1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BCACHEFS_H 3 #define _BCACHEFS_H 4 5 /* 6 * SOME HIGH LEVEL CODE DOCUMENTATION: 7 * 8 * Bcache mostly works with cache sets, cache devices, and backing devices. 9 * 10 * Support for multiple cache devices hasn't quite been finished off yet, but 11 * it's about 95% plumbed through. A cache set and its cache devices is sort of 12 * like a md raid array and its component devices. Most of the code doesn't care 13 * about individual cache devices, the main abstraction is the cache set. 14 * 15 * Multiple cache devices is intended to give us the ability to mirror dirty 16 * cached data and metadata, without mirroring clean cached data. 17 * 18 * Backing devices are different, in that they have a lifetime independent of a 19 * cache set. When you register a newly formatted backing device it'll come up 20 * in passthrough mode, and then you can attach and detach a backing device from 21 * a cache set at runtime - while it's mounted and in use. Detaching implicitly 22 * invalidates any cached data for that backing device. 23 * 24 * A cache set can have multiple (many) backing devices attached to it. 25 * 26 * There's also flash only volumes - this is the reason for the distinction 27 * between struct cached_dev and struct bcache_device. A flash only volume 28 * works much like a bcache device that has a backing device, except the 29 * "cached" data is always dirty. The end result is that we get thin 30 * provisioning with very little additional code. 31 * 32 * Flash only volumes work but they're not production ready because the moving 33 * garbage collector needs more work. More on that later. 34 * 35 * BUCKETS/ALLOCATION: 36 * 37 * Bcache is primarily designed for caching, which means that in normal 38 * operation all of our available space will be allocated. Thus, we need an 39 * efficient way of deleting things from the cache so we can write new things to 40 * it. 41 * 42 * To do this, we first divide the cache device up into buckets. A bucket is the 43 * unit of allocation; they're typically around 1 mb - anywhere from 128k to 2M+ 44 * works efficiently. 45 * 46 * Each bucket has a 16 bit priority, and an 8 bit generation associated with 47 * it. The gens and priorities for all the buckets are stored contiguously and 48 * packed on disk (in a linked list of buckets - aside from the superblock, all 49 * of bcache's metadata is stored in buckets). 50 * 51 * The priority is used to implement an LRU. We reset a bucket's priority when 52 * we allocate it or on cache it, and every so often we decrement the priority 53 * of each bucket. It could be used to implement something more sophisticated, 54 * if anyone ever gets around to it. 55 * 56 * The generation is used for invalidating buckets. Each pointer also has an 8 57 * bit generation embedded in it; for a pointer to be considered valid, its gen 58 * must match the gen of the bucket it points into. Thus, to reuse a bucket all 59 * we have to do is increment its gen (and write its new gen to disk; we batch 60 * this up). 61 * 62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that 63 * contain metadata (including btree nodes). 64 * 65 * THE BTREE: 66 * 67 * Bcache is in large part design around the btree. 68 * 69 * At a high level, the btree is just an index of key -> ptr tuples. 70 * 71 * Keys represent extents, and thus have a size field. Keys also have a variable 72 * number of pointers attached to them (potentially zero, which is handy for 73 * invalidating the cache). 74 * 75 * The key itself is an inode:offset pair. The inode number corresponds to a 76 * backing device or a flash only volume. The offset is the ending offset of the 77 * extent within the inode - not the starting offset; this makes lookups 78 * slightly more convenient. 79 * 80 * Pointers contain the cache device id, the offset on that device, and an 8 bit 81 * generation number. More on the gen later. 82 * 83 * Index lookups are not fully abstracted - cache lookups in particular are 84 * still somewhat mixed in with the btree code, but things are headed in that 85 * direction. 86 * 87 * Updates are fairly well abstracted, though. There are two different ways of 88 * updating the btree; insert and replace. 89 * 90 * BTREE_INSERT will just take a list of keys and insert them into the btree - 91 * overwriting (possibly only partially) any extents they overlap with. This is 92 * used to update the index after a write. 93 * 94 * BTREE_REPLACE is really cmpxchg(); it inserts a key into the btree iff it is 95 * overwriting a key that matches another given key. This is used for inserting 96 * data into the cache after a cache miss, and for background writeback, and for 97 * the moving garbage collector. 98 * 99 * There is no "delete" operation; deleting things from the index is 100 * accomplished by either by invalidating pointers (by incrementing a bucket's 101 * gen) or by inserting a key with 0 pointers - which will overwrite anything 102 * previously present at that location in the index. 