+/* SPDX-License-Identifier: GPL-2.0 */
/*
* Code for manipulating bucket marks for garbage collection.
*
for (_b = (_buckets)->b + (_buckets)->first_bucket; \
_b < (_buckets)->b + (_buckets)->nbuckets; _b++)
-#define bucket_cmpxchg(g, new, expr) \
-({ \
- u64 _v = READ_ONCE((g)->_mark.counter); \
- struct bucket_mark _old; \
- \
- do { \
- (new).counter = _old.counter = _v; \
- expr; \
- } while ((_v = cmpxchg(&(g)->_mark.counter, \
- _old.counter, \
- (new).counter)) != _old.counter);\
- _old; \
-})
-
-static inline struct bucket_array *bucket_array(struct bch_dev *ca)
+static inline void bucket_unlock(struct bucket *b)
{
- return rcu_dereference_check(ca->buckets,
+ smp_store_release(&b->lock, 0);
+}
+
+static inline void bucket_lock(struct bucket *b)
+{
+ while (xchg(&b->lock, 1))
+ cpu_relax();
+}
+
+static inline struct bucket_array *gc_bucket_array(struct bch_dev *ca)
+{
+ return rcu_dereference_check(ca->buckets_gc,
!ca->fs ||
- lockdep_is_held(&ca->fs->usage_lock) ||
+ percpu_rwsem_is_held(&ca->fs->mark_lock) ||
lockdep_is_held(&ca->fs->gc_lock) ||
lockdep_is_held(&ca->bucket_lock));
}
-static inline struct bucket *bucket(struct bch_dev *ca, size_t b)
+static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b)
{
- struct bucket_array *buckets = bucket_array(ca);
+ struct bucket_array *buckets = gc_bucket_array(ca);
BUG_ON(b < buckets->first_bucket || b >= buckets->nbuckets);
return buckets->b + b;
}
-static inline void bucket_io_clock_reset(struct bch_fs *c, struct bch_dev *ca,
- size_t b, int rw)
+static inline struct bucket_gens *bucket_gens(struct bch_dev *ca)
{
- bucket(ca, b)->io_time[rw] = c->bucket_clock[rw].hand;
+ return rcu_dereference_check(ca->bucket_gens,
+ !ca->fs ||
+ percpu_rwsem_is_held(&ca->fs->mark_lock) ||
+ lockdep_is_held(&ca->fs->gc_lock) ||
+ lockdep_is_held(&ca->bucket_lock));
}
-static inline u16 bucket_last_io(struct bch_fs *c, struct bucket *g, int rw)
+static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
{
- return c->bucket_clock[rw].hand - g->io_time[rw];
-}
+ struct bucket_gens *gens = bucket_gens(ca);
-/*
- * bucket_gc_gen() returns the difference between the bucket's current gen and
- * the oldest gen of any pointer into that bucket in the btree.
- */
-
-static inline u8 bucket_gc_gen(struct bch_dev *ca, size_t b)
-{
- return bucket(ca, b)->mark.gen - ca->oldest_gens[b];
+ BUG_ON(b < gens->first_bucket || b >= gens->nbuckets);
+ return gens->b + b;
}
static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
return sector_to_bucket(ca, ptr->offset);
}
-static inline struct bucket *PTR_BUCKET(struct bch_dev *ca,
- const struct bch_extent_ptr *ptr)
+static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca,
+ const struct bch_extent_ptr *ptr)
{
- return bucket(ca, PTR_BUCKET_NR(ca, ptr));
+ return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr));
}
-static inline struct bucket_mark ptr_bucket_mark(struct bch_dev *ca,
- const struct bch_extent_ptr *ptr)
+static inline enum bch_data_type ptr_data_type(const struct bkey *k,
+ const struct bch_extent_ptr *ptr)
{
- struct bucket_mark m;
-
- rcu_read_lock();
- m = READ_ONCE(bucket(ca, PTR_BUCKET_NR(ca, ptr))->mark);
- rcu_read_unlock();
+ if (k->type == KEY_TYPE_btree_ptr ||
+ k->type == KEY_TYPE_btree_ptr_v2)
+ return BCH_DATA_btree;
- return m;
+ return ptr->cached ? BCH_DATA_cached : BCH_DATA_user;
}
static inline int gen_cmp(u8 a, u8 b)
static inline u8 ptr_stale(struct bch_dev *ca,
const struct bch_extent_ptr *ptr)
{
- return gen_after(ptr_bucket_mark(ca, ptr).gen, ptr->gen);
-}
-
-/* bucket gc marks */
+ u8 ret;
-/* The dirty and cached sector counts saturate. If this occurs,
- * reference counting alone will not free the bucket, and a btree
- * GC must be performed. */
-#define GC_MAX_SECTORS_USED ((1U << 15) - 1)
-
-static inline unsigned bucket_sectors_used(struct bucket_mark mark)
-{
- return mark.dirty_sectors + mark.cached_sectors;
