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 buffer_head; 21 22 struct btrfs_io_geometry { 23 /* remaining bytes before crossing a stripe */ 24 u64 len; 25 /* offset of logical address in chunk */ 26 u64 offset; 27 /* length of single IO stripe */ 28 u64 stripe_len; 29 /* number of stripe where address falls */ 30 u64 stripe_nr; 31 /* offset of address in stripe */ 32 u64 stripe_offset; 33 /* offset of raid56 stripe into the chunk */ 34 u64 raid56_stripe_offset; 35 }; 36 37 /* 38 * Use sequence counter to get consistent device stat data on 39 * 32-bit processors. 40 */ 41 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 42 #include <linux/seqlock.h> 43 #define __BTRFS_NEED_DEVICE_DATA_ORDERED 44 #define btrfs_device_data_ordered_init(device) \ 45 seqcount_init(&device->data_seqcount) 46 #else 47 #define btrfs_device_data_ordered_init(device) do { } while (0) 48 #endif 49 50 #define BTRFS_DEV_STATE_WRITEABLE (0) 51 #define BTRFS_DEV_STATE_IN_FS_METADATA (1) 52 #define BTRFS_DEV_STATE_MISSING (2) 53 #define BTRFS_DEV_STATE_REPLACE_TGT (3) 54 #define BTRFS_DEV_STATE_FLUSH_SENT (4) 55 56 struct btrfs_device { 57 struct list_head dev_list; /* device_list_mutex */ 58 struct list_head dev_alloc_list; /* chunk mutex */ 59 struct list_head post_commit_list; /* chunk mutex */ 60 struct btrfs_fs_devices *fs_devices; 61 struct btrfs_fs_info *fs_info; 62 63 struct rcu_string *name; 64 65 u64 generation; 66 67 struct block_device *bdev; 68 69 /* the mode sent to blkdev_get */ 70 fmode_t mode; 71 72 unsigned long dev_state; 73 blk_status_t last_flush_error; 74 75 #ifdef __BTRFS_NEED_DEVICE_DATA_ORDERED 76 seqcount_t data_seqcount; 77 #endif 78 79 /* the internal btrfs device id */ 80 u64 devid; 81 82 /* size of the device in memory */ 83 u64 total_bytes; 84 85 /* size of the device on disk */ 86 u64 disk_total_bytes; 87 88 /* bytes used */ 89 u64 bytes_used; 90 91 /* optimal io alignment for this device */ 92 u32 io_align; 93 94 /* optimal io width for this device */ 95 u32 io_width; 96 /* type and info about this device */ 97 u64 type; 98 99 /* minimal io size for this device */ 100 u32 sector_size; 101 102 /* physical drive uuid (or lvm uuid) */ 103 u8 uuid[BTRFS_UUID_SIZE]; 104 105 /* 106 * size of the device on the current transaction 107 * 108 * This variant is update when committing the transaction, 109 * and protected by chunk mutex 110 */ 111 u64 commit_total_bytes; 112 113 /* bytes used on the current transaction */ 114 u64 commit_bytes_used; 115 116 /* for sending down flush barriers */ 117 struct bio *flush_bio; 118 struct completion flush_wait; 119 120 /* per-device scrub information */ 121 struct scrub_ctx *scrub_ctx; 122 123 /* readahead state */ 124 atomic_t reada_in_flight; 125 u64 reada_next; 126 struct reada_zone *reada_curr_zone; 127 struct radix_tree_root reada_zones; 128 struct radix_tree_root reada_extents; 129 130 /* disk I/O failure stats. For detailed description refer to 131 * enum btrfs_dev_stat_values in ioctl.h */ 132 int dev_stats_valid; 133 134 /* Counter to record the change of device stats */ 135 atomic_t dev_stats_ccnt; 136 atomic_t dev_stat_values[BTRFS_DEV_STAT_VALUES_MAX]; 137 138 struct extent_io_tree alloc_state; 139 140 struct completion kobj_unregister; 141 /* For sysfs/FSID/devinfo/devid/ */ 142 struct kobject devid_kobj; 143 }; 144 145 /* 146 * If we read those variants at the context of their own lock, we needn't 147 * use the following helpers, reading them directly is safe. 148 */ 149 #if BITS_PER_LONG==32 && defined(CONFIG_SMP) 150 #define BTRFS_DEVICE_GETSET_FUNCS(name) \ 151 static inline u64 \ 152 btrfs_device_get_##name(const struct btrfs_device *dev) \ 153 { \ 154 u64 size; \ 155 unsigned int seq; \ 156 \ 157 do { \ 158 seq = read_seqcount_begin(&dev->data_seqcount); \ 159 size = dev->name; \ 160 } while (read_seqcount_retry(&dev->data_seqcount, seq)); \ 161 return size; \ 162 } \ 163 \ 164 static inline void \ 165 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \ 166 { \ 167 preempt_disable(); \ 168 write_seqcount_begin(&dev->data_seqcount); \ 169 dev->name = size; \ 170 write_seqcount_end(&dev->data_seqcount); \ 171 preempt_enable(); \ 172 } 173 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION) 174 #define BTRFS_DEVICE_GETSET_FUNCS(name) \ 175 static inline u64 \ 176 btrfs_device_get_##name(const struct btrfs_device *dev) \ 177 { \ 178 u64 size; \ 179 \ 