1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #ifndef BTRFS_VOLUMES_H 7 #define BTRFS_VOLUMES_H 8 9 #include <linux/bio.h> 10 #include <linux/sort.h> 11 #include <linux/btrfs.h> 12 #include "async-thread.h" 13 14 #define BTRFS_MAX_DATA_CHUNK_SIZE (10ULL * SZ_1G) 15 16 extern struct mutex uuid_mutex; 17 18 #define BTRFS_STRIPE_LEN SZ_64K 19 20 struct btrfs_io_geometry { 21 /* remaining bytes before crossing a stripe */ 22 u64 len; 23 /* offset of logical address in chunk */ 24 u64 offset; 25 /* length of single IO stripe */ 26 u64 stripe_len; 27 /* number of stripe where address falls */ 28 u64 stripe_nr; 29 /* offset of address in stripe */ 30 u64 stripe_offset; 31 /* offset of raid56 stripe into the chunk */ 32 u64 raid56_stripe_offset; 33 }; 34 35 /* 36 * Use sequence counter to get consistent device stat data on 37 * 32-bit processors. 38 */ 39 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 40 #include <linux/seqlock.h> 41 #define __BTRFS_NEED_DEVICE_DATA_ORDERED 42 #define btrfs_device_data_ordered_init(device) \ 43 seqcount_init(&device->data_seqcount) 44 #else 45 #define btrfs_device_data_ordered_init(device) do { } while (0) 46 #endif 47 48 #define BTRFS_DEV_STATE_WRITEABLE (0) 49 #define BTRFS_DEV_STATE_IN_FS_METADATA (1) 50 #define BTRFS_DEV_STATE_MISSING (2) 51 #define BTRFS_DEV_STATE_REPLACE_TGT (3) 52 #define BTRFS_DEV_STATE_FLUSH_SENT (4) 53 #define BTRFS_DEV_STATE_NO_READA (5) 54 55 struct btrfs_zoned_device_info; 56 57 struct btrfs_device { 58 struct list_head dev_list; /* device_list_mutex */ 59 struct list_head dev_alloc_list; /* chunk mutex */ 60 struct list_head post_commit_list; /* chunk mutex */ 61 struct btrfs_fs_devices *fs_devices; 62 struct btrfs_fs_info *fs_info; 63 64 struct rcu_string __rcu *name; 65 66 u64 generation; 67 68 struct block_device *bdev; 69 70 struct btrfs_zoned_device_info *zone_info; 71 72 /* the mode sent to blkdev_get */ 73 fmode_t mode; 74 75 /* 76 * Device's major-minor number. Must be set even if the device is not 77 * opened (bdev == NULL), unless the device is missing. 78 */ 79 dev_t devt; 80 unsigned long dev_state; 81 blk_status_t last_flush_error; 82 83 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED 84 seqcount_t data_seqcount; 85 #endif 86 87 /* the internal btrfs device id */ 88 u64 devid; 89 90 /* size of the device in memory */ 91 u64 total_bytes; 92 93 /* size of the device on disk */ 94 u64 disk_total_bytes; 95 96 /* bytes used */ 97 u64 bytes_used; 98 99 /* optimal io alignment for this device */ 100 u32 io_align; 101 102 /* optimal io width for this device */ 103 u32 io_width; 104 /* type and info about this device */ 105 u64 type; 106 107 /* minimal io size for this device */ 108 u32 sector_size; 109 110 /* physical drive uuid (or lvm uuid) */ 111 u8 uuid[BTRFS_UUID_SIZE]; 112 113 /* 114 * size of the device on the current transaction 115 * 116 * This variant is update when committing the transaction, 117 * and protected by chunk mutex 118 */ 119 u64 commit_total_bytes; 120 121 /* bytes used on the current transaction */ 122 u64 commit_bytes_used; 123 124 /* for sending down flush barriers */ 125 struct bio *flush_bio; 126 struct completion flush_wait; 127 128 /* per-device scrub information */ 129 struct scrub_ctx *scrub_ctx; 130 131 /* disk I/O failure stats. For detailed description refer to 132 * enum btrfs_dev_stat_values in ioctl.h */ 133 int dev_stats_valid; 134 135 /* Counter to record the change of device stats */ 136 atomic_t dev_stats_ccnt; 137 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX]; 138 139 struct extent_io_tree alloc_state; 140 141 struct completion kobj_unregister; 142 /* For sysfs/FSID/devinfo/devid/ */ 143 struct kobject devid_kobj; 144 145 /* Bandwidth limit for scrub, in bytes */ 146 u64 scrub_speed_max; 147 }; 148 149 /* 150 * If we read those variants at the context of their own lock, we