diff options
| author | Kent Overstreet <kent.overstreet@gmail.com> | 2020-06-29 19:24:13 -0400 |
|---|---|---|
| committer | Kent Overstreet <kent.overstreet@gmail.com> | 2020-06-29 19:25:57 -0400 |
| commit | 217eb889b5c4ace879083617690d4b3d615de073 (patch) | |
| tree | 022151c655efd43e631e73a8f4e69ba7dc13a6ab | |
| parent | ce0522135ae04886e05b24bb48cc163f33bd8206 (diff) | |
Merge with b7020ae929 bcachefs: Refactor dio write code to reinit bch_write_op
72 files changed, 2998 insertions, 2181 deletions
diff --git a/fs/bcachefs/Makefile b/fs/bcachefs/Makefile index c7727d05cf49..d85ced62c0dd 100644 --- a/fs/bcachefs/Makefile +++ b/fs/bcachefs/Makefile @@ -13,6 +13,7 @@ bcachefs-y := \ btree_gc.o \ btree_io.o \ btree_iter.o \ + btree_key_cache.o \ btree_update_interior.o \ btree_update_leaf.o \ buckets.o \ diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c index c37945189c68..cb720ee04b86 100644 --- a/fs/bcachefs/alloc_background.c +++ b/fs/bcachefs/alloc_background.c @@ -4,6 +4,7 @@ #include "alloc_foreground.h" #include "btree_cache.h" #include "btree_io.h" +#include "btree_key_cache.h" #include "btree_update.h" #include "btree_update_interior.h" #include "btree_gc.h" @@ -208,29 +209,25 @@ void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, get_alloc_field(a.v, &d, i)); } -int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys) +static int bch2_alloc_read_fn(struct bch_fs *c, enum btree_id id, + unsigned level, struct bkey_s_c k) { - struct btree_trans trans; - struct btree_and_journal_iter iter; - struct bkey_s_c k; - struct bch_dev *ca; - unsigned i; - int ret = 0; - - bch2_trans_init(&trans, c, 0, 0); - - bch2_btree_and_journal_iter_init(&iter, &trans, journal_keys, - BTREE_ID_ALLOC, POS_MIN); - - while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { + if (!level) bch2_mark_key(c, k, 0, 0, NULL, 0, BTREE_TRIGGER_ALLOC_READ| BTREE_TRIGGER_NOATOMIC); - bch2_btree_and_journal_iter_advance(&iter); - } + return 0; +} - ret = bch2_trans_exit(&trans) ?: ret; +int bch2_alloc_read(struct bch_fs *c, struct journal_keys *journal_keys) +{ + struct bch_dev *ca; + unsigned i; + int ret = 0; + + ret = bch2_btree_and_journal_walk(c, journal_keys, BTREE_ID_ALLOC, + NULL, bch2_alloc_read_fn); if (ret) { bch_err(c, "error reading alloc info: %i", ret); return ret; @@ -280,6 +277,13 @@ static int bch2_alloc_write_key(struct btree_trans *trans, struct bkey_i_alloc *a; int ret; retry: + bch2_trans_begin(trans); + + ret = bch2_btree_key_cache_flush(trans, + BTREE_ID_ALLOC, iter->pos); + if (ret) + goto err; + k = bch2_btree_iter_peek_slot(iter); ret = bkey_err(k); if (ret) @@ -334,7 +338,7 @@ int bch2_alloc_write(struct bch_fs *c, unsigned flags, bool *wrote) BUG_ON(BKEY_ALLOC_VAL_U64s_MAX > 8); - bch2_trans_init(&trans, c, 0, 0); + bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0); iter = bch2_trans_get_iter(&trans, BTREE_ID_ALLOC, POS_MIN, BTREE_ITER_SLOTS|BTREE_ITER_INTENT); @@ -368,25 +372,6 @@ int bch2_alloc_write(struct bch_fs *c, unsigned flags, bool *wrote) return ret < 0 ? ret : 0; } -int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k) -{ - struct btree_trans trans; - struct btree_iter *iter; - int ret; - - bch2_trans_init(&trans, c, 0, 0); - - iter = bch2_trans_get_iter(&trans, BTREE_ID_ALLOC, k->k.p, - BTREE_ITER_SLOTS|BTREE_ITER_INTENT); - - ret = bch2_alloc_write_key(&trans, iter, - BTREE_INSERT_NOFAIL| - BTREE_INSERT_LAZY_RW| - BTREE_INSERT_JOURNAL_REPLAY); - bch2_trans_exit(&trans); - return ret < 0 ? ret : 0; -} - /* Bucket IO clocks: */ static void bch2_recalc_oldest_io(struct bch_fs *c, struct bch_dev *ca, int rw) @@ -516,6 +501,7 @@ static void bch2_bucket_clock_init(struct bch_fs *c, int rw) static int wait_buckets_available(struct bch_fs *c, struct bch_dev *ca) { unsigned long gc_count = c->gc_count; + u64 available; int ret = 0; ca->allocator_state = ALLOCATOR_BLOCKED; @@ -531,9 +517,11 @@ static int wait_buckets_available(struct bch_fs *c, struct bch_dev *ca) if (gc_count != c->gc_count) ca->inc_gen_really_needs_gc = 0; - if ((ssize_t) (dev_buckets_available(c, ca) - - ca->inc_gen_really_needs_gc) >= - (ssize_t) fifo_free(&ca->free_inc)) + available = max_t(s64, 0, dev_buckets_available(c, ca) - + ca->inc_gen_really_needs_gc); + + if (available > fifo_free(&ca->free_inc) || + (available && !fifo_full(&ca->free[RESERVE_BTREE]))) break; up_read(&c->gc_lock); @@ -844,10 +832,9 @@ static int bch2_invalidate_one_bucket2(struct btree_trans *trans, struct bkey_alloc_unpacked u; struct bucket *g; struct bucket_mark m; - struct bkey_s_c k; bool invalidating_cached_data; size_t b; - int ret; + int ret = 0; BUG_ON(!ca->alloc_heap.used || !ca->alloc_heap.data[0].nr); @@ -861,32 +848,53 @@ static int bch2_invalidate_one_bucket2(struct btree_trans *trans, BUG_ON(!fifo_push(&ca->free_inc, b)); - bch2_mark_alloc_bucket(c, ca, b, true, gc_pos_alloc(c, NULL), 0); + g = bucket(ca, b); + m = READ_ONCE(g->mark); + + invalidating_cached_data = m.cached_sectors != 0; + + /* + * If we're not invalidating cached data, we only increment the bucket + * gen in memory here, the incremented gen will be updated in the btree + * by bch2_trans_mark_pointer(): + */ + + if (!invalidating_cached_data) + bch2_invalidate_bucket(c, ca, b, &m); + else + bch2_mark_alloc_bucket(c, ca, b, true, gc_pos_alloc(c, NULL), 0); spin_unlock(&c->freelist_lock); percpu_up_read(&c->mark_lock); + if (!invalidating_cached_data) + goto out; + + /* + * If the read-only path is trying to shut down, we can't be generating + * new btree updates: + */ + if (test_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags)) { + ret = 1; + goto out; + } + BUG_ON(BKEY_ALLOC_VAL_U64s_MAX > 8); bch2_btree_iter_set_pos(iter, POS(ca->dev_idx, b)); retry: - k = bch2_btree_iter_peek_slot(iter); - ret = bkey_err(k); + ret = bch2_btree_iter_traverse(iter); if (ret) return ret; - /* - * The allocator has to start before journal replay is finished - thus, - * we have to trust the in memory bucket @m, not the version in the - * btree: - */ percpu_down_read(&c->mark_lock); - g = bucket(ca, b); + g = bucket(ca, iter->pos.offset); m = READ_ONCE(g->mark); u = alloc_mem_to_key(g, m); + percpu_up_read(&c->mark_lock); - invalidating_cached_data = m.cached_sectors != 0; + invalidating_cached_data = u.cached_sectors != 0; u.gen++; u.data_type = 0; @@ -919,7 +927,7 @@ retry: flags); if (ret == -EINTR) goto retry; - +out: if (!ret) { /* remove from alloc_heap: */ struct alloc_heap_entry e, *top = ca->alloc_heap.data; @@ -953,32 +961,7 @@ retry: percpu_up_read(&c->mark_lock); } - return ret; -} - -static bool bch2_invalidate_one_bucket(struct bch_fs *c, struct bch_dev *ca, - size_t bucket, u64 *flush_seq) -{ - struct bucket_mark m; - - percpu_down_read(&c->mark_lock); - spin_lock(&c->freelist_lock); - - bch2_invalidate_bucket(c, ca, bucket, &m); - - verify_not_on_freelist(c, ca, bucket); - BUG_ON(!fifo_push(&ca->free_inc, bucket)); - - spin_unlock(&c->freelist_lock); - - bucket_io_clock_reset(c, ca, bucket, READ); - bucket_io_clock_reset(c, ca, bucket, WRITE); - - percpu_up_read(&c->mark_lock); - - *flush_seq = max(*flush_seq, bucket_journal_seq(c, m)); - - return m.cached_sectors != 0; + return ret < 0 ? ret : 0; } /* @@ -995,7 +978,9 @@ static int bch2_invalidate_buckets(struct bch_fs *c, struct bch_dev *ca) iter = bch2_trans_get_iter(&trans, BTREE_ID_ALLOC, POS(ca->dev_idx, 0), - BTREE_ITER_SLOTS|BTREE_ITER_INTENT); + BTREE_ITER_CACHED| + BTREE_ITER_CACHED_NOFILL| + BTREE_ITER_INTENT); /* Only use nowait if we've already invalidated at least one bucket: */ while (!ret && @@ -1119,6 +1104,8 @@ static int bch2_allocator_thread(void *arg) while (1) { cond_resched(); + if (kthread_should_stop()) + break; pr_debug("discarding %zu invalidated buckets", fifo_used(&ca->free_inc)); @@ -1436,221 +1423,6 @@ int bch2_dev_allocator_start(struct bch_dev *ca) return 0; } -static bool flush_held_btree_writes(struct bch_fs *c) -{ - struct bucket_table *tbl; - struct rhash_head *pos; - struct btree *b; - bool nodes_unwritten; - size_t i; -again: - cond_resched(); - nodes_unwritten = false; - - if (bch2_journal_error(&c->journal)) - return true; - - rcu_read_lock(); - for_each_cached_btree(b, c, tbl, i, pos) - if (btree_node_need_write(b)) { - if (btree_node_may_write(b)) { - rcu_read_unlock(); - btree_node_lock_type(c, b, SIX_LOCK_read); - bch2_btree_node_write(c, b, SIX_LOCK_read); - six_unlock_read(&b->lock); - goto again; - } else { - nodes_unwritten = true; - } - } - rcu_read_unlock(); - - if (c->btree_roots_dirty) { - bch2_journal_meta(&c->journal); - goto again; - } - - return !nodes_unwritten && - !bch2_btree_interior_updates_nr_pending(c); -} - -static void allocator_start_issue_discards(struct bch_fs *c) -{ - struct bch_dev *ca; - unsigned dev_iter; - size_t bu; - - for_each_rw_member(ca, c, dev_iter) - while (fifo_pop(&ca->free_inc, bu)) - blkdev_issue_discard(ca->disk_sb.bdev, - bucket_to_sector(ca, bu), - ca->mi.bucket_size, GFP_NOIO, 0); -} - -static int resize_free_inc(struct bch_dev *ca) -{ - alloc_fifo free_inc; - - if (!fifo_full(&ca->free_inc)) - return 0; - - if (!init_fifo(&free_inc, - ca->free_inc.size * 2, - GFP_KERNEL)) - return -ENOMEM; - - fifo_move(&free_inc, &ca->free_inc); - swap(free_inc, ca->free_inc); - free_fifo(&free_inc); - return 0; -} - -static bool bch2_fs_allocator_start_fast(struct bch_fs *c) -{ - struct bch_dev *ca; - unsigned dev_iter; - bool ret = true; - - if (test_alloc_startup(c)) - return false; - - down_read(&c->gc_lock); - - /* Scan for buckets that are already invalidated: */ - for_each_rw_member(ca, c, dev_iter) { - struct bucket_array *buckets; - struct bucket_mark m; - long bu; - - down_read(&ca->bucket_lock); - buckets = bucket_array(ca); - - for (bu = buckets->first_bucket; - bu < buckets->nbuckets; bu++) { - m = READ_ONCE(buckets->b[bu].mark); - - if (!buckets->b[bu].gen_valid || - !is_available_bucket(m) || - m.cached_sectors || - (ca->buckets_nouse && - test_bit(bu, ca->buckets_nouse))) - continue; - - percpu_down_read(&c->mark_lock); - bch2_mark_alloc_bucket(c, ca, bu, true, - gc_pos_alloc(c, NULL), 0); - percpu_up_read(&c->mark_lock); - - fifo_push(&ca->free_inc, bu); - - discard_invalidated_buckets(c, ca); - - if (fifo_full(&ca->free[RESERVE_BTREE])) - break; - } - up_read(&ca->bucket_lock); - } - - up_read(&c->gc_lock); - - /* did we find enough buckets? */ - for_each_rw_member(ca, c, dev_iter) - if (!fifo_full(&ca->free[RESERVE_BTREE])) - ret = false; - - return ret; -} - -int bch2_fs_allocator_start(struct bch_fs *c) -{ - struct bch_dev *ca; - unsigned dev_iter; - u64 journal_seq = 0; - bool wrote; - long bu; - int ret = 0; - - if (!test_alloc_startup(c) && - bch2_fs_allocator_start_fast(c)) - return 0; - - pr_debug("not enough empty buckets; scanning for reclaimable buckets"); - - /* - * We're moving buckets to freelists _before_ they've been marked as - * invalidated on disk - we have to so that we can allocate new btree - * nodes to mark them as invalidated on disk. - * - * However, we can't _write_ to any of these buckets yet - they might - * have cached data in them, which is live until they're marked as - * invalidated on disk: - */ - set_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags); - - down_read(&c->gc_lock); - do { - wrote = false; - - for_each_rw_member(ca, c, dev_iter) { - find_reclaimable_buckets(c, ca); - - while (!fifo_full(&ca->free[RESERVE_BTREE]) && - (bu = next_alloc_bucket(ca)) >= 0) { - ret = resize_free_inc(ca); - if (ret) { - percpu_ref_put(&ca->io_ref); - up_read(&c->gc_lock); - goto err; - } - - bch2_invalidate_one_bucket(c, ca, bu, - &journal_seq); - - fifo_push(&ca->free[RESERVE_BTREE], bu); - } - } - - pr_debug("done scanning for reclaimable buckets"); - - /* - * XXX: it's possible for this to deadlock waiting on journal reclaim, - * since we're holding btree writes. What then? - */ - ret = bch2_alloc_write(c, - BTREE_INSERT_NOCHECK_RW| - BTREE_INSERT_USE_ALLOC_RESERVE| - BTREE_INSERT_NOWAIT, &wrote); - - /* - * If bch2_alloc_write() did anything, it may have used some - * buckets, and we need the RESERVE_BTREE freelist full - so we - * need to loop and scan again. - * And if it errored, it may have been because there weren't - * enough buckets, so just scan and loop again as long as it - * made some progress: - */ - } while (wrote); - up_read(&c->gc_lock); - - if (ret) - goto err; - - pr_debug("flushing journal"); - - ret = bch2_journal_flush(&c->journal); - if (ret) - goto err; - - pr_debug("issuing discards"); - allocator_start_issue_discards(c); -err: - clear_bit(BCH_FS_HOLD_BTREE_WRITES, &c->flags); - closure_wait_event(&c->btree_interior_update_wait, - flush_held_btree_writes(c)); - - return ret; -} - void bch2_fs_allocator_background_init(struct bch_fs *c) { spin_lock_init(&c->freelist_lock); diff --git a/fs/bcachefs/alloc_background.h b/fs/bcachefs/alloc_background.h index 501c444353fb..f6b9f27f0713 100644 --- a/fs/bcachefs/alloc_background.h +++ b/fs/bcachefs/alloc_background.h @@ -54,7 +54,6 @@ void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c); struct journal_keys; int bch2_alloc_read(struct bch_fs *, struct journal_keys *); -int bch2_alloc_replay_key(struct bch_fs *, struct bkey_i *); static inline void bch2_wake_allocator(struct bch_dev *ca) { @@ -70,8 +69,7 @@ static inline void bch2_wake_allocator(struct bch_dev *ca) static inline void verify_not_on_freelist(struct bch_fs *c, struct bch_dev *ca, size_t bucket) { - if (expensive_debug_checks(c) && - test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) { + if (expensive_debug_checks(c)) { size_t iter; long i; unsigned j; @@ -94,7 +92,6 @@ void bch2_dev_allocator_stop(struct bch_dev *); int bch2_dev_allocator_start(struct bch_dev *); int bch2_alloc_write(struct bch_fs *, unsigned, bool *); -int bch2_fs_allocator_start(struct bch_fs *); void bch2_fs_allocator_background_init(struct bch_fs *); #endif /* _BCACHEFS_ALLOC_BACKGROUND_H */ diff --git a/fs/bcachefs/alloc_foreground.c b/fs/bcachefs/alloc_foreground.c index 697d576802b6..979aba30bc9d 100644 --- a/fs/bcachefs/alloc_foreground.c +++ b/fs/bcachefs/alloc_foreground.c @@ -212,9 +212,9 @@ static inline unsigned open_buckets_reserved(enum alloc_reserve reserve) case RESERVE_ALLOC: return 0; case RESERVE_BTREE: - return BTREE_NODE_OPEN_BUCKET_RESERVE; + return OPEN_BUCKETS_COUNT / 4; default: - return BTREE_NODE_OPEN_BUCKET_RESERVE * 2; + return OPEN_BUCKETS_COUNT / 2; } } diff --git a/fs/bcachefs/alloc_types.h b/fs/bcachefs/alloc_types.h index 832568dc9551..4f1465077994 100644 --- a/fs/bcachefs/alloc_types.h +++ b/fs/bcachefs/alloc_types.h @@ -46,16 +46,22 @@ enum alloc_reserve { typedef FIFO(long) alloc_fifo; -/* Enough for 16 cache devices, 2 tiers and some left over for pipelining */ -#define OPEN_BUCKETS_COUNT 256 +#define OPEN_BUCKETS_COUNT 1024 #define WRITE_POINT_HASH_NR 32 #define WRITE_POINT_MAX 32 +typedef u16 open_bucket_idx_t; + struct open_bucket { spinlock_t lock; atomic_t pin; - u8 freelist; + open_bucket_idx_t freelist; + + /* + * When an open bucket has an ec_stripe attached, this is the index of + * the block in the stripe this open_bucket corresponds to: + */ u8 ec_idx; u8 type; unsigned valid:1; @@ -68,8 +74,8 @@ struct open_bucket { #define OPEN_BUCKET_LIST_MAX 15 struct open_buckets { - u8 nr; - u8 v[OPEN_BUCKET_LIST_MAX]; + open_bucket_idx_t nr; + open_bucket_idx_t v[OPEN_BUCKET_LIST_MAX]; }; struct dev_stripe_state { diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h index 289d7ae4e98c..893c89dbee60 100644 --- a/fs/bcachefs/bcachefs.h +++ b/fs/bcachefs/bcachefs.h @@ -190,6 +190,7 @@ #include <linux/percpu-rwsem.h> #include <linux/rhashtable.h> #include <linux/rwsem.h> +#include <linux/semaphore.h> #include <linux/seqlock.h> #include <linux/shrinker.h> #include <linux/types.h> @@ -338,7 +339,7 @@ enum bch_time_stats { #define BTREE_RESERVE_MAX (BTREE_MAX_DEPTH + (BTREE_MAX_DEPTH - 1)) /* Size of the freelist we allocate btree nodes from: */ -#define BTREE_NODE_RESERVE BTREE_RESERVE_MAX +#define BTREE_NODE_RESERVE (BTREE_RESERVE_MAX * 4) #define BTREE_NODE_OPEN_BUCKET_RESERVE (BTREE_RESERVE_MAX * BCH_REPLICAS_MAX) @@ -426,8 +427,8 @@ struct bch_dev { alloc_fifo free[RESERVE_NR]; alloc_fifo free_inc; - u8 open_buckets_partial[OPEN_BUCKETS_COUNT]; - unsigned open_buckets_partial_nr; + open_bucket_idx_t open_buckets_partial[OPEN_BUCKETS_COUNT]; + open_bucket_idx_t open_buckets_partial_nr; size_t fifo_last_bucket; @@ -477,9 +478,11 @@ struct bch_dev { enum { /* startup: */ BCH_FS_ALLOC_READ_DONE, - BCH_FS_ALLOCATOR_STARTED, + BCH_FS_ALLOC_CLEAN, BCH_FS_ALLOCATOR_RUNNING, + BCH_FS_ALLOCATOR_STOPPING, BCH_FS_INITIAL_GC_DONE, + BCH_FS_BTREE_INTERIOR_REPLAY_DONE, BCH_FS_FSCK_DONE, BCH_FS_STARTED, BCH_FS_RW, @@ -548,8 +551,8 @@ struct bch_fs { struct super_block *vfs_sb; char name[40]; - /* ro/rw, add/remove devices: */ - struct mutex state_lock; + /* ro/rw, add/remove/resize devices: */ + struct rw_semaphore state_lock; /* Counts outstanding writes, for clean transition to read-only */ struct percpu_ref writes; @@ -600,13 +603,10 @@ struct bch_fs { struct bio_set btree_bio; struct btree_root btree_roots[BTREE_ID_NR]; - bool btree_roots_dirty; struct mutex btree_root_lock; struct btree_cache btree_cache; - mempool_t btree_reserve_pool; - /* * Cache of allocated btree nodes - if we allocate a btree node and * don't use it, if we free it that space can't be reused until going @@ -624,8 +624,16 @@ struct bch_fs { struct mutex btree_interior_update_lock; struct closure_waitlist btree_interior_update_wait; + struct workqueue_struct *btree_interior_update_worker; + struct work_struct btree_interior_update_work; + + /* btree_iter.c: */ + struct mutex btree_trans_lock; + struct list_head btree_trans_list; mempool_t btree_iters_pool; + struct btree_key_cache btree_key_cache; + struct workqueue_struct *wq; /* copygc needs its own workqueue for index updates.. */ struct workqueue_struct *copygc_wq; @@ -682,8 +690,8 @@ struct bch_fs { struct closure_waitlist freelist_wait; u64 blocked_allocate; u64 blocked_allocate_open_bucket; - u8 open_buckets_freelist; - u8 open_buckets_nr_free; + open_bucket_idx_t open_buckets_freelist; + open_bucket_idx_t open_buckets_nr_free; struct closure_waitlist open_buckets_wait; struct open_bucket open_buckets[OPEN_BUCKETS_COUNT]; @@ -719,6 +727,7 @@ struct bch_fs { struct rw_semaphore gc_lock; /* IO PATH */ + struct semaphore io_in_flight; struct bio_set bio_read; struct bio_set bio_read_split; struct bio_set bio_write; @@ -732,7 +741,7 @@ struct bch_fs { ZSTD_parameters zstd_params; struct crypto_shash *sha256; - struct crypto_skcipher *chacha20; + struct crypto_sync_skcipher *chacha20; struct crypto_shash *poly1305; atomic64_t key_version; diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h index 616863ef77d4..f808e63a713d 100644 --- a/fs/bcachefs/bcachefs_format.h +++ b/fs/bcachefs/bcachefs_format.h @@ -1262,6 +1262,8 @@ LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); LE64_BITMASK(BCH_SB_HAS_ERRORS, struct bch_sb, flags[0], 60, 61); +LE64_BITMASK(BCH_SB_REFLINK, struct bch_sb, flags[0], 61, 62); + /* 61-64 unused */ LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); diff --git a/fs/bcachefs/bcachefs_ioctl.h b/fs/bcachefs/bcachefs_ioctl.h index ba8c75706bf1..d71157a3e073 100644 --- a/fs/bcachefs/bcachefs_ioctl.h +++ b/fs/bcachefs/bcachefs_ioctl.h @@ -275,9 +275,13 @@ struct bch_ioctl_dev_usage { __u32 bucket_size; __u64 nr_buckets; + __u64 available_buckets; __u64 buckets[BCH_DATA_NR]; __u64 sectors[BCH_DATA_NR]; + + __u64 ec_buckets; + __u64 ec_sectors; }; /* diff --git a/fs/bcachefs/bkey_methods.c b/fs/bcachefs/bkey_methods.c index 55ef4032b37c..36e0c5152b47 100644 --- a/fs/bcachefs/bkey_methods.c +++ b/fs/bcachefs/bkey_methods.c @@ -176,13 +176,17 @@ void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) { - pr_buf(out, "u64s %u type %s ", k->u64s, - bch2_bkey_types[k->type]); + if (k) { + pr_buf(out, "u64s %u type %s ", k->u64s, + bch2_bkey_types[k->type]); - bch2_bpos_to_text(out, k->p); + bch2_bpos_to_text(out, k->p); - pr_buf(out, " snap %u len %u ver %llu", - k->p.snapshot, k->size, k->version.lo); + pr_buf(out, " snap %u len %u ver %llu", + k->p.snapshot, k->size, k->version.lo); + } else { + pr_buf(out, "(null)"); + } } void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, @@ -198,8 +202,11 @@ void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) { bch2_bkey_to_text(out, k.k); - pr_buf(out, ": "); - bch2_val_to_text(out, c, k); + + if (k.k) { + pr_buf(out, ": "); + bch2_val_to_text(out, c, k); + } } void bch2_bkey_swab_val(struct bkey_s k) diff --git a/fs/bcachefs/bset.c b/fs/bcachefs/bset.c index 6360b2e8cf73..6fc91e6a35e8 100644 --- a/fs/bcachefs/bset.c +++ b/fs/bcachefs/bset.c @@ -64,21 +64,27 @@ struct bset_tree *bch2_bkey_to_bset(struct btree *b, struct bkey_packed *k) * by the time we actually do the insert will all be deleted. */ -void bch2_dump_bset(struct btree *b, struct bset *i, unsigned set) +void bch2_dump_bset(struct bch_fs *c, struct btree *b, + struct bset *i, unsigned set) { struct bkey_packed *_k, *_n; - struct bkey k, n; - char buf[120]; + struct bkey uk, n; + struct bkey_s_c k; + char buf[200]; if (!i->u64s) return; - for (_k = i->start, k = bkey_unpack_key(b, _k); + for (_k = i->start; _k < vstruct_last(i); - _k = _n, k = n) { + _k = _n) { _n = bkey_next_skip_noops(_k, vstruct_last(i)); - bch2_bkey_to_text(&PBUF(buf), &k); + k = bkey_disassemble(b, _k, &uk); + if (c) + bch2_bkey_val_to_text(&PBUF(buf), c, k); + else + bch2_bkey_to_text(&PBUF(buf), k.k); printk(KERN_ERR "block %u key %5zu: %s\n", set, _k->_data - i->_data, buf); @@ -87,31 +93,24 @@ void bch2_dump_bset(struct btree *b, struct bset *i, unsigned set) n = bkey_unpack_key(b, _n); - if (bkey_cmp(bkey_start_pos(&n), k.p) < 0) { + if (bkey_cmp(bkey_start_pos(&n), k.k->p) < 0) { printk(KERN_ERR "Key skipped backwards\n"); continue; } - /* - * Weird check for duplicate non extent keys: extents are - * deleted iff they have 0 size, so if it has zero size and it's - * not deleted these aren't extents: - */ - if (((!k.size && !bkey_deleted(&k)) || - (!n.size && !bkey_deleted(&n))) && - !bkey_deleted(&k) && - !bkey_cmp(n.p, k.p)) + if (!bkey_deleted(k.k) && + !bkey_cmp(n.p, k.k->p)) printk(KERN_ERR "Duplicate keys\n"); } } -void bch2_dump_btree_node(struct btree *b) +void bch2_dump_btree_node(struct bch_fs *c, struct btree *b) { struct bset_tree *t; console_lock(); for_each_bset(b, t) - bch2_dump_bset(b, bset(b, t), t - b->set); + bch2_dump_bset(c, b, bset(b, t), t - b->set); console_unlock(); } @@ -170,7 +169,7 @@ static void bch2_btree_node_iter_next_check(struct btree_node_iter *_iter, struct bkey nu = bkey_unpack_key(b, n); char buf1[80], buf2[80]; - bch2_dump_btree_node(b); + bch2_dump_btree_node(NULL, b); bch2_bkey_to_text(&PBUF(buf1), &ku); bch2_bkey_to_text(&PBUF(buf2), &nu); printk(KERN_ERR "out of order/overlapping:\n%s\n%s\n", @@ -248,7 +247,7 @@ void bch2_verify_insert_pos(struct btree *b, struct bkey_packed *where, char buf1[100]; char buf2[100]; - bch2_dump_btree_node(b); + bch2_dump_btree_node(NULL, b); bch2_bkey_to_text(&PBUF(buf1), &k1); bch2_bkey_to_text(&PBUF(buf2), &k2); @@ -269,7 +268,7 @@ void bch2_verify_insert_pos(struct btree *b, struct bkey_packed *where, char buf1[100]; char buf2[100]; - bch2_dump_btree_node(b); + bch2_dump_btree_node(NULL, b); bch2_bkey_to_text(&PBUF(buf1), &k1); bch2_bkey_to_text(&PBUF(buf2), &k2); diff --git a/fs/bcachefs/bset.h b/fs/bcachefs/bset.h index 7338ccbc8cbd..652ffed4adfb 100644 --- a/fs/bcachefs/bset.h +++ b/fs/bcachefs/bset.h @@ -600,8 +600,8 @@ void bch2_bfloat_to_text(struct printbuf *, struct btree *, /* Debug stuff */ -void bch2_dump_bset(struct btree *, struct bset *, unsigned); -void bch2_dump_btree_node(struct btree *); +void bch2_dump_bset(struct bch_fs *, struct btree *, struct bset *, unsigned); +void bch2_dump_btree_node(struct bch_fs *, struct btree *); void bch2_dump_btree_node_iter(struct btree *, struct btree_node_iter *); #ifdef CONFIG_BCACHEFS_DEBUG diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c index c12f8a6b5205..d3addd3a8964 100644 --- a/fs/bcachefs/btree_cache.c +++ b/fs/bcachefs/btree_cache.c @@ -28,7 +28,7 @@ void bch2_recalc_btree_reserve(struct bch_fs *c) for (i = 0; i < BTREE_ID_NR; i++) if (c->btree_roots[i].b) reserve += min_t(unsigned, 1, - c->btree_roots[i].b->level) * 8; + c->btree_roots[i].b->c.level) * 8; c->btree_cache.reserve = reserve; } @@ -72,24 +72,33 @@ static const struct rhashtable_params bch_btree_cache_params = { .obj_cmpfn = bch2_btree_cache_cmp_fn, }; -static void btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp) +static int __btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp) { - struct btree_cache *bc = &c->btree_cache; + BUG_ON(b->data || b->aux_data); b->data = kvpmalloc(btree_bytes(c), gfp); if (!b->data) - goto err; + return -ENOMEM; - if (bch2_btree_keys_alloc(b, btree_page_order(c), gfp)) - goto err; + if (bch2_btree_keys_alloc(b, btree_page_order(c), gfp)) { + kvpfree(b->data, btree_bytes(c)); + b->data = NULL; + return -ENOMEM; + } - bc->used++; - list_move(&b->list, &bc->freeable); - return; -err: - kvpfree(b->data, btree_bytes(c)); - b->data = NULL; - list_move(&b->list, &bc->freed); + return 0; +} + +static void btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp) +{ + struct btree_cache *bc = &c->btree_cache; + + if (!__btree_node_data_alloc(c, b, gfp)) { + bc->used++; + list_move(&b->list, &bc->freeable); + } else { + list_move(&b->list, &bc->freed); + } } static struct btree *btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp) @@ -99,7 +108,7 @@ static struct btree *btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp) return NULL; bkey_btree_ptr_init(&b->key); - six_lock_init(&b->lock); + six_lock_init(&b->c.lock); INIT_LIST_HEAD(&b->list); INIT_LIST_HEAD(&b->write_blocked); @@ -131,8 +140,8 @@ int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b, { int ret; - b->level = level; - b->btree_id = id; + b->c.level = level; + b->c.btree_id = id; mutex_lock(&bc->lock); ret = __bch2_btree_node_hash_insert(bc, b); @@ -163,10 +172,10 @@ static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush) lockdep_assert_held(&bc->lock); - if (!six_trylock_intent(&b->lock)) + if (!six_trylock_intent(&b->c.lock)) return -ENOMEM; - if (!six_trylock_write(&b->lock)) + if (!six_trylock_write(&b->c.lock)) goto out_unlock_intent; if (btree_node_noevict(b)) @@ -207,9 +216,9 @@ out: trace_btree_node_reap(c, b); return ret; out_unlock: - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); out_unlock_intent: - six_unlock_intent(&b->lock); + six_unlock_intent(&b->c.lock); ret = -ENOMEM; goto out; } @@ -241,7 +250,7 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink, return SHRINK_STOP; /* Return -1 if we can't do anything right now */ - if (sc->gfp_mask & __GFP_IO) + if (sc->gfp_mask & __GFP_FS) mutex_lock(&bc->lock); else if (!mutex_trylock(&bc->lock)) return -1; @@ -267,8 +276,8 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink, if (++i > 3 && !btree_node_reclaim(c, b)) { btree_node_data_free(c, b); - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); freed++; } } @@ -294,13 +303,13 @@ restart: mutex_unlock(&bc->lock); bch2_btree_node_hash_remove(bc, b); - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); if (freed >= nr) goto out; - if (sc->gfp_mask & __GFP_IO) + if (sc->gfp_mask & __GFP_FS) mutex_lock(&bc->lock); else if (!mutex_trylock(&bc->lock)) goto out; @@ -524,36 +533,47 @@ struct btree *bch2_btree_node_mem_alloc(struct bch_fs *c) */ list_for_each_entry(b, &bc->freeable, list) if (!btree_node_reclaim(c, b)) - goto out_unlock; + goto got_node; /* * We never free struct btree itself, just the memory that holds the on * disk node. Check the freed list before allocating a new one: */ list_for_each_entry(b, &bc->freed, list) - if (!btree_node_reclaim(c, b)) { - btree_node_data_alloc(c, b, __GFP_NOWARN|GFP_NOIO); - if (b->data) - goto out_unlock; + if (!btree_node_reclaim(c, b)) + goto got_node; + + b = NULL; +got_node: + if (b) + list_del_init(&b->list); + mutex_unlock(&bc->lock); - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); + if (!b) { + b = kzalloc(sizeof(struct btree), GFP_KERNEL); + if (!b) goto err; - } - b = btree_node_mem_alloc(c, __GFP_NOWARN|GFP_NOIO); - if (!b) - goto err; + bkey_btree_ptr_init(&b->key); + six_lock_init(&b->c.lock); + INIT_LIST_HEAD(&b->list); + INIT_LIST_HEAD(&b->write_blocked); + + BUG_ON(!six_trylock_intent(&b->c.lock)); + BUG_ON(!six_trylock_write(&b->c.lock)); + } + + if (!b->data) { + if (__btree_node_data_alloc(c, b, __GFP_NOWARN|GFP_KERNEL)) + goto err; + + mutex_lock(&bc->lock); + bc->used++; + mutex_unlock(&bc->lock); + } - BUG_ON(!six_trylock_intent(&b->lock)); - BUG_ON(!six_trylock_write(&b->lock)); -out_unlock: BUG_ON(btree_node_hashed(b)); BUG_ON(btree_node_write_in_flight(b)); - - list_del_init(&b->list); - mutex_unlock(&bc->lock); - memalloc_nofs_restore(flags); out: b->flags = 0; b->written = 0; @@ -566,8 +586,17 @@ out: bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc], start_time); + memalloc_nofs_restore(flags); return b; err: + mutex_lock(&bc->lock); + + if (b) { + list_add(&b->list, &bc->freed); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); + } + /* Try to cannibalize another cached btree node: */ if (bc->alloc_lock == current) { b = btree_node_cannibalize(c); @@ -581,6 +610,7 @@ err: } mutex_unlock(&bc->lock); + memalloc_nofs_restore(flags); return ERR_PTR(-ENOMEM); } @@ -619,8 +649,8 @@ static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c, list_add(&b->list, &bc->freeable); mutex_unlock(&bc->lock); - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); return NULL; } @@ -634,19 +664,27 @@ static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c, bch2_btree_node_read(c, b, sync); - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); if (!sync) { - six_unlock_intent(&b->lock); + six_unlock_intent(&b->c.lock); return NULL; } if (lock_type == SIX_LOCK_read) - six_lock_downgrade(&b->lock); + six_lock_downgrade(&b->c.lock); return b; } +static int lock_node_check_fn(struct six_lock *lock, void *p) +{ + struct btree *b = container_of(lock, struct btree, c.lock); + const struct bkey_i *k = p; + + return b->hash_val == btree_ptr_hash_val(k) ? 