FSID_SIZE); 1878 btrfs_mark_buffer_dirty(leaf); 1879 1880 ret = 0; 1881 out: 1882 btrfs_free_path(path); 1883 return ret; 1884 } 1885 1886 /* 1887 * Function to update ctime/mtime for a given device path. 1888 * Mainly used for ctime/mtime based probe like libblkid. 1889 */ 1890 static void update_dev_time(struct block_device *bdev) 1891 { 1892 struct inode *inode = bdev->bd_inode; 1893 struct timespec64 now; 1894 1895 /* Shouldn't happen but just in case. */ 1896 if (!inode) 1897 return; 1898 1899 now = current_time(inode); 1900 generic_update_time(inode, &now, S_MTIME | S_CTIME); 1901 } 1902 1903 static int btrfs_rm_dev_item(struct btrfs_device *device) 1904 { 1905 struct btrfs_root *root = device->fs_info->chunk_root; 1906 int ret; 1907 struct btrfs_path *path; 1908 struct btrfs_key key; 1909 struct btrfs_trans_handle *trans; 1910 1911 path = btrfs_alloc_path(); 1912 if (!path) 1913 return -ENOMEM; 1914 1915 trans = btrfs_start_transaction(root, 0); 1916 if (IS_ERR(trans)) { 1917 btrfs_free_path(path); 1918 return PTR_ERR(trans); 1919 } 1920 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 1921 key.type = BTRFS_DEV_ITEM_KEY; 1922 key.offset = device->devid; 1923 1924 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 1925 if (ret) { 1926 if (ret > 0) 1927 ret = -ENOENT; 1928 btrfs_abort_transaction(trans, ret); 1929 btrfs_end_transaction(trans); 1930 goto out; 1931 } 1932 1933 ret = btrfs_del_item(trans, root, path); 1934 if (ret) { 1935 btrfs_abort_transaction(trans, ret); 1936 btrfs_end_transaction(trans); 1937 } 1938 1939 out: 1940 btrfs_free_path(path); 1941 if (!ret) 1942 ret = btrfs_commit_transaction(trans); 1943 return ret; 1944 } 1945 1946 /* 1947 * Verify that @num_devices satisfies the RAID profile constraints in the whole 1948 * filesystem. It's up to the caller to adjust that number regarding eg. device 1949 * replace. 1950 */ 1951 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, 1952 u64 num_devices) 1953 { 1954 u64 all_avail; 1955 unsigned seq; 1956 int i; 1957 1958 do { 1959 seq = read_seqbegin(&fs_info->profiles_lock); 1960 1961 all_avail = fs_info->avail_data_alloc_bits | 1962 fs_info->avail_system_alloc_bits | 1963 fs_info->avail_metadata_alloc_bits; 1964 } while (read_seqretry(&fs_info->profiles_lock, seq)); 1965 1966 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { 1967 if (!(all_avail & btrfs_raid_array[i].bg_flag)) 1968 continue; 1969 1970 if (num_devices < btrfs_raid_array[i].devs_min) 1971 return btrfs_raid_array[i].mindev_error; 1972 } 1973 1974 return 0; 1975 } 1976 1977 static struct btrfs_device * btrfs_find_next_active_device( 1978 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) 1979 { 1980 struct btrfs_device *next_device; 1981 1982 list_for_each_entry(next_device, &fs_devs->devices, dev_list) { 1983 if (next_device != device && 1984 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) 1985 && next_device->bdev) 1986 return next_device; 1987 } 1988 1989 return NULL; 1990 } 1991 1992 /* 1993 * Helper function to check if the given device is part of s_bdev / latest_dev 1994 * and replace it with the provided or the next active device, in the context 1995 * where this function called, there should be always be another device (or 1996 * this_dev) which is active. 1997 */ 1998 void __cold btrfs_assign_next_active_device(struct btrfs_device *device, 1999 struct btrfs_device *next_device) 2000 { 2001 struct btrfs_fs_info *fs_info = device->fs_info; 2002 2003 if (!next_device) 2004 next_device = btrfs_find_next_active_device(fs_info->fs_devices, 2005 device); 2006 ASSERT(next_device); 2007 2008 if (fs_info->sb->s_bdev && 2009 (fs_info->sb->s_bdev == device->bdev)) 2010 fs_info->sb->s_bdev = next_device->bdev; 2011 2012 if (fs_info->fs_devices->latest_dev->bdev == device->bdev) 2013 fs_info->fs_devices->latest_dev = next_device; 2014 } 2015 2016 /* 2017 * Return btrfs_fs_devices::num_devices excluding the device that's being 2018 * currently replaced. 2019 */ 2020 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) 2021 { 2022 u64 num_devices = fs_info->fs_devices->num_devices; 2023 2024 down_read(&fs_info->dev_replace.rwsem); 2025 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { 2026 ASSERT(num_devices > 1); 2027 num_devices--; 2028 } 2029 up_read(&fs_info->dev_replace.rwsem); 2030 2031 return num_devices; 2032 } 2033 2034 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 2035 struct block_device *bdev, 2036 const char *device_path) 2037 { 2038 struct btrfs_super_block *disk_super; 2039 int copy_num; 2040 2041 if (!bdev) 2042 return; 2043 2044 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { 2045 struct page *page; 2046 int ret; 2047 2048 disk_super = btrfs_read_dev_one_super(bdev, copy_num); 2049 if (IS_ERR(disk_super)) 2050 continue; 2051 2052 if (bdev_is_zoned(bdev)) { 2053 btrfs_reset_sb_log_zones(bdev, copy_num); 2054 continue; 2055 } 2056 2057 memset(&disk_super->magic, 0, sizeof(disk_super->magic)); 2058 2059 page = virt_to_page(disk_super); 2060 set_page_dirty(page); 2061 lock_page(page); 2062 /* write_on_page() unlocks the page */ 2063 ret = write_one_page(page); 2064 if (ret) 2065 btrfs_warn(fs_info, 2066 "error clearing superblock number %d (%d)", 2067 copy_num, ret); 2068 btrfs_release_disk_super(disk_super); 2069 2070 } 2071 2072 /* Notify udev that device has changed */ 2073 btrfs_kobject_uevent(bdev, KOBJ_CHANGE); 2074 2075 /* Update ctime/mtime for device path for libblkid */ 2076 update_dev_time(bdev); 2077 } 2078 2079 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, 2080 u64 devid, struct block_device **bdev, fmode_t *mode) 2081 { 2082 struct btrfs_device *device; 2083 struct btrfs_fs_devices *cur_devices; 2084 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2085 u64 num_devices; 2086 int ret = 0; 2087 2088 /* 2089 * The device list in fs_devices is accessed without locks (neither 2090 * uuid_mutex nor device_list_mutex) as it won't change on a mounted 2091 * filesystem and another device rm cannot run. 2092 */ 2093 num_devices = btrfs_num_devices(fs_info); 2094 2095 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); 2096 if (ret) 2097 goto out; 2098 2099 device = btrfs_find_device_by_devspec(fs_info, devid, device_path); 2100 2101 if (IS_ERR(device)) { 2102 if (PTR_ERR(device) == -ENOENT && 2103 device_path && strcmp(device_path, "missing") == 0) 2104 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; 2105 else 2106 ret = PTR_ERR(device); 2107 goto out; 2108 } 2109 2110 if (btrfs_pinned_by_swapfile(fs_info, device)) { 2111 btrfs_warn_in_rcu(fs_info, 2112 "cannot remove device %s (devid %llu) due to active swapfile", 2113 rcu_str_deref(device->name), device->devid); 2114 ret = -ETXTBSY; 2115 goto out; 2116 } 2117 2118 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2119 ret = BTRFS_ERROR_DEV_TGT_REPLACE; 2120 goto out; 2121 } 2122 2123 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 2124 fs_info->fs_devices->rw_devices == 1) { 2125 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; 2126 goto out; 2127 } 2128 2129 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2130 mutex_lock(&fs_info->chunk_mutex); 2131 list_del_init(&device->dev_alloc_list); 2132 device->fs_devices->rw_devices--; 2133 mutex_unlock(&fs_info->chunk_mutex); 2134 } 2135 2136 ret = btrfs_shrink_device(device, 0); 2137 if (!ret) 2138 btrfs_reada_remove_dev(device); 2139 if (ret) 2140 goto error_undo; 2141 2142 /* 2143 * TODO: the superblock still includes this device in its num_devices 2144 * counter although write_all_supers() is not locked out. This 2145 * could give a filesystem state which requires a degraded mount. 2146 */ 2147 ret = btrfs_rm_dev_item(device); 2148 if (ret) 2149 goto error_undo; 2150 2151 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2152 btrfs_scrub_cancel_dev(device); 2153 2154 /* 2155 * the device list mutex makes sure that we don't change 2156 * the device list while someone else is writing out all 2157 * the device supers. Whoever is writing all supers, should 2158 * lock the device list mutex before getting the number of 2159 * devices in the super block (super_copy). Conversely, 2160 * whoever updates the number of devices in the super block 2161 * (super_copy) should hold the device list mutex. 2162 */ 2163 2164 /* 2165 * In normal cases the cur_devices == fs_devices. But in case 2166 * of deleting a seed device, the cur_devices should point to 2167 * its own fs_devices listed under the fs_devices->seed_list. 2168 */ 2169 cur_devices = device->fs_devices; 2170 mutex_lock(&fs_devices->device_list_mutex); 2171 list_del_rcu(&device->dev_list); 2172 2173 cur_devices->num_devices--; 2174 cur_devices->total_devices--; 2175 /* Update total_devices of the parent fs_devices if it's seed */ 2176 if (cur_devices != fs_devices) 2177 fs_devices->total_devices--; 2178 2179 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) 2180 cur_devices->missing_devices--; 2181 2182 btrfs_assign_next_active_device(device, NULL); 2183 2184 if (device->bdev) { 2185 cur_devices->open_devices--; 2186 /* remove sysfs entry */ 2187 btrfs_sysfs_remove_device(device); 2188 } 2189 2190 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; 2191 btrfs_set_super_num_devices(fs_info->super_copy, num_devices); 2192 mutex_unlock(&fs_devices->device_list_mutex); 2193 2194 /* 2195 * At this point, the device is zero sized and detached from the 2196 * devices list. All that's left is to zero out the old supers and 2197 * free the device. 2198 * 2199 * We cannot call btrfs_close_bdev() here because we're holding the sb 2200 * write lock, and blkdev_put() will pull in the ->open_mutex on the 2201 * block device and it's dependencies. Instead just flush the device 2202 * and let the caller do the final blkdev_put. 2203 */ 2204 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2205 btrfs_scratch_superblocks(fs_info, device->bdev, 2206 device->name->str); 2207 if (device->bdev) { 2208 sync_blockdev(device->bdev); 2209 invalidate_bdev(device->bdev); 2210 } 2211 } 2212 2213 *bdev = device->bdev; 2214 *mode = device->mode; 2215 synchronize_rcu(); 2216 btrfs_free_device(device); 2217 2218 /* 2219 * This can happen if cur_devices is the private seed devices list. We 2220 * cannot call close_fs_devices() here because it expects the uuid_mutex 2221 * to be held, but in fact we don't need that for the private 2222 * seed_devices, we can simply decrement cur_devices->opened and then 2223 * remove it from our list and free the fs_devices. 2224 */ 2225 if (cur_devices->num_devices == 0) { 2226 list_del_init(&cur_devices->seed_list); 2227 ASSERT(cur_devices->opened == 1); 2228 cur_devices->opened--; 2229 free_fs_devices(cur_devices); 2230 } 2231 2232 out: 2233 return ret; 2234 2235 error_undo: 2236 btrfs_reada_undo_remove_dev(device); 2237 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 2238 mutex_lock(&fs_info->chunk_mutex); 2239 list_add(&device->dev_alloc_list, 2240 &fs_devices->alloc_list); 2241 device->fs_devices->rw_devices++; 2242 mutex_unlock(&fs_info->chunk_mutex); 2243 } 2244 goto out; 2245 } 2246 2247 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) 2248 { 2249 struct btrfs_fs_devices *fs_devices; 2250 2251 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); 2252 2253 /* 2254 * in case of fs with no seed, srcdev->fs_devices will point 2255 * to fs_devices of fs_info. However when the dev being replaced is 2256 * a seed dev it will point to the seed's local fs_devices. In short 2257 * srcdev will have its correct fs_devices in both the cases. 2258 */ 2259 fs_devices = srcdev->fs_devices; 2260 2261 list_del_rcu(&srcdev->dev_list); 2262 list_del(&srcdev->dev_alloc_list); 2263 fs_devices->num_devices--; 2264 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) 2265 fs_devices->missing_devices--; 2266 2267 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) 2268 fs_devices->rw_devices--; 2269 2270 if (srcdev->bdev) 2271 fs_devices->open_devices--; 2272 } 2273 2274 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) 2275 { 2276 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; 2277 2278 mutex_lock(&uuid_mutex); 2279 2280 btrfs_close_bdev(srcdev); 2281 synchronize_rcu(); 2282 btrfs_free_device(srcdev); 2283 2284 /* if this is no devs we rather delete the fs_devices */ 2285 if (!fs_devices->num_devices) { 2286 /* 2287 * On a mounted FS, num_devices can't be zero unless it's a 2288 * seed. In case of a seed device being replaced, the replace 2289 * target added to the sprout FS, so there will be no more 2290 * device left under the seed FS. 2291 */ 2292 ASSERT(fs_devices->seeding); 2293 2294 list_del_init(&fs_devices->seed_list); 2295 close_fs_devices(fs_devices); 2296 free_fs_devices(fs_devices); 2297 } 2298 mutex_unlock(&uuid_mutex); 2299 } 2300 2301 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) 2302 { 2303 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; 2304 2305 mutex_lock(&fs_devices->device_list_mutex); 2306 2307 btrfs_sysfs_remove_device(tgtdev); 2308 2309 if (tgtdev->bdev) 2310 fs_devices->open_devices--; 2311 2312 fs_devices->num_devices--; 2313 2314 btrfs_assign_next_active_device(tgtdev, NULL); 2315 2316 list_del_rcu(&tgtdev->dev_list); 2317 2318 mutex_unlock(&fs_devices->device_list_mutex); 2319 2320 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, 2321 tgtdev->name->str); 2322 2323 btrfs_close_bdev(tgtdev); 2324 synchronize_rcu(); 2325 btrfs_free_device(tgtdev); 2326 } 2327 2328 /** 2329 * Populate args from device at path 2330 * 2331 * @fs_info: the filesystem 2332 * @args: the args to populate 2333 * @path: the path to the device 2334 * 2335 * This will read the super block of the device at @path and populate @args with 2336 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to 2337 * lookup a device to operate on, but need to do it before we take any locks. 2338 * This properly handles the special case of "missing" that a user may pass in, 2339 * and does some basic sanity checks. The caller must make sure that @path is 2340 * properly NUL terminated before calling in, and must call 2341 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and 2342 * uuid buffers. 2343 * 2344 * Return: 0 for success, -errno for failure 2345 */ 2346 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 2347 struct btrfs_dev_lookup_args *args, 2348 const char *path) 2349 { 2350 struct btrfs_super_block *disk_super; 2351 struct block_device *bdev; 2352 int ret; 2353 2354 if (!path || !path[0]) 2355 return -EINVAL; 2356 if (!strcmp(path, "missing")) { 2357 args->missing = true; 2358 return 0; 2359 } 2360 2361 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); 2362 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); 2363 if (!args->uuid || !args->fsid) { 2364 btrfs_put_dev_args_from_path(args); 2365 return -ENOMEM; 2366 } 2367 2368 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, 2369 &bdev, &disk_super); 2370 if (ret) 2371 return ret; 2372 args->devid = btrfs_stack_device_id(&disk_super->dev_item); 2373 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); 2374 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) 2375 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); 2376 else 2377 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); 2378 btrfs_release_disk_super(disk_super); 2379 blkdev_put(bdev, FMODE_READ); 2380 return 0; 2381 } 2382 2383 /* 2384 * Only use this jointly with btrfs_get_dev_args_from_path() because we will 2385 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables 2386 * that don't need to be freed. 2387 */ 2388 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) 2389 { 2390 kfree(args->uuid); 2391 kfree(args->fsid); 2392 args->uuid = NULL; 2393 args->fsid = NULL; 2394 } 2395 2396 struct btrfs_device *btrfs_find_device_by_devspec( 2397 struct btrfs_fs_info *fs_info, u64 devid, 2398 const char *device_path) 2399 { 2400 BTRFS_DEV_LOOKUP_ARGS(args); 2401 struct btrfs_device *device; 2402 int ret; 2403 2404 if (devid) { 2405 args.devid = devid; 2406 device = btrfs_find_device(fs_info->fs_devices, &args); 2407 if (!device) 2408 return ERR_PTR(-ENOENT); 2409 return device; 2410 } 2411 2412 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); 2413 if (ret) 2414 return ERR_PTR(ret); 2415 device = btrfs_find_device(fs_info->fs_devices, &args); 2416 btrfs_put_dev_args_from_path(&args); 2417 if (!device) 2418 return ERR_PTR(-ENOENT); 2419 return device; 2420 } 2421 2422 /* 2423 * does all the dirty work required for changing file system's UUID. 2424 */ 2425 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) 2426 { 2427 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2428 struct btrfs_fs_devices *old_devices; 2429 struct btrfs_fs_devices *seed_devices; 2430 struct btrfs_super_block *disk_super = fs_info->super_copy; 2431 struct btrfs_device *device; 2432 u64 super_flags; 2433 2434 lockdep_assert_held(&uuid_mutex); 2435 if (!fs_devices->seeding) 2436 return -EINVAL; 2437 2438 /* 2439 * Private copy of the seed devices, anchored at 2440 * fs_info->fs_devices->seed_list 2441 */ 2442 seed_devices = alloc_fs_devices(NULL, NULL); 2443 if (IS_ERR(seed_devices)) 2444 return PTR_ERR(seed_devices); 2445 2446 /* 2447 * It's necessary to retain a copy of the original seed fs_devices in 2448 * fs_uuids so that filesystems which have been seeded can successfully 2449 * reference the seed device from open_seed_devices. This also supports 2450 * multiple fs seed. 2451 */ 2452 old_devices = clone_fs_devices(fs_devices); 2453 if (IS_ERR(old_devices)) { 2454 kfree(seed_devices); 2455 return PTR_ERR(old_devices); 2456 } 2457 2458 list_add(&old_devices->fs_list, &fs_uuids); 2459 2460 memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); 2461 seed_devices->opened = 1; 2462 INIT_LIST_HEAD(&seed_devices->devices); 2463 INIT_LIST_HEAD(&seed_devices->alloc_list); 2464 mutex_init(&seed_devices->device_list_mutex); 2465 2466 mutex_lock(&fs_devices->device_list_mutex); 2467 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, 2468 synchronize_rcu); 2469 list_for_each_entry(device, &seed_devices->devices, dev_list) 2470 device->fs_devices = seed_devices; 2471 2472 fs_devices->seeding = false; 2473 fs_devices->num_devices = 0; 2474 fs_devices->open_devices = 0; 2475 fs_devices->missing_devices = 0; 2476 fs_devices->rotating = false; 2477 list_add(&seed_devices->seed_list, &fs_devices->seed_list); 2478 2479 generate_random_uuid(fs_devices->fsid); 2480 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); 2481 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); 2482 mutex_unlock(&fs_devices->device_list_mutex); 2483 2484 super_flags = btrfs_super_flags(disk_super) & 2485 ~BTRFS_SUPER_FLAG_SEEDING; 2486 btrfs_set_super_flags(disk_super, super_flags); 2487 2488 return 0; 2489 } 2490 2491 /* 2492 * Store the expected generation for seed devices in device items. 2493 */ 2494 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) 2495 { 2496 BTRFS_DEV_LOOKUP_ARGS(args); 2497 struct btrfs_fs_info *fs_info = trans->fs_info; 2498 struct btrfs_root *root = fs_info->chunk_root; 2499 struct btrfs_path *path; 2500 struct extent_buffer *leaf; 2501 struct btrfs_dev_item *dev_item; 2502 struct btrfs_device *device; 2503 struct btrfs_key key; 2504 u8 fs_uuid[BTRFS_FSID_SIZE]; 2505 u8 dev_uuid[BTRFS_UUID_SIZE]; 2506 int ret; 2507 2508 path = btrfs_alloc_path(); 2509 if (!path) 2510 return -ENOMEM; 2511 2512 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2513 key.offset = 0; 2514 key.type = BTRFS_DEV_ITEM_KEY; 2515 2516 while (1) { 2517 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2518 if (ret < 0) 2519 goto error; 2520 2521 leaf = path->nodes[0]; 2522 next_slot: 2523 if (path->slots[0] >= btrfs_header_nritems(leaf)) { 2524 ret = btrfs_next_leaf(root, path); 2525 if (ret > 0) 2526 break; 2527 if (ret < 0) 2528 goto error; 2529 leaf = path->nodes[0]; 2530 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2531 btrfs_release_path(path); 2532 continue; 2533 } 2534 2535 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 2536 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || 2537 key.type != BTRFS_DEV_ITEM_KEY) 2538 break; 2539 2540 dev_item = btrfs_item_ptr(leaf, path->slots[0], 2541 struct btrfs_dev_item); 2542 args.devid = btrfs_device_id(leaf, dev_item); 2543 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 2544 BTRFS_UUID_SIZE); 2545 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 2546 BTRFS_FSID_SIZE); 2547 args.uuid = dev_uuid; 2548 args.fsid = fs_uuid; 2549 device = btrfs_find_device(fs_info->fs_devices, &args); 2550 BUG_ON(!device); /* Logic error */ 2551 2552 if (device->fs_devices->seeding) { 2553 btrfs_set_device_generation(leaf, dev_item, 2554 device->generation); 2555 btrfs_mark_buffer_dirty(leaf); 2556 } 2557 2558 path->slots[0]++; 2559 goto next_slot; 2560 } 2561 ret = 0; 2562 error: 2563 btrfs_free_path(path); 2564 return ret; 2565 } 2566 2567 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) 2568 { 2569 struct btrfs_root *root = fs_info->dev_root; 2570 struct request_queue *q; 2571 struct btrfs_trans_handle *trans; 2572 struct btrfs_device *device; 2573 struct block_device *bdev; 2574 struct super_block *sb = fs_info->sb; 2575 struct rcu_string *name; 2576 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2577 u64 orig_super_total_bytes; 2578 u64 orig_super_num_devices; 2579 int seeding_dev = 0; 2580 int ret = 0; 2581 bool locked = false; 2582 2583 if (sb_rdonly(sb) && !fs_devices->seeding) 2584 return -EROFS; 2585 2586 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, 2587 fs_info->bdev_holder); 2588 if (IS_ERR(bdev)) 2589 return PTR_ERR(bdev); 2590 2591 if (!btrfs_check_device_zone_type(fs_info, bdev)) { 2592 ret = -EINVAL; 2593 goto error; 2594 } 2595 2596 if (fs_devices->seeding) { 2597 seeding_dev = 1; 2598 down_write(&sb->s_umount); 2599 mutex_lock(&uuid_mutex); 2600 locked = true; 2601 } 2602 2603 sync_blockdev(bdev); 2604 2605 rcu_read_lock(); 2606 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 2607 if (device->bdev == bdev) { 2608 ret = -EEXIST; 2609 rcu_read_unlock(); 2610 goto error; 2611 } 2612 } 2613 rcu_read_unlock(); 2614 2615 device = btrfs_alloc_device(fs_info, NULL, NULL); 2616 if (IS_ERR(device)) { 2617 /* we can safely leave the fs_devices entry around */ 2618 ret = PTR_ERR(device); 2619 goto error; 2620 } 2621 2622 name = rcu_string_strdup(device_path, GFP_KERNEL); 2623 if (!name) { 2624 ret = -ENOMEM; 2625 goto error_free_device; 2626 } 2627 rcu_assign_pointer(device->name, name); 2628 2629 device->fs_info = fs_info; 2630 device->bdev = bdev; 2631 2632 ret = btrfs_get_dev_zone_info(device); 2633 if (ret) 2634 goto error_free_device; 2635 2636 trans = btrfs_start_transaction(root, 0); 2637 if (IS_ERR(trans)) { 2638 ret = PTR_ERR(trans); 2639 goto error_free_zone; 2640 } 2641 2642 q = bdev_get_queue(bdev); 2643 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); 2644 device->generation = trans->transid; 2645 device->io_width = fs_info->sectorsize; 2646 device->io_align = fs_info->sectorsize; 2647 device->sector_size = fs_info->sectorsize; 2648 device->total_bytes = round_down(i_size_read(bdev->bd_inode), 2649 fs_info->sectorsize); 2650 device->disk_total_bytes = device->total_bytes; 2651 device->commit_total_bytes = device->total_bytes; 2652 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 2653 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 2654 device->mode = FMODE_EXCL; 2655 device->dev_stats_valid = 1; 2656 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); 2657 2658 if (seeding_dev) { 2659 btrfs_clear_sb_rdonly(sb); 2660 ret = btrfs_prepare_sprout(fs_info); 2661 if (ret) { 2662 btrfs_abort_transaction(trans, ret); 2663 goto error_trans; 2664 } 2665 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, 2666 device); 2667 } 2668 2669 device->fs_devices = fs_devices; 2670 2671 mutex_lock(&fs_devices->device_list_mutex); 2672 mutex_lock(&fs_info->chunk_mutex); 2673 list_add_rcu(&device->dev_list, &fs_devices->devices); 2674 list_add(&device->dev_alloc_list, &fs_devices->alloc_list); 2675 fs_devices->num_devices++; 2676 fs_devices->open_devices++; 2677 fs_devices->rw_devices++; 2678 fs_devices->total_devices++; 2679 fs_devices->total_rw_bytes += device->total_bytes; 2680 2681 atomic64_add(device->total_bytes, &fs_info->free_chunk_space); 2682 2683 if (!blk_queue_nonrot(q)) 2684 fs_devices->rotating = true; 2685 2686 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 2687 btrfs_set_super_total_bytes(fs_info->super_copy, 2688 round_down(orig_super_total_bytes + device->total_bytes, 2689 fs_info->sectorsize)); 2690 2691 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); 2692 btrfs_set_super_num_devices(fs_info->super_copy, 2693 orig_super_num_devices + 1); 2694 2695 /* 2696 * we've got more storage, clear any full flags on the space 2697 * infos 2698 */ 2699 btrfs_clear_space_info_full(fs_info); 2700 2701 mutex_unlock(&fs_info->chunk_mutex); 2702 2703 /* Add sysfs device entry */ 2704 btrfs_sysfs_add_device(device); 2705 2706 mutex_unlock(&fs_devices->device_list_mutex); 2707 2708 if (seeding_dev) { 2709 mutex_lock(&fs_info->chunk_mutex); 2710 ret = init_first_rw_device(trans); 2711 mutex_unlock(&fs_info->chunk_mutex); 2712 if (ret) { 2713 btrfs_abort_transaction(trans, ret); 2714 goto error_sysfs; 2715 } 2716 } 2717 2718 ret = btrfs_add_dev_item(trans, device); 2719 if (ret) { 2720 btrfs_abort_transaction(trans, ret); 2721 goto error_sysfs; 2722 } 2723 2724 if (seeding_dev) { 2725 ret = btrfs_finish_sprout(trans); 2726 if (ret) { 2727 btrfs_abort_transaction(trans, ret); 2728 goto error_sysfs; 2729 } 2730 2731 /* 2732 * fs_devices now represents the newly sprouted filesystem and 2733 * its fsid has been changed by btrfs_prepare_sprout 2734 */ 2735 btrfs_sysfs_update_sprout_fsid(fs_devices); 2736 } 2737 2738 ret = btrfs_commit_transaction(trans); 2739 2740 if (seeding_dev) { 2741 mutex_unlock(&uuid_mutex); 2742 up_write(&sb->s_umount); 2743 locked = false; 2744 2745 if (ret) /* transaction commit */ 2746 return ret; 2747 2748 ret = btrfs_relocate_sys_chunks(fs_info); 2749 if (ret < 0) 2750 btrfs_handle_fs_error(fs_info, ret, 2751 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); 2752 trans = btrfs_attach_transaction(root); 2753 if (IS_ERR(trans)) { 2754 if (PTR_ERR(trans) == -ENOENT) 2755 return 0; 2756 ret = PTR_ERR(trans); 2757 trans = NULL; 2758 goto error_sysfs; 2759 } 2760 ret = btrfs_commit_transaction(trans); 2761 } 2762 2763 /* 2764 * Now that we have written a new super block to this device, check all 2765 * other fs_devices list if device_path alienates any other scanned 2766 * device. 