bcache.h

   1/* SPDX-License-Identifier: GPL-2.0 */
   2#ifndef _BCACHE_H
   3#define _BCACHE_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#define pr_fmt(fmt) "bcache: %s() " fmt, __func__
 180
 181#include <linux/bio.h>
 182#include <linux/closure.h>
 183#include <linux/kobject.h>
 184#include <linux/list.h>
 185#include <linux/mutex.h>
 186#include <linux/rbtree.h>
 187#include <linux/rwsem.h>
 188#include <linux/refcount.h>
 189#include <linux/types.h>
 190#include <linux/workqueue.h>
 191#include <linux/kthread.h>
 192
 193#include "bcache_ondisk.h"
 194#include "bset.h"
 195#include "util.h"
 196
 197struct bucket {
 198	atomic_t	pin;
 199	uint16_t	prio;
 200	uint8_t		gen;
 201	uint8_t		last_gc; /* Most out of date gen in the btree */
 202	uint16_t	gc_mark; /* Bitfield used by GC. See below for field */
 203	uint16_t	reclaimable_in_gc:1;
 204};
 205
 206/*
 207 * I'd use bitfields for these, but I don't trust the compiler not to screw me
 208 * as multiple threads touch struct bucket without locking
 209 */
 210
 211BITMASK(GC_MARK,	 struct bucket, gc_mark, 0, 2);
 212#define GC_MARK_RECLAIMABLE	1
 213#define GC_MARK_DIRTY		2
 214#define GC_MARK_METADATA	3
 215#define GC_SECTORS_USED_SIZE	13
 216#define MAX_GC_SECTORS_USED	(~(~0ULL << GC_SECTORS_USED_SIZE))
 217BITMASK(GC_SECTORS_USED, struct bucket, gc_mark, 2, GC_SECTORS_USED_SIZE);
 218BITMASK(GC_MOVE, struct bucket, gc_mark, 15, 1);
 219
 220#include "journal.h"
 221#include "stats.h"
 222struct search;
 223struct btree;
 224struct keybuf;
 225
 226struct keybuf_key {
 227	struct rb_node		node;
 228	BKEY_PADDED(key);
 229	void			*private;
 230};
 231
 232struct keybuf {
 233	struct bkey		last_scanned;
 234	spinlock_t		lock;
 235
 236	/*
 237	 * Beginning and end of range in rb tree - so that we can skip taking
 238	 * lock and checking the rb tree when we need to check for overlapping
 239	 * keys.
 240	 */
 241	struct bkey		start;
 242	struct bkey		end;
 243
 244	struct rb_root		keys;
 245
 246#define KEYBUF_NR		500
 247	DECLARE_ARRAY_ALLOCATOR(struct keybuf_key, freelist, KEYBUF_NR);
 248};
 249
 250struct bcache_device {
 251	struct closure		cl;
 252
 253	struct kobject		kobj;
 254
 255	struct cache_set	*c;
 256	unsigned int		id;
 257#define BCACHEDEVNAME_SIZE	12
 258	char			name[BCACHEDEVNAME_SIZE];
 259
 260	struct gendisk		*disk;
 261
 262	unsigned long		flags;
 263#define BCACHE_DEV_CLOSING		0
 264#define BCACHE_DEV_DETACHING		1
 265#define BCACHE_DEV_UNLINK_DONE		2
 266#define BCACHE_DEV_WB_RUNNING		3
 267#define BCACHE_DEV_RATE_DW_RUNNING	4
 268	int			nr_stripes;
 269#define BCH_MIN_STRIPE_SZ		((4 << 20) >> SECTOR_SHIFT)
 270	unsigned int		stripe_size;
 271	atomic_t		*stripe_sectors_dirty;
 272	unsigned long		*full_dirty_stripes;
 273
 274	struct bio_set		bio_split;
 275
 276	struct bio_set		bio_detached;
 277
 278	unsigned int		data_csum:1;
 279
 280	int (*cache_miss)(struct btree *b, struct search *s,
 281			  struct bio *bio, unsigned int sectors);
 282	int (*ioctl)(struct bcache_device *d, blk_mode_t mode,
 283		     unsigned int cmd, unsigned long arg);
 284};
 285
 286struct io {
 287	/* Used to track sequential IO so it can be skipped */
 288	struct hlist_node	hash;
 289	struct list_head	lru;
 290
 291	unsigned long		jiffies;
 292	unsigned int		sequential;
 293	sector_t		last;
 294};
 295
 296enum stop_on_failure {
 297	BCH_CACHED_DEV_STOP_AUTO = 0,
 298	BCH_CACHED_DEV_STOP_ALWAYS,
 299	BCH_CACHED_DEV_STOP_MODE_MAX,
 300};
 301
 302struct cached_dev {
 303	struct list_head	list;
 304	struct bcache_device	disk;
 305	struct block_device	*bdev;
 306	struct file		*bdev_file;
 307
 308	struct cache_sb		sb;
 309	struct cache_sb_disk	*sb_disk;
 310	struct bio		sb_bio;
 311	struct bio_vec		sb_bv[1];
 312	struct closure		sb_write;
 313	struct semaphore	sb_write_mutex;
 314
 315	/* Refcount on the cache set. Always nonzero when we're caching. */
 316	refcount_t		count;
 317	struct work_struct	detach;
 318
 319	/*
 320	 * Device might not be running if it's dirty and the cache set hasn't
 321	 * showed up yet.
