1 files changed, 201 insertions, 55 deletions

diff --git a/fs/bcachefs/sb-members.h b/fs/bcachefs/sb-members.h
index be0a94183271..8d8a8a857648 100644
--- a/fs/bcachefs/sb-members.h
+++ b/fs/bcachefs/sb-members.h
@@ -3,6 +3,8 @@
 #define _BCACHEFS_SB_MEMBERS_H
 
 #include "darray.h"
+#include "bkey_types.h"
+#include "enumerated_ref.h"
 
 extern char * const bch2_member_error_strs[];
 
@@ -19,26 +21,22 @@ struct bch_member bch2_sb_member_get(struct bch_sb *sb, int i);
 
 static inline bool bch2_dev_is_online(struct bch_dev *ca)
 {
-	return !percpu_ref_is_zero(&ca->io_ref);
+	return !enumerated_ref_is_zero(&ca->io_ref[READ]);
 }
 
-static inline bool bch2_dev_is_readable(struct bch_dev *ca)
+static inline struct bch_dev *bch2_dev_rcu(struct bch_fs *, unsigned);
+
+static inline bool bch2_dev_idx_is_online(struct bch_fs *c, unsigned dev)
 {
-	return bch2_dev_is_online(ca) &&
-		ca->mi.state != BCH_MEMBER_STATE_failed;
+	guard(rcu)();
+	struct bch_dev *ca = bch2_dev_rcu(c, dev);
+	return ca && bch2_dev_is_online(ca);
 }
 
-static inline bool bch2_dev_get_ioref(struct bch_dev *ca, int rw)
+static inline bool bch2_dev_is_healthy(struct bch_dev *ca)
 {
-	if (!percpu_ref_tryget(&ca->io_ref))
-		return false;
-
-	if (ca->mi.state == BCH_MEMBER_STATE_rw ||
-	    (ca->mi.state == BCH_MEMBER_STATE_ro && rw == READ))
-		return true;
-
-	percpu_ref_put(&ca->io_ref);
-	return false;
+	return bch2_dev_is_online(ca) &&
+		ca->mi.state != BCH_MEMBER_STATE_failed;
 }
 
 static inline unsigned dev_mask_nr(const struct bch_devs_mask *devs)
@@ -104,16 +102,47 @@ static inline struct bch_dev *__bch2_next_dev(struct bch_fs *c, struct bch_dev *
 	for (struct bch_dev *_ca = NULL;				\
 	     (_ca = __bch2_next_dev((_c), _ca, (_mask)));)
 
-static inline struct bch_dev *bch2_get_next_dev(struct bch_fs *c, struct bch_dev *ca)
+#define for_each_online_member_rcu(_c, _ca)				\
+	for_each_member_device_rcu(_c, _ca, &(_c)->online_devs)
+
+#define for_each_rw_member_rcu(_c, _ca)					\
+	for_each_member_device_rcu(_c, _ca, &(_c)->rw_devs[BCH_DATA_free])
+
+static inline void bch2_dev_get(struct bch_dev *ca)
+{
+#ifdef CONFIG_BCACHEFS_DEBUG
+	BUG_ON(atomic_long_inc_return(&ca->ref) <= 1L);
+#else
+	percpu_ref_get(&ca->ref);
+#endif
+}
+
+static inline void __bch2_dev_put(struct bch_dev *ca)
+{
+#ifdef CONFIG_BCACHEFS_DEBUG
+	long r = atomic_long_dec_return(&ca->ref);
+	if (r < (long) !ca->dying)
+		panic("bch_dev->ref underflow, last put: %pS\n", (void *) ca->last_put);
+	ca->last_put = _THIS_IP_;
+	if (!r)
+		complete(&ca->ref_completion);
+#else
+	percpu_ref_put(&ca->ref);
+#endif
+}
+
+static inline void bch2_dev_put(struct bch_dev *ca)
 {
 	if (ca)
-		percpu_ref_put(&ca->ref);
+		__bch2_dev_put(ca);
+}
 
