brintos

brintos / linux-shallow public Read only

0
0
Text · 37.0 KiB · e94a83b Raw
1255 lines · c
1/* SPDX-License-Identifier: GPL-2.0 */2#ifndef _BCACHEFS_H3#define _BCACHEFS_H4 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, but11 * it's about 95% plumbed through. A cache set and its cache devices is sort of12 * like a md raid array and its component devices. Most of the code doesn't care13 * 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 dirty16 * cached data and metadata, without mirroring clean cached data.17 *18 * Backing devices are different, in that they have a lifetime independent of a19 * cache set. When you register a newly formatted backing device it'll come up20 * in passthrough mode, and then you can attach and detach a backing device from21 * a cache set at runtime - while it's mounted and in use. Detaching implicitly22 * 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 distinction27 * between struct cached_dev and struct bcache_device. A flash only volume28 * works much like a bcache device that has a backing device, except the29 * "cached" data is always dirty. The end result is that we get thin30 * provisioning with very little additional code.31 *32 * Flash only volumes work but they're not production ready because the moving33 * 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 normal38 * operation all of our available space will be allocated. Thus, we need an39 * efficient way of deleting things from the cache so we can write new things to40 * it.41 *42 * To do this, we first divide the cache device up into buckets. A bucket is the43 * 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 with47 * it. The gens and priorities for all the buckets are stored contiguously and48 * packed on disk (in a linked list of buckets - aside from the superblock, all49 * of bcache's metadata is stored in buckets).50 *51 * The priority is used to implement an LRU. We reset a bucket's priority when52 * we allocate it or on cache it, and every so often we decrement the priority53 * 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 857 * bit generation embedded in it; for a pointer to be considered valid, its gen58 * must match the gen of the bucket it points into.  Thus, to reuse a bucket all59 * we have to do is increment its gen (and write its new gen to disk; we batch60 * this up).61 *62 * Bcache is entirely COW - we never write twice to a bucket, even buckets that63 * 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 variable72 * number of pointers attached to them (potentially zero, which is handy for73 * invalidating the cache).74 *75 * The key itself is an inode:offset pair. The inode number corresponds to a76 * backing device or a flash only volume. The offset is the ending offset of the77 * extent within the inode - not the starting offset; this makes lookups78 * slightly more convenient.79 *80 * Pointers contain the cache device id, the offset on that device, and an 8 bit81 * generation number. More on the gen later.82 *83 * Index lookups are not fully abstracted - cache lookups in particular are84 * still somewhat mixed in with the btree code, but things are headed in that85 * direction.86 *87 * Updates are fairly well abstracted, though. There are two different ways of88 * 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 is92 * used to update the index after a write.93 *94 * BTREE_REPLACE is really cmpxchg(); it inserts a key into the btree iff it is95 * overwriting a key that matches another given key. This is used for inserting96 * data into the cache after a cache miss, and for background writeback, and for97 * the moving garbage collector.98 *99 * There is no "delete" operation; deleting things from the index is100 * accomplished by either by invalidating pointers (by incrementing a bucket's101 * gen) or by inserting a key with 0 pointers - which will overwrite anything102 * previously present at that location in the index.103 *104 * This means that there are always stale/invalid keys in the btree. They're105 * filtered out by the code that iterates through a btree node, and removed when106 * a btree node is rewritten.107 *108 * BTREE NODES:109 *110 * Our unit of allocation is a bucket, and we can't arbitrarily allocate and111 * 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 node114 * - no less than 1/4th - but a bucket still contains no more than a single115 * btree node. I'd actually like to change this, but for now we rely on the116 * 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 textbook119 * btree implementation.120 *121 * The way this is solved is that btree nodes are internally log structured; we122 * can append new keys to an existing btree node without rewriting it. This123 * 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 would126 * be expensive, and we have to distinguish between the keys we have written and127 * the keys we haven't. So to do a lookup in a btree node, we have to search128 * each sorted set. But we do merge written sets together lazily, so the cost of129 * these extra searches is quite low (normally most of the keys in a btree node130 * will be in one big set, and then there'll be one or two sets that are much131 * smaller).132 *133 * This log structure makes bcache's btree more of a hybrid between a134 * conventional btree and a compacting data structure, with some of the135 * advantages of both.136 *137 * GARBAGE COLLECTION:138 *139 * We can't just invalidate any bucket - it might contain dirty data or140 * metadata. If it once contained dirty data, other writes might overwrite it141 * 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 that145 * allocation can reuse buckets sooner when they've been mostly overwritten.146 *147 * It also does some things that are really internal to the btree148 * implementation. If a btree node contains pointers that are stale by more than149 * some threshold, it rewrites the btree node to avoid the bucket's generation150 * 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 strictly155 * order metadata writes so that the btree and everything else is consistent on156 * disk in the event of an unclean shutdown, and in fact bcache had writeback157 * caching (with recovery from unclean shutdown) before journalling was158 * implemented.159 *160 * Rather, the journal is purely a performance optimization; we can't complete a161 * write until we've updated the index on disk, otherwise the cache would be162 * inconsistent in the event of an unclean shutdown. This means that without the163 * journal, on random write workloads we constantly have to update all the leaf164 * nodes in the btree, and those writes will be mostly empty (appending at most165 * 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 reinsert169 * all the keys in the open journal entries. That means that when we're updating170 * a node in the btree, we can wait until a 4k block of keys fills up before171 * writing them out.172 *173 * For simplicity, we only journal updates to leaf nodes; updates to parent174 * nodes are rare enough (since our leaf nodes are huge) that it wasn't worth175 * 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_fmt180#ifdef __KERNEL__181#define pr_fmt(fmt) "bcachefs: %s() " fmt "\n", __func__182#else183#define pr_fmt(fmt) "%s() " fmt "\n", __func__184#endif185 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_DEBUG220#define BCH_WRITE_REF_DEBUG221#endif222 223#ifndef dynamic_fault224#define dynamic_fault(...)		0225#endif226 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, ...)					\232do {									\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_PREFIX246#endif247 248#ifdef BCACHEFS_LOG_PREFIX249 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#else258 259#define bch2_log_msg(_c, fmt)			fmt260#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#endif267 268#define bch2_fmt(_c, fmt)		bch2_log_msg(_c, fmt "\n")269 270void bch2_print_str(struct bch_fs *, const char *);271 272__printf(2, 3)273void bch2_print_opts(struct bch_opts *, const char *, ...);274 275__printf(2, 3)276void __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, ...)					\285do {									\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 325static 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)						\331do {									\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)				\337do {									\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, ...)				\343do {									\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, ...)					\350do {									\351	if ((c)->opts.verbose)						\352		bch_info(c, fmt, ##__VA_ARGS__);			\353} while (0)354 355#define pr_verbose_init(opts, fmt, ...)					\356do {									\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_DEBUG415#define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALL()416#else417#define BCH_DEBUG_PARAMS() BCH_DEBUG_PARAMS_ALWAYS()418#endif419 420#define BCH_DEBUG_PARAM(name, description) extern bool bch2_##name;421BCH_DEBUG_PARAMS()422#undef BCH_DEBUG_PARAM423 424#ifndef CONFIG_BCACHEFS_DEBUG425#define BCH_DEBUG_PARAM(name, description) static const __maybe_unused bool bch2_##name;426BCH_DEBUG_PARAMS_DEBUG()427#undef BCH_DEBUG_PARAM428#endif429 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 456enum bch_time_stats {457#define x(name) BCH_TIME_##name,458	BCH_TIME_STATS()459#undef x460	BCH_TIME_STAT_NR461};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		4U485 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 494struct btree;495 496struct io_count {497	u64			sectors[2][BCH_DATA_NR];498};499 500struct discard_in_flight {501	bool			in_progress:1;502	u64			bucket:63;503};504 505struct bch_dev {506	struct kobject		kobj;507#ifdef CONFIG_BCACHEFS_DEBUG508	atomic_long_t		ref;509	bool			dying;510	unsigned long		last_put;511#else512	struct percpu_ref	ref;513#endif514	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 superblock523	 * 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 and542	 * gc_gens_lock, for device resize - holding any is sufficient for543	 * access: Or rcu_read_lock(), but only for dev_ptr_stale():544	 */545	GENRADIX(struct bucket)	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_partial_buckets;559	unsigned		nr_btree_reserve;560 561	size_t			inc_gen_needs_gc;562	size_t			inc_gen_really_needs_gc;563	size_t			buckets_waiting_on_journal;564 565	struct work_struct	invalidate_work;566	struct work_struct	discard_work;567	struct mutex		discard_buckets_in_flight_lock;568	DARRAY(struct discard_in_flight)	discard_buckets_in_flight;569	struct work_struct	discard_fast_work;570 571	atomic64_t		rebalance_work;572 573	struct journal_device	journal;574	u64			prev_journal_sector;575 576	struct work_struct	io_error_work;577 578	/* The rest of this all shows up in sysfs */579	atomic64_t		cur_latency[2];580	struct bch2_time_stats_quantiles io_latency[2];581 582#define