103 * 104 * This means that there are always stale/invalid keys in the btree. They're 105 * filtered out by the code that iterates through a btree node, and removed when 106 * a btree node is rewritten. 107 * 108 * BTREE NODES: 109 * 110 * Our unit of allocation is a bucket, and we can't arbitrarily allocate and 111 * free smaller than a bucket - so, that's how big our btree nodes are. 112 * 113 * (If buckets are really big we'll only use part of the bucket for a btree node 114 * - no less than 1/4th - but a bucket still contains no more than a single 115 * btree node. I'd actually like to change this, but for now we rely on the 116 * bucket's gen for deleting btree nodes when we rewrite/split a node.) 117 * 118 * Anyways, btree nodes are big - big enough to be inefficient with a textbook 119 * btree implementation. 120 * 121 * The way this is solved is that btree nodes are internally log structured; we 122 * can append new keys to an existing btree node without rewriting it. This 123 * means each set of keys we write is sorted, but the node is not. 124 * 125 * We maintain this log structure in memory - keeping 1Mb of keys sorted would 126 * be expensive, and we have to distinguish between the keys we have written and 127 * the keys we haven't. So to do a lookup in a btree node, we have to search 128 * each sorted set. But we do merge written sets together lazily, so the cost of 129 * these extra searches is quite low (normally most of the keys in a btree node 130 * will be in one big set, and then there'll be one or two sets that are much 131 * smaller). 132 * 133 * This log structure makes bcache's btree more of a hybrid between a 134 * conventional btree and a compacting data structure, with some of the 135 * advantages of both. 136 * 137 * GARBAGE COLLECTION: 138 * 139 * We can't just invalidate any bucket - it might contain dirty data or 140 * metadata. If it once contained dirty data, other writes might overwrite it 141 * later, leaving no valid pointers into that bucket in the index. 142 * 143 * Thus, the primary purpose of garbage collection is to find buckets to reuse. 144 * It also counts how much valid data it each bucket currently contains, so that 145 * allocation can reuse buckets sooner when they've been mostly overwritten. 146 * 147 * It also does some things that are really internal to the btree 148 * implementation. If a btree node contains pointers that are stale by more than 149 * some threshold, it rewrites the btree node to avoid the bucket's generation 150 * wrapping around. It also merges adjacent btree nodes if they're empty enough. 151 * 152 * THE JOURNAL: 153 * 154 * Bcache's journal is not necessary for consistency; we always strictly 155 * order metadata writes so that the btree and everything else is consistent on 156 * disk in the event of an unclean shutdown, and in fact bcache had writeback 157 * caching (with recovery from unclean shutdown) before journalling was 158 * implemented. 159 * 160 * Rather, the journal is purely a performance optimization; we can't complete a 161 * write until we've updated the index on disk, otherwise the cache would be 162 * inconsistent in the event of an unclean shutdown. This means that without the 163 * journal, on random write workloads we constantly have to update all the leaf 164 * nodes in the btree, and those writes will be mostly empty (appending at most 165 * a few keys each) - highly inefficient in terms of amount of metadata writes, 166 * and it puts more strain on the various btree resorting/compacting code. 167 * 168 * The journal is just a log of keys we've inserted; on startup we just reinsert 169 * all the keys in the open journal entries. That means that when we're updating 170 * a node in the btree, we can wait until a 4k block of keys fills up before 171 * writing them out. 172 * 173 * For simplicity, we only journal updates to leaf nodes; updates to parent 174 * nodes are rare enough (since our leaf nodes are huge) that it wasn't worth 175 * the complexity to deal with journalling them (in particular, journal replay) 176 * - updates to non leaf nodes just happen synchronously (see btree_split()). 177 */ 178 179 #undef pr_fmt 180 #ifdef __KERNEL__ 181 #define pr_fmt(fmt) "bcachefs: %s() " fmt "\n", __func__ 182 #else 183 #define pr_fmt(fmt) "%s() " fmt "\n", __func__ 184 #endif 185 186 #include <linux/backing-dev-defs.h> 187 #include <linux/bug.h> 188 #include <linux/bio.h> 189 #include <linux/closure.h> 190 #include <linux/kobject.h> 191 #include <linux/list.h> 192 #include <linux/math64.h> 193 #include <linux/mutex.h> 194 #include <linux/percpu-refcount.h> 195 #include <linux/percpu-rwsem.h> 196 #include <linux/refcount.h> 197 #include <linux/rhashtable.h> 198 #include <linux/rwsem.h> 199 #include <linux/semaphore.h> 200 #include <linux/seqlock.h> 201 #include <linux/shrinker.h> 202 #include <linux/srcu.h> 203 #include <linux/types.h> 204 #include <linux/workqueue.h> 205 #include <linux/zstd.h> 206 207 #include "bcachefs_format.h" 208 #include "disk_accounting_types.h" 209 #include "errcode.h" 210 #include "fifo.h" 211 #include "nocow_locking_types.h" 212 #include "opts.h" 213 #include "recovery_passes_types.h" 214 #include "sb-errors_types.h" 215 #include "seqmutex.h" 216 #include "time_stats.h" 217 #include "util.h" 218 219 #ifdef CONFIG_BCACHEFS_DEBUG 220 #define BCH_WRITE_REF_DEBUG 221 #endif 222 223 #ifndef dynamic_fault 224 #define dynamic_fault(...) 