 322	 */
 323	atomic_t		running;
 324
 325	/*
 326	 * Writes take a shared lock from start to finish; scanning for dirty
 327	 * data to refill the rb tree requires an exclusive lock.
 328	 */
 329	struct rw_semaphore	writeback_lock;
 330
 331	/*
 332	 * Nonzero, and writeback has a refcount (d->count), iff there is dirty
 333	 * data in the cache. Protected by writeback_lock; must have an
 334	 * shared lock to set and exclusive lock to clear.
 335	 */
 336	atomic_t		has_dirty;
 337
 338#define BCH_CACHE_READA_ALL		0
 339#define BCH_CACHE_READA_META_ONLY	1
 340	unsigned int		cache_readahead_policy;
 341	struct bch_ratelimit	writeback_rate;
 342	struct delayed_work	writeback_rate_update;
 343
 344	/* Limit number of writeback bios in flight */
 345	struct semaphore	in_flight;
 346	struct task_struct	*writeback_thread;
 347	struct workqueue_struct	*writeback_write_wq;
 348
 349	struct keybuf		writeback_keys;
 350
 351	struct task_struct	*status_update_thread;
 352	/*
 353	 * Order the write-half of writeback operations strongly in dispatch
 354	 * order.  (Maintain LBA order; don't allow reads completing out of
 355	 * order to re-order the writes...)
 356	 */
 357	struct closure_waitlist writeback_ordering_wait;
 358	atomic_t		writeback_sequence_next;
 359
 360	/* For tracking sequential IO */
 361#define RECENT_IO_BITS	7
 362#define RECENT_IO	(1 << RECENT_IO_BITS)
 363	struct io		io[RECENT_IO];
 364	struct hlist_head	io_hash[RECENT_IO + 1];
 365	struct list_head	io_lru;
 366	spinlock_t		io_lock;
 367
 368	struct cache_accounting	accounting;
 369
 370	/* The rest of this all shows up in sysfs */
 371	unsigned int		sequential_cutoff;
 372
 373	unsigned int		io_disable:1;
 374	unsigned int		verify:1;
 375	unsigned int		bypass_torture_test:1;
 376
 377	unsigned int		partial_stripes_expensive:1;
 378	unsigned int		writeback_metadata:1;
 379	unsigned int		writeback_running:1;
 380	unsigned int		writeback_consider_fragment:1;
 381	unsigned char		writeback_percent;
 382	unsigned int		writeback_delay;
 383
 384	uint64_t		writeback_rate_target;
 385	int64_t			writeback_rate_proportional;
 386	int64_t			writeback_rate_integral;
 387	int64_t			writeback_rate_integral_scaled;
 388	int32_t			writeback_rate_change;
 389
 390	unsigned int		writeback_rate_update_seconds;
 391	unsigned int		writeback_rate_i_term_inverse;
 392	unsigned int		writeback_rate_p_term_inverse;
 393	unsigned int		writeback_rate_fp_term_low;
 394	unsigned int		writeback_rate_fp_term_mid;
 395	unsigned int		writeback_rate_fp_term_high;
 396	unsigned int		writeback_rate_minimum;
 397
 398	enum stop_on_failure	stop_when_cache_set_failed;
 399#define DEFAULT_CACHED_DEV_ERROR_LIMIT	64
 400	atomic_t		io_errors;
 401	unsigned int		error_limit;
 402	unsigned int		offline_seconds;
 403
 404	/*
 405	 * Retry to update writeback_rate if contention happens for
 406	 * down_read(dc->writeback_lock) in update_writeback_rate()
 407	 */
 408#define BCH_WBRATE_UPDATE_MAX_SKIPS	15
 409	unsigned int		rate_update_retry;
 410};
 411
 412enum alloc_reserve {
