X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fbuckets.h;h=bdf4fff9cb8a8a69e372a03daf9051231ae352bd;hb=1f78fed4693a5361f56508daac59bebd5b556379;hp=76ebe2ecb56ff834a442bf586150b4b21e2e0f90;hpb=bca8b084ad754afc54e628d9db7721b90d9480b7;p=bcachefs-tools-debian

diff --git a/libbcachefs/buckets.h b/libbcachefs/buckets.h
index 76ebe2e..bdf4fff 100644
--- a/libbcachefs/buckets.h
+++ b/libbcachefs/buckets.h
@@ -1,3 +1,4 @@
+/* SPDX-License-Identifier: GPL-2.0 */
 /*
  * Code for manipulating bucket marks for garbage collection.
  *
@@ -8,97 +9,104 @@
 #define _BUCKETS_H
 
 #include "buckets_types.h"
+#include "extents.h"
 #include "super.h"
 
 #define for_each_bucket(_b, _buckets)				\
 	for (_b = (_buckets)->b + (_buckets)->first_bucket;	\
 	     _b < (_buckets)->b + (_buckets)->nbuckets; _b++)
 
-#define bucket_cmpxchg(g, new, expr)				\
-({								\
-	u64 _v = atomic64_read(&(g)->_mark.v);			\
-	struct bucket_mark _old;				\
-								\
-	do {							\
-		(new).v.counter = _old.v.counter = _v;		\
-		expr;						\
-	} while ((_v = atomic64_cmpxchg(&(g)->_mark.v,		\
-			       _old.v.counter,			\
-			       (new).v.counter)) != _old.v.counter);\
-	_old;							\
-})
-
-static inline struct bucket_array *__bucket_array(struct bch_dev *ca,
-						  bool gc)
+static inline void bucket_unlock(struct bucket *b)
 {
-	return rcu_dereference_check(ca->buckets[gc],
-				     !ca->fs ||
-				     percpu_rwsem_is_held(&ca->fs->usage_lock) ||
-				     lockdep_is_held(&ca->fs->gc_lock) ||
-				     lockdep_is_held(&ca->bucket_lock));
+	smp_store_release(&b->lock, 0);
 }
 
-static inline struct bucket_array *bucket_array(struct bch_dev *ca)
+static inline void bucket_lock(struct bucket *b)
 {
-	return __bucket_array(ca, false);
+	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 ||
+				     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, bool gc)
+static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b)
 {
-	struct bucket_array *buckets = __bucket_array(ca, gc);
+	struct bucket_array *buckets = gc_bucket_array(ca);
 
 	BUG_ON(b < buckets->first_bucket || b >= buckets->nbuckets);
 	return buckets->b + b;
 }
 
-static inline struct bucket *bucket(struct bch_dev *ca, size_t b)
+static inline struct bucket_gens *bucket_gens(struct bch_dev *ca)
 {
-	return __bucket(ca, b, false);
+	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 void bucket_io_clock_reset(struct bch_fs *c, struct bch_dev *ca,
-					 size_t b, int rw)
+static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
 {
-	bucket(ca, b)->io_time[rw] = c->bucket_clock[rw].hand;
+	struct bucket_gens *gens = bucket_gens(ca);
+
+	BUG_ON(b < gens->first_bucket || b >= gens->nbuckets);
+	return gens->b + b;
 }
 
-static inline u16 bucket_last_io(struct bch_fs *c, struct bucket *g, int rw)
+static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
+				   const struct bch_extent_ptr *ptr)
 {
-	return c->bucket_clock[rw].hand - g->io_time[rw];
+	return sector_to_bucket(ca, ptr->offset);
 }
 
-/*
- * 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)
+static inline struct bpos PTR_BUCKET_POS(const struct bch_fs *c,
+				   const struct bch_extent_ptr *ptr)
 {
-	return bucket(ca, b)->mark.gen - ca->oldest_gens[b];
+	struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+
+	return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr));
 }
 
-static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
-				   const struct bch_extent_ptr *ptr)
+static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_fs *c,
+						const struct bch_extent_ptr *ptr,
+						u32 *bucket_offset)
 {
-	return sector_to_bucket(ca, ptr->offset);
+	struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+
+	return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_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;
+	if (bkey_is_btree_ptr(k))
+		return BCH_DATA_btree;
 