0 225 #endif 226 227 #define race_fault(...) dynamic_fault("bcachefs:race") 228 229 #define count_event(_c, _name) this_cpu_inc((_c)->counters[BCH_COUNTER_##_name]) 230 231 #define trace_and_count(_c, _name, ...) \ 232 do { \ 233 count_event(_c, _name); \ 234 trace_##_name(__VA_ARGS__); \ 235 } while (0) 236 237 #define bch2_fs_init_fault(name) \ 238 dynamic_fault("bcachefs:bch_fs_init:" name) 239 #define bch2_meta_read_fault(name) \ 240 dynamic_fault("bcachefs:meta:read:" name) 241 #define bch2_meta_write_fault(name) \ 242 dynamic_fault("bcachefs:meta:write:" name) 243 244 #ifdef __KERNEL__ 245 #define BCACHEFS_LOG_PREFIX 246 #endif 247 248 #ifdef BCACHEFS_LOG_PREFIX 249 250 #define bch2_log_msg(_c, fmt) "bcachefs (%s): " fmt, ((_c)->name) 251 #define bch2_fmt_dev(_ca, fmt) "bcachefs (%s): " fmt "\n", ((_ca)->name) 252 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "bcachefs (%s sector %llu): " fmt "\n", ((_ca)->name), (_offset) 253 #define bch2_fmt_inum(_c, _inum, fmt) "bcachefs (%s inum %llu): " fmt "\n", ((_c)->name), (_inum) 254 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \ 255 "bcachefs (%s inum %llu offset %llu): " fmt "\n", ((_c)->name), (_inum), (_offset) 256 257 #else 258 259 #define bch2_log_msg(_c, fmt) fmt 260 #define bch2_fmt_dev(_ca, fmt) "%s: " fmt "\n", ((_ca)->name) 261 #define bch2_fmt_dev_offset(_ca, _offset, fmt) "%s sector %llu: " fmt "\n", ((_ca)->name), (_offset) 262 #define bch2_fmt_inum(_c, _inum, fmt) "inum %llu: " fmt "\n", (_inum) 263 #define bch2_fmt_inum_offset(_c, _inum, _offset, fmt) \ 264 "inum %llu offset %llu: " fmt "\n", (_inum), (_offset) 265 266 #endif 267 268 #define bch2_fmt(_c, fmt) bch2_log_msg(_c, fmt "\n") 269 270 void bch2_print_str(struct bch_fs *, const char *); 271 272 __printf(2, 3) 273 void bch2_print_opts(struct bch_opts *, const char *, ...); 274 275 __printf(2, 3) 276 void __bch2_print(struct bch_fs *c, const char *fmt, ...); 277 278 #define maybe_dev_to_fs(_c) _Generic((_c), \ 279 struct bch_dev *: ((struct bch_dev *) (_c))->fs, \ 280 struct bch_fs *: (_c)) 281 282 #define bch2_print(_c, ...) __bch2_print(maybe_dev_to_fs(_c), __VA_ARGS__) 283 284 #define bch2_print_ratelimited(_c, ...) \ 285 do { \ 286 static DEFINE_RATELIMIT_STATE(_rs, \ 287 DEFAULT_RATELIMIT_INTERVAL, \ 288 DEFAULT_RATELIMIT_BURST); \ 289 \ 290 if (__ratelimit(&_rs)) \ 291 bch2_print(_c, __VA_ARGS__); \ 292 } while (0) 293 294 #define bch_info(c, fmt, ...) \ 295 bch2_print(c, KERN_INFO bch2_fmt(c, fmt), ##__VA_ARGS__) 296 #define bch_notice(c, fmt, ...) \ 297 bch2_print(c, KERN_NOTICE bch2_fmt(c, fmt), ##__VA_ARGS__) 298 #define bch_warn(c, fmt, ...) \ 299 bch2_print(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__) 300 #define bch_warn_ratelimited(c, fmt, ...) \ 301 bch2_print_ratelimited(c, KERN_WARNING bch2_fmt(c, fmt), ##__VA_ARGS__) 302 303 #define bch_err(c, fmt, ...) \ 304 bch2_print(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__) 305 #define bch_err_dev(ca, fmt, ...) \ 306 bch2_print(c, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__) 307 #define bch_err_dev_offset(ca, _offset, fmt, ...) \ 308 bch2_print(c, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__) 309 #define bch_err_inum(c, _inum, fmt, ...) \ 310 bch2_print(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__) 311 #define bch_err_inum_offset(c, _inum, _offset, fmt, ...) \ 312 bch2_print(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__) 313 314 #define bch_err_ratelimited(c, fmt, ...) \ 315 bch2_print_ratelimited(c, KERN_ERR bch2_fmt(c, fmt), ##__VA_ARGS__) 316 #define bch_err_dev_ratelimited(ca, fmt, ...) \ 317 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev(ca, fmt), ##__VA_ARGS__) 318 #define bch_err_dev_offset_ratelimited(ca, _offset, fmt, ...) \ 319 bch2_print_ratelimited(ca, KERN_ERR bch2_fmt_dev_offset(ca, _offset, fmt), ##__VA_ARGS__) 320 #define bch_err_inum_ratelimited(c, _inum, fmt, ...) \ 321 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum(c, _inum, fmt), ##__VA_ARGS__) 322 #define bch_err_inum_offset_ratelimited(c, _inum, _offset, fmt, ...) \ 323 bch2_print_ratelimited(c, KERN_ERR bch2_fmt_inum_offset(c, _inum, _offset, fmt), ##__VA_ARGS__) 324 325 static inline bool should_print_err(int err) 326 { 327 return err && !bch2_err_matches(err, BCH_ERR_transaction_restart); 328 } 329 330 #define bch_err_fn(_c, _ret) \ 331 do { \ 332 if (should_print_err(_ret)) \ 333 bch_err(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\ 334 } while (0) 335 336 #define bch_err_fn_ratelimited(_c, _ret) \ 337 do { \ 338 if (should_print_err(_ret)) \ 339 bch_err_ratelimited(_c, "%s(): error %s", __func__, bch2_err_str(_ret));\ 340 } while (0) 341 342 #define bch_err_msg(_c, _ret, _msg, ...) \ 343 do { \ 344 if (should_print_err(_ret)) \ 345 bch_err(_c, "%s(): error " _msg " %s", __func__, \ 346 ##__VA_ARGS__, bch2_err_str(_ret)); \ 347 } while (0) 