CONGESTED_MAX		1024583	atomic_t		congested;584	u64			congested_last;585 586	struct io_count __percpu *io_done;587};588 589/*590 * initial_gc_unfixed591 * error592 * topology error593 */594 595#define BCH_FS_FLAGS()			\596	x(new_fs)			\597	x(started)			\598	x(clean_recovery)		\599	x(btree_running)		\600	x(accounting_replay_done)	\601	x(may_go_rw)			\602	x(rw)				\603	x(was_rw)			\604	x(stopping)			\605	x(emergency_ro)			\606	x(going_ro)			\607	x(write_disable_complete)	\608	x(clean_shutdown)		\609	x(fsck_running)			\610	x(initial_gc_unfixed)		\611	x(need_delete_dead_snapshots)	\612	x(error)			\613	x(topology_error)		\614	x(errors_fixed)			\615	x(errors_not_fixed)		\616	x(no_invalid_checks)617 618enum bch_fs_flags {619#define x(n)		BCH_FS_##n,620	BCH_FS_FLAGS()621#undef x622};623 624struct btree_debug {625	unsigned		id;626};627 628#define BCH_TRANSACTIONS_NR 128629 630struct btree_transaction_stats {631	struct bch2_time_stats	duration;632	struct bch2_time_stats	lock_hold_times;633	struct mutex		lock;634	unsigned		nr_max_paths;635	unsigned		journal_entries_size;636	unsigned		max_mem;637	char			*max_paths_text;638};639 640struct bch_fs_pcpu {641	u64			sectors_available;642};643 644struct journal_seq_blacklist_table {645	size_t			nr;646	struct journal_seq_blacklist_table_entry {647		u64		start;648		u64		end;649		bool		dirty;650	}			entries[];651};652 653struct journal_keys {654	/* must match layout in darray_types.h */655	size_t			nr, size;656	struct journal_key {657		u64		journal_seq;658		u32		journal_offset;659		enum btree_id	btree_id:8;660		unsigned	level:8;661		bool		allocated;662		bool		overwritten;663		struct bkey_i	*k;664	}			*data;665	/*666	 * Gap buffer: instead of all the empty space in the array being at the667	 * end of the buffer - from @nr to @size - the empty space is at @gap.668	 * This means that sequential insertions are O(n) instead of O(n^2).669	 */670	size_t			gap;671	atomic_t		ref;672	bool			initial_ref_held;673};674 675struct btree_trans_buf {676	struct btree_trans	*trans;677};678 679#define BCACHEFS_ROOT_SUBVOL_INUM					\680	((subvol_inum) { BCACHEFS_ROOT_SUBVOL,	BCACHEFS_ROOT_INO })681 682#define BCH_WRITE_REFS()						\683	x(trans)							\684	x(write)							\685	x(promote)							\686	x(node_rewrite)							\687	x(stripe_create)						\688	x(stripe_delete)						\689	x(reflink)							\690	x(fallocate)							\691	x(fsync)							\692	x(dio_write)							\693	x(discard)							\694	x(discard_fast)							\695	x(invalidate)							\696	x(delete_dead_snapshots)					\697	x(gc_gens)							\698	x(snapshot_delete_pagecache)					\699	x(sysfs)							\700	x(btree_write_buffer)701 702enum bch_write_ref {703#define x(n) BCH_WRITE_REF_##n,704	BCH_WRITE_REFS()705#undef x706	BCH_WRITE_REF_NR,707};708 709struct bch_fs {710	struct closure		cl;711 712	struct list_head	list;713	struct kobject		kobj;714	struct kobject		counters_kobj;715	struct kobject		internal;716	struct kobject		opts_dir;717	struct kobject		time_stats;718	unsigned long		flags;719 720	int			minor;721	struct device		*chardev;722	struct super_block	*vfs_sb;723	dev_t			dev;724	char			name[40];725	struct stdio_redirect	*stdio;726	struct task_struct	*stdio_filter;727 728	/* ro/rw, add/remove/resize devices: */729	struct rw_semaphore	state_lock;730 731	/* Counts outstanding writes, for clean transition to read-only */732#ifdef BCH_WRITE_REF_DEBUG733	atomic_long_t		writes[BCH_WRITE_REF_NR];734#else735	struct percpu_ref	writes;736#endif737	/*738	 * Analagous to c->writes, for asynchronous ops that don't necessarily739	 * need fs to be read-write740	 */741	refcount_t		ro_ref;742	wait_queue_head_t	ro_ref_wait;743 744	struct