0 : -1; +} + /** * bch_btree_node_get - find a btree node in the cache and lock it, reading it * in from disk if necessary. @@ -719,13 +757,17 @@ lock_node: if (btree_node_read_locked(iter, level + 1)) btree_node_unlock(iter, level + 1); - if (!btree_node_lock(b, k->k.p, level, iter, lock_type)) + if (!btree_node_lock(b, k->k.p, level, iter, lock_type, + lock_node_check_fn, (void *) k)) { + if (b->hash_val != btree_ptr_hash_val(k)) + goto retry; return ERR_PTR(-EINTR); + } if (unlikely(b->hash_val != btree_ptr_hash_val(k) || - b->level != level || + b->c.level != level || race_fault())) { - six_unlock_type(&b->lock, lock_type); + six_unlock_type(&b->c.lock, lock_type); if (bch2_btree_node_relock(iter, level + 1)) goto retry; @@ -753,11 +795,11 @@ lock_node: set_btree_node_accessed(b); if (unlikely(btree_node_read_error(b))) { - six_unlock_type(&b->lock, lock_type); + six_unlock_type(&b->c.lock, lock_type); return ERR_PTR(-EIO); } - EBUG_ON(b->btree_id != iter->btree_id || + EBUG_ON(b->c.btree_id != iter->btree_id || BTREE_NODE_LEVEL(b->data) != level || bkey_cmp(b->data->max_key, k->k.p)); @@ -772,6 +814,7 @@ struct btree *bch2_btree_node_get_noiter(struct bch_fs *c, struct btree_cache *bc = &c->btree_cache; struct btree *b; struct bset_tree *t; + int ret; EBUG_ON(level >= BTREE_MAX_DEPTH); @@ -792,12 +835,14 @@ retry: return b; } else { lock_node: - six_lock_read(&b->lock); + ret = six_lock_read(&b->c.lock, lock_node_check_fn, (void *) k); + if (ret) + goto retry; if (unlikely(b->hash_val != btree_ptr_hash_val(k) || - b->btree_id != btree_id || - b->level != level)) { - six_unlock_read(&b->lock); + b->c.btree_id != btree_id || + b->c.level != level)) { + six_unlock_read(&b->c.lock); goto retry; } } @@ -821,11 +866,11 @@ lock_node: set_btree_node_accessed(b); if (unlikely(btree_node_read_error(b))) { - six_unlock_read(&b->lock); + six_unlock_read(&b->c.lock); return ERR_PTR(-EIO); } - EBUG_ON(b->btree_id != btree_id || + EBUG_ON(b->c.btree_id != btree_id || BTREE_NODE_LEVEL(b->data) != level || bkey_cmp(b->data->max_key, k->k.p)); @@ -843,18 +888,30 @@ struct btree *bch2_btree_node_get_sibling(struct bch_fs *c, struct bkey_packed *k; BKEY_PADDED(k) tmp; struct btree *ret = NULL; - unsigned level = b->level; + unsigned level = b->c.level; parent = btree_iter_node(iter, level + 1); if (!parent) return NULL; + /* + * There's a corner case where a btree_iter might have a node locked + * that is just outside its current pos - when + * bch2_btree_iter_set_pos_same_leaf() gets to the end of the node. + * + * But the lock ordering checks in __bch2_btree_node_lock() go off of + * iter->pos, not the node's key: so if the iterator is marked as + * needing to be traversed, we risk deadlock if we don't bail out here: + */ + if (iter->uptodate >= BTREE_ITER_NEED_TRAVERSE) + return ERR_PTR(-EINTR); + if (!bch2_btree_node_relock(iter, level + 1)) { ret = ERR_PTR(-EINTR); goto out; } - node_iter = iter->l[parent->level].iter; + node_iter = iter->l[parent->c.level].iter; k = bch2_btree_node_iter_peek_all(&node_iter, parent); BUG_ON(bkey_cmp_left_packed(parent, k, &b->key.k.p)); @@ -901,7 +958,7 @@ struct btree *bch2_btree_node_get_sibling(struct bch_fs *c, btree_iter_set_dirty(iter, BTREE_ITER_NEED_RELOCK); if (!IS_ERR(ret)) { - six_unlock_intent(&ret->lock); + six_unlock_intent(&ret->c.lock); ret = ERR_PTR(-EINTR); } } @@ -962,7 +1019,7 @@ void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c, pr_buf(out, "l %u %llu:%llu - %llu:%llu:\n" " ptrs: ", - b->level, + b->c.level, b->data->min_key.inode, b->data->min_key.offset, b->data->max_key.inode, diff --git a/fs/bcachefs/btree_cache.h b/fs/bcachefs/btree_cache.h index 98cca30778ea..2160012c734f 100644 --- a/fs/bcachefs/btree_cache.h +++ b/fs/bcachefs/btree_cache.h @@ -101,7 +101,7 @@ static inline unsigned btree_blocks(struct bch_fs *c) (BTREE_FOREGROUND_MERGE_THRESHOLD(c) + \ (BTREE_FOREGROUND_MERGE_THRESHOLD(c) << 2)) -#define btree_node_root(_c, _b) ((_c)->btree_roots[(_b)->btree_id].b) +#define btree_node_root(_c, _b) ((_c)->btree_roots[(_b)->c.btree_id].b) void bch2_btree_node_to_text(struct printbuf *, struct bch_fs *, struct btree *); diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c index 146f2428fe04..8771ef1f07cc 100644 --- a/fs/bcachefs/btree_gc.c +++ b/fs/bcachefs/btree_gc.c @@ -186,7 +186,7 @@ static int btree_gc_mark_node(struct bch_fs *c, struct btree *b, u8 *max_stale, bch2_btree_node_iter_advance(&iter, b); - if (b->level) { + if (b->c.level) { ret = bch2_gc_check_topology(c, k, &next_node_start, b->data->max_key, @@ -252,7 +252,7 @@ static int bch2_gc_btree(struct bch_fs *c, enum btree_id btree_id, if (!btree_node_fake(b)) ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key), &max_stale, initial); - gc_pos_set(c, gc_pos_btree_root(b->btree_id)); + gc_pos_set(c, gc_pos_btree_root(b->c.btree_id)); mutex_unlock(&c->btree_root_lock); return ret; @@ -280,7 +280,7 @@ static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b, if (ret) break; - if (b->level) { + if (b->c.level) { struct btree *child; BKEY_PADDED(k) tmp; @@ -296,16 +296,16 @@ static int bch2_gc_btree_init_recurse(struct bch_fs *c, struct btree *b, if (ret) break; - if (b->level > target_depth) { + if (b->c.level > target_depth) { child = bch2_btree_node_get_noiter(c, &tmp.k, - b->btree_id, b->level - 1); + b->c.btree_id, b->c.level - 1); ret = PTR_ERR_OR_ZERO(child); if (ret) break; ret = bch2_gc_btree_init_recurse(c, child, journal_keys, target_depth); - six_unlock_read(&child->lock); + six_unlock_read(&child->c.lock); if (ret) break; @@ -336,7 +336,7 @@ static int bch2_gc_btree_init(struct bch_fs *c, if (btree_node_fake(b)) return 0; - six_lock_read(&b->lock); + six_lock_read(&b->c.lock, NULL, NULL); if (fsck_err_on(bkey_cmp(b->data->min_key, POS_MIN), c, "btree root with incorrect min_key: %llu:%llu", b->data->min_key.inode, @@ -351,7 +351,7 @@ static int bch2_gc_btree_init(struct bch_fs *c, BUG(); } - if (b->level >= target_depth) + if (b->c.level >= target_depth) ret = bch2_gc_btree_init_recurse(c, b, journal_keys, target_depth); @@ -359,7 +359,7 @@ static int bch2_gc_btree_init(struct bch_fs *c, ret = bch2_gc_mark_key(c, bkey_i_to_s_c(&b->key), &max_stale, true); fsck_err: - six_unlock_read(&b->lock); + six_unlock_read(&b->c.lock); return ret; } @@ -464,6 +464,7 @@ static void bch2_mark_superblocks(struct bch_fs *c) mutex_unlock(&c->sb_lock); } +#if 0 /* Also see bch2_pending_btree_node_free_insert_done() */ static void bch2_mark_pending_btree_node_frees(struct bch_fs *c) { @@ -481,6 +482,7 @@ static void bch2_mark_pending_btree_node_frees(struct bch_fs *c) mutex_unlock(&c->btree_interior_update_lock); } +#endif static void bch2_mark_allocator_buckets(struct bch_fs *c) { @@ -579,8 +581,10 @@ static int bch2_gc_done(struct bch_fs *c, #define copy_bucket_field(_f) \ if (dst->b[b].mark._f != src->b[b].mark._f) { \ if (verify) \ - fsck_err(c, "dev %u bucket %zu has wrong " #_f \ + fsck_err(c, "bucket %u:%zu gen %u data type %s has wrong " #_f \ ": got %u, should be %u", i, b, \ + dst->b[b].mark.gen, \ + bch2_data_types[dst->b[b].mark.data_type],\ dst->b[b].mark._f, src->b[b].mark._f); \ dst->b[b]._mark._f = src->b[b].mark._f; \ } @@ -794,9 +798,14 @@ int bch2_gc(struct bch_fs *c, struct journal_keys *journal_keys, unsigned i, iter = 0; int ret; + lockdep_assert_held(&c->state_lock); trace_gc_start(c); down_write(&c->gc_lock); + + /* flush interior btree updates: */ + closure_wait_event(&c->btree_interior_update_wait, + !bch2_btree_interior_updates_nr_pending(c)); again: ret = bch2_gc_start(c, metadata_only); if (ret) @@ -808,7 +817,9 @@ again: if (ret) goto out; +#if 0 bch2_mark_pending_btree_node_frees(c); +#endif bch2_mark_allocator_buckets(c); c->gc_count++; @@ -874,6 +885,87 @@ out: return ret; } +/* + * For recalculating oldest gen, we only need to walk keys in leaf nodes; btree + * node pointers currently never have cached pointers that can become stale: + */ +static int bch2_gc_btree_gens(struct bch_fs *c, enum btree_id id) +{ + struct btree_trans trans; + struct btree_iter *iter; + struct bkey_s_c k; + int ret; + + bch2_trans_init(&trans, c, 0, 0); + + for_each_btree_key(&trans, iter, id, POS_MIN, BTREE_ITER_PREFETCH, k, ret) { + struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k); + const struct bch_extent_ptr *ptr; + + percpu_down_read(&c->mark_lock); + bkey_for_each_ptr(ptrs, ptr) { + struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); + struct bucket *g = PTR_BUCKET(ca, ptr, false); + + if (gen_after(g->gc_gen, ptr->gen)) + g->gc_gen = ptr->gen; + + if (gen_after(g->mark.gen, ptr->gen) > 32) { + /* rewrite btree node */ + + } + } + percpu_up_read(&c->mark_lock); + } + + bch2_trans_exit(&trans); + return ret; +} + +int bch2_gc_gens(struct bch_fs *c) +{ + struct bch_dev *ca; + struct bucket_array *buckets; + struct bucket *g; + unsigned i; + int ret; + + /* + * Ideally we would be using state_lock and not gc_lock here, but that + * introduces a deadlock in the RO path - we currently take the state + * lock at the start of going RO, thus the gc thread may get stuck: + */ + down_read(&c->gc_lock); + + for_each_member_device(ca, c, i) { + down_read(&ca->bucket_lock); + buckets = bucket_array(ca); + + for_each_bucket(g, buckets) + g->gc_gen = g->mark.gen; + up_read(&ca->bucket_lock); + } + + for (i = 0; i < BTREE_ID_NR; i++) + if (btree_node_type_needs_gc(i)) { + ret = bch2_gc_btree_gens(c, i); + if (ret) + goto err; + } + + for_each_member_device(ca, c, i) { + down_read(&ca->bucket_lock); + buckets = bucket_array(ca); + + for_each_bucket(g, buckets) + g->oldest_gen = g->gc_gen; + up_read(&ca->bucket_lock); + } +err: + up_read(&c->gc_lock); + return ret; +} + /* Btree coalescing */ static void recalc_packed_keys(struct btree *b) @@ -997,9 +1089,9 @@ static void bch2_coalesce_nodes(struct bch_fs *c, struct btree_iter *iter, set_btree_bset_end(n1, n1->set); - six_unlock_write(&n2->lock); + six_unlock_write(&n2->c.lock); bch2_btree_node_free_never_inserted(c, n2); - six_unlock_intent(&n2->lock); + six_unlock_intent(&n2->c.lock); memmove(new_nodes + i - 1, new_nodes + i, @@ -1033,7 +1125,9 @@ static void bch2_coalesce_nodes(struct bch_fs *c, struct btree_iter *iter, btree_node_reset_sib_u64s(n); bch2_btree_build_aux_trees(n); - six_unlock_write(&n->lock); + + bch2_btree_update_add_new_node(as, n); + six_unlock_write(&n->c.lock); bch2_btree_node_write(c, n, SIX_LOCK_intent); } @@ -1076,12 +1170,12 @@ next: BUG_ON(!bch2_keylist_empty(&keylist)); - BUG_ON(iter->l[old_nodes[0]->level].b != old_nodes[0]); + BUG_ON(iter->l[old_nodes[0]->c.level].b != old_nodes[0]); bch2_btree_iter_node_replace(iter, new_nodes[0]); for (i = 0; i < nr_new_nodes; i++) - bch2_open_buckets_put(c, &new_nodes[i]->ob); + bch2_btree_update_get_open_buckets(as, new_nodes[i]); /* Free the old nodes and update our sliding window */ for (i = 0; i < nr_old_nodes; i++) { @@ -1101,7 +1195,7 @@ next: } for (i = 0; i < nr_new_nodes; i++) - six_unlock_intent(&new_nodes[i]->lock); + six_unlock_intent(&new_nodes[i]->c.lock); bch2_btree_update_done(as); bch2_keylist_free(&keylist, NULL); @@ -1142,11 +1236,11 @@ static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id) for (i = 1; i < GC_MERGE_NODES; i++) { if (!merge[i] || - !six_relock_intent(&merge[i]->lock, lock_seq[i])) + !six_relock_intent(&merge[i]->c.lock, lock_seq[i])) break; - if (merge[i]->level != merge[0]->level) { - six_unlock_intent(&merge[i]->lock); + if (merge[i]->c.level != merge[0]->c.level) { + six_unlock_intent(&merge[i]->c.lock); break; } } @@ -1155,11 +1249,11 @@ static int bch2_coalesce_btree(struct bch_fs *c, enum btree_id btree_id) bch2_coalesce_nodes(c, iter, merge); for (i = 1; i < GC_MERGE_NODES && merge[i]; i++) { - lock_seq[i] = merge[i]->lock.state.seq; - six_unlock_intent(&merge[i]->lock); + lock_seq[i] = merge[i]->c.lock.state.seq; + six_unlock_intent(&merge[i]->c.lock); } - lock_seq[0] = merge[0]->lock.state.seq; + lock_seq[0] = merge[0]->c.lock.state.seq; if (kthread && kthread_should_stop()) { bch2_trans_exit(&trans); @@ -1247,7 +1341,14 @@ static int bch2_gc_thread(void *arg) last = atomic_long_read(&clock->now); last_kick = atomic_read(&c->kick_gc); + /* + * Full gc is currently incompatible with btree key cache: + */ +#if 0 ret = bch2_gc(c, NULL, false, false); +#else + ret = bch2_gc_gens(c); +#endif if (ret) bch_err(c, "btree gc failed: %i", ret); diff --git a/fs/bcachefs/btree_gc.h b/fs/bcachefs/btree_gc.h index bd5f2752954f..3694a3df62a8 100644 --- a/fs/bcachefs/btree_gc.h +++ b/fs/bcachefs/btree_gc.h @@ -8,6 +8,7 @@ void bch2_coalesce(struct bch_fs *); struct journal_keys; int bch2_gc(struct bch_fs *, struct journal_keys *, bool, bool); +int bch2_gc_gens(struct bch_fs *); void bch2_gc_thread_stop(struct bch_fs *); int bch2_gc_thread_start(struct bch_fs *); void bch2_mark_dev_superblock(struct bch_fs *, struct bch_dev *, unsigned); @@ -81,7 +82,7 @@ static inline struct gc_pos gc_pos_btree(enum btree_id id, */ static inline struct gc_pos gc_pos_btree_node(struct btree *b) { - return gc_pos_btree(b->btree_id, b->key.k.p, b->level); + return gc_pos_btree(b->c.btree_id, b->key.k.p, b->c.level); } /* diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c index 01a0ae0007c9..bb3aeccef67e 100644 --- a/fs/bcachefs/btree_io.c +++ b/fs/bcachefs/btree_io.c @@ -584,8 +584,8 @@ void bch2_btree_init_next(struct bch_fs *c, struct btree *b, struct btree_node_entry *bne; bool did_sort; - EBUG_ON(!(b->lock.state.seq & 1)); - EBUG_ON(iter && iter->l[b->level].b != b); + EBUG_ON(!(b->c.lock.state.seq & 1)); + EBUG_ON(iter && iter->l[b->c.level].b != b); did_sort = btree_node_compact(c, b, iter); @@ -620,7 +620,7 @@ static void bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset) bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, &bn->flags, bytes); - nonce = nonce_add(nonce, round_up(bytes, CHACHA20_BLOCK_SIZE)); + nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE)); } bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data, @@ -631,14 +631,14 @@ static void btree_err_msg(struct printbuf *out, struct bch_fs *c, struct btree *b, struct bset *i, unsigned offset, int write) { - pr_buf(out, "error validating btree node %s" - "at btree %u level %u/%u\n" - "pos %llu:%llu node offset %u", + pr_buf(out, "error validating btree node %sat btree %u level %u/%u\n" + "pos ", write ? "before write " : "", - b->btree_id, b->level, - c->btree_roots[b->btree_id].level, - b->key.k.p.inode, b->key.k.p.offset, - b->written); + b->c.btree_id, b->c.level, + c->btree_roots[b->c.btree_id].level); + bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key)); + + pr_buf(out, " node offset %u", b->written); if (i) pr_buf(out, " bset u64s %u", le16_to_cpu(i->u64s)); } @@ -747,11 +747,11 @@ static int validate_bset(struct bch_fs *c, struct btree *b, "incorrect sequence number (wrong btree node)"); } - btree_err_on(BTREE_NODE_ID(bn) != b->btree_id, + btree_err_on(BTREE_NODE_ID(bn) != b->c.btree_id, BTREE_ERR_MUST_RETRY, c, b, i, "incorrect btree id"); - btree_err_on(BTREE_NODE_LEVEL(bn) != b->level, + btree_err_on(BTREE_NODE_LEVEL(bn) != b->c.level, BTREE_ERR_MUST_RETRY, c, b, i, "incorrect level"); @@ -762,7 +762,7 @@ static int validate_bset(struct bch_fs *c, struct btree *b, } if (!write) - compat_btree_node(b->level, b->btree_id, version, + compat_btree_node(b->c.level, b->c.btree_id, version, BSET_BIG_ENDIAN(i), write, bn); if (b->key.k.type == KEY_TYPE_btree_ptr_v2) { @@ -783,7 +783,7 @@ static int validate_bset(struct bch_fs *c, struct btree *b, "incorrect max key"); if (write) - compat_btree_node(b->level, b->btree_id, version, + compat_btree_node(b->c.level, b->c.btree_id, version, BSET_BIG_ENDIAN(i), write, bn); /* XXX: ideally we would be validating min_key too */ @@ -805,7 +805,7 @@ static int validate_bset(struct bch_fs *c, struct btree *b, BTREE_ERR_FATAL, c, b, i, "invalid bkey format: %s", err); - compat_bformat(b->level, b->btree_id, version, + compat_bformat(b->c.level, b->c.btree_id, version, BSET_BIG_ENDIAN(i), write, &bn->format); } @@ -851,7 +851,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, /* XXX: validate k->u64s */ if (!write) - bch2_bkey_compat(b->level, b->btree_id, version, + bch2_bkey_compat(b->c.level, b->c.btree_id, version, BSET_BIG_ENDIAN(i), write, &b->format, k); @@ -874,7 +874,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, } if (write) - bch2_bkey_compat(b->level, b->btree_id, version, + bch2_bkey_compat(b->c.level, b->c.btree_id, version, BSET_BIG_ENDIAN(i), write, &b->format, k); @@ -897,7 +897,7 @@ static int validate_bset_keys(struct bch_fs *c, struct btree *b, bch2_bkey_to_text(&PBUF(buf1), &up); bch2_bkey_to_text(&PBUF(buf2), u.k); - bch2_dump_bset(b, i, 0); + bch2_dump_bset(c, b, i, 0); btree_err(BTREE_ERR_FATAL, c, b, i, "keys out of order: %s > %s", buf1, buf2); @@ -944,7 +944,8 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct btree *b, bool have_retry btree_err_on(b->data->keys.seq != bp->seq, BTREE_ERR_MUST_RETRY, c, b, NULL, - "got wrong btree node"); + "got wrong btree node (seq %llx want %llx)", + b->data->keys.seq, bp->seq); } while (b->written < c->opts.btree_node_size) { @@ -1279,8 +1280,8 @@ int bch2_btree_root_read(struct bch_fs *c, enum btree_id id, bch2_btree_set_root_for_read(c, b); err: - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); return ret; } @@ -1324,15 +1325,15 @@ static void bch2_btree_node_write_error(struct bch_fs *c, bch2_trans_init(&trans, c, 0, 0); - iter = bch2_trans_get_node_iter(&trans, b->btree_id, b->key.k.p, - BTREE_MAX_DEPTH, b->level, 0); + iter = bch2_trans_get_node_iter(&trans, b->c.btree_id, b->key.k.p, + BTREE_MAX_DEPTH, b->c.level, 0); retry: ret = bch2_btree_iter_traverse(iter); if (ret) goto err; /* has node been freed? */ - if (iter->l[b->level].b != b) { + if (iter->l[b->c.level].b != b) { /* node has been freed: */ BUG_ON(!btree_node_dying(b)); goto out; @@ -1763,18 +1764,18 @@ void bch2_btree_node_write(struct bch_fs *c, struct btree *b, BUG_ON(lock_type_held == SIX_LOCK_write); if (lock_type_held == SIX_LOCK_intent || - six_lock_tryupgrade(&b->lock)) { + six_lock_tryupgrade(&b->c.lock)) { __bch2_btree_node_write(c, b, SIX_LOCK_intent); /* don't cycle lock unnecessarily: */ if (btree_node_just_written(b) && - six_trylock_write(&b->lock)) { + six_trylock_write(&b->c.lock)) { bch2_btree_post_write_cleanup(c, b); - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); } if (lock_type_held == SIX_LOCK_read) - six_lock_downgrade(&b->lock); + six_lock_downgrade(&b->c.lock); } else { __bch2_btree_node_write(c, b, SIX_LOCK_read); } @@ -1844,7 +1845,7 @@ ssize_t bch2_dirty_btree_nodes_print(struct bch_fs *c, char *buf) b, (flags & (1 << BTREE_NODE_dirty)) != 0, (flags & (1 << BTREE_NODE_need_write)) != 0, - b->level, + b->c.level, b->written, !list_empty_careful(&b->write_blocked), b->will_make_reachable != 0, diff --git a/fs/bcachefs/btree_io.h b/fs/bcachefs/btree_io.h index 337d2bdd29e8..f3d7ec749b61 100644 --- a/fs/bcachefs/btree_io.h +++ b/fs/bcachefs/btree_io.h @@ -114,7 +114,7 @@ static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b, break; } - six_unlock_type(&b->lock, lock_held); + six_unlock_type(&b->c.lock, lock_held); btree_node_wait_on_io(b); btree_node_lock_type(c, b, lock_held); } diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c index 5528ba0f1d44..6fab76c3220c 100644 --- a/fs/bcachefs/btree_iter.c +++ b/fs/bcachefs/btree_iter.c @@ -4,22 +4,16 @@ #include "bkey_methods.h" #include "btree_cache.h" #include "btree_iter.h" +#include "btree_key_cache.h" #include "btree_locking.h" #include "btree_update.h" #include "debug.h" #include "extents.h" +#include "journal.h" #include <linux/prefetch.h> #include <trace/events/bcachefs.h> -#define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1) -#define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2) -#define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3) -#define BTREE_ITER_NO_NODE_UP ((struct btree *) 4) -#define BTREE_ITER_NO_NODE_DOWN ((struct btree *) 5) -#define BTREE_ITER_NO_NODE_INIT ((struct btree *) 6) -#define BTREE_ITER_NO_NODE_ERROR ((struct btree *) 7) - static inline bool is_btree_node(struct btree_iter *iter, unsigned l) { return l < BTREE_MAX_DEPTH && @@ -51,7 +45,7 @@ static inline bool btree_iter_pos_after_node(struct btree_iter *iter, static inline bool btree_iter_pos_in_node(struct btree_iter *iter, struct btree *b) { - return iter->btree_id == b->btree_id && + return iter->btree_id == b->c.btree_id && !btree_iter_pos_before_node(iter, b) && !btree_iter_pos_after_node(iter, b); } @@ -68,11 +62,11 @@ void __bch2_btree_node_lock_write(struct btree *b, struct btree_iter *iter) struct btree_iter *linked; unsigned readers = 0; - EBUG_ON(!btree_node_intent_locked(iter, b->level)); + EBUG_ON(!btree_node_intent_locked(iter, b->c.level)); trans_for_each_iter(iter->trans, linked) - if (linked->l[b->level].b == b && - btree_node_read_locked(linked, b->level)) + if (linked->l[b->c.level].b == b && + btree_node_read_locked(linked, b->c.level)) readers++; /* @@ -82,10 +76,10 @@ void __bch2_btree_node_lock_write(struct btree *b, struct btree_iter *iter) * locked: */ atomic64_sub(__SIX_VAL(read_lock, readers), - &b->lock.state.counter); + &b->c.lock.state.counter); btree_node_lock_type(iter->trans->c, b, SIX_LOCK_write); atomic64_add(__SIX_VAL(read_lock, readers), - &b->lock.state.counter); + &b->c.lock.state.counter); } bool __bch2_btree_node_relock(struct btree_iter *iter, unsigned level) @@ -99,9 +93,9 @@ bool __bch2_btree_node_relock(struct btree_iter *iter, unsigned level) if (race_fault()) return false; - if (six_relock_type(&b->lock, want, iter->l[level].lock_seq) || + if (six_relock_type(&b->c.lock, want, iter->l[level].lock_seq) || (btree_node_lock_seq_matches(iter, b, level) && - btree_node_lock_increment(iter, b, level, want))) { + btree_node_lock_increment(iter->trans, b, level, want))) { mark_btree_node_locked(iter, level, want); return true; } else { @@ -125,12 +119,12 @@ static bool bch2_btree_node_upgrade(struct btree_iter *iter, unsigned level) return false; if (btree_node_locked(iter, level) - ? six_lock_tryupgrade(&b->lock) - : six_relock_type(&b->lock, SIX_LOCK_intent, iter->l[level].lock_seq)) + ? six_lock_tryupgrade(&b->c.lock) + : six_relock_type(&b->c.lock, SIX_LOCK_intent, iter->l[level].lock_seq)) goto success; if (btree_node_lock_seq_matches(iter, b, level) && - btree_node_lock_increment(iter, b, level, BTREE_NODE_INTENT_LOCKED)) { + btree_node_lock_increment(iter->trans, b, level, BTREE_NODE_INTENT_LOCKED)) { btree_node_unlock(iter, level); goto success; } @@ -162,7 +156,7 @@ static inline bool btree_iter_get_locks(struct btree_iter *iter, ? 0 : (unsigned long) iter->l[l].b, is_btree_node(iter, l) - ? iter->l[l].b->lock.state.seq + ? iter->l[l].b->c.lock.state.seq : 0); fail_idx = l; @@ -191,24 +185,31 @@ static inline bool btree_iter_get_locks(struct btree_iter *iter, return iter->uptodate < BTREE_ITER_NEED_RELOCK; } +static struct bpos btree_node_pos(struct btree_bkey_cached_common *_b, + enum btree_iter_type type) +{ + return type != BTREE_ITER_CACHED + ? container_of(_b, struct btree, c)->key.k.p + : container_of(_b, struct bkey_cached, c)->key.pos; +} + /* Slowpath: */ bool __bch2_btree_node_lock(struct btree *b, struct bpos pos, - unsigned level, - struct btree_iter *iter, - enum six_lock_type type) + unsigned level, struct btree_iter *iter, + enum six_lock_type type, + six_lock_should_sleep_fn should_sleep_fn, + void *p) { + struct btree_trans *trans = iter->trans; struct btree_iter *linked; + u64 start_time = local_clock(); bool ret = true; /* Check if it's safe to block: */ - trans_for_each_iter(iter->trans, linked) { + trans_for_each_iter(trans, linked) { if (!linked->nodes_locked) continue; - /* * Must lock btree nodes in key order: */ - if (__btree_iter_cmp(iter->btree_id, pos, linked) < 0) - ret = false; - /* * Can't block taking an intent lock if we have _any_ nodes read * locked: @@ -223,13 +224,15 @@ bool __bch2_btree_node_lock(struct btree *b, struct bpos pos, */ if (type == SIX_LOCK_intent && linked->nodes_locked != linked->nodes_intent_locked) { - if (!(iter->trans->nounlock)) { + if (!(trans->nounlock)) { linked->locks_want = max_t(unsigned, linked->locks_want, __fls(linked->nodes_locked) + 1); - btree_iter_get_locks(linked, true, false); + if (!btree_iter_get_locks(linked, true, false)) + ret = false; + } else { + ret = false; } - ret = false; } /* @@ -239,14 +242,38 @@ bool __bch2_btree_node_lock(struct btree *b, struct bpos pos, */ if (linked->btree_id == iter->btree_id && level > __fls(linked->nodes_locked)) { - if (!(iter->trans->nounlock)) { + if (!(trans->nounlock)) { linked->locks_want = max(level + 1, max_t(unsigned, linked->locks_want, iter->locks_want)); - btree_iter_get_locks(linked, true, false); + if (!btree_iter_get_locks(linked, true, false)) + ret = false; + } else { + ret = false; } + } + + /* Must lock btree nodes in key order: */ + if ((cmp_int(iter->btree_id, linked->btree_id) ?: + -cmp_int(btree_iter_type(iter), btree_iter_type(linked))) < 0) + ret = false; + + if (iter->btree_id == linked->btree_id && + btree_node_locked(linked, level) && + bkey_cmp(pos, btree_node_pos((void *) linked->l[level].b, + btree_iter_type(linked))) <= 0) ret = false; + + /* + * Recheck if this is a node we already have locked - since one + * of the get_locks() calls might've successfully + * upgraded/relocked it: + */ + if (linked->l[level].b == b && + btree_node_locked_type(linked, level) >= type) { + six_lock_increment(&b->c.lock, type); + return true; } } @@ -255,7 +282,14 @@ bool __bch2_btree_node_lock(struct btree *b, struct bpos pos, return false; } - __btree_node_lock_type(iter->trans->c, b, type); + if (six_trylock_type(&b->c.lock, type)) + return true; + + if (six_lock_type(&b->c.lock, type, should_sleep_fn, p)) + return false; + + bch2_time_stats_update(&trans->c->times[lock_to_time_stat(type)], + start_time); return true; } @@ -266,7 +300,12 @@ static void bch2_btree_iter_verify_locks(struct btree_iter *iter) { unsigned l; - for (l = 0; btree_iter_node(iter, l); l++) { + if (!(iter->trans->iters_linked & (1ULL << iter->idx))) { + BUG_ON(iter->nodes_locked); + return; + } + + for (l = 0; is_btree_node(iter, l); l++) { if (iter->uptodate >= BTREE_ITER_NEED_RELOCK && !btree_node_locked(iter, l)) continue; @@ -280,7 +319,7 @@ void bch2_btree_trans_verify_locks(struct btree_trans *trans) { struct btree_iter *iter; - trans_for_each_iter(trans, iter) + trans_for_each_iter_all(trans, iter) bch2_btree_iter_verify_locks(iter); } #else @@ -288,7 +327,7 @@ static inline void bch2_btree_iter_verify_locks(struct btree_iter *iter) {} #endif __flatten -static bool bch2_btree_iter_relock(struct btree_iter *iter, bool trace) +bool bch2_btree_iter_relock(struct btree_iter *iter, bool trace) { return btree_iter_get_locks(iter, false, trace); } @@ -348,31 +387,20 @@ bool __bch2_btree_iter_upgrade_nounlock(struct btree_iter *iter, void __bch2_btree_iter_downgrade(struct btree_iter *iter, unsigned downgrade_to) { - struct btree_iter *linked; - unsigned l; - - /* - * We downgrade linked iterators as well because btree_iter_upgrade - * might have had to modify locks_want on linked iterators due to lock - * ordering: - */ - trans_for_each_iter(iter->trans, linked) { - unsigned new_locks_want = downgrade_to ?: - (linked->flags & BTREE_ITER_INTENT ? 1 : 0); + unsigned l, new_locks_want = downgrade_to ?: + (iter->flags & BTREE_ITER_INTENT ? 1 : 0); - if (linked->locks_want <= new_locks_want) - continue; - - linked->locks_want = new_locks_want; + if (iter->locks_want < downgrade_to) { + iter->locks_want = new_locks_want; - while (linked->nodes_locked && - (l = __fls(linked->nodes_locked)) >= linked->locks_want) { - if (l > linked->level) { - btree_node_unlock(linked, l); + while (iter->nodes_locked && + (l = __fls(iter->nodes_locked)) >= iter->locks_want) { + if (l > iter->level) { + btree_node_unlock(iter, l); } else { - if (btree_node_intent_locked(linked, l)) { - six_lock_downgrade(&linked->l[l].b->lock); - linked->nodes_intent_locked ^= 1 << l; + if (btree_node_intent_locked(iter, l)) { + six_lock_downgrade(&iter->l[l].b->c.lock); + iter->nodes_intent_locked ^= 1 << l; } break; } @@ -382,6 +410,14 @@ void __bch2_btree_iter_downgrade(struct btree_iter *iter, bch2_btree_trans_verify_locks(iter->trans); } +void bch2_trans_downgrade(struct btree_trans *trans) +{ + struct btree_iter *iter; + + trans_for_each_iter(trans, iter) + bch2_btree_iter_downgrade(iter); +} + /* Btree transaction locking: */ bool bch2_trans_relock(struct btree_trans *trans) @@ -408,6 +444,22 @@ void bch2_trans_unlock(struct btree_trans *trans) #ifdef CONFIG_BCACHEFS_DEBUG +static void bch2_btree_iter_verify_cached(struct btree_iter *iter) +{ + struct bkey_cached *ck; + bool locked = btree_node_locked(iter, 0); + + if (!bch2_btree_node_relock(iter, 0)) + return; + + ck = (void *) iter->l[0].b; + BUG_ON(ck->key.btree_id != iter->btree_id || + bkey_cmp(ck->key.pos, iter->pos)); + + if (!locked) + btree_node_unlock(iter, 0); +} + static void bch2_btree_iter_verify_level(struct btree_iter *iter, unsigned level) { @@ -422,6 +474,12 @@ static void bch2_btree_iter_verify_level(struct btree_iter *iter, if (!debug_check_iterators(iter->trans->c)) return; + if (btree_iter_type(iter) == BTREE_ITER_CACHED) { + if (!level) + bch2_btree_iter_verify_cached(iter); + return; + } + BUG_ON(iter->level < iter->min_depth); if (!btree_iter_node(iter, level)) @@ -513,7 +571,7 @@ void bch2_btree_trans_verify_iters(struct btree_trans *trans, struct btree *b) return; trans_for_each_iter_with_node(trans, b, iter) - bch2_btree_iter_verify_level(iter, b->level); + bch2_btree_iter_verify_level(iter, b->c.level); } #else @@ -544,7 +602,7 @@ static void __bch2_btree_iter_fix_key_modified(struct btree_iter *iter, struct btree *b, struct bkey_packed *where) { - struct btree_iter_level *l = &iter->l[b->level]; + struct btree_iter_level *l = &iter->l[b->c.level]; struct bpos pos = btree_iter_search_key(iter); if (where != bch2_btree_node_iter_peek_all(&l->iter, l->b)) @@ -564,7 +622,7 @@ void bch2_btree_iter_fix_key_modified(struct btree_iter *iter, trans_for_each_iter_with_node(iter->trans, b, linked) { __bch2_btree_iter_fix_key_modified(linked, b, where); - bch2_btree_iter_verify_level(linked, b->level); + bch2_btree_iter_verify_level(linked, b->c.level); } } @@ -634,7 +692,7 @@ fixup_done: */ if (!bch2_btree_node_iter_end(node_iter) && iter_current_key_modified && - (b->level || + (b->c.level || btree_node_type_is_extents(iter->btree_id))) { struct bset_tree *t; struct bkey_packed *k, *k2, *p; @@ -661,7 +719,7 @@ fixup_done: } } - if (!b->level && + if (!b->c.level && node_iter == &iter->l[0].iter && iter_current_key_modified) btree_iter_set_dirty(iter, BTREE_ITER_NEED_PEEK); @@ -677,7 +735,7 @@ void bch2_btree_node_iter_fix(struct btree_iter *iter, struct bset_tree *t = bch2_bkey_to_bset(b, where); struct btree_iter *linked; - if (node_iter != &iter->l[b->level].iter) { + if (node_iter != &iter->l[b->c.level].iter) { __bch2_btree_node_iter_fix(iter, b, node_iter, t, where, clobber_u64s, new_u64s); @@ -687,9 +745,9 @@ void bch2_btree_node_iter_fix(struct btree_iter *iter, trans_for_each_iter_with_node(iter->trans, b, linked) { __bch2_btree_node_iter_fix(linked, b, - &linked->l[b->level].iter, t, + &linked->l[b->c.level].iter, t, where, clobber_u64s, new_u64s); - bch2_btree_iter_verify_level(linked, b->level); + bch2_btree_iter_verify_level(linked, b->c.level); } } @@ -773,7 +831,7 @@ static void btree_iter_verify_new_node(struct btree_iter *iter, struct btree *b) if (!IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) return; - plevel = b->level + 1; + plevel = b->c.level + 1; if (!btree_iter_node(iter, plevel)) return; @@ -796,7 +854,7 @@ static void btree_iter_verify_new_node(struct btree_iter *iter, struct btree *b) } if (!parent_locked) - btree_node_unlock(iter, b->level + 1); + btree_node_unlock(iter, b->c.level + 1); } static inline void __btree_iter_init(struct btree_iter *iter, @@ -813,14 +871,16 @@ static inline void __btree_iter_init(struct btree_iter *iter, static inline void btree_iter_node_set(struct btree_iter *iter, struct btree *b) { + BUG_ON(btree_iter_type(iter) == BTREE_ITER_CACHED); + btree_iter_verify_new_node(iter, b); EBUG_ON(!btree_iter_pos_in_node(iter, b)); - EBUG_ON(b->lock.state.seq & 1); + EBUG_ON(b->c.lock.state.seq & 1); - iter->l[b->level].lock_seq = b->lock.state.seq; - iter->l[b->level].b = b; - __btree_iter_init(iter, b->level); + iter->l[b->c.level].lock_seq = b->c.lock.state.seq; + iter->l[b->c.level].b = b; + __btree_iter_init(iter, b->c.level); } /* @@ -833,18 +893,19 @@ void bch2_btree_iter_node_replace(struct btree_iter *iter, struct btree *b) struct btree_iter *linked; trans_for_each_iter(iter->trans, linked) - if (btree_iter_pos_in_node(linked, b)) { + if (btree_iter_type(linked) != BTREE_ITER_CACHED && + btree_iter_pos_in_node(linked, b)) { /* * bch2_btree_iter_node_drop() has already been called - * the old node we're replacing has already been * unlocked and the pointer invalidated */ - BUG_ON(btree_node_locked(linked, b->level)); + BUG_ON(btree_node_locked(linked, b->c.level)); - t = btree_lock_want(linked, b->level); + t = btree_lock_want(linked, b->c.level); if (t != BTREE_NODE_UNLOCKED) { - six_lock_increment(&b->lock, t); - mark_btree_node_locked(linked, b->level, t); + six_lock_increment(&b->c.lock, t); + mark_btree_node_locked(linked, b->c.level, t); } btree_iter_node_set(linked, b); @@ -854,7 +915,7 @@ void bch2_btree_iter_node_replace(struct btree_iter *iter, struct btree *b) void bch2_btree_iter_node_drop(struct btree_iter *iter, struct btree *b) { struct btree_iter *linked; - unsigned level = b->level; + unsigned level = b->c.level; trans_for_each_iter(iter->trans, linked) if (linked->l[level].b == b) { @@ -872,22 +933,30 @@ void bch2_btree_iter_reinit_node(struct btree_iter *iter, struct btree *b) struct btree_iter *linked; trans_for_each_iter_with_node(iter->trans, b, linked) - __btree_iter_init(linked, b->level); + __btree_iter_init(linked, b->c.level); +} + +static int lock_root_check_fn(struct six_lock *lock, void *p) +{ + struct btree *b = container_of(lock, struct btree, c.lock); + struct btree **rootp = p; + + return b == *rootp ? 0 : -1; } static inline int btree_iter_lock_root(struct btree_iter *iter, unsigned depth_want) { struct bch_fs *c = iter->trans->c; - struct btree *b; + struct btree *b, **rootp = &c->btree_roots[iter->btree_id].b; enum six_lock_type lock_type; unsigned i; EBUG_ON(iter->nodes_locked); while (1) { - b = READ_ONCE(c->btree_roots[iter->btree_id].b); - iter->level = READ_ONCE(b->level); + b = READ_ONCE(*rootp); + iter->level = READ_ONCE(b->c.level); if (unlikely(iter->level < depth_want)) { /* @@ -904,11 +973,12 @@ static inline int btree_iter_lock_root(struct btree_iter *iter, lock_type = __btree_lock_want(iter, iter->level); if (unlikely(!btree_node_lock(b, POS_MAX, iter->level, - iter, lock_type))) + iter, lock_type, + lock_root_check_fn, rootp))) return -EINTR; - if (likely(b == c->btree_roots[iter->btree_id].b && - b->level == iter->level && + if (likely(b == READ_ONCE(*rootp) && + b->c.level == iter->level && !race_fault())) { for (i = 0; i < iter->level; i++) iter->l[i].b = BTREE_ITER_NO_NODE_LOCK_ROOT; @@ -921,7 +991,7 @@ static inline int btree_iter_lock_root(struct btree_iter *iter, return 0; } - six_unlock_type(&b->lock, lock_type); + six_unlock_type(&b->c.lock, lock_type); } } @@ -1016,24 +1086,28 @@ static void btree_iter_up(struct btree_iter *iter) static int btree_iter_traverse_one(struct btree_iter *); -static int __btree_iter_traverse_all(struct btree_trans *trans, - struct btree_iter *orig_iter, int ret) +static int __btree_iter_traverse_all(struct btree_trans *trans, int ret) { struct bch_fs *c = trans->c; struct btree_iter *iter; u8 sorted[BTREE_ITER_MAX]; unsigned i, nr_sorted = 0; + if (trans->in_traverse_all) + return -EINTR; + + trans->in_traverse_all = true; +retry_all: + nr_sorted = 0; + trans_for_each_iter(trans, iter) - sorted[nr_sorted++] = iter - trans->iters; + sorted[nr_sorted++] = iter->idx; #define btree_iter_cmp_by_idx(_l, _r) \ btree_iter_cmp(&trans->iters[_l], &trans->iters[_r]) bubble_sort(sorted, nr_sorted, btree_iter_cmp_by_idx); #undef btree_iter_cmp_by_idx - -retry_all: bch2_trans_unlock(trans); if (unlikely(ret == -ENOMEM)) { @@ -1049,11 +1123,6 @@ retry_all: if (unlikely(ret == -EIO)) { trans->error = true; - if (orig_iter) { - orig_iter->flags |= BTREE_ITER_ERROR; - orig_iter->l[orig_iter->level].b = - BTREE_ITER_NO_NODE_ERROR; - } goto out; } @@ -1061,9 +1130,16 @@ retry_all: /* Now, redo traversals in correct order: */ for (i = 0; i < nr_sorted; i++) { - iter = &trans->iters[sorted[i]]; + unsigned idx = sorted[i]; + + /* + * sucessfully traversing one iterator can cause another to be + * unlinked, in btree_key_cache_fill() + */ + if (!