 413	RESERVE_BTREE,
 414	RESERVE_PRIO,
 415	RESERVE_MOVINGGC,
 416	RESERVE_NONE,
 417	RESERVE_NR,
 418};
 419
 420struct cache {
 421	struct cache_set	*set;
 422	struct cache_sb		sb;
 423	struct cache_sb_disk	*sb_disk;
 424	struct bio		sb_bio;
 425	struct bio_vec		sb_bv[1];
 426
 427	struct kobject		kobj;
 428	struct block_device	*bdev;
 429	struct file		*bdev_file;
 430
 431	struct task_struct	*alloc_thread;
 432
 433	struct closure		prio;
 434	struct prio_set		*disk_buckets;
 435
 436	/*
 437	 * When allocating new buckets, prio_write() gets first dibs - since we
 438	 * may not be allocate at all without writing priorities and gens.
 439	 * prio_last_buckets[] contains the last buckets we wrote priorities to
 440	 * (so gc can mark them as metadata), prio_buckets[] contains the
 441	 * buckets allocated for the next prio write.
 442	 */
 443	uint64_t		*prio_buckets;
 444	uint64_t		*prio_last_buckets;
 445
 446	/*
 447	 * free: Buckets that are ready to be used
 448	 *
 449	 * free_inc: Incoming buckets - these are buckets that currently have
 450	 * cached data in them, and we can't reuse them until after we write
 451	 * their new gen to disk. After prio_write() finishes writing the new
 452	 * gens/prios, they'll be moved to the free list.
 453	 */
 454	DECLARE_FIFO(long, free)[RESERVE_NR];
 455	DECLARE_FIFO(long, free_inc);
 456
 457	size_t			fifo_last_bucket;
 458
 459	/* Allocation stuff: */
 460	struct bucket		*buckets;
 461
 462	DECLARE_HEAP(struct bucket *, heap);
 463
 464	/*
 465	 * If nonzero, we know we aren't going to find any buckets to invalidate
 466	 * until a gc finishes - otherwise we could pointlessly burn a ton of
 467	 * cpu
 468	 */
 469	unsigned int		invalidate_needs_gc;
 470
 471	struct journal_device	journal;
 472
 473	/* The rest of this all shows up in sysfs */
 474#define IO_ERROR_SHIFT		20
 475	atomic_t		io_errors;
 476	atomic_t		io_count;
 477
 478	atomic_long_t		meta_sectors_written;
 479	atomic_long_t		btree_sectors_written;
 480	atomic_long_t		sectors_written;
 481};
 482
 483struct gc_stat {
 484	size_t			nodes;
 485	size_t			nodes_pre;
 486	size_t			key_bytes;
 487
 488	size_t			nkeys;
 489	uint64_t		data;	/* sectors */
 490	unsigned int		in_use; /* percent */
 491};
 492
 493/*
 494 * Flag bits, for how the cache set is shutting down, and what phase it's at:
 495 *
 496 * CACHE_SET_UNREGISTERING means we're not just shutting down, we're detaching
 497 * all the backing devices first (their cached data gets invalidated, and they
 498 * won't automatically reattach).
 499 *
 500 * CACHE_SET_STOPPING always gets set first when we're closing down a cache set;
 501 * we'll continue to run normally for awhile with CACHE_SET_STOPPING set (i.e.
 502 * flushing dirty data).
 503 *
 504 * CACHE_SET_RUNNING means all cache devices have been registered and journal
 505 * replay is complete.
 506 *
 507 * CACHE_SET_IO_DISABLE is set when bcache is stopping the whold cache set, all
 508 * external and internal I/O should be denied when this flag is set.
 509 *
 510 */
 511#define CACHE_SET_UNREGISTERING		0
 512#define	CACHE_SET_STOPPING		1
 513#define	CACHE_SET_RUNNING		2
 514#define CACHE_SET_IO_DISABLE		3
 515
 516struct cache_set {
 517	struct closure		cl;
 518
 519	struct list_head	list;
 520	struct kobject		kobj;
 521	struct kobject		internal;
 522	struct dentry		*debug;
 523	struct cache_accounting accounting;
 524
 525	unsigned long		flags;
 526	atomic_t		idle_counter;
 527	atomic_t		at_max_writeback_rate;
 528
 529	struct cache		*cache;
 530
 531	struct bcache_device	**devices;
 532	unsigned int		devices_max_used;
 533	atomic_t		attached_dev_nr;
 534	struct list_head	cached_devs;
 535	uint64_t		cached_dev_sectors;
 536	atomic_long_t		flash_dev_dirty_sectors;
 537	struct closure		caching;
 538
 539	struct closure		sb_write;
 540	struct semaphore	sb_write_mutex;
 541
 542	mempool_t		search;
 543	mempool_t		bio_meta;
 544	struct bio_set		bio_split;
 545
 546	/* For the btree cache */
 547	struct shrinker		*shrink;
 548
 549	/* For the btree cache and anything allocation related */
 550	struct mutex		bucket_lock;
 551
 552	/* log2(bucket_size), in sectors */
 553	unsigned short		bucket_bits;
 554
 555	/* log2(block_size), in sectors */
 556	unsigned short		block_bits;
 557
 558	/*
 559	 * Default number of pages for a new btree node - may be less than a
 560	 * full bucket
 561	 */
 562	unsigned int		btree_pages;
 563
 564	/*
 565	 * Lists of struct btrees; lru is the list for structs that have memory
 566	 * allocated for actual btree node, freed is for structs that do not.
 567	 *
 568	 * We never free a struct btree, except on shutdown - we just put it on
 569	 * the btree_cache_freed list and reuse it later. This simplifies the
 570	 * code, and it doesn't cost us much memory as the memory usage is
 571	 * dominated by buffers that hold the actual btree node data and those
 572	 * can be freed - and the number of struct btrees allocated is
 573	 * effectively bounded.
 574	 *
 575	 * btree_cache_freeable effectively is a small cache - we use it because
 576	 * high order page allocations can be rather expensive, and it's quite
 577	 * common to delete and allocate btree nodes in quick succession. It
 578	 * should never grow past ~2-3 nodes in practice.
 579	 */
 580	struct list_head	btree_cache;
 581	struct list_head	btree_cache_freeable;
 582	struct list_head	btree_cache_freed;
 583
 584	/* Number of elements in btree_cache + btree_cache_freeable lists */
 585	unsigned int		btree_cache_used;
 586
 587	/*
 588	 * If we need to allocate memory for a new btree node and that
 589	 * allocation fails, we can cannibalize another node in the btree cache
 590	 * to satisfy the allocation - lock to guarantee only one thread does
 591	 * this at a time:
 592	 */
 593	wait_queue_head_t	btree_cache_wait;
 594	struct task_struct	*btree_cache_alloc_lock;
 595	spinlock_t		btree_cannibalize_lock;
 596
 597	/*
 598	 * When we free a btree node, we increment the gen of the bucket the
 599	 * node is in - but we can't rewrite the prios and gens until we
 600	 * finished whatever it is we were doing, otherwise after a crash the
 601	 * btree node would be freed but for say a split, we might not have the
 602	 * pointers to the new nodes inserted into the btree yet.
 603	 *
 604	 * This is a refcount that blocks prio_write() until the new keys are
 605	 * written.
 606	 */
 607	atomic_t		prio_blocked;
 608	wait_queue_head_t	bucket_wait;
 609	atomic_t		bucket_wait_cnt;
 610
 611	/*
 612	 * For any bio we don't skip we subtract the number of sectors from
 613	 * rescale; when it hits 0 we rescale all the bucket priorities.
 614	 */
 615	atomic_t		rescale;
 616	/*
 617	 * used for GC, identify if any front side I/Os is inflight
 618	 */
 619	atomic_t		search_inflight;
 620	/*
 621	 * When we invalidate buckets, we use both the priority and the amount
 622	 * of good data to determine which buckets to reuse first - to weight
 623	 * those together consistently we keep track of the smallest nonzero
 624	 * priority of any bucket.