348 349 #define bch_verbose(c, fmt, ...) \ 350 do { \ 351 if ((c)->opts.verbose) \ 352 bch_info(c, fmt, ##__VA_ARGS__); \ 353 } while (0) 354 355 #define pr_verbose_init(opts, fmt, ...) \ 356 do { \ 357 if (opt_get(opts, verbose)) \ 358 pr_info(fmt, ##__VA_ARGS__); \ 359 } while (0) 360 361 /* Parameters that are useful for debugging, but should always be compiled in: */ 362 #define BCH_DEBUG_PARAMS_ALWAYS() \ 363 BCH_DEBUG_PARAM(key_merging_disabled, \ 364 "Disables merging of extents") \ 365 BCH_DEBUG_PARAM(btree_node_merging_disabled, \ 366 "Disables merging of btree nodes") \ 367 BCH_DEBUG_PARAM(btree_gc_always_rewrite, \ 368 "Causes mark and sweep to compact and rewrite every " \ 369 "btree node it traverses") \ 370 BCH_DEBUG_PARAM(btree_gc_rewrite_disabled, \ 371 "Disables rewriting of btree nodes during mark and sweep")\ 372 BCH_DEBUG_PARAM(btree_shrinker_disabled, \ 373 "Disables the shrinker callback for the btree node cache")\ 374 BCH_DEBUG_PARAM(verify_btree_ondisk, \ 375 "Reread btree nodes at various points to verify the " \ 376 "mergesort in the read path against modifications " \ 377 "done in memory") \ 378 BCH_DEBUG_PARAM(verify_all_btree_replicas, \ 379 "When reading btree nodes, read all replicas and " \ 380 "compare them") \ 381 BCH_DEBUG_PARAM(backpointers_no_use_write_buffer, \ 382 "Don't use the write buffer for backpointers, enabling "\ 383 "extra runtime checks") 384 385 /* Parameters that should only be compiled in debug mode: */ 386 #define BCH_DEBUG_PARAMS_DEBUG() \ 387 BCH_DEBUG_PARAM(expensive_debug_checks, \ 388 "Enables various runtime debugging checks that " \ 389 "significantly affect performance") \ 390 BCH_DEBUG_PARAM(debug_check_iterators, \ 391 "Enables extra verification for btree iterators") \ 392 BCH_DEBUG_PARAM(debug_check_btree_accounting, \ 393 "Verify btree accounting for keys within a node") \ 394 BCH_DEBUG_PARAM(journal_seq_verify, \ 395 "Store the journal sequence number in the version " \ 396 "number of every btree key, and verify that btree " \ 397 "update ordering is preserved during recovery") \ 398 BCH_DEBUG_PARAM(inject_invalid_keys, \ 399 "Store the journal sequence number in the version " \ 400 "number of every btree key, and verify that btree " \ 401 "update ordering is preserved during recovery") \ 402 BCH_DEBUG_PARAM(test_alloc_startup, \ 403 "Force allocator startup to use the slowpath where it" \ 404 "can't find enough free buckets without invalidating" \ 405 "cached data") \ 406 BCH_DEBUG_PARAM(force_reconstruct_read, \ 407 "Force reads to use the reconstruct path, when reading" \ 408 "from erasure coded extents") \ 409 BCH_DEBUG_PARAM(test_restart_gc, \ 410 "Test restarting mark and sweep gc when bucket gens change") 411 412 #define BCH_DEBUG_PARAMS_ALL() BCH_DEBUG_PARAMS_ALWAYS() BCH_DEBUG_PARAMS_DEBUG() 413 414 #ifdef CONFIG_BCACHEFS_DEBUG 415 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL() 416 #else 417 #define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS() 418 #endif 419 420 #define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name; 421 BCH_DEBUG_PARAMS() 422 #undef BCH_DEBUG_PARAM 423 424 #ifndef CONFIG_BCACHEFS_DEBUG 425 #define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name; 426 BCH_DEBUG_PARAMS_DEBUG() 427 #undef BCH_DEBUG_PARAM 428 #endif 429 430 #define BCH_TIME_STATS() \ 431 x(btree_node_mem_alloc) \ 432 x(btree_node_split) \ 433 x(btree_node_compact) \ 434 x(btree_node_merge) \ 435 x(btree_node_sort) \ 436 x(btree_node_read) \ 437 x(btree_node_read_done) \ 438 x(btree_interior_update_foreground) \ 439 x(btree_interior_update_total) \ 440 x(btree_gc) \ 441 x(data_write) \ 442 x(data_read) \ 443 x(data_promote) \ 444 x(journal_flush_write) \ 445 x(journal_noflush_write) \ 446 x(journal_flush_seq) \ 447 x(blocked_journal_low_on_space) \ 448 x(blocked_journal_low_on_pin) \ 449 x(blocked_journal_max_in_flight) \ 450 x(blocked_key_cache_flush) \ 451 x(blocked_allocate) \ 452 x(blocked_allocate_open_bucket) \ 453 x(blocked_write_buffer_full) \ 454 x(nocow_lock_contended) 455 456 enum bch_time_stats { 457 #define x(name) BCH_TIME_##name, 458 BCH_TIME_STATS() 459 #undef x 460 BCH_TIME_STAT_NR 461 }; 462 463 #include "alloc_types.h" 464 #include "btree_gc_types.h" 465 #include "btree_types.h" 466 #include "btree_node_scan_types.h" 467 #include "btree_write_buffer_types.h" 468 #include "buckets_types.h" 469 #include "buckets_waiting_for_journal_types.h" 470 #include "clock_types.h" 471 #include "disk_groups_types.h" 472 #include "ec_types.h" 473 #include "journal_types.h" 474 #include "keylist_types.h" 475 #include "quota_types.h" 476 #include "rebalance_types.h" 477 #include "replicas_types.h" 478 #include "sb-members_types.h" 479 #include "subvolume_types.h" 480 #include "super_types.h" 481 #include "thread_with_file_types.h" 482 483 /* Number of nodes btree coalesce will try to coalesce at once */ 484 #define GC_MERGE_NODES 4U 485 486 /* Maximum number of nodes we might need to allocate atomically: */ 487 #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1)) 488 489 /* Size of the freelist we allocate btree nodes from: */ 490 #define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4) 491 492 #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX) 493 494 struct btree; 495 496 struct io_count { 497 u64 sectors[2][BCH_DATA_NR]; 498 }; 499 500 struct discard_in_flight { 501 bool in_progress:1; 502 u64 bucket:63; 503 }; 504 505 struct bch_dev { 506 struct kobject kobj; 507 #ifdef CONFIG_BCACHEFS_DEBUG 508 atomic_long_t ref; 509 bool dying; 510 unsigned long last_put; 511 #else 512 struct percpu_ref ref; 513 #endif 514 struct completion ref_completion; 515 struct percpu_ref io_ref; 516 struct completion io_ref_completion; 517 518 struct bch_fs *fs; 519 520 u8 dev_idx; 521 /* 522 * Cached version of this device's member info from superblock 523 * Committed by bch2_write_super() -> bch_fs_mi_update() 524 */ 525 struct bch_member_cpu mi; 526 atomic64_t errors[BCH_MEMBER_ERROR_NR]; 527 528 __uuid_t uuid; 529 char name[BDEVNAME_SIZE]; 530 531 struct bch_sb_handle disk_sb; 532 struct bch_sb *sb_read_scratch; 533 int sb_write_error; 534 dev_t dev; 535 atomic_t flush_seq; 536 537 struct bch_devs_mask self; 538 539 /* 540 * Buckets: 541 * Per-bucket arrays are protected by c->mark_lock, bucket_lock and 542 * gc_gens_lock, for device resize - holding any is sufficient for 543 * access: Or rcu_read_lock(), but only for dev_ptr_stale(): 544 */ 545 struct bucket_array __rcu *buckets_gc; 546 struct bucket_gens __rcu *bucket_gens; 547 u8 *oldest_gen; 548 unsigned long *buckets_nouse; 549 struct rw_semaphore bucket_lock; 550 551 struct bch_dev_usage __percpu *usage; 552 553 /* Allocator: */ 554 u64 new_fs_bucket_idx; 555 u64 alloc_cursor[3]; 556 557 unsigned nr_open_buckets; 558 unsigned nr_btree_reserve; 559 560 size_t inc_gen_needs_gc; 561 size_t inc_gen_really_needs_gc; 562 size_t buckets_waiting_on_journal; 563 564 struct work_struct invalidate_work; 565 struct work_struct discard_work; 566 struct mutex discard_buckets_in_flight_lock; 567 DARRAY(struct discard_in_flight) discard_buckets_in_flight; 568 struct work_struct discard_fast_work; 569 570 atomic64_t rebalance_work; 571 572 struct journal_device journal; 573 u64 prev_journal_sector; 574 575 struct work_struct io_error_work; 576 577 /* The rest of this all shows up in sysfs */ 578 atomic64_t cur_latency[2]; 579 struct bch2_time_stats_quantiles io_latency[2]; 580 581 #define CONGESTED_MAX 1024 582 atomic_t congested; 583 u64 congested_last; 584 585 struct io_count __percpu *io_done; 586 }; 587 588 /* 589 * initial_gc_unfixed 590 * error 591 * topology error 592 */ 593 594 #define BCH_FS_FLAGS() \ 595 x(new_fs) \ 596 x(started) \ 597 x(btree_running) \ 598 x(accounting_replay_done) \ 599 x(may_go_rw) \ 600 x(rw) \ 601 x(was_rw) \ 602 x(stopping) \ 603 x(emergency_ro) \ 604 x(going_ro) \ 605 x(write_disable_complete) \ 606 x(clean_shutdown) \ 607 x(fsck_running) \ 608 x(initial_gc_unfixed) \ 609 x(need_delete_dead_snapshots) \ 610 x(error) \ 611 x(topology_error) \ 612 x(errors_fixed) \ 613 x(errors_not_fixed) \ 614 x(no_invalid_checks) 615 616 enum bch_fs_flags { 617 #define x(n) BCH_FS_##n, 618 BCH_FS_FLAGS() 619 #undef x 620 }; 621 622 struct btree_debug { 623 unsigned id; 624 }; 625 626 #define BCH_TRANSACTIONS_NR 128 627 628 struct btree_transaction_stats { 629 struct bch2_time_stats duration; 630 struct bch2_time_stats lock_hold_times; 631 struct mutex lock; 632 unsigned nr_max_paths; 633 unsigned journal_entries_size; 634 unsigned max_mem; 635 char *max_paths_text; 636 }; 637 638 struct bch_fs_pcpu { 639 u64 sectors_available; 640 }; 641 642 struct journal_seq_blacklist_table { 643 size_t nr; 644 struct journal_seq_blacklist_table_entry { 645 u64 start; 646 u64 end; 647 bool dirty; 648 } entries[]; 649 }; 650 651 struct journal_keys { 652 /* must match layout in darray_types.h */ 653 size_t nr, size; 654 struct journal_key { 655 u64 journal_seq; 656 u32 journal_offset; 657 enum btree_id btree_id:8; 658 unsigned level:8; 659 bool allocated; 660 bool overwritten; 661 struct bkey_i *k; 662 } *data; 663 /* 664 * Gap buffer: instead of all the empty space in the array being at the 665 * end of the buffer - from @nr to @size - the empty space is at @gap. 666 * This means that sequential insertions are O(n) instead of O(n^2). 667 */ 668 size_t gap; 669 atomic_t ref; 670 bool initial_ref_held; 671 }; 672 673 struct btree_trans_buf { 674 struct btree_trans *trans; 675 }; 676 677 #define BCACHEFS_ROOT_SUBVOL_INUM \ 678 ((subvol_inum) { BCACHEFS_ROOT_SUBVOL, BCACHEFS_ROOT_INO }) 679 680 #define BCH_WRITE_REFS() \ 681 x(trans) \ 682 x(write) \ 683 x(promote) \ 684 x(node_rewrite) \ 685 x(stripe_create) \ 686 x(stripe_delete) \ 687 x(reflink) \ 688 x(fallocate) \ 689 x(fsync) \ 690 x(dio_write) \ 691 x(discard) \ 692 x(discard_fast) \ 693 x(invalidate) \ 694 x(delete_dead_snapshots) \ 695 x(gc_gens) \ 696 x(snapshot_delete_pagecache) \ 697 x(sysfs) \ 698 x(btree_write_buffer) 699 700 enum bch_write_ref { 701 #define x(n) BCH_WRITE_REF_##n, 702 BCH_WRITE_REFS() 703 #undef x 704 BCH_WRITE_REF_NR, 705 }; 706 707 struct bch_fs { 708 struct closure cl; 709 710 struct list_head list; 711 struct kobject kobj; 712 struct kobject