(trans->iters_linked & (1ULL << idx))) + continue; - ret = btree_iter_traverse_one(iter); + ret = btree_iter_traverse_one(&trans->iters[idx]); if (ret) goto retry_all; } @@ -1078,12 +1154,14 @@ retry_all: } out: bch2_btree_cache_cannibalize_unlock(c); + + trans->in_traverse_all = false; return ret; } int bch2_btree_iter_traverse_all(struct btree_trans *trans) { - return __btree_iter_traverse_all(trans, NULL, 0); + return __btree_iter_traverse_all(trans, 0); } static inline bool btree_iter_good_node(struct btree_iter *iter, @@ -1128,9 +1206,6 @@ static int btree_iter_traverse_one(struct btree_iter *iter) { unsigned depth_want = iter->level; - if (unlikely(iter->level >= BTREE_MAX_DEPTH)) - return 0; - /* * if we need interior nodes locked, call btree_iter_relock() to make * sure we walk back up enough that we lock them: @@ -1139,9 +1214,15 @@ static int btree_iter_traverse_one(struct btree_iter *iter) iter->locks_want > 1) bch2_btree_iter_relock(iter, false); + if (btree_iter_type(iter) == BTREE_ITER_CACHED) + return bch2_btree_iter_traverse_cached(iter); + if (iter->uptodate < BTREE_ITER_NEED_RELOCK) return 0; + if (unlikely(iter->level >= BTREE_MAX_DEPTH)) + return 0; + /* * XXX: correctly using BTREE_ITER_UPTODATE should make using check_pos * here unnecessary @@ -1154,7 +1235,7 @@ static int btree_iter_traverse_one(struct btree_iter *iter) * * XXX correctly using BTREE_ITER_UPTODATE should make this unnecessary */ - if (btree_iter_node(iter, iter->level)) { + if (is_btree_node(iter, iter->level)) { BUG_ON(!btree_iter_pos_in_node(iter, iter->l[iter->level].b)); btree_iter_advance_to_pos(iter, &iter->l[iter->level], -1); @@ -1175,7 +1256,15 @@ static int btree_iter_traverse_one(struct btree_iter *iter) return 0; iter->level = depth_want; - iter->l[iter->level].b = BTREE_ITER_NO_NODE_DOWN; + + if (ret == -EIO) { + iter->flags |= BTREE_ITER_ERROR; + iter->l[iter->level].b = + BTREE_ITER_NO_NODE_ERROR; + } else { + iter->l[iter->level].b = + BTREE_ITER_NO_NODE_DOWN; + } return ret; } } @@ -1188,23 +1277,25 @@ static int btree_iter_traverse_one(struct btree_iter *iter) int __must_check __bch2_btree_iter_traverse(struct btree_iter *iter) { + struct btree_trans *trans = iter->trans; int ret; - ret = bch2_trans_cond_resched(iter->trans) ?: + ret = bch2_trans_cond_resched(trans) ?: btree_iter_traverse_one(iter); if (unlikely(ret)) - ret = __btree_iter_traverse_all(iter->trans, iter, ret); + ret = __btree_iter_traverse_all(trans, ret); return ret; } -static inline void bch2_btree_iter_checks(struct btree_iter *iter, - enum btree_iter_type type) +static inline void bch2_btree_iter_checks(struct btree_iter *iter) { + enum btree_iter_type type = btree_iter_type(iter); + EBUG_ON(iter->btree_id >= BTREE_ID_NR); - EBUG_ON(btree_iter_type(iter) != type); - BUG_ON(type == BTREE_ITER_KEYS && + BUG_ON((type == BTREE_ITER_KEYS || + type == BTREE_ITER_CACHED) && (bkey_cmp(iter->pos, bkey_start_pos(&iter->k)) < 0 || bkey_cmp(iter->pos, iter->k.p) > 0)); @@ -1219,7 +1310,8 @@ struct btree *bch2_btree_iter_peek_node(struct btree_iter *iter) struct btree *b; int ret; - bch2_btree_iter_checks(iter, BTREE_ITER_NODES); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_NODES); + bch2_btree_iter_checks(iter); if (iter->uptodate == BTREE_ITER_UPTODATE) return iter->l[iter->level].b; @@ -1247,7 +1339,8 @@ struct btree *bch2_btree_iter_next_node(struct btree_iter *iter) struct btree *b; int ret; - bch2_btree_iter_checks(iter, BTREE_ITER_NODES); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_NODES); + bch2_btree_iter_checks(iter); /* already got to end? */ if (!btree_iter_node(iter, iter->level)) @@ -1320,6 +1413,16 @@ void bch2_btree_iter_set_pos_same_leaf(struct btree_iter *iter, struct bpos new_ btree_iter_advance_to_pos(iter, l, -1); + /* + * XXX: + * keeping a node locked that's outside (even just outside) iter->pos + * breaks __bch2_btree_node_lock(). This seems to only affect + * bch2_btree_node_get_sibling so for now it's fixed there, but we + * should try to get rid of this corner case. + * + * (this behaviour is currently needed for BTREE_INSERT_NOUNLOCK) + */ + if (bch2_btree_node_iter_end(&l->iter) && btree_iter_pos_after_node(iter, l->b)) btree_iter_set_dirty(iter, BTREE_ITER_NEED_TRAVERSE); @@ -1332,6 +1435,13 @@ static void btree_iter_pos_changed(struct btree_iter *iter, int cmp) if (!cmp) goto out; + if (unlikely(btree_iter_type(iter) == BTREE_ITER_CACHED)) { + btree_node_unlock(iter, 0); + iter->l[0].b = BTREE_ITER_NO_NODE_UP; + btree_iter_set_dirty(iter, BTREE_ITER_NEED_TRAVERSE); + return; + } + l = btree_iter_up_until_good_node(iter, cmp); if (btree_iter_node(iter, l)) { @@ -1458,7 +1568,8 @@ struct bkey_s_c bch2_btree_iter_peek(struct btree_iter *iter) struct bkey_s_c k; int ret; - bch2_btree_iter_checks(iter, BTREE_ITER_KEYS); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_KEYS); + bch2_btree_iter_checks(iter); if (iter->uptodate == BTREE_ITER_UPTODATE && !bkey_deleted(&iter->k)) @@ -1545,7 +1656,8 @@ struct bkey_s_c bch2_btree_iter_peek_with_updates(struct btree_iter *iter) struct bkey_s_c k; int ret; - bch2_btree_iter_checks(iter, BTREE_ITER_KEYS); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_KEYS); + bch2_btree_iter_checks(iter); while (1) { ret = bch2_btree_iter_traverse(iter); @@ -1605,7 +1717,8 @@ struct bkey_s_c bch2_btree_iter_peek_prev(struct btree_iter *iter) struct bkey_s_c k; int ret; - bch2_btree_iter_checks(iter, BTREE_ITER_KEYS); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_KEYS); + bch2_btree_iter_checks(iter); if (iter->uptodate == BTREE_ITER_UPTODATE && !bkey_deleted(&iter->k)) @@ -1641,7 +1754,8 @@ struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *iter) { struct bpos pos = bkey_start_pos(&iter->k); - bch2_btree_iter_checks(iter, BTREE_ITER_KEYS); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_KEYS); + bch2_btree_iter_checks(iter); if (unlikely(!bkey_cmp(pos, POS_MIN))) return bkey_s_c_null; @@ -1722,7 +1836,8 @@ struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *iter) struct bkey_s_c k; int ret; - bch2_btree_iter_checks(iter, BTREE_ITER_KEYS); + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_KEYS); + bch2_btree_iter_checks(iter); if (iter->uptodate == BTREE_ITER_UPTODATE) return btree_iter_peek_uptodate(iter); @@ -1763,6 +1878,27 @@ struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *iter) return bch2_btree_iter_peek_slot(iter); } +struct bkey_s_c bch2_btree_iter_peek_cached(struct btree_iter *iter) +{ + struct bkey_cached *ck; + int ret; + + EBUG_ON(btree_iter_type(iter) != BTREE_ITER_CACHED); + bch2_btree_iter_checks(iter); + + ret = bch2_btree_iter_traverse(iter); + if (unlikely(ret)) + return bkey_s_c_err(ret); + + ck = (void *) iter->l[0].b; + + EBUG_ON(iter->btree_id != ck->key.btree_id || + bkey_cmp(iter->pos, ck->key.pos)); + BUG_ON(!ck->valid); + + return bkey_i_to_s_c(ck->k); +} + static inline void bch2_btree_iter_init(struct btree_trans *trans, struct btree_iter *iter, enum btree_id btree_id, struct bpos pos, unsigned flags) @@ -1912,7 +2048,7 @@ static struct btree_iter *btree_trans_iter_alloc(struct btree_trans *trans) struct btree_iter *iter; trans_for_each_iter(trans, iter) { - pr_err("iter: btree %s pos %llu:%llu%s%s%s %pf", + pr_err("iter: btree %s pos %llu:%llu%s%s%s %ps", bch2_btree_ids[iter->btree_id], iter->pos.inode, iter->pos.offset, @@ -1948,10 +2084,11 @@ static inline void btree_iter_copy(struct btree_iter *dst, *dst = *src; dst->idx = idx; + dst->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT; for (i = 0; i < BTREE_MAX_DEPTH; i++) if (btree_node_locked(dst, i)) - six_lock_increment(&dst->l[i].b->lock, + six_lock_increment(&dst->l[i].b->c.lock, __btree_lock_want(dst, i)); dst->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT; @@ -2006,8 +2143,9 @@ static struct btree_iter *__btree_trans_get_iter(struct btree_trans *trans, iter = best; } - iter->flags &= ~(BTREE_ITER_SLOTS|BTREE_ITER_INTENT|BTREE_ITER_PREFETCH); - iter->flags |= flags & (BTREE_ITER_SLOTS|BTREE_ITER_INTENT|BTREE_ITER_PREFETCH); + iter->flags &= ~BTREE_ITER_KEEP_UNTIL_COMMIT; + iter->flags &= ~BTREE_ITER_USER_FLAGS; + iter->flags |= flags & BTREE_ITER_USER_FLAGS; if (iter->flags & BTREE_ITER_INTENT) bch2_btree_iter_upgrade(iter, 1); @@ -2153,6 +2291,9 @@ void bch2_trans_reset(struct btree_trans *trans, unsigned flags) trans->nr_updates2 = 0; trans->mem_top = 0; + trans->extra_journal_entries = NULL; + trans->extra_journal_entry_u64s = 0; + if (trans->fs_usage_deltas) { trans->fs_usage_deltas->used = 0; memset(&trans->fs_usage_deltas->memset_start, 0, @@ -2189,12 +2330,27 @@ void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c, if (expected_mem_bytes) bch2_trans_preload_mem(trans, expected_mem_bytes); + +#ifdef CONFIG_BCACHEFS_DEBUG + trans->pid = current->pid; + mutex_lock(&c->btree_trans_lock); + list_add(&trans->list, &c->btree_trans_list); + mutex_unlock(&c->btree_trans_lock); +#endif } int bch2_trans_exit(struct btree_trans *trans) { bch2_trans_unlock(trans); +#ifdef CONFIG_BCACHEFS_DEBUG + mutex_lock(&trans->c->btree_trans_lock); + list_del(&trans->list); + mutex_unlock(&trans->c->btree_trans_lock); +#endif + + bch2_journal_preres_put(&trans->c->journal, &trans->journal_preres); + kfree(trans->fs_usage_deltas); kfree(trans->mem); if (trans->used_mempool) @@ -2207,6 +2363,69 @@ int bch2_trans_exit(struct btree_trans *trans) return trans->error ? -EIO : 0; } +static void bch2_btree_iter_node_to_text(struct printbuf *out, + struct btree_bkey_cached_common *_b, + enum btree_iter_type type) +{ + pr_buf(out, " %px l=%u %s:", + _b, _b->level, bch2_btree_ids[_b->btree_id]); + bch2_bpos_to_text(out, btree_node_pos(_b, type)); +} + +void bch2_btree_trans_to_text(struct printbuf *out, struct bch_fs *c) +{ +#ifdef CONFIG_BCACHEFS_DEBUG + struct btree_trans *trans; + struct btree_iter *iter; + struct btree *b; + unsigned l; + + mutex_lock(&c->btree_trans_lock); + list_for_each_entry(trans, &c->btree_trans_list, list) { + pr_buf(out, "%i %px %ps\n", trans->pid, trans, (void *) trans->ip); + + trans_for_each_iter(trans, iter) { + if (!iter->nodes_locked) + continue; + + pr_buf(out, " iter %u %s:", + iter->idx, + bch2_btree_ids[iter->btree_id]); + bch2_bpos_to_text(out, iter->pos); + pr_buf(out, "\n"); + + for (l = 0; l < BTREE_MAX_DEPTH; l++) { + if (btree_node_locked(iter, l)) { + pr_buf(out, " %s l=%u ", + btree_node_intent_locked(iter, l) ? "i" : "r", l); + bch2_btree_iter_node_to_text(out, + (void *) iter->l[l].b, + btree_iter_type(iter)); + pr_buf(out, "\n"); + } + } + } + + b = READ_ONCE(trans->locking); + if (b) { + pr_buf(out, " locking iter %u l=%u %s:", + trans->locking_iter_idx, + trans->locking_level, + bch2_btree_ids[trans->locking_btree_id]); + bch2_bpos_to_text(out, trans->locking_pos); + + + pr_buf(out, " node "); + bch2_btree_iter_node_to_text(out, + (void *) b, + btree_iter_type(&trans->iters[trans->locking_iter_idx])); + pr_buf(out, "\n"); + } + } + mutex_unlock(&c->btree_trans_lock); +#endif +} + void bch2_fs_btree_iter_exit(struct bch_fs *c) { mempool_exit(&c->btree_iters_pool); @@ -2216,6 +2435,9 @@ int bch2_fs_btree_iter_init(struct bch_fs *c) { unsigned nr = BTREE_ITER_MAX; + INIT_LIST_HEAD(&c->btree_trans_list); + mutex_init(&c->btree_trans_lock); + return mempool_init_kmalloc_pool(&c->btree_iters_pool, 1, sizeof(struct btree_iter) * nr + sizeof(struct btree_insert_entry) * nr + diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h index 6456787a8f77..bd9ec3ec9a92 100644 --- a/fs/bcachefs/btree_iter.h +++ b/fs/bcachefs/btree_iter.h @@ -27,13 +27,13 @@ static inline bool btree_node_lock_seq_matches(const struct btree_iter *iter, * that write lock. The lock sequence number is incremented by taking * and releasing write locks and is even when unlocked: */ - return iter->l[level].lock_seq >> 1 == b->lock.state.seq >> 1; + return iter->l[level].lock_seq >> 1 == b->c.lock.state.seq >> 1; } static inline struct btree *btree_node_parent(struct btree_iter *iter, struct btree *b) { - return btree_iter_node(iter, b->level + 1); + return btree_iter_node(iter, b->c.level + 1); } static inline bool btree_trans_has_multiple_iters(const struct btree_trans *trans) @@ -73,8 +73,8 @@ __trans_next_iter(struct btree_trans *trans, unsigned idx) static inline bool __iter_has_node(const struct btree_iter *iter, const struct btree *b) { - return iter->l[b->level].b == b && - btree_node_lock_seq_matches(iter, b, b->level); + return iter->l[b->c.level].b == b && + btree_node_lock_seq_matches(iter, b, b->c.level); } static inline struct btree_iter * @@ -110,6 +110,7 @@ void bch2_btree_node_iter_fix(struct btree_iter *, struct btree *, struct btree_node_iter *, struct bkey_packed *, unsigned, unsigned); +bool bch2_btree_iter_relock(struct btree_iter *, bool); bool bch2_trans_relock(struct btree_trans *); void bch2_trans_unlock(struct btree_trans *); @@ -136,6 +137,8 @@ static inline void bch2_btree_iter_downgrade(struct btree_iter *iter) __bch2_btree_iter_downgrade(iter, 0); } +void bch2_trans_downgrade(struct btree_trans *); + void bch2_btree_iter_node_replace(struct btree_iter *, struct btree *); void bch2_btree_iter_node_drop(struct btree_iter *, struct btree *); @@ -168,21 +171,18 @@ struct bkey_s_c bch2_btree_iter_prev(struct btree_iter *); struct bkey_s_c bch2_btree_iter_peek_slot(struct btree_iter *); struct bkey_s_c bch2_btree_iter_next_slot(struct btree_iter *); +struct bkey_s_c bch2_btree_iter_peek_cached(struct btree_iter *); + void bch2_btree_iter_set_pos_same_leaf(struct btree_iter *, struct bpos); void __bch2_btree_iter_set_pos(struct btree_iter *, struct bpos, bool); void bch2_btree_iter_set_pos(struct btree_iter *, struct bpos); -static inline int __btree_iter_cmp(enum btree_id id, - struct bpos pos, - const struct btree_iter *r) -{ - return cmp_int(id, r->btree_id) ?: bkey_cmp(pos, r->pos); -} - static inline int btree_iter_cmp(const struct btree_iter *l, const struct btree_iter *r) { - return __btree_iter_cmp(l->btree_id, l->pos, r); + return cmp_int(l->btree_id, r->btree_id) ?: + -cmp_int(btree_iter_type(l), btree_iter_type(r)) ?: + bkey_cmp(l->pos, r->pos); } /* @@ -216,9 +216,12 @@ static inline int bch2_trans_cond_resched(struct btree_trans *trans) static inline struct bkey_s_c __bch2_btree_iter_peek(struct btree_iter *iter, unsigned flags) { - return flags & BTREE_ITER_SLOTS - ? bch2_btree_iter_peek_slot(iter) - : bch2_btree_iter_peek(iter); + if ((flags & BTREE_ITER_TYPE) == BTREE_ITER_CACHED) + return bch2_btree_iter_peek_cached(iter); + else + return flags & BTREE_ITER_SLOTS + ? bch2_btree_iter_peek_slot(iter) + : bch2_btree_iter_peek(iter); } static inline struct bkey_s_c __bch2_btree_iter_next(struct btree_iter *iter, @@ -303,6 +306,8 @@ void *bch2_trans_kmalloc(struct btree_trans *, size_t); void bch2_trans_init(struct btree_trans *, struct bch_fs *, unsigned, size_t); int bch2_trans_exit(struct btree_trans *); +void bch2_btree_trans_to_text(struct printbuf *, struct bch_fs *); + void bch2_fs_btree_iter_exit(struct bch_fs *); int bch2_fs_btree_iter_init(struct bch_fs *); diff --git a/fs/bcachefs/btree_key_cache.c b/fs/bcachefs/btree_key_cache.c new file mode 100644 index 000000000000..d73cc8ddadac --- /dev/null +++ b/fs/bcachefs/btree_key_cache.c @@ -0,0 +1,519 @@ + +#include "bcachefs.h" +#include "btree_cache.h" +#include "btree_iter.h" +#include "btree_key_cache.h" +#include "btree_locking.h" +#include "btree_update.h" +#include "error.h" +#include "journal.h" +#include "journal_reclaim.h" + +#include <trace/events/bcachefs.h> + +static int bch2_btree_key_cache_cmp_fn(struct rhashtable_compare_arg *arg, + const void *obj) +{ + const struct bkey_cached *ck = obj; + const struct bkey_cached_key *key = arg->key; + + return cmp_int(ck->key.btree_id, key->btree_id) ?: + bkey_cmp(ck->key.pos, key->pos); +} + +static const struct rhashtable_params bch2_btree_key_cache_params = { + .head_offset = offsetof(struct bkey_cached, hash), + .key_offset = offsetof(struct bkey_cached, key), + .key_len = sizeof(struct bkey_cached_key), + .obj_cmpfn = bch2_btree_key_cache_cmp_fn, +}; + +__flatten +static inline struct bkey_cached * +btree_key_cache_find(struct bch_fs *c, enum btree_id btree_id, struct bpos pos) +{ + struct bkey_cached_key key = { + .btree_id = btree_id, + .pos = pos, + }; + + return rhashtable_lookup_fast(&c->btree_key_cache.table, &key, + bch2_btree_key_cache_params); +} + +static bool bkey_cached_lock_for_evict(struct bkey_cached *ck) +{ + if (!six_trylock_intent(&ck->c.lock)) + return false; + + if (!six_trylock_write(&ck->c.lock)) { + six_unlock_intent(&ck->c.lock); + return false; + } + + if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) { + six_unlock_write(&ck->c.lock); + six_unlock_intent(&ck->c.lock); + return false; + } + + return true; +} + +static void bkey_cached_evict(struct btree_key_cache *c, + struct bkey_cached *ck) +{ + BUG_ON(rhashtable_remove_fast(&c->table, &ck->hash, + bch2_btree_key_cache_params)); + memset(&ck->key, ~0, sizeof(ck->key)); +} + +static void bkey_cached_free(struct btree_key_cache *c, + struct bkey_cached *ck) +{ + list_move(&ck->list, &c->freed); + + kfree(ck->k); + ck->k = NULL; + ck->u64s = 0; + + six_unlock_write(&ck->c.lock); + six_unlock_intent(&ck->c.lock); +} + +static struct bkey_cached * +bkey_cached_alloc(struct btree_key_cache *c) +{ + struct bkey_cached *ck; + + list_for_each_entry(ck, &c->freed, list) + if (bkey_cached_lock_for_evict(ck)) + return ck; + + list_for_each_entry(ck, &c->clean, list) + if (bkey_cached_lock_for_evict(ck)) { + bkey_cached_evict(c, ck); + return ck; + } + + ck = kzalloc(sizeof(*ck), GFP_NOFS); + if (!ck) + return NULL; + + INIT_LIST_HEAD(&ck->list); + six_lock_init(&ck->c.lock); + BUG_ON(!six_trylock_intent(&ck->c.lock)); + BUG_ON(!six_trylock_write(&ck->c.lock)); + + return ck; +} + +static struct bkey_cached * +btree_key_cache_create(struct btree_key_cache *c, + enum btree_id btree_id, + struct bpos pos) +{ + struct bkey_cached *ck; + + ck = bkey_cached_alloc(c); + if (!ck) + return ERR_PTR(-ENOMEM); + + ck->c.level = 0; + ck->c.btree_id = btree_id; + ck->key.btree_id = btree_id; + ck->key.pos = pos; + ck->valid = false; + + BUG_ON(ck->flags); + + if (rhashtable_lookup_insert_fast(&c->table, + &ck->hash, + bch2_btree_key_cache_params)) { + /* We raced with another fill: */ + bkey_cached_free(c, ck); + return NULL; + } + + list_move(&ck->list, &c->clean); + six_unlock_write(&ck->c.lock); + + return ck; +} + +static int btree_key_cache_fill(struct btree_trans *trans, + struct btree_iter *ck_iter, + struct bkey_cached *ck) +{ + struct btree_iter *iter; + struct bkey_s_c k; + unsigned new_u64s = 0; + struct bkey_i *new_k = NULL; + int ret; + + iter = bch2_trans_get_iter(trans, ck->key.btree_id, + ck->key.pos, BTREE_ITER_SLOTS); + if (IS_ERR(iter)) + return PTR_ERR(iter); + + k = bch2_btree_iter_peek_slot(iter); + ret = bkey_err(k); + if (ret) { + bch2_trans_iter_put(trans, iter); + return ret; + } + + if (!bch2_btree_node_relock(ck_iter, 0)) { + bch2_trans_iter_put(trans, iter); + trace_transaction_restart_ip(trans->ip, _THIS_IP_); + return -EINTR; + } + + if (k.k->u64s > ck->u64s) { + new_u64s = roundup_pow_of_two(k.k->u64s); + new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOFS); + if (!new_k) { + bch2_trans_iter_put(trans, iter); + return -ENOMEM; + } + } + + bch2_btree_node_lock_write(ck_iter->l[0].b, ck_iter); + if (new_k) { + kfree(ck->k); + ck->u64s = new_u64s; + ck->k = new_k; + } + + bkey_reassemble(ck->k, k); + ck->valid = true; + bch2_btree_node_unlock_write(ck_iter->l[0].b, ck_iter); + + /* We're not likely to need this iterator again: */ + bch2_trans_iter_free(trans, iter); + + return 0; +} + +static int bkey_cached_check_fn(struct six_lock *lock, void *p) +{ + struct bkey_cached *ck = container_of(lock, struct bkey_cached, c.lock); + const struct btree_iter *iter = p; + + return ck->key.btree_id == iter->btree_id && + !bkey_cmp(ck->key.pos, iter->pos) ? 0 : -1; +} + +int bch2_btree_iter_traverse_cached(struct btree_iter *iter) +{ + struct btree_trans *trans = iter->trans; + struct bch_fs *c = trans->c; + struct bkey_cached *ck; + int ret = 0; + + BUG_ON(iter->level); + + if (btree_node_locked(iter, 0)) { + ck = (void *) iter->l[0].b; + goto fill; + } +retry: + ck = btree_key_cache_find(c, iter->btree_id, iter->pos); + if (!ck) { + if (iter->flags & BTREE_ITER_CACHED_NOCREATE) { + iter->l[0].b = NULL; + return 0; + } + + mutex_lock(&c->btree_key_cache.lock); + ck = btree_key_cache_create(&c->btree_key_cache, + iter->btree_id, iter->pos); + mutex_unlock(&c->btree_key_cache.lock); + + ret = PTR_ERR_OR_ZERO(ck); + if (ret) + goto err; + if (!ck) + goto retry; + + mark_btree_node_locked(iter, 0, SIX_LOCK_intent); + iter->locks_want = 1; + } else { + enum six_lock_type lock_want = __btree_lock_want(iter, 0); + + if (!btree_node_lock((void *) ck, iter->pos, 0, iter, lock_want, + bkey_cached_check_fn, iter)) { + if (ck->key.btree_id != iter->btree_id || + bkey_cmp(ck->key.pos, iter->pos)) { + goto retry; + } + + trace_transaction_restart_ip(trans->ip, _THIS_IP_); + ret = -EINTR; + goto err; + } + + if (ck->key.btree_id != iter->btree_id || + bkey_cmp(ck->key.pos, iter->pos)) { + six_unlock_type(&ck->c.lock, lock_want); + goto retry; + } + + mark_btree_node_locked(iter, 0, lock_want); + } + + iter->l[0].lock_seq = ck->c.lock.state.seq; + iter->l[0].b = (void *) ck; +fill: + if (!ck->valid && !(iter->flags & BTREE_ITER_CACHED_NOFILL)) { + if (!btree_node_intent_locked(iter, 0)) + bch2_btree_iter_upgrade(iter, 1); + if (!btree_node_intent_locked(iter, 0)) { + trace_transaction_restart_ip(trans->ip, _THIS_IP_); + ret = -EINTR; + goto err; + } + + ret = btree_key_cache_fill(trans, iter, ck); + if (ret) + goto err; + } + + iter->uptodate = BTREE_ITER_NEED_PEEK; + bch2_btree_iter_downgrade(iter); + return ret; +err: + if (ret != -EINTR) { + btree_node_unlock(iter, 0); + iter->flags |= BTREE_ITER_ERROR; + iter->l[0].b = BTREE_ITER_NO_NODE_ERROR; + } + return ret; +} + +static int btree_key_cache_flush_pos(struct btree_trans *trans, + struct bkey_cached_key key, + u64 journal_seq, + bool evict) +{ + struct bch_fs *c = trans->c; + struct journal *j = &c->journal; + struct btree_iter *c_iter = NULL, *b_iter = NULL; + struct bkey_cached *ck; + int ret; + + b_iter = bch2_trans_get_iter(trans, key.btree_id, key.pos, + BTREE_ITER_SLOTS| + BTREE_ITER_INTENT); + ret = PTR_ERR_OR_ZERO(b_iter); + if (ret) + goto out; + + c_iter = bch2_trans_get_iter(trans, key.btree_id, key.pos, + BTREE_ITER_CACHED| + BTREE_ITER_CACHED_NOFILL| + BTREE_ITER_CACHED_NOCREATE| + BTREE_ITER_INTENT); + ret = PTR_ERR_OR_ZERO(c_iter); + if (ret) + goto out; +retry: + ret = bch2_btree_iter_traverse(c_iter); + if (ret) + goto err; + + ck = (void *) c_iter->l[0].b; + if (!ck || + (journal_seq && ck->journal.seq != journal_seq)) + goto out; + + if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) { + if (!evict) + goto out; + goto evict; + } + + ret = bch2_btree_iter_traverse(b_iter) ?: + bch2_trans_update(trans, b_iter, ck->k, BTREE_TRIGGER_NORUN) ?: + bch2_trans_commit(trans, NULL, NULL, + BTREE_INSERT_NOUNLOCK| + BTREE_INSERT_NOCHECK_RW| + BTREE_INSERT_NOFAIL| + BTREE_INSERT_USE_RESERVE| + BTREE_INSERT_USE_ALLOC_RESERVE| + BTREE_INSERT_JOURNAL_RESERVED| + BTREE_INSERT_JOURNAL_RECLAIM); +err: + if (ret == -EINTR) + goto retry; + + BUG_ON(ret && !bch2_journal_error(j)); + + if (ret) + goto out; + + bch2_journal_pin_drop(j, &ck->journal); + bch2_journal_preres_put(j, &ck->res); + clear_bit(BKEY_CACHED_DIRTY, &ck->flags); + + if (!evict) { + mutex_lock(&c->btree_key_cache.lock); + list_move_tail(&ck->list, &c->btree_key_cache.clean); + mutex_unlock(&c->btree_key_cache.lock); + } else { +evict: + BUG_ON(!btree_node_intent_locked(c_iter, 0)); + + mark_btree_node_unlocked(c_iter, 0); + c_iter->l[0].b = NULL; + + six_lock_write(&ck->c.lock, NULL, NULL); + + mutex_lock(&c->btree_key_cache.lock); + bkey_cached_evict(&c->btree_key_cache, ck); + bkey_cached_free(&c->btree_key_cache, ck); + mutex_unlock(&c->btree_key_cache.lock); + } +out: + bch2_trans_iter_put(trans, b_iter); + bch2_trans_iter_put(trans, c_iter); + return ret; +} + +static void btree_key_cache_journal_flush(struct journal *j, + struct journal_entry_pin *pin, + u64 seq) +{ + struct bch_fs *c = container_of(j, struct bch_fs, journal); + struct bkey_cached *ck = + container_of(pin, struct bkey_cached, journal); + struct bkey_cached_key key; + struct btree_trans trans; + + six_lock_read(&ck->c.lock, NULL, NULL); + key = READ_ONCE(ck->key); + + if (ck->journal.seq != seq || + !test_bit(BKEY_CACHED_DIRTY, &ck->flags)) { + six_unlock_read(&ck->c.lock); + return; + } + six_unlock_read(&ck->c.lock); + + bch2_trans_init(&trans, c, 0, 0); + btree_key_cache_flush_pos(&trans, key, seq, false); + bch2_trans_exit(&trans); +} + +/* + * Flush and evict a key from the key cache: + */ +int bch2_btree_key_cache_flush(struct btree_trans *trans, + enum btree_id id, struct bpos pos) +{ + struct bch_fs *c = trans->c; + struct bkey_cached_key key = { id, pos }; + + /* Fastpath - assume it won't be found: */ + if (!btree_key_cache_find(c, id, pos)) + return 0; + + return btree_key_cache_flush_pos(trans, key, 0, true); +} + +bool bch2_btree_insert_key_cached(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_i *insert) +{ + struct bch_fs *c = trans->c; + struct bkey_cached *ck = (void *) iter->l[0].b; + + BUG_ON(insert->u64s > ck->u64s); + + if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) { + int difference; + + BUG_ON(jset_u64s(insert->u64s) > trans->journal_preres.u64s); + + difference = jset_u64s(insert->u64s) - ck->res.u64s; + if (difference > 0) { + trans->journal_preres.u64s -= difference; + ck->res.u64s += difference; + } + } + + bkey_copy(ck->k, insert); + ck->valid = true; + + if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) { + mutex_lock(&c->btree_key_cache.lock); + list_del_init(&ck->list); + + set_bit(BKEY_CACHED_DIRTY, &ck->flags); + mutex_unlock(&c->btree_key_cache.lock); + } + + bch2_journal_pin_update(&c->journal, trans->journal_res.seq, + &ck->journal, btree_key_cache_journal_flush); + return true; +} + +#ifdef CONFIG_BCACHEFS_DEBUG +void bch2_btree_key_cache_verify_clean(struct btree_trans *trans, + enum btree_id id, struct bpos pos) +{ + BUG_ON(btree_key_cache_find(trans->c, id, pos)); +} +#endif + +void bch2_fs_btree_key_cache_exit(struct btree_key_cache *c) +{ + struct bkey_cached *ck, *n; + + mutex_lock(&c->lock); + list_for_each_entry_safe(ck, n, &c->clean, list) { + kfree(ck->k); + kfree(ck); + } + list_for_each_entry_safe(ck, n, &c->freed, list) + kfree(ck); + mutex_unlock(&c->lock); + + rhashtable_destroy(&c->table); +} + +void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *c) +{ + mutex_init(&c->lock); + INIT_LIST_HEAD(&c->freed); + INIT_LIST_HEAD(&c->clean); +} + +int bch2_fs_btree_key_cache_init(struct btree_key_cache *c) +{ + return rhashtable_init(&c->table, &bch2_btree_key_cache_params); +} + +void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c) +{ + struct bucket_table *tbl; + struct bkey_cached *ck; + struct rhash_head *pos; + size_t i; + + mutex_lock(&c->lock); + tbl = rht_dereference_rcu(c->table.tbl, &c->table); + + for (i = 0; i < tbl->size; i++) { + rht_for_each_entry_rcu(ck, pos, tbl, i, hash) { + pr_buf(out, "%s:", + bch2_btree_ids[ck->key.btree_id]); + bch2_bpos_to_text(out, ck->key.pos); + + if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) + pr_buf(out, " journal seq %llu", ck->journal.seq); + pr_buf(out, "\n"); + } + } + mutex_unlock(&c->lock); +} diff --git a/fs/bcachefs/btree_key_cache.h b/fs/bcachefs/btree_key_cache.h new file mode 100644 index 000000000000..b1756c6c622c --- /dev/null +++ b/fs/bcachefs/btree_key_cache.h @@ -0,0 +1,25 @@ +#ifndef _BCACHEFS_BTREE_KEY_CACHE_H +#define _BCACHEFS_BTREE_KEY_CACHE_H + +int bch2_btree_iter_traverse_cached(struct btree_iter *); + +bool bch2_btree_insert_key_cached(struct btree_trans *, + struct btree_iter *, struct bkey_i *); +int bch2_btree_key_cache_flush(struct btree_trans *, + enum btree_id, struct bpos); +#ifdef CONFIG_BCACHEFS_DEBUG +void bch2_btree_key_cache_verify_clean(struct btree_trans *, + enum btree_id, struct bpos); +#else +static inline void +bch2_btree_key_cache_verify_clean(struct btree_trans *trans, + enum btree_id id, struct bpos pos) {} +#endif + +void bch2_fs_btree_key_cache_exit(struct btree_key_cache *); +void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *); +int bch2_fs_btree_key_cache_init(struct btree_key_cache *); + +void bch2_btree_key_cache_to_text(struct printbuf *, struct btree_key_cache *); + +#endif /* _BCACHEFS_BTREE_KEY_CACHE_H */ diff --git a/fs/bcachefs/btree_locking.h b/fs/bcachefs/btree_locking.h index 9081d3fc238a..81fbf3e18647 100644 --- a/fs/bcachefs/btree_locking.h +++ b/fs/bcachefs/btree_locking.h @@ -102,7 +102,7 @@ static inline void __btree_node_unlock(struct btree_iter *iter, unsigned level) EBUG_ON(level >= BTREE_MAX_DEPTH); if (lock_type != BTREE_NODE_UNLOCKED) - six_unlock_type(&iter->l[level].b->lock, lock_type); + six_unlock_type(&iter->l[level].b->c.lock, lock_type); mark_btree_node_unlocked(iter, level); } @@ -143,14 +143,14 @@ static inline void __btree_node_lock_type(struct bch_fs *c, struct btree *b, { u64 start_time = local_clock(); - six_lock_type(&b->lock, type); + six_lock_type(&b->c.lock, type, NULL, NULL); bch2_time_stats_update(&c->times[lock_to_time_stat(type)], start_time); } static inline void btree_node_lock_type(struct bch_fs *c, struct btree *b, enum six_lock_type type) { - if (!six_trylock_type(&b->lock, type)) + if (!six_trylock_type(&b->c.lock, type)) __btree_node_lock_type(c, b, type); } @@ -158,16 +158,16 @@ static inline void btree_node_lock_type(struct bch_fs *c, struct btree *b, * Lock a btree node if we already have it locked on one of our linked * iterators: */ -static inline bool btree_node_lock_increment(struct btree_iter *iter, +static inline bool btree_node_lock_increment(struct btree_trans *trans, struct btree *b, unsigned level, enum btree_node_locked_type want) { - struct btree_iter *linked; + struct btree_iter *iter; - trans_for_each_iter(iter->trans, linked) - if (linked->l[level].b == b && - btree_node_locked_type(linked, level) >= want) { - six_lock_increment(&b->lock, want); + trans_for_each_iter(trans, iter) + if (iter->l[level].b == b && + btree_node_locked_type(iter, level) >= want) { + six_lock_increment(&b->c.lock, want); return true; } @@ -175,18 +175,37 @@ static inline bool btree_node_lock_increment(struct btree_iter *iter, } bool __bch2_btree_node_lock(struct btree *, struct bpos, unsigned, - struct btree_iter *, enum six_lock_type); - -static inline bool btree_node_lock(struct btree *b, struct bpos pos, - unsigned level, - struct btree_iter *iter, - enum six_lock_type type) + struct btree_iter *, enum six_lock_type, + six_lock_should_sleep_fn, void *); + +static inline bool btree_node_lock(struct btree *b, + struct bpos pos, unsigned level, + struct btree_iter *iter, + enum six_lock_type type, + six_lock_should_sleep_fn should_sleep_fn, void *p) { - EBUG_ON(level >= BTREE_MAX_DEPTH); + struct btree_trans *trans = iter->trans; + bool ret; - return likely(six_trylock_type(&b->lock, type)) || - btree_node_lock_increment(iter, b, level, type) || - __bch2_btree_node_lock(b, pos, level, iter, type); + EBUG_ON(level >= BTREE_MAX_DEPTH); + EBUG_ON(!(trans->iters_linked & (1ULL << iter->idx))); + +#ifdef CONFIG_BCACHEFS_DEBUG + trans->locking = b; + trans->locking_iter_idx = iter->idx; + trans->locking_pos = pos; + trans->locking_btree_id = iter->btree_id; + trans->locking_level = level; +#endif + ret = likely(six_trylock_type(&b->c.lock, type)) || + btree_node_lock_increment(trans, b, level, type) || + __bch2_btree_node_lock(b, pos, level, iter, type, + should_sleep_fn, p); + +#ifdef CONFIG_BCACHEFS_DEBUG + trans->locking = NULL; +#endif + return ret; } bool __bch2_btree_node_relock(struct btree_iter *, unsigned); @@ -211,13 +230,13 @@ bch2_btree_node_unlock_write_inlined(struct btree *b, struct btree_iter *iter) { struct btree_iter *linked; - EBUG_ON(iter->l[b->level].b != b); - EBUG_ON(iter->l[b->level].lock_seq + 1 != b->lock.state.seq); + EBUG_ON(iter->l[b->c.level].b != b); + EBUG_ON(iter->l[b->c.level].lock_seq + 1 != b->c.lock.state.seq); trans_for_each_iter_with_node(iter->trans, b, linked) - linked->l[b->level].lock_seq += 2; + linked->l[b->c.level].lock_seq += 2; - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); } void bch2_btree_node_unlock_write(struct btree *, struct btree_iter *); @@ -226,10 +245,10 @@ void __bch2_btree_node_lock_write(struct btree *, struct btree_iter *); static inline void bch2_btree_node_lock_write(struct btree *b, struct btree_iter *iter) { - EBUG_ON(iter->l[b->level].b != b); - EBUG_ON(iter->l[b->level].lock_seq != b->lock.state.seq); + EBUG_ON(iter->l[b->c.level].b != b); + EBUG_ON(iter->l[b->c.level].lock_seq != b->c.lock.state.seq); - if (unlikely(!six_trylock_write(&b->lock))) + if (unlikely(!six_trylock_write(&b->c.lock))) __bch2_btree_node_lock_write(b, iter); } diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h index 732cdc35aa7c..16c4d058358b 100644 --- a/fs/bcachefs/btree_types.h +++ b/fs/bcachefs/btree_types.h @@ -60,17 +60,20 @@ struct btree_alloc { BKEY_PADDED(k); }; +struct btree_bkey_cached_common { + struct six_lock lock; + u8 level; + u8 btree_id; +}; + struct btree { - /* Hottest entries first */ + struct btree_bkey_cached_common c; + struct rhash_head hash; u64 hash_val; - struct six_lock lock; - unsigned long flags; u16 written; - u8 level; - u8 btree_id; u8 nsets; u8 nr_key_bits; @@ -180,6 +183,7 @@ struct btree_node_iter { enum btree_iter_type { BTREE_ITER_KEYS, BTREE_ITER_NODES, + BTREE_ITER_CACHED, }; #define BTREE_ITER_TYPE ((1 << 2) - 1) @@ -211,6 +215,15 @@ enum btree_iter_type { #define BTREE_ITER_IS_EXTENTS (1 << 6) #define BTREE_ITER_ERROR (1 << 7) #define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 8) +#define BTREE_ITER_CACHED_NOFILL (1 << 9) +#define BTREE_ITER_CACHED_NOCREATE (1 << 10) + +#define BTREE_ITER_USER_FLAGS \ + (BTREE_ITER_SLOTS \ + |BTREE_ITER_INTENT \ + |BTREE_ITER_PREFETCH \ + |BTREE_ITER_CACHED_NOFILL \ + |BTREE_ITER_CACHED_NOCREATE) enum btree_iter_uptodate { BTREE_ITER_UPTODATE = 0, @@ -219,6 +232,14 @@ enum btree_iter_uptodate { BTREE_ITER_NEED_TRAVERSE = 3, }; +#define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1) +#define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2) +#define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3) +#define BTREE_ITER_NO_NODE_UP ((struct btree *) 4) +#define BTREE_ITER_NO_NODE_DOWN ((struct btree *) 5) +#define BTREE_ITER_NO_NODE_INIT ((struct btree *) 6) +#define BTREE_ITER_NO_NODE_ERROR ((struct btree *) 7) + /* * @pos - iterator's current position * @level - current btree depth @@ -256,7 +277,8 @@ struct btree_iter { unsigned long ip_allocated; }; -static inline enum btree_iter_type btree_iter_type(struct btree_iter *iter) +static inline enum btree_iter_type +btree_iter_type(const struct btree_iter *iter) { return iter->flags & BTREE_ITER_TYPE; } @@ -266,6 +288,37 @@ static inline struct btree_iter_level *iter_l(struct btree_iter *iter) return iter->l + iter->level; } +struct btree_key_cache { + struct mutex lock; + struct rhashtable table; + struct list_head freed; + struct list_head clean; +}; + +struct bkey_cached_key { + u32 btree_id; + struct bpos pos; +} __attribute__((packed, aligned(4))); + +#define BKEY_CACHED_DIRTY 0 + +struct bkey_cached { + struct btree_bkey_cached_common c; + + unsigned long flags; + u8 u64s; + bool valid; + struct bkey_cached_key key; + + struct rhash_head hash; + struct list_head list; + + struct journal_preres res; + struct journal_entry_pin journal; + + struct bkey_i *k; +}; + struct btree_insert_entry { unsigned trigger_flags; unsigned trans_triggers_run:1; @@ -281,6 +334,15 @@ struct btree_insert_entry { struct btree_trans { struct bch_fs *c; +#ifdef CONFIG_BCACHEFS_DEBUG + struct list_head list; + struct btree *locking; + unsigned locking_iter_idx; + struct bpos locking_pos; + u8 locking_btree_id; + u8 locking_level; + pid_t pid; +#endif unsigned long ip; u64 iters_linked; @@ -295,6 +357,7 @@ struct btree_trans { unsigned error:1; unsigned nounlock:1; unsigned need_reset:1; + unsigned in_traverse_all:1; unsigned mem_top; unsigned mem_bytes; @@ -305,6 +368,10 @@ struct btree_trans { struct btree_insert_entry *updates2; /* update path: */ + struct jset_entry *extra_journal_entries; + unsigned extra_journal_entry_u64s; + struct journal_entry_pin *journal_pin; + struct journal_res journal_res; struct journal_preres journal_preres; u64 *journal_seq; @@ -482,7 +549,7 @@ static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_ /* Type of keys @b contains: */ static inline enum btree_node_type btree_node_type(struct btree *b) { - return __btree_node_type(b->level, b->btree_id); + return __btree_node_type(b->c.level, b->c.btree_id); } static inline bool btree_node_type_is_extents(enum btree_node_type type) @@ -501,6 +568,16 @@ static inline bool btree_node_is_extents(struct btree *b) return btree_node_type_is_extents(btree_node_type(b)); } +static inline enum btree_node_type btree_iter_key_type(struct btree_iter *iter) +{ + return __btree_node_type(iter->level, iter->btree_id); +} + +static inline bool btree_iter_is_extents(struct btree_iter *iter) +{ + return btree_node_type_is_extents(btree_iter_key_type(iter)); +} + #define BTREE_NODE_TYPE_HAS_TRIGGERS \ ((1U << BKEY_TYPE_EXTENTS)| \ (1U << BKEY_TYPE_ALLOC)| \ diff --git a/fs/bcachefs/btree_update.h b/fs/bcachefs/btree_update.h index 11f7d02de622..e0b1bde37484 100644 --- a/fs/bcachefs/btree_update.h +++ b/fs/bcachefs/btree_update.h @@ -23,6 +23,7 @@ enum btree_insert_flags { __BTREE_INSERT_USE_ALLOC_RESERVE, __BTREE_INSERT_JOURNAL_REPLAY, __BTREE_INSERT_JOURNAL_RESERVED, + __BTREE_INSERT_JOURNAL_RECLAIM, __BTREE_INSERT_NOWAIT, __BTREE_INSERT_GC_LOCK_HELD, __BCH_HASH_SET_MUST_CREATE, @@ -47,8 +48,12 @@ enum btree_insert_flags { /* Insert is for journal replay - don't get journal reservations: */ #define BTREE_INSERT_JOURNAL_REPLAY (1 << __BTREE_INSERT_JOURNAL_REPLAY) +/* Indicates that we have pre-reserved space in the journal: */ #define BTREE_INSERT_JOURNAL_RESERVED (1 << __BTREE_INSERT_JOURNAL_RESERVED) +/* Insert is being called from journal reclaim path: */ +#define BTREE_INSERT_JOURNAL_RECLAIM (1 << __BTREE_INSERT_JOURNAL_RECLAIM) + /* Don't block on allocation failure (for new btree nodes: */ #define BTREE_INSERT_NOWAIT (1 << __BTREE_INSERT_NOWAIT) #define BTREE_INSERT_GC_LOCK_HELD (1 << __BTREE_INSERT_GC_LOCK_HELD) diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c index 75b70187a954..a8cd6ffb6c7c 100644 --- a/fs/bcachefs/btree_update_interior.c +++ b/fs/bcachefs/btree_update_interior.c @@ -21,10 +21,6 @@ #include <linux/random.h> #include <trace/events/bcachefs.h> -static void btree_node_will_make_reachable(struct btree_update *, - struct btree *); -static void btree_update_drop_new_node(struct bch_fs *, struct btree *); - /* Debug code: */ /* @@ -39,7 +35,7 @@ static void btree_node_interior_verify(struct btree *b) struct bkey_s_c_btree_ptr_v2 bp; struct bkey unpacked; - BUG_ON(!b->level); + BUG_ON(!b->c.level); bch2_btree_node_iter_init_from_start(&iter, b); @@ -124,74 +120,6 @@ bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *b, /* Btree node freeing/allocation: */ -static bool btree_key_matches(struct bch_fs *c, - struct bkey_s_c l, - struct bkey_s_c r) -{ - struct bkey_ptrs_c ptrs1 = bch2_bkey_ptrs_c(l); - struct bkey_ptrs_c ptrs2 = bch2_bkey_ptrs_c(r); - const struct bch_extent_ptr *ptr1, *ptr2; - - bkey_for_each_ptr(ptrs1, ptr1) - bkey_for_each_ptr(ptrs2, ptr2) - if (ptr1->dev == ptr2->dev && - ptr1->gen == ptr2->gen && - ptr1->offset == ptr2->offset) - return true; - - return false; -} - -/* - * We're doing the index update that makes @b unreachable, update stuff to - * reflect that: - * - * Must be called _before_ btree_update_updated_root() or - * btree_update_updated_node: - */ -static void bch2_btree_node_free_index(struct btree_update *as, struct btree *b, - struct bkey_s_c k, - struct bch_fs_usage *stats) -{ - struct bch_fs *c = as->c; - struct pending_btree_node_free *d; - - for (d = as->pending; d < as->pending + as->nr_pending; d++) - if (!bkey_cmp(k.k->p, d->key.k.p) && - btree_key_matches(c, k, bkey_i_to_s_c(&d->key))) - goto found; - BUG(); -found: - BUG_ON(d->index_update_done); - d->index_update_done = true; - - /* - * We're dropping @k from the btree, but it's still live until the - * index update is persistent so we need to keep a reference around for - * mark and sweep to find - that's primarily what the - * btree_node_pending_free list is for. - * - * So here (when we set index_update_done = true), we're moving an - * existing reference to a different part of the larger "gc keyspace" - - * and the new position comes after the old position, since GC marks - * the pending free list after it walks the btree. - * - * If we move the reference while mark and sweep is _between_ the old - * and the new position, mark and sweep will see the reference twice - * and it'll get double accounted - so check for that here and subtract - * to cancel out one of mark and sweep's markings if necessary: - */ - - if (gc_pos_cmp(c->gc_pos, b - ? gc_pos_btree_node(b) - : gc_pos_btree_root(as->btree_id)) >= 0 && - gc_pos_cmp(c->gc_pos, gc_phase(GC_PHASE_PENDING_DELETE)) < 0) - bch2_mark_key_locked(c, bkey_i_to_s_c(&d->key), - 0, 0, NULL, 0, - BTREE_TRIGGER_OVERWRITE| - BTREE_TRIGGER_GC); -} - static void __btree_node_free(struct bch_fs *c, struct btree *b) { trace_btree_node_free(c, b); @@ -207,6 +135,8 @@ static void __btree_node_free(struct bch_fs *c, struct btree *b) bch2_btree_node_hash_remove(&c->btree_cache, b); + six_lock_wakeup_all(&b->c.lock); + mutex_lock(&c->btree_cache.lock); list_move(&b->list, &c->btree_cache.freeable); mutex_unlock(&c->btree_cache.lock); @@ -216,15 +146,13 @@ void bch2_btree_node_free_never_inserted(struct bch_fs *c, struct btree *b) { struct open_buckets ob = b->ob; - btree_update_drop_new_node(c, b); - b->ob.nr = 0; clear_btree_node_dirty(b); btree_node_lock_type(c, b, SIX_LOCK_write); __btree_node_free(c, b); - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); bch2_open_buckets_put(c, &ob); } @@ -235,39 +163,12 @@ void bch2_btree_node_free_inmem(struct bch_fs *c, struct btree *b, struct btree_iter *linked; trans_for_each_iter(iter->trans, linked) - BUG_ON(linked->l[b->level].b == b); + BUG_ON(linked->l[b->c.level].b == b); - /* - * Is this a node that isn't reachable on disk yet? - * - * Nodes that aren't reachable yet have writes blocked until they're - * reachable - now that we've cancelled any pending writes and moved - * things waiting on that write to wait on this update, we can drop this - * node from the list of nodes that the other update is making - * reachable, prior to freeing it: - */ - btree_update_drop_new_node(c, b); - - six_lock_write(&b->lock); + six_lock_write(&b->c.lock, NULL, NULL); __btree_node_free(c, b); - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); -} - -static void bch2_btree_node_free_ondisk(struct bch_fs *c, - struct pending_btree_node_free *pending, - u64 journal_seq) -{ - BUG_ON(!pending->index_update_done); - - bch2_mark_key(c, bkey_i_to_s_c(&pending->key), - 0, 0, NULL, journal_seq, BTREE_TRIGGER_OVERWRITE); - - if (gc_visited(c, gc_phase(GC_PHASE_PENDING_DELETE))) - bch2_mark_key(c, bkey_i_to_s_c(&pending->key), - 0, 0, NULL, journal_seq, - BTREE_TRIGGER_OVERWRITE| - BTREE_TRIGGER_GC); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); } static struct btree *__bch2_btree_node_alloc(struct bch_fs *c, @@ -357,17 +258,17 @@ static struct btree *bch2_btree_node_alloc(struct btree_update *as, unsigned lev int ret; BUG_ON(level >= BTREE_MAX_DEPTH); - BUG_ON(!as->reserve->nr); + BUG_ON(!as->nr_prealloc_nodes); - b = as->reserve->b[--as->reserve->nr]; + b = as->prealloc_nodes[--as->nr_prealloc_nodes]; set_btree_node_accessed(b); set_btree_node_dirty(b); set_btree_node_need_write(b); bch2_bset_init_first(b, &b->data->keys); - b->level = level; - b->btree_id = as->btree_id; + b->c.level = level; + b->c.btree_id = as->btree_id; memset(&b->nr, 0, sizeof(b->nr)); b->data->magic = cpu_to_le64(bset_magic(c)); @@ -394,8 +295,6 @@ static struct btree *bch2_btree_node_alloc(struct btree_update *as, unsigned lev bch2_btree_build_aux_trees(b); - btree_node_will_make_reachable(as, b); - ret = bch2_btree_node_hash_insert(&c->btree_cache, b, level, as->btree_id); BUG_ON(ret); @@ -422,7 +321,7 @@ struct btree *__bch2_btree_node_alloc_replacement(struct btree_update *as, { struct btree *n; - n = bch2_btree_node_alloc(as, b->level); + n = bch2_btree_node_alloc(as, b->c.level); SET_BTREE_NODE_SEQ(n->data, BTREE_NODE_SEQ(b->data) + 1); @@ -466,21 +365,22 @@ static struct btree *__btree_root_alloc(struct btree_update *as, unsigned level) btree_node_set_format(b, b->data->format); bch2_btree_build_aux_trees(b); - six_unlock_write(&b->lock); + bch2_btree_update_add_new_node(as, b); + six_unlock_write(&b->c.lock); return b; } -static void bch2_btree_reserve_put(struct bch_fs *c, struct btree_reserve *reserve) +static void bch2_btree_reserve_put(struct btree_update *as) { - bch2_disk_reservation_put(c, &reserve->disk_res); + struct bch_fs *c = as->c; mutex_lock(&c->btree_reserve_cache_lock); - while (reserve->nr) { - struct btree *b = reserve->b[--reserve->nr]; + while (as->nr_prealloc_nodes) { + struct btree *b = as->prealloc_nodes[--as->nr_prealloc_nodes]; - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); if (c->btree_reserve_cache_nr < ARRAY_SIZE(c->btree_reserve_cache)) { @@ -496,42 +396,20 @@ static void bch2_btree_reserve_put(struct bch_fs *c, struct btree_reserve *reser btree_node_lock_type(c, b, SIX_LOCK_write); __btree_node_free(c, b); - six_unlock_write(&b->lock); + six_unlock_write(&b->c.lock); - six_unlock_intent(&b->lock); + six_unlock_intent(&b->c.lock); } mutex_unlock(&c->btree_reserve_cache_lock); - - mempool_free(reserve, &c->btree_reserve_pool); } -static struct btree_reserve *bch2_btree_reserve_get(struct bch_fs *c, - unsigned nr_nodes, - unsigned flags, - struct closure *cl) +static int bch2_btree_reserve_get(struct btree_update *as, unsigned nr_nodes, + unsigned flags, struct closure *cl) { - struct btree_reserve *reserve; + struct bch_fs *c = as->c; struct btree *b; - struct disk_reservation disk_res = { 0, 0 }; - unsigned sectors = nr_nodes * c->opts.btree_node_size; - int ret, disk_res_flags = 0; - - if (flags & BTREE_INSERT_NOFAIL) - disk_res_flags |= BCH_DISK_RESERVATION_NOFAIL; - - /* - * This check isn't necessary for correctness - it's just to potentially - * prevent us from doing a lot of work that'll end up being wasted: - */ - ret = bch2_journal_error(&c->journal); - if (ret) - return ERR_PTR(ret); - - if (bch2_disk_reservation_get(c, &disk_res, sectors, - c->opts.metadata_replicas, - disk_res_flags)) - return ERR_PTR(-ENOSPC); + int ret; BUG_ON(nr_nodes > BTREE_RESERVE_MAX); @@ -540,18 +418,11 @@ static struct btree_reserve *bch2_btree_reserve_get(struct bch_fs *c, * open bucket reserve: */ ret = bch2_btree_cache_cannibalize_lock(c, cl); - if (ret) { - bch2_disk_reservation_put(c, &disk_res); - return ERR_PTR(ret); - } - - reserve = mempool_alloc(&c->btree_reserve_pool, GFP_NOIO); - - reserve->disk_res = disk_res; - reserve->nr = 0; + if (ret) + return ret; - while (reserve->nr < nr_nodes) { - b = __bch2_btree_node_alloc(c, &disk_res, + while (as->nr_prealloc_nodes < nr_nodes) { + b = __bch2_btree_node_alloc(c, &as->disk_res, flags & BTREE_INSERT_NOWAIT ? NULL : cl, flags); if (IS_ERR(b)) { @@ -563,21 +434,20 @@ static struct btree_reserve *bch2_btree_reserve_get(struct bch_fs *c, if (ret) goto err_free; - reserve->b[reserve->nr++] = b; + as->prealloc_nodes[as->nr_prealloc_nodes++] = b; } bch2_btree_cache_cannibalize_unlock(c); - return reserve; + return 0; err_free: - bch2_btree_reserve_put(c, reserve); bch2_btree_cache_cannibalize_unlock(c); trace_btree_reserve_get_fail(c, nr_nodes, cl); - return ERR_PTR(ret); + return ret; } /* Asynchronous interior node update machinery */ -static void __bch2_btree_update_free(struct btree_update *as) +static void bch2_btree_update_free(struct btree_update *as) { struct bch_fs *c = as->c; @@ -585,14 +455,13 @@ static void __bch2_btree_update_free(struct btree_update *as) bch2_journal_pin_drop(&c->journal, &as->journal); bch2_journal_pin_flush(&c->journal, &as->journal); + bch2_disk_reservation_put(c, &as->disk_res); + bch2_btree_reserve_put(as); - BUG_ON(as->nr_new_nodes || as->nr_pending); - - if (as->reserve) - bch2_btree_reserve_put(c, as->reserve); - + mutex_lock(&c->btree_interior_update_lock); list_del(&as->unwritten_list); list_del(&as->list); + mutex_unlock(&c->btree_interior_update_lock); closure_debug_destroy(&as->cl); mempool_free(as, &c->btree_interior_update_pool); @@ -600,37 +469,59 @@ static void __bch2_btree_update_free(struct btree_update *as) closure_wake_up(&c->btree_interior_update_wait); } -static void bch2_btree_update_free(struct btree_update *as) +static void btree_update_will_delete_key(struct btree_update *as, + struct bkey_i *k) { - struct bch_fs *c = as->c; + BUG_ON(bch2_keylist_u64s(&as->old_keys) + k->k.u64s > + ARRAY_SIZE(as->_old_keys)); + bch2_keylist_add(&as->old_keys, k); +} - mutex_lock(&c->btree_interior_update_lock); - __bch2_btree_update_free(as); - mutex_unlock(&c->btree_interior_update_lock); +static void btree_update_will_add_key(struct btree_update *as, + struct bkey_i *k) +{ + BUG_ON(bch2_keylist_u64s(&as->new_keys) + k->k.u64s > + ARRAY_SIZE(as->_new_keys)); + bch2_keylist_add(&as->new_keys, k); } -static inline bool six_trylock_intentwrite(struct six_lock *lock) +/* + * The transactional part of an interior btree node update, where we journal the + * update we did to the interior node and update alloc info: + */ +static int btree_update_nodes_written_trans(struct btree_trans *trans, + struct btree_update *as) { - if (!six_trylock_intent(lock)) - return false; + struct bkey_i *k; + int ret; + + trans->extra_journal_entries = (void *) &as->journal_entries[0]; + trans->extra_journal_entry_u64s = as->journal_u64s; + trans->journal_pin = &as->journal; - if (!six_trylock_write(lock)) { - six_unlock_intent(lock); - return false; + for_each_keylist_key(&as->new_keys, k) { + ret = bch2_trans_mark_key(trans, bkey_i_to_s_c(k), + 0, 0, BTREE_TRIGGER_INSERT); + if (ret) + return ret; } - return true; + for_each_keylist_key(&as->old_keys, k) { + ret = bch2_trans_mark_key(trans, bkey_i_to_s_c(k), + 0, 0, BTREE_TRIGGER_OVERWRITE); + if (ret) + return ret; + } + + return 0; } -static void btree_update_nodes_written(struct closure *cl) +static void btree_update_nodes_written(struct btree_update *as) { - struct btree_update *as = container_of(cl, struct btree_update, cl); - struct btree *nodes_need_write[BTREE_MAX_DEPTH * 2 + GC_MERGE_NODES + 1]; - unsigned nr_nodes_need_write; - struct journal_res res = { 0 }; struct bch_fs *c = as->c; - struct btree_root *r; - struct btree *b; + struct btree *b = as->b; + u64 journal_seq = 0; + unsigned i; int ret; /* @@ -638,78 +529,28 @@ static void btree_update_nodes_written(struct closure *cl) * to child nodes that weren't written yet: now, the child nodes have * been written so we can write out the update to the interior node. */ - mutex_lock(&c->btree_interior_update_lock); - as->nodes_written = true; -again: - nr_nodes_need_write = 0; - as = list_first_entry_or_null(&c->btree_interior_updates_unwritten, - struct btree_update, unwritten_list); - if (!as || !as->nodes_written) { - mutex_unlock(&c->btree_interior_update_lock); - return; - } - - b = as->b; - if (b && !six_trylock_intentwrite(&b->lock)) { - mutex_unlock(&c->btree_interior_update_lock); - - btree_node_lock_type(c, b, SIX_LOCK_intent); - six_lock_write(&b->lock); - - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); - - mutex_lock(&c->btree_interior_update_lock); - goto again; - } - - ret = bch2_journal_res_get(&c->journal, &res, as->journal_u64s, - JOURNAL_RES_GET_NONBLOCK| - JOURNAL_RES_GET_RESERVED); - if (ret == -EAGAIN) { - unsigned u64s = as->journal_u64s; - - if (b) { - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); - } - - mutex_unlock(&c->btree_interior_update_lock); - - ret = bch2_journal_res_get(&c->journal, &res, u64s, - JOURNAL_RES_GET_CHECK| - JOURNAL_RES_GET_RESERVED); - if (!ret) { - mutex_lock(&c->btree_interior_update_lock); - goto again; - } - } - - if (!ret) { - struct journal_buf *buf = &c->journal.buf[res.idx]; - struct jset_entry *entry = vstruct_idx(buf->data, res.offset); - - res.offset += as->journal_u64s; - res.u64s -= as->journal_u64s; - memcpy_u64s(entry, as->journal_entries, as->journal_u64s); - } else { - /* - * On journal error we have to run most of the normal path so - * that shutdown works - unblocking btree node writes in - * particular and writing them if needed - except for - * journalling the update: - */ - - BUG_ON(!bch2_journal_error(&c->journal)); - } - - switch (as->mode) { - case BTREE_INTERIOR_NO_UPDATE: - BUG(); - case BTREE_INTERIOR_UPDATING_NODE: - /* @b is the node we did the final insert into: */ + /* + * We can't call into journal reclaim here: we'd block on the journal + * reclaim lock, but we may need to release the open buckets we have + * pinned in order for other btree updates to make forward progress, and + * journal reclaim does btree updates when flushing bkey_cached entries, + * which may require allocations as well. + */ + ret = bch2_trans_do(c, &as->disk_res, &journal_seq, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_USE_RESERVE| + BTREE_INSERT_USE_ALLOC_RESERVE| + BTREE_INSERT_NOCHECK_RW| + BTREE_INSERT_JOURNAL_RECLAIM| + BTREE_INSERT_JOURNAL_RESERVED, + btree_update_nodes_written_trans(&trans, as)); + BUG_ON(ret && !bch2_journal_error(&c->journal)); + + if (b) { /* + * @b is the node we did the final insert into: + * * On failure to get a journal reservation, we still have to * unblock the write and allow most of the write path to happen * so that shutdown works, but the i->journal_seq mechanism @@ -719,83 +560,90 @@ again: * we're in journal error state: */ + btree_node_lock_type(c, b, SIX_LOCK_intent); + btree_node_lock_type(c, b, SIX_LOCK_write); + mutex_lock(&c->btree_interior_update_lock); + list_del(&as->write_blocked_list); - if (!ret) { + if (!ret && as->b == b) { struct bset *i = btree_bset_last(b); + BUG_ON(!b->c.level); + BUG_ON(!btree_node_dirty(b)); + i->journal_seq = cpu_to_le64( - max(res.seq, + max(journal_seq, le64_to_cpu(i->journal_seq))); - bch2_btree_add_journal_pin(c, b, res.seq); + bch2_btree_add_journal_pin(c, b, journal_seq); } - nodes_need_write[nr_nodes_need_write++] = b; - - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); - break; - - case BTREE_INTERIOR_UPDATING_AS: - BUG_ON(b); - break; - - case BTREE_INTERIOR_UPDATING_ROOT: - r = &c->btree_roots[as->btree_id]; - - BUG_ON(b); + mutex_unlock(&c->btree_interior_update_lock); + six_unlock_write(&b->c.lock); - mutex_lock(&c->btree_root_lock); - bkey_copy(&r->key, as->parent_keys.keys); - r->level = as->level; - r->alive = true; - c->btree_roots_dirty = true; - mutex_unlock(&c->btree_root_lock); - break; + btree_node_write_if_need(c, b, SIX_LOCK_intent); + six_unlock_intent(&b->c.lock); } bch2_journal_pin_drop(&c->journal, &as->journal); - bch2_journal_res_put(&c->journal, &res); bch2_journal_preres_put(&c->journal, &as->journal_preres); - while (as->nr_new_nodes) { - b = as->new_nodes[--as->nr_new_nodes]; + mutex_lock(&c->btree_interior_update_lock); + for (i = 0; i < as->nr_new_nodes; i++) { + b = as->new_nodes[i]; BUG_ON(b->will_make_reachable != (unsigned long) as); b->will_make_reachable = 0; + } + mutex_unlock(&c->btree_interior_update_lock); - nodes_need_write[nr_nodes_need_write++] = b; + for (i = 0; i < as->nr_new_nodes; i++) { + b = as->new_nodes[i]; + + btree_node_lock_type(c, b, SIX_LOCK_read); + btree_node_write_if_need(c, b, SIX_LOCK_read); + six_unlock_read(&b->c.lock); } - while (as->nr_pending) - bch2_btree_node_free_ondisk(c, - &as->pending[--as->nr_pending], res.seq); + for (i = 0; i < as->nr_open_buckets; i++) + bch2_open_bucket_put(c, c->open_buckets + as->open_buckets[i]); - __bch2_btree_update_free(as); - /* - * for flush_held_btree_writes() waiting on updates to flush or - * nodes to be writeable: - */ - closure_wake_up(&c->btree_interior_update_wait); + bch2_btree_update_free(as); +} - /* - * Can't take btree node locks while holding btree_interior_update_lock: - * */ - mutex_unlock(&c->btree_interior_update_lock); +static void btree_interior_update_work(struct work_struct *work) +{ + struct bch_fs *c = + container_of(work, struct bch_fs, btree_interior_update_work); + struct btree_update *as; - /* Do btree writes after dropping journal res/locks: */ - while (nr_nodes_need_write) { - b = nodes_need_write[--nr_nodes_need_write]; + while (1) { + mutex_lock(&c->btree_interior_update_lock); + as = list_first_entry_or_null(&c->btree_interior_updates_unwritten, + struct btree_update, unwritten_list); + if (as && !as->nodes_written) + as = NULL; + mutex_unlock(&c->btree_interior_update_lock); - btree_node_lock_type(c, b, SIX_LOCK_read); - bch2_btree_node_write_cond(c, b, btree_node_need_write(b)); - six_unlock_read(&b->lock); + if (!as) + break; + + btree_update_nodes_written(as); } +} + +static void btree_update_set_nodes_written(struct closure *cl) +{ + struct btree_update *as = container_of(cl, struct btree_update, cl); + struct bch_fs *c = as->c; mutex_lock(&c->btree_interior_update_lock); - goto again; + as->nodes_written = true; + mutex_unlock(&c->btree_interior_update_lock); + + queue_work(c->btree_interior_update_worker, &c->btree_interior_update_work); } /* @@ -814,7 +662,6 @@ static void btree_update_updated_node(struct btree_update *as, struct btree *b) as->mode = BTREE_INTERIOR_UPDATING_NODE; as->b = b; - as->level = b->level; list_add(&as->write_blocked_list, &b->write_blocked); mutex_unlock(&c->btree_interior_update_lock); @@ -845,25 +692,45 @@ static void btree_update_reparent(struct btree_update *as, static void btree_update_updated_root(struct btree_update *as, struct btree *b) { + struct bkey_i *insert = &b->key; struct bch_fs *c = as->c; BUG_ON(as->mode != BTREE_INTERIOR_NO_UPDATE); - BUG_ON(!bch2_keylist_empty(&as->parent_keys)); + + BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) > + ARRAY_SIZE(as->journal_entries)); + + as->journal_u64s += + journal_entry_set((void *) &as->journal_entries[as->journal_u64s], + BCH_JSET_ENTRY_btree_root, + b->c.btree_id, b->c.level, + insert, insert->k.u64s); mutex_lock(&c->btree_interior_update_lock); list_add_tail(&as->unwritten_list, &c->btree_interior_updates_unwritten); as->mode = BTREE_INTERIOR_UPDATING_ROOT; - as->level = b->level; - bch2_keylist_add(&as->parent_keys, &b->key); mutex_unlock(&c->btree_interior_update_lock); } -static void btree_node_will_make_reachable(struct btree_update *as, - struct btree *b) +/* + * bch2_btree_update_add_new_node: + * + * This causes @as to wait on @b to be written, before it gets to + * bch2_btree_update_nodes_written + * + * Additionally, it sets b->will_make_reachable to prevent any additional writes + * to @b from happening besides the first until @b is reachable on disk + * + * And it adds @b to the list of @as's new nodes, so that we can update sector + * counts in bch2_btree_update_nodes_written: + */ +void bch2_btree_update_add_new_node(struct btree_update *as, struct btree *b) { struct bch_fs *c = as->c; + closure_get(&as->cl); + mutex_lock(&c->btree_interior_update_lock); BUG_ON(as->nr_new_nodes >= ARRAY_SIZE(as->new_nodes)); BUG_ON(b->will_make_reachable); @@ -871,10 +738,14 @@ static void btree_node_will_make_reachable(struct btree_update *as, as->new_nodes[as->nr_new_nodes++] = b; b->will_make_reachable = 1UL|(unsigned long) as; - closure_get(&as->cl); mutex_unlock(&c->btree_interior_update_lock); + + btree_update_will_add_key(as, &b->key); } +/* + * returns true if @b was a new node + */ static void btree_update_drop_new_node(struct bch_fs *c, struct btree *b) { struct btree_update *as; @@ -882,6 +753,11 @@ static void btree_update_drop_new_node(struct bch_fs *c, struct btree *b) unsigned i; mutex_lock(&c->btree_interior_update_lock); + /* + * When b->will_make_reachable != 0, it owns a ref on as->cl that's + * dropped when it gets written by bch2_btree_complete_write - the + * xchg() is for synchronization with bch2_btree_complete_write: + */ v = xchg(&b->will_make_reachable, 0); as = (struct btree_update *) (v & ~1UL); @@ -903,25 +779,11 @@ found: closure_put(&as->cl); } -static void btree_interior_update_add_node_reference(struct btree_update *as, - struct btree *b) +void bch2_btree_update_get_open_buckets(struct btree_update *as, struct btree *b) { - struct bch_fs *c = as->c; - struct pending_btree_node_free *d; - - mutex_lock(&c->btree_interior_update_lock); - - /* Add this node to the list of nodes being freed: */ - BUG_ON(as->nr_pending >= ARRAY_SIZE(as->pending)); - - d = &as->pending[as->nr_pending++]; - d->index_update_done = false; - d->seq = b->data->keys.seq; - d->btree_id = b->btree_id; - d->level = b->level; - bkey_copy(&d->key, &b->key); - - mutex_unlock(&c->btree_interior_update_lock); + while (b->ob.nr) + as->open_buckets[as->nr_open_buckets++] = + b->ob.v[--b->ob.nr]; } /* @@ -941,8 +803,6 @@ void bch2_btree_interior_update_will_free_node(struct btree_update *as, if (btree_node_fake(b)) return; - btree_interior_update_add_node_reference(as, b); - mutex_lock(&c->btree_interior_update_lock); /* @@ -954,7 +814,7 @@ void bch2_btree_interior_update_will_free_node(struct btree_update *as, * operations complete */ list_for_each_entry_safe(p, n, &b->write_blocked, write_blocked_list) { - list_del(&p->write_blocked_list); + list_del_init(&p->write_blocked_list); btree_update_reparent(as, p); /* @@ -984,16 +844,28 @@ void bch2_btree_interior_update_will_free_node(struct btree_update *as, bch2_journal_pin_drop(&c->journal, &w->journal); mutex_unlock(&c->btree_interior_update_lock); + + /* + * Is this a node that isn't reachable on disk yet? + * + * Nodes that aren't reachable yet have writes blocked until they're + * reachable - now that we've cancelled any pending writes and moved + * things waiting on that write to wait on this update, we can drop this + * node from the list of nodes that the other update is making + * reachable, prior to freeing it: + */ + btree_update_drop_new_node(c, b); + + btree_update_will_delete_key(as, &b->key); } void bch2_btree_update_done(struct btree_update *as) { BUG_ON(as->mode == BTREE_INTERIOR_NO_UPDATE); - bch2_btree_reserve_put(as->c, as->reserve); - as->reserve = NULL; + bch2_btree_reserve_put(as); - continue_at(&as->cl, btree_update_nodes_written, system_freezable_wq); + continue_at(&as->cl, btree_update_set_nodes_written, system_freezable_wq); } struct btree_update * @@ -1002,60 +874,79 @@ bch2_btree_update_start(struct btree_trans *trans, enum btree_id id, struct closure *cl) { struct bch_fs *c = trans->c; - struct journal_preres journal_preres = { 0 }; - struct btree_reserve *reserve; struct btree_update *as; - int ret; + int disk_res_flags = (flags & BTREE_INSERT_NOFAIL) + ? BCH_DISK_RESERVATION_NOFAIL : 0; + int journal_flags = (flags & BTREE_INSERT_JOURNAL_RESERVED) + ? JOURNAL_RES_GET_RECLAIM : 0; + int ret = 0; + + /* + * This check isn't necessary for correctness - it's just to potentially + * prevent us from doing a lot of work that'll end up being wasted: + */ + ret = bch2_journal_error(&c->journal); + if (ret) + return ERR_PTR(ret); + + as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOIO); + memset(as, 0, sizeof(*as)); + closure_init(&as->cl, NULL); + as->c = c; + as->mode = BTREE_INTERIOR_NO_UPDATE; + as->btree_id = id; + INIT_LIST_HEAD(&as->list); + INIT_LIST_HEAD(&as->unwritten_list); + INIT_LIST_HEAD(&as->write_blocked_list); + bch2_keylist_init(&as->old_keys, as->_old_keys); + bch2_keylist_init(&as->new_keys, as->_new_keys); + bch2_keylist_init(&as->parent_keys, as->inline_keys); - ret = bch2_journal_preres_get(&c->journal, &journal_preres, + ret = bch2_journal_preres_get(&c->journal, &as->journal_preres, BTREE_UPDATE_JOURNAL_RES, - JOURNAL_RES_GET_NONBLOCK); + journal_flags|JOURNAL_RES_GET_NONBLOCK); if (ret == -EAGAIN) { if (flags & BTREE_INSERT_NOUNLOCK) return ERR_PTR(-EINTR); bch2_trans_unlock(trans); - ret = bch2_journal_preres_get(&c->journal, &journal_preres, - BTREE_UPDATE_JOURNAL_RES, 0); + ret = bch2_journal_preres_get(&c->journal, &as->journal_preres, + BTREE_UPDATE_JOURNAL_RES, + journal_flags); if (ret) return ERR_PTR(ret); if (!bch2_trans_relock(trans)) { - bch2_journal_preres_put(&c->journal, &journal_preres); - return ERR_PTR(-EINTR); + ret = -EINTR; + goto err; } } - reserve = bch2_btree_reserve_get(c, nr_nodes, flags, cl); - if (IS_ERR(reserve)) { - bch2_journal_preres_put(&c->journal, &journal_preres); - return ERR_CAST(reserve); - } - - as = mempool_alloc(&c->btree_interior_update_pool, GFP_NOIO); - memset(as, 0, sizeof(*as)); - closure_init(&as->cl, NULL); - as->c = c; - as->mode = BTREE_INTERIOR_NO_UPDATE; - as->btree_id = id; - as->reserve = reserve; - INIT_LIST_HEAD(&as->write_blocked_list); - INIT_LIST_HEAD(&as->unwritten_list); - as->journal_preres = journal_preres; + ret = bch2_disk_reservation_get(c, &as->disk_res, + nr_nodes * c->opts.btree_node_size, + c->opts.metadata_replicas, + disk_res_flags); + if (ret) + goto err; - bch2_keylist_init(&as->parent_keys, as->inline_keys); + ret = bch2_btree_reserve_get(as, nr_nodes, flags, cl); + if (ret) + goto err; mutex_lock(&c->btree_interior_update_lock); list_add_tail(&as->list, &c->btree_interior_update_list); mutex_unlock(&c->btree_interior_update_lock); return as; +err: + bch2_btree_update_free(as); + return ERR_PTR(ret); } /* Btree root updates: */ -static void __bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b) +static void bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b) { /* Root nodes cannot be reaped */ mutex_lock(&c->btree_cache.lock); @@ -1064,7 +955,7 @@ static void __bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b) mutex_lock(&c->btree_root_lock); BUG_ON(btree_node_root(c, b) && - (b->level < btree_node_root(c, b)->level || + (b->c.level < btree_node_root(c, b)->c.level || !btree_node_dying(btree_node_root(c, b)))); btree_node_root(c, b) = b; @@ -1073,38 +964,6 @@ static void __bch2_btree_set_root_inmem(struct bch_fs *c, struct btree *b) bch2_recalc_btree_reserve(c); } -static void bch2_btree_set_root_inmem(struct btree_update *as, struct btree *b) -{ - struct bch_fs *c = as->c; - struct btree *old = btree_node_root(c, b); - struct bch_fs_usage *fs_usage; - - __bch2_btree_set_root_inmem(c, b); - - mutex_lock(&c->btree_interior_update_lock); - percpu_down_read(&c->mark_lock); - fs_usage = bch2_fs_usage_scratch_get(c); - - bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key), - 0, 0, fs_usage, 0, - BTREE_TRIGGER_INSERT); - if (gc_visited(c, gc_pos_btree_root(b->btree_id))) - bch2_mark_key_locked(c, bkey_i_to_s_c(&b->key), - 0, 0, NULL, 0, - BTREE_TRIGGER_INSERT| - BTREE_TRIGGER_GC); - - if (old && !btree_node_fake(old)) - bch2_btree_node_free_index(as, NULL, - bkey_i_to_s_c(&old->key), - fs_usage); - bch2_fs_usage_apply(c, fs_usage, &as->reserve->disk_res, 0); - - bch2_fs_usage_scratch_put(c, fs_usage); - percpu_up_read(&c->mark_lock); - mutex_unlock(&c->btree_interior_update_lock); -} - /** * bch_btree_set_root - update the root in memory and on disk * @@ -1135,7 +994,7 @@ static void bch2_btree_set_root(struct btree_update *as, struct btree *b, */ bch2_btree_node_lock_write(old, iter); - bch2_btree_set_root_inmem(as, b); + bch2_btree_set_root_inmem(c, b); btree_update_updated_root(as, b); @@ -1156,57 +1015,21 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as, struct btree *b struct bkey_i *insert, struct btree_node_iter *node_iter) { - struct bch_fs *c = as->c; - struct bch_fs_usage *fs_usage; - struct jset_entry *entry; struct bkey_packed *k; - struct bkey tmp; BUG_ON(as->journal_u64s + jset_u64s(insert->k.u64s) > ARRAY_SIZE(as->journal_entries)); - entry = (void *) &as->journal_entries[as->journal_u64s]; - memset(entry, 0, sizeof(*entry)); - entry->u64s = cpu_to_le16(insert->k.u64s); - entry->type = BCH_JSET_ENTRY_btree_keys; - entry->btree_id = b->btree_id; - entry->level = b->level; - memcpy_u64s_small(entry->_data, insert, insert->k.u64s); - as->journal_u64s += jset_u64s(insert->k.u64s); - - mutex_lock(&c->btree_interior_update_lock); - percpu_down_read(&c->mark_lock); - fs_usage = bch2_fs_usage_scratch_get(c); - - bch2_mark_key_locked(c, bkey_i_to_s_c(insert), - 0, 0, fs_usage, 0, - BTREE_TRIGGER_INSERT); - - if (gc_visited(c, gc_pos_btree_node(b))) - bch2_mark_key_locked(c, bkey_i_to_s_c(insert), - 0, 0, NULL, 0, - BTREE_TRIGGER_INSERT| - BTREE_TRIGGER_GC); + as->journal_u64s += + journal_entry_set((void *) &as->journal_entries[as->journal_u64s], + BCH_JSET_ENTRY_btree_keys, + b->c.btree_id, b->c.level, + insert, insert->k.u64s); while ((k = bch2_btree_node_iter_peek_all(node_iter, b)) && bkey_iter_pos_cmp(b, k, &insert->k.p) < 0) bch2_btree_node_iter_advance(node_iter, b); - /* - * If we're overwriting, look up pending delete and mark so that gc - * marks it on the pending delete list: - */ - if (k && !bkey_cmp_packed(b, k, &insert->k)) - bch2_btree_node_free_index(as, b, - bkey_disassemble(b, k, &tmp), - fs_usage); - - bch2_fs_usage_apply(c, fs_usage, &as->reserve->disk_res, 0); - - bch2_fs_usage_scratch_put(c, fs_usage); - percpu_up_read(&c->mark_lock); - mutex_unlock(&c->btree_interior_update_lock); - bch2_btree_bset_insert_key(iter, b, node_iter, insert); set_btree_node_dirty(b); set_btree_node_need_write(b); @@ -1225,7 +1048,8 @@ static struct btree *__btree_split_node(struct btree_update *as, struct bset *set1, *set2; struct bkey_packed *k, *prev = NULL; - n2 = bch2_btree_node_alloc(as, n1->level); + n2 = bch2_btree_node_alloc(as, n1->c.level); + bch2_btree_update_add_new_node(as, n2); n2->data->max_key = n1->data->max_key; n2->data->format = n1->format; @@ -1293,7 +1117,7 @@ static struct btree *__btree_split_node(struct btree_update *as, bch2_verify_btree_nr_keys(n1); bch2_verify_btree_nr_keys(n2); - if (n1->level) { + if (n1->c.level) { btree_node_interior_verify(n1); btree_node_interior_verify(n2); } @@ -1321,14 +1145,6 @@ static void btree_split_insert_keys(struct btree_update *as, struct btree *b, struct bkey_packed *src, *dst, *n; struct bset *i; - /* - * XXX - * - * these updates must be journalled - * - * oops - */ - BUG_ON(btree_node_type(b) != BKEY_TYPE_BTREE); bch2_btree_node_iter_init(&node_iter, b, &k->k.p); @@ -1375,11 +1191,12 @@ static void btree_split(struct btree_update *as, struct btree *b, u64 start_time = local_clock(); BUG_ON(!parent && (b != btree_node_root(c, b))); - BUG_ON(!btree_node_intent_locked(iter, btree_node_root(c, b)->level)); + BUG_ON(!btree_node_intent_locked(iter, btree_node_root(c, b)->c.level)); bch2_btree_interior_update_will_free_node(as, b); n1 = bch2_btree_node_alloc_replacement(as, b); + bch2_btree_update_add_new_node(as, n1); if (keys) btree_split_insert_keys(as, n1, iter, keys); @@ -1391,8 +1208,8 @@ static void btree_split(struct btree_update *as, struct btree *b, bch2_btree_build_aux_trees(n2); bch2_btree_build_aux_trees(n1); - six_unlock_write(&n2->lock); - six_unlock_write(&n1->lock); + six_unlock_write(&n2->c.lock); + six_unlock_write(&n1->c.lock); bch2_btree_node_write(c, n2, SIX_LOCK_intent); @@ -1406,7 +1223,7 @@ static void btree_split(struct btree_update *as, struct btree *b, if (!parent) { /* Depth increases, make a new root */ - n3 = __btree_root_alloc(as, b->level + 1); + n3 = __btree_root_alloc(as, b->c.level + 1); n3->sib_u64s[0] = U16_MAX; n3->sib_u64s[1] = U16_MAX; @@ -1419,7 +1236,7 @@ static void btree_split(struct btree_update *as, struct btree *b, trace_btree_compact(c, b); bch2_btree_build_aux_trees(n1); - six_unlock_write(&n1->lock); + six_unlock_write(&n1->c.lock); if (parent) bch2_keylist_add(&as->parent_keys, &n1->key); @@ -1439,15 +1256,15 @@ static void btree_split(struct btree_update *as, struct btree *b, bch2_btree_set_root(as, n1, iter); } - bch2_open_buckets_put(c, &n1->ob); + bch2_btree_update_get_open_buckets(as, n1); if (n2) - bch2_open_buckets_put(c, &n2->ob); + bch2_btree_update_get_open_buckets(as, n2); if (n3) - bch2_open_buckets_put(c, &n3->ob); + bch2_btree_update_get_open_buckets(as, n3); /* Successful split, update the iterator to point to the new nodes: */ - six_lock_increment(&b->lock, SIX_LOCK_intent); + six_lock_increment(&b->c.lock, SIX_LOCK_intent); bch2_btree_iter_node_drop(iter, b); if (n3) bch2_btree_iter_node_replace(iter, n3); @@ -1464,10 +1281,10 @@ static void btree_split(struct btree_update *as, struct btree *b, bch2_btree_node_free_inmem(c, b, iter); if (n3) - six_unlock_intent(&n3->lock); + six_unlock_intent(&n3->c.lock); if (n2) - six_unlock_intent(&n2->lock); - six_unlock_intent(&n1->lock); + six_unlock_intent(&n2->c.lock); + six_unlock_intent(&n1->c.lock); bch2_btree_trans_verify_locks(iter->trans); @@ -1485,7 +1302,7 @@ bch2_btree_insert_keys_interior(struct btree_update *as, struct btree *b, struct bkey_packed *k; /* Don't screw up @iter's position: */ - node_iter = iter->l[b->level].iter; + node_iter = iter->l[b->c.level].iter; /* * btree_split(), btree_gc_coalesce() will insert keys before @@ -1502,7 +1319,7 @@ bch2_btree_insert_keys_interior(struct btree_update *as, struct btree *b, btree_update_updated_node(as, b); trans_for_each_iter_with_node(iter->trans, b, linked) - bch2_btree_node_iter_peek(&linked->l[b->level].iter, b); + bch2_btree_node_iter_peek(&linked->l[b->c.level].iter, b); bch2_btree_trans_verify_iters(iter->trans, b); } @@ -1528,8 +1345,8 @@ void bch2_btree_insert_node(struct btree_update *as, struct btree *b, int old_live_u64s = b->nr.live_u64s; int live_u64s_added, u64s_added; - BUG_ON(!btree_node_intent_locked(iter, btree_node_root(c, b)->level)); - BUG_ON(!b->level); + BUG_ON(!btree_node_intent_locked(iter, btree_node_root(c, b)->c.level)); + BUG_ON(!b->c.level); BUG_ON(!as || as->b); bch2_verify_keylist_sorted(keys); @@ -1538,7 +1355,7 @@ void bch2_btree_insert_node(struct btree_update *as, struct btree *b, bch2_btree_node_lock_for_insert(c, b, iter); - if (!bch2_btree_node_insert_fits(c, b, bch_keylist_u64s(keys))) { + if (!bch2_btree_node_insert_fits(c, b, bch2_keylist_u64s(keys))) { bch2_btree_node_unlock_write(b, iter); goto split; } @@ -1566,7 +1383,7 @@ void bch2_btree_insert_node(struct btree_update *as, struct btree *b, * the btree iterator yet, so the merge path's unlock/wait/relock dance * won't work: */ - bch2_foreground_maybe_merge(c, iter, b->level, + bch2_foreground_maybe_merge(c, iter, b->c.level, flags|BTREE_INSERT_NOUNLOCK); return; split: @@ -1581,14 +1398,14 @@ int bch2_btree_split_leaf(struct bch_fs *c, struct btree_iter *iter, struct btree_update *as; struct closure cl; int ret = 0; - struct btree_iter *linked; + struct btree_insert_entry *i; /* * We already have a disk reservation and open buckets pinned; this * allocation must not block: */ - trans_for_each_iter(trans, linked) - if (linked->btree_id == BTREE_ID_EXTENTS) + trans_for_each_update(trans, i) + if (btree_node_type_needs_gc(i->iter->btree_id)) flags |= BTREE_INSERT_USE_RESERVE; closure_init_stack(&cl); @@ -1718,7 +1535,7 @@ retry: b->sib_u64s[sib] = sib_u64s; if (b->sib_u64s[sib] > BTREE_FOREGROUND_MERGE_THRESHOLD(c)) { - six_unlock_intent(&m->lock); + six_unlock_intent(&m->c.lock); goto out; } @@ -1748,7 +1565,8 @@ retry: bch2_btree_interior_update_will_free_node(as, b); bch2_btree_interior_update_will_free_node(as, m); - n = bch2_btree_node_alloc(as, b->level); + n = bch2_btree_node_alloc(as, b->c.level); + bch2_btree_update_add_new_node(as, n); btree_set_min(n, prev->data->min_key); btree_set_max(n, next->data->max_key); @@ -1760,7 +1578,7 @@ retry: bch2_btree_sort_into(c, n, next); bch2_btree_build_aux_trees(n); - six_unlock_write(&n->lock); + six_unlock_write(&n->c.lock); bkey_init(&delete.k); delete.k.p = prev->key.k.p; @@ -1771,9 +1589,9 @@ retry: bch2_btree_insert_node(as, parent, iter, &as->parent_keys, flags); - bch2_open_buckets_put(c, &n->ob); + bch2_btree_update_get_open_buckets(as, n); - six_lock_increment(&b->lock, SIX_LOCK_intent); + six_lock_increment(&b->c.lock, SIX_LOCK_intent); bch2_btree_iter_node_drop(iter, b); bch2_btree_iter_node_drop(iter, m); @@ -1784,7 +1602,7 @@ retry: bch2_btree_node_free_inmem(c, b, iter); bch2_btree_node_free_inmem(c, m, iter); - six_unlock_intent(&n->lock); + six_unlock_intent(&n->c.lock); bch2_btree_update_done(as); @@ -1806,7 +1624,7 @@ out: return; err_cycle_gc_lock: - six_unlock_intent(&m->lock); + six_unlock_intent(&m->c.lock); if (flags & BTREE_INSERT_NOUNLOCK) goto out; @@ -1819,7 +1637,7 @@ err_cycle_gc_lock: goto err; err_unlock: - six_unlock_intent(&m->lock); + six_unlock_intent(&m->c.lock); if (!