-}
+ rcu_read_lock();
+ ret = gen_after(*bucket_gen(ca, PTR_BUCKET_NR(ca, ptr)), ptr->gen);
+ rcu_read_unlock();
-static inline bool bucket_unused(struct bucket_mark mark)
-{
- return !mark.owned_by_allocator &&
- !mark.data_type &&
- !bucket_sectors_used(mark);
+ return ret;
}
/* Device usage: */
-struct bch_dev_usage __bch2_dev_usage_read(struct bch_dev *);
-struct bch_dev_usage bch2_dev_usage_read(struct bch_fs *, struct bch_dev *);
+struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *);
static inline u64 __dev_buckets_available(struct bch_dev *ca,
- struct bch_dev_usage stats)
+ struct bch_dev_usage stats,
+ enum alloc_reserve reserve)
{
- u64 total = ca->mi.nbuckets - ca->mi.first_bucket;
+ s64 total = ca->mi.nbuckets - ca->mi.first_bucket;
+ s64 reserved = 0;
+
+ switch (reserve) {
+ case RESERVE_none:
+ reserved += ca->mi.nbuckets >> 6;
+ fallthrough;
+ case RESERVE_movinggc:
+ reserved += ca->nr_btree_reserve;
+ fallthrough;
+ case RESERVE_btree:
+ reserved += ca->nr_btree_reserve;
+ fallthrough;
+ case RESERVE_btree_movinggc:
+ break;
+ default:
+ BUG();
+ }
if (WARN_ONCE(stats.buckets_unavailable > total,
- "buckets_unavailable overflow\n"))
+ "buckets_unavailable overflow (%llu > %llu)\n",
+ stats.buckets_unavailable, total))
return 0;
- return total - stats.buckets_unavailable;
+ return max_t(s64, 0,
+ total -
+ stats.buckets_unavailable -
+ ca->nr_open_buckets -
+ reserved);
}
-/*
- * Number of reclaimable buckets - only for use by the allocator thread:
- */
-static inline u64 dev_buckets_available(struct bch_fs *c, struct bch_dev *ca)
+static inline u64 dev_buckets_available(struct bch_dev *ca,
+ enum alloc_reserve reserve)
{
- return __dev_buckets_available(ca, bch2_dev_usage_read(c, ca));
+ return __dev_buckets_available(ca, bch2_dev_usage_read(ca), reserve);
}
-static inline u64 __dev_buckets_free(struct bch_dev *ca,
- struct bch_dev_usage stats)
+/* Filesystem usage: */
+
+static inline unsigned fs_usage_u64s(struct bch_fs *c)
{
- return __dev_buckets_available(ca, stats) +
- fifo_used(&ca->free[RESERVE_NONE]) +
- fifo_used(&ca->free_inc);
+ return sizeof(struct bch_fs_usage) / sizeof(u64) +
+ READ_ONCE(c->replicas.nr);
}
-static inline u64 dev_buckets_free(struct bch_fs *c, struct bch_dev *ca)
+static inline unsigned dev_usage_u64s(void)
{
- return __dev_buckets_free(ca, bch2_dev_usage_read(c, ca));
+ return sizeof(struct bch_dev_usage) / sizeof(u64);
}
-/* Filesystem usage: */
+u64 bch2_fs_usage_read_one(struct bch_fs *, u64 *);
-static inline enum bch_data_type s_alloc_to_data_type(enum s_alloc s)
-{
- switch (s) {
- case S_META:
- return BCH_DATA_BTREE;
- case S_DIRTY:
- return BCH_DATA_USER;
- default:
- BUG();
- }
-}
+struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *);
-struct bch_fs_usage __bch2_fs_usage_read(struct bch_fs *);
-struct bch_fs_usage bch2_fs_usage_read(struct bch_fs *);
-void bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *,
- struct disk_reservation *, struct gc_pos);
+void bch2_fs_usage_acc_to_base(struct bch_fs *, unsigned);
-u64 __bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage);
-u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage);
-u64 bch2_fs_sectors_free(struct bch_fs *, struct bch_fs_usage);
+void bch2_fs_usage_to_text(struct printbuf *,
+ struct bch_fs *, struct bch_fs_usage_online *);
-static inline bool is_available_bucket(struct bucket_mark mark)
-{
- return (!mark.owned_by_allocator &&
- !mark.dirty_sectors &&
- !mark.nouse);
-}
+u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage_online *);
-static inline bool bucket_needs_journal_commit(struct bucket_mark m,
- u16 last_seq_ondisk)
-{
- return m.journal_seq_valid &&
- ((s16) m.journal_seq - (s16) last_seq_ondisk > 0);
-}
+struct bch_fs_usage_short
+bch2_fs_usage_read_short(struct bch_fs *);
-void bch2_bucket_seq_cleanup(struct bch_fs *);
+/* key/bucket marking: */