-	rcu_read_lock();
-	m = READ_ONCE(bucket(ca, PTR_BUCKET_NR(ca, ptr))->mark);
-	rcu_read_unlock();
+	return ptr->cached ? BCH_DATA_cached : BCH_DATA_user;
+}
+
+static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
+{
+	EBUG_ON(sectors < 0);
 
-	return m;
+	return crc_is_compressed(p.crc)
+		? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size,
+				   p.crc.uncompressed_size)
+		: sectors;
 }
 
 static inline int gen_cmp(u8 a, u8 b)
@@ -120,126 +128,196 @@ static inline int gen_after(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);
-}
+	u8 ret;
 
-/* bucket gc marks */
-
-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 *, bool);
-struct bch_dev_usage bch2_dev_usage_read(struct bch_fs *, struct bch_dev *);
-
-static inline u64 __dev_buckets_available(struct bch_dev *ca,
-					  struct bch_dev_usage stats)
+void bch2_dev_usage_read_fast(struct bch_dev *, struct bch_dev_usage *);
+static inline struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca)
 {
-	u64 total = ca->mi.nbuckets - ca->mi.first_bucket;
+	struct bch_dev_usage ret;
+
+	bch2_dev_usage_read_fast(ca, &ret);
+	return ret;
+}
 
-	if (WARN_ONCE(stats.buckets_unavailable > total,
-		      "buckets_unavailable overflow (%llu > %llu)\n",
-		      stats.buckets_unavailable, total))
-		return 0;
+void bch2_dev_usage_init(struct bch_dev *);
 
-	return total - stats.buckets_unavailable;
+static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum alloc_reserve reserve)
+{
+	s64 reserved = 0;
+
+	switch (reserve) {
+	case RESERVE_NR:
+		unreachable();
+	case RESERVE_stripe:
+		reserved += ca->mi.nbuckets >> 6;
+		fallthrough;
+	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;
+	}
+
+	return 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_free(struct bch_dev *ca,
+				   struct bch_dev_usage usage,
+				   enum alloc_reserve reserve)
 {
-	return __dev_buckets_available(ca, bch2_dev_usage_read(c, ca));
+	return max_t(s64, 0,
+		     usage.d[BCH_DATA_free].buckets -
+		     ca->nr_open_buckets -
+		     bch2_dev_buckets_reserved(ca, reserve));
 }
 
-static inline u64 __dev_buckets_free(struct bch_dev *ca,
-				     struct bch_dev_usage stats)
+static inline u64 __dev_buckets_available(struct bch_dev *ca,
+					  struct bch_dev_usage usage,
+					  enum alloc_reserve reserve)
 {
-	return __dev_buckets_available(ca, stats) +
-		fifo_used(&ca->free[RESERVE_NONE]) +
-		fifo_used(&ca->free_inc);
+	return max_t(s64, 0,
+		       usage.d[BCH_DATA_free].buckets
+		     + usage.d[BCH_DATA_cached].buckets
+		     + usage.d[BCH_DATA_need_gc_gens].buckets
+		     + usage.d[BCH_DATA_need_discard].buckets
+		     - ca->nr_open_buckets
+		     - bch2_dev_buckets_reserved(ca, reserve));
 }
 
-static inline u64 dev_buckets_free(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_free(ca, bch2_dev_usage_read(c, ca));
+	return __dev_buckets_available(ca, bch2_dev_usage_read(ca), reserve);
 }
 
 /* Filesystem usage: */
 
-struct bch_fs_usage __bch2_fs_usage_read(struct bch_fs *, bool);
-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);
+static inline unsigned __fs_usage_u64s(unsigned nr_replicas)
+{
+	return sizeof(struct bch_fs_usage) / sizeof(u64) + nr_replicas;
+}
 
-u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage);
+static inline unsigned fs_usage_u64s(struct bch_fs *c)
+{
+	return __fs_usage_u64s(READ_ONCE(c->replicas.nr));
+}
 
-static inline u64 bch2_fs_sectors_free(struct bch_fs *c,
-				       struct bch_fs_usage stats)
+static inline unsigned __fs_usage_online_u64s(unsigned nr_replicas)
 {
-	return c->capacity - bch2_fs_sectors_used(c, stats);
+	return sizeof(struct bch_fs_usage_online) / sizeof(u64) + nr_replicas;
 }
 
