bcachefs: Add new alloc fields

prep work for persistent alloc info

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

5 files changed, 166 insertions, 104 deletions

diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index 885aff511f97..bd3070539e28 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -23,6 +23,13 @@
 #include <linux/sched/task.h>
 #include <linux/sort.h>
 
+static const char * const bch2_alloc_field_names[] = {
+#define x(name, bytes) #name,
+	BCH_ALLOC_FIELDS()
+#undef x
+	NULL
+};
+
 static void bch2_recalc_oldest_io(struct bch_fs *, struct bch_dev *, int);
 
 /* Ratelimiting/PD controllers */
@@ -62,14 +69,73 @@ static void pd_controllers_update(struct work_struct *work)
 
 /* Persistent alloc info: */
 
+static inline u64 get_alloc_field(const struct bch_alloc *a,
+				  const void **p, unsigned field)
+{
+	unsigned bytes = BCH_ALLOC_FIELD_BYTES[field];
+	u64 v;
+
+	if (!(a->fields & (1 << field)))
+		return 0;
+
+	switch (bytes) {
+	case 1:
+		v = *((const u8 *) *p);
+		break;
+	case 2:
+		v = le16_to_cpup(*p);
+		break;
+	case 4:
+		v = le32_to_cpup(*p);
+		break;
+	case 8:
+		v = le64_to_cpup(*p);
+		break;
+	default:
+		BUG();
+	}
+
+	*p += bytes;
+	return v;
+}
+
+static inline void put_alloc_field(struct bkey_i_alloc *a, void **p,
+				   unsigned field, u64 v)
+{
+	unsigned bytes = BCH_ALLOC_FIELD_BYTES[field];
+
+	if (!v)
+		return;
+
+	a->v.fields |= 1 << field;
+
+	switch (bytes) {
+	case 1:
+		*((u8 *) *p) = v;
+		break;
+	case 2:
+		*((__le16 *) *p) = cpu_to_le16(v);
+		break;
+	case 4:
+		*((__le32 *) *p) = cpu_to_le32(v);
+		break;
+	case 8:
+		*((__le64 *) *p) = cpu_to_le64(v);
+		break;
+	default:
+		BUG();
+	}
+
+	*p += bytes;
+}
+
 static unsigned bch_alloc_val_u64s(const struct bch_alloc *a)
 {
-	unsigned bytes = offsetof(struct bch_alloc, data);
+	unsigned i, bytes = offsetof(struct bch_alloc, data);
 
-	if (a->fields & (1 << BCH_ALLOC_FIELD_READ_TIME))
-		bytes += 2;
-	if (a->fields & (1 << BCH_ALLOC_FIELD_WRITE_TIME))
-		bytes += 2;
+	for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_FIELD_BYTES); i++)
+		if (a->fields & (1 << i))
+			bytes += BCH_ALLOC_FIELD_BYTES[i];
 
 	return DIV_ROUND_UP(bytes, sizeof(u64));
 }
@@ -93,58 +159,55 @@ void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c,
 			struct bkey_s_c k)
 {
 	struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
+	const void *d = a.v->data;
+	unsigned i;
 
 	pr_buf(out, "gen %u", a.v->gen);
+
+	for (i = 0; i < BCH_ALLOC_FIELD_NR; i++)
+		if (a.v->fields & (1 << i))
+			pr_buf(out, " %s %llu",
+			       bch2_alloc_field_names[i],
+			       get_alloc_field(a.v, &d, i));
 }
 
-static inline unsigned get_alloc_field(const u8 **p, unsigned bytes)
+static void __alloc_read_key(struct bucket *g, const struct bch_alloc *a)
 {
-	unsigned v;
-
-	switch (bytes) {
-	case 1:
-		v = **p;
-		break;
-	case 2:
-		v = le16_to_cpup((void *) *p);
-		break;
-	case 4:
-		v = le32_to_cpup((void *) *p);
-		break;
-	default:
-		BUG();
-	}
-
-	*p += bytes;
-	return v;
+	const void *d = a->data;
+	unsigned idx = 0;
+
+	g->_mark.gen		= a->gen;
+	g->gen_valid		= 1;
+	g->io_time[READ]	= get_alloc_field(a, &d, idx++);
+	g->io_time[WRITE]	= get_alloc_field(a, &d, idx++);
+	g->_mark.data_type	= get_alloc_field(a, &d, idx++);
+	g->_mark.dirty_sectors	= get_alloc_field(a, &d, idx++);
+	g->_mark.cached_sectors	= get_alloc_field(a, &d, idx++);
 }
 
-static inline void put_alloc_field(u8 **p, unsigned bytes, unsigned v)
+static void __alloc_write_key(struct bkey_i_alloc *a, struct bucket *g)
 {
-	switch (bytes) {
-	case 1:
-		**p = v;
-		break;
-	case 2:
-		*((__le16 *) *p) = cpu_to_le16(v);
-		break;
-	case 4:
-		*((__le32 *) *p) = cpu_to_le32(v);
-		break;
-	default:
-		BUG();
-	}
+	struct bucket_mark m = READ_ONCE(g->mark);
+	unsigned idx = 0;
+	void *d = a->v.data;
 