-	rcu_read_lock();
+static inline struct bch_dev *bch2_get_next_dev(struct bch_fs *c, struct bch_dev *ca)
+{
+	guard(rcu)();
+	bch2_dev_put(ca);
 	if ((ca = __bch2_next_dev(c, ca, NULL)))
-		percpu_ref_get(&ca->ref);
-	rcu_read_unlock();
-
+		bch2_dev_get(ca);
 	return ca;
 }
 
@@ -129,86 +158,179 @@ static inline struct bch_dev *bch2_get_next_dev(struct bch_fs *c, struct bch_dev
 
 static inline struct bch_dev *bch2_get_next_online_dev(struct bch_fs *c,
 						       struct bch_dev *ca,
-						       unsigned state_mask)
+						       unsigned state_mask,
+						       int rw, unsigned ref_idx)
 {
+	guard(rcu)();
 	if (ca)
-		percpu_ref_put(&ca->io_ref);
+		enumerated_ref_put(&ca->io_ref[rw], ref_idx);
 
-	rcu_read_lock();
 	while ((ca = __bch2_next_dev(c, ca, NULL)) &&
 	       (!((1 << ca->mi.state) & state_mask) ||
-		!percpu_ref_tryget(&ca->io_ref)))
+		!enumerated_ref_tryget(&ca->io_ref[rw], ref_idx)))
 		;
-	rcu_read_unlock();
 
 	return ca;
 }
 
-#define __for_each_online_member(_c, _ca, state_mask)			\
+#define __for_each_online_member(_c, _ca, state_mask, rw, ref_idx)	\
 	for (struct bch_dev *_ca = NULL;				\
-	     (_ca = bch2_get_next_online_dev(_c, _ca, state_mask));)
+	     (_ca = bch2_get_next_online_dev(_c, _ca, state_mask, rw, ref_idx));)
 
-#define for_each_online_member(c, ca)					\
-	__for_each_online_member(c, ca, ~0)
+#define for_each_online_member(c, ca, ref_idx)				\
+	__for_each_online_member(c, ca, ~0, READ, ref_idx)
 
-#define for_each_rw_member(c, ca)					\
-	__for_each_online_member(c, ca, BIT(BCH_MEMBER_STATE_rw))
+#define for_each_rw_member(c, ca, ref_idx)					\
+	__for_each_online_member(c, ca, BIT(BCH_MEMBER_STATE_rw), WRITE, ref_idx)
 
-#define for_each_readable_member(c, ca)				\
-	__for_each_online_member(c, ca,	BIT( BCH_MEMBER_STATE_rw)|BIT(BCH_MEMBER_STATE_ro))
+#define for_each_readable_member(c, ca, ref_idx)				\
+	__for_each_online_member(c, ca,	BIT( BCH_MEMBER_STATE_rw)|BIT(BCH_MEMBER_STATE_ro), READ, ref_idx)
 
-/*
- * If a key exists that references a device, the device won't be going away and
- * we can omit rcu_read_lock():
- */
-static inline struct bch_dev *bch_dev_bkey_exists(const struct bch_fs *c, unsigned idx)
+static inline bool bch2_dev_exists(const struct bch_fs *c, unsigned dev)
+{
+	return dev < c->sb.nr_devices && c->devs[dev];
+}
+
+static inline bool bucket_valid(const struct bch_dev *ca, u64 b)
+{
+	return b - ca->mi.first_bucket < ca->mi.nbuckets_minus_first;
+}
+
+static inline struct bch_dev *bch2_dev_have_ref(const struct bch_fs *c, unsigned dev)
 {
-	EBUG_ON(idx >= c->sb.nr_devices || !c->devs[idx]);
+	EBUG_ON(!bch2_dev_exists(c, dev));
 
-	return rcu_dereference_check(c->devs[idx], 1);
+	return rcu_dereference_check(c->devs[dev], 1);
 }
 