work_struct	read_only_work;745 746	struct bch_dev __rcu	*devs[BCH_SB_MEMBERS_MAX];747 748	struct bch_accounting_mem accounting;749 750	struct bch_replicas_cpu replicas;751	struct bch_replicas_cpu replicas_gc;752	struct mutex		replicas_gc_lock;753 754	struct journal_entry_res btree_root_journal_res;755	struct journal_entry_res clock_journal_res;756 757	struct bch_disk_groups_cpu __rcu *disk_groups;758 759	struct bch_opts		opts;760 761	/* Updated by bch2_sb_update():*/762	struct {763		__uuid_t	uuid;764		__uuid_t	user_uuid;765 766		u16		version;767		u16		version_min;768		u16		version_upgrade_complete;769 770		u8		nr_devices;771		u8		clean;772 773		u8		encryption_type;774 775		u64		time_base_lo;776		u32		time_base_hi;777		unsigned	time_units_per_sec;778		unsigned	nsec_per_time_unit;779		u64		features;780		u64		compat;781		unsigned long	errors_silent[BITS_TO_LONGS(BCH_FSCK_ERR_MAX)];782		u64		btrees_lost_data;783	}			sb;784 785 786	struct bch_sb_handle	disk_sb;787 788	unsigned short		block_bits;	/* ilog2(block_size) */789 790	u16			btree_foreground_merge_threshold;791 792	struct closure		sb_write;793	struct mutex		sb_lock;794 795	/* snapshot.c: */796	struct snapshot_table __rcu *snapshots;797	struct mutex		snapshot_table_lock;798	struct rw_semaphore	snapshot_create_lock;799 800	struct work_struct	snapshot_delete_work;801	struct work_struct	snapshot_wait_for_pagecache_and_delete_work;802	snapshot_id_list	snapshots_unlinked;803	struct mutex		snapshots_unlinked_lock;804 805	/* BTREE CACHE */806	struct bio_set		btree_bio;807	struct workqueue_struct	*btree_read_complete_wq;808	struct workqueue_struct	*btree_write_submit_wq;809 810	struct btree_root	btree_roots_known[BTREE_ID_NR];811	DARRAY(struct btree_root) btree_roots_extra;812	struct mutex		btree_root_lock;813 814	struct btree_cache	btree_cache;815 816	/*817	 * Cache of allocated btree nodes - if we allocate a btree node and818	 * don't use it, if we free it that space can't be reused until going819	 * _all_ the way through the allocator (which exposes us to a livelock820	 * when allocating btree reserves fail halfway through) - instead, we821	 * can stick them here:822	 */823	struct btree_alloc	btree_reserve_cache[BTREE_NODE_RESERVE * 2];824	unsigned		btree_reserve_cache_nr;825	struct mutex		btree_reserve_cache_lock;826 827	mempool_t		btree_interior_update_pool;828	struct list_head	btree_interior_update_list;829	struct list_head	btree_interior_updates_unwritten;830	struct mutex		btree_interior_update_lock;831	struct closure_waitlist	btree_interior_update_wait;832 833	struct workqueue_struct	*btree_interior_update_worker;834	struct work_struct	btree_interior_update_work;835 836	struct workqueue_struct	*btree_node_rewrite_worker;837 838	struct list_head	pending_node_rewrites;839	struct mutex		pending_node_rewrites_lock;840 841	/* btree_io.c: */842	spinlock_t		btree_write_error_lock;843	struct btree_write_stats {844		atomic64_t	nr;845		atomic64_t	bytes;846	}			btree_write_stats[BTREE_WRITE_TYPE_NR];847 848	/* btree_iter.c: */849	struct seqmutex		btree_trans_lock;850	struct list_head	btree_trans_list;851	mempool_t		btree_trans_pool;852	mempool_t		btree_trans_mem_pool;853	struct btree_trans_buf  __percpu	*btree_trans_bufs;854 855	struct srcu_struct	btree_trans_barrier;856	bool			btree_trans_barrier_initialized;857 858	struct btree_key_cache	btree_key_cache;859	unsigned		btree_key_cache_btrees;860 861	struct btree_write_buffer btree_write_buffer;862 863	struct workqueue_struct	*btree_update_wq;864	struct workqueue_struct	*btree_io_complete_wq;865	/* copygc needs its own workqueue for index updates.. */866	struct workqueue_struct	