needn't 151 * use the following helpers, reading them directly is safe. 152 */ 153 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 154 #define BTRFS_DEVICE_GETSET_FUNCS(name) \ 155 static inline u64 \ 156 btrfs_device_get_##name(const struct btrfs_device *dev) \ 157 { \ 158 u64 size; \ 159 unsigned int seq; \ 160 \ 161 do { \ 162 seq = read_seqcount_begin(&dev->data_seqcount); \ 163 size = dev->name; \ 164 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \ 165 return size; \ 166 } \ 167 \ 168 static inline void \ 169 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \ 170 { \ 171 preempt_disable(); \ 172 write_seqcount_begin(&dev->data_seqcount); \ 173 dev->name = size; \ 174 write_seqcount_end(&dev->data_seqcount); \ 175 preempt_enable(); \ 176 } 177 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 178 #define BTRFS_DEVICE_GETSET_FUNCS(name) \ 179 static inline u64 \ 180 btrfs_device_get_##name(const struct btrfs_device *dev) \ 181 { \ 182 u64 size; \ 183 \ 184 preempt_disable(); \ 185 size = dev->name; \ 186 preempt_enable(); \ 187 return size; \ 188 } \ 189 \ 190 static inline void \ 191 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \ 192 { \ 193 preempt_disable(); \ 194 dev->name = size; \ 195 preempt_enable(); \ 196 } 197 #else 198 #define BTRFS_DEVICE_GETSET_FUNCS(name) \ 199 static inline u64 \ 200 btrfs_device_get_##name(const struct btrfs_device *dev) \ 201 { \ 202 return dev->name; \ 203 } \ 204 \ 205 static inline void \ 206 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \ 207 { \ 208 dev->name = size; \ 209 } 210 #endif 211 212 BTRFS_DEVICE_GETSET_FUNCS(total_bytes); 213 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes); 214 BTRFS_DEVICE_GETSET_FUNCS(bytes_used); 215 216 enum btrfs_chunk_allocation_policy { 217 BTRFS_CHUNK_ALLOC_REGULAR, 218 BTRFS_CHUNK_ALLOC_ZONED, 219 }; 220 221 /* 222 * Read policies for mirrored block group profiles, read picks the stripe based 223 * on these policies. 224 */ 225 enum btrfs_read_policy { 226 /* Use process PID to choose the stripe */ 227 BTRFS_READ_POLICY_PID, 228 BTRFS_NR_READ_POLICY, 229 }; 230 231 struct btrfs_fs_devices { 232 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */ 233 u8 metadata_uuid[BTRFS_FSID_SIZE]; 234 bool fsid_change; 235 struct list_head fs_list; 236 237 /* 238 * Number of devices under this fsid including missing and 239 * replace-target device and excludes seed devices. 240 */ 241 u64 num_devices; 242 243 /* 244 * The number of devices that successfully opened, including 245 * replace-target, excludes seed devices. 246 */ 247 u64 open_devices; 248 249 /* The number of devices that are under the chunk allocation list. */ 250 u64 rw_devices; 251 252 /* Count of missing devices under this fsid excluding seed device. */ 253 u64 missing_devices; 254 u64 total_rw_bytes; 255 256 /* 257 * Count of devices from btrfs_super_block::num_devices for this fsid, 258 * which includes the seed device, excludes the transient replace-target 259 * device. 260 */ 261 u64 total_devices; 262 263 /* Highest generation number of seen devices */ 264 u64 latest_generation; 265 266 /* 267 * The mount device or a device with highest generation after removal 268 * or replace. 269 */ 270 struct btrfs_device *latest_dev; 271 272 /* all of the devices in the FS, protected by a mutex 273 * so we can safely walk it to write out the supers without 274 * worrying about add/remove by the multi-device code. 275 * Scrubbing super can kick off supers writing by holding 276 * this mutex lock. 277 */ 278 struct mutex device_list_mutex; 279 280 /* List of all devices, protected by device_list_mutex */ 281 struct list_head devices; 282 283 /* 284 * Devices which can satisfy space allocation. Protected by 285 * chunk_mutex 286 */ 287 struct list_head alloc_list; 288 289 struct list_head seed_list; 290 bool seeding; 291 292 int opened; 293 294 /* set when we find or add a device that doesn't have the 295 * nonrot flag set 296 */ 297 bool rotating; 298 299 struct btrfs_fs_info *fs_info; 300 /* sysfs kobjects */ 301 struct kobject fsid_kobj; 302 struct kobject *devices_kobj; 303 struct kobject *devinfo_kobj; 304 struct completion kobj_unregister; 305 306 enum btrfs_chunk_allocation_policy chunk_alloc_policy; 307 308 /* Policy used to read the mirrored stripes */ 309 enum btrfs_read_policy read_policy; 310 }; 311 312 #define BTRFS_BIO_INLINE_CSUM_SIZE 64 313 314 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ 315 - sizeof(struct btrfs_chunk)) \ 316 / sizeof(struct btrfs_stripe) + 1) 317 318 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ 319 - 2 * sizeof(struct btrfs_disk_key) \ 320 - 2 * sizeof(struct btrfs_chunk)) \ 321 / sizeof(struct btrfs_stripe) + 1) 322 323 /* 324 * Additional info to pass along bio. 325 * 326 * Mostly for btrfs specific features like csum and mirror_num. 327 */ 328 struct btrfs_bio { 329 unsigned int mirror_num; 330 331 /* @device is for stripe IO submission. */ 332 struct btrfs_device *device; 333 u8 *csum; 334 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE]; 335 struct bvec_iter iter; 336 337 /* 338 * This member must come last, bio_alloc_bioset will allocate enough 339 * bytes for entire btrfs_bio but relies on bio being last. 340 */ 341 struct bio bio; 342 }; 343 344 static inline struct btrfs_bio *btrfs_bio(struct bio *bio) 345 { 346 return container_of(bio, struct btrfs_bio, bio); 347 } 348 349 static inline void btrfs_bio_free_csum(struct btrfs_bio *bbio) 350 { 351 if (bbio->csum != bbio->csum_inline) { 352 kfree(bbio->csum); 353 bbio->csum = NULL; 354 } 355 } 356 357 struct btrfs_io_stripe { 358 struct btrfs_device *dev; 359 u64 physical; 360 u64 length; /* only used for discard mappings */ 361 }; 362 363 /* 364 * Context for IO subsmission for device stripe. 365 * 366 * - Track the unfinished mirrors for mirror based profiles 367 * Mirror based profiles are SINGLE/DUP/RAID1/RAID10. 368 * 369 * - Contain the logical -> physical mapping info 370 * Used by submit_stripe_bio() for mapping logical bio 371 * into physical device address. 372 * 373 * - Contain device replace info 374 * Used by handle_ops_on_dev_replace() to copy logical bios 375 * into the new device. 376 * 377 * - Contain RAID56 full stripe logical bytenrs 378 */ 379 struct btrfs_io_context { 380 refcount_t refs; 381 atomic_t stripes_pending; 382 struct btrfs_fs_info *fs_info; 383 u64 map_type; /* get from map_lookup->type */ 384 bio_end_io_t *end_io; 385 struct bio *orig_bio; 386 void *private; 387 atomic_t error; 388 int max_errors; 389 int num_stripes; 390 int mirror_num; 391 int num_tgtdevs; 392 int *tgtdev_map; 393 /* 394 * logical block numbers for the start of each stripe 395 * The last one or two are p/q. These are sorted, 396 * so raid_map[0] is the start of our full stripe 397 */ 398 u64 *raid_map; 399 struct btrfs_io_stripe stripes[]; 400 }; 401 402 struct btrfs_device_info { 403 struct btrfs_device *dev; 404 u64 dev_offset; 405 u64 max_avail; 406 u64 total_avail; 407 }; 408 409 struct btrfs_raid_attr { 410 u8 sub_stripes; /* sub_stripes info for map */ 411 u8 dev_stripes; /* stripes per dev */ 412 u8 devs_max; /* max devs to use */ 413 u8 devs_min; /* min devs needed */ 414 u8 tolerated_failures; /* max tolerated fail devs */ 415 u8 devs_increment; /* ndevs has to be a multiple of this */ 416 u8 ncopies; /* how many copies to data has */ 417 u8 nparity; /* number of stripes worth of bytes to store 418 * parity information */ 419 u8 mindev_error; /* error code if min devs requisite is unmet */ 420 const char raid_name[8]; /* name of the raid */ 421 u64 bg_flag; /* block group flag of the raid */ 422 }; 423 424 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES]; 425 426 struct map_lookup { 427 u64 type; 428 int io_align; 429 int io_width; 430 u64 stripe_len; 431 int num_stripes; 432 int sub_stripes; 433 int verified_stripes; /* For mount time dev extent verification */ 434 struct btrfs_io_stripe stripes[]; 435 }; 436 437 #define map_lookup_size(n) (sizeof(struct map_lookup) + \ 438 (sizeof(struct btrfs_io_stripe) * (n))) 439 440 struct btrfs_balance_args; 441 struct btrfs_balance_progress; 442 struct btrfs_balance_control { 443 struct btrfs_balance_args data; 444 struct btrfs_balance_args meta; 445 struct btrfs_balance_args sys; 446 447 u64 flags; 448 449 struct btrfs_balance_progress stat; 450 }; 451 452 /* 453 * Search for a given device by the set parameters 454 */ 455 struct btrfs_dev_lookup_args { 456 u64 devid; 457 u8 *uuid; 458 u8 *fsid; 459 bool missing; 460 }; 461 462 /* We have to initialize to -1 because BTRFS_DEV_REPLACE_DEVID is 0 */ 463 #define BTRFS_DEV_LOOKUP_ARGS_INIT { .devid = (u64)-1 } 464 465 #define BTRFS_DEV_LOOKUP_ARGS(name) \ 466 struct btrfs_dev_lookup_args name = BTRFS_DEV_LOOKUP_ARGS_INIT 467 468 enum btrfs_map_op { 469 BTRFS_MAP_READ, 470 BTRFS_MAP_WRITE, 471 BTRFS_MAP_DISCARD, 472 BTRFS_MAP_GET_READ_MIRRORS, 473 }; 474 475 static inline enum btrfs_map_op btrfs_op(struct bio *bio) 476 { 477 switch (bio_op(bio)) { 478 case REQ_OP_DISCARD: 479 return BTRFS_MAP_DISCARD; 480 case REQ_OP_WRITE: 481 case REQ_OP_ZONE_APPEND: 482 return BTRFS_MAP_WRITE; 483 default: 484 WARN_ON_ONCE(1); 485 fallthrough; 486 case REQ_OP_READ: 487 return BTRFS_MAP_READ; 488 } 489 } 490 491 void btrfs_get_bioc(struct btrfs_io_context *bioc); 492 void btrfs_put_bioc(struct btrfs_io_context *bioc); 493 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 494 u64 logical, u64 *length, 495 struct btrfs_io_context **bioc_ret, int mirror_num); 496 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 497 u64 logical, u64 *length, 498 struct btrfs_io_context **bioc_ret); 499 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *map, 500 enum btrfs_map_op op, u64 logical, 501 struct btrfs_io_geometry *io_geom); 502 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info); 503 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info); 504 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, 505 u64 type); 506 void btrfs_mapping_tree_free(struct extent_map_tree *tree); 507 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 508 int mirror_num); 509 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 510 fmode_t flags, void *holder); 511 struct btrfs_device *btrfs_scan_one_device(const char *path, 512 fmode_t flags, void *holder); 513 int btrfs_forget_devices(dev_t devt); 514 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices); 515 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices); 516 void btrfs_assign_next_active_device(struct btrfs_device *device, 517 struct btrfs_device *this_dev); 518 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, 519 u64 devid, 520 const char *devpath); 521 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, 522 struct btrfs_dev_lookup_args *args, 523 const char *path); 524 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 525 const u64 *devid, 526 const u8 *uuid); 527 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args); 528 void btrfs_free_device(struct btrfs_device *device); 529 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 530 struct btrfs_dev_lookup_args *args, 531 struct block_device **bdev, fmode_t *mode); 532 void __exit btrfs_cleanup_fs_uuids(void); 533 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len); 534 int btrfs_grow_device(struct btrfs_trans_handle *trans, 535 struct btrfs_device *device, u64 new_size); 536 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, 537 const struct btrfs_dev_lookup_args *args); 538 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size); 539 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path); 540 int