counters_kobj; 713 struct kobject internal; 714 struct kobject opts_dir; 715 struct kobject time_stats; 716 unsigned long flags; 717 718 int minor; 719 struct device *chardev; 720 struct super_block *vfs_sb; 721 dev_t dev; 722 char name[40]; 723 struct stdio_redirect *stdio; 724 struct task_struct *stdio_filter; 725 726 /* ro/rw, add/remove/resize devices: */ 727 struct rw_semaphore state_lock; 728 729 /* Counts outstanding writes, for clean transition to read-only */ 730 #ifdef BCH_WRITE_REF_DEBUG 731 atomic_long_t writes[BCH_WRITE_REF_NR]; 732 #else 733 struct percpu_ref writes; 734 #endif 735 /* 736 * Analagous to c->writes, for asynchronous ops that don't necessarily 737 * need fs to be read-write 738 */ 739 refcount_t ro_ref; 740 wait_queue_head_t ro_ref_wait; 741 742 struct work_struct read_only_work; 743 744 struct bch_dev __rcu *devs[BCH_SB_MEMBERS_MAX]; 745 746 struct bch_accounting_mem accounting; 747 748 struct bch_replicas_cpu replicas; 749 struct bch_replicas_cpu replicas_gc; 750 struct mutex replicas_gc_lock; 751 752 struct journal_entry_res btree_root_journal_res; 753 struct journal_entry_res clock_journal_res; 754 755 struct bch_disk_groups_cpu __rcu *disk_groups; 756 757 struct bch_opts opts; 758 759 /* Updated by bch2_sb_update():*/ 760 struct { 761 __uuid_t uuid; 762 __uuid_t user_uuid; 763 764 u16 version; 765 u16 version_min; 766 u16 version_upgrade_complete; 767 768 u8 nr_devices; 769 u8 clean; 770 771 u8 encryption_type; 772 773 u64 time_base_lo; 774 u32 time_base_hi; 775 unsigned time_units_per_sec; 776 unsigned nsec_per_time_unit; 777 u64 features; 778 u64 compat; 779 unsigned long errors_silent[BITS_TO_LONGS(BCH_SB_ERR_MAX)]; 780 u64 btrees_lost_data; 781 } sb; 782 783 784 struct bch_sb_handle disk_sb; 785 786 unsigned short block_bits; /* ilog2(block_size) */ 787 788 u16 btree_foreground_merge_threshold; 789 790 struct closure sb_write; 791 struct mutex sb_lock; 792 793 /* snapshot.c: */ 794 struct snapshot_table __rcu *snapshots; 795 struct mutex snapshot_table_lock; 796 struct rw_semaphore snapshot_create_lock; 797 798 struct work_struct snapshot_delete_work; 799 struct work_struct snapshot_wait_for_pagecache_and_delete_work; 800 snapshot_id_list snapshots_unlinked; 801 struct mutex snapshots_unlinked_lock; 802 803 /* BTREE CACHE */ 804 struct bio_set btree_bio; 805 struct workqueue_struct *btree_read_complete_wq; 806 struct workqueue_struct *btree_write_submit_wq; 807 808 struct btree_root btree_roots_known[BTREE_ID_NR]; 809 DARRAY(struct btree_root) btree_roots_extra; 810 struct mutex btree_root_lock; 811 812 struct btree_cache btree_cache; 813 814 /* 815 * Cache of allocated btree nodes - if we allocate a btree node and 816 * don't use it, if we free it that space can't be reused until going 817 * _all_ the way through the allocator (which exposes us to a livelock 818 * when allocating btree reserves fail halfway through) - instead, we 819 * can stick them here: 820 */ 821 struct btree_alloc btree_reserve_cache[BTREE_NODE_RESERVE * 2]; 822 unsigned btree_reserve_cache_nr; 823 struct mutex btree_reserve_cache_lock; 824 825 mempool_t btree_interior_update_pool; 826 struct list_head btree_interior_update_list; 827 struct list_head btree_interior_updates_unwritten; 828 struct mutex btree_interior_update_lock; 829 struct closure_waitlist btree_interior_update_wait; 830 831 struct workqueue_struct *btree_interior_update_worker; 832 struct work_struct btree_interior_update_work; 833 834 struct workqueue_struct *btree_node_rewrite_worker; 835 836 struct list_head pending_node_rewrites; 837 struct mutex pending_node_rewrites_lock; 838 839 /* btree_io.c: */ 840 spinlock_t btree_write_error_lock; 841 struct btree_write_stats { 842 atomic64_t nr; 843 atomic64_t bytes; 844 } btree_write_stats[BTREE_WRITE_TYPE_NR]; 845 846 /* btree_iter.c: */ 847 struct seqmutex btree_trans_lock; 848 struct list_head btree_trans_list; 849 mempool_t btree_trans_pool; 850 mempool_t btree_trans_mem_pool; 851 struct btree_trans_buf __percpu *btree_trans_bufs; 852 853 struct srcu_struct btree_trans_barrier; 854 bool btree_trans_barrier_initialized; 855 856 struct btree_key_cache btree_key_cache; 857 unsigned btree_key_cache_btrees; 858 859 struct btree_write_buffer btree_write_buffer; 860 861 struct workqueue_struct *btree_update_wq; 862 struct workqueue_struct *btree_io_complete_wq; 863 /* copygc needs its own workqueue for index updates.. */ 864 struct workqueue_struct *copygc_wq; 865 /* 866 * Use a dedicated wq for write ref holder tasks. Required to avoid 867 * dependency problems with other wq tasks that can block on ref 868 * draining, such as read-only transition. 