*copygc_wq;867	/*868	 * Use a dedicated wq for write ref holder tasks. Required to avoid869	 * dependency problems with other wq tasks that can block on ref870	 * draining, such as read-only transition.871	 */872	struct workqueue_struct *write_ref_wq;873 874	/* ALLOCATION */875	struct bch_devs_mask	rw_devs[BCH_DATA_NR];876	unsigned long		rw_devs_change_count;877 878	u64			capacity; /* sectors */879	u64			reserved; /* sectors */880 881	/*882	 * When capacity _decreases_ (due to a disk being removed), we883	 * increment capacity_gen - this invalidates outstanding reservations884	 * and forces them to be revalidated885	 */886	u32			capacity_gen;887	unsigned		bucket_size_max;888 889	atomic64_t		sectors_available;890	struct mutex		sectors_available_lock;891 892	struct bch_fs_pcpu __percpu	*pcpu;893 894	struct percpu_rw_semaphore	mark_lock;895 896	seqcount_t			usage_lock;897	struct bch_fs_usage_base __percpu *usage;898	u64 __percpu		*online_reserved;899 900	unsigned long		allocator_last_stuck;901 902	struct io_clock		io_clock[2];903 904	/* JOURNAL SEQ BLACKLIST */905	struct journal_seq_blacklist_table *906				journal_seq_blacklist_table;907 908	/* ALLOCATOR */909	spinlock_t		freelist_lock;910	struct closure_waitlist	freelist_wait;911 912	open_bucket_idx_t	open_buckets_freelist;913	open_bucket_idx_t	open_buckets_nr_free;914	struct closure_waitlist	open_buckets_wait;915	struct open_bucket	open_buckets[OPEN_BUCKETS_COUNT];916	open_bucket_idx_t	open_buckets_hash[OPEN_BUCKETS_COUNT];917 918	open_bucket_idx_t	open_buckets_partial[OPEN_BUCKETS_COUNT];919	open_bucket_idx_t	open_buckets_partial_nr;920 921	struct write_point	btree_write_point;922	struct write_point	rebalance_write_point;923 924	struct write_point	write_points[WRITE_POINT_MAX];925	struct hlist_head	write_points_hash[WRITE_POINT_HASH_NR];926	struct mutex		write_points_hash_lock;927	unsigned		write_points_nr;928 929	struct buckets_waiting_for_journal buckets_waiting_for_journal;930 931	/* GARBAGE COLLECTION */932	struct work_struct	gc_gens_work;933	unsigned long		gc_count;934 935	enum btree_id		gc_gens_btree;936	struct bpos		gc_gens_pos;937 938	/*939	 * Tracks GC's progress - everything in the range [ZERO_KEY..gc_cur_pos]940	 * has been marked by GC.941	 *942	 * gc_cur_phase is a superset of btree_ids (BTREE_ID_extents etc.)943	 *944	 * Protected by gc_pos_lock. Only written to by GC thread, so GC thread945	 * can read without a lock.946	 */947	seqcount_t		gc_pos_lock;948	struct gc_pos		gc_pos;949 950	/*951	 * The allocation code needs gc_mark in struct bucket to be correct, but952	 * it's not while a gc is in progress.953	 */954	struct rw_semaphore	gc_lock;955	struct mutex		gc_gens_lock;956 957	/* IO PATH */958	struct semaphore	io_in_flight;959	struct bio_set		bio_read;960	struct bio_set		bio_read_split;961	struct bio_set		bio_write;962	struct bio_set		replica_set;963	struct mutex		bio_bounce_pages_lock;964	mempool_t		bio_bounce_pages;965	struct bucket_nocow_lock_table966				nocow_locks;967	struct rhashtable	promote_table;968 969	mempool_t		compression_bounce[2];970	mempool_t		compress_workspace[BCH_COMPRESSION_TYPE_NR];971	mempool_t		decompress_workspace;972	size_t			zstd_workspace_size;973 974	struct crypto_shash	*sha256;975	struct crypto_sync_skcipher *chacha20;976	struct crypto_shash	*poly1305;977 978	atomic64_t		key_version;979 980	mempool_t		large_bkey_pool;981 982	/* MOVE.C */983	struct list_head	moving_context_list;984	struct mutex		moving_context_lock;985 986	/* REBALANCE */987	struct bch_fs_rebalance	rebalance;988 989	/* COPYGC */990	struct task_struct	*copygc_thread;991	struct write_point	copygc_write_point;992	