2767 * We can ignore the return value as it typically returns -EINVAL and 2768 * only succeeds if the device was an alien. 2769 */ 2770 btrfs_forget_devices(device_path); 2771 2772 /* Update ctime/mtime for blkid or udev */ 2773 update_dev_time(bdev); 2774 2775 return ret; 2776 2777 error_sysfs: 2778 btrfs_sysfs_remove_device(device); 2779 mutex_lock(&fs_info->fs_devices->device_list_mutex); 2780 mutex_lock(&fs_info->chunk_mutex); 2781 list_del_rcu(&device->dev_list); 2782 list_del(&device->dev_alloc_list); 2783 fs_info->fs_devices->num_devices--; 2784 fs_info->fs_devices->open_devices--; 2785 fs_info->fs_devices->rw_devices--; 2786 fs_info->fs_devices->total_devices--; 2787 fs_info->fs_devices->total_rw_bytes -= device->total_bytes; 2788 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); 2789 btrfs_set_super_total_bytes(fs_info->super_copy, 2790 orig_super_total_bytes); 2791 btrfs_set_super_num_devices(fs_info->super_copy, 2792 orig_super_num_devices); 2793 mutex_unlock(&fs_info->chunk_mutex); 2794 mutex_unlock(&fs_info->fs_devices->device_list_mutex); 2795 error_trans: 2796 if (seeding_dev) 2797 btrfs_set_sb_rdonly(sb); 2798 if (trans) 2799 btrfs_end_transaction(trans); 2800 error_free_zone: 2801 btrfs_destroy_dev_zone_info(device); 2802 error_free_device: 2803 btrfs_free_device(device); 2804 error: 2805 blkdev_put(bdev, FMODE_EXCL); 2806 if (locked) { 2807 mutex_unlock(&uuid_mutex); 2808 up_write(&sb->s_umount); 2809 } 2810 return ret; 2811 } 2812 2813 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, 2814 struct btrfs_device *device) 2815 { 2816 int ret; 2817 struct btrfs_path *path; 2818 struct btrfs_root *root = device->fs_info->chunk_root; 2819 struct btrfs_dev_item *dev_item; 2820 struct extent_buffer *leaf; 2821 struct btrfs_key key; 2822 2823 path = btrfs_alloc_path(); 2824 if (!path) 2825 return -ENOMEM; 2826 2827 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 2828 key.type = BTRFS_DEV_ITEM_KEY; 2829 key.offset = device->devid; 2830 2831 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 2832 if (ret < 0) 2833 goto out; 2834 2835 if (ret > 0) { 2836 ret = -ENOENT; 2837 goto out; 2838 } 2839 2840 leaf = path->nodes[0]; 2841 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); 2842 2843 btrfs_set_device_id(leaf, dev_item, device->devid); 2844 btrfs_set_device_type(leaf, dev_item, device->type); 2845 btrfs_set_device_io_align(leaf, dev_item, device->io_align); 2846 btrfs_set_device_io_width(leaf, dev_item, device->io_width); 2847 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); 2848 btrfs_set_device_total_bytes(leaf, dev_item, 2849 btrfs_device_get_disk_total_bytes(device)); 2850 btrfs_set_device_bytes_used(leaf, dev_item, 2851 btrfs_device_get_bytes_used(device)); 2852 btrfs_mark_buffer_dirty(leaf); 2853 2854 out: 2855 btrfs_free_path(path); 2856 return ret; 2857 } 2858 2859 int btrfs_grow_device(struct btrfs_trans_handle *trans, 2860 struct btrfs_device *device, u64 new_size) 2861 { 2862 struct btrfs_fs_info *fs_info = device->fs_info; 2863 struct btrfs_super_block *super_copy = fs_info->super_copy; 2864 u64 old_total; 2865 u64 diff; 2866 2867 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 2868 return -EACCES; 2869 2870 new_size = round_down(new_size, fs_info->sectorsize); 2871 2872 mutex_lock(&fs_info->chunk_mutex); 2873 old_total = btrfs_super_total_bytes(super_copy); 2874 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); 2875 2876 if (new_size <= device->total_bytes || 2877 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2878 mutex_unlock(&fs_info->chunk_mutex); 2879 return -EINVAL; 2880 } 2881 2882 btrfs_set_super_total_bytes(super_copy, 2883 round_down(old_total + diff, fs_info->sectorsize)); 2884 device->fs_devices->total_rw_bytes += diff; 2885 2886 btrfs_device_set_total_bytes(device, new_size); 2887 btrfs_device_set_disk_total_bytes(device, new_size); 2888 btrfs_clear_space_info_full(device->fs_info); 2889 if (list_empty(&device->post_commit_list)) 2890 list_add_tail(&device->post_commit_list, 2891 &trans->transaction->dev_update_list); 2892 mutex_unlock(&fs_info->chunk_mutex); 2893 2894 return btrfs_update_device(trans, device); 2895 } 2896 2897 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 2898 { 2899 struct btrfs_fs_info *fs_info = trans->fs_info; 2900 struct btrfs_root *root = fs_info->chunk_root; 2901 int ret; 2902 struct btrfs_path *path; 2903 struct btrfs_key key; 2904 2905 path = btrfs_alloc_path(); 2906 if (!path) 2907 return -ENOMEM; 2908 2909 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 2910 key.offset = chunk_offset; 2911 key.type = BTRFS_CHUNK_ITEM_KEY; 2912 2913 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2914 if (ret < 0) 2915 goto out; 2916 else if (ret > 0) { /* Logic error or corruption */ 2917 btrfs_handle_fs_error(fs_info, -ENOENT, 2918 "Failed lookup while freeing chunk."); 2919 ret = -ENOENT; 2920 goto out; 2921 } 2922 2923 ret = btrfs_del_item(trans, root, path); 2924 if (ret < 0) 2925 btrfs_handle_fs_error(fs_info, ret, 2926 "Failed to delete chunk item."); 2927 out: 2928 btrfs_free_path(path); 2929 return ret; 2930 } 2931 2932 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 2933 { 2934 struct btrfs_super_block *super_copy = fs_info->super_copy; 2935 struct btrfs_disk_key *disk_key; 2936 struct btrfs_chunk *chunk; 2937 u8 *ptr; 2938 int ret = 0; 2939 u32 num_stripes; 2940 u32 array_size; 2941 u32 len = 0; 2942 u32 cur; 2943 struct btrfs_key key; 2944 2945 lockdep_assert_held(&fs_info->chunk_mutex); 2946 array_size = btrfs_super_sys_array_size(super_copy); 2947 2948 ptr = super_copy->sys_chunk_array; 2949 cur = 0; 2950 2951 while (cur < array_size) { 2952 disk_key = (struct btrfs_disk_key *)ptr; 2953 btrfs_disk_key_to_cpu(&key, disk_key); 2954 2955 len = sizeof(*disk_key); 2956 2957 if (key.type == BTRFS_CHUNK_ITEM_KEY) { 2958 chunk = (struct btrfs_chunk *)(ptr + len); 2959 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 2960 len += btrfs_chunk_item_size(num_stripes); 2961 } else { 2962 ret = -EIO; 2963 break; 2964 } 2965 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && 2966 key.offset == chunk_offset) { 2967 memmove(ptr, ptr + len, array_size - (cur + len)); 2968 array_size -= len; 2969 btrfs_set_super_sys_array_size(super_copy, array_size); 2970 } else { 2971 ptr += len; 2972 cur += len; 2973 } 2974 } 2975 return ret; 2976 } 2977 2978 /* 2979 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. 2980 * @logical: Logical block offset in bytes. 2981 * @length: Length of extent in bytes. 2982 * 2983 * Return: Chunk mapping or ERR_PTR. 2984 */ 2985 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 2986 u64 logical, u64 length) 2987 { 2988 struct extent_map_tree *em_tree; 2989 struct extent_map *em; 2990 2991 em_tree = &fs_info->mapping_tree; 2992 read_lock(&em_tree->lock); 2993 em = lookup_extent_mapping(em_tree, logical, length); 2994 read_unlock(&em_tree->lock); 2995 2996 if (!em) { 2997 btrfs_crit(fs_info, "unable to find logical %llu length %llu", 2998 logical, length); 2999 return ERR_PTR(-EINVAL); 3000 } 3001 3002 if (em->start > logical || em->start + em->len < logical) { 3003 btrfs_crit(fs_info, 3004 "found a bad mapping, wanted %llu-%llu, found %llu-%llu", 3005 logical, length, em->start, em->start + em->len); 3006 free_extent_map(em); 3007 return ERR_PTR(-EINVAL); 3008 } 3009 3010 /* callers are responsible for dropping em's ref. */ 3011 return em; 3012 } 3013 3014 static int remove_chunk_item(struct btrfs_trans_handle *trans, 3015 struct map_lookup *map, u64 chunk_offset) 3016 { 3017 int i; 3018 3019 /* 3020 * Removing chunk items and updating the device items in the chunks btree 3021 * requires holding the chunk_mutex. 3022 * See the comment at btrfs_chunk_alloc() for the details. 3023 */ 3024 lockdep_assert_held(&trans->fs_info->chunk_mutex); 3025 3026 for (i = 0; i < map->num_stripes; i++) { 3027 int ret; 3028 3029 ret = btrfs_update_device(trans, map->stripes[i].dev); 3030 if (ret) 3031 return ret; 3032 } 3033 3034 return btrfs_free_chunk(trans, chunk_offset); 3035 } 3036 3037 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) 3038 { 3039 struct btrfs_fs_info *fs_info = trans->fs_info; 3040 struct extent_map *em; 3041 struct map_lookup *map; 3042 u64 dev_extent_len = 0; 3043 int i, ret = 0; 3044 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3045 3046 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 3047 if (IS_ERR(em)) { 3048 /* 3049 * This is a logic error, but we don't want to just rely on the 3050 * user having built with ASSERT enabled, so if ASSERT doesn't 3051 * do anything we still error out. 3052 */ 3053 ASSERT(0); 3054 return PTR_ERR(em); 3055 } 3056 map = em->map_lookup; 3057 3058 /* 3059 * First delete the device extent items from the devices btree. 3060 * We take the device_list_mutex to avoid racing with the finishing phase 3061 * of a device replace operation. See the comment below before acquiring 3062 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex 3063 * because that can result in a deadlock when deleting the device extent 3064 * items from the devices btree - COWing an extent buffer from the btree 3065 * may result in allocating a new metadata chunk, which would attempt to 3066 * lock again fs_info->chunk_mutex. 3067 */ 3068 mutex_lock(&fs_devices->device_list_mutex); 3069 for (i = 0; i < map->num_stripes; i++) { 3070 struct btrfs_device *device = map->stripes[i].dev; 3071 ret = btrfs_free_dev_extent(trans, device, 3072 map->stripes[i].physical, 3073 &dev_extent_len); 3074 if (ret) { 3075 mutex_unlock(&fs_devices->device_list_mutex); 3076 btrfs_abort_transaction(trans, ret); 3077 goto out; 3078 } 3079 3080 if (device->bytes_used > 0) { 3081 mutex_lock(&fs_info->chunk_mutex); 3082 btrfs_device_set_bytes_used(device, 3083 device->bytes_used - dev_extent_len); 3084 atomic64_add(dev_extent_len, &fs_info->free_chunk_space); 3085 btrfs_clear_space_info_full(fs_info); 3086 mutex_unlock(&fs_info->chunk_mutex); 3087 } 3088 } 3089 mutex_unlock(&fs_devices->device_list_mutex); 3090 3091 /* 3092 * We acquire fs_info->chunk_mutex for 2 reasons: 3093 * 3094 * 1) Just like with the first phase of the chunk allocation, we must 3095 * reserve system space, do all chunk btree updates and deletions, and 3096 * update the system chunk array in the superblock while holding this 3097 * mutex. This is for similar reasons as explained on the comment at 3098 * the top of btrfs_chunk_alloc(); 3099 * 3100 * 2) Prevent races with the final phase of a device replace operation 3101 * that replaces the device object associated with the map's stripes, 3102 * because the device object's id can change at any time during that 3103 * final phase of the device replace operation 3104 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 3105 * replaced device and then see it with an ID of 3106 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating 3107 * the device item, which does not exists on the chunk btree. 3108 * The finishing phase of device replace acquires both the 3109 * device_list_mutex and the chunk_mutex, in that order, so we are 3110 * safe by just acquiring the chunk_mutex. 3111 */ 3112 trans->removing_chunk = true; 3113 mutex_lock(&fs_info->chunk_mutex); 3114 3115 check_system_chunk(trans, map->type); 3116 3117 ret = remove_chunk_item(trans, map, chunk_offset); 3118 /* 3119 * Normally we should not get -ENOSPC since we reserved space before 3120 * through the call to check_system_chunk(). 3121 * 3122 * Despite our system space_info having enough free space, we may not 3123 * be able to allocate extents from its block groups, because all have 3124 * an incompatible profile, which will force us to allocate a new system 3125 * block group with the right profile, or right after we called 3126 * check_system_space() above, a scrub turned the only system block group 3127 * with enough free space into RO mode. 3128 * This is explained with more detail at do_chunk_alloc(). 3129 * 3130 * So if we get -ENOSPC, allocate a new system chunk and retry once. 3131 */ 3132 if (ret == -ENOSPC) { 3133 const u64 sys_flags = btrfs_system_alloc_profile(fs_info); 3134 struct btrfs_block_group *sys_bg; 3135 3136 sys_bg = btrfs_create_chunk(trans, sys_flags); 3137 if (IS_ERR(sys_bg)) { 3138 ret = PTR_ERR(sys_bg); 3139 btrfs_abort_transaction(trans, ret); 3140 goto out; 3141 } 3142 3143 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); 3144 if (ret) { 3145 btrfs_abort_transaction(trans, ret); 3146 goto out; 3147 } 3148 3149 ret = remove_chunk_item(trans, map, chunk_offset); 3150 if (ret) { 3151 btrfs_abort_transaction(trans, ret); 3152 goto out; 3153 } 3154 } else if (ret) { 3155 btrfs_abort_transaction(trans, ret); 3156 goto out; 3157 } 3158 3159 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); 3160 3161 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 3162 ret = btrfs_del_sys_chunk(fs_info, chunk_offset); 3163 if (ret) { 3164 btrfs_abort_transaction(trans, ret); 3165 goto out; 3166 } 3167 } 3168 3169 mutex_unlock(&fs_info->chunk_mutex); 3170 trans->removing_chunk = false; 3171 3172 /* 3173 * We are done with chunk btree updates and deletions, so release the 3174 * system space we previously reserved (with check_system_chunk()). 3175 */ 3176 btrfs_trans_release_chunk_metadata(trans); 3177 3178 ret = btrfs_remove_block_group(trans, chunk_offset, em); 3179 if (ret) { 3180 btrfs_abort_transaction(trans, ret); 3181 goto out; 3182 } 3183 3184 out: 3185 if (trans->removing_chunk) { 3186 mutex_unlock(&fs_info->chunk_mutex); 3187 trans->removing_chunk = false; 3188 } 3189 /* once for us */ 3190 free_extent_map(em); 3191 return ret; 3192 } 3193 3194 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) 3195 { 3196 struct btrfs_root *root = fs_info->chunk_root; 3197 struct btrfs_trans_handle *trans; 3198 struct btrfs_block_group *block_group; 3199 u64 length; 3200 int ret; 3201 3202 /* 3203 * Prevent races with automatic removal of unused block groups. 3204 * After we relocate and before we remove the chunk with offset 3205 * chunk_offset, automatic removal of the block group can kick in, 3206 * resulting in a failure when calling btrfs_remove_chunk() below. 3207 * 3208 * Make sure to acquire this mutex before doing a tree search (dev 3209 * or chunk trees) to find chunks. Otherwise the cleaner kthread might 3210 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after 3211 * we release the path used to search the chunk/dev tree and before 3212 * the current task acquires this mutex and calls us. 3213 */ 3214 lockdep_assert_held(&fs_info->reclaim_bgs_lock); 3215 3216 /* step one, relocate all the extents inside this chunk */ 3217 btrfs_scrub_pause(fs_info); 3218 ret = btrfs_relocate_block_group(fs_info, chunk_offset); 3219 btrfs_scrub_continue(fs_info); 3220 if (ret) 3221 return ret; 3222 3223 block_group = btrfs_lookup_block_group(fs_info, chunk_offset); 3224 if (!block_group) 3225 return -ENOENT; 3226 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); 3227 length = block_group->length; 3228 btrfs_put_block_group(block_group); 3229 3230 /* 3231 * On a zoned file system, discard the whole block group, this will 3232 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If 3233 * resetting the zone fails, don't treat it as a fatal problem from the 3234 * filesystem's point of view. 3235 */ 3236 if (btrfs_is_zoned(fs_info)) { 3237 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); 3238 if (ret) 3239 btrfs_info(fs_info, 3240 "failed to reset zone %llu after relocation", 3241 chunk_offset); 3242 } 3243 3244 trans = btrfs_start_trans_remove_block_group(root->fs_info, 3245 chunk_offset); 3246 if (IS_ERR(trans)) { 3247 ret = PTR_ERR(trans); 3248 btrfs_handle_fs_error(root->fs_info, ret, NULL); 3249 return ret; 3250 } 3251 3252 /* 3253 * step two, delete the device extents and the 3254 * chunk tree entries 3255 */ 3256 ret = btrfs_remove_chunk(trans, chunk_offset); 3257 btrfs_end_transaction(trans); 3258 return ret; 3259 } 3260 3261 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) 3262 { 3263 struct btrfs_root *chunk_root = fs_info->chunk_root; 3264 struct btrfs_path *path; 3265 struct extent_buffer *leaf; 3266 struct btrfs_chunk *chunk; 3267 struct btrfs_key key; 3268 struct btrfs_key found_key; 3269 u64 chunk_type; 3270 bool retried = false; 3271 int failed = 0; 3272 int ret; 3273 3274 path = btrfs_alloc_path(); 3275 if (!path) 3276 return -ENOMEM; 3277 3278 again: 3279 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3280 key.offset = (u64)-1; 3281 key.type = BTRFS_CHUNK_ITEM_KEY; 3282 3283 while (1) { 3284 mutex_lock(&fs_info->reclaim_bgs_lock); 3285 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3286 if (ret < 0) { 3287 mutex_unlock(&fs_info->reclaim_bgs_lock); 3288 goto error; 3289 } 3290 BUG_ON(ret == 0); /* Corruption */ 3291 3292 ret = btrfs_previous_item(chunk_root, path, key.objectid, 3293 key.type); 3294 if (ret) 3295 mutex_unlock(&fs_info->reclaim_bgs_lock); 3296 if (ret < 0) 3297 goto error; 3298 if (ret > 0) 3299 break; 3300 3301 leaf = path->nodes[0]; 3302 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3303 3304 chunk = btrfs_item_ptr(leaf, path->slots[0], 3305 struct btrfs_chunk); 3306 chunk_type = btrfs_chunk_type(leaf, chunk); 3307 btrfs_release_path(path); 3308 3309 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { 3310 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3311 if (ret == -ENOSPC) 3312 failed++; 3313 else 3314 BUG_ON(ret); 3315 } 3316 mutex_unlock(&fs_info->reclaim_bgs_lock); 3317 3318 if (found_key.offset == 0) 3319 break; 3320 key.offset = found_key.offset - 1; 3321 } 3322 ret = 0; 3323 if (failed && !retried) { 3324 failed = 0; 3325 retried = true; 3326 goto again; 3327 } else if (WARN_ON(failed && retried)) { 3328 ret = -ENOSPC; 3329 } 3330 error: 3331 btrfs_free_path(path); 3332 return ret; 3333 } 3334 3335 /* 3336 * return 1 : allocate a data chunk successfully, 3337 * return <0: errors during allocating a data chunk, 3338 * return 0 : no need to allocate a data chunk. 3339 */ 3340 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, 3341 u64 chunk_offset) 3342 { 3343 struct btrfs_block_group *cache; 3344 u64 bytes_used; 3345 u64 chunk_type; 3346 3347 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3348 ASSERT(cache); 3349 chunk_type = cache->flags; 3350 btrfs_put_block_group(cache); 3351 3352 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) 3353 return 0; 3354 3355 spin_lock(&fs_info->data_sinfo->lock); 3356 bytes_used = fs_info->data_sinfo->bytes_used; 3357 spin_unlock(&fs_info->data_sinfo->lock); 3358 3359 if (!bytes_used) { 3360 struct btrfs_trans_handle *trans; 3361 int ret; 3362 3363 trans = btrfs_join_transaction(fs_info->tree_root); 3364 if (IS_ERR(trans)) 3365 return PTR_ERR(trans); 3366 3367 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); 3368 btrfs_end_transaction(trans); 3369 if (ret < 0) 3370 return ret; 3371 return 1; 3372 } 3373 3374 return 0; 3375 } 3376 3377 static int insert_balance_item(struct btrfs_fs_info *fs_info, 3378 struct btrfs_balance_control *bctl) 3379 { 3380 struct btrfs_root *root = fs_info->tree_root; 3381 struct btrfs_trans_handle *trans; 3382 struct btrfs_balance_item *item; 3383 struct btrfs_disk_balance_args disk_bargs; 3384 struct btrfs_path *path; 3385 struct extent_buffer *leaf; 3386 struct btrfs_key key; 3387 int ret, err; 3388 3389 path = btrfs_alloc_path(); 3390 if (!path) 3391 return -ENOMEM; 3392 3393 trans = btrfs_start_transaction(root, 0); 3394 if (IS_ERR(trans)) { 3395 btrfs_free_path(path); 3396 return PTR_ERR(trans); 3397 } 3398 3399 key.objectid = BTRFS_BALANCE_OBJECTID; 3400 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3401 key.offset = 0; 3402 3403 ret = btrfs_insert_empty_item(trans, root, path, &key, 3404 sizeof(*item)); 3405 if (ret) 3406 goto out; 3407 3408 leaf = path->nodes[0]; 3409 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 3410 3411 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 3412 3413 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); 3414 btrfs_set_balance_data(leaf, item, &disk_bargs); 3415 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); 3416 btrfs_set_balance_meta(leaf, item, &disk_bargs); 3417 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); 3418 btrfs_set_balance_sys(leaf, item, &disk_bargs); 3419 3420 btrfs_set_balance_flags(leaf, item, bctl->flags); 3421 3422 btrfs_mark_buffer_dirty(leaf); 3423 out: 3424 btrfs_free_path(path); 3425 err = btrfs_commit_transaction(trans); 3426 if (err && !ret) 3427 ret = err; 3428 return ret; 3429 } 3430 3431 static int del_balance_item(struct btrfs_fs_info *fs_info) 3432 { 3433 struct btrfs_root *root = fs_info->tree_root; 3434 struct btrfs_trans_handle *trans; 3435 struct btrfs_path *path; 3436 struct btrfs_key key; 3437 int ret, err; 3438 3439 path = btrfs_alloc_path(); 3440 if (!path) 3441 return -ENOMEM; 3442 3443 trans = btrfs_start_transaction_fallback_global_rsv(root, 0); 3444 if (IS_ERR(trans)) { 3445 btrfs_free_path(path); 3446 return PTR_ERR(trans); 3447 } 3448 3449 key.objectid = BTRFS_BALANCE_OBJECTID; 3450 key.type = BTRFS_TEMPORARY_ITEM_KEY; 3451 key.offset = 0; 3452 3453 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 3454 if (ret < 0) 3455 goto out; 3456 if (ret > 0) { 3457 ret = -ENOENT; 3458 goto out; 3459 } 3460 3461 ret = btrfs_del_item(trans, root, path); 3462 out: 3463 btrfs_free_path(path); 3464 err = btrfs_commit_transaction(trans); 3465 if (err && !ret) 3466 ret = err; 3467 return ret; 3468 } 3469 3470 /* 3471 * This is a heuristic used to reduce the number of chunks balanced on 3472 * resume after balance was interrupted. 