btrfs_balance(struct btrfs_fs_info *fs_info, 541 struct btrfs_balance_control *bctl, 542 struct btrfs_ioctl_balance_args *bargs); 543 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf); 544 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info); 545 int btrfs_recover_balance(struct btrfs_fs_info *fs_info); 546 int btrfs_pause_balance(struct btrfs_fs_info *fs_info); 547 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset); 548 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info); 549 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info); 550 int btrfs_uuid_scan_kthread(void *data); 551 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset); 552 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 553 u64 *start, u64 *max_avail); 554 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index); 555 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 556 struct btrfs_ioctl_get_dev_stats *stats); 557 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info); 558 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info); 559 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans); 560 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev); 561 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev); 562 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev); 563 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, 564 u64 logical, u64 len); 565 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 566 u64 logical); 567 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, 568 struct btrfs_block_group *bg); 569 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset); 570 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 571 u64 logical, u64 length); 572 void btrfs_release_disk_super(struct btrfs_super_block *super); 573 574 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev, 575 int index) 576 { 577 atomic_inc(dev->dev_stat_values + index); 578 /* 579 * This memory barrier orders stores updating statistics before stores 580 * updating dev_stats_ccnt. 581 * 582 * It pairs with smp_rmb() in btrfs_run_dev_stats(). 583 */ 584 smp_mb__before_atomic(); 585 atomic_inc(&dev->dev_stats_ccnt); 586 } 587 588 static inline int btrfs_dev_stat_read(struct btrfs_device *dev, 589 int index) 590 { 591 return atomic_read(dev->dev_stat_values + index); 592 } 593 594 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev, 595 int index) 596 { 597 int ret; 598 599 ret = atomic_xchg(dev->dev_stat_values + index, 0); 600 /* 601 * atomic_xchg implies a full memory barriers as per atomic_t.txt: 602 * - RMW operations that have a return value are fully ordered; 603 * 604 * This implicit memory barriers is paired with the smp_rmb in 605 * btrfs_run_dev_stats 606 */ 607 atomic_inc(&dev->dev_stats_ccnt); 608 return ret; 609 } 610 611 static inline void btrfs_dev_stat_set(struct btrfs_device *dev, 612 int index, unsigned long val) 613 { 614 atomic_set(dev->dev_stat_values + index, val); 615 /* 616 * This memory barrier orders stores updating statistics before stores 617 * updating dev_stats_ccnt. 618 * 619 * It pairs with smp_rmb() in btrfs_run_dev_stats(). 620 */ 621 smp_mb__before_atomic(); 622 atomic_inc(&dev->dev_stats_ccnt); 623 } 624 625 void btrfs_commit_device_sizes(struct btrfs_transaction *trans); 626 627 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void); 628 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 629 struct btrfs_device *failing_dev); 630 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, 631 struct block_device *bdev, 632 const char *device_path); 633 634 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags); 635 int btrfs_bg_type_to_factor(u64 flags); 636 const char *btrfs_bg_type_to_raid_name(u64 flags); 637 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info); 638 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical); 639 640 #endif 641