s64			copygc_wait_at;993	s64			copygc_wait;994	bool			copygc_running;995	wait_queue_head_t	copygc_running_wq;996 997	/* STRIPES: */998	GENRADIX(struct stripe) stripes;999	GENRADIX(struct gc_stripe) gc_stripes;1000 1001	struct hlist_head	ec_stripes_new[32];1002	spinlock_t		ec_stripes_new_lock;1003 1004	ec_stripes_heap		ec_stripes_heap;1005	struct mutex		ec_stripes_heap_lock;1006 1007	/* ERASURE CODING */1008	struct list_head	ec_stripe_head_list;1009	struct mutex		ec_stripe_head_lock;1010 1011	struct list_head	ec_stripe_new_list;1012	struct mutex		ec_stripe_new_lock;1013	wait_queue_head_t	ec_stripe_new_wait;1014 1015	struct work_struct	ec_stripe_create_work;1016	u64			ec_stripe_hint;1017 1018	struct work_struct	ec_stripe_delete_work;1019 1020	struct bio_set		ec_bioset;1021 1022	/* REFLINK */1023	reflink_gc_table	reflink_gc_table;1024	size_t			reflink_gc_nr;1025 1026	/* fs.c */1027	struct list_head	vfs_inodes_list;1028	struct mutex		vfs_inodes_lock;1029	struct rhashtable	vfs_inodes_table;1030 1031	/* VFS IO PATH - fs-io.c */1032	struct bio_set		writepage_bioset;1033	struct bio_set		dio_write_bioset;1034	struct bio_set		dio_read_bioset;1035	struct bio_set		nocow_flush_bioset;1036 1037	/* QUOTAS */1038	struct bch_memquota_type quotas[QTYP_NR];1039 1040	/* RECOVERY */1041	u64			journal_replay_seq_start;1042	u64			journal_replay_seq_end;1043	/*1044	 * Two different uses:1045	 * "Has this fsck pass?" - i.e. should this type of error be an1046	 * emergency read-only1047	 * And, in certain situations fsck will rewind to an earlier pass: used1048	 * for signaling to the toplevel code which pass we want to run now.1049	 */1050	enum bch_recovery_pass	curr_recovery_pass;1051	/* bitmask of recovery passes that we actually ran */1052	u64			recovery_passes_complete;1053	/* never rewinds version of curr_recovery_pass */1054	enum bch_recovery_pass	recovery_pass_done;1055	struct semaphore	online_fsck_mutex;1056 1057	/* DEBUG JUNK */1058	struct dentry		*fs_debug_dir;1059	struct dentry		*btree_debug_dir;1060	struct btree_debug	btree_debug[BTREE_ID_NR];1061	struct btree		*verify_data;1062	struct btree_node	*verify_ondisk;1063	struct mutex		verify_lock;1064 1065	u64			*unused_inode_hints;1066	unsigned		inode_shard_bits;1067 1068	/*1069	 * A btree node on disk could have too many bsets for an iterator to fit1070	 * on the stack - have to dynamically allocate them1071	 */1072	mempool_t		fill_iter;1073 1074	mempool_t		btree_bounce_pool;1075 1076	struct journal		journal;1077	GENRADIX(struct journal_replay *) journal_entries;1078	u64			journal_entries_base_seq;1079	struct journal_keys	journal_keys;1080	struct list_head	journal_iters;1081 1082	struct find_btree_nodes	found_btree_nodes;1083 1084	u64			last_bucket_seq_cleanup;1085 1086	u64			counters_on_mount[BCH_COUNTER_NR];1087	u64 __percpu		*counters;1088 1089	unsigned		copy_gc_enabled:1;1090 1091	struct bch2_time_stats	times[BCH_TIME_STAT_NR];1092 1093	struct btree_transaction_stats btree_transaction_stats[BCH_TRANSACTIONS_NR];1094 1095	/* ERRORS */1096	struct list_head	fsck_error_msgs;1097	struct mutex		fsck_error_msgs_lock;1098	bool			fsck_alloc_msgs_err;1099 1100	bch_sb_errors_cpu	fsck_error_counts;1101	struct mutex		fsck_error_counts_lock;1102};1103 1104extern struct wait_queue_head bch2_read_only_wait;1105 1106static inline void bch2_write_ref_get(struct bch_fs *c, enum bch_write_ref ref)1107{1108#ifdef BCH_WRITE_REF_DEBUG1109	atomic_long_inc(&c->writes[ref]);1110#else1111	percpu_ref_get(&c->writes);1112#endif1113}1114 1115static inline bool __bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)1116{1117#ifdef BCH_WRITE_REF_DEBUG1118	