bcachefs: LRU btree

This implements new persistent LRUs, to be used for buckets containing
cached data, as well as stripes ordered by time when a block became
empty.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

fs/bcachefs/Makefile

@@ -41,6 +41,7 @@ bcachefs-y		:=	\
 	journal_sb.o		\
 	journal_seq_blacklist.o	\
 	keylist.o		\
+	lru.o			\
 	migrate.o		\
 	move.o			\
 	movinggc.o		\

fs/bcachefs/bcachefs.h

@@ -391,6 +391,7 @@ enum gc_phase {
 	GC_PHASE_BTREE_reflink,
 	GC_PHASE_BTREE_subvolumes,
 	GC_PHASE_BTREE_snapshots,
+	GC_PHASE_BTREE_lru,
 
 	GC_PHASE_PENDING_DELETE,
 };

fs/bcachefs/bcachefs_format.h

@@ -352,7 +352,8 @@ static inline void bkey_init(struct bkey *k)
 	x(snapshot,		22)			\
 	x(inode_v2,		23)			\
 	x(alloc_v3,		24)			\
-	x(set,			25)
+	x(set,			25)			\
+	x(lru,			26)
 
 enum bch_bkey_type {
 #define x(name, nr) KEY_TYPE_##name	= nr,
@@ -1024,6 +1025,15 @@ LE32_BITMASK(BCH_SNAPSHOT_DELETED,	struct bch_snapshot, flags,  0,  1)
 /* True if a subvolume points to this snapshot node: */
 LE32_BITMASK(BCH_SNAPSHOT_SUBVOL,	struct bch_snapshot, flags,  1,  2)
 
+/* LRU btree: */
+
+struct bch_lru {
+	struct bch_val		v;
+	__le64			idx;
+} __attribute__((packed, aligned(8)));
+
+#define LRU_ID_STRIPES		(1U << 16)
+
 /* Optional/variable size superblock sections: */
 
 struct bch_sb_field {
@@ -1838,7 +1848,8 @@ LE32_BITMASK(JSET_NO_FLUSH,	struct jset, flags, 5, 6);
 	x(stripes,	6)			\
 	x(reflink,	7)			\
 	x(subvolumes,	8)			\
-	x(snapshots,	9)
+	x(snapshots,	9)			\
+	x(lru,		10)
 
 enum btree_id {
 #define x(kwd, val) BTREE_ID_##kwd = val,

fs/bcachefs/bkey_methods.c

@@ -9,6 +9,7 @@
 #include "error.h"
 #include "extents.h"
 #include "inode.h"
+#include "lru.h"
 #include "quota.h"
 #include "reflink.h"
 #include "subvolume.h"
@@ -165,6 +166,9 @@ static unsigned bch2_key_types_allowed[] = {
 	[BKEY_TYPE_snapshots] =
 		(1U << KEY_TYPE_deleted)|
 		(1U << KEY_TYPE_snapshot),
+	[BKEY_TYPE_lru] =
+		(1U << KEY_TYPE_deleted)|
+		(1U << KEY_TYPE_lru),
 	[BKEY_TYPE_btree] =
 		(1U << KEY_TYPE_deleted)|
 		(1U << KEY_TYPE_btree_ptr)|

fs/bcachefs/lru.c

@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "btree_iter.h"
+#include "btree_update.h"
+#include "error.h"
+#include "lru.h"
+
+const char *bch2_lru_invalid(const struct bch_fs *c, struct bkey_s_c k)
+{
+	const struct bch_lru *lru = bkey_s_c_to_lru(k).v;
+
+	if (bkey_val_bytes(k.k) < sizeof(*lru))
+		return "incorrect value size";
+
+	return NULL;
+}
+
+void bch2_lru_to_text(struct printbuf *out, struct bch_fs *c,
+		      struct bkey_s_c k)
+{
+	const struct bch_lru *lru = bkey_s_c_to_lru(k).v;
+
+	pr_buf(out, "idx %llu", le64_to_cpu(lru->idx));
+}
+
+static int lru_delete(struct btree_trans *trans, u64 id, u64 idx, u64 time)
+{
+	struct bch_fs *c = trans->c;
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	u64 existing_idx;
+	int ret = 0;
+
+	if (!time)
+		return 0;
+
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_lru,
+			     POS(id, time),
+			     BTREE_ITER_INTENT|
+			     BTREE_ITER_WITH_UPDATES);
+	k = bch2_btree_iter_peek_slot(&iter);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
+
+	if (k.k->type != KEY_TYPE_lru) {
+		bch2_fs_inconsistent(c,
+			"pointer to nonexistent lru %llu:%llu",
+			id, time);
+		ret = -EIO;
+		goto err;
+	}
+
+	existing_idx = le64_to_cpu(bkey_s_c_to_lru(k).v->idx);
+	if (existing_idx != idx) {
+		bch2_fs_inconsistent(c,
+			"lru %llu:%llu with wrong backpointer: got %llu, should be %llu",
+			id, time, existing_idx, idx);
+		ret = -EIO;
+		goto err;
+	}
+
+	ret = bch2_btree_delete_at(trans, &iter, 0);
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+static int lru_set(struct btree_trans *trans, u64 lru_id, u64 idx, u64 *time)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	struct bkey_i_lru *lru;
+	int ret = 0;
+
+	if (!*time)
+		return 0;
+
+	for_each_btree_key_norestart(trans, iter, BTREE_ID_lru,
+			POS(lru_id, *time),
+			BTREE_ITER_SLOTS|
+			BTREE_ITER_INTENT|
+			BTREE_ITER_WITH_UPDATES, k, ret)
+		if (bkey_deleted(k.k))
+			break;
+
+	if (ret)
+		goto err;
+
+	BUG_ON(iter.pos.inode != lru_id);
+	*time = iter.pos.offset;
+
+	lru = bch2_trans_kmalloc(trans, sizeof(*lru));
+	ret = PTR_ERR_OR_ZERO(lru);
+	if (ret)
+		goto err;
+
+	bkey_lru_init(&lru->k_i);
+	lru->k.p	= iter.pos;
+	lru->v.idx	= cpu_to_le64(idx);
+
+	ret = bch2_trans_update(trans, &iter, &lru->k_i, 0);
+	if (ret)
+		goto err;
+err:
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+int bch2_lru_change(struct btree_trans *trans, u64 id, u64 idx,
+		    u64 old_time, u64 *new_time)
+{
+	if (old_time == *new_time)
+		return 0;
+
+	return  lru_delete(trans, id, idx, old_time) ?:
+		lru_set(trans, id, idx, new_time);
+}

fs/bcachefs/lru.h

@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_LRU_H
+#define _BCACHEFS_LRU_H
+
+const char *bch2_lru_invalid(const struct bch_fs *, struct bkey_s_c);
+void bch2_lru_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+
+#define bch2_bkey_ops_lru (struct bkey_ops) {	\
+	.key_invalid	= bch2_lru_invalid,	\
+	.val_to_text	= bch2_lru_to_text,	\
+}
+
+int bch2_lru_change(struct btree_trans *, u64, u64, u64, u64 *);
+
+#endif /* _BCACHEFS_LRU_H */