 625	 */
 626	uint16_t		min_prio;
 627
 628	/*
 629	 * max(gen - last_gc) for all buckets. When it gets too big we have to
 630	 * gc to keep gens from wrapping around.
 631	 */
 632	uint8_t			need_gc;
 633	struct gc_stat		gc_stats;
 634	size_t			nbuckets;
 635	size_t			avail_nbuckets;
 636
 637	struct task_struct	*gc_thread;
 638	/* Where in the btree gc currently is */
 639	struct bkey		gc_done;
 640
 641	/*
 642	 * For automatical garbage collection after writeback completed, this
 643	 * varialbe is used as bit fields,
 644	 * - 0000 0001b (BCH_ENABLE_AUTO_GC): enable gc after writeback
 645	 * - 0000 0010b (BCH_DO_AUTO_GC):     do gc after writeback
 646	 * This is an optimization for following write request after writeback
 647	 * finished, but read hit rate dropped due to clean data on cache is
 648	 * discarded. Unless user explicitly sets it via sysfs, it won't be
 649	 * enabled.
 650	 */
 651#define BCH_ENABLE_AUTO_GC	1
 652#define BCH_DO_AUTO_GC		2
 653	uint8_t			gc_after_writeback;
 654
 655	/*
 656	 * The allocation code needs gc_mark in struct bucket to be correct, but
 657	 * it's not while a gc is in progress. Protected by bucket_lock.
 658	 */
 659	int			gc_mark_valid;
 660
 661	/* Counts how many sectors bio_insert has added to the cache */
 662	atomic_t		sectors_to_gc;
 663	wait_queue_head_t	gc_wait;
 664
 665	struct keybuf		moving_gc_keys;
 666	/* Number of moving GC bios in flight */
 667	struct semaphore	moving_in_flight;
 668
 669	struct workqueue_struct	*moving_gc_wq;
 670
 671	struct btree		*root;
 672
 673#ifdef CONFIG_BCACHE_DEBUG
 674	struct btree		*verify_data;
 675	struct bset		*verify_ondisk;
 676	struct mutex		verify_lock;
 677#endif
 678
 679	uint8_t			set_uuid[16];
 680	unsigned int		nr_uuids;
 681	struct uuid_entry	*uuids;
 682	BKEY_PADDED(uuid_bucket);
 683	struct closure		uuid_write;
 684	struct semaphore	uuid_write_mutex;
 685
 686	/*
 687	 * A btree node on disk could have too many bsets for an iterator to fit
 688	 * on the stack - have to dynamically allocate them.
 689	 * bch_cache_set_alloc() will make sure the pool can allocate iterators
 690	 * equipped with enough room that can host
 691	 *     (sb.bucket_size / sb.block_size)
 692	 * btree_iter_sets, which is more than static MAX_BSETS.
 693	 */
 694	mempool_t		fill_iter;
 695
 696	struct bset_sort_state	sort;
 697
 698	/* List of buckets we're currently writing data to */
 699	struct list_head	data_buckets;
 700	spinlock_t		data_bucket_lock;
 701
 702	struct journal		journal;
 703
 704#define CONGESTED_MAX		1024
 705	unsigned int		congested_last_us;
 706	atomic_t		congested;
 707
 708	/* The rest of this all shows up in sysfs */
 709	unsigned int		congested_read_threshold_us;
 710	unsigned int		congested_write_threshold_us;
 711
 712	struct time_stats	btree_gc_time;
 713	struct time_stats	btree_split_time;
 714	struct time_stats	btree_read_time;
 715
 716	atomic_long_t		cache_read_races;
 717	atomic_long_t		writeback_keys_done;
 718	atomic_long_t		writeback_keys_failed;
 719
 720	atomic_long_t		reclaim;
 721	atomic_long_t		reclaimed_journal_buckets;
 722	atomic_long_t		flush_write;
 723
 724	enum			{
 725		ON_ERROR_UNREGISTER,
 726		ON_ERROR_PANIC,
 727	}			on_error;
 728#define DEFAULT_IO_ERROR_LIMIT 8
 729	unsigned int		error_limit;
 730	unsigned int		error_decay;
 731
 732	unsigned short		journal_delay_ms;
 733	bool			expensive_debug_checks;
 734	unsigned int		verify:1;
 735	unsigned int		key_merging_disabled:1;
 736	unsigned int		gc_always_rewrite:1;
 737	unsigned int		shrinker_disabled:1;
 738	unsigned int		copy_gc_enabled:1;
 739	unsigned int		idle_max_writeback_rate_enabled:1;
 740
 741#define BUCKET_HASH_BITS	12
 742	struct hlist_head	bucket_hash[1 << BUCKET_HASH_BITS];
 743};
 744
 745struct bbio {
 746	unsigned int		submit_time_us;
 747	union {
 748		struct bkey	key;
 749		uint64_t	_pad[3];
 750		/*
 751		 * We only need pad = 3 here because we only ever carry around a
 752		 * single pointer - i.e. the pointer we're doing io to/from.