return !test_bit(BCH_FS_going_ro, &c->flags) &&1119		atomic_long_inc_not_zero(&c->writes[ref]);1120#else1121	return percpu_ref_tryget(&c->writes);1122#endif1123}1124 1125static inline bool bch2_write_ref_tryget(struct bch_fs *c, enum bch_write_ref ref)1126{1127#ifdef BCH_WRITE_REF_DEBUG1128	return !test_bit(BCH_FS_going_ro, &c->flags) &&1129		atomic_long_inc_not_zero(&c->writes[ref]);1130#else1131	return percpu_ref_tryget_live(&c->writes);1132#endif1133}1134 1135static inline void bch2_write_ref_put(struct bch_fs *c, enum bch_write_ref ref)1136{1137#ifdef BCH_WRITE_REF_DEBUG1138	long v = atomic_long_dec_return(&c->writes[ref]);1139 1140	BUG_ON(v < 0);1141	if (v)1142		return;1143	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)1144		if (atomic_long_read(&c->writes[i]))1145			return;1146 1147	set_bit(BCH_FS_write_disable_complete, &c->flags);1148	wake_up(&bch2_read_only_wait);1149#else1150	percpu_ref_put(&c->writes);1151#endif1152}1153 1154static inline bool bch2_ro_ref_tryget(struct bch_fs *c)1155{1156	if (test_bit(BCH_FS_stopping, &c->flags))1157		return false;1158 1159	return refcount_inc_not_zero(&c->ro_ref);1160}1161 1162static inline void bch2_ro_ref_put(struct bch_fs *c)1163{1164	if (refcount_dec_and_test(&c->ro_ref))1165		wake_up(&c->ro_ref_wait);1166}1167 1168static inline void bch2_set_ra_pages(struct bch_fs *c, unsigned ra_pages)1169{1170#ifndef NO_BCACHEFS_FS1171	if (c->vfs_sb)1172		c->vfs_sb->s_bdi->ra_pages = ra_pages;1173#endif1174}1175 1176static inline unsigned bucket_bytes(const struct bch_dev *ca)1177{1178	return ca->mi.bucket_size << 9;1179}1180 1181static inline unsigned block_bytes(const struct bch_fs *c)1182{1183	return c->opts.block_size;1184}1185 1186static inline unsigned block_sectors(const struct bch_fs *c)1187{1188	return c->opts.block_size >> 9;1189}1190 1191static inline bool btree_id_cached(const struct bch_fs *c, enum btree_id btree)1192{1193	return c->btree_key_cache_btrees & (1U << btree);1194}1195 1196static inline struct timespec64 bch2_time_to_timespec(const struct bch_fs *c, s64 time)1197{1198	struct timespec64 t;1199	s64 sec;1200	s32 rem;1201 1202	time += c->sb.time_base_lo;1203 1204	sec = div_s64_rem(time, c->sb.time_units_per_sec, &rem);1205 1206	set_normalized_timespec64(&t, sec, rem * (s64)c->sb.nsec_per_time_unit);1207 1208	return t;1209}1210 1211static inline s64 timespec_to_bch2_time(const struct bch_fs *c, struct timespec64 ts)1212{1213	return (ts.tv_sec * c->sb.time_units_per_sec +1214		(int) ts.tv_nsec / c->sb.nsec_per_time_unit) - c->sb.time_base_lo;1215}1216 1217static inline s64 bch2_current_time(const struct bch_fs *c)1218{1219	struct timespec64 now;1220 1221	ktime_get_coarse_real_ts64(&now);1222	return timespec_to_bch2_time(c, now);1223}1224 1225static inline u64 bch2_current_io_time(const struct bch_fs *c, int rw)1226{1227	return max(1ULL, (u64) atomic64_read(&c->io_clock[rw].now) & LRU_TIME_MAX);1228}1229 1230static inline struct stdio_redirect *bch2_fs_stdio_redirect(struct bch_fs *c)1231{1232	struct stdio_redirect *stdio = c->stdio;1233 1234	if (c->stdio_filter && c->stdio_filter != current)1235		stdio = NULL;1236	return stdio;1237}1238 1239static inline unsigned metadata_replicas_required(struct bch_fs *c)1240{1241	return min(c->opts.metadata_replicas,1242		   c->opts.metadata_replicas_required);1243}1244 1245static inline unsigned data_replicas_required(struct bch_fs *c)1246{1247	return min(c->opts.data_replicas,1248		   c->opts.data_replicas_required);1249}1250 1251#define BKEY_PADDED_ONSTACK(key, pad)				\1252	struct { struct bkey_i key; __u64 key ## _pad[pad]; }1253 1254#endif /* _BCACHEFS_H */1255