869 */ 870 struct workqueue_struct *write_ref_wq; 871 872 /* ALLOCATION */ 873 struct bch_devs_mask rw_devs[BCH_DATA_NR]; 874 875 u64 capacity; /* sectors */ 876 u64 reserved; /* sectors */ 877 878 /* 879 * When capacity _decreases_ (due to a disk being removed), we 880 * increment capacity_gen - this invalidates outstanding reservations 881 * and forces them to be revalidated 882 */ 883 u32 capacity_gen; 884 unsigned bucket_size_max; 885 886 atomic64_t sectors_available; 887 struct mutex sectors_available_lock; 888 889 struct bch_fs_pcpu __percpu *pcpu; 890 891 struct percpu_rw_semaphore mark_lock; 892 893 seqcount_t usage_lock; 894 struct bch_fs_usage_base __percpu *usage; 895 u64 __percpu *online_reserved; 896 897 unsigned long allocator_last_stuck; 898 899 struct io_clock io_clock[2]; 900 901 /* JOURNAL SEQ BLACKLIST */ 902 struct journal_seq_blacklist_table * 903 journal_seq_blacklist_table; 904 905 /* ALLOCATOR */ 906 spinlock_t freelist_lock; 907 struct closure_waitlist freelist_wait; 908 909 open_bucket_idx_t open_buckets_freelist; 910 open_bucket_idx_t open_buckets_nr_free; 911 struct closure_waitlist open_buckets_wait; 912 struct open_bucket open_buckets[OPEN_BUCKETS_COUNT]; 913 open_bucket_idx_t open_buckets_hash[OPEN_BUCKETS_COUNT]; 914 915 open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT]; 916 open_bucket_idx_t open_buckets_partial_nr; 917 918 struct write_point btree_write_point; 919 struct write_point rebalance_write_point; 920 921 struct write_point write_points[WRITE_POINT_MAX]; 922 struct hlist_head write_points_hash[WRITE_POINT_HASH_NR]; 923 struct mutex write_points_hash_lock; 924 unsigned write_points_nr; 925 926 struct buckets_waiting_for_journal buckets_waiting_for_journal; 927 928 /* GARBAGE COLLECTION */ 929 struct work_struct gc_gens_work; 930 unsigned long gc_count; 931 932 enum btree_id gc_gens_btree; 933 struct bpos gc_gens_pos; 934 935 /* 936 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos] 937 * has been marked by GC. 938 * 939 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.) 940 * 941 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread 942 * can read without a lock. 943 */ 944 seqcount_t gc_pos_lock; 945 struct gc_pos gc_pos; 946 947 /* 948 * The allocation code needs gc_mark in struct bucket to be correct, but 949 * it's not while a gc is in progress. 950 */ 951 struct rw_semaphore gc_lock; 952 struct mutex gc_gens_lock; 953 954 /* IO PATH */ 955 struct semaphore io_in_flight; 956 struct bio_set bio_read; 957 struct bio_set bio_read_split; 958 struct bio_set bio_write; 959 struct bio_set replica_set; 960 struct mutex bio_bounce_pages_lock; 961 mempool_t bio_bounce_pages; 962 struct bucket_nocow_lock_table 963 nocow_locks; 964 struct rhashtable promote_table; 965 966 mempool_t compression_bounce[2]; 967 mempool_t compress_workspace[BCH_COMPRESSION_TYPE_NR]; 968 mempool_t decompress_workspace; 969 size_t zstd_workspace_size; 970 971 struct crypto_shash *sha256; 972 struct crypto_sync_skcipher *chacha20; 973 struct crypto_shash *poly1305; 974 975 atomic64_t key_version; 976 977 mempool_t large_bkey_pool; 978 979 /* MOVE.C */ 980 struct list_head moving_context_list; 981 struct mutex moving_context_lock; 982 983 /* REBALANCE */ 984 struct bch_fs_rebalance rebalance; 985 986 /* COPYGC */ 987 struct task_struct *copygc_thread; 988 struct write_point copygc_write_point; 989 s64 copygc_wait_at; 990 s64 copygc_wait; 991 bool copygc_running; 992 wait_queue_head_t copygc_running_wq; 993 994 /* STRIPES: */ 995 GENRADIX(struct stripe) stripes; 996 GENRADIX(struct gc_stripe) gc_stripes; 997 998 struct hlist_head ec_stripes_new[32]; 999 spinlock_t ec_stripes_new_lock; 1000 1001 ec_stripes_heap ec_stripes_heap; 1002 struct mutex ec_stripes_heap_lock; 1003 1004 /* ERASURE CODING */ 1005 struct list_head ec_stripe_head_list; 1006 struct mutex ec_stripe_head_lock; 1007 1008 struct list_head ec_stripe_new_list; 1009 struct mutex ec_stripe_new_lock; 1010 wait_queue_head_t ec_stripe_new_wait; 1011 1012 struct work_struct ec_stripe_create_work; 1013 u64 ec_stripe_hint; 1014 1015 struct work_struct ec_stripe_delete_work; 1016 1017 struct bio_set ec_bioset; 1018 1019 /* REFLINK */ 1020 reflink_gc_table reflink_gc_table; 1021 size_t reflink_gc_nr; 1022 1023 /* fs.c */ 1024 struct list_head vfs_inodes_list; 1025 struct mutex vfs_inodes_lock; 1026 1027 /* VFS IO PATH - fs-io.c */ 1028 struct bio_set writepage_bioset; 1029 struct bio_set dio_write_bioset; 1030 struct bio_set dio_read_bioset; 1031 struct bio_set nocow_flush_bioset; 1032 1033 /* QUOTAS */ 1034 struct bch_memquota_type quotas[QTYP_NR]; 1035 1036 /* RECOVERY */ 1037 u64 journal_replay_seq_start; 1038 u64 journal_replay_seq_end; 1039 /* 1040 * Two different uses: 1041 * "Has this fsck pass?" - i.e. should this type of error be an 1042 * emergency read-only 1043 * And, in certain situations fsck will rewind to an earlier pass: used 1044 * for signaling to the toplevel code which pass we want to run now. 1045 */ 1046 enum bch_recovery_pass curr_recovery_pass; 1047 /* bitmap of explicitly enabled recovery passes: */ 1048 u64 recovery_passes_explicit; 1049 /* bitmask of recovery passes that we actually ran */ 1050 u64 recovery_passes_complete; 1051 /* never rewinds version of curr_recovery_pass */ 1052 enum bch_recovery_pass recovery_pass_done; 1053 struct semaphore online_fsck_mutex; 1054 1055 /* DEBUG JUNK */ 1056 struct dentry *fs_debug_dir; 1057 struct dentry *btree_debug_dir; 1058 struct btree_debug btree_debug[BTREE_ID_NR]; 1059 struct btree *verify_data; 1060 struct btree_node *verify_ondisk; 1061 