 753		 */
 754	};
 755	struct bio		bio;
 756};
 757
 758struct detached_dev_io_private {
 759	struct bcache_device	*d;
 760	unsigned long		start_time;
 761	struct bio		*orig_bio;
 762	struct bio              bio;
 763};
 764
 765#define BTREE_PRIO		USHRT_MAX
 766#define INITIAL_PRIO		32768U
 767
 768#define btree_bytes(c)		((c)->btree_pages * PAGE_SIZE)
 769#define btree_blocks(b)							\
 770	((unsigned int) (KEY_SIZE(&b->key) >> (b)->c->block_bits))
 771
 772#define btree_default_blocks(c)						\
 773	((unsigned int) ((PAGE_SECTORS * (c)->btree_pages) >> (c)->block_bits))
 774
 775#define bucket_bytes(ca)	((ca)->sb.bucket_size << 9)
 776#define block_bytes(ca)		((ca)->sb.block_size << 9)
 777
 778static inline unsigned int meta_bucket_pages(struct cache_sb *sb)
 779{
 780	unsigned int n, max_pages;
 781
 782	max_pages = min_t(unsigned int,
 783			  __rounddown_pow_of_two(USHRT_MAX) / PAGE_SECTORS,
 784			  MAX_ORDER_NR_PAGES);
 785
 786	n = sb->bucket_size / PAGE_SECTORS;
 787	if (n > max_pages)
 788		n = max_pages;
 789
 790	return n;
 791}
 792
 793static inline unsigned int meta_bucket_bytes(struct cache_sb *sb)
 794{
 795	return meta_bucket_pages(sb) << PAGE_SHIFT;
 796}
 797
 798#define prios_per_bucket(ca)						\
 799	((meta_bucket_bytes(&(ca)->sb) - sizeof(struct prio_set)) /	\
 800	 sizeof(struct bucket_disk))
 801
 802#define prio_buckets(ca)						\
 803	DIV_ROUND_UP((size_t) (ca)->sb.nbuckets, prios_per_bucket(ca))
 804
 805static inline size_t sector_to_bucket(struct cache_set *c, sector_t s)
 806{
 807	return s >> c->bucket_bits;
 808}
 809
 810static inline sector_t bucket_to_sector(struct cache_set *c, size_t b)
 811{
 812	return ((sector_t) b) << c->bucket_bits;
 813}
 814
 815static inline sector_t bucket_remainder(struct cache_set *c, sector_t s)
 816{
 817	return s & (c->cache->sb.bucket_size - 1);
 818}
 819
 820static inline size_t PTR_BUCKET_NR(struct cache_set *c,
 821				   const struct bkey *k,
 822				   unsigned int ptr)
 823{
 824	return sector_to_bucket(c, PTR_OFFSET(k, ptr));
 825}
 826
 827static inline struct bucket *PTR_BUCKET(struct cache_set *c,
 828					const struct bkey *k,
 829					unsigned int ptr)
 830{
 831	return c->cache->buckets + PTR_BUCKET_NR(c, k, ptr);
 832}
 833
 834static inline uint8_t gen_after(uint8_t a, uint8_t b)
 835{
 836	uint8_t r = a - b;
 837
 838	return r > 128U ? 0 : r;
 839}
 840
 841static inline uint8_t ptr_stale(struct cache_set *c, const struct bkey *k,
 842				unsigned int i)
 843{
 844	return gen_after(PTR_BUCKET(c, k, i)->gen, PTR_GEN(k, i));
 845}
 846
 847static inline bool ptr_available(struct cache_set *c, const struct bkey *k,
 848				 unsigned int i)
 849{
 850	return (PTR_DEV(k, i) < MAX_CACHES_PER_SET) && c->cache;
 851}
 852
 853/* Btree key macros */
 854
 855/*
 856 * This is used for various on disk data structures - cache_sb, prio_set, bset,
 857 * jset: The checksum is _always_ the first 8 bytes of these structs
 858 */