+
+void bch2_fs_usage_initialize(struct bch_fs *);
-bool bch2_invalidate_bucket(struct bch_fs *, struct bch_dev *,
- size_t, struct bucket_mark *);
-void bch2_mark_alloc_bucket(struct bch_fs *, struct bch_dev *,
- size_t, bool, struct gc_pos, unsigned);
void bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
size_t, enum bch_data_type, unsigned,
struct gc_pos, unsigned);
-#define BCH_BUCKET_MARK_NOATOMIC (1 << 0)
-#define BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE (1 << 1)
-#define BCH_BUCKET_MARK_GC_WILL_VISIT (1 << 2)
-#define BCH_BUCKET_MARK_GC_LOCK_HELD (1 << 3)
+int bch2_mark_alloc(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_extent(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_stripe(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_inode(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_reservation(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_reflink_p(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
+
+int bch2_trans_mark_extent(struct btree_trans *, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_stripe(struct btree_trans *, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_inode(struct btree_trans *, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_reservation(struct btree_trans *, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_reflink_p(struct btree_trans *, struct bkey_s_c, struct bkey_i *, unsigned);
+
+int bch2_mark_key(struct btree_trans *, struct bkey_s_c, struct bkey_s_c, unsigned);
-void bch2_mark_key(struct bch_fs *, struct bkey_s_c, s64, bool, struct gc_pos,
- struct bch_fs_usage *, u64, unsigned);
+int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c,
+ struct bkey_i *, unsigned);
-void bch2_recalc_sectors_available(struct bch_fs *);
+static inline int bch2_trans_mark_old(struct btree_trans *trans,
+ struct bkey_s_c old, unsigned flags)
+{
+ struct bkey_i deleted;
+
+ bkey_init(&deleted.k);
+ deleted.k.p = old.k->p;
+
+ return bch2_trans_mark_key(trans, old, &deleted,
+ BTREE_TRIGGER_OVERWRITE|flags);
+}
-void __bch2_disk_reservation_put(struct bch_fs *, struct disk_reservation *);
+static inline int bch2_trans_mark_new(struct btree_trans *trans,
+ struct bkey_i *new, unsigned flags)
+{
+ struct bkey_i deleted;
+
+ bkey_init(&deleted.k);
+ deleted.k.p = new->k.p;
+
+ return bch2_trans_mark_key(trans, bkey_i_to_s_c(&deleted), new,
+ BTREE_TRIGGER_INSERT|flags);
+}
+
+int bch2_trans_fs_usage_apply(struct btree_trans *, struct replicas_delta_list *);
+
+int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *,
+ size_t, enum bch_data_type, unsigned);
+int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *);
+
+/* disk reservations: */
static inline void bch2_disk_reservation_put(struct bch_fs *c,
struct disk_reservation *res)
{
- if (res->sectors)
- __bch2_disk_reservation_put(c, res);
+ this_cpu_sub(*c->online_reserved, res->sectors);
+ res->sectors = 0;
}
#define BCH_DISK_RESERVATION_NOFAIL (1 << 0)
-#define BCH_DISK_RESERVATION_GC_LOCK_HELD (1 << 1)
-#define BCH_DISK_RESERVATION_BTREE_LOCKS_HELD (1 << 2)
int bch2_disk_reservation_add(struct bch_fs *,
- struct disk_reservation *,
- unsigned, int);
+ struct disk_reservation *,
+ u64, int);
static inline struct disk_reservation
bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas)
static inline int bch2_disk_reservation_get(struct bch_fs *c,
struct disk_reservation *res,
- unsigned sectors,
- unsigned nr_replicas,
+ u64 sectors, unsigned nr_replicas,
int flags)
{
*res = bch2_disk_reservation_init(c, nr_replicas);
return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags);
}
+#define RESERVE_FACTOR 6
+
+static inline u64 avail_factor(u64 r)
+{
+ return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1);
+}
+
int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64);
void bch2_dev_buckets_free(struct bch_dev *);
int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *);