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