 859#define csum_set(i)							\
 860	bch_crc64(((void *) (i)) + sizeof(uint64_t),			\
 861		  ((void *) bset_bkey_last(i)) -			\
 862		  (((void *) (i)) + sizeof(uint64_t)))
 863
 864/* Error handling macros */
 865
 866#define btree_bug(b, ...)						\
 867do {									\
 868	if (bch_cache_set_error((b)->c, __VA_ARGS__))			\
 869		dump_stack();						\
 870} while (0)
 871
 872#define cache_bug(c, ...)						\
 873do {									\
 874	if (bch_cache_set_error(c, __VA_ARGS__))			\
 875		dump_stack();						\
 876} while (0)
 877
 878#define btree_bug_on(cond, b, ...)					\
 879do {									\
 880	if (cond)							\
 881		btree_bug(b, __VA_ARGS__);				\
 882} while (0)
 883
 884#define cache_bug_on(cond, c, ...)					\
 885do {									\
 886	if (cond)							\
 887		cache_bug(c, __VA_ARGS__);				\
 888} while (0)
 889
 890#define cache_set_err_on(cond, c, ...)					\
 891do {									\
 892	if (cond)							\
 893		bch_cache_set_error(c, __VA_ARGS__);			\
 894} while (0)
 895
 896/* Looping macros */
 897
 898#define for_each_bucket(b, ca)						\
 899	for (b = (ca)->buckets + (ca)->sb.first_bucket;			\
 900	     b < (ca)->buckets + (ca)->sb.nbuckets; b++)
 901
 902static inline void cached_dev_put(struct cached_dev *dc)
 903{
 904	if (refcount_dec_and_test(&dc->count))
 905		schedule_work(&dc->detach);
 906}
 907
 908static inline bool cached_dev_get(struct cached_dev *dc)
 909{
 910	if (!refcount_inc_not_zero(&dc->count))
 911		return false;
 912
 913	/* Paired with the mb in cached_dev_attach */
 914	smp_mb__after_atomic();
 915	return true;
 916}
 917
 918/*
 919 * bucket_gc_gen() returns the difference between the bucket's current gen and
 920 * the oldest gen of any pointer into that bucket in the btree (last_gc).
 921 */
 922
 923static inline uint8_t bucket_gc_gen(struct bucket *b)
 924{
 925	return b->gen - b->last_gc;
 926}
 927
 928#define BUCKET_GC_GEN_MAX	96U
 929
 930#define kobj_attribute_write(n, fn)					\
 931	static struct kobj_attribute ksysfs_##n = __ATTR(n, 0200, NULL, fn)
 932
 933#define kobj_attribute_rw(n, show, store)				\
 934	static struct kobj_attribute ksysfs_##n =			\
 935		__ATTR(n, 0600, show, store)
 936
 937static inline void wake_up_allocators(struct cache_set *c)
 938{
 939	struct cache *ca = c->cache;
 940
 941	wake_up_process(ca->alloc_thread);
 942}
 943
 944static inline void closure_bio_submit(struct cache_set *c,
 945				      struct bio *bio,
 946				      struct closure *cl)
 947{
 948	closure_get(cl);
 949	if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags))) {
 950		bio->bi_status = BLK_STS_IOERR;
 951		bio_endio(bio);
 952		return;
 953	}
 954	submit_bio_noacct(bio);
 955}
 956
 957/*
 958 * Prevent the kthread exits directly, and make sure when kthread_stop()
 959 * is called to stop a kthread, it is still alive. If a kthread might be
 960 * stopped by CACHE_SET_IO_DISABLE bit set, wait_for_kthread_stop() is
 961 * necessary before the kthread returns.
 962 */
 963static inline void wait_for_kthread_stop(void)
 964{
 965	while (!kthread_should_stop()) {
 966		set_current_state(TASK_INTERRUPTIBLE);
 967		schedule();
 968	}
 969}
 970
 971/* Forward declarations */
 972
 973void bch_count_backing_io_errors(struct cached_dev *dc, struct bio *bio);
 974void bch_count_io_errors(struct cache *ca, blk_status_t error,
 975			 int is_read, const char *m);
 976void bch_bbio_count_io_errors(struct cache_set *c, struct bio *bio,
 977			      blk_status_t error, const char *m);
 978void bch_bbio_endio(struct cache_set *c, struct bio *bio,
 979		    blk_status_t error, const char *m);
 980void bch_bbio_free(struct bio *bio, struct cache_set *c);
 981struct bio *bch_bbio_alloc(struct cache_set *c);
 982
 983void __bch_submit_bbio(struct bio *bio, struct cache_set *c);
 984void bch_submit_bbio(struct bio *bio, struct cache_set *c,
 985		     struct bkey *k, unsigned int ptr);
 986
 987uint8_t bch_inc_gen(struct cache *ca, struct bucket *b);
 988void bch_rescale_priorities(struct cache_set *c, int sectors);
 989
 990bool bch_can_invalidate_bucket(struct cache *ca, struct bucket *b);
 991void __bch_invalidate_one_bucket(struct cache *ca, struct bucket *b);
 992
 993void __bch_bucket_free(struct cache *ca, struct bucket *b);
 994void bch_bucket_free(struct cache_set *c, struct bkey *k);
 995
 996long bch_bucket_alloc(struct cache *ca, unsigned int reserve, bool wait);
 997int __bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
 998			   struct bkey *k, bool wait);
 999int bch_bucket_alloc_set(struct cache_set *c, unsigned int reserve,
1000			 struct bkey *k, bool wait);
1001bool bch_alloc_sectors(struct cache_set *c, struct bkey *k,
1002		       unsigned int sectors, unsigned int write_point,
1003		       unsigned int write_prio, bool wait);
1004bool bch_cached_dev_error(struct cached_dev *dc);
1005
1006__printf(2, 3)
1007bool bch_cache_set_error(struct cache_set *c, const char *fmt, ...);
1008
1009int bch_prio_write(struct cache *ca, bool wait);
1010void bch_write_bdev_super(struct cached_dev *dc, struct closure *parent);
1011
1012extern struct workqueue_struct *bcache_wq;
1013extern struct workqueue_struct *bch_journal_wq;
1014extern struct workqueue_struct *bch_flush_wq;
1015extern struct mutex bch_register_lock;
1016extern struct list_head bch_cache_sets;
1017
1018extern const struct kobj_type bch_cached_dev_ktype;
1019extern const struct kobj_type bch_flash_dev_ktype;
1020extern const struct kobj_type bch_cache_set_ktype;
1021extern const struct kobj_type bch_cache_set_internal_ktype;
1022extern const struct kobj_type bch_cache_ktype;
1023
1024void bch_cached_dev_release(struct kobject *kobj);
1025void bch_flash_dev_release(struct kobject *kobj);
1026void bch_cache_set_release(struct kobject *kobj);
1027void bch_cache_release(struct kobject *kobj);
1028
1029int bch_uuid_write(struct cache_set *c);
1030void bcache_write_super(struct cache_set *c);
1031
1032int bch_flash_dev_create(struct cache_set *c, uint64_t size);
1033
1034int bch_cached_dev_attach(struct cached_dev *dc, struct cache_set *c,
1035			  uint8_t *set_uuid);
1036void bch_cached_dev_detach(struct cached_dev *dc);
1037int bch_cached_dev_run(struct cached_dev *dc);
1038void bcache_device_stop(struct bcache_device *d);
1039
1040void bch_cache_set_unregister(struct cache_set *c);
1041void bch_cache_set_stop(struct cache_set *c);
1042
1043struct cache_set *bch_cache_set_alloc(struct cache_sb *sb);
1044void bch_btree_cache_free(struct cache_set *c);
1045int bch_btree_cache_alloc(struct cache_set *c);
1046void bch_moving_init_cache_set(struct cache_set *c);
1047int bch_open_buckets_alloc(struct cache_set *c);
1048void bch_open_buckets_free(struct cache_set *c);
1049
1050int bch_cache_allocator_start(struct cache *ca);
1051
1052void bch_debug_exit(void);
1053void bch_debug_init(void);
1054void bch_request_exit(void);
1055int bch_request_init(void);
1056void bch_btree_exit(void);
1057int bch_btree_init(void);
1058
1059#endif /* _BCACHE_H */