-static inline bool is_available_bucket(struct bucket_mark mark)
+static inline unsigned fs_usage_online_u64s(struct bch_fs *c)
 {
-	return (!mark.owned_by_allocator &&
-		!mark.dirty_sectors &&
-		!mark.stripe &&
-		!mark.nouse);
+	return __fs_usage_online_u64s(READ_ONCE(c->replicas.nr));
 }
 
-static inline bool bucket_needs_journal_commit(struct bucket_mark m,
-					       u16 last_seq_ondisk)
+static inline unsigned dev_usage_u64s(void)
 {
-	return m.journal_seq_valid &&
-		((s16) m.journal_seq - (s16) last_seq_ondisk > 0);
+	return sizeof(struct bch_dev_usage) / sizeof(u64);
 }
 
-void bch2_bucket_seq_cleanup(struct bch_fs *);
+u64 bch2_fs_usage_read_one(struct bch_fs *, u64 *);
+
+struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *);
 
-void 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);
+void bch2_fs_usage_acc_to_base(struct bch_fs *, unsigned);
 
-#define BCH_BUCKET_MARK_NOATOMIC		(1 << 0)
-#define BCH_BUCKET_MARK_GC			(1 << 1)
+void bch2_fs_usage_to_text(struct printbuf *,
+			   struct bch_fs *, struct bch_fs_usage_online *);
 
-void bch2_mark_key(struct bch_fs *, enum bkey_type, struct bkey_s_c,
-		   bool, s64, struct gc_pos,
-		   struct bch_fs_usage *, u64, unsigned);
-void bch2_mark_update(struct btree_insert *, struct btree_insert_entry *);
+u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage_online *);
 
-void __bch2_disk_reservation_put(struct bch_fs *, struct disk_reservation *);
+struct bch_fs_usage_short
+bch2_fs_usage_read_short(struct bch_fs *);
+
+/* key/bucket marking: */
+
+void bch2_fs_usage_initialize(struct bch_fs *);
+
+int bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
+			      size_t, enum bch_data_type, unsigned,
+			      struct gc_pos, unsigned);
+
+int bch2_mark_alloc(struct btree_trans *, enum btree_id, unsigned,
+		    struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_extent(struct btree_trans *, enum btree_id, unsigned,
+		     struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_stripe(struct btree_trans *, enum btree_id, unsigned,
+		     struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_inode(struct btree_trans *, enum btree_id, unsigned,
+		    struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_reservation(struct btree_trans *, enum btree_id, unsigned,
+			  struct bkey_s_c, struct bkey_s_c, unsigned);
+int bch2_mark_reflink_p(struct btree_trans *, enum btree_id, unsigned,
+			struct bkey_s_c, struct bkey_s_c, unsigned);
+
+int bch2_trans_mark_extent(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_stripe(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_inode(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_reservation(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
+int bch2_trans_mark_reflink_p(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
+
+void bch2_trans_fs_usage_revert(struct btree_trans *, struct replicas_delta_list *);
+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);
+	if (res->sectors) {
+		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);
+int __bch2_disk_reservation_add(struct bch_fs *,
+				struct disk_reservation *,
+				u64, int);
+
+static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
+					    u64 sectors, int flags)
+{
+#ifdef __KERNEL__
+	u64 old, new;
+
+	do {
+		old = this_cpu_read(c->pcpu->sectors_available);
+		if (sectors > old)
+			return __bch2_disk_reservation_add(c, res, sectors, flags);
+
+		new = old - sectors;
+	} while (this_cpu_cmpxchg(c->pcpu->sectors_available, old, new) != old);
+
+	this_cpu_add(*c->online_reserved, sectors);
+	res->sectors			+= sectors;
+	return 0;
+#else
+	return __bch2_disk_reservation_add(c, res, sectors, flags);
+#endif
+}
 
 static inline struct disk_reservation
 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas)
@@ -256,8 +334,7 @@ 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);
@@ -265,6 +342,13 @@ static inline int bch2_disk_reservation_get(struct bch_fs *c,
 	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 *);