(flags & BTREE_INSERT_GC_LOCK_HELD)) up_read(&c->gc_lock); err: @@ -1859,9 +1677,10 @@ static int __btree_node_rewrite(struct bch_fs *c, struct btree_iter *iter, bch2_btree_interior_update_will_free_node(as, b); n = bch2_btree_node_alloc_replacement(as, b); + bch2_btree_update_add_new_node(as, n); bch2_btree_build_aux_trees(n); - six_unlock_write(&n->lock); + six_unlock_write(&n->c.lock); trace_btree_gc_rewrite_node(c, b); @@ -1874,13 +1693,13 @@ static int __btree_node_rewrite(struct bch_fs *c, struct btree_iter *iter, bch2_btree_set_root(as, n, iter); } - bch2_open_buckets_put(c, &n->ob); + bch2_btree_update_get_open_buckets(as, n); - six_lock_increment(&b->lock, SIX_LOCK_intent); + six_lock_increment(&b->c.lock, SIX_LOCK_intent); bch2_btree_iter_node_drop(iter, b); bch2_btree_iter_node_replace(iter, n); bch2_btree_node_free_inmem(c, b, iter); - six_unlock_intent(&n->lock); + six_unlock_intent(&n->c.lock); bch2_btree_update_done(as); return 0; @@ -1949,56 +1768,15 @@ static void __bch2_btree_node_update_key(struct bch_fs *c, struct btree *parent; int ret; - /* - * Two corner cases that need to be thought about here: - * - * @b may not be reachable yet - there might be another interior update - * operation waiting on @b to be written, and we're gonna deliver the - * write completion to that interior update operation _before_ - * persisting the new_key update - * - * That ends up working without us having to do anything special here: - * the reason is, we do kick off (and do the in memory updates) for the - * update for @new_key before we return, creating a new interior_update - * operation here. - * - * The new interior update operation here will in effect override the - * previous one. The previous one was going to terminate - make @b - * reachable - in one of two ways: - * - updating the btree root pointer - * In that case, - * no, this doesn't work. argh. - */ - - if (b->will_make_reachable) - as->must_rewrite = true; - - btree_interior_update_add_node_reference(as, b); - - /* - * XXX: the rest of the update path treats this like we're actually - * inserting a new node and deleting the existing node, so the - * reservation needs to include enough space for @b - * - * that is actually sketch as fuck though and I am surprised the code - * seems to work like that, definitely need to go back and rework it - * into something saner. - * - * (I think @b is just getting double counted until the btree update - * finishes and "deletes" @b on disk) - */ - ret = bch2_disk_reservation_add(c, &as->reserve->disk_res, - c->opts.btree_node_size * - bch2_bkey_nr_ptrs(bkey_i_to_s_c(new_key)), - BCH_DISK_RESERVATION_NOFAIL); - BUG_ON(ret); + btree_update_will_delete_key(as, &b->key); + btree_update_will_add_key(as, new_key); parent = btree_node_parent(iter, b); if (parent) { if (new_hash) { bkey_copy(&new_hash->key, new_key); ret = bch2_btree_node_hash_insert(&c->btree_cache, - new_hash, b->level, b->btree_id); + new_hash, b->c.level, b->c.btree_id); BUG_ON(ret); } @@ -2019,44 +1797,18 @@ static void __bch2_btree_node_update_key(struct bch_fs *c, bkey_copy(&b->key, new_key); } } else { - struct bch_fs_usage *fs_usage; - BUG_ON(btree_node_root(c, b) != b); bch2_btree_node_lock_write(b, iter); + bkey_copy(&b->key, new_key); - mutex_lock(&c->btree_interior_update_lock); - percpu_down_read(&c->mark_lock); - fs_usage = bch2_fs_usage_scratch_get(c); - - bch2_mark_key_locked(c, bkey_i_to_s_c(new_key), - 0, 0, fs_usage, 0, - BTREE_TRIGGER_INSERT); - if (gc_visited(c, gc_pos_btree_root(b->btree_id))) - bch2_mark_key_locked(c, bkey_i_to_s_c(new_key), - 0, 0, NULL, 0, - BTREE_TRIGGER_INSERT|| - BTREE_TRIGGER_GC); - - bch2_btree_node_free_index(as, NULL, - bkey_i_to_s_c(&b->key), - fs_usage); - bch2_fs_usage_apply(c, fs_usage, &as->reserve->disk_res, 0); - - bch2_fs_usage_scratch_put(c, fs_usage); - percpu_up_read(&c->mark_lock); - mutex_unlock(&c->btree_interior_update_lock); - - if (btree_ptr_hash_val(new_key) != b->hash_val) { + if (btree_ptr_hash_val(&b->key) != b->hash_val) { mutex_lock(&c->btree_cache.lock); bch2_btree_node_hash_remove(&c->btree_cache, b); - bkey_copy(&b->key, new_key); ret = __bch2_btree_node_hash_insert(&c->btree_cache, b); BUG_ON(ret); mutex_unlock(&c->btree_cache.lock); - } else { - bkey_copy(&b->key, new_key); } btree_update_updated_root(as, b); @@ -2150,8 +1902,8 @@ err: list_move(&new_hash->list, &c->btree_cache.freeable); mutex_unlock(&c->btree_cache.lock); - six_unlock_write(&new_hash->lock); - six_unlock_intent(&new_hash->lock); + six_unlock_write(&new_hash->c.lock); + six_unlock_intent(&new_hash->c.lock); } up_read(&c->gc_lock); closure_sync(&cl); @@ -2171,7 +1923,7 @@ void bch2_btree_set_root_for_read(struct bch_fs *c, struct btree *b) { BUG_ON(btree_node_root(c, b)); - __bch2_btree_set_root_inmem(c, b); + bch2_btree_set_root_inmem(c, b); } void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id) @@ -2191,8 +1943,8 @@ void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id) bch2_btree_cache_cannibalize_unlock(c); set_btree_node_fake(b); - b->level = 0; - b->btree_id = id; + b->c.level = 0; + b->c.btree_id = id; bkey_btree_ptr_init(&b->key); b->key.k.p = POS_MAX; @@ -2207,13 +1959,14 @@ void bch2_btree_root_alloc(struct bch_fs *c, enum btree_id id) b->data->format = bch2_btree_calc_format(b); btree_node_set_format(b, b->data->format); - ret = bch2_btree_node_hash_insert(&c->btree_cache, b, b->level, b->btree_id); + ret = bch2_btree_node_hash_insert(&c->btree_cache, b, + b->c.level, b->c.btree_id); BUG_ON(ret); - __bch2_btree_set_root_inmem(c, b); + bch2_btree_set_root_inmem(c, b); - six_unlock_write(&b->lock); - six_unlock_intent(&b->lock); + six_unlock_write(&b->c.lock); + six_unlock_intent(&b->c.lock); } ssize_t bch2_btree_updates_print(struct bch_fs *c, char *buf) @@ -2246,3 +1999,75 @@ size_t bch2_btree_interior_updates_nr_pending(struct bch_fs *c) return ret; } + +void bch2_journal_entries_to_btree_roots(struct bch_fs *c, struct jset *jset) +{ + struct btree_root *r; + struct jset_entry *entry; + + mutex_lock(&c->btree_root_lock); + + vstruct_for_each(jset, entry) + if (entry->type == BCH_JSET_ENTRY_btree_root) { + r = &c->btree_roots[entry->btree_id]; + r->level = entry->level; + r->alive = true; + bkey_copy(&r->key, &entry->start[0]); + } + + mutex_unlock(&c->btree_root_lock); +} + +struct jset_entry * +bch2_btree_roots_to_journal_entries(struct bch_fs *c, + struct jset_entry *start, + struct jset_entry *end) +{ + struct jset_entry *entry; + unsigned long have = 0; + unsigned i; + + for (entry = start; entry < end; entry = vstruct_next(entry)) + if (entry->type == BCH_JSET_ENTRY_btree_root) + __set_bit(entry->btree_id, &have); + + mutex_lock(&c->btree_root_lock); + + for (i = 0; i < BTREE_ID_NR; i++) + if (c->btree_roots[i].alive && !test_bit(i, &have)) { + journal_entry_set(end, + BCH_JSET_ENTRY_btree_root, + i, c->btree_roots[i].level, + &c->btree_roots[i].key, + c->btree_roots[i].key.u64s); + end = vstruct_next(end); + } + + mutex_unlock(&c->btree_root_lock); + + return end; +} + +void bch2_fs_btree_interior_update_exit(struct bch_fs *c) +{ + if (c->btree_interior_update_worker) + destroy_workqueue(c->btree_interior_update_worker); + mempool_exit(&c->btree_interior_update_pool); +} + +int bch2_fs_btree_interior_update_init(struct bch_fs *c) +{ + mutex_init(&c->btree_reserve_cache_lock); + INIT_LIST_HEAD(&c->btree_interior_update_list); + INIT_LIST_HEAD(&c->btree_interior_updates_unwritten); + mutex_init(&c->btree_interior_update_lock); + INIT_WORK(&c->btree_interior_update_work, btree_interior_update_work); + + c->btree_interior_update_worker = + alloc_workqueue("btree_update", WQ_UNBOUND|WQ_MEM_RECLAIM, 1); + if (!c->btree_interior_update_worker) + return -ENOMEM; + + return mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1, + sizeof(struct btree_update)); +} diff --git a/fs/bcachefs/btree_update_interior.h b/fs/bcachefs/btree_update_interior.h index 2fddf5d31eb9..4a5b9dcfbdd0 100644 --- a/fs/bcachefs/btree_update_interior.h +++ b/fs/bcachefs/btree_update_interior.h @@ -6,34 +6,13 @@ #include "btree_locking.h" #include "btree_update.h" -struct btree_reserve { - struct disk_reservation disk_res; - unsigned nr; - struct btree *b[BTREE_RESERVE_MAX]; -}; - void __bch2_btree_calc_format(struct bkey_format_state *, struct btree *); bool bch2_btree_node_format_fits(struct bch_fs *c, struct btree *, struct bkey_format *); -/* Btree node freeing/allocation: */ - -/* - * Tracks a btree node that has been (or is about to be) freed in memory, but - * has _not_ yet been freed on disk (because the write that makes the new - * node(s) visible and frees the old hasn't completed yet) - */ -struct pending_btree_node_free { - bool index_update_done; - - __le64 seq; - enum btree_id btree_id; - unsigned level; - __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX); -}; +#define BTREE_UPDATE_NODES_MAX ((BTREE_MAX_DEPTH - 2) * 2 + GC_MERGE_NODES) -#define BTREE_UPDATE_JOURNAL_RES \ - ((BKEY_BTREE_PTR_U64s_MAX + 1) * (BTREE_MAX_DEPTH - 1) * 2) +#define BTREE_UPDATE_JOURNAL_RES (BTREE_UPDATE_NODES_MAX * (BKEY_BTREE_PTR_U64s_MAX + 1)) /* * Tracks an in progress split/rewrite of a btree node and the update to the @@ -72,9 +51,8 @@ struct btree_update { unsigned nodes_written:1; enum btree_id btree_id; - u8 level; - struct btree_reserve *reserve; + struct disk_reservation disk_res; struct journal_preres journal_preres; /* @@ -96,17 +74,28 @@ struct btree_update { */ struct journal_entry_pin journal; - /* - * Nodes being freed: - * Protected by c->btree_node_pending_free_lock - */ - struct pending_btree_node_free pending[BTREE_MAX_DEPTH + GC_MERGE_NODES]; - unsigned nr_pending; + /* Preallocated nodes we reserve when we start the update: */ + struct btree *prealloc_nodes[BTREE_UPDATE_NODES_MAX]; + unsigned nr_prealloc_nodes; + + /* Nodes being freed: */ + struct keylist old_keys; + u64 _old_keys[BTREE_UPDATE_NODES_MAX * + BKEY_BTREE_PTR_VAL_U64s_MAX]; + + /* Nodes being added: */ + struct keylist new_keys; + u64 _new_keys[BTREE_UPDATE_NODES_MAX * + BKEY_BTREE_PTR_VAL_U64s_MAX]; /* New nodes, that will be made reachable by this update: */ - struct btree *new_nodes[BTREE_MAX_DEPTH * 2 + GC_MERGE_NODES]; + struct btree *new_nodes[BTREE_UPDATE_NODES_MAX]; unsigned nr_new_nodes; + open_bucket_idx_t open_buckets[BTREE_UPDATE_NODES_MAX * + BCH_REPLICAS_MAX]; + open_bucket_idx_t nr_open_buckets; + unsigned journal_u64s; u64 journal_entries[BTREE_UPDATE_JOURNAL_RES]; @@ -120,14 +109,12 @@ struct btree_update { u64 inline_keys[BKEY_BTREE_PTR_U64s_MAX * 3]; }; -#define for_each_pending_btree_node_free(c, as, p) \ - list_for_each_entry(as, &c->btree_interior_update_list, list) \ - for (p = as->pending; p < as->pending + as->nr_pending; p++) - void bch2_btree_node_free_inmem(struct bch_fs *, struct btree *, struct btree_iter *); void bch2_btree_node_free_never_inserted(struct bch_fs *, struct btree *); +void bch2_btree_update_get_open_buckets(struct btree_update *, struct btree *); + struct btree *__bch2_btree_node_alloc_replacement(struct btree_update *, struct btree *, struct bkey_format); @@ -139,6 +126,7 @@ bch2_btree_update_start(struct btree_trans *, enum btree_id, unsigned, void bch2_btree_interior_update_will_free_node(struct btree_update *, struct btree *); +void bch2_btree_update_add_new_node(struct btree_update *, struct btree *); void bch2_btree_insert_node(struct btree_update *, struct btree *, struct btree_iter *, struct keylist *, @@ -185,7 +173,7 @@ void bch2_btree_root_alloc(struct bch_fs *, enum btree_id); static inline unsigned btree_update_reserve_required(struct bch_fs *c, struct btree *b) { - unsigned depth = btree_node_root(c, b)->level + 1; + unsigned depth = btree_node_root(c, b)->c.level + 1; /* * Number of nodes we might have to allocate in a worst case btree @@ -193,9 +181,9 @@ static inline unsigned btree_update_reserve_required(struct bch_fs *c, * a new root, unless we're already at max depth: */ if (depth < BTREE_MAX_DEPTH) - return (depth - b->level) * 2 + 1; + return (depth - b->c.level) * 2 + 1; else - return (depth - b->level) * 2 - 1; + return (depth - b->c.level) * 2 - 1; } static inline void btree_node_reset_sib_u64s(struct btree *b) @@ -333,4 +321,11 @@ ssize_t bch2_btree_updates_print(struct bch_fs *, char *); size_t bch2_btree_interior_updates_nr_pending(struct bch_fs *); +void bch2_journal_entries_to_btree_roots(struct bch_fs *, struct jset *); +struct jset_entry *bch2_btree_roots_to_journal_entries(struct bch_fs *, + struct jset_entry *, struct jset_entry *); + +void bch2_fs_btree_interior_update_exit(struct bch_fs *); +int bch2_fs_btree_interior_update_init(struct bch_fs *); + #endif /* _BCACHEFS_BTREE_UPDATE_INTERIOR_H */ diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c index 7faf98fd2f64..6e9688d0bb77 100644 --- a/fs/bcachefs/btree_update_leaf.c +++ b/fs/bcachefs/btree_update_leaf.c @@ -6,6 +6,7 @@ #include "btree_gc.h" #include "btree_io.h" #include "btree_iter.h" +#include "btree_key_cache.h" #include "btree_locking.h" #include "buckets.h" #include "debug.h" @@ -32,6 +33,9 @@ inline void bch2_btree_node_lock_for_insert(struct bch_fs *c, struct btree *b, { bch2_btree_node_lock_write(b, iter); + if (btree_iter_type(iter) == BTREE_ITER_CACHED) + return; + if (unlikely(btree_node_just_written(b)) && bch2_btree_post_write_cleanup(c, b)) bch2_btree_iter_reinit_node(iter, b); @@ -135,7 +139,7 @@ static void __btree_node_flush(struct journal *j, struct journal_entry_pin *pin, btree_node_lock_type(c, b, SIX_LOCK_read); bch2_btree_node_write_cond(c, b, (btree_current_write(b) == w && w->journal.seq == seq)); - six_unlock_read(&b->lock); + six_unlock_read(&b->c.lock); } static void btree_node_flush0(struct journal *j, struct journal_entry_pin *pin, u64 seq) @@ -159,71 +163,35 @@ inline void bch2_btree_add_journal_pin(struct bch_fs *c, : btree_node_flush1); } -static inline void __btree_journal_key(struct btree_trans *trans, - enum btree_id btree_id, - struct bkey_i *insert) -{ - struct journal *j = &trans->c->journal; - u64 seq = trans->journal_res.seq; - bool needs_whiteout = insert->k.needs_whiteout; - - /* ick */ - insert->k.needs_whiteout = false; - bch2_journal_add_keys(j, &trans->journal_res, - btree_id, insert); - insert->k.needs_whiteout = needs_whiteout; - - bch2_journal_set_has_inode(j, &trans->journal_res, - insert->k.p.inode); - - if (trans->journal_seq) - *trans->journal_seq = seq; -} - -static void bch2_btree_journal_key(struct btree_trans *trans, - struct btree_iter *iter, - struct bkey_i *insert) -{ - struct bch_fs *c = trans->c; - struct journal *j = &c->journal; - struct btree *b = iter_l(iter)->b; - - EBUG_ON(trans->journal_res.ref != - !(trans->flags & BTREE_INSERT_JOURNAL_REPLAY)); - - if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) { - __btree_journal_key(trans, iter->btree_id, insert); - btree_bset_last(b)->journal_seq = - cpu_to_le64(trans->journal_res.seq); - } - - bch2_btree_add_journal_pin(c, b, - likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY)) - ? trans->journal_res.seq - : j->replay_journal_seq); - - if (unlikely(!btree_node_dirty(b))) - set_btree_node_dirty(b); -} - /** * btree_insert_key - insert a key one key into a leaf node */ -static void btree_insert_key_leaf(struct btree_trans *trans, +static bool btree_insert_key_leaf(struct btree_trans *trans, struct btree_iter *iter, struct bkey_i *insert) { struct bch_fs *c = trans->c; struct btree *b = iter_l(iter)->b; struct bset_tree *t = bset_tree_last(b); + struct bset *i = bset(b, t); int old_u64s = bset_u64s(t); int old_live_u64s = b->nr.live_u64s; int live_u64s_added, u64s_added; - insert->k.needs_whiteout = false; + EBUG_ON(!iter->level && + !test_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags)); + + if (unlikely(!bch2_btree_bset_insert_key(iter, b, + &iter_l(iter)->iter, insert))) + return false; + + i->journal_seq = cpu_to_le64(max(trans->journal_res.seq, + le64_to_cpu(i->journal_seq))); + + bch2_btree_add_journal_pin(c, b, trans->journal_res.seq); - if (likely(bch2_btree_bset_insert_key(iter, b, &iter_l(iter)->iter, insert))) - bch2_btree_journal_key(trans, iter, insert); + if (unlikely(!btree_node_dirty(b))) + set_btree_node_dirty(b); live_u64s_added = (int) b->nr.live_u64s - old_live_u64s; u64s_added = (int) bset_u64s(t) - old_u64s; @@ -238,8 +206,11 @@ static void btree_insert_key_leaf(struct btree_trans *trans, bch2_btree_iter_reinit_node(iter, b); trace_btree_insert_key(c, b, insert); + return true; } +/* Cached btree updates: */ + /* Normal update interface: */ static inline void btree_insert_entry_checks(struct btree_trans *trans, @@ -310,7 +281,7 @@ btree_key_can_insert(struct btree_trans *trans, if (unlikely(btree_node_old_extent_overwrite(b))) return BTREE_INSERT_BTREE_NODE_FULL; - ret = !(iter->flags & BTREE_ITER_IS_EXTENTS) + ret = !btree_iter_is_extents(iter) ? BTREE_INSERT_OK : bch2_extent_can_insert(trans, iter, insert); if (ret) @@ -322,11 +293,60 @@ btree_key_can_insert(struct btree_trans *trans, return BTREE_INSERT_OK; } +static enum btree_insert_ret +btree_key_can_insert_cached(struct btree_trans *trans, + struct btree_iter *iter, + struct bkey_i *insert, + unsigned *u64s) +{ + struct bkey_cached *ck = (void *) iter->l[0].b; + unsigned new_u64s; + struct bkey_i *new_k; + + BUG_ON(iter->level); + + if (*u64s <= ck->u64s) + return BTREE_INSERT_OK; + + new_u64s = roundup_pow_of_two(*u64s); + new_k = krealloc(ck->k, new_u64s * sizeof(u64), GFP_NOFS); + if (!new_k) + return -ENOMEM; + + ck->u64s = new_u64s; + ck->k = new_k; + return BTREE_INSERT_OK; +} + static inline void do_btree_insert_one(struct btree_trans *trans, struct btree_iter *iter, struct bkey_i *insert) { - btree_insert_key_leaf(trans, iter, insert); + struct bch_fs *c = trans->c; + struct journal *j = &c->journal; + bool did_work; + + EBUG_ON(trans->journal_res.ref != + !(trans->flags & BTREE_INSERT_JOURNAL_REPLAY)); + + insert->k.needs_whiteout = false; + + did_work = (btree_iter_type(iter) != BTREE_ITER_CACHED) + ? btree_insert_key_leaf(trans, iter, insert) + : bch2_btree_insert_key_cached(trans, iter, insert); + if (!did_work) + return; + + if (likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) { + bch2_journal_add_keys(j, &trans->journal_res, + iter->btree_id, insert); + + bch2_journal_set_has_inode(j, &trans->journal_res, + insert->k.p.inode); + + if (trans->journal_seq) + *trans->journal_seq = trans->journal_res.seq; + } } static inline bool iter_has_trans_triggers(struct btree_iter *iter) @@ -351,10 +371,16 @@ static noinline void bch2_trans_mark_gc(struct btree_trans *trans) struct bch_fs *c = trans->c; struct btree_insert_entry *i; - trans_for_each_update(trans, i) - if (gc_visited(c, gc_pos_btree_node(iter_l(i->iter)->b))) + trans_for_each_update(trans, i) { + /* + * XXX: synchronization of cached update triggers with gc + */ + BUG_ON(btree_iter_type(i->iter) == BTREE_ITER_CACHED); + + if (gc_visited(c, gc_pos_btree_node(i->iter->l[0].b))) bch2_mark_update(trans, i->iter, i->k, NULL, i->trigger_flags|BTREE_TRIGGER_GC); + } } static inline int @@ -387,7 +413,9 @@ bch2_trans_commit_write_locked(struct btree_trans *trans, u64s = 0; u64s += i->k->k.u64s; - ret = btree_key_can_insert(trans, i->iter, i->k, &u64s); + ret = btree_iter_type(i->iter) != BTREE_ITER_CACHED + ? btree_key_can_insert(trans, i->iter, i->k, &u64s) + : btree_key_can_insert_cached(trans, i->iter, i->k, &u64s); if (ret) { *stopped_at = i; return ret; @@ -411,6 +439,17 @@ bch2_trans_commit_write_locked(struct btree_trans *trans, JOURNAL_RES_GET_NONBLOCK); if (ret) goto err; + } else { + trans->journal_res.seq = c->journal.replay_journal_seq; + } + + if (unlikely(trans->extra_journal_entry_u64s)) { + memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res), + trans->extra_journal_entries, + trans->extra_journal_entry_u64s); + + trans->journal_res.offset += trans->extra_journal_entry_u64s; + trans->journal_res.u64s -= trans->extra_journal_entry_u64s; } /* @@ -472,7 +511,9 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans, ret = bch2_journal_preres_get(&trans->c->journal, &trans->journal_preres, trans->journal_preres_u64s, - JOURNAL_RES_GET_NONBLOCK); + JOURNAL_RES_GET_NONBLOCK| + ((trans->flags & BTREE_INSERT_JOURNAL_RECLAIM) + ? JOURNAL_RES_GET_RECLAIM : 0)); if (unlikely(ret == -EAGAIN)) ret = bch2_trans_journal_preres_get_cold(trans, trans->journal_preres_u64s); @@ -486,7 +527,7 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans, * or anything else that might call bch2_trans_relock(), since that * would just retake the read locks: */ - trans_for_each_iter_all(trans, iter) { + trans_for_each_iter(trans, iter) { if (iter->nodes_locked != iter->nodes_intent_locked) { EBUG_ON(iter->flags & BTREE_ITER_KEEP_UNTIL_COMMIT); EBUG_ON(trans->iters_live & (1ULL << iter->idx)); @@ -511,6 +552,10 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans, bch2_btree_node_unlock_write_inlined(iter_l(i->iter)->b, i->iter); + if (!ret && trans->journal_pin) + bch2_journal_pin_add(&trans->c->journal, trans->journal_res.seq, + trans->journal_pin, NULL); + /* * Drop journal reservation after dropping write locks, since dropping * the journal reservation may kick off a journal write: @@ -524,14 +569,14 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans, trans->nounlock = true; trans_for_each_update2(trans, i) - if (!same_leaf_as_prev(trans, i)) + if (btree_iter_type(i->iter) != BTREE_ITER_CACHED && + !same_leaf_as_prev(trans, i)) bch2_foreground_maybe_merge(trans->c, i->iter, 0, trans->flags); trans->nounlock = false; - trans_for_each_update2(trans, i) - bch2_btree_iter_downgrade(i->iter); + bch2_trans_downgrade(trans); return 0; } @@ -800,7 +845,7 @@ int __bch2_trans_commit(struct btree_trans *trans) memset(&trans->journal_preres, 0, sizeof(trans->journal_preres)); - trans->journal_u64s = 0; + trans->journal_u64s = trans->extra_journal_entry_u64s; trans->journal_preres_u64s = 0; if (!(trans->flags & BTREE_INSERT_NOCHECK_RW) && @@ -810,6 +855,14 @@ int __bch2_trans_commit(struct btree_trans *trans) return ret; } +#ifdef CONFIG_BCACHEFS_DEBUG + trans_for_each_update(trans, i) + if (btree_iter_type(i->iter) != BTREE_ITER_CACHED && + !(i->trigger_flags & BTREE_TRIGGER_NORUN)) + bch2_btree_key_cache_verify_clean(trans, + i->iter->btree_id, i->iter->pos); +#endif + /* * Running triggers will append more updates to the list of updates as * we're walking it: @@ -818,9 +871,9 @@ int __bch2_trans_commit(struct btree_trans *trans) trans_trigger_run = false; trans_for_each_update(trans, i) { - if (unlikely(i->iter->uptodate > BTREE_ITER_NEED_PEEK)) { + if (unlikely(i->iter->uptodate > BTREE_ITER_NEED_PEEK && + (ret = bch2_btree_iter_traverse(i->iter)))) { trace_trans_restart_traverse(trans->ip); - ret = -EINTR; goto out; } @@ -882,7 +935,8 @@ int __bch2_trans_commit(struct btree_trans *trans) BUG_ON(i->iter->locks_want < 1); u64s = jset_u64s(i->k->k.u64s); - if (0) + if (btree_iter_type(i->iter) == BTREE_ITER_CACHED && + likely(!(trans->flags & BTREE_INSERT_JOURNAL_REPLAY))) trans->journal_preres_u64s += u64s; trans->journal_u64s += u64s; } diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c index 2e1df04c760d..0ec194b93c71 100644 --- a/fs/bcachefs/buckets.c +++ b/fs/bcachefs/buckets.c @@ -374,6 +374,11 @@ static inline int is_fragmented_bucket(struct bucket_mark m, return 0; } +static inline int bucket_stripe_sectors(struct bucket_mark m) +{ + return m.stripe ? m.dirty_sectors : 0; +} + static inline enum bch_data_type bucket_type(struct bucket_mark m) { return m.cached_sectors && !m.dirty_sectors @@ -441,33 +446,35 @@ static void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca, struct bucket_mark old, struct bucket_mark new, bool gc) { - struct bch_dev_usage *dev_usage; + struct bch_dev_usage *u; percpu_rwsem_assert_held(&c->mark_lock); preempt_disable(); - dev_usage = this_cpu_ptr(ca->usage[gc]); + u = this_cpu_ptr(ca->usage[gc]); if (bucket_type(old)) - account_bucket(fs_usage, dev_usage, bucket_type(old), + account_bucket(fs_usage, u, bucket_type(old), -1, -ca->mi.bucket_size); if (bucket_type(new)) - account_bucket(fs_usage, dev_usage, bucket_type(new), + account_bucket(fs_usage, u, bucket_type(new), 1, ca->mi.bucket_size); - dev_usage->buckets_alloc += + u->buckets_alloc += (int) new.owned_by_allocator - (int) old.owned_by_allocator; - dev_usage->buckets_ec += - (int) new.stripe - (int) old.stripe; - dev_usage->buckets_unavailable += + u->buckets_unavailable += is_unavailable_bucket(new) - is_unavailable_bucket(old); - dev_usage->sectors[old.data_type] -= old.dirty_sectors; - dev_usage->sectors[new.data_type] += new.dirty_sectors; - dev_usage->sectors[BCH_DATA_CACHED] += + u->buckets_ec += (int) new.stripe - (int) old.stripe; + u->sectors_ec += bucket_stripe_sectors(new) - + bucket_stripe_sectors(old); + + u->sectors[old.data_type] -= old.dirty_sectors; + u->sectors[new.data_type] += new.dirty_sectors; + u->sectors[BCH_DATA_CACHED] += (int) new.cached_sectors - (int) old.cached_sectors; - dev_usage->sectors_fragmented += + u->sectors_fragmented += is_fragmented_bucket(new, ca) - is_fragmented_bucket(old, ca); preempt_enable(); @@ -778,29 +785,31 @@ static int bch2_mark_alloc(struct bch_fs *c, struct bkey_s_c k, }) static int __bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca, - size_t b, enum bch_data_type type, + size_t b, enum bch_data_type data_type, unsigned sectors, bool gc) { struct bucket *g = __bucket(ca, b, gc); struct bucket_mark old, new; bool overflow; - BUG_ON(type != BCH_DATA_SB && - type != BCH_DATA_JOURNAL); + BUG_ON(data_type != BCH_DATA_SB && + data_type != BCH_DATA_JOURNAL); old = bucket_cmpxchg(g, new, ({ - new.data_type = type; + new.data_type = data_type; overflow = checked_add(new.dirty_sectors, sectors); })); bch2_fs_inconsistent_on(old.data_type && - old.data_type != type, c, + old.data_type != data_type, c, "different types of data in same bucket: %s, %s", bch2_data_types[old.data_type], - bch2_data_types[type]); + bch2_data_types[data_type]); bch2_fs_inconsistent_on(overflow, c, - "bucket sector count overflow: %u + %u > U16_MAX", + "bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > U16_MAX", + ca->dev_idx, b, new.gen, + bch2_data_types[old.data_type ?: data_type], old.dirty_sectors, sectors); if (c) @@ -916,58 +925,117 @@ static void bucket_set_stripe(struct bch_fs *c, } } -static bool bch2_mark_pointer(struct bch_fs *c, - struct extent_ptr_decoded p, - s64 sectors, enum bch_data_type data_type, - struct bch_fs_usage *fs_usage, - u64 journal_seq, unsigned flags) +static int __mark_pointer(struct bch_fs *c, struct bkey_s_c k, + struct extent_ptr_decoded p, + s64 sectors, enum bch_data_type ptr_data_type, + u8 bucket_gen, u8 *bucket_data_type, + u16 *dirty_sectors, u16 *cached_sectors) +{ + u16 *dst_sectors = !p.ptr.cached + ? dirty_sectors + : cached_sectors; + u16 orig_sectors = *dst_sectors; + char buf[200]; + + if (gen_after(p.ptr.gen, bucket_gen)) { + bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, + "bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(bch_dev_bkey_exists(c, p.ptr.dev), &p.ptr), + bucket_gen, + bch2_data_types[*bucket_data_type ?: ptr_data_type], + p.ptr.gen, + (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); + return -EIO; + } + + if (gen_cmp(bucket_gen, p.ptr.gen) >= 96U) { + bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, + "bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(bch_dev_bkey_exists(c, p.ptr.dev), &p.ptr), + bucket_gen, + bch2_data_types[*bucket_data_type ?: ptr_data_type], + p.ptr.gen, + (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); + return -EIO; + } + + if (bucket_gen != p.ptr.gen && !p.ptr.cached) { + bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, + "bucket %u:%zu gen %u data type %s: stale dirty ptr (gen %u)\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(bch_dev_bkey_exists(c, p.ptr.dev), &p.ptr), + bucket_gen, + bch2_data_types[*bucket_data_type ?: ptr_data_type], + p.ptr.gen, + (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); + return -EIO; + } + + if (bucket_gen != p.ptr.gen) + return 1; + + if (*bucket_data_type && *bucket_data_type != ptr_data_type) { + bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, + "bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(bch_dev_bkey_exists(c, p.ptr.dev), &p.ptr), + bucket_gen, + bch2_data_types[*bucket_data_type], + bch2_data_types[ptr_data_type], + (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); + return -EIO; + } + + if (checked_add(*dst_sectors, sectors)) { + bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, + "bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U16_MAX\n" + "while marking %s", + p.ptr.dev, PTR_BUCKET_NR(bch_dev_bkey_exists(c, p.ptr.dev), &p.ptr), + bucket_gen, + bch2_data_types[*bucket_data_type ?: ptr_data_type], + orig_sectors, sectors, + (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf)); + return -EIO; + } + + *bucket_data_type = *dirty_sectors || *cached_sectors + ? ptr_data_type : 0; + return 0; +} + +static int bch2_mark_pointer(struct bch_fs *c, struct bkey_s_c k, + struct extent_ptr_decoded p, + s64 sectors, enum bch_data_type data_type, + struct bch_fs_usage *fs_usage, + u64 journal_seq, unsigned flags) { bool gc = flags & BTREE_TRIGGER_GC; struct bucket_mark old, new; struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev); struct bucket *g = PTR_BUCKET(ca, &p.ptr, gc); - bool overflow; + u8 bucket_data_type; u64 v; + int ret; v = atomic64_read(&g->_mark.v); do { new.v.counter = old.v.counter = v; + bucket_data_type = new.data_type; - /* - * Check this after reading bucket mark to guard against - * the allocator invalidating a bucket after we've already - * checked the gen - */ - if (gen_after(p.ptr.gen, new.gen)) { - bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "pointer gen in the future"); - return true; - } - - if (new.gen != p.ptr.gen) { - /* XXX write repair