3473 */ 3474 static void update_balance_args(struct btrfs_balance_control *bctl) 3475 { 3476 /* 3477 * Turn on soft mode for chunk types that were being converted. 3478 */ 3479 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) 3480 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; 3481 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) 3482 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; 3483 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) 3484 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; 3485 3486 /* 3487 * Turn on usage filter if is not already used. The idea is 3488 * that chunks that we have already balanced should be 3489 * reasonably full. Don't do it for chunks that are being 3490 * converted - that will keep us from relocating unconverted 3491 * (albeit full) chunks. 3492 */ 3493 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && 3494 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3495 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3496 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; 3497 bctl->data.usage = 90; 3498 } 3499 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && 3500 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3501 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3502 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; 3503 bctl->sys.usage = 90; 3504 } 3505 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && 3506 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3507 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { 3508 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; 3509 bctl->meta.usage = 90; 3510 } 3511 } 3512 3513 /* 3514 * Clear the balance status in fs_info and delete the balance item from disk. 3515 */ 3516 static void reset_balance_state(struct btrfs_fs_info *fs_info) 3517 { 3518 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3519 int ret; 3520 3521 BUG_ON(!fs_info->balance_ctl); 3522 3523 spin_lock(&fs_info->balance_lock); 3524 fs_info->balance_ctl = NULL; 3525 spin_unlock(&fs_info->balance_lock); 3526 3527 kfree(bctl); 3528 ret = del_balance_item(fs_info); 3529 if (ret) 3530 btrfs_handle_fs_error(fs_info, ret, NULL); 3531 } 3532 3533 /* 3534 * Balance filters. Return 1 if chunk should be filtered out 3535 * (should not be balanced). 3536 */ 3537 static int chunk_profiles_filter(u64 chunk_type, 3538 struct btrfs_balance_args *bargs) 3539 { 3540 chunk_type = chunk_to_extended(chunk_type) & 3541 BTRFS_EXTENDED_PROFILE_MASK; 3542 3543 if (bargs->profiles & chunk_type) 3544 return 0; 3545 3546 return 1; 3547 } 3548 3549 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, 3550 struct btrfs_balance_args *bargs) 3551 { 3552 struct btrfs_block_group *cache; 3553 u64 chunk_used; 3554 u64 user_thresh_min; 3555 u64 user_thresh_max; 3556 int ret = 1; 3557 3558 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3559 chunk_used = cache->used; 3560 3561 if (bargs->usage_min == 0) 3562 user_thresh_min = 0; 3563 else 3564 user_thresh_min = div_factor_fine(cache->length, 3565 bargs->usage_min); 3566 3567 if (bargs->usage_max == 0) 3568 user_thresh_max = 1; 3569 else if (bargs->usage_max > 100) 3570 user_thresh_max = cache->length; 3571 else 3572 user_thresh_max = div_factor_fine(cache->length, 3573 bargs->usage_max); 3574 3575 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) 3576 ret = 0; 3577 3578 btrfs_put_block_group(cache); 3579 return ret; 3580 } 3581 3582 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, 3583 u64 chunk_offset, struct btrfs_balance_args *bargs) 3584 { 3585 struct btrfs_block_group *cache; 3586 u64 chunk_used, user_thresh; 3587 int ret = 1; 3588 3589 cache = btrfs_lookup_block_group(fs_info, chunk_offset); 3590 chunk_used = cache->used; 3591 3592 if (bargs->usage_min == 0) 3593 user_thresh = 1; 3594 else if (bargs->usage > 100) 3595 user_thresh = cache->length; 3596 else 3597 user_thresh = div_factor_fine(cache->length, bargs->usage); 3598 3599 if (chunk_used < user_thresh) 3600 ret = 0; 3601 3602 btrfs_put_block_group(cache); 3603 return ret; 3604 } 3605 3606 static int chunk_devid_filter(struct extent_buffer *leaf, 3607 struct btrfs_chunk *chunk, 3608 struct btrfs_balance_args *bargs) 3609 { 3610 struct btrfs_stripe *stripe; 3611 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3612 int i; 3613 3614 for (i = 0; i < num_stripes; i++) { 3615 stripe = btrfs_stripe_nr(chunk, i); 3616 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) 3617 return 0; 3618 } 3619 3620 return 1; 3621 } 3622 3623 static u64 calc_data_stripes(u64 type, int num_stripes) 3624 { 3625 const int index = btrfs_bg_flags_to_raid_index(type); 3626 const int ncopies = btrfs_raid_array[index].ncopies; 3627 const int nparity = btrfs_raid_array[index].nparity; 3628 3629 return (num_stripes - nparity) / ncopies; 3630 } 3631 3632 /* [pstart, pend) */ 3633 static int chunk_drange_filter(struct extent_buffer *leaf, 3634 struct btrfs_chunk *chunk, 3635 struct btrfs_balance_args *bargs) 3636 { 3637 struct btrfs_stripe *stripe; 3638 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3639 u64 stripe_offset; 3640 u64 stripe_length; 3641 u64 type; 3642 int factor; 3643 int i; 3644 3645 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) 3646 return 0; 3647 3648 type = btrfs_chunk_type(leaf, chunk); 3649 factor = calc_data_stripes(type, num_stripes); 3650 3651 for (i = 0; i < num_stripes; i++) { 3652 stripe = btrfs_stripe_nr(chunk, i); 3653 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) 3654 continue; 3655 3656 stripe_offset = btrfs_stripe_offset(leaf, stripe); 3657 stripe_length = btrfs_chunk_length(leaf, chunk); 3658 stripe_length = div_u64(stripe_length, factor); 3659 3660 if (stripe_offset < bargs->pend && 3661 stripe_offset + stripe_length > bargs->pstart) 3662 return 0; 3663 } 3664 3665 return 1; 3666 } 3667 3668 /* [vstart, vend) */ 3669 static int chunk_vrange_filter(struct extent_buffer *leaf, 3670 struct btrfs_chunk *chunk, 3671 u64 chunk_offset, 3672 struct btrfs_balance_args *bargs) 3673 { 3674 if (chunk_offset < bargs->vend && 3675 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) 3676 /* at least part of the chunk is inside this vrange */ 3677 return 0; 3678 3679 return 1; 3680 } 3681 3682 static int chunk_stripes_range_filter(struct extent_buffer *leaf, 3683 struct btrfs_chunk *chunk, 3684 struct btrfs_balance_args *bargs) 3685 { 3686 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 3687 3688 if (bargs->stripes_min <= num_stripes 3689 && num_stripes <= bargs->stripes_max) 3690 return 0; 3691 3692 return 1; 3693 } 3694 3695 static int chunk_soft_convert_filter(u64 chunk_type, 3696 struct btrfs_balance_args *bargs) 3697 { 3698 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 3699 return 0; 3700 3701 chunk_type = chunk_to_extended(chunk_type) & 3702 BTRFS_EXTENDED_PROFILE_MASK; 3703 3704 if (bargs->target == chunk_type) 3705 return 1; 3706 3707 return 0; 3708 } 3709 3710 static int should_balance_chunk(struct extent_buffer *leaf, 3711 struct btrfs_chunk *chunk, u64 chunk_offset) 3712 { 3713 struct btrfs_fs_info *fs_info = leaf->fs_info; 3714 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3715 struct btrfs_balance_args *bargs = NULL; 3716 u64 chunk_type = btrfs_chunk_type(leaf, chunk); 3717 3718 /* type filter */ 3719 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & 3720 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { 3721 return 0; 3722 } 3723 3724 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3725 bargs = &bctl->data; 3726 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3727 bargs = &bctl->sys; 3728 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3729 bargs = &bctl->meta; 3730 3731 /* profiles filter */ 3732 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && 3733 chunk_profiles_filter(chunk_type, bargs)) { 3734 return 0; 3735 } 3736 3737 /* usage filter */ 3738 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && 3739 chunk_usage_filter(fs_info, chunk_offset, bargs)) { 3740 return 0; 3741 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && 3742 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { 3743 return 0; 3744 } 3745 3746 /* devid filter */ 3747 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && 3748 chunk_devid_filter(leaf, chunk, bargs)) { 3749 return 0; 3750 } 3751 3752 /* drange filter, makes sense only with devid filter */ 3753 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && 3754 chunk_drange_filter(leaf, chunk, bargs)) { 3755 return 0; 3756 } 3757 3758 /* vrange filter */ 3759 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && 3760 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { 3761 return 0; 3762 } 3763 3764 /* stripes filter */ 3765 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && 3766 chunk_stripes_range_filter(leaf, chunk, bargs)) { 3767 return 0; 3768 } 3769 3770 /* soft profile changing mode */ 3771 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && 3772 chunk_soft_convert_filter(chunk_type, bargs)) { 3773 return 0; 3774 } 3775 3776 /* 3777 * limited by count, must be the last filter 3778 */ 3779 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { 3780 if (bargs->limit == 0) 3781 return 0; 3782 else 3783 bargs->limit--; 3784 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { 3785 /* 3786 * Same logic as the 'limit' filter; the minimum cannot be 3787 * determined here because we do not have the global information 3788 * about the count of all chunks that satisfy the filters. 3789 */ 3790 if (bargs->limit_max == 0) 3791 return 0; 3792 else 3793 bargs->limit_max--; 3794 } 3795 3796 return 1; 3797 } 3798 3799 static int __btrfs_balance(struct btrfs_fs_info *fs_info) 3800 { 3801 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3802 struct btrfs_root *chunk_root = fs_info->chunk_root; 3803 u64 chunk_type; 3804 struct btrfs_chunk *chunk; 3805 struct btrfs_path *path = NULL; 3806 struct btrfs_key key; 3807 struct btrfs_key found_key; 3808 struct extent_buffer *leaf; 3809 int slot; 3810 int ret; 3811 int enospc_errors = 0; 3812 bool counting = true; 3813 /* The single value limit and min/max limits use the same bytes in the */ 3814 u64 limit_data = bctl->data.limit; 3815 u64 limit_meta = bctl->meta.limit; 3816 u64 limit_sys = bctl->sys.limit; 3817 u32 count_data = 0; 3818 u32 count_meta = 0; 3819 u32 count_sys = 0; 3820 int chunk_reserved = 0; 3821 3822 path = btrfs_alloc_path(); 3823 if (!path) { 3824 ret = -ENOMEM; 3825 goto error; 3826 } 3827 3828 /* zero out stat counters */ 3829 spin_lock(&fs_info->balance_lock); 3830 memset(&bctl->stat, 0, sizeof(bctl->stat)); 3831 spin_unlock(&fs_info->balance_lock); 3832 again: 3833 if (!counting) { 3834 /* 3835 * The single value limit and min/max limits use the same bytes 3836 * in the 3837 */ 3838 bctl->data.limit = limit_data; 3839 bctl->meta.limit = limit_meta; 3840 bctl->sys.limit = limit_sys; 3841 } 3842 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 3843 key.offset = (u64)-1; 3844 key.type = BTRFS_CHUNK_ITEM_KEY; 3845 3846 while (1) { 3847 if ((!counting && atomic_read(&fs_info->balance_pause_req)) || 3848 atomic_read(&fs_info->balance_cancel_req)) { 3849 ret = -ECANCELED; 3850 goto error; 3851 } 3852 3853 mutex_lock(&fs_info->reclaim_bgs_lock); 3854 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); 3855 if (ret < 0) { 3856 mutex_unlock(&fs_info->reclaim_bgs_lock); 3857 goto error; 3858 } 3859 3860 /* 3861 * this shouldn't happen, it means the last relocate 3862 * failed 3863 */ 3864 if (ret == 0) 3865 BUG(); /* FIXME break ? */ 3866 3867 ret = btrfs_previous_item(chunk_root, path, 0, 3868 BTRFS_CHUNK_ITEM_KEY); 3869 if (ret) { 3870 mutex_unlock(&fs_info->reclaim_bgs_lock); 3871 ret = 0; 3872 break; 3873 } 3874 3875 leaf = path->nodes[0]; 3876 slot = path->slots[0]; 3877 btrfs_item_key_to_cpu(leaf, &found_key, slot); 3878 3879 if (found_key.objectid != key.objectid) { 3880 mutex_unlock(&fs_info->reclaim_bgs_lock); 3881 break; 3882 } 3883 3884 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 3885 chunk_type = btrfs_chunk_type(leaf, chunk); 3886 3887 if (!counting) { 3888 spin_lock(&fs_info->balance_lock); 3889 bctl->stat.considered++; 3890 spin_unlock(&fs_info->balance_lock); 3891 } 3892 3893 ret = should_balance_chunk(leaf, chunk, found_key.offset); 3894 3895 btrfs_release_path(path); 3896 if (!ret) { 3897 mutex_unlock(&fs_info->reclaim_bgs_lock); 3898 goto loop; 3899 } 3900 3901 if (counting) { 3902 mutex_unlock(&fs_info->reclaim_bgs_lock); 3903 spin_lock(&fs_info->balance_lock); 3904 bctl->stat.expected++; 3905 spin_unlock(&fs_info->balance_lock); 3906 3907 if (chunk_type & BTRFS_BLOCK_GROUP_DATA) 3908 count_data++; 3909 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) 3910 count_sys++; 3911 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) 3912 count_meta++; 3913 3914 goto loop; 3915 } 3916 3917 /* 3918 * Apply limit_min filter, no need to check if the LIMITS 3919 * filter is used, limit_min is 0 by default 3920 */ 3921 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && 3922 count_data < bctl->data.limit_min) 3923 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && 3924 count_meta < bctl->meta.limit_min) 3925 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && 3926 count_sys < bctl->sys.limit_min)) { 3927 mutex_unlock(&fs_info->reclaim_bgs_lock); 3928 goto loop; 3929 } 3930 3931 if (!chunk_reserved) { 3932 /* 3933 * We may be relocating the only data chunk we have, 3934 * which could potentially end up with losing data's 3935 * raid profile, so lets allocate an empty one in 3936 * advance. 3937 */ 3938 ret = btrfs_may_alloc_data_chunk(fs_info, 3939 found_key.offset); 3940 if (ret < 0) { 3941 mutex_unlock(&fs_info->reclaim_bgs_lock); 3942 goto error; 3943 } else if (ret == 1) { 3944 chunk_reserved = 1; 3945 } 3946 } 3947 3948 ret = btrfs_relocate_chunk(fs_info, found_key.offset); 3949 mutex_unlock(&fs_info->reclaim_bgs_lock); 3950 if (ret == -ENOSPC) { 3951 enospc_errors++; 3952 } else if (ret == -ETXTBSY) { 3953 btrfs_info(fs_info, 3954 "skipping relocation of block group %llu due to active swapfile", 3955 found_key.offset); 3956 ret = 0; 3957 } else if (ret) { 3958 goto error; 3959 } else { 3960 spin_lock(&fs_info->balance_lock); 3961 bctl->stat.completed++; 3962 spin_unlock(&fs_info->balance_lock); 3963 } 3964 loop: 3965 if (found_key.offset == 0) 3966 break; 3967 key.offset = found_key.offset - 1; 3968 } 3969 3970 if (counting) { 3971 btrfs_release_path(path); 3972 counting = false; 3973 goto again; 3974 } 3975 error: 3976 btrfs_free_path(path); 3977 if (enospc_errors) { 3978 btrfs_info(fs_info, "%d enospc errors during balance", 3979 enospc_errors); 3980 if (!ret) 3981 ret = -ENOSPC; 3982 } 3983 3984 return ret; 3985 } 3986 3987 /** 3988 * alloc_profile_is_valid - see if a given profile is valid and reduced 3989 * @flags: profile to validate 3990 * @extended: if true @flags is treated as an extended profile 3991 */ 3992 static int alloc_profile_is_valid(u64 flags, int extended) 3993 { 3994 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : 3995 BTRFS_BLOCK_GROUP_PROFILE_MASK); 3996 3997 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; 3998 3999 /* 1) check that all other bits are zeroed */ 4000 if (flags & ~mask) 4001 return 0; 4002 4003 /* 2) see if profile is reduced */ 4004 if (flags == 0) 4005 return !extended; /* "0" is valid for usual profiles */ 4006 4007 return has_single_bit_set(flags); 4008 } 4009 4010 static inline int balance_need_close(struct btrfs_fs_info *fs_info) 4011 { 4012 /* cancel requested || normal exit path */ 4013 return atomic_read(&fs_info->balance_cancel_req) || 4014 (atomic_read(&fs_info->balance_pause_req) == 0 && 4015 atomic_read(&fs_info->balance_cancel_req) == 0); 4016 } 4017 4018 /* 4019 * Validate target profile against allowed profiles and return true if it's OK. 4020 * Otherwise print the error message and return false. 4021 */ 4022 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, 4023 const struct btrfs_balance_args *bargs, 4024 u64 allowed, const char *type) 4025 { 4026 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) 4027 return true; 4028 4029 if (fs_info->sectorsize < PAGE_SIZE && 4030 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) { 4031 btrfs_err(fs_info, 4032 "RAID56 is not yet supported for sectorsize %u with page size %lu", 4033 fs_info->sectorsize, PAGE_SIZE); 4034 return false; 4035 } 4036 /* Profile is valid and does not have bits outside of the allowed set */ 4037 if (alloc_profile_is_valid(bargs->target, 1) && 4038 (bargs->target & ~allowed) == 0) 4039 return true; 4040 4041 btrfs_err(fs_info, "balance: invalid convert %s profile %s", 4042 type, btrfs_bg_type_to_raid_name(bargs->target)); 4043 return false; 4044 } 4045 4046 /* 4047 * Fill @buf with textual description of balance filter flags @bargs, up to 4048 * @size_buf including the terminating null. The output may be trimmed if it 4049 * does not fit into the provided buffer. 4050 */ 4051 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, 4052 u32 size_buf) 4053 { 4054 int ret; 4055 u32 size_bp = size_buf; 4056 char *bp = buf; 4057 u64 flags = bargs->flags; 4058 char tmp_buf[128] = {'\0'}; 4059 4060 if (!flags) 4061 return; 4062 4063 #define CHECK_APPEND_NOARG(a) \ 4064 do { \ 4065 ret = snprintf(bp, size_bp, (a)); \ 4066 if (ret < 0 || ret >= size_bp) \ 4067 goto out_overflow; \ 4068 size_bp -= ret; \ 4069 bp += ret; \ 4070 } while (0) 4071 4072 #define CHECK_APPEND_1ARG(a, v1) \ 4073 do { \ 4074 ret = snprintf(bp, size_bp, (a), (v1)); \ 4075 if (ret < 0 || ret >= size_bp) \ 4076 goto out_overflow; \ 4077 size_bp -= ret; \ 4078 bp += ret; \ 4079 } while (0) 4080 4081 #define CHECK_APPEND_2ARG(a, v1, v2) \ 4082 do { \ 4083 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ 4084 if (ret < 0 || ret >= size_bp) \ 4085 goto out_overflow; \ 4086 size_bp -= ret; \ 4087 bp += ret; \ 4088 } while (0) 4089 4090 if (flags & BTRFS_BALANCE_ARGS_CONVERT) 4091 CHECK_APPEND_1ARG("convert=%s,", 4092 btrfs_bg_type_to_raid_name(bargs->target)); 4093 4094 if (flags & BTRFS_BALANCE_ARGS_SOFT) 4095 CHECK_APPEND_NOARG("soft,"); 4096 4097 if (flags & BTRFS_BALANCE_ARGS_PROFILES) { 4098 btrfs_describe_block_groups(bargs->profiles, tmp_buf, 4099 sizeof(tmp_buf)); 4100 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); 4101 } 4102 4103 if (flags & BTRFS_BALANCE_ARGS_USAGE) 4104 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); 4105 4106 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) 4107 CHECK_APPEND_2ARG("usage=%u..%u,", 4108 bargs->usage_min, bargs->usage_max); 4109 4110 if (flags & BTRFS_BALANCE_ARGS_DEVID) 4111 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); 4112 4113 if (flags & BTRFS_BALANCE_ARGS_DRANGE) 4114 CHECK_APPEND_2ARG("drange=%llu..%llu,", 4115 bargs->pstart, bargs->pend); 4116 4117 if (flags & BTRFS_BALANCE_ARGS_VRANGE) 4118 CHECK_APPEND_2ARG("vrange=%llu..%llu,", 4119 bargs->vstart, bargs->vend); 4120 4121 if (flags & BTRFS_BALANCE_ARGS_LIMIT) 4122 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); 4123 4124 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) 4125 CHECK_APPEND_2ARG("limit=%u..%u,", 4126 bargs->limit_min, bargs->limit_max); 4127 4128 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) 4129 CHECK_APPEND_2ARG("stripes=%u..%u,", 4130 bargs->stripes_min, bargs->stripes_max); 4131 4132 #undef CHECK_APPEND_2ARG 4133 #undef CHECK_APPEND_1ARG 4134 #undef CHECK_APPEND_NOARG 4135 4136 out_overflow: 4137 4138 if (size_bp < size_buf) 4139 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ 4140 else 4141 buf[0] = '\0'; 4142 } 4143 4144 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) 4145 { 4146 u32 size_buf = 1024; 4147 char tmp_buf[192] = {'\0'}; 4148 char *buf; 4149 char *bp; 4150 u32 size_bp = size_buf; 4151 int ret; 4152 struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4153 4154 buf = kzalloc(size_buf, GFP_KERNEL); 4155 if (!buf) 4156 return; 4157 4158 bp = buf; 4159 4160 #define CHECK_APPEND_1ARG(a, v1) \ 4161 do { \ 4162 ret = snprintf(bp, size_bp, (a), (v1)); \ 4163 if (ret < 0 || ret >= size_bp) \ 4164 goto out_overflow; \ 4165 size_bp -= ret; \ 4166 bp += ret; \ 4167 } while (0) 4168 4169 if (bctl->flags & BTRFS_BALANCE_FORCE) 4170 CHECK_APPEND_1ARG("%s", "-f "); 4171 4172 if (bctl->flags & BTRFS_BALANCE_DATA) { 4173 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); 4174 CHECK_APPEND_1ARG("-d%s ", tmp_buf); 4175 } 4176 4177 if (bctl->flags & BTRFS_BALANCE_METADATA) { 4178 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); 4179 CHECK_APPEND_1ARG("-m%s ", tmp_buf); 4180 } 4181 4182 if (bctl->flags & BTRFS_BALANCE_SYSTEM) { 4183 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); 4184 CHECK_APPEND_1ARG("-s%s ", tmp_buf); 4185 } 4186 4187 #undef CHECK_APPEND_1ARG 4188 4189 out_overflow: 4190 4191 if (size_bp < size_buf) 4192 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ 4193 btrfs_info(fs_info, "balance: %s %s", 4194 (bctl->flags & BTRFS_BALANCE_RESUME) ? 4195 "resume" : "start", buf); 4196 4197 kfree(buf); 4198 } 4199 4200 /* 4201 * Should be called with balance mutexe held 4202 */ 4203 int btrfs_balance(struct btrfs_fs_info *fs_info, 4204 struct btrfs_balance_control *bctl, 4205 struct btrfs_ioctl_balance_args *bargs) 4206 { 4207 u64 meta_target, data_target; 4208 u64 allowed; 4209 int mixed = 0; 4210 int ret; 4211 u64 num_devices; 4212 unsigned seq; 4213 bool reducing_redundancy; 4214 int i; 4215 4216 if (btrfs_fs_closing(fs_info) || 4217 atomic_read(&fs_info->balance_pause_req) || 4218 btrfs_should_cancel_balance(fs_info)) { 4219 ret = -EINVAL; 4220 goto out; 4221 } 4222 4223 allowed = btrfs_super_incompat_flags(fs_info->super_copy); 4224 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 4225 mixed = 1; 4226 4227 /* 4228 * In case of mixed groups both data and meta should be picked, 4229 * and identical options should be given for both of them. 4230 */ 4231 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; 4232 if (mixed && (bctl->flags & allowed)) { 4233 if (!(bctl->flags & BTRFS_BALANCE_DATA) || 4234 !(bctl->flags & BTRFS_BALANCE_METADATA) || 4235 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { 4236 btrfs_err(fs_info, 4237 "balance: mixed groups data and metadata options must be the same"); 4238 ret = -EINVAL; 4239 goto out; 4240 } 4241 } 4242 4243 /* 4244 * rw_devices will not change at the moment, device add/delete/replace 4245 * are exclusive 4246 */ 4247 num_devices = fs_info->fs_devices->rw_devices; 4248 4249 /* 4250 * SINGLE profile on-disk has no profile bit, but in-memory we have a 4251 * special bit for it, to make it easier to distinguish. Thus we need 4252 * to set it manually, or balance would refuse the profile. 4253 */ 4254 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; 4255 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) 4256 if (num_devices >= btrfs_raid_array[i].devs_min) 4257 allowed |= btrfs_raid_array[i].bg_flag; 4258 4259 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || 4260 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || 4261 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { 4262 ret = -EINVAL; 4263 goto out; 4264 } 4265 4266 /* 4267 * Allow to reduce metadata or system integrity only if force set for 4268 * profiles with redundancy (copies, parity) 4269 */ 4270 allowed = 0; 4271 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { 4272 if (btrfs_raid_array[i].ncopies >= 2 || 4273 btrfs_raid_array[i].tolerated_failures >= 1) 4274 allowed |= btrfs_raid_array[i].bg_flag; 4275 } 4276 do { 4277 seq = read_seqbegin(&fs_info->profiles_lock); 4278 4279 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4280 (fs_info->avail_system_alloc_bits & allowed) && 4281 !(bctl->sys.target & allowed)) || 4282 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && 4283 (fs_info->avail_metadata_alloc_bits & allowed) && 4284 !(bctl->meta.target & allowed))) 4285 reducing_redundancy = true; 4286 else 4287 reducing_redundancy = false; 4288 4289 /* if we're not converting, the target field is uninitialized */ 4290 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4291 bctl->meta.target : fs_info->avail_metadata_alloc_bits; 4292 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? 4293 bctl->data.target : fs_info->avail_data_alloc_bits; 4294 } while (read_seqretry(&fs_info->profiles_lock, seq)); 4295 4296 if (reducing_redundancy) { 4297 if (bctl->flags & BTRFS_BALANCE_FORCE) { 4298 btrfs_info(fs_info, 4299 "balance: force reducing metadata redundancy"); 4300 } else { 4301 btrfs_err(fs_info, 4302 "balance: reduces metadata redundancy, use --force if you want this"); 4303 ret = -EINVAL; 4304 goto out; 4305 } 4306 } 4307 4308 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < 4309 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { 4310 btrfs_warn(fs_info, 4311 "balance: metadata profile %s has lower redundancy than data profile %s", 4312 btrfs_bg_type_to_raid_name(meta_target), 4313 btrfs_bg_type_to_raid_name(data_target)); 4314 } 4315 4316 ret = insert_balance_item(fs_info, bctl); 4317 if (ret && ret != -EEXIST) 4318 goto out; 4319 4320 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { 4321 BUG_ON(ret == -EEXIST); 4322 BUG_ON(fs_info->balance_ctl); 4323 spin_lock(&fs_info->balance_lock); 4324 fs_info->balance_ctl = bctl; 4325 spin_unlock(&fs_info->balance_lock); 4326 } else { 4327 BUG_ON(ret != -EEXIST); 4328 spin_lock(&fs_info->balance_lock); 4329 update_balance_args(bctl); 4330 spin_unlock(&fs_info->balance_lock); 4331 } 4332 4333 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4334 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4335 describe_balance_start_or_resume(fs_info); 4336 mutex_unlock(&fs_info->balance_mutex); 4337 4338 ret = __btrfs_balance(fs_info); 4339 4340 mutex_lock(&fs_info->balance_mutex); 4341 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) 4342 btrfs_info(fs_info, "balance: paused"); 4343 /* 4344 * Balance can be canceled by: 4345 * 4346 * - Regular cancel request 4347 * Then ret == -ECANCELED and balance_cancel_req > 0 4348 * 4349 * - Fatal signal to "btrfs" process 4350 * Either the signal caught by wait_reserve_ticket() and callers 4351 * got -EINTR, or caught by btrfs_should_cancel_balance() and 4352 * got -ECANCELED. 4353 * Either way, in this case balance_cancel_req = 0, and 4354 * ret == -EINTR or ret == -ECANCELED. 4355 * 4356 * So here we only check the return value to catch canceled balance. 4357 */ 4358 else if (ret == -ECANCELED || ret == -EINTR) 4359 btrfs_info(fs_info, "balance: canceled"); 4360 else 4361 btrfs_info(fs_info, "balance: ended with status: %d", ret); 4362 4363 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); 4364 4365 if (bargs) { 4366 memset(bargs, 0, sizeof(*bargs)); 4367 btrfs_update_ioctl_balance_args(fs_info, bargs); 4368 } 4369 4370 if ((ret && ret != -ECANCELED && ret != -ENOSPC) || 4371 balance_need_close(fs_info)) { 4372 reset_balance_state(fs_info); 4373 btrfs_exclop_finish(fs_info); 4374 } 4375 4376 wake_up(&fs_info->balance_wait_q); 4377 4378 return ret; 4379 out: 4380 if (bctl->flags & BTRFS_BALANCE_RESUME) 4381 reset_balance_state(fs_info); 4382 else 4383 kfree(bctl); 4384 btrfs_exclop_finish(fs_info); 4385 4386 return ret; 4387 } 4388 4389 static int balance_kthread(void *data) 4390 { 4391 struct btrfs_fs_info *fs_info = data; 4392 int ret = 0; 4393 4394 mutex_lock(&fs_info->balance_mutex); 4395 if (fs_info->balance_ctl) 4396 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); 4397 mutex_unlock(&fs_info->balance_mutex); 4398 4399 return ret; 4400 } 4401 4402 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) 4403 { 4404 struct task_struct *tsk; 4405 4406 mutex_lock(&fs_info->balance_mutex); 4407 if (!fs_info->balance_ctl) { 4408 mutex_unlock(&fs_info->balance_mutex); 4409 return 0; 4410 } 4411 mutex_unlock(&fs_info->balance_mutex); 4412 4413 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { 4414 btrfs_info(fs_info, "balance: resume skipped"); 4415 return 0; 4416 } 4417 4418 /* 4419 * A ro->rw remount sequence should continue with the paused balance 4420 * regardless of who pauses it, system or the user as of now, so set 4421 * the resume flag. 4422 */ 4423 spin_lock(&fs_info->balance_lock); 4424 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; 4425 spin_unlock(&fs_info->balance_lock); 4426 4427 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); 4428 return PTR_ERR_OR_ZERO(tsk); 4429 } 4430 4431 int btrfs_recover_balance(struct btrfs_fs_info *fs_info) 4432 { 4433 struct btrfs_balance_control *bctl; 4434 struct btrfs_balance_item *item; 4435 struct btrfs_disk_balance_args disk_bargs; 4436 struct btrfs_path *path; 4437 struct extent_buffer *leaf; 4438 struct btrfs_key key; 4439 int ret; 4440 4441 path = btrfs_alloc_path(); 4442 if (!path) 4443 return -ENOMEM; 4444 4445 key.objectid = BTRFS_BALANCE_OBJECTID; 4446 key.type = BTRFS_TEMPORARY_ITEM_KEY; 4447 key.offset = 0; 4448 4449 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 4450 if (ret < 0) 4451 goto out; 4452 if (ret > 0) { /* ret = -ENOENT; */ 4453 ret = 0; 4454 goto out; 4455 } 4456 4457 bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4458 if (!bctl) { 4459 ret = -ENOMEM; 4460 goto out; 4461 } 4462 4463 leaf = path->nodes[0]; 4464 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); 4465 4466 bctl->flags = btrfs_balance_flags(leaf, item); 4467 bctl->flags |= BTRFS_BALANCE_RESUME; 4468 4469 btrfs_balance_data(leaf, item, &disk_bargs); 4470 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); 4471 btrfs_balance_meta(leaf, item, &disk_bargs); 4472 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); 4473 btrfs_balance_sys(leaf, item, &disk_bargs); 4474 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); 4475 4476 /* 4477 * This should never happen, as the paused balance state is recovered 4478 * during mount without any chance of other exclusive ops to collide. 4479 * 4480 * This gives the exclusive op status to balance and keeps in paused 4481 * state until user intervention (cancel or umount). If the ownership 4482 * cannot be assigned, show a message but do not fail. The balance 4483 * is in a paused state and must have fs_info::balance_ctl properly 4484 * set up. 4485 */ 4486 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) 4487 btrfs_warn(fs_info, 4488 "balance: cannot set exclusive op status, resume manually"); 4489 4490 btrfs_release_path(path); 4491 4492 mutex_lock(&fs_info->balance_mutex); 4493 BUG_ON(fs_info->balance_ctl); 4494 spin_lock(&fs_info->balance_lock); 4495 fs_info->balance_ctl = bctl; 4496 spin_unlock(&fs_info->balance_lock); 4497 mutex_unlock(&fs_info->balance_mutex); 4498 out: 4499 btrfs_free_path(path); 4500 return ret; 4501 } 4502 4503 int btrfs_pause_balance(struct btrfs_fs_info *fs_info) 4504 { 4505 int ret = 0; 4506 4507 mutex_lock(&fs_info->balance_mutex); 4508 if (!fs_info->balance_ctl) { 4509 mutex_unlock(&fs_info->balance_mutex); 4510 return -ENOTCONN; 4511 } 4512 4513 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4514 atomic_inc(&fs_info->balance_pause_req); 4515 mutex_unlock(&fs_info->balance_mutex); 4516 4517 wait_event(fs_info->balance_wait_q, 4518 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4519 4520 mutex_lock(&fs_info->balance_mutex); 4521 /* we are good with balance_ctl ripped off from under us */ 4522 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4523 atomic_dec(&fs_info->balance_pause_req); 4524 } else { 4525 ret = -ENOTCONN; 4526 } 4527 4528 mutex_unlock(&fs_info->balance_mutex); 4529 return ret; 4530 } 4531 4532 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) 4533 { 4534 mutex_lock(&fs_info->balance_mutex); 4535 if (!fs_info->balance_ctl) { 4536 mutex_unlock(&fs_info->balance_mutex); 4537 return -ENOTCONN; 4538 } 4539 4540 /* 4541 * A paused balance with the item stored on disk can be resumed at 4542 * mount time if the mount is read-write. Otherwise it's still paused 4543 * and we must not allow cancelling as it deletes the item. 4544 */ 4545 if (sb_rdonly(fs_info->sb)) { 4546 mutex_unlock(&fs_info->balance_mutex); 4547 return -EROFS; 4548 } 4549 4550 atomic_inc(&fs_info->balance_cancel_req); 4551 /* 4552 * if we are running just wait and return, balance item is 4553 * deleted in btrfs_balance in this case 4554 */ 4555 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4556 mutex_unlock(&fs_info->balance_mutex); 4557 wait_event(fs_info->balance_wait_q, 4558 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4559 mutex_lock(&fs_info->balance_mutex); 4560 } else { 4561 mutex_unlock(&fs_info->balance_mutex); 4562 /* 4563 * Lock released to allow other waiters to continue, we'll 4564 * reexamine the status again. 4565 */ 4566 mutex_lock(&fs_info->balance_mutex); 4567 4568 if (fs_info->balance_ctl) { 4569 reset_balance_state(fs_info); 4570 btrfs_exclop_finish(fs_info); 4571 btrfs_info(fs_info, "balance: canceled"); 4572 } 4573 } 4574 4575 BUG_ON(fs_info->balance_ctl || 4576 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); 4577 atomic_dec(&fs_info->balance_cancel_req); 4578 mutex_unlock(&fs_info->balance_mutex); 4579 return 0; 4580 } 4581 4582 int btrfs_uuid_scan_kthread(void *data) 4583 { 4584 struct btrfs_fs_info *fs_info = data; 4585 struct btrfs_root *root = fs_info->tree_root; 4586 struct btrfs_key key; 4587 struct btrfs_path *path = NULL; 4588 int ret = 0; 4589 struct extent_buffer *eb; 4590 int slot; 4591 struct btrfs_root_item root_item; 4592 u32 item_size; 4593 struct btrfs_trans_handle *trans = NULL; 4594 bool closing = false; 4595 4596 path = btrfs_alloc_path(); 4597 if (!path) { 4598 ret = -ENOMEM; 4599 goto out; 4600 } 4601 4602 key.objectid = 0; 4603 key.type = BTRFS_ROOT_ITEM_KEY; 4604 key.offset = 0; 4605 4606 while (1) { 4607 if (btrfs_fs_closing(fs_info)) { 4608 closing = true; 4609 break; 4610 } 4611 ret = btrfs_search_forward(root, &key, path, 4612 BTRFS_OLDEST_GENERATION); 4613 if (ret) { 4614 if (ret > 0) 4615 ret = 0; 4616 break; 4617 } 4618 4619 if (key.type != BTRFS_ROOT_ITEM_KEY || 4620 (key.objectid < BTRFS_FIRST_FREE_OBJECTID && 4621 key.objectid != BTRFS_FS_TREE_OBJECTID) || 4622 key.objectid > BTRFS_LAST_FREE_OBJECTID) 4623 goto skip; 4624 4625 eb = path->nodes[0]; 4626 slot = path->slots[0]; 4627 item_size = btrfs_item_size_nr(eb, slot); 4628 if (item_size < sizeof(root_item)) 4629 goto skip; 4630 4631 read_extent_buffer(eb, &root_item, 4632 btrfs_item_ptr_offset(eb, slot), 4633 (int)sizeof(root_item)); 4634 if (btrfs_root_refs(&root_item) == 0) 4635 goto skip; 4636 4637 if (!btrfs_is_empty_uuid(root_item.uuid) || 4638 !btrfs_is_empty_uuid(root_item.received_uuid)) { 4639 if (trans) 4640 goto update_tree; 4641 4642 btrfs_release_path(path); 4643 /* 4644 * 1 - subvol uuid item 4645 * 1 - received_subvol uuid item 4646 */ 4647 trans = btrfs_start_transaction(fs_info->uuid_root, 2); 4648 if (IS_ERR(trans)) { 4649 ret = PTR_ERR(trans); 4650 break; 4651 } 4652 continue; 4653 } else { 4654 goto skip; 4655 } 4656 update_tree: 4657 btrfs_release_path(path); 4658 if (!btrfs_is_empty_uuid(root_item.uuid)) { 4659 ret = btrfs_uuid_tree_add(trans, root_item.uuid, 4660 BTRFS_UUID_KEY_SUBVOL, 4661 key.objectid); 4662 if (ret < 0) { 4663 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4664 ret); 4665 break; 4666 } 4667 } 4668 4669 if (!btrfs_is_empty_uuid(root_item.received_uuid)) { 4670 ret = btrfs_uuid_tree_add(trans, 4671 root_item.received_uuid, 4672 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4673 key.objectid); 4674 if (ret < 0) { 4675 btrfs_warn(fs_info, "uuid_tree_add failed %d", 4676 ret); 4677 break; 4678 } 4679 } 4680 4681 skip: 4682 btrfs_release_path(path); 4683 if (trans) { 4684 ret = btrfs_end_transaction(trans); 4685 trans = NULL; 4686 if (ret) 4687 break; 4688 } 4689 4690 if (key.offset < (u64)-1) { 4691 key.offset++; 4692 } else if (key.type < BTRFS_ROOT_ITEM_KEY) { 4693 key.offset = 0; 4694 key.type = BTRFS_ROOT_ITEM_KEY; 4695 } else if (key.objectid < (u64)-1) { 4696 key.offset = 0; 4697 key.type = BTRFS_ROOT_ITEM_KEY; 4698 key.objectid++; 4699 } else { 4700 break; 4701 } 4702 cond_resched(); 4703 } 4704 4705 out: 4706 btrfs_free_path(path); 4707 if (trans && !IS_ERR(trans)) 4708 btrfs_end_transaction(trans); 4709 if (ret) 4710 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); 4711 else if (!closing) 4712 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); 4713 up(&fs_info->uuid_tree_rescan_sem); 4714 return 0; 4715 } 4716 4717 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) 4718 { 4719 struct btrfs_trans_handle *trans; 4720 struct btrfs_root *tree_root = fs_info->tree_root; 4721 struct btrfs_root *uuid_root; 4722 struct task_struct *task; 4723 int ret; 4724 4725 /* 4726 * 1 - root node 4727 * 1 - root item 4728 */ 4729 trans = btrfs_start_transaction(tree_root, 2); 4730 if (IS_ERR(trans)) 4731 return PTR_ERR(trans); 4732 4733 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); 4734 if (IS_ERR(uuid_root)) { 4735 ret = PTR_ERR(uuid_root); 4736 btrfs_abort_transaction(trans, ret); 4737 btrfs_end_transaction(trans); 4738 return ret; 4739 } 4740 4741 fs_info->uuid_root = uuid_root; 4742 4743 ret = btrfs_commit_transaction(trans); 4744 if (ret) 4745 return ret; 4746 4747 down(&fs_info->uuid_tree_rescan_sem); 4748 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); 4749 if (IS_ERR(task)) { 4750 /* fs_info->update_uuid_tree_gen remains 0 in all error case */ 4751 btrfs_warn(fs_info, "failed to start uuid_scan task"); 4752 up(&fs_info->uuid_tree_rescan_sem); 4753 return PTR_ERR(task); 4754 } 4755 4756 return 0; 4757 } 4758 4759 /* 4760 * shrinking a device means finding all of the device extents past 4761 * the new size, and then following the back refs to the chunks. 4762 * The chunk relocation code actually frees the device extent 4763 */ 4764 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) 4765 { 4766 struct btrfs_fs_info *fs_info = device->fs_info; 4767 struct btrfs_root *root = fs_info->dev_root; 4768 struct btrfs_trans_handle *trans; 4769 struct btrfs_dev_extent *dev_extent = NULL; 4770 struct btrfs_path *path; 4771 u64 length; 4772 u64 chunk_offset; 4773 int ret; 4774 int slot; 4775 int failed = 0; 4776 bool retried = false; 4777 struct extent_buffer *l; 4778 struct btrfs_key key; 4779 struct btrfs_super_block *super_copy = fs_info->super_copy; 4780 u64 old_total = btrfs_super_total_bytes(super_copy); 4781 u64 old_size = btrfs_device_get_total_bytes(device); 4782 u64 diff; 4783 u64 start; 4784 4785 new_size = round_down(new_size, fs_info->sectorsize); 4786 start = new_size; 4787 diff = round_down(old_size - new_size, fs_info->sectorsize); 4788 4789 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 4790 return -EINVAL; 4791 4792 path = btrfs_alloc_path(); 4793 if (!path) 4794 return -ENOMEM; 4795 4796 path->reada = READA_BACK; 4797 4798 trans = btrfs_start_transaction(root, 0); 4799 if (IS_ERR(trans)) { 4800 btrfs_free_path(path); 4801 return PTR_ERR(trans); 4802 } 4803 4804 mutex_lock(&fs_info->chunk_mutex); 4805 4806 btrfs_device_set_total_bytes(device, new_size); 4807 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 4808 device->fs_devices->total_rw_bytes -= diff; 4809 atomic64_sub(diff, &fs_info->free_chunk_space); 4810 } 4811 4812 /* 4813 * Once the device's size has been set to the new size, ensure all 4814 * in-memory chunks are synced to disk so that the loop below sees them 4815 * and relocates them accordingly. 4816 */ 4817 if (contains_pending_extent(device, &start, diff)) { 4818 mutex_unlock(&fs_info->chunk_mutex); 4819 ret = btrfs_commit_transaction(trans); 4820 if (ret) 4821 goto done; 4822 } else { 4823 mutex_unlock(&fs_info->chunk_mutex); 4824 btrfs_end_transaction(trans); 4825 } 4826 4827 again: 4828 key.objectid = device->devid; 4829 key.offset = (u64)-1; 4830 key.type = BTRFS_DEV_EXTENT_KEY; 4831 4832 do { 4833 mutex_lock(&fs_info->reclaim_bgs_lock); 4834 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4835 if (ret < 0) { 4836 mutex_unlock(&fs_info->reclaim_bgs_lock); 4837 goto done; 4838 } 4839 4840 ret = btrfs_previous_item(root, path, 0, key.type); 4841 if (ret) { 4842 mutex_unlock(&fs_info->reclaim_bgs_lock); 4843 if (ret < 0) 4844 goto done; 4845 ret = 0; 4846 btrfs_release_path(path); 4847 break; 4848 } 4849 4850 l = path->nodes[0]; 4851 slot = path->slots[0]; 4852 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 4853 4854 if (key.objectid != device->devid) { 4855 mutex_unlock(&fs_info->reclaim_bgs_lock); 4856 btrfs_release_path(path); 4857 break; 4858 } 4859 4860 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); 4861 length = btrfs_dev_extent_length(l, dev_extent); 4862 4863 if (key.offset + length <= new_size) { 4864 mutex_unlock(&fs_info->reclaim_bgs_lock); 4865 btrfs_release_path(path); 4866 break; 4867 } 4868 4869 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); 4870 btrfs_release_path(path); 4871 4872 /* 4873 * We may be relocating the only data chunk we have, 4874 * which could potentially end up with losing data's 4875 * raid profile, so lets allocate an empty one in 4876 * advance. 4877 */ 4878 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); 4879 if (ret < 0) { 4880 mutex_unlock(&fs_info->reclaim_bgs_lock); 4881 goto done; 4882 } 4883 4884 ret = btrfs_relocate_chunk(fs_info, chunk_offset); 4885 mutex_unlock(&fs_info->reclaim_bgs_lock); 4886 if (ret == -ENOSPC) { 4887 failed++; 4888 } else if (ret) { 4889 if (ret == -ETXTBSY) { 4890 btrfs_warn(fs_info, 4891 "could not shrink block group %llu due to active swapfile", 4892 chunk_offset); 4893 } 4894 goto done; 4895 } 4896 } while (key.offset-- > 0); 4897 4898 if (failed && !retried) { 4899 failed = 0; 4900 retried = true; 4901 goto again; 4902 } else if (failed && retried) { 4903 ret = -ENOSPC; 4904 goto done; 4905 } 4906 4907 /* Shrinking succeeded, else we would be at "done". */ 4908 trans = btrfs_start_transaction(root, 0); 4909 if (IS_ERR(trans)) { 4910 ret = PTR_ERR(trans); 4911 goto done; 4912 } 4913 4914 mutex_lock(&fs_info->chunk_mutex); 4915 /* Clear all state bits beyond the shrunk device size */ 4916 clear_extent_bits(&device->alloc_state, new_size, (u64)-1, 4917 CHUNK_STATE_MASK); 4918 4919 btrfs_device_set_disk_total_bytes(device, new_size); 4920 if (list_empty(&device->post_commit_list)) 4921 list_add_tail(&device->post_commit_list, 4922 &trans->transaction->dev_update_list); 4923 4924 WARN_ON(diff > old_total); 4925 btrfs_set_super_total_bytes(super_copy, 4926 round_down(old_total - diff, fs_info->sectorsize)); 4927 mutex_unlock(&fs_info->chunk_mutex); 4928 4929 /* Now btrfs_update_device() will change the on-disk size. */ 4930 ret = btrfs_update_device(trans, device); 4931 if (ret < 0) { 4932 btrfs_abort_transaction(trans, ret); 4933 btrfs_end_transaction(trans); 4934 } else { 4935 ret = btrfs_commit_transaction(trans); 4936 } 4937 done: 4938 btrfs_free_path(path); 4939 if (ret) { 4940 mutex_lock(&fs_info->chunk_mutex); 4941 btrfs_device_set_total_bytes(device, old_size); 4942 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) 4943 device->fs_devices->total_rw_bytes += diff; 4944 atomic64_add(diff, &fs_info->free_chunk_space); 4945 mutex_unlock(&fs_info->chunk_mutex); 4946 } 4947 return ret; 4948 } 4949 4950 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, 4951 struct btrfs_key *key, 4952 struct btrfs_chunk *chunk, int item_size) 4953 { 4954 struct btrfs_super_block *super_copy = fs_info->super_copy; 4955 struct btrfs_disk_key disk_key; 4956 u32 array_size; 4957 u8 *ptr; 4958 4959 lockdep_assert_held(&fs_info->chunk_mutex); 4960 4961 array_size = btrfs_super_sys_array_size(super_copy); 4962 if (array_size + item_size + sizeof(disk_key) 4963 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) 4964 return -EFBIG; 4965 4966 ptr = super_copy->sys_chunk_array + array_size; 4967 btrfs_cpu_key_to_disk(&disk_key, key); 4968 memcpy(ptr, &disk_key, sizeof(disk_key)); 4969 ptr += sizeof(disk_key); 4970 memcpy(ptr, chunk, item_size); 4971 item_size += sizeof(disk_key); 4972 btrfs_set_super_sys_array_size(super_copy, array_size + item_size); 4973 4974 return 0; 4975 } 4976 4977 /* 4978 * sort the devices in descending order by max_avail, total_avail 4979 */ 4980 static int btrfs_cmp_device_info(const void *a, const void *b) 4981 { 4982 const struct btrfs_device_info *di_a = a; 4983 const struct btrfs_device_info *di_b = b; 4984 4985 if (di_a->max_avail > di_b->max_avail) 4986 return -1; 4987 if (di_a->max_avail < di_b->max_avail) 4988 return 1; 4989 if (di_a->total_avail > di_b->total_avail) 4990 return -1; 4991 if (di_a->total_avail < di_b->total_avail) 4992 return 1; 4993 return 0; 4994 } 4995 4996 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) 4997 { 4998 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) 4999 return; 5000 5001 btrfs_set_fs_incompat(info, RAID56); 5002 } 5003 5004 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) 5005 { 5006 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) 5007 return; 5008 5009 btrfs_set_fs_incompat(info, RAID1C34); 5010 } 5011 5012 /* 5013 * Structure used internally for btrfs_create_chunk() function. 5014 * Wraps needed parameters. 5015 */ 5016 struct alloc_chunk_ctl { 5017 u64 start; 5018 u64 type; 5019 /* Total number of stripes to allocate */ 5020 int num_stripes; 5021 /* sub_stripes info for map */ 5022 int sub_stripes; 5023 /* Stripes per device */ 5024 int dev_stripes; 5025 /* Maximum number of devices to use */ 5026 int devs_max; 5027 /* Minimum number of devices to use */ 5028 int devs_min; 5029 /* ndevs has to be a multiple of this */ 5030 int devs_increment; 5031 /* Number of copies */ 5032 int ncopies; 5033 /* Number of stripes worth of bytes to store parity information */ 5034 int nparity; 5035 u64 max_stripe_size; 5036 u64 max_chunk_size; 5037 u64 dev_extent_min; 5038 u64 stripe_size; 5039 u64 chunk_size; 5040 int ndevs; 5041 }; 5042 5043 static void init_alloc_chunk_ctl_policy_regular( 5044 struct btrfs_fs_devices *fs_devices, 5045 struct alloc_chunk_ctl *ctl) 5046 { 5047 u64 type = ctl->type; 5048 5049 if (type & BTRFS_BLOCK_GROUP_DATA) { 5050 ctl->max_stripe_size = SZ_1G; 5051 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; 5052 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5053 /* For larger filesystems, use larger metadata chunks */ 5054 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) 5055 ctl->max_stripe_size = SZ_1G; 5056 else 5057 ctl->max_stripe_size = SZ_256M; 5058 ctl->max_chunk_size = ctl->max_stripe_size; 5059 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5060 ctl->max_stripe_size = SZ_32M; 5061 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5062 ctl->devs_max = min_t(int, ctl->devs_max, 5063 BTRFS_MAX_DEVS_SYS_CHUNK); 5064 } else { 5065 BUG(); 5066 } 5067 5068 /* We don't want a chunk larger than 10% of writable space */ 5069 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), 5070 ctl->max_chunk_size); 5071 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; 5072 } 5073 5074 static void init_alloc_chunk_ctl_policy_zoned( 5075 struct btrfs_fs_devices *fs_devices, 5076 struct alloc_chunk_ctl *ctl) 5077 { 5078 u64 zone_size = fs_devices->fs_info->zone_size; 5079 u64 limit; 5080 int min_num_stripes = ctl->devs_min * ctl->dev_stripes; 5081 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; 5082 u64 min_chunk_size = min_data_stripes * zone_size; 5083 u64 type = ctl->type; 5084 5085 ctl->max_stripe_size = zone_size; 5086 if (type & BTRFS_BLOCK_GROUP_DATA) { 5087 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, 5088 zone_size); 5089 } else if (type & BTRFS_BLOCK_GROUP_METADATA) { 5090 ctl->max_chunk_size = ctl->max_stripe_size; 5091 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { 5092 ctl->max_chunk_size = 2 * ctl->max_stripe_size; 5093 ctl->devs_max = min_t(int, ctl->devs_max, 5094 BTRFS_MAX_DEVS_SYS_CHUNK); 5095 } else { 5096 BUG(); 5097 } 5098 5099 /* We don't want a chunk larger than 10% of writable space */ 5100 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), 5101 zone_size), 5102 min_chunk_size); 5103 ctl->max_chunk_size = min(limit, ctl->max_chunk_size); 5104 ctl->dev_extent_min = zone_size * ctl->dev_stripes; 5105 } 5106 5107 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, 5108 struct alloc_chunk_ctl *ctl) 5109 { 5110 int index = btrfs_bg_flags_to_raid_index(ctl->type); 5111 5112 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; 5113 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; 5114 ctl->devs_max = btrfs_raid_array[index].devs_max; 5115 if (!ctl->devs_max) 5116 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); 5117 ctl->devs_min = btrfs_raid_array[index].devs_min; 5118 ctl->devs_increment = btrfs_raid_array[index].devs_increment; 5119 ctl->ncopies = btrfs_raid_array[index].ncopies; 5120 ctl->nparity = btrfs_raid_array[index].nparity; 5121 ctl->ndevs = 0; 5122 5123 switch (fs_devices->chunk_alloc_policy) { 5124 case BTRFS_CHUNK_ALLOC_REGULAR: 5125 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); 5126 break; 5127 case BTRFS_CHUNK_ALLOC_ZONED: 5128 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); 5129 break; 5130 default: 5131 BUG(); 5132 } 5133 } 5134 5135 static int gather_device_info(struct btrfs_fs_devices *fs_devices, 5136 struct alloc_chunk_ctl *ctl, 5137 struct btrfs_device_info *devices_info) 5138 { 5139 struct btrfs_fs_info *info = fs_devices->fs_info; 5140 struct btrfs_device *device; 5141 u64 total_avail; 5142 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; 5143 int ret; 5144 int ndevs = 0; 5145 u64 max_avail; 5146 u64 dev_offset; 5147 5148 /* 5149 * in the first pass through the devices list, we gather information 5150 * about the available holes on each device. 5151 */ 5152 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 5153 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 5154 WARN(1, KERN_ERR 5155 "BTRFS: read-only device in alloc_list\n"); 5156 continue; 5157 } 5158 5159 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 5160 &device->dev_state) || 5161 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) 5162 continue; 5163 5164 if (device->total_bytes > device->bytes_used) 5165 total_avail = device->total_bytes - device->bytes_used; 5166 else 5167 total_avail = 0; 5168 5169 /* If there is no space on this device, skip it. */ 5170 if (total_avail < ctl->dev_extent_min) 5171 continue; 5172 5173 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, 5174 &max_avail); 5175 if (ret && ret != -ENOSPC) 5176 return ret; 5177 5178 if (ret == 0) 5179 max_avail = dev_extent_want; 5180 5181 if (max_avail < ctl->dev_extent_min) { 5182 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5183 btrfs_debug(info, 5184 "%s: devid %llu has no free space, have=%llu want=%llu", 5185 __func__, device->devid, max_avail, 5186 ctl->dev_extent_min); 5187 continue; 5188 } 5189 5190 if (ndevs == fs_devices->rw_devices) { 5191 WARN(1, "%s: found more than %llu devices\n", 5192 __func__, fs_devices->rw_devices); 5193 break; 5194 } 5195 devices_info[ndevs].dev_offset = dev_offset; 5196 devices_info[ndevs].max_avail = max_avail; 5197 devices_info[ndevs].total_avail = total_avail; 5198 devices_info[ndevs].dev = device; 5199 ++ndevs; 5200 } 5201 ctl->ndevs = ndevs; 5202 5203 /* 5204 * now sort the devices by hole size / available space 5205 */ 5206 sort(devices_info, ndevs, sizeof(struct btrfs_device_info), 5207 btrfs_cmp_device_info, NULL); 5208 5209 return 0; 5210 } 5211 5212 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, 5213 struct btrfs_device_info *devices_info) 5214 { 5215 /* Number of stripes that count for block group size */ 5216 int data_stripes; 5217 5218 /* 5219 * The primary goal is to maximize the number of stripes, so use as 5220 * many devices as possible, even if the stripes are not maximum sized. 5221 * 5222 * The DUP profile stores more than one stripe per device, the 5223 * max_avail is the total size so we have to adjust. 5224 */ 5225 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, 5226 ctl->dev_stripes); 5227 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5228 5229 /* This will have to be fixed for RAID1 and RAID10 over more drives */ 5230 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5231 5232 /* 5233 * Use the number of data stripes to figure out how big this chunk is 5234 * really going to be in terms of logical address space, and compare 5235 * that answer with the max chunk size. If it's higher, we try to 5236 * reduce stripe_size. 5237 */ 5238 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5239 /* 5240 * Reduce stripe_size, round it up to a 16MB boundary again and 5241 * then use it, unless it ends up being even bigger than the 5242 * previous value we had already. 5243 */ 5244 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, 5245 data_stripes), SZ_16M), 5246 ctl->stripe_size); 5247 } 5248 5249 /* Align to BTRFS_STRIPE_LEN */ 5250 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); 5251 ctl->chunk_size = ctl->stripe_size * data_stripes; 5252 5253 return 0; 5254 } 5255 5256 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, 5257 struct btrfs_device_info *devices_info) 5258 { 5259 u64 zone_size = devices_info[0].dev->zone_info->zone_size; 5260 /* Number of stripes that count for block group size */ 5261 int data_stripes; 5262 5263 /* 5264 * It should hold because: 5265 * dev_extent_min == dev_extent_want == zone_size * dev_stripes 5266 */ 5267 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); 5268 5269 ctl->stripe_size = zone_size; 5270 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5271 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5272 5273 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ 5274 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { 5275 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, 5276 ctl->stripe_size) + ctl->nparity, 5277 ctl->dev_stripes); 5278 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; 5279 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; 5280 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); 5281 } 5282 5283 ctl->chunk_size = ctl->stripe_size * data_stripes; 5284 5285 return 0; 5286 } 5287 5288 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, 5289 struct alloc_chunk_ctl *ctl, 5290 struct btrfs_device_info *devices_info) 5291 { 5292 struct btrfs_fs_info *info = fs_devices->fs_info; 5293 5294 /* 5295 * Round down to number of usable stripes, devs_increment can be any 5296 * number so we can't use round_down() that requires power of 2, while 5297 * rounddown is safe. 5298 */ 5299 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); 5300 5301 if (ctl->ndevs < ctl->devs_min) { 5302 if (btrfs_test_opt(info, ENOSPC_DEBUG)) { 5303 btrfs_debug(info, 5304 "%s: not enough devices with free space: have=%d minimum required=%d", 5305 __func__, ctl->ndevs, ctl->devs_min); 5306 } 5307 return -ENOSPC; 5308 } 5309 5310 ctl->ndevs = min(ctl->ndevs, ctl->devs_max); 5311 5312 switch (fs_devices->chunk_alloc_policy) { 5313 case BTRFS_CHUNK_ALLOC_REGULAR: 5314 return decide_stripe_size_regular(ctl, devices_info); 5315 case BTRFS_CHUNK_ALLOC_ZONED: 5316 return decide_stripe_size_zoned(ctl, devices_info); 5317 default: 5318 BUG(); 5319 } 5320 } 5321 5322 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, 5323 struct alloc_chunk_ctl *ctl, 5324 struct btrfs_device_info *devices_info) 5325 { 5326 struct btrfs_fs_info *info = trans->fs_info; 5327 struct map_lookup *map = NULL; 5328 struct extent_map_tree *em_tree; 5329 struct btrfs_block_group *block_group; 5330 struct extent_map *em; 5331 u64 start = ctl->start; 5332 u64 type = ctl->type; 5333 int ret; 5334 int i; 5335 int j; 5336 5337 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); 5338 if (!map) 5339 return ERR_PTR(-ENOMEM); 5340 map->num_stripes = ctl->num_stripes; 5341 5342 for (i = 0; i < ctl->ndevs; ++i) { 5343 for (j = 0; j < ctl->dev_stripes; ++j) { 5344 int s = i * ctl->dev_stripes + j; 5345 map->stripes[s].dev = devices_info[i].dev; 5346 map->stripes[s].physical = devices_info[i].dev_offset + 5347 j * ctl->stripe_size; 5348 } 5349 } 5350 map->stripe_len = BTRFS_STRIPE_LEN; 5351 map->io_align = BTRFS_STRIPE_LEN; 5352 map->io_width = BTRFS_STRIPE_LEN; 5353 map->type = type; 5354 map->sub_stripes = ctl->sub_stripes; 5355 5356 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); 5357 5358 em = alloc_extent_map(); 5359 if (!em) { 5360 kfree(map); 5361 return ERR_PTR(-ENOMEM); 5362 } 5363 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 5364 em->map_lookup = map; 5365 em->start = start; 5366 em->len = ctl->chunk_size; 5367 em->block_start = 0; 5368 em->block_len = em->len; 5369 em->orig_block_len = ctl->stripe_size; 5370 5371 em_tree = &info->mapping_tree; 5372 write_lock(&em_tree->lock); 5373 ret = add_extent_mapping(em_tree, em, 0); 5374 if (ret) { 5375 write_unlock(&em_tree->lock); 5376 free_extent_map(em); 5377 return ERR_PTR(ret); 5378 } 5379 write_unlock(&em_tree->lock); 5380 5381 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); 5382 if (IS_ERR(block_group)) 5383 goto error_del_extent; 5384 5385 for (i = 0; i < map->num_stripes; i++) { 5386 struct btrfs_device *dev = map->stripes[i].dev; 5387 5388 btrfs_device_set_bytes_used(dev, 5389 dev->bytes_used + ctl->stripe_size); 5390 if (list_empty(&dev->post_commit_list)) 5391 list_add_tail(&dev->post_commit_list, 5392 &trans->transaction->dev_update_list); 5393 } 5394 5395 atomic64_sub(ctl->stripe_size * map->num_stripes, 5396 &info->free_chunk_space); 5397 5398 free_extent_map(em); 5399 check_raid56_incompat_flag(info, type); 5400 check_raid1c34_incompat_flag(info, type); 5401 5402 return block_group; 5403 5404 error_del_extent: 5405 write_lock(&em_tree->lock); 5406 remove_extent_mapping(em_tree, em); 5407 write_unlock(&em_tree->lock); 5408 5409 /* One for our allocation */ 5410 free_extent_map(em); 5411 /* One for the tree reference */ 5412 free_extent_map(em); 5413 5414 return block_group; 5415 } 5416 5417 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 5418 u64 type) 5419 { 5420 struct btrfs_fs_info *info = trans->fs_info; 5421 struct btrfs_fs_devices *fs_devices = info->fs_devices; 5422 struct btrfs_device_info *devices_info = NULL; 5423 struct alloc_chunk_ctl ctl; 5424 struct btrfs_block_group *block_group; 5425 int ret; 5426 5427 lockdep_assert_held(&info->chunk_mutex); 5428 5429 if (!alloc_profile_is_valid(type, 0)) { 5430 ASSERT(0); 5431 return ERR_PTR(-EINVAL); 5432 } 5433 5434 if (list_empty(&fs_devices->alloc_list)) { 5435 if (btrfs_test_opt(info, ENOSPC_DEBUG)) 5436 btrfs_debug(info, "%s: no writable device", __func__); 5437 return ERR_PTR(-ENOSPC); 5438 } 5439 5440 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { 5441 btrfs_err(info, "invalid chunk type 0x%llx requested", type); 5442 ASSERT(0); 5443 return ERR_PTR(-EINVAL); 5444 } 5445 5446 ctl.start = find_next_chunk(info); 5447 ctl.type = type; 5448 init_alloc_chunk_ctl(fs_devices, &ctl); 5449 5450 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), 5451 GFP_NOFS); 5452 if (!devices_info) 5453 return ERR_PTR(-ENOMEM); 5454 5455 ret = gather_device_info(fs_devices, &ctl, devices_info); 5456 if (ret < 0) { 5457 block_group = ERR_PTR(ret); 5458 goto out; 5459 } 5460 5461 ret = decide_stripe_size(fs_devices, &ctl, devices_info); 5462 if (ret < 0) { 5463 block_group = ERR_PTR(ret); 5464 goto out; 5465 } 5466 5467 block_group = create_chunk(trans, &ctl, devices_info); 5468 5469 out: 5470 kfree(devices_info); 5471 return block_group; 5472 } 5473 5474 /* 5475 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the 5476 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system 5477 * chunks. 5478 * 5479 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation 5480 * phases. 5481 */ 5482 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 5483 struct btrfs_block_group *bg) 5484 { 5485 struct btrfs_fs_info *fs_info = trans->fs_info; 5486 struct btrfs_root *extent_root = fs_info->extent_root; 5487 struct btrfs_root *chunk_root = fs_info->chunk_root; 5488 struct btrfs_key key; 5489 struct btrfs_chunk *chunk; 5490 struct btrfs_stripe *stripe; 5491 struct extent_map *em; 5492 struct map_lookup *map; 5493 size_t item_size; 5494 int i; 5495 int ret; 5496 5497 /* 5498 * We take the chunk_mutex for 2 reasons: 5499 * 5500 * 1) Updates and insertions in the chunk btree must be done while holding 5501 * the chunk_mutex, as well as updating the system chunk array in the 5502 * superblock. See the comment on top of btrfs_chunk_alloc() for the 5503 * details; 5504 * 5505 * 2) To prevent races with the final phase of a device replace operation 5506 * that replaces the device object associated with the map's stripes, 5507 * because the device object's id can change at any time during that 5508 * final phase of the device replace operation 5509 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the 5510 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, 5511 * which would cause a failure when updating the device item, which does 5512 * not exists, or persisting a stripe of the chunk item with such ID. 5513 * Here we can't use the device_list_mutex because our caller already 5514 * has locked the chunk_mutex, and the final phase of device replace 5515 * acquires both mutexes - first the device_list_mutex and then the 5516 * chunk_mutex. Using any of those two mutexes protects us from a 5517 * concurrent device replace. 5518 */ 5519 lockdep_assert_held(&fs_info->chunk_mutex); 5520 5521 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); 5522 if (IS_ERR(em)) { 5523 ret = PTR_ERR(em); 5524 btrfs_abort_transaction(trans, ret); 5525 return ret; 5526 } 5527 5528 map = em->map_lookup; 5529 item_size = btrfs_chunk_item_size(map->num_stripes); 5530 5531 chunk = kzalloc(item_size, GFP_NOFS); 5532 if (!chunk) { 5533 ret = -ENOMEM; 5534 btrfs_abort_transaction(trans, ret); 5535 goto out; 5536 } 5537 5538 for (i = 0; i < map->num_stripes; i++) { 5539 struct btrfs_device *device = map->stripes[i].dev; 5540 5541 ret = btrfs_update_device(trans, device); 5542 if (ret) 5543 goto out; 5544 } 5545 5546 stripe = &chunk->stripe; 5547 for (i = 0; i < map->num_stripes; i++) { 5548 struct btrfs_device *device = map->stripes[i].dev; 5549 const u64 dev_offset = map->stripes[i].physical; 5550 5551 btrfs_set_stack_stripe_devid(stripe, device->devid); 5552 btrfs_set_stack_stripe_offset(stripe, dev_offset); 5553 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); 5554 stripe++; 5555 } 5556 5557 btrfs_set_stack_chunk_length(chunk, bg->length); 5558 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); 5559 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); 5560 btrfs_set_stack_chunk_type(chunk, map->type); 5561 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); 5562 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); 5563 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); 5564 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); 5565 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); 5566 5567 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; 5568 key.type = BTRFS_CHUNK_ITEM_KEY; 5569 key.offset = bg->start; 5570 5571 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); 5572 if (ret) 5573 goto out; 5574 5575 bg->chunk_item_inserted = 1; 5576 5577 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { 5578 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); 5579 if (ret) 5580 goto out; 5581 } 5582 5583 out: 5584 kfree(chunk); 5585 free_extent_map(em); 5586 return ret; 5587 } 5588 5589 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) 5590 { 5591 struct btrfs_fs_info *fs_info = trans->fs_info; 5592 u64 alloc_profile; 5593 struct btrfs_block_group *meta_bg; 5594 struct btrfs_block_group *sys_bg; 5595 5596 /* 5597 * When adding a new device for sprouting, the seed device is read-only 5598 * so we must first allocate a metadata and a system chunk. But before 5599 * adding the block group items to the extent, device and chunk btrees, 5600 * we must first: 5601 * 5602 * 1) Create both chunks without doing any changes to the btrees, as 5603 * otherwise we would get -ENOSPC since the block groups from the 5604 * seed device are read-only; 5605 * 5606 * 2) Add the device item for the new sprout device - finishing the setup 5607 * of a new block group requires updating the device item in the chunk 5608 * btree, so it must exist when we attempt to do it. The previous step 5609 * ensures this does not fail with -ENOSPC. 5610 * 5611 * After that we can add the block group items to their btrees: 5612 * update existing device item in the chunk btree, add a new block group 5613 * item to the extent btree, add a new chunk item to the chunk btree and 5614 * finally add the new device extent items to the devices btree. 5615 */ 5616 5617 alloc_profile = btrfs_metadata_alloc_profile(fs_info); 5618 meta_bg = btrfs_create_chunk(trans, alloc_profile); 5619 if (IS_ERR(meta_bg)) 5620 return PTR_ERR(meta_bg); 5621 5622 alloc_profile = btrfs_system_alloc_profile(fs_info); 5623 sys_bg = btrfs_create_chunk(trans, alloc_profile); 5624 if (IS_ERR(sys_bg)) 5625 return PTR_ERR(sys_bg); 5626 5627 return 0; 5628 } 5629 5630 static inline int btrfs_chunk_max_errors(struct map_lookup *map) 5631 { 5632 const int index = btrfs_bg_flags_to_raid_index(map->type); 5633 5634 return btrfs_raid_array[index].tolerated_failures; 5635 } 5636 5637 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) 5638 { 5639 struct extent_map *em; 5640 struct map_lookup *map; 5641 int miss_ndevs = 0; 5642 int i; 5643 bool ret = true; 5644 5645 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); 5646 if (IS_ERR(em)) 5647 return false; 5648 5649 map = em->map_lookup; 5650 for (i = 0; i < map->num_stripes; i++) { 5651 if (test_bit(BTRFS_DEV_STATE_MISSING, 5652 &map->stripes[i].dev->dev_state)) { 5653 miss_ndevs++; 5654 continue; 5655 } 5656 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, 5657 &map->stripes[i].dev->dev_state)) { 5658 ret = false; 5659 goto end; 5660 } 5661 } 5662 5663 /* 5664 * If the number of missing devices is larger than max errors, we can 5665 * not write the data into that chunk successfully. 5666 */ 5667 if (miss_ndevs > btrfs_chunk_max_errors(map)) 5668 ret = false; 5669 end: 5670 free_extent_map(em); 5671 return ret; 5672 } 5673 5674 void btrfs_mapping_tree_free(struct extent_map_tree *tree) 5675 { 5676 struct extent_map *em; 5677 5678 while (1) { 5679 write_lock(&tree->lock); 5680 em = lookup_extent_mapping(tree, 0, (u64)-1); 5681 if (em) 5682 remove_extent_mapping(tree, em); 5683 write_unlock(&tree->lock); 5684 if (!em) 5685 break; 5686 /* once for us */ 5687 free_extent_map(em); 5688 /* once for the tree */ 5689 free_extent_map(em); 5690 } 5691 } 5692 5693 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5694 { 5695 struct extent_map *em; 5696 struct map_lookup *map; 5697 int ret; 5698 5699 em = btrfs_get_chunk_map(fs_info, logical, len); 5700 if (IS_ERR(em)) 5701 /* 5702 * We could return errors for these cases, but that could get 5703 * ugly and we'd probably do the same thing which is just not do 5704 * anything else and exit, so return 1 so the callers don't try 5705 * to use other copies. 5706 */ 5707 return 1; 5708 5709 map = em->map_lookup; 5710 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) 5711 ret = map->num_stripes; 5712 else if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5713 ret = map->sub_stripes; 5714 else if (map->type & BTRFS_BLOCK_GROUP_RAID5) 5715 ret = 2; 5716 else if (map->type & BTRFS_BLOCK_GROUP_RAID6) 5717 /* 5718 * There could be two corrupted data stripes, we need 5719 * to loop retry in order to rebuild the correct data. 5720 * 5721 * Fail a stripe at a time on every retry except the 5722 * stripe under reconstruction. 5723 */ 5724 ret = map->num_stripes; 5725 else 5726 ret = 1; 5727 free_extent_map(em); 5728 5729 down_read(&fs_info->dev_replace.rwsem); 5730 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && 5731 fs_info->dev_replace.tgtdev) 5732 ret++; 5733 up_read(&fs_info->dev_replace.rwsem); 5734 5735 return ret; 5736 } 5737 5738 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 5739 u64 logical) 5740 { 5741 struct extent_map *em; 5742 struct map_lookup *map; 5743 unsigned long len = fs_info->sectorsize; 5744 5745 em = btrfs_get_chunk_map(fs_info, logical, len); 5746 5747 if (!WARN_ON(IS_ERR(em))) { 5748 map = em->map_lookup; 5749 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5750 len = map->stripe_len * nr_data_stripes(map); 5751 free_extent_map(em); 5752 } 5753 return len; 5754 } 5755 5756 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) 5757 { 5758 struct extent_map *em; 5759 struct map_lookup *map; 5760 int ret = 0; 5761 5762 em = btrfs_get_chunk_map(fs_info, logical, len); 5763 5764 if(!WARN_ON(IS_ERR(em))) { 5765 map = em->map_lookup; 5766 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) 5767 ret = 1; 5768 free_extent_map(em); 5769 } 5770 return ret; 5771 } 5772 5773 static int find_live_mirror(struct btrfs_fs_info *fs_info, 5774 struct map_lookup *map, int first, 5775 int dev_replace_is_ongoing) 5776 { 5777 int i; 5778 int num_stripes; 5779 int preferred_mirror; 5780 int tolerance; 5781 struct btrfs_device *srcdev; 5782 5783 ASSERT((map->type & 5784 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); 5785 5786 if (map->type & BTRFS_BLOCK_GROUP_RAID10) 5787 num_stripes = map->sub_stripes; 5788 else 5789 num_stripes = map->num_stripes; 5790 5791 switch (fs_info->fs_devices->read_policy) { 5792 default: 5793 /* Shouldn't happen, just warn and use pid instead of failing */ 5794 btrfs_warn_rl(fs_info, 5795 "unknown read_policy type %u, reset to pid", 5796 fs_info->fs_devices->read_policy); 5797 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; 5798 fallthrough; 5799 case BTRFS_READ_POLICY_PID: 5800 preferred_mirror = first + (current->pid % num_stripes); 5801 break; 5802 } 5803 5804 if (dev_replace_is_ongoing && 5805 fs_info->dev_replace.cont_reading_from_srcdev_mode == 5806 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) 5807 srcdev = fs_info->dev_replace.srcdev; 5808 else 5809 srcdev = NULL; 5810 5811 /* 5812 * try to avoid the drive that is the source drive for a 5813 * dev-replace procedure, only choose it if no other non-missing 5814 * mirror is available 5815 */ 5816 for (tolerance = 0; tolerance < 2; tolerance++) { 5817 if (map->stripes[preferred_mirror].dev->bdev && 5818 (tolerance || map->stripes[preferred_mirror].dev != srcdev)) 5819 return preferred_mirror; 5820 for (i = first; i < first + num_stripes; i++) { 5821 if (map->stripes[i].dev->bdev && 5822 (tolerance || map->stripes[i].dev != srcdev)) 5823 return i; 5824 } 5825 } 5826 5827 /* we couldn't find one that doesn't fail. Just return something 5828 * and the io error handling code will clean up eventually 5829 */ 5830 return preferred_mirror; 5831 } 5832 5833 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ 5834 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) 5835 { 5836 int i; 5837 int again = 1; 5838 5839 while (again) { 5840 again = 0; 5841 for (i = 0; i < num_stripes - 1; i++) { 5842 /* Swap if parity is on a smaller index */ 5843 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { 5844 swap(bioc->stripes[i], bioc->stripes[i + 1]); 5845 swap(bioc->raid_map[i], bioc->raid_map[i + 1]); 5846 again = 1; 5847 } 5848 } 5849 } 5850 } 5851 5852 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, 5853 int total_stripes, 5854 int real_stripes) 5855 { 5856 struct btrfs_io_context *bioc = kzalloc( 5857 /* The size of btrfs_io_context */ 5858 sizeof(struct btrfs_io_context) + 5859 /* Plus the variable array for the stripes */ 5860 sizeof(struct btrfs_io_stripe) * (total_stripes) + 5861 /* Plus the variable array for the tgt dev */ 5862 sizeof(int) * (real_stripes) + 5863 /* 5864 * Plus the raid_map, which includes both the tgt dev 5865 * and the stripes. 5866 */ 5867 sizeof(u64) * (total_stripes), 5868 GFP_NOFS|__GFP_NOFAIL); 5869 5870 atomic_set(&bioc->error, 0); 5871 refcount_set(&bioc->refs, 1); 5872 5873 bioc->fs_info = fs_info; 5874 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); 5875 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); 5876 5877 return bioc; 5878 } 5879 5880 void btrfs_get_bioc(struct btrfs_io_context *bioc) 5881 { 5882 WARN_ON(!refcount_read(&bioc->refs)); 5883 refcount_inc(&bioc->refs); 5884 } 5885 5886 void btrfs_put_bioc(struct btrfs_io_context *bioc) 5887 { 5888 if (!bioc) 5889 return; 5890 if (refcount_dec_and_test(&bioc->refs)) 5891 kfree(bioc); 5892 } 5893 5894 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ 5895 /* 5896 * Please note that, discard won't be sent to target device of device 5897 * replace. 5898 */ 5899 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, 5900 u64 logical, u64 *length_ret, 5901 struct btrfs_io_context **bioc_ret) 5902 { 5903 struct extent_map *em; 5904 struct map_lookup *map; 5905 struct btrfs_io_context *bioc; 5906 u64 length = *length_ret; 5907 u64 offset; 5908 u64 stripe_nr; 5909 u64 stripe_nr_end; 5910 u64 stripe_end_offset; 5911 u64 stripe_cnt; 5912 u64 stripe_len; 5913 u64 stripe_offset; 5914 u64 num_stripes; 5915 u32 stripe_index; 5916 u32 factor = 0; 5917 u32 sub_stripes = 0; 5918 u64 stripes_per_dev = 0; 5919 u32 remaining_stripes = 0; 5920 u32 last_stripe = 0; 5921 int ret = 0; 5922 int i; 5923 5924 /* Discard always returns a bioc. */ 5925 ASSERT(bioc_ret); 5926 5927 em = btrfs_get_chunk_map(fs_info, logical, length); 5928 if (IS_ERR(em)) 5929 return PTR_ERR(em); 5930 5931 map = em->map_lookup; 5932 /* we don't discard raid56 yet */ 5933 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 5934 ret = -EOPNOTSUPP; 5935 goto out; 5936 } 5937 5938 offset = logical - em->start; 5939 length = min_t(u64, em->start + em->len - logical, length); 5940 *length_ret = length; 5941 5942 stripe_len = map->stripe_len; 5943 /* 5944 * stripe_nr counts the total number of stripes we have to stride 5945 * to get to this block 5946 */ 5947 stripe_nr = div64_u64(offset, stripe_len); 5948 5949 /* stripe_offset is the offset of this block in its stripe */ 5950 stripe_offset = offset - stripe_nr * stripe_len; 5951 5952 stripe_nr_end = round_up(offset + length, map->stripe_len); 5953 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); 5954 stripe_cnt = stripe_nr_end - stripe_nr; 5955 stripe_end_offset = stripe_nr_end * map->stripe_len - 5956 (offset + length); 5957 /* 5958 * after this, stripe_nr is the number of stripes on this 5959 * device we have to walk to find the data, and stripe_index is 5960 * the number of our device in the stripe array 5961 */ 5962 num_stripes = 1; 5963 stripe_index = 0; 5964 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 5965 BTRFS_BLOCK_GROUP_RAID10)) { 5966 if (map->type & BTRFS_BLOCK_GROUP_RAID0) 5967 sub_stripes = 1; 5968 else 5969 sub_stripes = map->sub_stripes; 5970 5971 factor = map->num_stripes / sub_stripes; 5972 num_stripes = min_t(u64, map->num_stripes, 5973 sub_stripes * stripe_cnt); 5974 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 5975 stripe_index *= sub_stripes; 5976 stripes_per_dev = div_u64_rem(stripe_cnt, factor, 5977 &remaining_stripes); 5978 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); 5979 last_stripe *= sub_stripes; 5980 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | 5981 BTRFS_BLOCK_GROUP_DUP)) { 5982 num_stripes = map->num_stripes; 5983 } else { 5984 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 5985 &stripe_index); 5986 } 5987 5988 bioc = alloc_btrfs_io_context(fs_info, num_stripes, 0); 5989 if (!bioc) { 5990 ret = -ENOMEM; 5991 goto out; 5992 } 5993 5994 for (i = 0; i < num_stripes; i++) { 5995 bioc->stripes[i].physical = 5996 map->stripes[stripe_index].physical + 5997 stripe_offset + stripe_nr * map->stripe_len; 5998 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 5999 6000 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | 6001 BTRFS_BLOCK_GROUP_RAID10)) { 6002 bioc->stripes[i].length = stripes_per_dev * 6003 map->stripe_len; 6004 6005 if (i / sub_stripes < remaining_stripes) 6006 bioc->stripes[i].length += map->stripe_len; 6007 6008 /* 6009 * Special for the first stripe and 6010 * the last stripe: 6011 * 6012 * |-------|...|-------| 6013 * |----------| 6014 * off end_off 6015 */ 6016 if (i < sub_stripes) 6017 bioc->stripes[i].length -= stripe_offset; 6018 6019 if (stripe_index >= last_stripe && 6020 stripe_index <= (last_stripe + 6021 sub_stripes - 1)) 6022 bioc->stripes[i].length -= stripe_end_offset; 6023 6024 if (i == sub_stripes - 1) 6025 stripe_offset = 0; 6026 } else { 6027 bioc->stripes[i].length = length; 6028 } 6029 6030 stripe_index++; 6031 if (stripe_index == map->num_stripes) { 6032 stripe_index = 0; 6033 stripe_nr++; 6034 } 6035 } 6036 6037 *bioc_ret = bioc; 6038 bioc->map_type = map->type; 6039 bioc->num_stripes = num_stripes; 6040 out: 6041 free_extent_map(em); 6042 return ret; 6043 } 6044 6045 /* 6046 * In dev-replace case, for repair case (that's the only case where the mirror 6047 * is selected explicitly when calling btrfs_map_block), blocks left of the 6048 * left cursor can also be read from the target drive. 6049 * 6050 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the 6051 * array of stripes. 6052 * For READ, it also needs to be supported using the same mirror number. 6053 * 6054 * If the requested block is not left of the left cursor, EIO is returned. This 6055 * can happen because btrfs_num_copies() returns one more in the dev-replace 6056 * case. 6057 */ 6058 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, 6059 u64 logical, u64 length, 6060 u64 srcdev_devid, int *mirror_num, 6061 u64 *physical) 6062 { 6063 struct btrfs_io_context *bioc = NULL; 6064 int num_stripes; 6065 int index_srcdev = 0; 6066 int found = 0; 6067 u64 physical_of_found = 0; 6068 int i; 6069 int ret = 0; 6070 6071 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, 6072 logical, &length, &bioc, 0, 0); 6073 if (ret) { 6074 ASSERT(bioc == NULL); 6075 return ret; 6076 } 6077 6078 num_stripes = bioc->num_stripes; 6079 if (*mirror_num > num_stripes) { 6080 /* 6081 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, 6082 * that means that the requested area is not left of the left 6083 * cursor 6084 */ 6085 btrfs_put_bioc(bioc); 6086 return -EIO; 6087 } 6088 6089 /* 6090 * process the rest of the function using the mirror_num of the source 6091 * drive. Therefore look it up first. At the end, patch the device 6092 * pointer to the one of the target drive. 6093 */ 6094 for (i = 0; i < num_stripes; i++) { 6095 if (bioc->stripes[i].dev->devid != srcdev_devid) 6096 continue; 6097 6098 /* 6099 * In case of DUP, in order to keep it simple, only add the 6100 * mirror with the lowest physical address 6101 */ 6102 if (found && 6103 physical_of_found <= bioc->stripes[i].physical) 6104 continue; 6105 6106 index_srcdev = i; 6107 found = 1; 6108 physical_of_found = bioc->stripes[i].physical; 6109 } 6110 6111 btrfs_put_bioc(bioc); 6112 6113 ASSERT(found); 6114 if (!found) 6115 return -EIO; 6116 6117 *mirror_num = index_srcdev + 1; 6118 *physical = physical_of_found; 6119 return ret; 6120 } 6121 6122 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) 6123 { 6124 struct btrfs_block_group *cache; 6125 bool ret; 6126 6127 /* Non zoned filesystem does not use "to_copy" flag */ 6128 if (!btrfs_is_zoned(fs_info)) 6129 return false; 6130 6131 cache = btrfs_lookup_block_group(fs_info, logical); 6132 6133 spin_lock(&cache->lock); 6134 ret = cache->to_copy; 6135 spin_unlock(&cache->lock); 6136 6137 btrfs_put_block_group(cache); 6138 return ret; 6139 } 6140 6141 static void handle_ops_on_dev_replace(enum btrfs_map_op op, 6142 struct btrfs_io_context **bioc_ret, 6143 struct btrfs_dev_replace *dev_replace, 6144 u64 logical, 6145 int *num_stripes_ret, int *max_errors_ret) 6146 { 6147 struct btrfs_io_context *bioc = *bioc_ret; 6148 u64 srcdev_devid = dev_replace->srcdev->devid; 6149 int tgtdev_indexes = 0; 6150 int num_stripes = *num_stripes_ret; 6151 int max_errors = *max_errors_ret; 6152 int i; 6153 6154 if (op == BTRFS_MAP_WRITE) { 6155 int index_where_to_add; 6156 6157 /* 6158 * A block group which have "to_copy" set will eventually 6159 * copied by dev-replace process. We can avoid cloning IO here. 6160 */ 6161 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) 6162 return; 6163 6164 /* 6165 * duplicate the write operations while the dev replace 6166 * procedure is running. Since the copying of the old disk to 6167 * the new disk takes place at run time while the filesystem is 6168 * mounted writable, the regular write operations to the old 6169 * disk have to be duplicated to go to the new disk as well. 6170 * 6171 * Note that device->missing is handled by the caller, and that 6172 * the write to the old disk is already set up in the stripes 6173 * array. 6174 */ 6175 index_where_to_add = num_stripes; 6176 for (i = 0; i < num_stripes; i++) { 6177 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6178 /* write to new disk, too */ 6179 struct btrfs_io_stripe *new = 6180 bioc->stripes + index_where_to_add; 6181 struct btrfs_io_stripe *old = 6182 bioc->stripes + i; 6183 6184 new->physical = old->physical; 6185 new->length = old->length; 6186 new->dev = dev_replace->tgtdev; 6187 bioc->tgtdev_map[i] = index_where_to_add; 6188 index_where_to_add++; 6189 max_errors++; 6190 tgtdev_indexes++; 6191 } 6192 } 6193 num_stripes = index_where_to_add; 6194 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { 6195 int index_srcdev = 0; 6196 int found = 0; 6197 u64 physical_of_found = 0; 6198 6199 /* 6200 * During the dev-replace procedure, the target drive can also 6201 * be used to read data in case it is needed to repair a corrupt 6202 * block elsewhere. This is possible if the requested area is 6203 * left of the left cursor. In this area, the target drive is a 6204 * full copy of the source drive. 6205 */ 6206 for (i = 0; i < num_stripes; i++) { 6207 if (bioc->stripes[i].dev->devid == srcdev_devid) { 6208 /* 6209 * In case of DUP, in order to keep it simple, 6210 * only add the mirror with the lowest physical 6211 * address 6212 */ 6213 if (found && 6214 physical_of_found <= bioc->stripes[i].physical) 6215 continue; 6216 index_srcdev = i; 6217 found = 1; 6218 physical_of_found = bioc->stripes[i].physical; 6219 } 6220 } 6221 if (found) { 6222 struct btrfs_io_stripe *tgtdev_stripe = 6223 bioc->stripes + num_stripes; 6224 6225 tgtdev_stripe->physical = physical_of_found; 6226 tgtdev_stripe->length = 6227 bioc->stripes[index_srcdev].length; 6228 tgtdev_stripe->dev = dev_replace->tgtdev; 6229 bioc->tgtdev_map[index_srcdev] = num_stripes; 6230 6231 tgtdev_indexes++; 6232 num_stripes++; 6233 } 6234 } 6235 6236 *num_stripes_ret = num_stripes; 6237 *max_errors_ret = max_errors; 6238 bioc->num_tgtdevs = tgtdev_indexes; 6239 *bioc_ret = bioc; 6240 } 6241 6242 static bool need_full_stripe(enum btrfs_map_op op) 6243 { 6244 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); 6245 } 6246 6247 /* 6248 * Calculate the geometry of a particular (address, len) tuple. This 6249 * information is used to calculate how big a particular bio can get before it 6250 * straddles a stripe. 6251 * 6252 * @fs_info: the filesystem 6253 * @em: mapping containing the logical extent 6254 * @op: type of operation - write or read 6255 * @logical: address that we want to figure out the geometry of 6256 * @io_geom: pointer used to return values 6257 * 6258 * Returns < 0 in case a chunk for the given logical address cannot be found, 6259 * usually shouldn't happen unless @logical is corrupted, 0 otherwise. 6260 */ 6261 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, 6262 enum btrfs_map_op op, u64 logical, 6263 struct btrfs_io_geometry *io_geom) 6264 { 6265 struct map_lookup *map; 6266 u64 len; 6267 u64 offset; 6268 u64 stripe_offset; 6269 u64 stripe_nr; 6270 u64 stripe_len; 6271 u64 raid56_full_stripe_start = (u64)-1; 6272 int data_stripes; 6273 6274 ASSERT(op != BTRFS_MAP_DISCARD); 6275 6276 map = em->map_lookup; 6277 /* Offset of this logical address in the chunk */ 6278 offset = logical - em->start; 6279 /* Len of a stripe in a chunk */ 6280 stripe_len = map->stripe_len; 6281 /* Stripe where this block falls in */ 6282 stripe_nr = div64_u64(offset, stripe_len); 6283 /* Offset of stripe in the chunk */ 6284 stripe_offset = stripe_nr * stripe_len; 6285 if (offset < stripe_offset) { 6286 btrfs_crit(fs_info, 6287 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", 6288 stripe_offset, offset, em->start, logical, stripe_len); 6289 return -EINVAL; 6290 } 6291 6292 /* stripe_offset is the offset of this block in its stripe */ 6293 stripe_offset = offset - stripe_offset; 6294 data_stripes = nr_data_stripes(map); 6295 6296 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { 6297 u64 max_len = stripe_len - stripe_offset; 6298 6299 /* 6300 * In case of raid56, we need to know the stripe aligned start 6301 */ 6302 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6303 unsigned long full_stripe_len = stripe_len * data_stripes; 6304 raid56_full_stripe_start = offset; 6305 6306 /* 6307 * Allow a write of a full stripe, but make sure we 6308 * don't allow straddling of stripes 6309 */ 6310 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, 6311 full_stripe_len); 6312 raid56_full_stripe_start *= full_stripe_len; 6313 6314 /* 6315 * For writes to RAID[56], allow a full stripeset across 6316 * all disks. For other RAID types and for RAID[56] 6317 * reads, just allow a single stripe (on a single disk). 6318 */ 6319 if (op == BTRFS_MAP_WRITE) { 6320 max_len = stripe_len * data_stripes - 6321 (offset - raid56_full_stripe_start); 6322 } 6323 } 6324 len = min_t(u64, em->len - offset, max_len); 6325 } else { 6326 len = em->len - offset; 6327 } 6328 6329 io_geom->len = len; 6330 io_geom->offset = offset; 6331 io_geom->stripe_len = stripe_len; 6332 io_geom->stripe_nr = stripe_nr; 6333 io_geom->stripe_offset = stripe_offset; 6334 io_geom->raid56_stripe_offset = raid56_full_stripe_start; 6335 6336 return 0; 6337 } 6338 6339 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, 6340 enum btrfs_map_op op, 6341 u64 logical, u64 *length, 6342 struct btrfs_io_context **bioc_ret, 6343 int mirror_num, int need_raid_map) 6344 { 6345 struct extent_map *em; 6346 struct map_lookup *map; 6347 u64 stripe_offset; 6348 u64 stripe_nr; 6349 u64 stripe_len; 6350 u32 stripe_index; 6351 int data_stripes; 6352 int i; 6353 int ret = 0; 6354 int num_stripes; 6355 int max_errors = 0; 6356 int tgtdev_indexes = 0; 6357 struct btrfs_io_context *bioc = NULL; 6358 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 6359 int dev_replace_is_ongoing = 0; 6360 int num_alloc_stripes; 6361 int patch_the_first_stripe_for_dev_replace = 0; 6362 u64 physical_to_patch_in_first_stripe = 0; 6363 u64 raid56_full_stripe_start = (u64)-1; 6364 struct btrfs_io_geometry geom; 6365 6366 ASSERT(bioc_ret); 6367 ASSERT(op != BTRFS_MAP_DISCARD); 6368 6369 em = btrfs_get_chunk_map(fs_info, logical, *length); 6370 ASSERT(!IS_ERR(em)); 6371 6372 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); 6373 if (ret < 0) 6374 return ret; 6375 6376 map = em->map_lookup; 6377 6378 *length = geom.len; 6379 stripe_len = geom.stripe_len; 6380 stripe_nr = geom.stripe_nr; 6381 stripe_offset = geom.stripe_offset; 6382 raid56_full_stripe_start = geom.raid56_stripe_offset; 6383 data_stripes = nr_data_stripes(map); 6384 6385 down_read(&dev_replace->rwsem); 6386 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 6387 /* 6388 * Hold the semaphore for read during the whole operation, write is 6389 * requested at commit time but must wait. 6390 */ 6391 if (!dev_replace_is_ongoing) 6392 up_read(&dev_replace->rwsem); 6393 6394 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && 6395 !need_full_stripe(op) && dev_replace->tgtdev != NULL) { 6396 ret = get_extra_mirror_from_replace(fs_info, logical, *length, 6397 dev_replace->srcdev->devid, 6398 &mirror_num, 6399 &physical_to_patch_in_first_stripe); 6400 if (ret) 6401 goto out; 6402 else 6403 patch_the_first_stripe_for_dev_replace = 1; 6404 } else if (mirror_num > map->num_stripes) { 6405 mirror_num = 0; 6406 } 6407 6408 num_stripes = 1; 6409 stripe_index = 0; 6410 if (map->type & BTRFS_BLOCK_GROUP_RAID0) { 6411 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6412 &stripe_index); 6413 if (!need_full_stripe(op)) 6414 mirror_num = 1; 6415 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { 6416 if (need_full_stripe(op)) 6417 num_stripes = map->num_stripes; 6418 else if (mirror_num) 6419 stripe_index = mirror_num - 1; 6420 else { 6421 stripe_index = find_live_mirror(fs_info, map, 0, 6422 dev_replace_is_ongoing); 6423 mirror_num = stripe_index + 1; 6424 } 6425 6426 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { 6427 if (need_full_stripe(op)) { 6428 num_stripes = map->num_stripes; 6429 } else if (mirror_num) { 6430 stripe_index = mirror_num - 1; 6431 } else { 6432 mirror_num = 1; 6433 } 6434 6435 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { 6436 u32 factor = map->num_stripes / map->sub_stripes; 6437 6438 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 6439 stripe_index *= map->sub_stripes; 6440 6441 if (need_full_stripe(op)) 6442 num_stripes = map->sub_stripes; 6443 else if (mirror_num) 6444 stripe_index += mirror_num - 1; 6445 else { 6446 int old_stripe_index = stripe_index; 6447 stripe_index = find_live_mirror(fs_info, map, 6448 stripe_index, 6449 dev_replace_is_ongoing); 6450 mirror_num = stripe_index - old_stripe_index + 1; 6451 } 6452 6453 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 6454 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { 6455 /* push stripe_nr back to the start of the full stripe */ 6456 stripe_nr = div64_u64(raid56_full_stripe_start, 6457 stripe_len * data_stripes); 6458 6459 /* RAID[56] write or recovery. Return all stripes */ 6460 num_stripes = map->num_stripes; 6461 max_errors = nr_parity_stripes(map); 6462 6463 *length = map->stripe_len; 6464 stripe_index = 0; 6465 stripe_offset = 0; 6466 } else { 6467 /* 6468 * Mirror #0 or #1 means the original data block. 6469 * Mirror #2 is RAID5 parity block. 6470 * Mirror #3 is RAID6 Q block. 6471 */ 6472 stripe_nr = div_u64_rem(stripe_nr, 6473 data_stripes, &stripe_index); 6474 if (mirror_num > 1) 6475 stripe_index = data_stripes + mirror_num - 2; 6476 6477 /* We distribute the parity blocks across stripes */ 6478 div_u64_rem(stripe_nr + stripe_index, map->num_stripes, 6479 &stripe_index); 6480 if (!need_full_stripe(op) && mirror_num <= 1) 6481 mirror_num = 1; 6482 } 6483 } else { 6484 /* 6485 * after this, stripe_nr is the number of stripes on this 6486 * device we have to walk to find the data, and stripe_index is 6487 * the number of our device in the stripe array 6488 */ 6489 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, 6490 &stripe_index); 6491 mirror_num = stripe_index + 1; 6492 } 6493 if (stripe_index >= map->num_stripes) { 6494 btrfs_crit(fs_info, 6495 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", 6496 stripe_index, map->num_stripes); 6497 ret = -EINVAL; 6498 goto out; 6499 } 6500 6501 num_alloc_stripes = num_stripes; 6502 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { 6503 if (op == BTRFS_MAP_WRITE) 6504 num_alloc_stripes <<= 1; 6505 if (op == BTRFS_MAP_GET_READ_MIRRORS) 6506 num_alloc_stripes++; 6507 tgtdev_indexes = num_stripes; 6508 } 6509 6510 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); 6511 if (!bioc) { 6512 ret = -ENOMEM; 6513 goto out; 6514 } 6515 6516 for (i = 0; i < num_stripes; i++) { 6517 bioc->stripes[i].physical = map->stripes[stripe_index].physical + 6518 stripe_offset + stripe_nr * map->stripe_len; 6519 bioc->stripes[i].dev = map->stripes[stripe_index].dev; 6520 stripe_index++; 6521 } 6522 6523 /* Build raid_map */ 6524 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && 6525 (need_full_stripe(op) || mirror_num > 1)) { 6526 u64 tmp; 6527 unsigned rot; 6528 6529 /* Work out the disk rotation on this stripe-set */ 6530 div_u64_rem(stripe_nr, num_stripes, &rot); 6531 6532 /* Fill in the logical address of each stripe */ 6533 tmp = stripe_nr * data_stripes; 6534 for (i = 0; i < data_stripes; i++) 6535 bioc->raid_map[(i + rot) % num_stripes] = 6536 em->start + (tmp + i) * map->stripe_len; 6537 6538 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; 6539 if (map->type & BTRFS_BLOCK_GROUP_RAID6) 6540 bioc->raid_map[(i + rot + 1) % num_stripes] = 6541 RAID6_Q_STRIPE; 6542 6543 sort_parity_stripes(bioc, num_stripes); 6544 } 6545 6546 if (need_full_stripe(op)) 6547 max_errors = btrfs_chunk_max_errors(map); 6548 6549 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && 6550 need_full_stripe(op)) { 6551 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, 6552 &num_stripes, &max_errors); 6553 } 6554 6555 *bioc_ret = bioc; 6556 bioc->map_type = map->type; 6557 bioc->num_stripes = num_stripes; 6558 bioc->max_errors = max_errors; 6559 bioc->mirror_num = mirror_num; 6560 6561 /* 6562 * this is the case that REQ_READ && dev_replace_is_ongoing && 6563 * mirror_num == num_stripes + 1 && dev_replace target drive is 6564 * available as a mirror 6565 */ 6566 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { 6567 WARN_ON(num_stripes > 1); 6568 bioc->stripes[0].dev = dev_replace->tgtdev; 6569 bioc->stripes[0].physical = physical_to_patch_in_first_stripe; 6570 bioc->mirror_num = map->num_stripes + 1; 6571 } 6572 out: 6573 if (dev_replace_is_ongoing) { 6574 lockdep_assert_held(&dev_replace->rwsem); 6575 /* Unlock and let waiting writers proceed */ 6576 up_read(&dev_replace->rwsem); 6577 } 6578 free_extent_map(em); 6579 return ret; 6580 } 6581 6582 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6583 u64 logical, u64 *length, 6584 struct btrfs_io_context **bioc_ret, int mirror_num) 6585 { 6586 if (op == BTRFS_MAP_DISCARD) 6587 return __btrfs_map_block_for_discard(fs_info, logical, 6588 length, bioc_ret); 6589 6590 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 6591 mirror_num, 0); 6592 } 6593 6594 /* For Scrub/replace */ 6595 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 6596 u64 logical, u64 *length, 6597 struct btrfs_io_context **bioc_ret) 6598 { 6599 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, 0, 1); 6600 } 6601 6602 static inline void btrfs_end_bioc(struct btrfs_io_context *bioc, struct bio *bio) 6603 { 6604 bio->bi_private = bioc->private; 6605 bio->bi_end_io = bioc->end_io; 6606 bio_endio(bio); 6607 6608 btrfs_put_bioc(bioc); 6609 } 6610 6611 static void btrfs_end_bio(struct bio *bio) 6612 { 6613 struct btrfs_io_context *bioc = bio->bi_private; 6614 int is_orig_bio = 0; 6615 6616 if (bio->bi_status) { 6617 atomic_inc(&bioc->error); 6618 if (bio->bi_status == BLK_STS_IOERR || 6619 bio->bi_status == BLK_STS_TARGET) { 6620 struct btrfs_device *dev = btrfs_bio(bio)->device; 6621 6622 ASSERT(dev->bdev); 6623 if (btrfs_op(bio) == BTRFS_MAP_WRITE) 6624 btrfs_dev_stat_inc_and_print(dev, 6625 BTRFS_DEV_STAT_WRITE_ERRS); 6626 else if (!(bio->bi_opf & REQ_RAHEAD)) 6627 btrfs_dev_stat_inc_and_print(dev, 6628 BTRFS_DEV_STAT_READ_ERRS); 6629 if (bio->bi_opf & REQ_PREFLUSH) 6630 btrfs_dev_stat_inc_and_print(dev, 6631 BTRFS_DEV_STAT_FLUSH_ERRS); 6632 } 6633 } 6634 6635 if (bio == bioc->orig_bio) 6636 is_orig_bio = 1; 6637 6638 btrfs_bio_counter_dec(bioc->fs_info); 6639 6640 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6641 if (!is_orig_bio) { 6642 bio_put(bio); 6643 bio = bioc->orig_bio; 6644 } 6645 6646 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6647 /* only send an error to the higher layers if it is 6648 * beyond the tolerance of the btrfs bio 6649 */ 6650 if (atomic_read(&bioc->error) > bioc->max_errors) { 6651 bio->bi_status = BLK_STS_IOERR; 6652 } else { 6653 /* 6654 * this bio is actually up to date, we didn't 6655 * go over the max number of errors 6656 */ 6657 bio->bi_status = BLK_STS_OK; 6658 } 6659 6660 btrfs_end_bioc(bioc, bio); 6661 } else if (!is_orig_bio) { 6662 bio_put(bio); 6663 } 6664 } 6665 6666 static void submit_stripe_bio(struct btrfs_io_context *bioc, struct bio *bio, 6667 u64 physical, struct btrfs_device *dev) 6668 { 6669 struct btrfs_fs_info *fs_info = bioc->fs_info; 6670 6671 bio->bi_private = bioc; 6672 btrfs_bio(bio)->device = dev; 6673 bio->bi_end_io = btrfs_end_bio; 6674 bio->bi_iter.bi_sector = physical >> 9; 6675 /* 6676 * For zone append writing, bi_sector must point the beginning of the 6677 * zone 6678 */ 6679 if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 6680 if (btrfs_dev_is_sequential(dev, physical)) { 6681 u64 zone_start = round_down(physical, fs_info->zone_size); 6682 6683 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; 6684 } else { 6685 bio->bi_opf &= ~REQ_OP_ZONE_APPEND; 6686 bio->bi_opf |= REQ_OP_WRITE; 6687 } 6688 } 6689 btrfs_debug_in_rcu(fs_info, 6690 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", 6691 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, 6692 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), 6693 dev->devid, bio->bi_iter.bi_size); 6694 bio_set_dev(bio, dev->bdev); 6695 6696 btrfs_bio_counter_inc_noblocked(fs_info); 6697 6698 btrfsic_submit_bio(bio); 6699 } 6700 6701 static void bioc_error(struct btrfs_io_context *bioc, struct bio *bio, u64 logical) 6702 { 6703 atomic_inc(&bioc->error); 6704 if (atomic_dec_and_test(&bioc->stripes_pending)) { 6705 /* Should be the original bio. */ 6706 WARN_ON(bio != bioc->orig_bio); 6707 6708 btrfs_bio(bio)->mirror_num = bioc->mirror_num; 6709 bio->bi_iter.bi_sector = logical >> 9; 6710 if (atomic_read(&bioc->error) > bioc->max_errors) 6711 bio->bi_status = BLK_STS_IOERR; 6712 else 6713 bio->bi_status = BLK_STS_OK; 6714 btrfs_end_bioc(bioc, bio); 6715 } 6716 } 6717 6718 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 6719 int mirror_num) 6720 { 6721 struct btrfs_device *dev; 6722 struct bio *first_bio = bio; 6723 u64 logical = bio->bi_iter.bi_sector << 9; 6724 u64 length = 0; 6725 u64 map_length; 6726 int ret; 6727 int dev_nr; 6728 int total_devs; 6729 struct btrfs_io_context *bioc = NULL; 6730 6731 length = bio->bi_iter.bi_size; 6732 map_length = length; 6733 6734 btrfs_bio_counter_inc_blocked(fs_info); 6735 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, 6736 &map_length, &bioc, mirror_num, 1); 6737 if (ret) { 6738 btrfs_bio_counter_dec(fs_info); 6739 return errno_to_blk_status(ret); 6740 } 6741 6742 total_devs = bioc->num_stripes; 6743 bioc->orig_bio = first_bio; 6744 bioc->private = first_bio->bi_private; 6745 bioc->end_io = first_bio->bi_end_io; 6746 atomic_set(&bioc->stripes_pending, bioc->num_stripes); 6747 6748 if ((bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && 6749 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { 6750 /* In this case, map_length has been set to the length of 6751 a single stripe; not the whole write */ 6752 if (btrfs_op(bio) == BTRFS_MAP_WRITE) { 6753 ret = raid56_parity_write(bio, bioc, map_length); 6754 } else { 6755 ret = raid56_parity_recover(bio, bioc, map_length, 6756 mirror_num, 1); 6757 } 6758 6759 btrfs_bio_counter_dec(fs_info); 6760 return errno_to_blk_status(ret); 6761 } 6762 6763 if (map_length < length) { 6764 btrfs_crit(fs_info, 6765 "mapping failed logical %llu bio len %llu len %llu", 6766 logical, length, map_length); 6767 BUG(); 6768 } 6769 6770 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { 6771 dev = bioc->stripes[dev_nr].dev; 6772 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, 6773 &dev->dev_state) || 6774 (btrfs_op(first_bio) == BTRFS_MAP_WRITE && 6775 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { 6776 bioc_error(bioc, first_bio, logical); 6777 continue; 6778 } 6779 6780 if (dev_nr < total_devs - 1) 6781 bio = btrfs_bio_clone(first_bio); 6782 else 6783 bio = first_bio; 6784 6785 submit_stripe_bio(bioc, bio, bioc->stripes[dev_nr].physical, dev); 6786 } 6787 btrfs_bio_counter_dec(fs_info); 6788 return BLK_STS_OK; 6789 } 6790 6791 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, 6792 const struct btrfs_fs_devices *fs_devices) 6793 { 6794 if (args->fsid == NULL) 6795 return true; 6796 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) 6797 return true; 6798 return false; 6799 } 6800 6801 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, 6802 const struct btrfs_device *device) 6803 { 6804 ASSERT((args->devid != (u64)-1) || args->missing); 6805 6806 if ((args->devid != (u64)-1) && device->devid != args->devid) 6807 return false; 6808 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) 6809 return false; 6810 if (!args->missing) 6811 return true; 6812 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && 6813 !device->bdev) 6814 return true; 6815 return false; 6816 } 6817 6818 /* 6819 * Find a device specified by @devid or @uuid in the list of @fs_devices, or 6820 * return NULL. 6821 * 6822 * If devid and uuid are both specified, the match must be exact, otherwise 6823 * only devid is used. 6824 */ 6825 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 6826 const struct btrfs_dev_lookup_args *args) 6827 { 6828 struct btrfs_device *device; 6829 struct btrfs_fs_devices *seed_devs; 6830 6831 if (dev_args_match_fs_devices(args, fs_devices)) { 6832 list_for_each_entry(device, &fs_devices->devices, dev_list) { 6833 if (dev_args_match_device(args, device)) 6834 return device; 6835 } 6836 } 6837 6838 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 6839 if (!dev_args_match_fs_devices(args, seed_devs)) 6840 continue; 6841 list_for_each_entry(device, &seed_devs->devices, dev_list) { 6842 if (dev_args_match_device(args, device)) 6843 return device; 6844 } 6845 } 6846 6847 return NULL; 6848 } 6849 6850 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, 6851 u64 devid, u8 *dev_uuid) 6852 { 6853 struct btrfs_device *device; 6854 unsigned int nofs_flag; 6855 6856 /* 6857 * We call this under the chunk_mutex, so we want to use NOFS for this 6858 * allocation, however we don't want to change btrfs_alloc_device() to 6859 * always do NOFS because we use it in a lot of other GFP_KERNEL safe 6860 * places. 6861 */ 6862 nofs_flag = memalloc_nofs_save(); 6863 device = btrfs_alloc_device(NULL, &devid, dev_uuid); 6864 memalloc_nofs_restore(nofs_flag); 6865 if (IS_ERR(device)) 6866 return device; 6867 6868 list_add(&device->dev_list, &fs_devices->devices); 6869 device->fs_devices = fs_devices; 6870 fs_devices->num_devices++; 6871 6872 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 6873 fs_devices->missing_devices++; 6874 6875 return device; 6876 } 6877 6878 /** 6879 * btrfs_alloc_device - allocate struct btrfs_device 6880 * @fs_info: used only for generating a new devid, can be NULL if 6881 * devid is provided (i.e. @devid != NULL). 6882 * @devid: a pointer to devid for this device. If NULL a new devid 6883 * is generated. 6884 * @uuid: a pointer to UUID for this device. If NULL a new UUID 6885 * is generated. 6886 * 6887 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() 6888 * on error. Returned struct is not linked onto any lists and must be 6889 * destroyed with btrfs_free_device. 6890 */ 6891 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 6892 const u64 *devid, 6893 const u8 *uuid) 6894 { 6895 struct btrfs_device *dev; 6896 u64 tmp; 6897 6898 if (WARN_ON(!devid && !fs_info)) 6899 return ERR_PTR(-EINVAL); 6900 6901 dev = kzalloc(sizeof(*dev), GFP_KERNEL); 6902 if (!dev) 6903 return ERR_PTR(-ENOMEM); 6904 6905 /* 6906 * Preallocate a bio that's always going to be used for flushing device 6907 * barriers and matches the device lifespan 6908 */ 6909 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0); 6910 if (!dev->flush_bio) { 6911 kfree(dev); 6912 return ERR_PTR(-ENOMEM); 6913 } 6914 6915 INIT_LIST_HEAD(&dev->dev_list); 6916 INIT_LIST_HEAD(&dev->dev_alloc_list); 6917 INIT_LIST_HEAD(&dev->post_commit_list); 6918 6919 atomic_set(&dev->reada_in_flight, 0); 6920 atomic_set(&dev->dev_stats_ccnt, 0); 6921 btrfs_device_data_ordered_init(dev); 6922 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6923 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); 6924 extent_io_tree_init(fs_info, &dev->alloc_state, 6925 IO_TREE_DEVICE_ALLOC_STATE, NULL); 6926 6927 if (devid) 6928 tmp = *devid; 6929 else { 6930 int ret; 6931 6932 ret = find_next_devid(fs_info, &tmp); 6933 if (ret) { 6934 btrfs_free_device(dev); 6935 return ERR_PTR(ret); 6936 } 6937 } 6938 dev->devid = tmp; 6939 6940 if (uuid) 6941 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); 6942 else 6943 generate_random_uuid(dev->uuid); 6944 6945 return dev; 6946 } 6947 6948 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, 6949 u64 devid, u8 *uuid, bool error) 6950 { 6951 if (error) 6952 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", 6953 devid, uuid); 6954 else 6955 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", 6956 devid, uuid); 6957 } 6958 6959 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) 6960 { 6961 const int data_stripes = calc_data_stripes(type, num_stripes); 6962 6963 return div_u64(chunk_len, data_stripes); 6964 } 6965 6966 #if BITS_PER_LONG == 32 6967 /* 6968 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE 6969 * can't be accessed on 32bit systems. 6970 * 6971 * This function do mount time check to reject the fs if it already has 6972 * metadata chunk beyond that limit. 6973 */ 6974 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6975 u64 logical, u64 length, u64 type) 6976 { 6977 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6978 return 0; 6979 6980 if (logical + length < MAX_LFS_FILESIZE) 6981 return 0; 6982 6983 btrfs_err_32bit_limit(fs_info); 6984 return -EOVERFLOW; 6985 } 6986 6987 /* 6988 * This is to give early warning for any metadata chunk reaching 6989 * BTRFS_32BIT_EARLY_WARN_THRESHOLD. 6990 * Although we can still access the metadata, it's not going to be possible 6991 * once the limit is reached. 6992 */ 6993 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, 6994 u64 logical, u64 length, u64 type) 6995 { 6996 if (!(type & BTRFS_BLOCK_GROUP_METADATA)) 6997 return; 6998 6999 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) 7000 return; 7001 7002 btrfs_warn_32bit_limit(fs_info); 7003 } 7004 #endif 7005 7006 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, 7007 struct btrfs_chunk *chunk) 7008 { 7009 BTRFS_DEV_LOOKUP_ARGS(args); 7010 struct btrfs_fs_info *fs_info = leaf->fs_info; 7011 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7012 struct map_lookup *map; 7013 struct extent_map *em; 7014 u64 logical; 7015 u64 length; 7016 u64 devid; 7017 u64 type; 7018 u8 uuid[BTRFS_UUID_SIZE]; 7019 int num_stripes; 7020 int ret; 7021 int i; 7022 7023 logical = key->offset; 7024 length = btrfs_chunk_length(leaf, chunk); 7025 type = btrfs_chunk_type(leaf, chunk); 7026 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 7027 7028 #if BITS_PER_LONG == 32 7029 ret = check_32bit_meta_chunk(fs_info, logical, length, type); 7030 if (ret < 0) 7031 return ret; 7032 warn_32bit_meta_chunk(fs_info, logical, length, type); 7033 #endif 7034 7035 /* 7036 * Only need to verify chunk item if we're reading from sys chunk array, 7037 * as chunk item in tree block is already verified by tree-checker. 7038 */ 7039 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { 7040 ret = btrfs_check_chunk_valid(leaf, chunk, logical); 7041 if (ret) 7042 return ret; 7043 } 7044 7045 read_lock(&map_tree->lock); 7046 em = lookup_extent_mapping(map_tree, logical, 1); 7047 read_unlock(&map_tree->lock); 7048 7049 /* already mapped? */ 7050 if (em && em->start <= logical && em->start + em->len > logical) { 7051 free_extent_map(em); 7052 return 0; 7053 } else if (em) { 7054 free_extent_map(em); 7055 } 7056 7057 em = alloc_extent_map(); 7058 if (!em) 7059 return -ENOMEM; 7060 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); 7061 if (!map) { 7062 free_extent_map(em); 7063 return -ENOMEM; 7064 } 7065 7066 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); 7067 em->map_lookup = map; 7068 em->start = logical; 7069 em->len = length; 7070 em->orig_start = 0; 7071 em->block_start = 0; 7072 em->block_len = em->len; 7073 7074 map->num_stripes = num_stripes; 7075 map->io_width = btrfs_chunk_io_width(leaf, chunk); 7076 map->io_align = btrfs_chunk_io_align(leaf, chunk); 7077 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 7078 map->type = type; 7079 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 7080 map->verified_stripes = 0; 7081 em->orig_block_len = calc_stripe_length(type, em->len, 7082 map->num_stripes); 7083 for (i = 0; i < num_stripes; i++) { 7084 map->stripes[i].physical = 7085 btrfs_stripe_offset_nr(leaf, chunk, i); 7086 devid = btrfs_stripe_devid_nr(leaf, chunk, i); 7087 args.devid = devid; 7088 read_extent_buffer(leaf, uuid, (unsigned long) 7089 btrfs_stripe_dev_uuid_nr(chunk, i), 7090 BTRFS_UUID_SIZE); 7091 args.uuid = uuid; 7092 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); 7093 if (!map->stripes[i].dev && 7094 !btrfs_test_opt(fs_info, DEGRADED)) { 7095 free_extent_map(em); 7096 btrfs_report_missing_device(fs_info, devid, uuid, true); 7097 return -ENOENT; 7098 } 7099 if (!map->stripes[i].dev) { 7100 map->stripes[i].dev = 7101 add_missing_dev(fs_info->fs_devices, devid, 7102 uuid); 7103 if (IS_ERR(map->stripes[i].dev)) { 7104 free_extent_map(em); 7105 btrfs_err(fs_info, 7106 "failed to init missing dev %llu: %ld", 7107 devid, PTR_ERR(map->stripes[i].dev)); 7108 return PTR_ERR(map->stripes[i].dev); 7109 } 7110 btrfs_report_missing_device(fs_info, devid, uuid, false); 7111 } 7112 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, 7113 &(map->stripes[i].dev->dev_state)); 7114 7115 } 7116 7117 write_lock(&map_tree->lock); 7118 ret = add_extent_mapping(map_tree, em, 0); 7119 write_unlock(&map_tree->lock); 7120 if (ret < 0) { 7121 btrfs_err(fs_info, 7122 "failed to add chunk map, start=%llu len=%llu: %d", 7123 em->start, em->len, ret); 7124 } 7125 free_extent_map(em); 7126 7127 return ret; 7128 } 7129 7130 static void fill_device_from_item(struct extent_buffer *leaf, 7131 struct btrfs_dev_item *dev_item, 7132 struct btrfs_device *device) 7133 { 7134 unsigned long ptr; 7135 7136 device->devid = btrfs_device_id(leaf, dev_item); 7137 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); 7138 device->total_bytes = device->disk_total_bytes; 7139 device->commit_total_bytes = device->disk_total_bytes; 7140 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); 7141 device->commit_bytes_used = device->bytes_used; 7142 device->type = btrfs_device_type(leaf, dev_item); 7143 device->io_align = btrfs_device_io_align(leaf, dev_item); 7144 device->io_width = btrfs_device_io_width(leaf, dev_item); 7145 device->sector_size = btrfs_device_sector_size(leaf, dev_item); 7146 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); 7147 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); 7148 7149 ptr = btrfs_device_uuid(dev_item); 7150 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); 7151 } 7152 7153 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, 7154 u8 *fsid) 7155 { 7156 struct btrfs_fs_devices *fs_devices; 7157 int ret; 7158 7159 lockdep_assert_held(&uuid_mutex); 7160 ASSERT(fsid); 7161 7162 /* This will match only for multi-device seed fs */ 7163 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) 7164 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) 7165 return fs_devices; 7166 7167 7168 fs_devices = find_fsid(fsid, NULL); 7169 if (!fs_devices) { 7170 if (!btrfs_test_opt(fs_info, DEGRADED)) 7171 return ERR_PTR(-ENOENT); 7172 7173 fs_devices = alloc_fs_devices(fsid, NULL); 7174 if (IS_ERR(fs_devices)) 7175 return fs_devices; 7176 7177 fs_devices->seeding = true; 7178 fs_devices->opened = 1; 7179 return fs_devices; 7180 } 7181 7182 /* 7183 * Upon first call for a seed fs fsid, just create a private copy of the 7184 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list 7185 */ 7186 fs_devices = clone_fs_devices(fs_devices); 7187 if (IS_ERR(fs_devices)) 7188 return fs_devices; 7189 7190 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); 7191 if (ret) { 7192 free_fs_devices(fs_devices); 7193 return ERR_PTR(ret); 7194 } 7195 7196 if (!fs_devices->seeding) { 7197 close_fs_devices(fs_devices); 7198 free_fs_devices(fs_devices); 7199 return ERR_PTR(-EINVAL); 7200 } 7201 7202 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); 7203 7204 return fs_devices; 7205 } 7206 7207 static int read_one_dev(struct extent_buffer *leaf, 7208 struct btrfs_dev_item *dev_item) 7209 { 7210 BTRFS_DEV_LOOKUP_ARGS(args); 7211 struct btrfs_fs_info *fs_info = leaf->fs_info; 7212 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7213 struct btrfs_device *device; 7214 u64 devid; 7215 int ret; 7216 u8 fs_uuid[BTRFS_FSID_SIZE]; 7217 u8 dev_uuid[BTRFS_UUID_SIZE]; 7218 7219 devid = args.devid = btrfs_device_id(leaf, dev_item); 7220 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), 7221 BTRFS_UUID_SIZE); 7222 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), 7223 BTRFS_FSID_SIZE); 7224 args.uuid = dev_uuid; 7225 args.fsid = fs_uuid; 7226 7227 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { 7228 fs_devices = open_seed_devices(fs_info, fs_uuid); 7229 if (IS_ERR(fs_devices)) 7230 return PTR_ERR(fs_devices); 7231 } 7232 7233 device = btrfs_find_device(fs_info->fs_devices, &args); 7234 if (!device) { 7235 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7236 btrfs_report_missing_device(fs_info, devid, 7237 dev_uuid, true); 7238 return -ENOENT; 7239 } 7240 7241 device = add_missing_dev(fs_devices, devid, dev_uuid); 7242 if (IS_ERR(device)) { 7243 btrfs_err(fs_info, 7244 "failed to add missing dev %llu: %ld", 7245 devid, PTR_ERR(device)); 7246 return PTR_ERR(device); 7247 } 7248 btrfs_report_missing_device(fs_info, devid, dev_uuid, false); 7249 } else { 7250 if (!device->bdev) { 7251 if (!btrfs_test_opt(fs_info, DEGRADED)) { 7252 btrfs_report_missing_device(fs_info, 7253 devid, dev_uuid, true); 7254 return -ENOENT; 7255 } 7256 btrfs_report_missing_device(fs_info, devid, 7257 dev_uuid, false); 7258 } 7259 7260 if (!device->bdev && 7261 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { 7262 /* 7263 * this happens when a device that was properly setup 7264 * in the device info lists suddenly goes bad. 7265 * device->bdev is NULL, and so we have to set 7266 * device->missing to one here 7267 */ 7268 device->fs_devices->missing_devices++; 7269 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); 7270 } 7271 7272 /* Move the device to its own fs_devices */ 7273 if (device->fs_devices != fs_devices) { 7274 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, 7275 &device->dev_state)); 7276 7277 list_move(&device->dev_list, &fs_devices->devices); 7278 device->fs_devices->num_devices--; 7279 fs_devices->num_devices++; 7280 7281 device->fs_devices->missing_devices--; 7282 fs_devices->missing_devices++; 7283 7284 device->fs_devices = fs_devices; 7285 } 7286 } 7287 7288 if (device->fs_devices != fs_info->fs_devices) { 7289 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); 7290 if (device->generation != 7291 btrfs_device_generation(leaf, dev_item)) 7292 return -EINVAL; 7293 } 7294 7295 fill_device_from_item(leaf, dev_item, device); 7296 if (device->bdev) { 7297 u64 max_total_bytes = i_size_read(device->bdev->bd_inode); 7298 7299 if (device->total_bytes > max_total_bytes) { 7300 btrfs_err(fs_info, 7301 "device total_bytes should be at most %llu but found %llu", 7302 max_total_bytes, device->total_bytes); 7303 return -EINVAL; 7304 } 7305 } 7306 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); 7307 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && 7308 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 7309 device->fs_devices->total_rw_bytes += device->total_bytes; 7310 atomic64_add(device->total_bytes - device->bytes_used, 7311 &fs_info->free_chunk_space); 7312 } 7313 ret = 0; 7314 return ret; 7315 } 7316 7317 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) 7318 { 7319 struct btrfs_root *root = fs_info->tree_root; 7320 struct btrfs_super_block *super_copy = fs_info->super_copy; 7321 struct extent_buffer *sb; 7322 struct btrfs_disk_key *disk_key; 7323 struct btrfs_chunk *chunk; 7324 u8 *array_ptr; 7325 unsigned long sb_array_offset; 7326 int ret = 0; 7327 u32 num_stripes; 7328 u32 array_size; 7329 u32 len = 0; 7330 u32 cur_offset; 7331 u64 type; 7332 struct btrfs_key key; 7333 7334 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); 7335 /* 7336 * This will create extent buffer of nodesize, superblock size is 7337 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will 7338 * overallocate but we can keep it as-is, only the first page is used. 7339 */ 7340 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, 7341 root->root_key.objectid, 0); 7342 if (IS_ERR(sb)) 7343 return PTR_ERR(sb); 7344 set_extent_buffer_uptodate(sb); 7345 /* 7346 * The sb extent buffer is artificial and just used to read the system array. 7347 * set_extent_buffer_uptodate() call does not properly mark all it's 7348 * pages up-to-date when the page is larger: extent does not cover the 7349 * whole page and consequently check_page_uptodate does not find all 7350 * the page's extents up-to-date (the hole beyond sb), 7351 * write_extent_buffer then triggers a WARN_ON. 7352 * 7353 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, 7354 * but sb spans only this function. Add an explicit SetPageUptodate call 7355 * to silence the warning eg. on PowerPC 64. 7356 */ 7357 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) 7358 SetPageUptodate(sb->pages[0]); 7359 7360 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); 7361 array_size = btrfs_super_sys_array_size(super_copy); 7362 7363 array_ptr = super_copy->sys_chunk_array; 7364 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); 7365 cur_offset = 0; 7366 7367 while (cur_offset < array_size) { 7368 disk_key = (struct btrfs_disk_key *)array_ptr; 7369 len = sizeof(*disk_key); 7370 if (cur_offset + len > array_size) 7371 goto out_short_read; 7372 7373 btrfs_disk_key_to_cpu(&key, disk_key); 7374 7375 array_ptr += len; 7376 sb_array_offset += len; 7377 cur_offset += len; 7378 7379 if (key.type != BTRFS_CHUNK_ITEM_KEY) { 7380 btrfs_err(fs_info, 7381 "unexpected item type %u in sys_array at offset %u", 7382 (u32)key.type, cur_offset); 7383 ret = -EIO; 7384 break; 7385 } 7386 7387 chunk = (struct btrfs_chunk *)sb_array_offset; 7388 /* 7389 * At least one btrfs_chunk with one stripe must be present, 7390 * exact stripe count check comes afterwards 7391 */ 7392 len = btrfs_chunk_item_size(1); 7393 if (cur_offset + len > array_size) 7394 goto out_short_read; 7395 7396 num_stripes = btrfs_chunk_num_stripes(sb, chunk); 7397 if (!num_stripes) { 7398 btrfs_err(fs_info, 7399 "invalid number of stripes %u in sys_array at offset %u", 7400 num_stripes, cur_offset); 7401 ret = -EIO; 7402 break; 7403 } 7404 7405 type = btrfs_chunk_type(sb, chunk); 7406 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { 7407 btrfs_err(fs_info, 7408 "invalid chunk type %llu in sys_array at offset %u", 7409 type, cur_offset); 7410 ret = -EIO; 7411 break; 7412 } 7413 7414 len = btrfs_chunk_item_size(num_stripes); 7415 if (cur_offset + len > array_size) 7416 goto out_short_read; 7417 7418 ret = read_one_chunk(&key, sb, chunk); 7419 if (ret) 7420 break; 7421 7422 array_ptr += len; 7423 sb_array_offset += len; 7424 cur_offset += len; 7425 } 7426 clear_extent_buffer_uptodate(sb); 7427 free_extent_buffer_stale(sb); 7428 return ret; 7429 7430 out_short_read: 7431 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", 7432 len, cur_offset); 7433 clear_extent_buffer_uptodate(sb); 7434 free_extent_buffer_stale(sb); 7435 return -EIO; 7436 } 7437 7438 /* 7439 * Check if all chunks in the fs are OK for read-write degraded mount 7440 * 7441 * If the @failing_dev is specified, it's accounted as missing. 7442 * 7443 * Return true if all chunks meet the minimal RW mount requirements. 7444 * Return false if any chunk doesn't meet the minimal RW mount requirements. 