-static inline struct bch_dev *bch_dev_locked(struct bch_fs *c, unsigned idx)
+static inline struct bch_dev *bch2_dev_locked(struct bch_fs *c, unsigned dev)
 {
-	EBUG_ON(idx >= c->sb.nr_devices || !c->devs[idx]);
+	EBUG_ON(!bch2_dev_exists(c, dev));
 
-	return rcu_dereference_protected(c->devs[idx],
+	return rcu_dereference_protected(c->devs[dev],
 					 lockdep_is_held(&c->sb_lock) ||
 					 lockdep_is_held(&c->state_lock));
 }
 
-/* XXX kill, move to struct bch_fs */
-static inline struct bch_devs_mask bch2_online_devs(struct bch_fs *c)
+static inline struct bch_dev *bch2_dev_rcu_noerror(struct bch_fs *c, unsigned dev)
+{
+	return c && dev < c->sb.nr_devices
+		? rcu_dereference(c->devs[dev])
+		: NULL;
+}
+
+int bch2_dev_missing_bkey(struct bch_fs *, struct bkey_s_c, unsigned);
+
+void bch2_dev_missing_atomic(struct bch_fs *, unsigned);
+
+static inline struct bch_dev *bch2_dev_rcu(struct bch_fs *c, unsigned dev)
+{
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, dev);
+	if (unlikely(!ca))
+		bch2_dev_missing_atomic(c, dev);
+	return ca;
+}
+
+static inline struct bch_dev *bch2_dev_tryget_noerror(struct bch_fs *c, unsigned dev)
+{
+	guard(rcu)();
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, dev);
+	if (ca)
+		bch2_dev_get(ca);
+	return ca;
+}
+
+static inline struct bch_dev *bch2_dev_tryget(struct bch_fs *c, unsigned dev)
+{
+	struct bch_dev *ca = bch2_dev_tryget_noerror(c, dev);
+	if (unlikely(!ca))
+		bch2_dev_missing_atomic(c, dev);
+	return ca;
+}
+
+static inline struct bch_dev *bch2_dev_bucket_tryget_noerror(struct bch_fs *c, struct bpos bucket)
+{
+	struct bch_dev *ca = bch2_dev_tryget_noerror(c, bucket.inode);
+	if (ca && unlikely(!bucket_valid(ca, bucket.offset))) {
+		bch2_dev_put(ca);
+		ca = NULL;
+	}
+	return ca;
+}
+
+void bch2_dev_bucket_missing(struct bch_dev *, u64);
+
+static inline struct bch_dev *bch2_dev_bucket_tryget(struct bch_fs *c, struct bpos bucket)
+{
+	struct bch_dev *ca = bch2_dev_tryget(c, bucket.inode);
+	if (ca && unlikely(!bucket_valid(ca, bucket.offset))) {
+		bch2_dev_bucket_missing(ca, bucket.offset);
+		bch2_dev_put(ca);
+		ca = NULL;
+	}
+	return ca;
+}
+
+static inline struct bch_dev *bch2_dev_iterate_noerror(struct bch_fs *c, struct bch_dev *ca, unsigned dev_idx)
+{
+	if (ca && ca->dev_idx == dev_idx)
+		return ca;
+	bch2_dev_put(ca);
+	return bch2_dev_tryget_noerror(c, dev_idx);
+}
+
+static inline struct bch_dev *bch2_dev_iterate(struct bch_fs *c, struct bch_dev *ca, unsigned dev_idx)
+{
+	if (ca && ca->dev_idx == dev_idx)
+		return ca;
+	bch2_dev_put(ca);
+	return bch2_dev_tryget(c, dev_idx);
+}
+
+static inline struct bch_dev *bch2_dev_get_ioref(struct bch_fs *c, unsigned dev,
+						 int rw, unsigned ref_idx)
 {
-	struct bch_devs_mask devs;
+	might_sleep();
+
+	guard(rcu)();
+	struct bch_dev *ca = bch2_dev_rcu(c, dev);
+	if (!ca || !enumerated_ref_tryget(&ca->io_ref[rw], ref_idx))
+		return NULL;
+
+	if (ca->mi.state == BCH_MEMBER_STATE_rw ||
+	    (ca->mi.state == BCH_MEMBER_STATE_ro && rw == READ))
+		return ca;
 