180 preempt_disable(); \ 181 size = dev->name; \ 182 preempt_enable(); \ 183 return size; \ 184 } \ 185 \ 186 static inline void \ 187 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \ 188 { \ 189 preempt_disable(); \ 190 dev->name = size; \ 191 preempt_enable(); \ 192 } 193 #else 194 #define BTRFS_DEVICE_GETSET_FUNCS(name) \ 195 static inline u64 \ 196 btrfs_device_get_##name(const struct btrfs_device *dev) \ 197 { \ 198 return dev->name; \ 199 } \ 200 \ 201 static inline void \ 202 btrfs_device_set_##name(struct btrfs_device *dev, u64 size) \ 203 { \ 204 dev->name = size; \ 205 } 206 #endif 207 208 BTRFS_DEVICE_GETSET_FUNCS(total_bytes); 209 BTRFS_DEVICE_GETSET_FUNCS(disk_total_bytes); 210 BTRFS_DEVICE_GETSET_FUNCS(bytes_used); 211 212 struct btrfs_fs_devices { 213 u8 fsid[BTRFS_FSID_SIZE]; /* FS specific uuid */ 214 u8 metadata_uuid[BTRFS_FSID_SIZE]; 215 bool fsid_change; 216 struct list_head fs_list; 217 218 u64 num_devices; 219 u64 open_devices; 220 u64 rw_devices; 221 u64 missing_devices; 222 u64 total_rw_bytes; 223 u64 total_devices; 224 225 /* Highest generation number of seen devices */ 226 u64 latest_generation; 227 228 struct block_device *latest_bdev; 229 230 /* all of the devices in the FS, protected by a mutex 231 * so we can safely walk it to write out the supers without 232 * worrying about add/remove by the multi-device code. 233 * Scrubbing super can kick off supers writing by holding 234 * this mutex lock. 235 */ 236 struct mutex device_list_mutex; 237 238 /* List of all devices, protected by device_list_mutex */ 239 struct list_head devices; 240 241 /* 242 * Devices which can satisfy space allocation. Protected by 243 * chunk_mutex 244 */ 245 struct list_head alloc_list; 246 247 struct btrfs_fs_devices *seed; 248 bool seeding; 249 250 int opened; 251 252 /* set when we find or add a device that doesn't have the 253 * nonrot flag set 254 */ 255 bool rotating; 256 257 struct btrfs_fs_info *fs_info; 258 /* sysfs kobjects */ 259 struct kobject fsid_kobj; 260 struct kobject *devices_kobj; 261 struct kobject *devinfo_kobj; 262 struct completion kobj_unregister; 263 }; 264 265 #define BTRFS_BIO_INLINE_CSUM_SIZE 64 266 267 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ 268 - sizeof(struct btrfs_chunk)) \ 269 / sizeof(struct btrfs_stripe) + 1) 270 271 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ 272 - 2 * sizeof(struct btrfs_disk_key) \ 273 - 2 * sizeof(struct btrfs_chunk)) \ 274 / sizeof(struct btrfs_stripe) + 1) 275 276 /* 277 * we need the mirror number and stripe index to be passed around 278 * the call chain while we are processing end_io (especially errors). 279 * Really, what we need is a btrfs_bio structure that has this info 280 * and is properly sized with its stripe array, but we're not there 281 * quite yet. We have our own btrfs bioset, and all of the bios 282 * we allocate are actually btrfs_io_bios. We'll cram as much of 283 * struct btrfs_bio as we can into this over time. 284 */ 285 struct btrfs_io_bio { 286 unsigned int mirror_num; 287 unsigned int stripe_index; 288 u64 logical; 289 u8 *csum; 290 u8 csum_inline[BTRFS_BIO_INLINE_CSUM_SIZE]; 291 struct bvec_iter iter; 292 /* 293 * This member must come last, bio_alloc_bioset will allocate enough 294 * bytes for entire btrfs_io_bio but relies on bio being last. 295 */ 296 struct bio bio; 297 }; 298 299 static inline struct btrfs_io_bio *btrfs_io_bio(struct bio *bio) 300 { 301 return container_of(bio, struct btrfs_io_bio, bio); 302 } 303 304 static inline void btrfs_io_bio_free_csum(struct btrfs_io_bio *io_bio) 305 { 306 if (io_bio->csum != io_bio->csum_inline) { 307 kfree(io_bio->csum); 308 io_bio->csum = NULL; 309 } 310 } 311 312 struct btrfs_bio_stripe { 313 struct btrfs_device *dev; 314 u64 physical; 315 u64 length; /* only used for discard mappings */ 316 }; 317 318 struct btrfs_bio { 319 refcount_t refs; 320 atomic_t stripes_pending; 321 struct btrfs_fs_info *fs_info; 322 u64 map_type; /* get from map_lookup->type */ 323 bio_end_io_t *end_io; 324 struct bio *orig_bio; 325 void *private; 326 atomic_t error; 327 int max_errors; 328 int num_stripes; 329 int mirror_num; 330 int num_tgtdevs; 331 int *tgtdev_map; 332 /* 333 * logical block numbers for the start of each stripe 334 * The last one or two are p/q. These are sorted, 335 * so raid_map[0] is the start of our full stripe 336 */ 337 u64 *raid_map; 338 struct btrfs_bio_stripe stripes[]; 339 }; 340 341 struct btrfs_device_info { 342 struct btrfs_device *dev; 343 u64 dev_offset; 344 u64 max_avail; 345 u64 total_avail; 346 }; 347 348 struct btrfs_raid_attr { 349 u8 sub_stripes; /* sub_stripes info for map */ 350 u8 dev_stripes; /* stripes per dev */ 351 u8 devs_max; /* max devs to use */ 352 u8 devs_min; /* min devs needed */ 353 u8 tolerated_failures; /* max tolerated fail devs */ 354 u8 devs_increment; /* ndevs has to be a multiple of this */ 355 u8 ncopies; /* how many copies to data has */ 356 u8 nparity; /* number of stripes worth of bytes to store 357 * parity information */ 358 u8 mindev_error; /* error code if min devs requisite is unmet */ 359 const char raid_name[8]; /* name of the raid */ 360 u64 bg_flag; /* block group flag of the raid */ 361 }; 362 363 extern const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES]; 364 365 struct map_lookup { 366 u64 type; 367 int io_align; 368 int io_width; 369 u64 stripe_len; 370 int num_stripes; 371 int sub_stripes; 372 int verified_stripes; /* For mount time dev extent verification */ 373 struct btrfs_bio_stripe stripes[]; 374 }; 375 376 #define map_lookup_size(n) (sizeof(struct map_lookup) + \ 377 (sizeof(struct btrfs_bio_stripe) * (n))) 378 379 struct btrfs_balance_args; 380 struct btrfs_balance_progress; 381 struct btrfs_balance_control { 382 struct btrfs_balance_args data; 383 struct btrfs_balance_args meta; 384 struct btrfs_balance_args sys; 385 386 u64 flags; 387 388 struct btrfs_balance_progress stat; 389 }; 390 391 enum btrfs_map_op { 392 BTRFS_MAP_READ, 393 BTRFS_MAP_WRITE, 394 BTRFS_MAP_DISCARD, 395 BTRFS_MAP_GET_READ_MIRRORS, 396 }; 397 398 static inline enum btrfs_map_op btrfs_op(struct bio *bio) 399 { 400 switch (bio_op(bio)) { 401 case REQ_OP_DISCARD: 402 return BTRFS_MAP_DISCARD; 403 case REQ_OP_WRITE: 404 return BTRFS_MAP_WRITE; 405 default: 406 WARN_ON_ONCE(1); 407 /* fall through */ 408 case REQ_OP_READ: 409 return BTRFS_MAP_READ; 410 } 411 } 412 413 void btrfs_get_bbio(struct btrfs_bio *bbio); 414 void btrfs_put_bbio(struct btrfs_bio *bbio); 415 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 416 u64 logical, u64 *length, 417 struct btrfs_bio **bbio_ret, int mirror_num); 418 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 419 u64 logical, u64 *length, 420 struct btrfs_bio **bbio_ret); 421 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, 422 u64 logical, u64 len, struct btrfs_io_geometry *io_geom); 423 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info); 424 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info); 425 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type); 426 void btrfs_mapping_tree_free(struct extent_map_tree *tree); 427 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, 428 int mirror_num); 429 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, 430 fmode_t flags, void *holder); 431 struct btrfs_device *btrfs_scan_one_device(const char *path, 432 fmode_t flags, void *holder); 433 int btrfs_forget_devices(const char *path); 434 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices); 435 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step); 436 void btrfs_assign_next_active_device(struct btrfs_device *device, 437 struct btrfs_device *this_dev); 438 struct btrfs_device *btrfs_find_device_by_devspec(struct btrfs_fs_info *fs_info, 439 u64 devid, 440 const char *devpath); 441 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, 442 const u64 *devid, 443 const u8 *uuid); 444 void btrfs_free_device(struct btrfs_device *device); 445 int btrfs_rm_device(struct btrfs_fs_info *fs_info, 446 const char *device_path, u64 devid); 447 void __exit btrfs_cleanup_fs_uuids(void); 448 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len); 449 int btrfs_grow_device(struct btrfs_trans_handle *trans, 450 struct btrfs_device *device, u64 new_size); 451 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, 452 u64 devid, u8 *uuid, u8 *fsid, bool seed); 453 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size); 454 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *path); 455 int btrfs_balance(struct btrfs_fs_info *fs_info, 456 struct btrfs_balance_control *bctl, 457 struct btrfs_ioctl_balance_args *bargs); 458 void btrfs_describe_block_groups(u64 flags, char *buf, u32 size_buf); 459 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info); 460 int btrfs_recover_balance(struct btrfs_fs_info *fs_info); 461 int btrfs_pause_balance(struct btrfs_fs_info *fs_info); 462 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info); 463 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info); 464 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info); 465 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset); 466 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, 467 u64 *start, u64 *max_avail); 468 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index); 469 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, 470 struct btrfs_ioctl_get_dev_stats *stats); 471 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info); 472 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info); 473 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans); 474 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev); 475 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev); 476 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev); 477 void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path); 478 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, 479 u64 logical, u64 len); 480 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, 481 u64 logical); 482 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, 483 u64 chunk_offset, u64 chunk_size); 484 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset); 485 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, 486 u64 logical, u64 length); 487 488 static inline void btrfs_dev_stat_inc(struct btrfs_device *dev, 489 int index) 490 { 491 atomic_inc(dev->dev_stat_values + index); 492 /* 493 * This memory barrier orders stores updating statistics before stores 494 * updating dev_stats_ccnt. 495 * 496 * It pairs with smp_rmb() in btrfs_run_dev_stats(). 497 */ 498 smp_mb__before_atomic(); 499 atomic_inc(&dev->dev_stats_ccnt); 500 } 501 502 static inline int btrfs_dev_stat_read(struct btrfs_device *dev, 503 int index) 504 { 505 return atomic_read(dev->dev_stat_values + index); 506 } 507 508 static inline int btrfs_dev_stat_read_and_reset(struct btrfs_device *dev, 509 int index) 510 { 511 int ret; 512 513 ret = atomic_xchg(dev->dev_stat_values + index, 0); 514 /* 515 * atomic_xchg implies a full memory barriers as per atomic_t.txt: 516 * - RMW operations that have a return value are fully ordered; 517 * 518 * This implicit memory barriers is paired with the smp_rmb in 519 * btrfs_run_dev_stats 520 */ 521 atomic_inc(&dev->dev_stats_ccnt); 522 return ret; 523 } 524 525 static inline void btrfs_dev_stat_set(struct btrfs_device *dev, 526 int index, unsigned long val) 527 { 528 atomic_set(dev->dev_stat_values + index, val); 529 /* 530 * This memory barrier orders stores updating statistics before stores 531 * updating dev_stats_ccnt. 532 * 533 * It pairs with smp_rmb() in btrfs_run_dev_stats(). 534 */ 535 smp_mb__before_atomic(); 536 atomic_inc(&dev->dev_stats_ccnt); 537 } 538 539 /* 540 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which 541 * can be used as index to access btrfs_raid_array[]. 542 */ 543 static inline enum btrfs_raid_types btrfs_bg_flags_to_raid_index(u64 flags) 544 { 545 if (flags & BTRFS_BLOCK_GROUP_RAID10) 546 return BTRFS_RAID_RAID10; 547 else if (flags & BTRFS_BLOCK_GROUP_RAID1) 548 return BTRFS_RAID_RAID1; 549 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3) 550 return BTRFS_RAID_RAID1C3; 551 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4) 552 return BTRFS_RAID_RAID1C4; 553 else if (flags & BTRFS_BLOCK_GROUP_DUP) 554 return BTRFS_RAID_DUP; 555 else if (flags & BTRFS_BLOCK_GROUP_RAID0) 556 return BTRFS_RAID_RAID0; 557 else if (flags & BTRFS_BLOCK_GROUP_RAID5) 558 return BTRFS_RAID_RAID5; 559 else if (flags & BTRFS_BLOCK_GROUP_RAID6) 560 return BTRFS_RAID_RAID6; 561 562 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ 563 } 564 565 void btrfs_commit_device_sizes(struct btrfs_transaction *trans); 566 567 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void); 568 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info); 569 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info); 570 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, 571 struct btrfs_device *failing_dev); 572 573 int btrfs_bg_type_to_factor(u64 flags); 574 const char *btrfs_bg_type_to_raid_name(u64 flags); 575 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info); 576 577 #endif 578