code for this */ - if (!p.ptr.cached && - test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)) - bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "stale dirty pointer"); - return true; - } - - if (!p.ptr.cached) - overflow = checked_add(new.dirty_sectors, sectors); - else - overflow = checked_add(new.cached_sectors, sectors); + ret = __mark_pointer(c, k, p, sectors, data_type, new.gen, + &bucket_data_type, + &new.dirty_sectors, + &new.cached_sectors); + if (ret) + return ret; - if (!new.dirty_sectors && - !new.cached_sectors) { - new.data_type = 0; + new.data_type = bucket_data_type; - if (journal_seq) { - new.journal_seq_valid = 1; - new.journal_seq = journal_seq; - } - } else { - new.data_type = data_type; + if (journal_seq) { + new.journal_seq_valid = 1; + new.journal_seq = journal_seq; } if (flags & BTREE_TRIGGER_NOATOMIC) { @@ -978,25 +1046,11 @@ static bool bch2_mark_pointer(struct bch_fs *c, old.v.counter, new.v.counter)) != old.v.counter); - if (old.data_type && old.data_type != data_type) - bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "bucket %u:%zu gen %u different types of data in same bucket: %s, %s", - p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr), - new.gen, - bch2_data_types[old.data_type], - bch2_data_types[data_type]); - - bch2_fs_inconsistent_on(overflow, c, - "bucket sector count overflow: %u + %lli > U16_MAX", - !p.ptr.cached - ? old.dirty_sectors - : old.cached_sectors, sectors); - bch2_dev_usage_update(c, ca, fs_usage, old, new, gc); BUG_ON(!gc && bucket_became_unavailable(old, new)); - return false; + return 0; } static int bch2_mark_stripe_ptr(struct bch_fs *c, @@ -1060,6 +1114,7 @@ static int bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k, struct extent_ptr_decoded p; struct bch_replicas_padded r; s64 dirty_sectors = 0; + bool stale; int ret; r.e.data_type = data_type; @@ -1072,8 +1127,13 @@ static int bch2_mark_extent(struct bch_fs *c, struct bkey_s_c k, s64 disk_sectors = data_type == BCH_DATA_BTREE ? sectors : ptr_disk_sectors_delta(p, offset, sectors, flags); - bool stale = bch2_mark_pointer(c, p, disk_sectors, data_type, - fs_usage, journal_seq, flags); + + ret = bch2_mark_pointer(c, k, p, disk_sectors, data_type, + fs_usage, journal_seq, flags); + if (ret < 0) + return ret; + + stale = ret > 0; if (p.ptr.cached) { if (!stale) @@ -1175,7 +1235,7 @@ static int bch2_mark_stripe(struct bch_fs *c, struct bkey_s_c k, return 0; } -int bch2_mark_key_locked(struct bch_fs *c, +static int bch2_mark_key_locked(struct bch_fs *c, struct bkey_s_c k, unsigned offset, s64 sectors, struct bch_fs_usage *fs_usage, @@ -1307,8 +1367,8 @@ int bch2_mark_update(struct btree_trans *trans, unsigned flags) { struct bch_fs *c = trans->c; - struct btree *b = iter->l[0].b; - struct btree_node_iter node_iter = iter->l[0].iter; + struct btree *b = iter_l(iter)->b; + struct btree_node_iter node_iter = iter_l(iter)->iter; struct bkey_packed *_k; int ret = 0; @@ -1370,45 +1430,49 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans, disk_res_sectors); trans_for_each_update(trans, i) { - struct btree_iter *iter = i->iter; - struct btree *b = iter->l[0].b; - struct btree_node_iter node_iter = iter->l[0].iter; - struct bkey_packed *_k; - pr_err("while inserting"); bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(i->k)); pr_err("%s", buf); pr_err("overlapping with"); - node_iter = iter->l[0].iter; - while ((_k = bch2_btree_node_iter_peek(&node_iter, b))) { - struct bkey unpacked; - struct bkey_s_c k; + if (btree_iter_type(i->iter) != BTREE_ITER_CACHED) { + struct btree *b = iter_l(i->iter)->b; + struct btree_node_iter node_iter = iter_l(i->iter)->iter; + struct bkey_packed *_k; - k = bkey_disassemble(b, _k, &unpacked); + while ((_k = bch2_btree_node_iter_peek(&node_iter, b))) { + struct bkey unpacked; + struct bkey_s_c k; - if (btree_node_is_extents(b) - ? bkey_cmp(i->k->k.p, bkey_start_pos(k.k)) <= 0 - : bkey_cmp(i->k->k.p, k.k->p)) - break; + pr_info("_k %px format %u", _k, _k->format); + k = bkey_disassemble(b, _k, &unpacked); - bch2_bkey_val_to_text(&PBUF(buf), c, k); - pr_err("%s", buf); + if (btree_node_is_extents(b) + ? bkey_cmp(i->k->k.p, bkey_start_pos(k.k)) <= 0 + : bkey_cmp(i->k->k.p, k.k->p)) + break; + + bch2_bkey_val_to_text(&PBUF(buf), c, k); + pr_err("%s", buf); + + bch2_btree_node_iter_advance(&node_iter, b); + } + } else { + struct bkey_cached *ck = (void *) i->iter->l[0].b; - bch2_btree_node_iter_advance(&node_iter, b); + bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(ck->k)); + pr_err("%s", buf); } } } /* trans_mark: */ -static int trans_get_key(struct btree_trans *trans, - enum btree_id btree_id, struct bpos pos, - struct btree_iter **iter, - struct bkey_s_c *k) +static struct btree_iter *trans_get_update(struct btree_trans *trans, + enum btree_id btree_id, struct bpos pos, + struct bkey_s_c *k) { struct btree_insert_entry *i; - int ret; trans_for_each_update(trans, i) if (i->iter->btree_id == btree_id && @@ -1416,17 +1480,33 @@ static int trans_get_key(struct btree_trans *trans, ? bkey_cmp(pos, bkey_start_pos(&i->k->k)) >= 0 && bkey_cmp(pos, i->k->k.p) < 0 : !bkey_cmp(pos, i->iter->pos))) { - *iter = i->iter; - *k = bkey_i_to_s_c(i->k); - return 1; + *k = bkey_i_to_s_c(i->k); + return i->iter; } + return NULL; +} + +static int trans_get_key(struct btree_trans *trans, + enum btree_id btree_id, struct bpos pos, + struct btree_iter **iter, + struct bkey_s_c *k) +{ + unsigned flags = btree_id != BTREE_ID_ALLOC + ? BTREE_ITER_SLOTS + : BTREE_ITER_CACHED; + int ret; + + *iter = trans_get_update(trans, btree_id, pos, k); + if (*iter) + return 1; + *iter = bch2_trans_get_iter(trans, btree_id, pos, - BTREE_ITER_SLOTS|BTREE_ITER_INTENT); + flags|BTREE_ITER_INTENT); if (IS_ERR(*iter)) return PTR_ERR(*iter); - *k = bch2_btree_iter_peek_slot(*iter); + *k = __bch2_btree_iter_peek(*iter, flags); ret = bkey_err(*k); if (ret) bch2_trans_iter_put(trans, *iter); @@ -1434,85 +1514,44 @@ static int trans_get_key(struct btree_trans *trans, } static int bch2_trans_mark_pointer(struct btree_trans *trans, - struct extent_ptr_decoded p, + struct bkey_s_c k, struct extent_ptr_decoded p, s64 sectors, enum bch_data_type data_type) { struct bch_fs *c = trans->c; struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev); + struct bpos pos = POS(p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr)); struct btree_iter *iter; - struct bkey_s_c k; + struct bkey_s_c k_a; struct bkey_alloc_unpacked u; struct bkey_i_alloc *a; - u16 *dst_sectors, orig_sectors; + struct bucket *g; int ret; - ret = trans_get_key(trans, BTREE_ID_ALLOC, - POS(p.ptr.dev, PTR_BUCKET_NR(ca, &p.ptr)), - &iter, &k); - if (ret < 0) - return ret; - - if (!ret && unlikely(!test_bit(BCH_FS_ALLOC_WRITTEN, &c->flags))) { - /* - * During journal replay, and if gc repairs alloc info at - * runtime, the alloc info in the btree might not be up to date - * yet - so, trust the in memory mark: - */ - struct bucket *g; - struct bucket_mark m; - - percpu_down_read(&c->mark_lock); - g = bucket(ca, iter->pos.offset); - m = READ_ONCE(g->mark); - u = alloc_mem_to_key(g, m); - percpu_up_read(&c->mark_lock); + iter = trans_get_update(trans, BTREE_ID_ALLOC, pos, &k_a); + if (iter) { + u = bch2_alloc_unpack(k_a); } else { - /* - * Unless we're already updating that key: - */ - if (k.k->type != KEY_TYPE_alloc) { - bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "pointer to nonexistent bucket %llu:%llu", - iter->pos.inode, iter->pos.offset); - ret = -1; + iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, pos, + BTREE_ITER_CACHED| + BTREE_ITER_CACHED_NOFILL| + BTREE_ITER_INTENT); + if (IS_ERR(iter)) + return PTR_ERR(iter); + + ret = bch2_btree_iter_traverse(iter); + if (ret) goto out; - } - - u = bch2_alloc_unpack(k); - } - - if (gen_after(u.gen, p.ptr.gen)) { - ret = 1; - goto out; - } - if (u.data_type && u.data_type != data_type) { - bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK, - "bucket %llu:%llu gen %u different types of data in same bucket: %s, %s", - iter->pos.inode, iter->pos.offset, - u.gen, - bch2_data_types[u.data_type], - bch2_data_types[data_type]); - ret = -1; - goto out; + percpu_down_read(&c->mark_lock); + g = bucket(ca, pos.offset); + u = alloc_mem_to_key(g, READ_ONCE(g->mark)); + percpu_up_read(&c->mark_lock); } - dst_sectors = !p.ptr.cached - ? &u.dirty_sectors - : &u.cached_sectors; - orig_sectors = *dst_sectors; - - if (checked_add(*dst_sectors, sectors)) { - bch2_fs_inconsistent(c, - "bucket sector count overflow: %u + %lli > U16_MAX", - orig_sectors, sectors); - /* return an error indicating that we need full fsck */ - ret = -EIO; + ret = __mark_pointer(c, k, p, sectors, data_type, u.gen, &u.data_type, + &u.dirty_sectors, &u.cached_sectors); + if (ret) goto out; - } - - u.data_type = u.dirty_sectors || u.cached_sectors - ? data_type : 0; a = bch2_trans_kmalloc(trans, BKEY_ALLOC_U64s_MAX * 8); ret = PTR_ERR_OR_ZERO(a); @@ -1520,7 +1559,7 @@ static int bch2_trans_mark_pointer(struct btree_trans *trans, goto out; bkey_alloc_init(&a->k_i); - a->k.p = iter->pos; + a->k.p = pos; bch2_alloc_pack(a, u); bch2_trans_update(trans, iter, &a->k_i, 0); out: @@ -1597,7 +1636,7 @@ static int bch2_trans_mark_extent(struct btree_trans *trans, ? sectors : ptr_disk_sectors_delta(p, offset, sectors, flags); - ret = bch2_trans_mark_pointer(trans, p, disk_sectors, + ret = bch2_trans_mark_pointer(trans, k, p, disk_sectors, data_type); if (ret < 0) return ret; @@ -1774,8 +1813,8 @@ int bch2_trans_mark_update(struct btree_trans *trans, struct bkey_i *insert, unsigned flags) { - struct btree *b = iter->l[0].b; - struct btree_node_iter node_iter = iter->l[0].iter; + struct btree *b = iter_l(iter)->b; + struct btree_node_iter node_iter = iter_l(iter)->iter; struct bkey_packed *_k; int ret; @@ -1793,6 +1832,13 @@ int bch2_trans_mark_update(struct btree_trans *trans, if (unlikely(flags & BTREE_TRIGGER_NOOVERWRITES)) return 0; + if (btree_iter_type(iter) == BTREE_ITER_CACHED) { + struct bkey_cached *ck = (void *) iter->l[0].b; + + return bch2_trans_mark_key(trans, bkey_i_to_s_c(ck->k), + 0, 0, BTREE_TRIGGER_OVERWRITE); + } + while ((_k = bch2_btree_node_iter_peek(&node_iter, b))) { struct bkey unpacked; struct bkey_s_c k; @@ -2009,8 +2055,10 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) swap(ca->buckets_nouse, buckets_nouse); - if (resize) + if (resize) { percpu_up_write(&c->mark_lock); + up_write(&c->gc_lock); + } spin_lock(&c->freelist_lock); for (i = 0; i < RESERVE_NR; i++) { @@ -2029,10 +2077,8 @@ int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) nbuckets = ca->mi.nbuckets; - if (resize) { + if (resize) up_write(&ca->bucket_lock); - up_write(&c->gc_lock); - } if (start_copygc && bch2_copygc_start(c, ca)) diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h index 765650ce9d0a..97265fe90e96 100644 --- a/fs/bcachefs/buckets.h +++ b/fs/bcachefs/buckets.h @@ -259,8 +259,6 @@ void bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *, size_t, enum bch_data_type, unsigned, struct gc_pos, unsigned); -int bch2_mark_key_locked(struct bch_fs *, struct bkey_s_c, unsigned, s64, - struct bch_fs_usage *, u64, unsigned); int bch2_mark_key(struct bch_fs *, struct bkey_s_c, unsigned, s64, struct bch_fs_usage *, u64, unsigned); int bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *, diff --git a/fs/bcachefs/buckets_types.h b/fs/bcachefs/buckets_types.h index f3ff4a18b1fd..53f22726893d 100644 --- a/fs/bcachefs/buckets_types.h +++ b/fs/bcachefs/buckets_types.h @@ -39,6 +39,7 @@ struct bucket { u16 io_time[2]; u8 oldest_gen; + u8 gc_gen; unsigned gen_valid:1; }; @@ -52,12 +53,14 @@ struct bucket_array { struct bch_dev_usage { u64 buckets[BCH_DATA_NR]; u64 buckets_alloc; - u64 buckets_ec; u64 buckets_unavailable; /* _compressed_ sectors: */ u64 sectors[BCH_DATA_NR]; u64 sectors_fragmented; + + u64 buckets_ec; + u64 sectors_ec; }; struct bch_fs_usage { diff --git a/fs/bcachefs/chardev.c b/fs/bcachefs/chardev.c index 5028d0dcc2d6..3af521947502 100644 --- a/fs/bcachefs/chardev.c +++ b/fs/bcachefs/chardev.c @@ -470,9 +470,12 @@ static long bch2_ioctl_dev_usage(struct bch_fs *c, src = bch2_dev_usage_read(c, ca); - arg.state = ca->mi.state; - arg.bucket_size = ca->mi.bucket_size; - arg.nr_buckets = ca->mi.nbuckets - ca->mi.first_bucket; + arg.state = ca->mi.state; + arg.bucket_size = ca->mi.bucket_size; + arg.nr_buckets = ca->mi.nbuckets - ca->mi.first_bucket; + arg.available_buckets = arg.nr_buckets - src.buckets_unavailable; + arg.ec_buckets = src.buckets_ec; + arg.ec_sectors = src.sectors_ec; for (i = 0; i < BCH_DATA_NR; i++) { arg.buckets[i] = src.buckets[i]; diff --git a/fs/bcachefs/checksum.c b/fs/bcachefs/checksum.c index a01073e54a33..3d88719ba86c 100644 --- a/fs/bcachefs/checksum.c +++ b/fs/bcachefs/checksum.c @@ -10,7 +10,7 @@ #include <linux/random.h> #include <linux/scatterlist.h> #include <crypto/algapi.h> -#include <crypto/chacha20.h> +#include <crypto/chacha.h> #include <crypto/hash.h> #include <crypto/poly1305.h> #include <crypto/skcipher.h> @@ -68,21 +68,21 @@ static u64 bch2_checksum_update(unsigned type, u64 crc, const void *data, size_t } } -static inline void do_encrypt_sg(struct crypto_skcipher *tfm, +static inline void do_encrypt_sg(struct crypto_sync_skcipher *tfm, struct nonce nonce, struct scatterlist *sg, size_t len) { - SKCIPHER_REQUEST_ON_STACK(req, tfm); + SYNC_SKCIPHER_REQUEST_ON_STACK(req, tfm); int ret; - skcipher_request_set_tfm(req, tfm); + skcipher_request_set_sync_tfm(req, tfm); skcipher_request_set_crypt(req, sg, sg, len, nonce.d); ret = crypto_skcipher_encrypt(req); BUG_ON(ret); } -static inline void do_encrypt(struct crypto_skcipher *tfm, +static inline void do_encrypt(struct crypto_sync_skcipher *tfm, struct nonce nonce, void *buf, size_t len) { @@ -95,8 +95,8 @@ static inline void do_encrypt(struct crypto_skcipher *tfm, int bch2_chacha_encrypt_key(struct bch_key *key, struct nonce nonce, void *buf, size_t len) { - struct crypto_skcipher *chacha20 = - crypto_alloc_skcipher("chacha20", 0, 0); + struct crypto_sync_skcipher *chacha20 = + crypto_alloc_sync_skcipher("chacha20", 0, 0); int ret; if (!chacha20) { @@ -104,7 +104,8 @@ int bch2_chacha_encrypt_key(struct bch_key *key, struct nonce nonce, return PTR_ERR(chacha20); } - ret = crypto_skcipher_setkey(chacha20, (void *) key, sizeof(*key)); + ret = crypto_skcipher_setkey(&chacha20->base, + (void *) key, sizeof(*key)); if (ret) { pr_err("crypto_skcipher_setkey() error: %i", ret); goto err; @@ -112,7 +113,7 @@ int bch2_chacha_encrypt_key(struct bch_key *key, struct nonce nonce, do_encrypt(chacha20, nonce, buf, len); err: - crypto_free_skcipher(chacha20); + crypto_free_sync_skcipher(chacha20); return ret; } @@ -199,7 +200,7 @@ static struct bch_csum __bch2_checksum_bio(struct bch_fs *c, unsigned type, kunmap_atomic(p); } #else - __bio_for_each_contig_segment(bv, bio, *iter, *iter) + __bio_for_each_bvec(bv, bio, *iter, *iter) crc = bch2_checksum_update(type, crc, page_address(bv.bv_page) + bv.bv_offset, bv.bv_len); @@ -224,7 +225,7 @@ static struct bch_csum __bch2_checksum_bio(struct bch_fs *c, unsigned type, kunmap_atomic(p); } #else - __bio_for_each_contig_segment(bv, bio, *iter, *iter) + __bio_for_each_bvec(bv, bio, *iter, *iter) crypto_shash_update(desc, page_address(bv.bv_page) + bv.bv_offset, bv.bv_len); @@ -463,7 +464,7 @@ err: static int bch2_alloc_ciphers(struct bch_fs *c) { if (!c->chacha20) - c->chacha20 = crypto_alloc_skcipher("chacha20", 0, 0); + c->chacha20 = crypto_alloc_sync_skcipher("chacha20", 0, 0); if (IS_ERR(c->chacha20)) { bch_err(c, "error requesting chacha20 module: %li", PTR_ERR(c->chacha20)); @@ -546,7 +547,7 @@ int bch2_enable_encryption(struct bch_fs *c, bool keyed) goto err; } - ret = crypto_skcipher_setkey(c->chacha20, + ret = crypto_skcipher_setkey(&c->chacha20->base, (void *) &key.key, sizeof(key.key)); if (ret) goto err; @@ -574,7 +575,7 @@ void bch2_fs_encryption_exit(struct bch_fs *c) if (!IS_ERR_OR_NULL(c->poly1305)) crypto_free_shash(c->poly1305); if (!IS_ERR_OR_NULL(c->chacha20)) - crypto_free_skcipher(c->chacha20); + crypto_free_sync_skcipher(c->chacha20); if (!IS_ERR_OR_NULL(c->sha256)) crypto_free_shash(c->sha256); } @@ -606,7 +607,7 @@ int bch2_fs_encryption_init(struct bch_fs *c) if (ret) goto out; - ret = crypto_skcipher_setkey(c->chacha20, + ret = crypto_skcipher_setkey(&c->chacha20->base, (void *) &key.key, sizeof(key.key)); if (ret) goto out; diff --git a/fs/bcachefs/checksum.h b/fs/bcachefs/checksum.h index 833537cc8fd0..24dee8039d57 100644 --- a/fs/bcachefs/checksum.h +++ b/fs/bcachefs/checksum.h @@ -7,7 +7,7 @@ #include "super-io.h" #include <linux/crc64.h> -#include <crypto/chacha20.h> +#include <crypto/chacha.h> static inline bool bch2_checksum_mergeable(unsigned type) { @@ -138,9 +138,9 @@ static inline bool bch2_crc_cmp(struct bch_csum l, struct bch_csum r) /* for skipping ahead and encrypting/decrypting at an offset: */ static inline struct nonce nonce_add(struct nonce nonce, unsigned offset) { - EBUG_ON(offset & (CHACHA20_BLOCK_SIZE - 1)); + EBUG_ON(offset & (CHACHA_BLOCK_SIZE - 1)); - le32_add_cpu(&nonce.d[0], offset / CHACHA20_BLOCK_SIZE); + le32_add_cpu(&nonce.d[0], offset / CHACHA_BLOCK_SIZE); return nonce; } diff --git a/fs/bcachefs/clock.c b/fs/bcachefs/clock.c index d9de0d1302e2..a9f5d5696622 100644 --- a/fs/bcachefs/clock.c +++ b/fs/bcachefs/clock.c @@ -162,7 +162,7 @@ ssize_t bch2_io_timers_show(struct io_clock *clock, char *buf) now = atomic_long_read(&clock->now); for (i = 0; i < clock->timers.used; i++) - pr_buf(&out, "%pf:\t%li\n", + pr_buf(&out, "%ps:\t%li\n", clock->timers.data[i]->fn, clock->timers.data[i]->expire - now); spin_unlock(&clock->timer_lock); diff --git a/fs/bcachefs/compress.c b/fs/bcachefs/compress.c index c431b1e79702..3d75527d2d81 100644 --- a/fs/bcachefs/compress.c +++ b/fs/bcachefs/compress.c @@ -7,6 +7,7 @@ #include "super-io.h" #include <linux/lz4.h> +#include <linux/sched/mm.h> #include <linux/zlib.h> #include <linux/zstd.h> @@ -45,7 +46,7 @@ static bool bio_phys_contig(struct bio *bio, struct bvec_iter start) struct bvec_iter iter; void *expected_start = NULL; - __bio_for_each_segment(bv, bio, iter, start) { + __bio_for_each_bvec(bv, bio, iter, start) { if (expected_start && expected_start != page_address(bv.bv_page) + bv.bv_offset) return false; @@ -63,7 +64,7 @@ static struct bbuf __bio_map_or_bounce(struct bch_fs *c, struct bio *bio, struct bbuf ret; struct bio_vec bv; struct bvec_iter iter; - unsigned nr_pages = 0; + unsigned nr_pages = 0, flags; struct page *stack_pages[16]; struct page **pages = NULL; void *data; @@ -103,7 +104,10 @@ static struct bbuf __bio_map_or_bounce(struct bch_fs *c, struct bio *bio, __bio_for_each_segment(bv, bio, iter, start) pages[nr_pages++] = bv.bv_page; + flags = memalloc_nofs_save(); data = vmap(pages, nr_pages, VM_MAP, PAGE_KERNEL); + memalloc_nofs_restore(flags); + if (pages != stack_pages) kfree(pages); @@ -603,7 +607,7 @@ have_compressed: } if (!mempool_initialized(&c->decompress_workspace)) { - ret = mempool_init_kmalloc_pool( + ret = mempool_init_kvpmalloc_pool( &c->decompress_workspace, 1, decompress_workspace_size); if (ret) diff --git a/fs/bcachefs/debug.c b/fs/bcachefs/debug.c index 69b123bad83b..aa10591a3b1a 100644 --- a/fs/bcachefs/debug.c +++ b/fs/bcachefs/debug.c @@ -52,8 +52,8 @@ void __bch2_btree_verify(struct bch_fs *c, struct btree *b) bkey_copy(&v->key, &b->key); v->written = 0; - v->level = b->level; - v->btree_id = b->btree_id; + v->c.level = b->c.level; + v->c.btree_id = b->c.btree_id; bch2_btree_keys_init(v, &c->expensive_debug_checks); if (bch2_bkey_pick_read_device(c, bkey_i_to_s_c(&b->key), @@ -97,10 +97,10 @@ void __bch2_btree_verify(struct bch_fs *c, struct btree *b) console_lock(); printk(KERN_ERR "*** in memory:\n"); - bch2_dump_bset(b, inmemory, 0); + bch2_dump_bset(c, b, inmemory, 0); printk(KERN_ERR "*** read back in:\n"); - bch2_dump_bset(v, sorted, 0); + bch2_dump_bset(c, v, sorted, 0); while (offset < b->written) { if (!offset ) { @@ -117,7 +117,7 @@ void __bch2_btree_verify(struct bch_fs *c, struct btree *b) } printk(KERN_ERR "*** on disk block %u:\n", offset); - bch2_dump_bset(b, i, offset); + bch2_dump_bset(c, b, i, offset); offset += sectors; } diff --git a/fs/bcachefs/dirent.c b/fs/bcachefs/dirent.c index ae5c9fd8d9f7..f34bfda8ab0d 100644 --- a/fs/bcachefs/dirent.c +++ b/fs/bcachefs/dirent.c @@ -104,7 +104,7 @@ void bch2_dirent_to_text(struct printbuf *out, struct bch_fs *c, bch_scnmemcpy(out, d.v->d_name, bch2_dirent_name_bytes(d)); - pr_buf(out, " -> %llu", d.v->d_inum); + pr_buf(out, " -> %llu type %u", d.v->d_inum, d.v->d_type); } static struct bkey_i_dirent *dirent_create_key(struct btree_trans *trans, diff --git a/fs/bcachefs/ec.c b/fs/bcachefs/ec.c index 933945b65925..9442d6e4041c 100644 --- a/fs/bcachefs/ec.c +++ b/fs/bcachefs/ec.c @@ -1273,38 +1273,28 @@ int bch2_stripes_write(struct bch_fs *c, unsigned flags, bool *wrote) return ret; } -int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys) +static int bch2_stripes_read_fn(struct bch_fs *c, enum btree_id id, + unsigned level, struct bkey_s_c k) { - struct btree_trans trans; - struct btree_and_journal_iter iter; - struct bkey_s_c k; - int ret; - - ret = bch2_fs_ec_start(c); - if (ret) - return ret; - - bch2_trans_init(&trans, c, 0, 0); - - bch2_btree_and_journal_iter_init(&iter, &trans, journal_keys, - BTREE_ID_EC, POS_MIN); - + int ret = 0; - while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { - bch2_mark_key(c, k, 0, 0, NULL, 0, - BTREE_TRIGGER_ALLOC_READ| - BTREE_TRIGGER_NOATOMIC); + if (k.k->type == KEY_TYPE_stripe) + ret = __ec_stripe_mem_alloc(c, k.k->p.offset, GFP_KERNEL) ?: + bch2_mark_key(c, k, 0, 0, NULL, 0, + BTREE_TRIGGER_ALLOC_READ| + BTREE_TRIGGER_NOATOMIC); - bch2_btree_and_journal_iter_advance(&iter); - } + return ret; +} - ret = bch2_trans_exit(&trans) ?: ret; - if (ret) { +int bch2_stripes_read(struct bch_fs *c, struct journal_keys *journal_keys) +{ + int ret = bch2_btree_and_journal_walk(c, journal_keys, BTREE_ID_EC, + NULL, bch2_stripes_read_fn); + if (ret) bch_err(c, "error reading stripes: %i", ret); - return ret; - } - return 0; + return ret; } int bch2_ec_mem_alloc(struct bch_fs *c, bool gc) @@ -1343,11 +1333,6 @@ int bch2_ec_mem_alloc(struct bch_fs *c, bool gc) return 0; } -int bch2_fs_ec_start(struct bch_fs *c) -{ - return bch2_ec_mem_alloc(c, false); -} - void bch2_fs_ec_exit(struct bch_fs *c) { struct ec_stripe_head *h; diff --git a/fs/bcachefs/ec.h b/fs/bcachefs/ec.h index cf67abd48490..4dfaac034886 100644 --- a/fs/bcachefs/ec.h +++ b/fs/bcachefs/ec.h @@ -157,8 +157,6 @@ int bch2_stripes_write(struct bch_fs *, unsigned, bool *); int bch2_ec_mem_alloc(struct bch_fs *, bool); -int bch2_fs_ec_start(struct bch_fs *); - void bch2_fs_ec_exit(struct bch_fs *); int bch2_fs_ec_init(struct bch_fs *); diff --git a/fs/bcachefs/error.c b/fs/bcachefs/error.c index 5a5cfee623e2..cd46706fb6f5 100644 --- a/fs/bcachefs/error.c +++ b/fs/bcachefs/error.c @@ -37,7 +37,7 @@ void bch2_io_error_work(struct work_struct *work) struct bch_fs *c = ca->fs; bool dev; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_RO, BCH_FORCE_IF_DEGRADED); if (dev @@ -47,7 +47,7 @@ void bch2_io_error_work(struct work_struct *work) bch_err(ca, "too many IO errors, setting %s RO", dev ? "device" : "filesystem"); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); } void bch2_io_error(struct bch_dev *ca) @@ -85,7 +85,7 @@ enum fsck_err_ret bch2_fsck_err(struct bch_fs *c, unsigned flags, if (s->fmt == fmt) goto found; - s = kzalloc(sizeof(*s), GFP_KERNEL); + s = kzalloc(sizeof(*s), GFP_NOFS); if (!s) { if (!c->fsck_alloc_err) bch_err(c, "kmalloc err, cannot ratelimit fsck errs"); diff --git a/fs/bcachefs/error.h b/fs/bcachefs/error.h index de319794ccd1..94b53312fbbd 100644 --- a/fs/bcachefs/error.h +++ b/fs/bcachefs/error.h @@ -102,6 +102,7 @@ struct fsck_err_state { #define FSCK_CAN_IGNORE (1 << 1) #define FSCK_NEED_FSCK (1 << 2) +__printf(3, 4) __cold enum fsck_err_ret bch2_fsck_err(struct bch_fs *, unsigned, const char *, ...); void bch2_flush_fsck_errs(struct bch_fs *); diff --git a/fs/bcachefs/extent_update.c b/fs/bcachefs/extent_update.c index d0af1bc17018..fd011df3cb99 100644 --- a/fs/bcachefs/extent_update.c +++ b/fs/bcachefs/extent_update.c @@ -76,7 +76,8 @@ static int count_iters_for_insert(struct btree_trans *trans, if (*nr_iters >= max_iters) { struct bpos pos = bkey_start_pos(k.k); - pos.offset += r_k.k->p.offset - idx; + pos.offset += min_t(u64, k.k->size, + r_k.k->p.offset - idx); *end = bpos_min(*end, pos); ret = 1; diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c index d1a4ab04fbbf..251d4af773a5 100644 --- a/fs/bcachefs/extents.c +++ b/fs/bcachefs/extents.c @@ -220,7 +220,7 @@ void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c, { struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k); - pr_buf(out, "seq %llu sectors %u written %u min_key ", + pr_buf(out, "seq %llx sectors %u written %u min_key ", le64_to_cpu(bp.v->seq), le16_to_cpu(bp.v->sectors), le16_to_cpu(bp.v->sectors_written)); diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c index ab2d808eea43..162aa55d5b20 100644 --- a/fs/bcachefs/fs-io.c +++ b/fs/bcachefs/fs-io.c @@ -54,6 +54,7 @@ struct dio_write { sync:1, free_iov:1; struct quota_res quota_res; + u64 written; struct iov_iter iter; struct iovec inline_vecs[2]; @@ -602,7 +603,7 @@ int bch2_migrate_page(struct address_space *mapping, struct page *newpage, EBUG_ON(!PageLocked(page)); EBUG_ON(!PageLocked(newpage)); - ret = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0); + ret = migrate_page_move_mapping(mapping, newpage, page, 0); if (ret != MIGRATEPAGE_SUCCESS) return ret; @@ -627,10 +628,10 @@ int bch2_migrate_page(struct address_space *mapping, struct page *newpage, static void bch2_readpages_end_io(struct bio *bio) { + struct bvec_iter_all iter; struct bio_vec *bv; - unsigned i; - bio_for_each_segment_all(bv, bio, i) { + bio_for_each_segment_all(bv, bio, iter) { struct page *page = bv->bv_page; if (!bio->bi_status) { @@ -782,11 +783,8 @@ static void readpage_bio_extend(struct readpages_iter *iter, if (!get_more) break; - rcu_read_lock(); - page = radix_tree_lookup(&iter->mapping->i_pages, page_offset); - rcu_read_unlock(); - - if (page && !radix_tree_exceptional_entry(page)) + page = xa_load(&iter->mapping->i_pages, page_offset); + if (page && !xa_is_value(page)) break; page = __page_cache_alloc(readahead_gfp_mask(iter->mapping)); @@ -848,7 +846,7 @@ retry: sectors = k.k->size - offset_into_extent; ret = bch2_read_indirect_extent(trans, - &offset_into_extent, sk.k); + &offset_into_extent, &sk); if (ret) break; @@ -1037,32 +1035,33 @@ static void bch2_writepage_io_done(struct closure *cl) struct bch_writepage_io, cl); struct bch_fs *c = io->op.c; struct bio *bio = &io->op.wbio.bio; + struct bvec_iter_all iter; struct bio_vec *bvec; - unsigned i, j; + unsigned i; if (io->op.error) { - bio_for_each_segment_all(bvec, bio, i) { + bio_for_each_segment_all(bvec, bio, iter) { struct bch_page_state *s; SetPageError(bvec->bv_page); - mapping_set_error(io->inode->v.i_mapping, -EIO); + mapping_set_error(bvec->bv_page->mapping, -EIO); s = __bch2_page_state(bvec->bv_page); spin_lock(&s->lock); - for (j = 0; j < PAGE_SECTORS; j++) - s->s[j].nr_replicas = 0; + for (i = 0; i < PAGE_SECTORS; i++) + s->s[i].nr_replicas = 0; spin_unlock(&s->lock); } } if (io->op.flags & BCH_WRITE_WROTE_DATA_INLINE) { - bio_for_each_segment_all(bvec, bio, i) { + bio_for_each_segment_all(bvec, bio, iter) { struct bch_page_state *s; s = __bch2_page_state(bvec->bv_page); spin_lock(&s->lock); - for (j = 0; j < PAGE_SECTORS; j++) - s->s[j].nr_replicas = 0; + for (i = 0; i < PAGE_SECTORS; i++) + s->s[i].nr_replicas = 0; spin_unlock(&s->lock); } } @@ -1086,7 +1085,7 @@ static void bch2_writepage_io_done(struct closure *cl) */ i_sectors_acct(c, io->inode, NULL, io->op.i_sectors_delta); - bio_for_each_segment_all(bvec, bio, i) { + bio_for_each_segment_all(bvec, bio, iter) { struct bch_page_state *s = __bch2_page_state(bvec->bv_page); if (atomic_dec_and_test(&s->write_count)) @@ -1240,7 +1239,7 @@ do_io: if (w->io && (w->io->op.res.nr_replicas != nr_replicas_this_write || - bio_full(&w->io->op.wbio.bio) || + bio_full(&w->io->op.wbio.bio, PAGE_SIZE) || w->io->op.wbio.bio.bi_iter.bi_size + (sectors << 9) >= (BIO_MAX_PAGES * PAGE_SIZE) || bio_end_sector(&w->io->op.wbio.bio) != sector)) @@ -1798,17 +1797,19 @@ ssize_t bch2_read_iter(struct kiocb *iocb, struct iov_iter *iter) /* O_DIRECT writes */ +static void bch2_dio_write_loop_async(struct bch_write_op *); + static long bch2_dio_write_loop(struct dio_write *dio) { bool kthread = (current->flags & PF_KTHREAD) != 0; - struct bch_fs *c = dio->op.c; struct kiocb *req = dio->req; struct address_space *mapping = req->ki_filp->f_mapping; struct bch_inode_info *inode = file_bch_inode(req->ki_filp); + struct bch_fs *c = inode->v.i_sb->s_fs_info; struct bio *bio = &dio->op.wbio.bio; + struct bvec_iter_all iter; struct bio_vec *bv; - unsigned i, unaligned; - u64 new_i_size; + unsigned unaligned; bool sync = dio->sync; long ret; @@ -1849,14 +1850,30 @@ static long bch2_dio_write_loop(struct dio_write *dio) * bio_iov_iter_get_pages was only able to get < * blocksize worth of pages: */ - bio_for_each_segment_all(bv, bio, i) + bio_for_each_segment_all(bv, bio, iter) put_page(bv->bv_page); ret = -EFAULT; goto err; } - dio->op.pos = POS(inode->v.i_ino, - (req->ki_pos >> 9) + dio->op.written); + bch2_write_op_init(&dio->op, c, io_opts(c, &inode->ei_inode)); + dio->op.end_io = bch2_dio_write_loop_async; + dio->op.target = dio->op.opts.foreground_target; + op_journal_seq_set(&dio->op, &inode->ei_journal_seq); + dio->op.write_point = writepoint_hashed((unsigned long) current); + dio->op.nr_replicas = dio->op.opts.data_replicas; + dio->op.pos = POS(inode->v.i_ino, (u64) req->ki_pos >> 9); + + if ((req->ki_flags & IOCB_DSYNC) && + !c->opts.journal_flush_disabled) + dio->op.flags |= BCH_WRITE_FLUSH; + + ret = bch2_disk_reservation_get(c, &dio->op.res, bio_sectors(bio), + dio->op.opts.data_replicas, 0); + if (unlikely(ret) && + !bch2_check_range_allocated(c, dio->op.pos, + bio_sectors(bio), dio->op.opts.data_replicas)) + goto err; task_io_account_write(bio->bi_iter.bi_size); @@ -1888,16 +1905,15 @@ do_io: loop: i_sectors_acct(c, inode, &dio->quota_res, dio->op.i_sectors_delta); - dio->op.i_sectors_delta = 0; - - new_i_size = req->ki_pos + ((u64) dio->op.written << 9); + req->ki_pos += (u64) dio->op.written << 9; + dio->written += dio->op.written; spin_lock(&inode->v.i_lock); - if (new_i_size > inode->v.i_size) - i_size_write(&inode->v, new_i_size); + if (req->ki_pos > inode->v.i_size) + i_size_write(&inode->v, req->ki_pos); spin_unlock(&inode->v.i_lock); - bio_for_each_segment_all(bv, bio, i) + bio_for_each_segment_all(bv, bio, iter) put_page(bv->bv_page); if (!dio->iter.count || dio->op.error) break; @@ -1906,10 +1922,9 @@ loop: reinit_completion(&dio->done); } - ret = dio->op.error ?: ((long) dio->op.written << 9); + ret = dio->op.error ?: ((long) dio->written << 9); err: bch2_pagecache_block_put(&inode->ei_pagecache_lock); - bch2_disk_reservation_put(c, &dio->op.res); bch2_quota_reservation_put(c, inode, &dio->quota_res); if (dio->free_iov) @@ -1944,7 +1959,6 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) struct address_space *mapping = file->f_mapping; struct bch_inode_info *inode = file_bch_inode(file); struct bch_fs *c = inode->v.i_sb->s_fs_info; - struct bch_io_opts opts = io_opts(c, &inode->ei_inode); struct dio_write *dio; struct bio *bio; bool locked = true, extending; @@ -1992,35 +2006,14 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) dio->sync = is_sync_kiocb(req) || extending; dio->free_iov = false; dio->quota_res.sectors = 0; + dio->written = 0; dio->iter = *iter; - bch2_write_op_init(&dio->op, c, opts); - dio->op.end_io = bch2_dio_write_loop_async; - dio->op.target = opts.foreground_target; - op_journal_seq_set(&dio->op, &inode->ei_journal_seq); - dio->op.write_point = writepoint_hashed((unsigned long) current); - dio->op.flags |= BCH_WRITE_NOPUT_RESERVATION; - - if ((req->ki_flags & IOCB_DSYNC) && - !c->opts.journal_flush_disabled) - dio->op.flags |= BCH_WRITE_FLUSH; - ret = bch2_quota_reservation_add(c, inode, &dio->quota_res, iter->count >> 9, true); if (unlikely(ret)) goto err_put_bio; - dio->op.nr_replicas = dio->op.opts.data_replicas; - - ret = bch2_disk_reservation_get(c, &dio->op.res, iter->count >> 9, - dio->op.opts.data_replicas, 0); - if (unlikely(ret) && - !bch2_check_range_allocated(c, POS(inode->v.i_ino, - req->ki_pos >> 9), - iter->count >> 9, - dio->op.opts.data_replicas)) - goto err_put_bio; - ret = write_invalidate_inode_pages_range(mapping, req->ki_pos, req->ki_pos + iter->count - 1); @@ -2031,12 +2024,9 @@ ssize_t bch2_direct_write(struct kiocb *req, struct iov_iter *iter) err: if (locked) inode_unlock(&inode->v); - if (ret > 0) - req->ki_pos += ret; return ret; err_put_bio: bch2_pagecache_block_put(&inode->ei_pagecache_lock); - bch2_disk_reservation_put(c, &dio->op.res); bch2_quota_reservation_put(c, inode, &dio->quota_res); bio_put(bio); inode_dio_end(&inode->v); @@ -2834,235 +2824,6 @@ static void mark_range_unallocated(struct bch_inode_info *inode, } while (index <= end_index); } -static int generic_access_check_limits(struct file *file, loff_t pos, - loff_t *count) -{ - struct inode *inode = file->f_mapping->host; - loff_t max_size = inode->i_sb->s_maxbytes; - - if (!