-	*p += bytes;
+	a->v.fields	= 0;
+	a->v.gen	= m.gen;
+
+	d = a->v.data;
+	put_alloc_field(a, &d, idx++, g->io_time[READ]);
+	put_alloc_field(a, &d, idx++, g->io_time[WRITE]);
+	put_alloc_field(a, &d, idx++, m.data_type);
+	put_alloc_field(a, &d, idx++, m.dirty_sectors);
+	put_alloc_field(a, &d, idx++, m.cached_sectors);
+
+	set_bkey_val_bytes(&a->k, (void *) d - (void *) &a->v);
 }
 
 static void bch2_alloc_read_key(struct bch_fs *c, struct bkey_s_c k)
 {
 	struct bch_dev *ca;
 	struct bkey_s_c_alloc a;
-	struct bucket_mark new;
-	struct bucket *g;
-	const u8 *d;
 
 	if (k.k->type != KEY_TYPE_alloc)
 		return;
@@ -156,19 +219,7 @@ static void bch2_alloc_read_key(struct bch_fs *c, struct bkey_s_c k)
 		return;
 
 	percpu_down_read(&c->usage_lock);
-
-	g = bucket(ca, a.k->p.offset);
-	bucket_cmpxchg(g, new, ({
-		new.gen = a.v->gen;
-		new.gen_valid = 1;
-	}));
-
-	d = a.v->data;
-	if (a.v->fields & (1 << BCH_ALLOC_FIELD_READ_TIME))
-		g->io_time[READ] = get_alloc_field(&d, 2);
-	if (a.v->fields & (1 << BCH_ALLOC_FIELD_WRITE_TIME))
-		g->io_time[WRITE] = get_alloc_field(&d, 2);
-
+	__alloc_read_key(bucket(ca, a.k->p.offset), a.v);
 	percpu_up_read(&c->usage_lock);
 }
 
@@ -222,28 +273,21 @@ static int __bch2_alloc_write_key(struct bch_fs *c, struct bch_dev *ca,
 				  size_t b, struct btree_iter *iter,
 				  u64 *journal_seq, unsigned flags)
 {
-	struct bucket_mark m;
-	__BKEY_PADDED(k, DIV_ROUND_UP(sizeof(struct bch_alloc), 8)) alloc_key;
-	struct bucket *g;
-	struct bkey_i_alloc *a;
+#if 0
+	__BKEY_PADDED(k, BKEY_ALLOC_VAL_U64s_MAX) alloc_key;
+#else
+	/* hack: */
+	__BKEY_PADDED(k, 8) alloc_key;
+#endif
+	struct bkey_i_alloc *a = bkey_alloc_init(&alloc_key.k);
 	int ret;
-	u8 *d;
 
-	percpu_down_read(&c->usage_lock);
-	g = bucket(ca, b);
+	BUG_ON(BKEY_ALLOC_VAL_U64s_MAX > 8);
 
-	m = READ_ONCE(g->mark);
-	a = bkey_alloc_init(&alloc_key.k);
-	a->k.p		= POS(ca->dev_idx, b);
-	a->v.fields	= 0;
-	a->v.gen	= m.gen;
-	set_bkey_val_u64s(&a->k, bch_alloc_val_u64s(&a->v));
+	a->k.p = POS(ca->dev_idx, b);
 
-	d = a->v.data;
-	if (a->v.fields & (1 << BCH_ALLOC_FIELD_READ_TIME))
-		put_alloc_field(&d, 2, g->io_time[READ]);
-	if (a->v.fields & (1 << BCH_ALLOC_FIELD_WRITE_TIME))
-		put_alloc_field(&d, 2, g->io_time[WRITE]);
+	percpu_down_read(&c->usage_lock);
+	__alloc_write_key(a, bucket(ca, b));
 	percpu_up_read(&c->usage_lock);
 
 	bch2_btree_iter_cond_resched(iter);
@@ -1295,7 +1339,7 @@ static int __bch2_fs_allocator_start(struct bch_fs *c)
 		     bu < buckets->nbuckets; bu++) {
 			m = READ_ONCE(buckets->b[bu].mark);
 
-			if (!m.gen_valid ||
+			if (!buckets->b[bu].gen_valid ||
 			    !is_available_bucket(m) ||
 			    m.cached_sectors)
 				continue;
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index d774ddf6cbb3..e9ae7e6d53b2 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -619,8 +619,6 @@ struct bch_fs {
 
 	struct percpu_rw_semaphore usage_lock;
 
-	struct closure_waitlist	freelist_wait;
-
 	/*
 	 * When we invalidate buckets, we use both the priority and the amount
 	 * of good data to determine which buckets to reuse first - to weight
@@ -633,6 +631,7 @@ struct bch_fs {
 