7445 */ 7446 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 7447 struct btrfs_device *failing_dev) 7448 { 7449 struct extent_map_tree *map_tree = &fs_info->mapping_tree; 7450 struct extent_map *em; 7451 u64 next_start = 0; 7452 bool ret = true; 7453 7454 read_lock(&map_tree->lock); 7455 em = lookup_extent_mapping(map_tree, 0, (u64)-1); 7456 read_unlock(&map_tree->lock); 7457 /* No chunk at all? Return false anyway */ 7458 if (!em) { 7459 ret = false; 7460 goto out; 7461 } 7462 while (em) { 7463 struct map_lookup *map; 7464 int missing = 0; 7465 int max_tolerated; 7466 int i; 7467 7468 map = em->map_lookup; 7469 max_tolerated = 7470 btrfs_get_num_tolerated_disk_barrier_failures( 7471 map->type); 7472 for (i = 0; i < map->num_stripes; i++) { 7473 struct btrfs_device *dev = map->stripes[i].dev; 7474 7475 if (!dev || !dev->bdev || 7476 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || 7477 dev->last_flush_error) 7478 missing++; 7479 else if (failing_dev && failing_dev == dev) 7480 missing++; 7481 } 7482 if (missing > max_tolerated) { 7483 if (!failing_dev) 7484 btrfs_warn(fs_info, 7485 "chunk %llu missing %d devices, max tolerance is %d for writable mount", 7486 em->start, missing, max_tolerated); 7487 free_extent_map(em); 7488 ret = false; 7489 goto out; 7490 } 7491 next_start = extent_map_end(em); 7492 free_extent_map(em); 7493 7494 read_lock(&map_tree->lock); 7495 em = lookup_extent_mapping(map_tree, next_start, 7496 (u64)(-1) - next_start); 7497 read_unlock(&map_tree->lock); 7498 } 7499 out: 7500 return ret; 7501 } 7502 7503 static void readahead_tree_node_children(struct extent_buffer *node) 7504 { 7505 int i; 7506 const int nr_items = btrfs_header_nritems(node); 7507 7508 for (i = 0; i < nr_items; i++) 7509 btrfs_readahead_node_child(node, i); 7510 } 7511 7512 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) 7513 { 7514 struct btrfs_root *root = fs_info->chunk_root; 7515 struct btrfs_path *path; 7516 struct extent_buffer *leaf; 7517 struct btrfs_key key; 7518 struct btrfs_key found_key; 7519 int ret; 7520 int slot; 7521 u64 total_dev = 0; 7522 u64 last_ra_node = 0; 7523 7524 path = btrfs_alloc_path(); 7525 if (!path) 7526 return -ENOMEM; 7527 7528 /* 7529 * uuid_mutex is needed only if we are mounting a sprout FS 7530 * otherwise we don't need it. 7531 */ 7532 mutex_lock(&uuid_mutex); 7533 7534 /* 7535 * It is possible for mount and umount to race in such a way that 7536 * we execute this code path, but open_fs_devices failed to clear 7537 * total_rw_bytes. We certainly want it cleared before reading the 7538 * device items, so clear it here. 7539 */ 7540 fs_info->fs_devices->total_rw_bytes = 0; 7541 7542 /* 7543 * Read all device items, and then all the chunk items. All 7544 * device items are found before any chunk item (their object id 7545 * is smaller than the lowest possible object id for a chunk 7546 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). 7547 */ 7548 key.objectid = BTRFS_DEV_ITEMS_OBJECTID; 7549 key.offset = 0; 7550 key.type = 0; 7551 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 7552 if (ret < 0) 7553 goto error; 7554 while (1) { 7555 struct extent_buffer *node; 7556 7557 leaf = path->nodes[0]; 7558 slot = path->slots[0]; 7559 if (slot >= btrfs_header_nritems(leaf)) { 7560 ret = btrfs_next_leaf(root, path); 7561 if (ret == 0) 7562 continue; 7563 if (ret < 0) 7564 goto error; 7565 break; 7566 } 7567 /* 7568 * The nodes on level 1 are not locked but we don't need to do 7569 * that during mount time as nothing else can access the tree 7570 */ 7571 node = path->nodes[1]; 7572 if (node) { 7573 if (last_ra_node != node->start) { 7574 readahead_tree_node_children(node); 7575 last_ra_node = node->start; 7576 } 7577 } 7578 btrfs_item_key_to_cpu(leaf, &found_key, slot); 7579 if (found_key.type == BTRFS_DEV_ITEM_KEY) { 7580 struct btrfs_dev_item *dev_item; 7581 dev_item = btrfs_item_ptr(leaf, slot, 7582 struct btrfs_dev_item); 7583 ret = read_one_dev(leaf, dev_item); 7584 if (ret) 7585 goto error; 7586 total_dev++; 7587 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { 7588 struct btrfs_chunk *chunk; 7589 7590 /* 7591 * We are only called at mount time, so no need to take 7592 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, 7593 * we always lock first fs_info->chunk_mutex before 7594 * acquiring any locks on the chunk tree. This is a 7595 * requirement for chunk allocation, see the comment on 7596 * top of btrfs_chunk_alloc() for details. 7597 */ 7598 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); 7599 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 7600 ret = read_one_chunk(&found_key, leaf, chunk); 7601 if (ret) 7602 goto error; 7603 } 7604 path->slots[0]++; 7605 } 7606 7607 /* 7608 * After loading chunk tree, we've got all device information, 7609 * do another round of validation checks. 7610 */ 7611 if (total_dev != fs_info->fs_devices->total_devices) { 7612 btrfs_err(fs_info, 7613 "super_num_devices %llu mismatch with num_devices %llu found here", 7614 btrfs_super_num_devices(fs_info->super_copy), 7615 total_dev); 7616 ret = -EINVAL; 7617 goto error; 7618 } 7619 if (btrfs_super_total_bytes(fs_info->super_copy) < 7620 fs_info->fs_devices->total_rw_bytes) { 7621 btrfs_err(fs_info, 7622 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", 7623 btrfs_super_total_bytes(fs_info->super_copy), 7624 fs_info->fs_devices->total_rw_bytes); 7625 ret = -EINVAL; 7626 goto error; 7627 } 7628 ret = 0; 7629 error: 7630 mutex_unlock(&uuid_mutex); 7631 7632 btrfs_free_path(path); 7633 return ret; 7634 } 7635 7636 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) 7637 { 7638 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7639 struct btrfs_device *device; 7640 7641 fs_devices->fs_info = fs_info; 7642 7643 mutex_lock(&fs_devices->device_list_mutex); 7644 list_for_each_entry(device, &fs_devices->devices, dev_list) 7645 device->fs_info = fs_info; 7646 7647 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7648 list_for_each_entry(device, &seed_devs->devices, dev_list) 7649 device->fs_info = fs_info; 7650 7651 seed_devs->fs_info = fs_info; 7652 } 7653 mutex_unlock(&fs_devices->device_list_mutex); 7654 } 7655 7656 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, 7657 const struct btrfs_dev_stats_item *ptr, 7658 int index) 7659 { 7660 u64 val; 7661 7662 read_extent_buffer(eb, &val, 7663 offsetof(struct btrfs_dev_stats_item, values) + 7664 ((unsigned long)ptr) + (index * sizeof(u64)), 7665 sizeof(val)); 7666 return val; 7667 } 7668 7669 static void btrfs_set_dev_stats_value(struct extent_buffer *eb, 7670 struct btrfs_dev_stats_item *ptr, 7671 int index, u64 val) 7672 { 7673 write_extent_buffer(eb, &val, 7674 offsetof(struct btrfs_dev_stats_item, values) + 7675 ((unsigned long)ptr) + (index * sizeof(u64)), 7676 sizeof(val)); 7677 } 7678 7679 static int btrfs_device_init_dev_stats(struct btrfs_device *device, 7680 struct btrfs_path *path) 7681 { 7682 struct btrfs_dev_stats_item *ptr; 7683 struct extent_buffer *eb; 7684 struct btrfs_key key; 7685 int item_size; 7686 int i, ret, slot; 7687 7688 if (!device->fs_info->dev_root) 7689 return 0; 7690 7691 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7692 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7693 key.offset = device->devid; 7694 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); 7695 if (ret) { 7696 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7697 btrfs_dev_stat_set(device, i, 0); 7698 device->dev_stats_valid = 1; 7699 btrfs_release_path(path); 7700 return ret < 0 ? ret : 0; 7701 } 7702 slot = path->slots[0]; 7703 eb = path->nodes[0]; 7704 item_size = btrfs_item_size_nr(eb, slot); 7705 7706 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); 7707 7708 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7709 if (item_size >= (1 + i) * sizeof(__le64)) 7710 btrfs_dev_stat_set(device, i, 7711 btrfs_dev_stats_value(eb, ptr, i)); 7712 else 7713 btrfs_dev_stat_set(device, i, 0); 7714 } 7715 7716 device->dev_stats_valid = 1; 7717 btrfs_dev_stat_print_on_load(device); 7718 btrfs_release_path(path); 7719 7720 return 0; 7721 } 7722 7723 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) 7724 { 7725 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; 7726 struct btrfs_device *device; 7727 struct btrfs_path *path = NULL; 7728 int ret = 0; 7729 7730 path = btrfs_alloc_path(); 7731 if (!path) 7732 return -ENOMEM; 7733 7734 mutex_lock(&fs_devices->device_list_mutex); 7735 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7736 ret = btrfs_device_init_dev_stats(device, path); 7737 if (ret) 7738 goto out; 7739 } 7740 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { 7741 list_for_each_entry(device, &seed_devs->devices, dev_list) { 7742 ret = btrfs_device_init_dev_stats(device, path); 7743 if (ret) 7744 goto out; 7745 } 7746 } 7747 out: 7748 mutex_unlock(&fs_devices->device_list_mutex); 7749 7750 btrfs_free_path(path); 7751 return ret; 7752 } 7753 7754 static int update_dev_stat_item(struct btrfs_trans_handle *trans, 7755 struct btrfs_device *device) 7756 { 7757 struct btrfs_fs_info *fs_info = trans->fs_info; 7758 struct btrfs_root *dev_root = fs_info->dev_root; 7759 struct btrfs_path *path; 7760 struct btrfs_key key; 7761 struct extent_buffer *eb; 7762 struct btrfs_dev_stats_item *ptr; 7763 int ret; 7764 int i; 7765 7766 key.objectid = BTRFS_DEV_STATS_OBJECTID; 7767 key.type = BTRFS_PERSISTENT_ITEM_KEY; 7768 key.offset = device->devid; 7769 7770 path = btrfs_alloc_path(); 7771 if (!path) 7772 return -ENOMEM; 7773 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); 7774 if (ret < 0) { 7775 btrfs_warn_in_rcu(fs_info, 7776 "error %d while searching for dev_stats item for device %s", 7777 ret, rcu_str_deref(device->name)); 7778 goto out; 7779 } 7780 7781 if (ret == 0 && 7782 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { 7783 /* need to delete old one and insert a new one */ 7784 ret = btrfs_del_item(trans, dev_root, path); 7785 if (ret != 0) { 7786 btrfs_warn_in_rcu(fs_info, 7787 "delete too small dev_stats item for device %s failed %d", 7788 rcu_str_deref(device->name), ret); 7789 goto out; 7790 } 7791 ret = 1; 7792 } 7793 7794 if (ret == 1) { 7795 /* need to insert a new item */ 7796 btrfs_release_path(path); 7797 ret = btrfs_insert_empty_item(trans, dev_root, path, 7798 &key, sizeof(*ptr)); 7799 if (ret < 0) { 7800 btrfs_warn_in_rcu(fs_info, 7801 "insert dev_stats item for device %s failed %d", 7802 rcu_str_deref(device->name), ret); 7803 goto out; 7804 } 7805 } 7806 7807 eb = path->nodes[0]; 7808 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); 7809 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7810 btrfs_set_dev_stats_value(eb, ptr, i, 7811 btrfs_dev_stat_read(device, i)); 7812 btrfs_mark_buffer_dirty(eb); 7813 7814 out: 7815 btrfs_free_path(path); 7816 return ret; 7817 } 7818 7819 /* 7820 * called from commit_transaction. Writes all changed device stats to disk. 7821 */ 7822 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) 7823 { 7824 struct btrfs_fs_info *fs_info = trans->fs_info; 7825 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7826 struct btrfs_device *device; 7827 int stats_cnt; 7828 int ret = 0; 7829 7830 mutex_lock(&fs_devices->device_list_mutex); 7831 list_for_each_entry(device, &fs_devices->devices, dev_list) { 7832 stats_cnt = atomic_read(&device->dev_stats_ccnt); 7833 if (!device->dev_stats_valid || stats_cnt == 0) 7834 continue; 7835 7836 7837 /* 7838 * There is a LOAD-LOAD control dependency between the value of 7839 * dev_stats_ccnt and updating the on-disk values which requires 7840 * reading the in-memory counters. Such control dependencies 7841 * require explicit read memory barriers. 7842 * 7843 * This memory barriers pairs with smp_mb__before_atomic in 7844 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full 7845 * barrier implied by atomic_xchg in 7846 * btrfs_dev_stats_read_and_reset 7847 */ 7848 smp_rmb(); 7849 7850 ret = update_dev_stat_item(trans, device); 7851 if (!ret) 7852 atomic_sub(stats_cnt, &device->dev_stats_ccnt); 7853 } 7854 mutex_unlock(&fs_devices->device_list_mutex); 7855 7856 return ret; 7857 } 7858 7859 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) 7860 { 7861 btrfs_dev_stat_inc(dev, index); 7862 btrfs_dev_stat_print_on_error(dev); 7863 } 7864 7865 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) 7866 { 7867 if (!dev->dev_stats_valid) 7868 return; 7869 btrfs_err_rl_in_rcu(dev->fs_info, 7870 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7871 rcu_str_deref(dev->name), 7872 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7873 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7874 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7875 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7876 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7877 } 7878 7879 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) 7880 { 7881 int i; 7882 7883 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7884 if (btrfs_dev_stat_read(dev, i) != 0) 7885 break; 7886 if (i == BTRFS_DEV_STAT_VALUES_MAX) 7887 return; /* all values == 0, suppress message */ 7888 7889 btrfs_info_in_rcu(dev->fs_info, 7890 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", 7891 rcu_str_deref(dev->name), 7892 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), 7893 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), 7894 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), 7895 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), 7896 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); 7897 } 7898 7899 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 7900 struct btrfs_ioctl_get_dev_stats *stats) 7901 { 7902 BTRFS_DEV_LOOKUP_ARGS(args); 7903 struct btrfs_device *dev; 7904 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 7905 int i; 7906 7907 mutex_lock(&fs_devices->device_list_mutex); 7908 args.devid = stats->devid; 7909 dev = btrfs_find_device(fs_info->fs_devices, &args); 7910 mutex_unlock(&fs_devices->device_list_mutex); 7911 7912 if (!dev) { 7913 btrfs_warn(fs_info, "get dev_stats failed, device not found"); 7914 return -ENODEV; 7915 } else if (!dev->dev_stats_valid) { 7916 btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); 7917 return -ENODEV; 7918 } else if (stats->flags & BTRFS_DEV_STATS_RESET) { 7919 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { 7920 if (stats->nr_items > i) 7921 stats->values[i] = 7922 btrfs_dev_stat_read_and_reset(dev, i); 7923 else 7924 btrfs_dev_stat_set(dev, i, 0); 7925 } 7926 btrfs_info(fs_info, "device stats zeroed by %s (%d)", 7927 current->comm, task_pid_nr(current)); 7928 } else { 7929 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) 7930 if (stats->nr_items > i) 7931 stats->values[i] = btrfs_dev_stat_read(dev, i); 7932 } 7933 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) 7934 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; 7935 return 0; 7936 } 7937 7938 /* 7939 * Update the size and bytes used for each device where it changed. This is 7940 * delayed since we would otherwise get errors while writing out the 7941 * superblocks. 7942 * 7943 * Must be invoked during transaction commit. 7944 */ 7945 void btrfs_commit_device_sizes(struct btrfs_transaction *trans) 7946 { 7947 struct btrfs_device *curr, *next; 7948 7949 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); 7950 7951 if (list_empty(&trans->dev_update_list)) 7952 return; 7953 7954 /* 7955 * We don't need the device_list_mutex here. This list is owned by the 7956 * transaction and the transaction must complete before the device is 7957 * released. 7958 */ 7959 mutex_lock(&trans->fs_info->chunk_mutex); 7960 list_for_each_entry_safe(curr, next, &trans->dev_update_list, 7961 post_commit_list) { 7962 list_del_init(&curr->post_commit_list); 7963 curr->commit_total_bytes = curr->disk_total_bytes; 7964 curr->commit_bytes_used = curr->bytes_used; 7965 } 7966 mutex_unlock(&trans->fs_info->chunk_mutex); 7967 } 7968 7969 /* 7970 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. 7971 */ 7972 int btrfs_bg_type_to_factor(u64 flags) 7973 { 7974 const int index = btrfs_bg_flags_to_raid_index(flags); 7975 7976 return btrfs_raid_array[index].ncopies; 7977 } 7978 7979 7980 7981 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, 7982 u64 chunk_offset, u64 devid, 7983 u64 physical_offset, u64 physical_len) 7984 { 7985 struct btrfs_dev_lookup_args args = { .devid = devid }; 7986 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 7987 struct extent_map *em; 7988 struct map_lookup *map; 7989 struct btrfs_device *dev; 7990 u64 stripe_len; 7991 bool found = false; 7992 int ret = 0; 7993 int i; 7994 7995 read_lock(&em_tree->lock); 7996 em = lookup_extent_mapping(em_tree, chunk_offset, 1); 7997 read_unlock(&em_tree->lock); 7998 7999 if (!em) { 8000 btrfs_err(fs_info, 8001 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", 8002 physical_offset, devid); 8003 ret = -EUCLEAN; 8004 goto out; 8005 } 8006 8007 map = em->map_lookup; 8008 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); 8009 if (physical_len != stripe_len) { 8010 btrfs_err(fs_info, 8011 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", 8012 physical_offset, devid, em->start, physical_len, 8013 stripe_len); 8014 ret = -EUCLEAN; 8015 goto out; 8016 } 8017 8018 for (i = 0; i < map->num_stripes; i++) { 8019 if (map->stripes[i].dev->devid == devid && 8020 map->stripes[i].physical == physical_offset) { 8021 found = true; 8022 if (map->verified_stripes >= map->num_stripes) { 8023 btrfs_err(fs_info, 8024 "too many dev extents for chunk %llu found", 8025 em->start); 8026 ret = -EUCLEAN; 8027 goto out; 8028 } 8029 map->verified_stripes++; 8030 break; 8031 } 8032 } 8033 if (!found) { 8034 btrfs_err(fs_info, 8035 "dev extent physical offset %llu devid %llu has no corresponding chunk", 8036 physical_offset, devid); 8037 ret = -EUCLEAN; 8038 } 8039 8040 /* Make sure no dev extent is beyond device boundary */ 8041 dev = btrfs_find_device(fs_info->fs_devices, &args); 8042 if (!dev) { 8043 btrfs_err(fs_info, "failed to find devid %llu", devid); 8044 ret = -EUCLEAN; 8045 goto out; 8046 } 8047 8048 if (physical_offset + physical_len > dev->disk_total_bytes) { 8049 btrfs_err(fs_info, 8050 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", 8051 devid, physical_offset, physical_len, 8052 dev->disk_total_bytes); 8053 ret = -EUCLEAN; 8054 goto out; 8055 } 8056 8057 if (dev->zone_info) { 8058 u64 zone_size = dev->zone_info->zone_size; 8059 8060 if (!IS_ALIGNED(physical_offset, zone_size) || 8061 !IS_ALIGNED(physical_len, zone_size)) { 8062 btrfs_err(fs_info, 8063 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", 8064 devid, physical_offset, physical_len); 8065 ret = -EUCLEAN; 8066 goto out; 8067 } 8068 } 8069 8070 out: 8071 free_extent_map(em); 8072 return ret; 8073 } 8074 8075 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) 8076 { 8077 struct extent_map_tree *em_tree = &fs_info->mapping_tree; 8078 struct extent_map *em; 8079 struct rb_node *node; 8080 int ret = 0; 8081 8082 read_lock(&em_tree->lock); 8083 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { 8084 em = rb_entry(node, struct extent_map, rb_node); 8085 if (em->map_lookup->num_stripes != 8086 em->map_lookup->verified_stripes) { 8087 btrfs_err(fs_info, 8088 "chunk %llu has missing dev extent, have %d expect %d", 8089 em->start, em->map_lookup->verified_stripes, 8090 em->map_lookup->num_stripes); 8091 ret = -EUCLEAN; 8092 goto out; 8093 } 8094 } 8095 out: 8096 read_unlock(&em_tree->lock); 8097 return ret; 8098 } 8099 8100 /* 8101 * Ensure that all dev extents are mapped to correct chunk, otherwise 8102 * later chunk allocation/free would cause unexpected behavior. 8103 * 8104 * NOTE: This will iterate through the whole device tree, which should be of 8105 * the same size level as the chunk tree. This slightly increases mount time. 8106 */ 8107 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) 8108 { 8109 struct btrfs_path *path; 8110 struct btrfs_root *root = fs_info->dev_root; 8111 struct btrfs_key key; 8112 u64 prev_devid = 0; 8113 u64 prev_dev_ext_end = 0; 8114 int ret = 0; 8115 8116 /* 8117 * We don't have a dev_root because we mounted with ignorebadroots and 8118 * failed to load the root, so we want to skip the verification in this 8119 * case for sure. 8120 * 8121 * However if the dev root is fine, but the tree itself is corrupted 8122 * we'd still fail to mount. This verification is only to make sure 8123 * writes can happen safely, so instead just bypass this check 8124 * completely in the case of IGNOREBADROOTS. 8125 */ 8126 if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) 8127 return 0; 8128 8129 key.objectid = 1; 8130 key.type = BTRFS_DEV_EXTENT_KEY; 8131 key.offset = 0; 8132 8133 path = btrfs_alloc_path(); 8134 if (!path) 8135 return -ENOMEM; 8136 8137 path->reada = READA_FORWARD; 8138 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 8139 if (ret < 0) 8140 goto out; 8141 8142 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 8143 ret = btrfs_next_leaf(root, path); 8144 if (ret < 0) 8145 goto out; 8146 /* No dev extents at all? Not good */ 8147 if (ret > 0) { 8148 ret = -EUCLEAN; 8149 goto out; 8150 } 8151 } 8152 while (1) { 8153 struct extent_buffer *leaf = path->nodes[0]; 8154 struct btrfs_dev_extent *dext; 8155 int slot = path->slots[0]; 8156 u64 chunk_offset; 8157 u64 physical_offset; 8158 u64 physical_len; 8159 u64 devid; 8160 8161 btrfs_item_key_to_cpu(leaf, &key, slot); 8162 if (key.type != BTRFS_DEV_EXTENT_KEY) 8163 break; 8164 devid = key.objectid; 8165 physical_offset = key.offset; 8166 8167 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 8168 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); 8169 physical_len = btrfs_dev_extent_length(leaf, dext); 8170 8171 /* Check if this dev extent overlaps with the previous one */ 8172 if (devid == prev_devid && physical_offset < prev_dev_ext_end) { 8173 btrfs_err(fs_info, 8174 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", 8175 devid, physical_offset, prev_dev_ext_end); 8176 ret = -EUCLEAN; 8177 goto out; 8178 } 8179 8180 ret = verify_one_dev_extent(fs_info, chunk_offset, devid, 8181 physical_offset, physical_len); 8182 if (ret < 0) 8183 goto out; 8184 prev_devid = devid; 8185 prev_dev_ext_end = physical_offset + physical_len; 8186 8187 ret = btrfs_next_item(root, path); 8188 if (ret < 0) 8189 goto out; 8190 if (ret > 0) { 8191 ret = 0; 8192 break; 8193 } 8194 } 8195 8196 /* Ensure all chunks have corresponding dev extents */ 8197 ret = verify_chunk_dev_extent_mapping(fs_info); 8198 out: 8199 btrfs_free_path(path); 8200 return ret; 8201 } 8202 8203 /* 8204 * Check whether the given block group or device is pinned by any inode being 8205 * used as a swapfile. 8206 */ 8207 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) 8208 { 8209 struct btrfs_swapfile_pin *sp; 8210 struct rb_node *node; 8211 8212 spin_lock(&fs_info->swapfile_pins_lock); 8213 node = fs_info->swapfile_pins.rb_node; 8214 while (node) { 8215 sp = rb_entry(node, struct btrfs_swapfile_pin, node); 8216 if (ptr < sp->ptr) 8217 node = node->rb_left; 8218 else if (ptr > sp->ptr) 8219 node = node->rb_right; 8220 else 8221 break; 8222 } 8223 spin_unlock(&fs_info->swapfile_pins_lock); 8224 return node != NULL; 8225 } 8226 8227 static int relocating_repair_kthread(void *data) 8228 { 8229 struct btrfs_block_group *cache = (struct btrfs_block_group *)data; 8230 struct btrfs_fs_info *fs_info = cache->fs_info; 8231 u64 target; 8232 int ret = 0; 8233 8234 target = cache->start; 8235 btrfs_put_block_group(cache); 8236 8237 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 8238 btrfs_info(fs_info, 8239 "zoned: skip relocating block group %llu to repair: EBUSY", 8240 target); 8241 return -EBUSY; 8242 } 8243 8244 mutex_lock(&fs_info->reclaim_bgs_lock); 8245 8246 /* Ensure block group still exists */ 8247 cache = btrfs_lookup_block_group(fs_info, target); 8248 if (!cache) 8249 goto out; 8250 8251 if (!cache->relocating_repair) 8252 goto out; 8253 8254 ret = btrfs_may_alloc_data_chunk(fs_info, target); 8255 if (ret < 0) 8256 goto out; 8257 8258 btrfs_info(fs_info, 8259 "zoned: relocating block group %llu to repair IO failure", 8260 target); 8261 ret = btrfs_relocate_chunk(fs_info, target); 8262 8263 out: 8264 if (cache) 8265 btrfs_put_block_group(cache); 8266 mutex_unlock(&fs_info->reclaim_bgs_lock); 8267 btrfs_exclop_finish(fs_info); 8268 8269 return ret; 8270 } 8271 8272 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) 8273 { 8274 struct btrfs_block_group *cache; 8275 8276 /* Do not attempt to repair in degraded state */ 8277 if (btrfs_test_opt(fs_info, DEGRADED)) 8278 return 0; 8279 8280 cache = btrfs_lookup_block_group(fs_info, logical); 8281 if (!cache) 8282 return 0; 8283 8284 spin_lock(&cache->lock); 8285 if (cache->relocating_repair) { 8286 spin_unlock(&cache->lock); 8287 btrfs_put_block_group(cache); 8288 return 0; 8289 } 8290 cache->relocating_repair = 1; 8291 spin_unlock(&cache->lock); 8292 8293 kthread_run(relocating_repair_kthread, cache, 8294 "btrfs-relocating-repair"); 8295 8296 return 0; 8297 } 8298