(file->f_flags & O_LARGEFILE)) - max_size = MAX_NON_LFS; - - if (unlikely(pos >= max_size)) - return -EFBIG; - *count = min(*count, max_size - pos); - return 0; -} - -static int generic_write_check_limits(struct file *file, loff_t pos, - loff_t *count) -{ - loff_t limit = rlimit(RLIMIT_FSIZE); - - if (limit != RLIM_INFINITY) { - if (pos >= limit) { - send_sig(SIGXFSZ, current, 0); - return -EFBIG; - } - *count = min(*count, limit - pos); - } - - return generic_access_check_limits(file, pos, count); -} - -static int generic_remap_checks(struct file *file_in, loff_t pos_in, - struct file *file_out, loff_t pos_out, - loff_t *req_count, unsigned int remap_flags) -{ - struct inode *inode_in = file_in->f_mapping->host; - struct inode *inode_out = file_out->f_mapping->host; - uint64_t count = *req_count; - uint64_t bcount; - loff_t size_in, size_out; - loff_t bs = inode_out->i_sb->s_blocksize; - int ret; - - /* The start of both ranges must be aligned to an fs block. */ - if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs)) - return -EINVAL; - - /* Ensure offsets don't wrap. */ - if (pos_in + count < pos_in || pos_out + count < pos_out) - return -EINVAL; - - size_in = i_size_read(inode_in); - size_out = i_size_read(inode_out); - - /* Dedupe requires both ranges to be within EOF. */ - if ((remap_flags & REMAP_FILE_DEDUP) && - (pos_in >= size_in || pos_in + count > size_in || - pos_out >= size_out || pos_out + count > size_out)) - return -EINVAL; - - /* Ensure the infile range is within the infile. */ - if (pos_in >= size_in) - return -EINVAL; - count = min(count, size_in - (uint64_t)pos_in); - - ret = generic_access_check_limits(file_in, pos_in, &count); - if (ret) - return ret; - - ret = generic_write_check_limits(file_out, pos_out, &count); - if (ret) - return ret; - - /* - * If the user wanted us to link to the infile's EOF, round up to the - * next block boundary for this check. - * - * Otherwise, make sure the count is also block-aligned, having - * already confirmed the starting offsets' block alignment. - */ - if (pos_in + count == size_in) { - bcount = ALIGN(size_in, bs) - pos_in; - } else { - if (!IS_ALIGNED(count, bs)) - count = ALIGN_DOWN(count, bs); - bcount = count; - } - - /* Don't allow overlapped cloning within the same file. */ - if (inode_in == inode_out && - pos_out + bcount > pos_in && - pos_out < pos_in + bcount) - return -EINVAL; - - /* - * We shortened the request but the caller can't deal with that, so - * bounce the request back to userspace. - */ - if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) - return -EINVAL; - - *req_count = count; - return 0; -} - -static int generic_remap_check_len(struct inode *inode_in, - struct inode *inode_out, - loff_t pos_out, - loff_t *len, - unsigned int remap_flags) -{ - u64 blkmask = i_blocksize(inode_in) - 1; - loff_t new_len = *len; - - if ((*len & blkmask) == 0) - return 0; - - if ((remap_flags & REMAP_FILE_DEDUP) || - pos_out + *len < i_size_read(inode_out)) - new_len &= ~blkmask; - - if (new_len == *len) - return 0; - - if (remap_flags & REMAP_FILE_CAN_SHORTEN) { - *len = new_len; - return 0; - } - - return (remap_flags & REMAP_FILE_DEDUP) ? -EBADE : -EINVAL; -} - -static int generic_remap_file_range_prep(struct file *file_in, loff_t pos_in, - struct file *file_out, loff_t pos_out, - loff_t *len, unsigned int remap_flags) -{ - struct inode *inode_in = file_inode(file_in); - struct inode *inode_out = file_inode(file_out); - bool same_inode = (inode_in == inode_out); - int ret; - - /* Don't touch certain kinds of inodes */ - if (IS_IMMUTABLE(inode_out)) - return -EPERM; - - if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) - return -ETXTBSY; - - /* Don't reflink dirs, pipes, sockets... */ - if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) - return -EISDIR; - if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) - return -EINVAL; - - /* Zero length dedupe exits immediately; reflink goes to EOF. */ - if (*len == 0) { - loff_t isize = i_size_read(inode_in); - - if ((remap_flags & REMAP_FILE_DEDUP) || pos_in == isize) - return 0; - if (pos_in > isize) - return -EINVAL; - *len = isize - pos_in; - if (*len == 0) - return 0; - } - - /* Check that we don't violate system file offset limits. */ - ret = generic_remap_checks(file_in, pos_in, file_out, pos_out, len, - remap_flags); - if (ret) - return ret; - - /* Wait for the completion of any pending IOs on both files */ - inode_dio_wait(inode_in); - if (!same_inode) - inode_dio_wait(inode_out); - - ret = filemap_write_and_wait_range(inode_in->i_mapping, - pos_in, pos_in + *len - 1); - if (ret) - return ret; - - ret = filemap_write_and_wait_range(inode_out->i_mapping, - pos_out, pos_out + *len - 1); - if (ret) - return ret; - - /* - * Check that the extents are the same. - */ - if (remap_flags & REMAP_FILE_DEDUP) { - bool is_same = false; - - ret = vfs_dedupe_file_range_compare(inode_in, pos_in, - inode_out, pos_out, *len, &is_same); - if (ret) - return ret; - if (!is_same) - return -EBADE; - } - - ret = generic_remap_check_len(inode_in, inode_out, pos_out, len, - remap_flags); - if (ret) - return ret; - - /* If can't alter the file contents, we're done. */ - if (!(remap_flags & REMAP_FILE_DEDUP)) { - /* Update the timestamps, since we can alter file contents. */ - if (!(file_out->f_mode & FMODE_NOCMTIME)) { - ret = file_update_time(file_out); - if (ret) - return ret; - } - - /* - * Clear the security bits if the process is not being run by - * root. This keeps people from modifying setuid and setgid - * binaries. - */ - ret = file_remove_privs(file_out); - if (ret) - return ret; - } - - return 0; -} - loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, struct file *file_dst, loff_t pos_dst, loff_t len, unsigned remap_flags) @@ -3074,6 +2835,9 @@ loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src, u64 aligned_len; loff_t ret = 0; + if (!c->opts.reflink) + return -EOPNOTSUPP; + if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY)) return -EINVAL; @@ -3255,7 +3019,7 @@ static loff_t page_hole_offset(struct address_space *mapping, loff_t offset) loff_t ret = -1; page = find_lock_entry(mapping, index); - if (!page || radix_tree_exception(page)) + if (!page || xa_is_value(page)) return offset; pg_offset = __page_hole_offset(page, offset & (PAGE_SIZE - 1)); diff --git a/fs/bcachefs/fs-io.h b/fs/bcachefs/fs-io.h index 1b593ea707d5..7063556d289b 100644 --- a/fs/bcachefs/fs-io.h +++ b/fs/bcachefs/fs-io.h @@ -35,10 +35,6 @@ int bch2_fsync(struct file *, loff_t, loff_t, int); int bch2_truncate(struct bch_inode_info *, struct iattr *); long bch2_fallocate_dispatch(struct file *, int, loff_t, loff_t); -#define REMAP_FILE_ADVISORY (0) -#define REMAP_FILE_DEDUP (1 << 0) -#define REMAP_FILE_CAN_SHORTEN (1 << 1) - loff_t bch2_remap_file_range(struct file *, loff_t, struct file *, loff_t, loff_t, unsigned); diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c index b1445bdc3e9d..a47923d67f7a 100644 --- a/fs/bcachefs/fs.c +++ b/fs/bcachefs/fs.c @@ -889,7 +889,7 @@ retry: sectors = k.k->size - offset_into_extent; ret = bch2_read_indirect_extent(&trans, - &offset_into_extent, cur.k); + &offset_into_extent, &cur); if (ret) break; @@ -966,15 +966,6 @@ static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx) return bch2_readdir(c, inode->v.i_ino, ctx); } -static int bch2_clone_file_range(struct file *file_src, loff_t pos_src, - struct file *file_dst, loff_t pos_dst, - u64 len) -{ - return bch2_remap_file_range(file_src, pos_src, - file_dst, pos_dst, - len, 0); -} - static const struct file_operations bch_file_operations = { .llseek = bch2_llseek, .read_iter = bch2_read_iter, @@ -992,7 +983,7 @@ static const struct file_operations bch_file_operations = { #ifdef CONFIG_COMPAT .compat_ioctl = bch2_compat_fs_ioctl, #endif - .clone_file_range = bch2_clone_file_range, + .remap_file_range = bch2_remap_file_range, }; static const struct inode_operations bch_file_inode_operations = { @@ -1324,16 +1315,16 @@ static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * cons if (IS_ERR(c)) return c; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); if (!test_bit(BCH_FS_STARTED, &c->flags)) { - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); closure_put(&c->cl); pr_err("err mounting %s: incomplete filesystem", dev_name); return ERR_PTR(-EINVAL); } - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); set_bit(BCH_FS_BDEV_MOUNTED, &c->flags); return c; @@ -1382,7 +1373,7 @@ static int bch2_remount(struct super_block *sb, int *flags, char *data) return ret; if (opts.read_only != c->opts.read_only) { - mutex_lock(&c->state_lock); + down_write(&c->state_lock); if (opts.read_only) { bch2_fs_read_only(c); @@ -1392,7 +1383,7 @@ static int bch2_remount(struct super_block *sb, int *flags, char *data) ret = bch2_fs_read_write(c); if (ret) { bch_err(c, "error going rw: %i", ret); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return -EINVAL; } @@ -1401,7 +1392,7 @@ static int bch2_remount(struct super_block *sb, int *flags, char *data) c->opts.read_only = opts.read_only; - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); } if (opts.errors >= 0) @@ -1523,7 +1514,7 @@ static struct dentry *bch2_mount(struct file_system_type *fs_type, sb->s_bdi->congested_fn = bch2_congested; sb->s_bdi->congested_data = c; - sb->s_bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE; + sb->s_bdi->ra_pages = VM_READAHEAD_PAGES; for_each_online_member(ca, c, i) { struct block_device *bdev = ca->disk_sb.bdev; diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c index 3ab621c62c43..c6ca5968a2e0 100644 --- a/fs/bcachefs/fsck.c +++ b/fs/bcachefs/fsck.c @@ -1169,7 +1169,7 @@ static int check_inode_nlink(struct bch_fs *c, } if (!S_ISDIR(u->bi_mode) && link->dir_count) { - need_fsck_err(c, "non directory with subdirectories", + need_fsck_err(c, "non directory with subdirectories (inum %llu)", u->bi_inum); return 0; } diff --git a/fs/bcachefs/io.c b/fs/bcachefs/io.c index b2684803c0d9..8c441050b02a 100644 --- a/fs/bcachefs/io.c +++ b/fs/bcachefs/io.c @@ -124,10 +124,10 @@ void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw) void bch2_bio_free_pages_pool(struct bch_fs *c, struct bio *bio) { + struct bvec_iter_all iter; struct bio_vec *bv; - unsigned i; - bio_for_each_segment_all(bv, bio, i) + bio_for_each_segment_all(bv, bio, iter) if (bv->bv_page != ZERO_PAGE(0)) mempool_free(bv->bv_page, &c->bio_bounce_pages); bio->bi_vcnt = 0; @@ -493,13 +493,15 @@ static void bch2_write_done(struct closure *cl) if (!op->error && (op->flags & BCH_WRITE_FLUSH)) op->error = bch2_journal_error(&c->journal); - if (!(op->flags & BCH_WRITE_NOPUT_RESERVATION)) - bch2_disk_reservation_put(c, &op->res); + bch2_disk_reservation_put(c, &op->res); percpu_ref_put(&c->writes); bch2_keylist_free(&op->insert_keys, op->inline_keys); bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time); + if (!(op->flags & BCH_WRITE_FROM_INTERNAL)) + up(&c->io_in_flight); + if (op->end_io) { EBUG_ON(cl->parent); closure_debug_destroy(cl); @@ -1258,6 +1260,12 @@ void bch2_write(struct closure *cl) goto err; } + /* + * Can't ratelimit copygc - we'd deadlock: + */ + if (!(op->flags & BCH_WRITE_FROM_INTERNAL)) + down(&c->io_in_flight); + bch2_increment_clock(c, bio_sectors(bio), WRITE); data_len = min_t(u64, bio->bi_iter.bi_size, @@ -1272,8 +1280,7 @@ void bch2_write(struct closure *cl) continue_at_nobarrier(cl, __bch2_write, NULL); return; err: - if (!(op->flags & BCH_WRITE_NOPUT_RESERVATION)) - bch2_disk_reservation_put(c, &op->res); + bch2_disk_reservation_put(c, &op->res); if (op->end_io) { EBUG_ON(cl->parent); @@ -1641,7 +1648,7 @@ retry: sectors = k.k->size - offset_into_extent; ret = bch2_read_indirect_extent(&trans, - &offset_into_extent, sk.k); + &offset_into_extent, &sk); if (ret) break; @@ -1943,14 +1950,14 @@ static void bch2_read_endio(struct bio *bio) int __bch2_read_indirect_extent(struct btree_trans *trans, unsigned *offset_into_extent, - struct bkey_i *orig_k) + struct bkey_on_stack *orig_k) { struct btree_iter *iter; struct bkey_s_c k; u64 reflink_offset; int ret; - reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k)->v.idx) + + reflink_offset = le64_to_cpu(bkey_i_to_reflink_p(orig_k->k)->v.idx) + *offset_into_extent; iter = bch2_trans_get_iter(trans, BTREE_ID_REFLINK, @@ -1973,7 +1980,7 @@ int __bch2_read_indirect_extent(struct btree_trans *trans, } *offset_into_extent = iter->pos.offset - bkey_start_offset(k.k); - bkey_reassemble(orig_k, k); + bkey_on_stack_reassemble(orig_k, trans->c, k); err: bch2_trans_iter_put(trans, iter); return ret; @@ -2273,7 +2280,7 @@ retry: k = bkey_i_to_s_c(sk.k); ret = bch2_read_indirect_extent(&trans, - &offset_into_extent, sk.k); + &offset_into_extent, &sk); if (ret) goto err; diff --git a/fs/bcachefs/io.h b/fs/bcachefs/io.h index c4c847306345..0ad293bd6295 100644 --- a/fs/bcachefs/io.h +++ b/fs/bcachefs/io.h @@ -3,6 +3,7 @@ #define _BCACHEFS_IO_H #include "checksum.h" +#include "bkey_on_stack.h" #include "io_types.h" #define to_wbio(_bio) \ @@ -29,14 +30,13 @@ enum bch_write_flags { BCH_WRITE_PAGES_STABLE = (1 << 4), BCH_WRITE_PAGES_OWNED = (1 << 5), BCH_WRITE_ONLY_SPECIFIED_DEVS = (1 << 6), - BCH_WRITE_NOPUT_RESERVATION = (1 << 7), - BCH_WRITE_WROTE_DATA_INLINE = (1 << 8), - BCH_WRITE_FROM_INTERNAL = (1 << 9), + BCH_WRITE_WROTE_DATA_INLINE = (1 << 7), + BCH_WRITE_FROM_INTERNAL = (1 << 8), /* Internal: */ - BCH_WRITE_JOURNAL_SEQ_PTR = (1 << 10), - BCH_WRITE_SKIP_CLOSURE_PUT = (1 << 11), - BCH_WRITE_DONE = (1 << 12), + BCH_WRITE_JOURNAL_SEQ_PTR = (1 << 9), + BCH_WRITE_SKIP_CLOSURE_PUT = (1 << 10), + BCH_WRITE_DONE = (1 << 11), }; static inline u64 *op_journal_seq(struct bch_write_op *op) @@ -110,13 +110,13 @@ struct cache_promote_op; struct extent_ptr_decoded; int __bch2_read_indirect_extent(struct btree_trans *, unsigned *, - struct bkey_i *); + struct bkey_on_stack *); static inline int bch2_read_indirect_extent(struct btree_trans *trans, unsigned *offset_into_extent, - struct bkey_i *k) + struct bkey_on_stack *k) { - return k->k.type == KEY_TYPE_reflink_p + return k->k->k.type == KEY_TYPE_reflink_p ? __bch2_read_indirect_extent(trans, offset_into_extent, k) : 0; } diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c index 0a4538b3dc60..b4f7b61ba9ac 100644 --- a/fs/bcachefs/journal.c +++ b/fs/bcachefs/journal.c @@ -428,9 +428,10 @@ int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res, static bool journal_preres_available(struct journal *j, struct journal_preres *res, - unsigned new_u64s) + unsigned new_u64s, + unsigned flags) { - bool ret = bch2_journal_preres_get_fast(j, res, new_u64s); + bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags); if (!ret) bch2_journal_reclaim_work(&j->reclaim_work.work); @@ -440,13 +441,14 @@ static bool journal_preres_available(struct journal *j, int __bch2_journal_preres_get(struct journal *j, struct journal_preres *res, - unsigned new_u64s) + unsigned new_u64s, + unsigned flags) { int ret; closure_wait_event(&j->preres_wait, (ret = bch2_journal_error(j)) || - journal_preres_available(j, res, new_u64s)); + journal_preres_available(j, res, new_u64s, flags)); return ret; } @@ -959,15 +961,12 @@ void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca) void bch2_fs_journal_stop(struct journal *j) { - struct bch_fs *c = container_of(j, struct bch_fs, journal); - bch2_journal_flush_all_pins(j); wait_event(j->wait, journal_entry_close(j)); /* do we need to write another journal entry? */ - if (test_bit(JOURNAL_NOT_EMPTY, &j->flags) || - c->btree_roots_dirty) + if (test_bit(JOURNAL_NOT_EMPTY, &j->flags)) bch2_journal_meta(j); journal_quiesce(j); @@ -988,9 +987,8 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq, u64 last_seq = cur_seq, nr, seq; if (!list_empty(journal_entries)) - last_seq = le64_to_cpu(list_first_entry(journal_entries, - struct journal_replay, - list)->j.seq); + last_seq = le64_to_cpu(list_last_entry(journal_entries, + struct journal_replay, list)->j.last_seq); nr = cur_seq - last_seq; @@ -1019,8 +1017,10 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq, list_for_each_entry(i, journal_entries, list) { seq = le64_to_cpu(i->j.seq); + BUG_ON(seq >= cur_seq); - BUG_ON(seq < last_seq || seq >= cur_seq); + if (seq < last_seq) + continue; journal_seq_pin(j, seq)->devs = i->devs; } @@ -1238,14 +1238,14 @@ ssize_t bch2_journal_print_pins(struct journal *j, char *buf) i, atomic_read(&pin_list->count)); list_for_each_entry(pin, &pin_list->list, list) - pr_buf(&out, "\t%p %pf\n", + pr_buf(&out, "\t%px %ps\n", pin, pin->flush); if (!list_empty(&pin_list->flushed)) pr_buf(&out, "flushed:\n"); list_for_each_entry(pin, &pin_list->flushed, list) - pr_buf(&out, "\t%p %pf\n", + pr_buf(&out, "\t%px %ps\n", pin, pin->flush); } spin_unlock(&j->lock); diff --git a/fs/bcachefs/journal.h b/fs/bcachefs/journal.h index 1ba8b62b9f2b..30de6d96188e 100644 --- a/fs/bcachefs/journal.h +++ b/fs/bcachefs/journal.h @@ -199,27 +199,39 @@ bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s) return entry; } +static inline struct jset_entry * +journal_res_entry(struct journal *j, struct journal_res *res) +{ + return vstruct_idx(j->buf[res->idx].data, res->offset); +} + +static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type, + enum btree_id id, unsigned level, + const void *data, unsigned u64s) +{ + memset(entry, 0, sizeof(*entry)); + entry->u64s = cpu_to_le16(u64s); + entry->type = type; + entry->btree_id = id; + entry->level = level; + memcpy_u64s_small(entry->_data, data, u64s); + + return jset_u64s(u64s); +} + static inline void bch2_journal_add_entry(struct journal *j, struct journal_res *res, unsigned type, enum btree_id id, unsigned level, const void *data, unsigned u64s) { - struct journal_buf *buf = &j->buf[res->idx]; - struct jset_entry *entry = vstruct_idx(buf->data, res->offset); - unsigned actual = jset_u64s(u64s); + unsigned actual = journal_entry_set(journal_res_entry(j, res), + type, id, level, data, u64s); EBUG_ON(!res->ref); EBUG_ON(actual > res->u64s); res->offset += actual; res->u64s -= actual; - - memset(entry, 0, sizeof(*entry)); - entry->u64s = cpu_to_le16(u64s); - entry->type = type; - entry->btree_id = id; - entry->level = level; - memcpy_u64s(entry->_data, data, u64s); } static inline void bch2_journal_add_keys(struct journal *j, struct journal_res *res, @@ -269,7 +281,7 @@ static inline void bch2_journal_res_put(struct journal *j, if (!res->ref) return; - lock_release(&j->res_map, 0, _THIS_IP_); + lock_release(&j->res_map, _THIS_IP_); while (res->u64s) bch2_journal_add_entry(j, res, @@ -287,6 +299,7 @@ int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *, #define JOURNAL_RES_GET_NONBLOCK (1 << 0) #define JOURNAL_RES_GET_CHECK (1 << 1) #define JOURNAL_RES_GET_RESERVED (1 << 2) +#define JOURNAL_RES_GET_RECLAIM (1 << 3) static inline int journal_res_get_fast(struct journal *j, struct journal_res *res, @@ -394,11 +407,12 @@ static inline void bch2_journal_preres_put(struct journal *j, } int __bch2_journal_preres_get(struct journal *, - struct journal_preres *, unsigned); + struct journal_preres *, unsigned, unsigned); static inline int bch2_journal_preres_get_fast(struct journal *j, struct journal_preres *res, - unsigned new_u64s) + unsigned new_u64s, + unsigned flags) { int d = new_u64s - res->u64s; union journal_preres_state old, new; @@ -409,7 +423,15 @@ static inline int bch2_journal_preres_get_fast(struct journal *j, new.reserved += d; - if (new.reserved > new.remaining) + /* + * If we're being called from the journal reclaim path, we have + * to unconditionally give out the pre-reservation, there's + * nothing else sensible we can do - otherwise we'd recurse back + * into the reclaim path and deadlock: + */ + + if (!(flags & JOURNAL_RES_GET_RECLAIM) && + new.reserved > new.remaining) return 0; } while ((v = atomic64_cmpxchg(&j->prereserved.counter, old.v, new.v)) != old.v); @@ -426,13 +448,13 @@ static inline int bch2_journal_preres_get(struct journal *j, if (new_u64s <= res->u64s) return 0; - if (bch2_journal_preres_get_fast(j, res, new_u64s)) + if (bch2_journal_preres_get_fast(j, res, new_u64s, flags)) return 0; if (flags & JOURNAL_RES_GET_NONBLOCK) return -EAGAIN; - return __bch2_journal_preres_get(j, res, new_u64s); + return __bch2_journal_preres_get(j, res, new_u64s, flags); } /* journal_entry_res: */ diff --git a/fs/bcachefs/journal_io.c b/fs/bcachefs/journal_io.c index 39bb2154cce1..b7625285b3ad 100644 --- a/fs/bcachefs/journal_io.c +++ b/fs/bcachefs/journal_io.c @@ -2,6 +2,7 @@ #include "bcachefs.h" #include "alloc_foreground.h" #include "btree_io.h" +#include "btree_update_interior.h" #include "buckets.h" #include "checksum.h" #include "error.h" @@ -40,19 +41,21 @@ static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca, list)->j.last_seq : 0; - /* Is this entry older than the range we need? */ - if (le64_to_cpu(j->seq) < le64_to_cpu(last_seq)) { - ret = JOURNAL_ENTRY_ADD_OUT_OF_RANGE; - goto out; - } + if (!c->opts.read_entire_journal) { + /* Is this entry older than the range we need? */ + if (le64_to_cpu(j->seq) < le64_to_cpu(last_seq)) { + ret = JOURNAL_ENTRY_ADD_OUT_OF_RANGE; + goto out; + } - /* Drop entries we don't need anymore */ - list_for_each_entry_safe(i, pos, jlist->head, list) { - if (le64_to_cpu(i->j.seq) >= le64_to_cpu(j->last_seq)) - break; - list_del(&i->list); - kvpfree(i, offsetof(struct journal_replay, j) + - vstruct_bytes(&i->j)); + /* Drop entries we don't need anymore */ + list_for_each_entry_safe(i, pos, jlist->head, list) { + if (le64_to_cpu(i->j.seq) >= le64_to_cpu(j->last_seq)) + break; + list_del(&i->list); + kvpfree(i, offsetof(struct journal_replay, j) + + vstruct_bytes(&i->j)); + } } list_for_each_entry_reverse(i, jlist->head, list) { @@ -993,8 +996,23 @@ void bch2_journal_write(struct closure *cl) j->write_start_time = local_clock(); - start = vstruct_last(jset); - end = bch2_journal_super_entries_add_common(c, start, + /* + * New btree roots are set by journalling them; when the journal entry + * gets written we have to propagate them to c->btree_roots + * + * But, every journal entry we write has to contain all the btree roots + * (at least for now); so after we copy btree roots to c->btree_roots we + * have to get any missing btree roots and add them to this journal + * entry: + */ + + bch2_journal_entries_to_btree_roots(c, jset); + + start = end = vstruct_last(jset); + + end = bch2_btree_roots_to_journal_entries(c, jset->start, end); + + end = bch2_journal_super_entries_add_common(c, end, le64_to_cpu(jset->seq)); u64s = (u64 *) end - (u64 *) start; BUG_ON(u64s > j->entry_u64s_reserved); diff --git a/fs/bcachefs/journal_reclaim.c b/fs/bcachefs/journal_reclaim.c index d34434f62454..4811ab9f879e 100644 --- a/fs/bcachefs/journal_reclaim.c +++ b/fs/bcachefs/journal_reclaim.c @@ -28,18 +28,10 @@ unsigned bch2_journal_dev_buckets_available(struct journal *j, struct journal_device *ja, enum journal_space_from from) { - struct bch_fs *c = container_of(j, struct bch_fs, journal); unsigned available = (journal_space_from(ja, from) - ja->cur_idx - 1 + ja->nr) % ja->nr; /* - * Allocator startup needs some journal space before we can do journal - * replay: - */ - if (available && test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) - --available; - - /* * Don't use the last bucket unless writing the new last_seq * will make another bucket available: */ @@ -330,7 +322,7 @@ static void bch2_journal_pin_add_locked(struct journal *j, u64 seq, __journal_pin_drop(j, pin); - BUG_ON(!atomic_read(&pin_list->count)); + BUG_ON(!atomic_read(&pin_list->count) && seq == journal_last_seq(j)); atomic_inc(&pin_list->count); pin->seq = seq; @@ -354,6 +346,37 @@ void __bch2_journal_pin_add(struct journal *j, u64 seq, journal_wake(j); } +void bch2_journal_pin_update(struct journal *j, u64 seq, + struct journal_entry_pin *pin, + journal_pin_flush_fn flush_fn) +{ + if (journal_pin_active(pin) && pin->seq < seq) + return; + + spin_lock(&j->lock); + + if (pin->seq != seq) { + bch2_journal_pin_add_locked(j, seq, pin, flush_fn); + } else { + struct journal_entry_pin_list *pin_list = + journal_seq_pin(j, seq); + + /* + * If the pin is already pinning the right sequence number, it + * still might've already been flushed: + */ + list_move(&pin->list, &pin_list->list); + } + + spin_unlock(&j->lock); + + /* + * If the journal is currently full, we might want to call flush_fn + * immediately: + */ + journal_wake(j); +} + void bch2_journal_pin_copy(struct journal *j, struct journal_entry_pin *dst, struct journal_entry_pin *src, @@ -393,6 +416,9 @@ journal_get_next_pin(struct journal *j, u64 max_seq, u64 *seq) struct journal_entry_pin_list *pin_list; struct journal_entry_pin *ret = NULL; + if (!test_bit(JOURNAL_RECLAIM_STARTED, &j->flags)) + return NULL; + spin_lock(&j->lock); fifo_for_each_entry_ptr(pin_list, &j->pin, *seq) @@ -413,10 +439,12 @@ journal_get_next_pin(struct journal *j, u64 max_seq, u64 *seq) return ret; } -static void journal_flush_pins(struct journal *j, u64 seq_to_flush, +/* returns true if we did work */ +static bool journal_flush_pins(struct journal *j, u64 seq_to_flush, unsigned min_nr) { struct journal_entry_pin *pin; + bool ret = false; u64 seq; lockdep_assert_held(&j->reclaim_lock); @@ -431,7 +459,10 @@ static void journal_flush_pins(struct journal *j, u64 seq_to_flush, BUG_ON(j->flush_in_progress != pin); j->flush_in_progress = NULL; wake_up(&j->pin_flush_wait); + ret = true; } + + return ret; } /** @@ -523,7 +554,8 @@ void bch2_journal_reclaim_work(struct work_struct *work) mutex_unlock(&j->reclaim_lock); } -static int journal_flush_done(struct journal *j, u64 seq_to_flush) +static int journal_flush_done(struct journal *j, u64 seq_to_flush, + bool *did_work) { int ret; @@ -533,7 +565,7 @@ static int journal_flush_done(struct journal *j, u64 seq_to_flush) mutex_lock(&j->reclaim_lock); - journal_flush_pins(j, seq_to_flush, 0); + *did_work = journal_flush_pins(j, seq_to_flush, 0); spin_lock(&j->lock); /* @@ -551,12 +583,17 @@ static int journal_flush_done(struct journal *j, u64 seq_to_flush) return ret; } -void bch2_journal_flush_pins(struct journal *j, u64 seq_to_flush) +bool bch2_journal_flush_pins(struct journal *j, u64 seq_to_flush) { + bool did_work = false; + if (!test_bit(JOURNAL_STARTED, &j->flags)) - return; + return false; + + closure_wait_event(&j->async_wait, + journal_flush_done(j, seq_to_flush, &did_work)); - closure_wait_event(&j->async_wait, journal_flush_done(j, seq_to_flush)); + return did_work; } int bch2_journal_flush_device_pins(struct journal *j, int dev_idx) diff --git a/fs/bcachefs/journal_reclaim.h b/fs/bcachefs/journal_reclaim.h index 883a0a5680af..8128907a7623 100644 --- a/fs/bcachefs/journal_reclaim.h +++ b/fs/bcachefs/journal_reclaim.h @@ -38,10 +38,14 @@ static inline void bch2_journal_pin_add(struct journal *j, u64 seq, struct journal_entry_pin *pin, journal_pin_flush_fn flush_fn) { - if (unlikely(!journal_pin_active(pin))) + if (unlikely(!journal_pin_active(pin) || pin->seq > seq)) __bch2_journal_pin_add(j, seq, pin, flush_fn); } +void bch2_journal_pin_update(struct journal *, u64, + struct journal_entry_pin *, + journal_pin_flush_fn); + void bch2_journal_pin_copy(struct journal *, struct journal_entry_pin *, struct journal_entry_pin *, @@ -53,11 +57,11 @@ void bch2_journal_do_discards(struct journal *); void bch2_journal_reclaim(struct journal *); void bch2_journal_reclaim_work(struct work_struct *); -void bch2_journal_flush_pins(struct journal *, u64); +bool bch2_journal_flush_pins(struct journal *, u64); -static inline void bch2_journal_flush_all_pins(struct journal *j) +static inline bool bch2_journal_flush_all_pins(struct journal *j) { - bch2_journal_flush_pins(j, U64_MAX); + return bch2_journal_flush_pins(j, U64_MAX); } int bch2_journal_flush_device_pins(struct journal *, int); diff --git a/fs/bcachefs/journal_types.h b/fs/bcachefs/journal_types.h index 8eea12a03c06..154b51b891d3 100644 --- a/fs/bcachefs/journal_types.h +++ b/fs/bcachefs/journal_types.h @@ -125,6 +125,7 @@ union journal_preres_state { enum { JOURNAL_REPLAY_DONE, JOURNAL_STARTED, + JOURNAL_RECLAIM_STARTED, JOURNAL_NEED_WRITE, JOURNAL_NOT_EMPTY, JOURNAL_MAY_GET_UNRESERVED, diff --git a/fs/bcachefs/keylist.c b/fs/bcachefs/keylist.c index 5da54ced9cad..864dfaa67b7a 100644 --- a/fs/bcachefs/keylist.c +++ b/fs/bcachefs/keylist.c @@ -6,7 +6,7 @@ int bch2_keylist_realloc(struct keylist *l, u64 *inline_u64s, size_t nr_inline_u64s, size_t new_u64s) { - size_t oldsize = bch_keylist_u64s(l); + size_t oldsize = bch2_keylist_u64s(l); size_t newsize = oldsize + new_u64s; u64 *old_buf = l->keys_p == inline_u64s ? NULL : l->keys_p; u64 *new_keys; @@ -52,7 +52,7 @@ void bch2_keylist_pop_front(struct keylist *l) memmove_u64s_down(l->keys, bkey_next(l->keys), - bch_keylist_u64s(l)); + bch2_keylist_u64s(l)); } #ifdef CONFIG_BCACHEFS_DEBUG diff --git a/fs/bcachefs/keylist.h b/fs/bcachefs/keylist.h index a7ff86b08abc..195799bb20bc 100644 --- a/fs/bcachefs/keylist.h +++ b/fs/bcachefs/keylist.h @@ -36,14 +36,14 @@ static inline bool bch2_keylist_empty(struct keylist *l) return l->top == l->keys; } -static inline size_t bch_keylist_u64s(struct keylist *l) +static inline size_t bch2_keylist_u64s(struct keylist *l) { return l->top_p - l->keys_p; } static inline size_t bch2_keylist_bytes(struct keylist *l) { - return bch_keylist_u64s(l) * sizeof(u64); + return bch2_keylist_u64s(l) * sizeof(u64); } static inline struct bkey_i *bch2_keylist_front(struct keylist *l) diff --git a/fs/bcachefs/migrate.c b/fs/bcachefs/migrate.c index e26fa1608f39..96c8690adc5b 100644 --- a/fs/bcachefs/migrate.c +++ b/fs/bcachefs/migrate.c @@ -151,15 +151,8 @@ retry: } /* flush relevant btree updates */ - while (1) { - closure_wait_event(&c->btree_interior_update_wait, - !bch2_btree_interior_updates_nr_pending(c) || - c->btree_roots_dirty); - if (c->btree_roots_dirty) - bch2_journal_meta(&c->journal); - if (!bch2_btree_interior_updates_nr_pending(c)) - break; - } + closure_wait_event(&c->btree_interior_update_wait, + !bch2_btree_interior_updates_nr_pending(c)); ret = 0; err: diff --git a/fs/bcachefs/move.c b/fs/bcachefs/move.c index b82cd23fc848..b42350f9e9fb 100644 --- a/fs/bcachefs/move.c +++ b/fs/bcachefs/move.c @@ -178,9 +178,12 @@ next: } continue; nomatch: - if (m->ctxt) + if (m->ctxt) { + BUG_ON(k.k->p.offset <= iter->pos.offset); + atomic64_inc(&m->ctxt->stats->keys_raced); atomic64_add(k.k->p.offset - iter->pos.offset, &m->ctxt->stats->sectors_raced); + } atomic_long_inc(&c->extent_migrate_raced); trace_move_race(&new->k); bch2_btree_iter_next_slot(iter); @@ -313,12 +316,12 @@ static void move_free(struct closure *cl) { struct moving_io *io = container_of(cl, struct moving_io, cl); struct moving_context *ctxt = io->write.ctxt; + struct bvec_iter_all iter; struct bio_vec *bv; - unsigned i; bch2_disk_reservation_put(io->write.op.c, &io->write.op.res); - bio_for_each_segment_all(bv, &io->write.op.wbio.bio, i) + bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter) if (bv->bv_page) __free_page(bv->bv_page); @@ -775,14 +778,8 @@ int bch2_data_job(struct bch_fs *c, ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret; - while (1) { - closure_wait_event(&c->btree_interior_update_wait, - !bch2_btree_interior_updates_nr_pending(c) || - c->btree_roots_dirty); - if (!bch2_btree_interior_updates_nr_pending(c)) - break; - bch2_journal_meta(&c->journal); - } + closure_wait_event(&c->btree_interior_update_wait, + !bch2_btree_interior_updates_nr_pending(c)); ret = bch2_replicas_gc2(c) ?: ret; diff --git a/fs/bcachefs/move_types.h b/fs/bcachefs/move_types.h index 6788170d3f95..fc0de165af9f 100644 --- a/fs/bcachefs/move_types.h +++ b/fs/bcachefs/move_types.h @@ -8,6 +8,7 @@ struct bch_move_stats { struct bpos pos; atomic64_t keys_moved; + atomic64_t keys_raced; atomic64_t sectors_moved; atomic64_t sectors_seen; atomic64_t sectors_raced; diff --git a/fs/bcachefs/movinggc.c b/fs/bcachefs/movinggc.c index e9cb2304576f..0a87cd7405dd 100644 --- a/fs/bcachefs/movinggc.c +++ b/fs/bcachefs/movinggc.c @@ -78,7 +78,17 @@ static bool __copygc_pred(struct bch_dev *ca, ssize_t i = eytzinger0_find_le(h->data, h->used, sizeof(h->data[0]), bucket_offset_cmp, &search); +#if 0 + /* eytzinger search verify code: */ + ssize_t j = -1, k; + for (k = 0; k < h->used; k++) + if (h->data[k].offset <= ptr->offset && + (j < 0 || h->data[k].offset > h->data[j].offset)) + j = k; + + BUG_ON(i != j); +#endif return (i >= 0 && ptr->offset < h->data[i].offset + ca->mi.bucket_size && ptr->gen == h->data[i].gen); @@ -203,9 +213,12 @@ static void bch2_copygc(struct bch_fs *c, struct bch_dev *ca) if (sectors_not_moved && !ret) bch_warn_ratelimited(c, - "copygc finished but %llu/%llu sectors, %llu/%llu buckets not moved", + "copygc finished but %llu/%llu sectors, %llu/%llu buckets not moved (move stats: moved %llu sectors, raced %llu keys, %llu sectors)", sectors_not_moved, sectors_to_move, - buckets_not_moved, buckets_to_move); + buckets_not_moved, buckets_to_move, + atomic64_read(&move_stats.sectors_moved), + atomic64_read(&move_stats.keys_raced), + atomic64_read(&move_stats.sectors_raced)); trace_copygc(ca, atomic64_read(&move_stats.sectors_moved), sectors_not_moved, diff --git a/fs/bcachefs/opts.h b/fs/bcachefs/opts.h index ba4903352343..3b051e7a8f1d 100644 --- a/fs/bcachefs/opts.h +++ b/fs/bcachefs/opts.h @@ -207,6 +207,11 @@ enum opt_type { OPT_BOOL(), \ BCH_SB_PRJQUOTA, false, \ NULL, "Enable project quotas") \ + x(reflink, u8, \ + OPT_FORMAT|OPT_MOUNT|OPT_RUNTIME, \ + OPT_BOOL(), \ + BCH_SB_REFLINK, true, \ + NULL, "Enable reflink support") \ x(degraded, u8, \ OPT_MOUNT, \ OPT_BOOL(), \ @@ -260,6 +265,11 @@ enum opt_type { OPT_BOOL(), \ NO_SB_OPT, false, \ NULL, "Don't free journal entries/keys after startup")\ + x(read_entire_journal, u8, \ + 0, \ + OPT_BOOL(), \ + NO_SB_OPT, false, \ + NULL, "Read all journal entries, not just dirty ones")\ x(noexcl, u8, \ OPT_MOUNT, \ OPT_BOOL(), \ diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c index a4d0eec2ea3e..41b864dcdc39 100644 --- a/fs/bcachefs/recovery.c +++ b/fs/bcachefs/recovery.c @@ -188,7 +188,79 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *i iter->b = b; bch2_btree_node_iter_init_from_start(&iter->node_iter, iter->b); bch2_journal_iter_init(&iter->journal, journal_keys, - b->btree_id, b->level, b->data->min_key); + b->c.btree_id, b->c.level, b->data->min_key); +} + +/* Walk btree, overlaying keys from the journal: */ + +static int bch2_btree_and_journal_walk_recurse(struct bch_fs *c, struct btree *b, + struct journal_keys *journal_keys, + enum btree_id btree_id, + btree_walk_node_fn node_fn, + btree_walk_key_fn key_fn) +{ + struct btree_and_journal_iter iter; + struct bkey_s_c k; + int ret = 0; + + bch2_btree_and_journal_iter_init_node_iter(&iter, journal_keys, b); + + while ((k = bch2_btree_and_journal_iter_peek(&iter)).k) { + ret = key_fn(c, btree_id, b->c.level, k); + if (ret) + break; + + if (b->c.level) { + struct btree *child; + BKEY_PADDED(k) tmp; + + bkey_reassemble(&tmp.k, k); + k = bkey_i_to_s_c(&tmp.k); + + bch2_btree_and_journal_iter_advance(&iter); + + if (b->c.level > 0) { + child = bch2_btree_node_get_noiter(c, &tmp.k, + b->c.btree_id, b->c.level - 1); + ret = PTR_ERR_OR_ZERO(child); + if (ret) + break; + + ret = (node_fn ? node_fn(c, b) : 0) ?: + bch2_btree_and_journal_walk_recurse(c, child, + journal_keys, btree_id, node_fn, key_fn); + six_unlock_read(&child->c.lock); + + if (ret) + break; + } + } else { + bch2_btree_and_journal_iter_advance(&iter); + } + } + + return ret; +} + +int bch2_btree_and_journal_walk(struct bch_fs *c, struct journal_keys *journal_keys, + enum btree_id btree_id, + btree_walk_node_fn node_fn, + btree_walk_key_fn key_fn) +{ + struct btree *b = c->btree_roots[btree_id].b; + int ret = 0; + + if (btree_node_fake(b)) + return 0; + + six_lock_read(&b->c.lock, NULL, NULL); + ret = (node_fn ? node_fn(c, b) : 0) ?: + bch2_btree_and_journal_walk_recurse(c, b, journal_keys, btree_id, + node_fn, key_fn) ?: + key_fn(c, btree_id, b->c.level + 1, bkey_i_to_s_c(&b->key)); + six_unlock_read(&b->c.lock); + + return ret; } /* sort and dedup all keys in the journal: */ @@ -220,17 +292,6 @@ static int journal_sort_key_cmp(const void *_l, const void *_r) cmp_int(l->journal_offset, r->journal_offset); } -static