 	/* ALLOCATOR */
 	spinlock_t		freelist_lock;
+	struct closure_waitlist	freelist_wait;
 	u8			open_buckets_freelist;
 	u8			open_buckets_nr_free;
 	struct closure_waitlist	open_buckets_wait;
diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h
index 801156b74335..162a0a307f1b 100644
--- a/fs/bcachefs/bcachefs_format.h
+++ b/fs/bcachefs/bcachefs_format.h
@@ -807,11 +807,6 @@ struct bch_xattr {
 
 /* Bucket/allocation information: */
 
-enum {
-	BCH_ALLOC_FIELD_READ_TIME	= 0,
-	BCH_ALLOC_FIELD_WRITE_TIME	= 1,
-};
-
 struct bch_alloc {
 	struct bch_val		v;
 	__u8			fields;
@@ -819,6 +814,32 @@ struct bch_alloc {
 	__u8			data[];
 } __attribute__((packed, aligned(8)));
 
+#define BCH_ALLOC_FIELDS()			\
+	x(read_time, 2)				\
+	x(write_time, 2)			\
+	x(data_type, 1)				\
+	x(dirty_sectors, 2)			\
+	x(cached_sectors, 2)
+
+enum {
+#define x(name, bytes) BCH_ALLOC_FIELD_##name,
+	BCH_ALLOC_FIELDS()
+#undef x
+	BCH_ALLOC_FIELD_NR
+};
+
+static const unsigned BCH_ALLOC_FIELD_BYTES[] = {
+#define x(name, bytes) [BCH_ALLOC_FIELD_##name] = bytes,
+	BCH_ALLOC_FIELDS()
+#undef x
+};
+
+#define x(name, bytes) + bytes
+static const unsigned BKEY_ALLOC_VAL_U64s_MAX =
+	DIV_ROUND_UP(offsetof(struct bch_alloc, data)
+		     BCH_ALLOC_FIELDS(), sizeof(u64));
+#undef x
+
 /* Quotas: */
 
 enum quota_types {
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index 65cf64f22522..997e72aa4cb1 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -144,12 +144,12 @@ static int bch2_gc_mark_key(struct bch_fs *c, struct bkey_s_c k,
 			size_t b = PTR_BUCKET_NR(ca, ptr);
 			struct bucket *g = PTR_BUCKET(ca, ptr);
 
-			if (mustfix_fsck_err_on(!g->mark.gen_valid, c,
+			if (mustfix_fsck_err_on(!g->gen_valid, c,
 					"found ptr with missing gen in alloc btree,\n"
 					"type %u gen %u",
 					k.k->type, ptr->gen)) {
 				g->_mark.gen = ptr->gen;
-				g->_mark.gen_valid = 1;
+				g->gen_valid = 1;
 				set_bit(b, ca->buckets_dirty);
 			}
 
@@ -157,7 +157,7 @@ static int bch2_gc_mark_key(struct bch_fs *c, struct bkey_s_c k,
 					"%u ptr gen in the future: %u > %u",
 					k.k->type, ptr->gen, g->mark.gen)) {
 				g->_mark.gen = ptr->gen;
-				g->_mark.gen_valid = 1;
+				g->gen_valid = 1;
 				set_bit(b, ca->buckets_dirty);
 				set_bit(BCH_FS_FIXED_GENS, &c->flags);
 			}
diff --git a/fs/bcachefs/buckets_types.h b/fs/bcachefs/buckets_types.h
index 9f7812c69bbc..a47a7856eee4 100644
--- a/fs/bcachefs/buckets_types.h
+++ b/fs/bcachefs/buckets_types.h
@@ -9,28 +9,25 @@
 
 struct bucket_mark {
 	union {
-	struct {
-		atomic64_t	v;
-	};
+	atomic64_t	v;
 
 	struct {
-		u8		gen;
-		u8		data_type:3,
-				gen_valid:1,
-				owned_by_allocator:1,
-				nouse:1,
-				journal_seq_valid:1,
-				stripe:1;
-		u16		dirty_sectors;
-		u16		cached_sectors;
-
-		/*
-		 * low bits of journal sequence number when this bucket was most
-		 * recently modified: if journal_seq_valid is set, this bucket
-		 * can't be reused until the journal sequence number written to
-		 * disk is >= the bucket's journal sequence number:
-		 */
-		u16		journal_seq;
+	u8		gen;
+	u8		data_type:3,
+			owned_by_allocator:1,
+			nouse:1,
+			journal_seq_valid:1,
+			stripe:1;
+	u16		dirty_sectors;
+	u16		cached_sectors;
+
+	/*
+	 * low bits of journal sequence number when this bucket was most
+	 * recently modified: if journal_seq_valid is set, this bucket can't be
+	 * reused until the journal sequence number written to disk is >= the
+	 * bucket's journal sequence number:
+	 */
+	u16		journal_seq;
 	};
 	};
 };
@@ -42,6 +39,7 @@ struct bucket {
 	};
 
 	u16				io_time[2];
+	unsigned			gen_valid:1;
 };
 
 struct bucket_array {