-	memset(&devs, 0, sizeof(devs));
-	for_each_online_member(c, ca)
-		__set_bit(ca->dev_idx, devs.d);
-	return devs;
+	enumerated_ref_put(&ca->io_ref[rw], ref_idx);
+	return NULL;
 }
 
 extern const struct bch_sb_field_ops bch_sb_field_ops_members_v1;
 extern const struct bch_sb_field_ops bch_sb_field_ops_members_v2;
 
-static inline bool bch2_member_exists(struct bch_member *m)
+static inline bool bch2_member_alive(struct bch_member *m)
 {
-	return !bch2_is_zero(&m->uuid, sizeof(m->uuid));
+	return  !bch2_is_zero(&m->uuid, sizeof(m->uuid)) &&
+		!uuid_equal(&m->uuid, &BCH_SB_MEMBER_DELETED_UUID);
 }
 
-static inline bool bch2_dev_exists(struct bch_sb *sb, unsigned dev)
+static inline bool bch2_member_exists(struct bch_sb *sb, unsigned dev)
 {
 	if (dev < sb->nr_devices) {
 		struct bch_member m = bch2_sb_member_get(sb, dev);
-		return bch2_member_exists(&m);
+		return bch2_member_alive(&m);
 	}
 	return false;
 }
 
+unsigned bch2_sb_nr_devices(const struct bch_sb *);
+
 static inline struct bch_member_cpu bch2_mi_to_cpu(struct bch_member *mi)
 {
 	return (struct bch_member_cpu) {
 		.nbuckets	= le64_to_cpu(mi->nbuckets),
+		.nbuckets_minus_first = le64_to_cpu(mi->nbuckets) -
+			le16_to_cpu(mi->first_bucket),
 		.first_bucket	= le16_to_cpu(mi->first_bucket),
 		.bucket_size	= le16_to_cpu(mi->bucket_size),
 		.group		= BCH_MEMBER_GROUP(mi),
@@ -219,7 +341,10 @@ static inline struct bch_member_cpu bch2_mi_to_cpu(struct bch_member *mi)
 			? BCH_MEMBER_DURABILITY(mi) - 1
 			: 1,
 		.freespace_initialized = BCH_MEMBER_FREESPACE_INITIALIZED(mi),
-		.valid		= bch2_member_exists(mi),
+		.resize_on_mount	= BCH_MEMBER_RESIZE_ON_MOUNT(mi),
+		.valid		= bch2_member_alive(mi),
+		.btree_bitmap_shift	= mi->btree_bitmap_shift,
+		.btree_allocated_bitmap = le64_to_cpu(mi->btree_allocated_bitmap),
 	};
 }
 
@@ -228,4 +353,25 @@ void bch2_sb_members_from_cpu(struct bch_fs *);
 void bch2_dev_io_errors_to_text(struct printbuf *, struct bch_dev *);
 void bch2_dev_errors_reset(struct bch_dev *);
 
+static inline bool bch2_dev_btree_bitmap_marked_sectors(struct bch_dev *ca, u64 start, unsigned sectors)
+{
+	u64 end = start + sectors;
+
+	if (end > 64ULL << ca->mi.btree_bitmap_shift)
+		return false;
+
+	for (unsigned bit = start >> ca->mi.btree_bitmap_shift;
+	     (u64) bit << ca->mi.btree_bitmap_shift < end;
+	     bit++)
+		if (!(ca->mi.btree_allocated_bitmap & BIT_ULL(bit)))
+			return false;
+	return true;
+}
+
+bool bch2_dev_btree_bitmap_marked(struct bch_fs *, struct bkey_s_c);
+void bch2_dev_btree_bitmap_mark(struct bch_fs *, struct bkey_s_c);
+
+int bch2_sb_member_alloc(struct bch_fs *);
+void bch2_sb_members_clean_deleted(struct bch_fs *);
+
 #endif /* _BCACHEFS_SB_MEMBERS_H */