int journal_sort_seq_cmp(const void *_l, const void *_r) -{ - const struct journal_key *l = _l; - const struct journal_key *r = _r; - - return cmp_int(r->level, l->level) ?: - cmp_int(l->journal_seq, r->journal_seq) ?: - cmp_int(l->btree_id, r->btree_id) ?: - bkey_cmp(l->k->k.p, r->k->k.p); -} - void bch2_journal_keys_free(struct journal_keys *keys) { kvfree(keys->d); @@ -247,20 +308,30 @@ static struct journal_keys journal_keys_sort(struct list_head *journal_entries) struct journal_key *src, *dst; size_t nr_keys = 0; - list_for_each_entry(p, journal_entries, list) + if (list_empty(journal_entries)) + return keys; + + keys.journal_seq_base = + le64_to_cpu(list_last_entry(journal_entries, + struct journal_replay, list)->j.last_seq); + + list_for_each_entry(p, journal_entries, list) { + if (le64_to_cpu(p->j.seq) < keys.journal_seq_base) + continue; + for_each_jset_key(k, _n, entry, &p->j) nr_keys++; + } - keys.journal_seq_base = - le64_to_cpu(list_first_entry(journal_entries, - struct journal_replay, - list)->j.seq); keys.d = kvmalloc(sizeof(keys.d[0]) * nr_keys, GFP_KERNEL); if (!keys.d) goto err; - list_for_each_entry(p, journal_entries, list) + list_for_each_entry(p, journal_entries, list) { + if (le64_to_cpu(p->j.seq) < keys.journal_seq_base) + continue; + for_each_jset_key(k, _n, entry, &p->j) keys.d[keys.nr++] = (struct journal_key) { .btree_id = entry->btree_id, @@ -270,6 +341,7 @@ static struct journal_keys journal_keys_sort(struct list_head *journal_entries) keys.journal_seq_base, .journal_offset = k->_data - p->j._data, }; + } sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_key_cmp, NULL); @@ -435,11 +507,48 @@ static int bch2_journal_replay_key(struct bch_fs *c, enum btree_id id, __bch2_journal_replay_key(&trans, id, level, k)); } +static int __bch2_alloc_replay_key(struct btree_trans *trans, struct bkey_i *k) +{ + struct btree_iter *iter; + int ret; + + iter = bch2_trans_get_iter(trans, BTREE_ID_ALLOC, k->k.p, + BTREE_ITER_CACHED| + BTREE_ITER_CACHED_NOFILL| + BTREE_ITER_INTENT); + ret = PTR_ERR_OR_ZERO(iter) ?: + bch2_trans_update(trans, iter, k, BTREE_TRIGGER_NORUN); + bch2_trans_iter_put(trans, iter); + return ret; +} + +static int bch2_alloc_replay_key(struct bch_fs *c, struct bkey_i *k) +{ + return bch2_trans_do(c, NULL, NULL, + BTREE_INSERT_NOFAIL| + BTREE_INSERT_USE_RESERVE| + BTREE_INSERT_LAZY_RW| + BTREE_INSERT_JOURNAL_REPLAY, + __bch2_alloc_replay_key(&trans, k)); +} + +static int journal_sort_seq_cmp(const void *_l, const void *_r) +{ + const struct journal_key *l = _l; + const struct journal_key *r = _r; + + return cmp_int(r->level, l->level) ?: + cmp_int(l->journal_seq, r->journal_seq) ?: + cmp_int(l->btree_id, r->btree_id) ?: + bkey_cmp(l->k->k.p, r->k->k.p); +} + static int bch2_journal_replay(struct bch_fs *c, struct journal_keys keys) { struct journal *j = &c->journal; struct journal_key *i; + u64 seq; int ret; sort(keys.d, keys.nr, sizeof(keys.d[0]), journal_sort_seq_cmp, NULL); @@ -447,26 +556,63 @@ static int bch2_journal_replay(struct bch_fs *c, if (keys.nr) replay_now_at(j, keys.journal_seq_base); + seq = j->replay_journal_seq; + + /* + * First replay updates to the alloc btree - these will only update the + * btree key cache: + */ for_each_journal_key(keys, i) { - if (!i->level) - replay_now_at(j, keys.journal_seq_base + i->journal_seq); + cond_resched(); - if (i->level) - ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k); - if (i->btree_id == BTREE_ID_ALLOC) + if (!i->level && i->btree_id == BTREE_ID_ALLOC) { + j->replay_journal_seq = keys.journal_seq_base + i->journal_seq; ret = bch2_alloc_replay_key(c, i->k); - else if (i->k->k.size) - ret = bch2_extent_replay_key(c, i->btree_id, i->k); - else - ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k); + if (ret) + goto err; + } + } - if (ret) { - bch_err(c, "journal replay: error %d while replaying key", - ret); - return ret; + /* + * Next replay updates to interior btree nodes: + */ + for_each_journal_key(keys, i) { + cond_resched(); + + if (i->level) { + j->replay_journal_seq = keys.journal_seq_base + i->journal_seq; + ret = bch2_journal_replay_key(c, i->btree_id, i->level, i->k); + if (ret) + goto err; } + } + + /* + * Now that the btree is in a consistent state, we can start journal + * reclaim (which will be flushing entries from the btree key cache back + * to the btree: + */ + set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags); + set_bit(JOURNAL_RECLAIM_STARTED, &j->flags); + + j->replay_journal_seq = seq; + /* + * Now replay leaf node updates: + */ + for_each_journal_key(keys, i) { cond_resched(); + + if (i->level || i->btree_id == BTREE_ID_ALLOC) + continue; + + replay_now_at(j, keys.journal_seq_base + i->journal_seq); + + ret = i->k->k.size + ? bch2_extent_replay_key(c, i->btree_id, i->k) + : bch2_journal_replay_key(c, i->btree_id, i->level, i->k); + if (ret) + goto err; } replay_now_at(j, j->replay_journal_seq_end); @@ -475,6 +621,9 @@ static int bch2_journal_replay(struct bch_fs *c, bch2_journal_set_replay_done(j); bch2_journal_flush_all_pins(j); return bch2_journal_error(j); +err: + bch_err(c, "journal replay: error %d while replaying key", ret); + return ret; } static bool journal_empty(struct list_head *journal) @@ -496,6 +645,9 @@ verify_journal_entries_not_blacklisted_or_missing(struct bch_fs *c, int ret = 0; list_for_each_entry(i, journal, list) { + if (le64_to_cpu(i->j.seq) < start_seq) + continue; + fsck_err_on(seq != le64_to_cpu(i->j.seq), c, "journal entries %llu-%llu missing! (replaying %llu-%llu)", seq, le64_to_cpu(i->j.seq) - 1, @@ -691,6 +843,7 @@ static int verify_superblock_clean(struct bch_fs *c, "superblock read clock doesn't match journal after clean shutdown"); for (i = 0; i < BTREE_ID_NR; i++) { + char buf1[200], buf2[200]; struct bkey_i *k1, *k2; unsigned l1 = 0, l2 = 0; @@ -706,7 +859,11 @@ static int verify_superblock_clean(struct bch_fs *c, k1->k.u64s != k2->k.u64s || memcmp(k1, k2, bkey_bytes(k1)) || l1 != l2, c, - "superblock btree root doesn't match journal after clean shutdown"); + "superblock btree root %u doesn't match journal after clean shutdown\n" + "sb: l=%u %s\n" + "journal: l=%u %s\n", i, + l1, (bch2_bkey_val_to_text(&PBUF(buf1), c, bkey_i_to_s_c(k1)), buf1), + l2, (bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(k2)), buf2)); } fsck_err: return ret; @@ -1077,12 +1234,24 @@ int bch2_fs_initialize(struct bch_fs *c) bch2_mark_dev_superblock(c, ca, 0); mutex_unlock(&c->sb_lock); + mutex_lock(&c->sb_lock); + c->disk_sb.sb->version = c->disk_sb.sb->version_min = + le16_to_cpu(bcachefs_metadata_version_current); + c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink; + c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL; + + bch2_write_super(c); + mutex_unlock(&c->sb_lock); + set_bit(BCH_FS_ALLOC_READ_DONE, &c->flags); set_bit(BCH_FS_INITIAL_GC_DONE, &c->flags); for (i = 0; i < BTREE_ID_NR; i++) bch2_btree_root_alloc(c, i); + set_bit(BCH_FS_BTREE_INTERIOR_REPLAY_DONE, &c->flags); + set_bit(JOURNAL_RECLAIM_STARTED, &c->journal.flags); + err = "unable to allocate journal buckets"; for_each_online_member(ca, c, i) { ret = bch2_dev_journal_alloc(ca); @@ -1135,11 +1304,6 @@ int bch2_fs_initialize(struct bch_fs *c) goto err; mutex_lock(&c->sb_lock); - c->disk_sb.sb->version = c->disk_sb.sb->version_min = - le16_to_cpu(bcachefs_metadata_version_current); - c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_atomic_nlink; - c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALL; - SET_BCH_SB_INITIALIZED(c->disk_sb.sb, true); SET_BCH_SB_CLEAN(c->disk_sb.sb, false); diff --git a/fs/bcachefs/recovery.h b/fs/bcachefs/recovery.h index 19f2f172a26b..a66827c9addf 100644 --- a/fs/bcachefs/recovery.h +++ b/fs/bcachefs/recovery.h @@ -44,6 +44,13 @@ void bch2_btree_and_journal_iter_init_node_iter(struct btree_and_journal_iter *, struct journal_keys *, struct btree *); +typedef int (*btree_walk_node_fn)(struct bch_fs *c, struct btree *b); +typedef int (*btree_walk_key_fn)(struct bch_fs *c, enum btree_id id, + unsigned level, struct bkey_s_c k); + +int bch2_btree_and_journal_walk(struct bch_fs *, struct journal_keys *, enum btree_id, + btree_walk_node_fn, btree_walk_key_fn); + void bch2_journal_keys_free(struct journal_keys *); void bch2_journal_entries_free(struct list_head *); diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c index 2f223be74926..3c473f1380a6 100644 --- a/fs/bcachefs/reflink.c +++ b/fs/bcachefs/reflink.c @@ -167,6 +167,9 @@ s64 bch2_remap_range(struct bch_fs *c, u64 src_done, dst_done; int ret = 0, ret2 = 0; + if (!c->opts.reflink) + return -EOPNOTSUPP; + if (!percpu_ref_tryget(&c->writes)) return -EROFS; diff --git a/fs/bcachefs/super-io.c b/fs/bcachefs/super-io.c index 6596764c8421..f2be64c869df 100644 --- a/fs/bcachefs/super-io.c +++ b/fs/bcachefs/super-io.c @@ -1,6 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 #include "bcachefs.h" +#include "btree_update_interior.h" #include "buckets.h" #include "checksum.h" #include "disk_groups.h" @@ -955,7 +956,6 @@ int bch2_fs_mark_dirty(struct bch_fs *c) mutex_lock(&c->sb_lock); SET_BCH_SB_CLEAN(c->disk_sb.sb, false); - c->disk_sb.sb->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA); c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite; c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates; c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_updates_journalled; @@ -989,27 +989,8 @@ bch2_journal_super_entries_add_common(struct bch_fs *c, struct jset_entry *entry, u64 journal_seq) { - struct btree_root *r; unsigned i; - mutex_lock(&c->btree_root_lock); - - for (r = c->btree_roots; - r < c->btree_roots + BTREE_ID_NR; - r++) - if (r->alive) { - entry_init_u64s(entry, r->key.u64s + 1); - entry->btree_id = r - c->btree_roots; - entry->level = r->level; - entry->type = BCH_JSET_ENTRY_btree_root; - bkey_copy(&entry->start[0], &r->key); - - entry = vstruct_next(entry); - } - c->btree_roots_dirty = false; - - mutex_unlock(&c->btree_root_lock); - percpu_down_write(&c->mark_lock); if (!journal_seq) { @@ -1110,6 +1091,7 @@ void bch2_fs_mark_clean(struct bch_fs *c) entry = sb_clean->start; entry = bch2_journal_super_entries_add_common(c, entry, 0); + entry = bch2_btree_roots_to_journal_entries(c, entry, entry); BUG_ON((void *) entry > vstruct_end(&sb_clean->field)); memset(entry, 0, diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c index d347389771e0..0cdf285e4ffd 100644 --- a/fs/bcachefs/super.c +++ b/fs/bcachefs/super.c @@ -13,6 +13,7 @@ #include "bkey_sort.h" #include "btree_cache.h" #include "btree_gc.h" +#include "btree_key_cache.h" #include "btree_update_interior.h" #include "btree_io.h" #include "chardev.h" @@ -207,7 +208,7 @@ int bch2_congested(void *data, int bdi_bits) static void __bch2_fs_read_only(struct bch_fs *c) { struct bch_dev *ca; - bool wrote; + bool wrote = false; unsigned i, clean_passes = 0; int ret; @@ -224,48 +225,68 @@ static void __bch2_fs_read_only(struct bch_fs *c) */ bch2_journal_flush_all_pins(&c->journal); + /* + * If the allocator threads didn't all start up, the btree updates to + * write out alloc info aren't going to work: + */ if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags)) - goto allocator_not_running; + goto nowrote_alloc; - do { - wrote = false; + bch_verbose(c, "writing alloc info"); + /* + * This should normally just be writing the bucket read/write clocks: + */ + ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?: + bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote); + bch_verbose(c, "writing alloc info complete"); - ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?: - bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote); + if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags)) + bch2_fs_inconsistent(c, "error writing out alloc info %i", ret); - if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags)) - bch2_fs_inconsistent(c, "error writing out alloc info %i", ret); + if (ret) + goto nowrote_alloc; - if (ret) - break; + bch_verbose(c, "flushing journal and stopping allocators"); - for_each_member_device(ca, c, i) - bch2_dev_allocator_quiesce(c, ca); + bch2_journal_flush_all_pins(&c->journal); + set_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags); - bch2_journal_flush_all_pins(&c->journal); + do { + clean_passes++; + + if (bch2_journal_flush_all_pins(&c->journal)) + clean_passes = 0; /* - * We need to explicitly wait on btree interior updates to complete - * before stopping the journal, flushing all journal pins isn't - * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree - * interior updates have to drop their journal pin before they're - * fully complete: + * In flight interior btree updates will generate more journal + * updates and btree updates (alloc btree): */ - closure_wait_event(&c->btree_interior_update_wait, - !bch2_btree_interior_updates_nr_pending(c)); + if (bch2_btree_interior_updates_nr_pending(c)) { + closure_wait_event(&c->btree_interior_update_wait, + !bch2_btree_interior_updates_nr_pending(c)); + clean_passes = 0; + } + flush_work(&c->btree_interior_update_work); - clean_passes = wrote ? 0 : clean_passes + 1; + if (bch2_journal_flush_all_pins(&c->journal)) + clean_passes = 0; } while (clean_passes < 2); -allocator_not_running: + bch_verbose(c, "flushing journal and stopping allocators complete"); + + set_bit(BCH_FS_ALLOC_CLEAN, &c->flags); +nowrote_alloc: + closure_wait_event(&c->btree_interior_update_wait, + !bch2_btree_interior_updates_nr_pending(c)); + flush_work(&c->btree_interior_update_work); + for_each_member_device(ca, c, i) bch2_dev_allocator_stop(ca); clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags); + clear_bit(BCH_FS_ALLOCATOR_STOPPING, &c->flags); bch2_fs_journal_stop(&c->journal); - /* XXX: mark super that alloc info is persistent */ - /* * the journal kicks off btree writes via reclaim - wait for in flight * writes after stopping journal: @@ -338,8 +359,11 @@ void bch2_fs_read_only(struct bch_fs *c) !test_bit(BCH_FS_ERROR, &c->flags) && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) && test_bit(BCH_FS_STARTED, &c->flags) && - !c->opts.norecovery) + test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) && + !c->opts.norecovery) { + bch_verbose(c, "marking filesystem clean"); bch2_fs_mark_clean(c); + } clear_bit(BCH_FS_RW, &c->flags); } @@ -349,9 +373,9 @@ static void bch2_fs_read_only_work(struct work_struct *work) struct bch_fs *c = container_of(work, struct bch_fs, read_only_work); - mutex_lock(&c->state_lock); + down_write(&c->state_lock); bch2_fs_read_only(c); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); } static void bch2_fs_read_only_async(struct bch_fs *c) @@ -426,20 +450,12 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early) if (ret) goto err; + clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags); + for_each_rw_member(ca, c, i) bch2_dev_allocator_add(c, ca); bch2_recalc_capacity(c); - if (!test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) { - ret = bch2_fs_allocator_start(c); - if (ret) { - bch_err(c, "error initializing allocator"); - goto err; - } - - set_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags); - } - for_each_rw_member(ca, c, i) { ret = bch2_dev_allocator_start(ca); if (ret) { @@ -494,7 +510,9 @@ static void bch2_fs_free(struct bch_fs *c) bch2_fs_ec_exit(c); bch2_fs_encryption_exit(c); bch2_fs_io_exit(c); + bch2_fs_btree_interior_update_exit(c); bch2_fs_btree_iter_exit(c); + bch2_fs_btree_key_cache_exit(&c->btree_key_cache); bch2_fs_btree_cache_exit(c); bch2_fs_journal_exit(&c->journal); bch2_io_clock_exit(&c->io_clock[WRITE]); @@ -511,8 +529,6 @@ static void bch2_fs_free(struct bch_fs *c) mempool_exit(&c->large_bkey_pool); mempool_exit(&c->btree_bounce_pool); bioset_exit(&c->btree_bio); - mempool_exit(&c->btree_interior_update_pool); - mempool_exit(&c->btree_reserve_pool); mempool_exit(&c->fill_iter); percpu_ref_exit(&c->writes); kfree(c->replicas.entries); @@ -551,9 +567,9 @@ void bch2_fs_stop(struct bch_fs *c) cancel_work_sync(&c->journal_seq_blacklist_gc_work); - mutex_lock(&c->state_lock); + down_write(&c->state_lock); bch2_fs_read_only(c); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); for_each_member_device(ca, c, i) if (ca->kobj.state_in_sysfs && @@ -625,7 +641,7 @@ static const char *bch2_fs_online(struct bch_fs *c) bch2_opts_create_sysfs_files(&c->opts_dir)) return "error creating sysfs objects"; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); err = "error creating sysfs objects"; __for_each_member_device(ca, c, i, NULL) @@ -635,7 +651,7 @@ static const char *bch2_fs_online(struct bch_fs *c) list_add(&c->list, &bch_fs_list); err = NULL; err: - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return err; } @@ -657,7 +673,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) c->minor = -1; c->disk_sb.fs_sb = true; - mutex_init(&c->state_lock); + init_rwsem(&c->state_lock); mutex_init(&c->sb_lock); mutex_init(&c->replicas_gc_lock); mutex_init(&c->btree_root_lock); @@ -668,6 +684,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) for (i = 0; i < BCH_TIME_STAT_NR; i++) bch2_time_stats_init(&c->times[i]); + bch2_fs_btree_key_cache_init_early(&c->btree_key_cache); bch2_fs_allocator_background_init(c); bch2_fs_allocator_foreground_init(c); bch2_fs_rebalance_init(c); @@ -675,11 +692,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) INIT_LIST_HEAD(&c->list); - INIT_LIST_HEAD(&c->btree_interior_update_list); - INIT_LIST_HEAD(&c->btree_interior_updates_unwritten); - mutex_init(&c->btree_reserve_cache_lock); - mutex_init(&c->btree_interior_update_lock); - mutex_init(&c->usage_scratch_lock); mutex_init(&c->bio_bounce_pages_lock); @@ -705,6 +717,8 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) seqcount_init(&c->usage_lock); + sema_init(&c->io_in_flight, 64); + c->copy_gc_enabled = 1; c->rebalance.enabled = 1; c->promote_whole_extents = true; @@ -752,10 +766,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) || percpu_ref_init(&c->writes, bch2_writes_disabled, PERCPU_REF_INIT_DEAD, GFP_KERNEL) || - mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1, - sizeof(struct btree_reserve)) || - mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1, - sizeof(struct btree_update)) || mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || bioset_init(&c->btree_bio, 1, max(offsetof(struct btree_read_bio, bio), @@ -770,7 +780,9 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) bch2_fs_journal_init(&c->journal) || bch2_fs_replicas_init(c) || bch2_fs_btree_cache_init(c) || + bch2_fs_btree_key_cache_init(&c->btree_key_cache) || bch2_fs_btree_iter_init(c) || + bch2_fs_btree_interior_update_init(c) || bch2_fs_io_init(c) || bch2_fs_encryption_init(c) || bch2_fs_compress_init(c) || @@ -855,7 +867,7 @@ int bch2_fs_start(struct bch_fs *c) unsigned i; int ret = -EINVAL; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); BUG_ON(test_bit(BCH_FS_STARTED, &c->flags)); @@ -905,7 +917,7 @@ int bch2_fs_start(struct bch_fs *c) print_mount_opts(c); ret = 0; out: - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return ret; err: switch (ret) { @@ -1405,22 +1417,47 @@ int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, { int ret; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); ret = __bch2_dev_set_state(c, ca, new_state, flags); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return ret; } /* Device add/removal: */ +int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca) +{ + struct btree_trans trans; + size_t i; + int ret; + + bch2_trans_init(&trans, c, 0, 0); + + for (i = 0; i < ca->mi.nbuckets; i++) { + ret = bch2_btree_key_cache_flush(&trans, + BTREE_ID_ALLOC, POS(ca->dev_idx, i)); + if (ret) + break; + } + bch2_trans_exit(&trans); + + if (ret) + return ret; + + return bch2_btree_delete_range(c, BTREE_ID_ALLOC, + POS(ca->dev_idx, 0), + POS(ca->dev_idx + 1, 0), + NULL); +} + int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) { struct bch_sb_field_members *mi; unsigned dev_idx = ca->dev_idx, data; int ret = -EINVAL; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); /* * We consume a reference to ca->ref, regardless of whether we succeed @@ -1447,10 +1484,7 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) goto err; } - ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC, - POS(ca->dev_idx, 0), - POS(ca->dev_idx + 1, 0), - NULL); + ret = bch2_dev_remove_alloc(c, ca); if (ret) { bch_err(ca, "Remove failed, error deleting alloc info"); goto err; @@ -1510,13 +1544,13 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) bch2_write_super(c); mutex_unlock(&c->sb_lock); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return 0; err: if (ca->mi.state == BCH_MEMBER_STATE_RW && !percpu_ref_is_zero(&ca->io_ref)) __bch2_dev_read_write(c, ca); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return ret; } @@ -1592,7 +1626,7 @@ int bch2_dev_add(struct bch_fs *c, const char *path) dev_usage_clear(ca); - mutex_lock(&c->state_lock); + down_write(&c->state_lock); mutex_lock(&c->sb_lock); err = "insufficient space in new superblock"; @@ -1653,12 +1687,12 @@ have_slot: goto err_late; } - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return 0; err_unlock: mutex_unlock(&c->sb_lock); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); err: if (ca) bch2_dev_free(ca); @@ -1681,11 +1715,11 @@ int bch2_dev_online(struct bch_fs *c, const char *path) const char *err; int ret; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); ret = bch2_read_super(path, &opts, &sb); if (ret) { - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return ret; } @@ -1716,10 +1750,10 @@ int bch2_dev_online(struct bch_fs *c, const char *path) bch2_write_super(c); mutex_unlock(&c->sb_lock); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return 0; err: - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); bch2_free_super(&sb); bch_err(c, "error bringing %s online: %s", path, err); return -EINVAL; @@ -1727,23 +1761,23 @@ err: int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) { - mutex_lock(&c->state_lock); + down_write(&c->state_lock); if (!bch2_dev_is_online(ca)) { bch_err(ca, "Already offline"); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return 0; } if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) { bch_err(ca, "Cannot offline required disk"); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return -EINVAL; } __bch2_dev_offline(c, ca); - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return 0; } @@ -1752,7 +1786,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) struct bch_member *mi; int ret = 0; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); if (nbuckets < ca->mi.nbuckets) { bch_err(ca, "Cannot shrink yet"); @@ -1783,7 +1817,7 @@ int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) bch2_recalc_capacity(c); err: - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); return ret; } @@ -1862,13 +1896,13 @@ struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, goto err; err = "bch2_dev_online() error"; - mutex_lock(&c->state_lock); + down_write(&c->state_lock); for (i = 0; i < nr_devices; i++) if (bch2_dev_attach_bdev(c, &sb[i])) { - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); goto err_print; } - mutex_unlock(&c->state_lock); + up_write(&c->state_lock); err = "insufficient devices"; if (!bch2_fs_may_start(c)) diff --git a/fs/bcachefs/sysfs.c b/fs/bcachefs/sysfs.c index d78ffcc0e8a4..c169d282a1f9 100644 --- a/fs/bcachefs/sysfs.c +++ b/fs/bcachefs/sysfs.c @@ -14,6 +14,7 @@ #include "btree_cache.h" #include "btree_io.h" #include "btree_iter.h" +#include "btree_key_cache.h" #include "btree_update.h" #include "btree_update_interior.h" #include "btree_gc.h" @@ -134,7 +135,6 @@ do { \ write_attribute(trigger_journal_flush); write_attribute(trigger_btree_coalesce); write_attribute(trigger_gc); -write_attribute(trigger_alloc_write); write_attribute(prune_cache); rw_attribute(btree_gc_periodic); @@ -166,6 +166,8 @@ read_attribute(journal_debug); read_attribute(journal_pins); read_attribute(btree_updates); read_attribute(dirty_btree_nodes); +read_attribute(btree_key_cache); +read_attribute(btree_transactions); read_attribute(internal_uuid); @@ -402,6 +404,20 @@ SHOW(bch2_fs) if (attr == &sysfs_dirty_btree_nodes) return bch2_dirty_btree_nodes_print(c, buf); + if (attr == &sysfs_btree_key_cache) { + struct printbuf out = _PBUF(buf, PAGE_SIZE); + + bch2_btree_key_cache_to_text(&out, &c->btree_key_cache); + return out.pos - buf; + } + + if (attr == &sysfs_btree_transactions) { + struct printbuf out = _PBUF(buf, PAGE_SIZE); + + bch2_btree_trans_to_text(&out, c); + return out.pos - buf; + } + if (attr == &sysfs_compression_stats) return bch2_compression_stats(c, buf); @@ -420,7 +436,7 @@ SHOW(bch2_fs) return 0; } -STORE(__bch2_fs) +STORE(bch2_fs) { struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); @@ -478,13 +494,17 @@ STORE(__bch2_fs) if (attr == &sysfs_trigger_btree_coalesce) bch2_coalesce(c); - if (attr == &sysfs_trigger_gc) + if (attr == &sysfs_trigger_gc) { + /* + * Full gc is currently incompatible with btree key cache: + */ +#if 0 + down_read(&c->state_lock); bch2_gc(c, NULL, false, false); - - if (attr == &sysfs_trigger_alloc_write) { - bool wrote; - - bch2_alloc_write(c, 0, &wrote); + up_read(&c->state_lock); +#else + bch2_gc_gens(c); +#endif } if (attr == &sysfs_prune_cache) { @@ -494,6 +514,7 @@ STORE(__bch2_fs) sc.nr_to_scan = strtoul_or_return(buf); c->btree_cache.shrink.scan_objects(&c->btree_cache.shrink, &sc); } + #ifdef CONFIG_BCACHEFS_TESTS if (attr == &sysfs_perf_test) { char *tmp = kstrdup(buf, GFP_KERNEL), *p = tmp; @@ -515,17 +536,6 @@ STORE(__bch2_fs) #endif return size; } - -STORE(bch2_fs) -{ - struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); - - mutex_lock(&c->state_lock); - size = __bch2_fs_store(kobj, attr, buf, size); - mutex_unlock(&c->state_lock); - - return size; -} SYSFS_OPS(bch2_fs); struct attribute *bch2_fs_files[] = { @@ -571,6 +581,8 @@ struct attribute *bch2_fs_internal_files[] = { &sysfs_journal_pins, &sysfs_btree_updates, &sysfs_dirty_btree_nodes, + &sysfs_btree_key_cache, + &sysfs_btree_transactions, &sysfs_read_realloc_races, &sysfs_extent_migrate_done, @@ -579,7 +591,6 @@ struct attribute *bch2_fs_internal_files[] = { &sysfs_trigger_journal_flush, &sysfs_trigger_btree_coalesce, &sysfs_trigger_gc, - &sysfs_trigger_alloc_write, &sysfs_prune_cache, &sysfs_copy_gc_enabled, @@ -835,6 +846,7 @@ static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf) " meta: %llu\n" " user: %llu\n" " cached: %llu\n" + " erasure coded: %llu\n" " fragmented: %llu\n" " copygc threshold: %llu\n" "freelist_wait: %s\n" @@ -861,6 +873,7 @@ static ssize_t show_dev_alloc_debug(struct bch_dev *ca, char *buf) stats.sectors[BCH_DATA_BTREE], stats.sectors[BCH_DATA_USER], stats.sectors[BCH_DATA_CACHED], + stats.sectors_ec, stats.sectors_fragmented, ca->copygc_threshold, c->freelist_wait.list.first ? "waiting" : "empty", diff --git a/fs/bcachefs/util.h b/fs/bcachefs/util.h index 2b19a0038045..0128daba5970 100644 --- a/fs/bcachefs/util.h +++ b/fs/bcachefs/util.h @@ -664,35 +664,6 @@ static inline void memset_u64s_tail(void *s, int c, unsigned bytes) memset(s + bytes, c, rem); } -static inline struct bio_vec next_contig_bvec(struct bio *bio, - struct bvec_iter *iter) -{ - struct bio_vec bv = bio_iter_iovec(bio, *iter); - - bio_advance_iter(bio, iter, bv.bv_len); -#ifndef CONFIG_HIGHMEM - while (iter->bi_size) { - struct bio_vec next = bio_iter_iovec(bio, *iter); - - if (page_address(bv.bv_page) + bv.bv_offset + bv.bv_len != - page_address(next.bv_page) + next.bv_offset) - break; - - bv.bv_len += next.bv_len; - bio_advance_iter(bio, iter, next.bv_len); - } -#endif - return bv; -} - -#define __bio_for_each_contig_segment(bv, bio, iter, start) \ - for (iter = (start); \ - (iter).bi_size && \ - ((bv = next_contig_bvec((bio), &(iter))), 1);) - -#define bio_for_each_contig_segment(bv, bio, iter) \ - __bio_for_each_contig_segment(bv, bio, iter, (bio)->bi_iter) - void sort_cmp_size(void *base, size_t num, size_t size, int (*cmp_func)(const void *, const void *, size_t), void (*swap_func)(void *, void *, size_t)); diff --git a/include/linux/six.h b/include/linux/six.h index 0fb1b2f49345..a16e94f482e9 100644 --- a/include/linux/six.h +++ b/include/linux/six.h @@ -115,6 +115,8 @@ struct six_lock { #endif }; +typedef int (*six_lock_should_sleep_fn)(struct six_lock *lock, void *); + static __always_inline void __six_lock_init(struct six_lock *lock, const char *name, struct lock_class_key *key) @@ -141,7 +143,7 @@ do { \ #define __SIX_LOCK(type) \ bool six_trylock_##type(struct six_lock *); \ bool six_relock_##type(struct six_lock *, u32); \ -void six_lock_##type(struct six_lock *); \ +int six_lock_##type(struct six_lock *, six_lock_should_sleep_fn, void *);\ void six_unlock_##type(struct six_lock *); __SIX_LOCK(read) @@ -167,14 +169,15 @@ static inline bool six_trylock_type(struct six_lock *lock, enum six_lock_type ty } static inline bool six_relock_type(struct six_lock *lock, enum six_lock_type type, - unsigned seq) + unsigned seq) { SIX_LOCK_DISPATCH(type, six_relock, lock, seq); } -static inline void six_lock_type(struct six_lock *lock, enum six_lock_type type) +static inline int six_lock_type(struct six_lock *lock, enum six_lock_type type, + six_lock_should_sleep_fn should_sleep_fn, void *p) { - SIX_LOCK_DISPATCH(type, six_lock, lock); + SIX_LOCK_DISPATCH(type, six_lock, lock, should_sleep_fn, p); } static inline void six_unlock_type(struct six_lock *lock, enum six_lock_type type) @@ -189,4 +192,6 @@ bool six_trylock_convert(struct six_lock *, enum six_lock_type, void six_lock_increment(struct six_lock *, enum six_lock_type); +void six_lock_wakeup_all(struct six_lock *); + #endif /* _LINUX_SIX_H */ diff --git a/include/trace/events/bcachefs.h b/include/trace/events/bcachefs.h index 01a9cc736cab..bafbccafae30 100644 --- a/include/trace/events/bcachefs.h +++ b/include/trace/events/bcachefs.h @@ -144,8 +144,8 @@ DECLARE_EVENT_CLASS(btree_node, TP_fast_assign( memcpy(__entry->uuid, c->sb.user_uuid.b, 16); - __entry->level = b->level; - __entry->id = b->btree_id; + __entry->level = b->c.level; + __entry->id = b->c.btree_id; __entry->inode = b->key.k.p.inode; __entry->offset = b->key.k.p.offset; ), @@ -262,7 +262,7 @@ TRACE_EVENT(btree_insert_key, ), TP_fast_assign( - __entry->id = b->btree_id; + __entry->id = b->c.btree_id; __entry->inode = k->k.p.inode; __entry->offset = k->k.p.offset; __entry->size = k->k.size; diff --git a/kernel/locking/six.c b/kernel/locking/six.c index 9fa58b6fadc9..49d46ed2e18e 100644 --- a/kernel/locking/six.c +++ b/kernel/locking/six.c @@ -15,7 +15,7 @@ #endif #define six_acquire(l, t) lock_acquire(l, 0, t, 0, 0, NULL, _RET_IP_) -#define six_release(l) lock_release(l, 0, _RET_IP_) +#define six_release(l) lock_release(l, _RET_IP_) struct six_lock_vals { /* Value we add to the lock in order to take the lock: */ @@ -108,7 +108,8 @@ static bool __six_trylock_type(struct six_lock *lock, enum six_lock_type type) if (!do_six_trylock_type(lock, type)) return false; - six_acquire(&lock->dep_map, 1); + if (type != SIX_LOCK_write) + six_acquire(&lock->dep_map, 1); return true; } @@ -130,7 +131,8 @@ static bool __six_relock_type(struct six_lock *lock, enum six_lock_type type, old.v + l[type].lock_val)) != old.v); six_set_owner(lock, type, old); - six_acquire(&lock->dep_map, 1); + if (type != SIX_LOCK_write) + six_acquire(&lock->dep_map, 1); return true; } @@ -265,15 +267,21 @@ static inline bool six_optimistic_spin(struct six_lock *lock, enum six_lock_type #endif noinline -static void __six_lock_type_slowpath(struct six_lock *lock, enum six_lock_type type) +static int __six_lock_type_slowpath(struct six_lock *lock, enum six_lock_type type, + six_lock_should_sleep_fn should_sleep_fn, void *p) { const struct six_lock_vals l[] = LOCK_VALS; union six_lock_state old, new; struct six_lock_waiter wait; + int ret = 0; u64 v; + ret = should_sleep_fn ? should_sleep_fn(lock, p) : 0; + if (ret) + return ret; + if (six_optimistic_spin(lock, type)) - return; + return 0; lock_contended(&lock->dep_map, _RET_IP_); @@ -290,6 +298,10 @@ static void __six_lock_type_slowpath(struct six_lock *lock, enum six_lock_type t raw_spin_unlock(&lock->wait_lock); } + ret = should_sleep_fn ? should_sleep_fn(lock, p) : 0; + if (ret) + break; + v = READ_ONCE(lock->state.v); do { new.v = old.v = v; @@ -309,7 +321,8 @@ static void __six_lock_type_slowpath(struct six_lock *lock, enum six_lock_type t schedule(); } - six_set_owner(lock, type, old); + if (!ret) + six_set_owner(lock, type, old); __set_current_state(TASK_RUNNING); @@ -318,17 +331,28 @@ static void __six_lock_type_slowpath(struct six_lock *lock, enum six_lock_type t list_del_init(&wait.list); raw_spin_unlock(&lock->wait_lock); } + + return ret; } __always_inline -static void __six_lock_type(struct six_lock *lock, enum six_lock_type type) +static int __six_lock_type(struct six_lock *lock, enum six_lock_type type, + six_lock_should_sleep_fn should_sleep_fn, void *p) { - six_acquire(&lock->dep_map, 0); + int ret; - if (!do_six_trylock_type(lock, type)) - __six_lock_type_slowpath(lock, type); + if (type != SIX_LOCK_write) + six_acquire(&lock->dep_map, 0); + + ret = do_six_trylock_type(lock, type) ? 0 + : __six_lock_type_slowpath(lock, type, should_sleep_fn, p); - lock_acquired(&lock->dep_map, _RET_IP_); + if (ret && type != SIX_LOCK_write) + six_release(&lock->dep_map); + if (!ret) + lock_acquired(&lock->dep_map, _RET_IP_); + + return ret; } static inline void six_lock_wakeup(struct six_lock *lock, @@ -382,7 +406,8 @@ static void __six_unlock_type(struct six_lock *lock, enum six_lock_type type) EBUG_ON(type == SIX_LOCK_write && !(lock->state.v & __SIX_LOCK_HELD_intent)); - six_release(&lock->dep_map); + if (type != SIX_LOCK_write) + six_release(&lock->dep_map); if (type == SIX_LOCK_intent) { EBUG_ON(lock->owner != current); @@ -413,9 +438,10 @@ bool six_relock_##type(struct six_lock *lock, u32 seq) \ } \ EXPORT_SYMBOL_GPL(six_relock_##type); \ \ -void six_lock_##type(struct six_lock *lock) \ +int six_lock_##type(struct six_lock *lock, \ + six_lock_should_sleep_fn should_sleep_fn, void *p) \ { \ - __six_lock_type(lock, SIX_LOCK_##type); \ + return __six_lock_type(lock, SIX_LOCK_##type, should_sleep_fn, p);\ } \ EXPORT_SYMBOL_GPL(six_lock_##type); \ \ @@ -510,3 +536,18 @@ void six_lock_increment(struct six_lock *lock, enum six_lock_type type) } } EXPORT_SYMBOL_GPL(six_lock_increment); + +void six_lock_wakeup_all(struct six_lock *lock) +{ + struct six_lock_waiter *w; + + raw_spin_lock(&lock->wait_lock); + + list_for_each_entry(w, &lock->wait_list[0], list) + wake_up_process(w->task); + list_for_each_entry(w, &lock->wait_list[1], list) + wake_up_process(w->task); + + raw_spin_unlock(&lock->wait_lock); +} +EXPORT_SYMBOL_GPL(six_lock_wakeup_all); |