struct mutex verify_lock; 1062 1063 u64 *unused_inode_hints; 1064 unsigned inode_shard_bits; 1065 1066 /* 1067 * A btree node on disk could have too many bsets for an iterator to fit 1068 * on the stack - have to dynamically allocate them 1069 */ 1070 mempool_t fill_iter; 1071 1072 mempool_t btree_bounce_pool; 1073 1074 struct journal journal; 1075 GENRADIX(struct journal_replay *) journal_entries; 1076 u64 journal_entries_base_seq; 1077 struct journal_keys journal_keys; 1078 struct list_head journal_iters; 1079 1080 struct find_btree_nodes found_btree_nodes; 1081 1082 u64 last_bucket_seq_cleanup; 1083 1084 u64 counters_on_mount[BCH_COUNTER_NR]; 1085 u64 __percpu *counters; 1086 1087 unsigned copy_gc_enabled:1; 1088 bool promote_whole_extents; 1089 1090 struct bch2_time_stats times[BCH_TIME_STAT_NR]; 1091 1092 struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR]; 1093 1094 /* ERRORS */ 1095 struct list_head fsck_error_msgs; 1096 struct mutex fsck_error_msgs_lock; 1097 bool fsck_alloc_msgs_err; 1098 1099 bch_sb_errors_cpu fsck_error_counts; 1100 struct mutex fsck_error_counts_lock; 1101 }; 1102 1103 extern struct wait_queue_head bch2_read_only_wait; 1104 1105 static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref) 1106 { 1107 #ifdef BCH_WRITE_REF_DEBUG 1108 atomic_long_inc(&c->writes[ref]); 1109 #else 1110 percpu_ref_get(&c->writes); 1111 #endif 1112 } 1113 1114 static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref) 1115 { 1116 #ifdef BCH_WRITE_REF_DEBUG 1117 return !test_bit(BCH_FS_going_ro, &c->flags) && 1118 atomic_long_inc_not_zero(&c->writes[ref]); 1119 #else 1120 return percpu_ref_tryget(&c->writes); 1121 #endif 1122 } 1123 1124 static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref) 1125 { 1126 #ifdef BCH_WRITE_REF_DEBUG 1127 return !test_bit(BCH_FS_going_ro, &c->flags) && 1128 atomic_long_inc_not_zero(&c->writes[ref]); 1129 #else 1130 return percpu_ref_tryget_live(&c->writes); 1131 #endif 1132 } 1133 1134 static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref) 1135 { 1136 #ifdef BCH_WRITE_REF_DEBUG 1137 long v = atomic_long_dec_return(&c->writes[ref]); 1138 1139 BUG_ON(v < 0); 1140 if (v) 1141 return; 1142 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) 1143 if (atomic_long_read(&c->writes[i])) 1144 return; 1145 1146 set_bit(BCH_FS_write_disable_complete, &c->flags); 1147 wake_up(&bch2_read_only_wait); 1148 #else 1149 percpu_ref_put(&c->writes); 1150 #endif 1151 } 1152 1153 static inline bool bch2_ro_ref_tryget(struct bch_fs *c) 1154 { 1155 if (test_bit(BCH_FS_stopping, &c->flags)) 1156 return false; 1157 1158 return refcount_inc_not_zero(&c->ro_ref); 1159 } 1160 1161 static inline void bch2_ro_ref_put(struct bch_fs *c) 1162 { 1163 if (refcount_dec_and_test(&c->ro_ref)) 1164 wake_up(&c->ro_ref_wait); 1165 } 1166 1167 static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages) 1168 { 1169 #ifndef NO_BCACHEFS_FS 1170 if (c->vfs_sb) 1171 c->vfs_sb->s_bdi->ra_pages = ra_pages; 1172 #endif 1173 } 1174 1175 static inline unsigned bucket_bytes(const struct bch_dev *ca) 1176 { 1177 return ca->mi.bucket_size << 9; 1178 } 1179 1180 static inline unsigned block_bytes(const struct bch_fs *c) 1181 { 1182 return c->opts.block_size; 1183 } 1184 1185 static inline unsigned block_sectors(const struct bch_fs *c) 1186 { 1187 return c->opts.block_size >> 9; 1188 } 1189 1190 static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree) 1191 { 1192 return c->btree_key_cache_btrees & (1U << btree); 1193 } 1194 1195 static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time) 1196 { 1197 struct timespec64 t; 1198 s32 rem; 1199 1200 time += c->sb.time_base_lo; 1201 1202 t.tv_sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem); 1203 t.tv_nsec = rem * c->sb.nsec_per_time_unit; 1204 return t; 1205 } 1206 1207 static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts) 1208 { 1209 return (ts.tv_sec * c->sb.time_units_per_sec + 1210 (int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo; 1211 } 1212 1213 static inline s64 bch2_current_time(const struct bch_fs *c) 1214 { 1215 struct timespec64 now; 1216 1217 ktime_get_coarse_real_ts64(&now); 1218 return timespec_to_bch2_time(c, now); 1219 } 1220 1221 static inline u64 bch2_current_io_time(const struct bch_fs *c, int rw) 1222 { 1223 return max(1ULL, (u64) atomic64_read(&c->io_clock[rw].now) & LRU_TIME_MAX); 1224 } 1225 1226 static inline struct stdio_redirect *bch2_fs_stdio_redirect(struct bch_fs *c) 1227 { 1228 struct stdio_redirect *stdio = c->stdio; 1229 1230 if (c->stdio_filter && c->stdio_filter != current) 1231 stdio = NULL; 1232 return stdio; 1233 } 1234 1235 static inline unsigned metadata_replicas_required(struct bch_fs *c) 1236 { 1237 return min(c->opts.metadata_replicas, 1238 c->opts.metadata_replicas_required); 1239 } 1240 1241 static inline unsigned data_replicas_required(struct bch_fs *c) 1242 { 1243 return min(c->opts.data_replicas, 1244 c->opts.data_replicas_required); 1245 } 1246 1247 #define BKEY_PADDED_ONSTACK(key, pad) \ 1248 struct { struct bkey_i key; __u64 key ## _pad[pad]; } 1249 1250 #endif /* _BCACHEFS_H */ 1251