X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Finode.c;h=823a1ddec5aca57983b123fc13df6be2ba940aaa;hb=13f53aa228c83731226f4a359983215f1f7c2a47;hp=8c3d4431de109478905c46a7be67d1962601e4a7;hpb=0c7db4eca3e6519043c10288cb41f8a0ee634a0b;p=bcachefs-tools-debian

diff --git a/libbcachefs/inode.c b/libbcachefs/inode.c
index 8c3d443..823a1dd 100644
--- a/libbcachefs/inode.c
+++ b/libbcachefs/inode.c
@@ -1,18 +1,25 @@
+// SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "btree_key_cache.h"
 #include "bkey_methods.h"
 #include "btree_update.h"
 #include "error.h"
 #include "extents.h"
 #include "inode.h"
-#include "io.h"
-#include "keylist.h"
+#include "str_hash.h"
+#include "varint.h"
 
 #include <linux/random.h>
 
 #include <asm/unaligned.h>
 
-#define FIELD_BYTES()						\
+const char * const bch2_inode_opts[] = {
+#define x(name, ...)	#name,
+	BCH_INODE_OPTS()
+#undef  x
+	NULL,
+};
 
 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
 static const u8 bits_table[8] = {
@@ -83,21 +90,17 @@ static int inode_decode_field(const u8 *in, const u8 *end,
 	return bytes;
 }
 
-void bch2_inode_pack(struct bkey_inode_buf *packed,
-		     const struct bch_inode_unpacked *inode)
+static noinline void bch2_inode_pack_v1(struct bkey_inode_buf *packed,
+					const struct bch_inode_unpacked *inode)
 {
-	u8 *out = packed->inode.v.fields;
+	struct bkey_i_inode *k = &packed->inode;
+	u8 *out = k->v.fields;
 	u8 *end = (void *) &packed[1];
 	u8 *last_nonzero_field = out;
 	unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
+	unsigned bytes;
 
-	bkey_inode_init(&packed->inode.k_i);
-	packed->inode.k.p.inode		= inode->bi_inum;
-	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
-	packed->inode.v.bi_flags	= cpu_to_le32(inode->bi_flags);
-	packed->inode.v.bi_mode		= cpu_to_le16(inode->bi_mode);
-
-#define BCH_INODE_FIELD(_name, _bits)					\
+#define x(_name, _bits)							\
 	out += inode_encode_field(out, end, 0, inode->_name);		\
 	nr_fields++;							\
 									\
@@ -107,17 +110,78 @@ void bch2_inode_pack(struct bkey_inode_buf *packed,
 	}
 
 	BCH_INODE_FIELDS()
-#undef  BCH_INODE_FIELD
+#undef  x
 
 	out = last_nonzero_field;
 	nr_fields = last_nonzero_fieldnr;
 
-	set_bkey_val_bytes(&packed->inode.k, out - (u8 *) &packed->inode.v);
-	memset(out, 0,
-	       (u8 *) &packed->inode.v +
-	       bkey_val_bytes(&packed->inode.k) - out);
+	bytes = out - (u8 *) &packed->inode.v;
+	set_bkey_val_bytes(&packed->inode.k, bytes);
+	memset_u64s_tail(&packed->inode.v, 0, bytes);
+
+	SET_INODE_NR_FIELDS(&k->v, nr_fields);
+}
 
-	SET_INODE_NR_FIELDS(&packed->inode.v, nr_fields);
+static void bch2_inode_pack_v2(struct bkey_inode_buf *packed,
+			       const struct bch_inode_unpacked *inode)
+{
+	struct bkey_i_inode *k = &packed->inode;
+	u8 *out = k->v.fields;
+	u8 *end = (void *) &packed[1];
+	u8 *last_nonzero_field = out;
+	unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
+	unsigned bytes;
+	int ret;
+
+#define x(_name, _bits)							\
+	nr_fields++;							\
+									\
+	if (inode->_name) {						\
+		ret = bch2_varint_encode(out, inode->_name);		\
+		out += ret;						\
+									\
+		if (_bits > 64)						\
+			*out++ = 0;					\
+									\
+		last_nonzero_field = out;				\
+		last_nonzero_fieldnr = nr_fields;			\
+	} else {							\
+		*out++ = 0;						\
+									\
+		if (_bits > 64)						\
+			*out++ = 0;					\
+	}
+
+	BCH_INODE_FIELDS()
+#undef  x
+	BUG_ON(out > end);
+
+	out = last_nonzero_field;
+	nr_fields = last_nonzero_fieldnr;
+
+	bytes = out - (u8 *) &packed->inode.v;
+	set_bkey_val_bytes(&packed->inode.k, bytes);
+	memset_u64s_tail(&packed->inode.v, 0, bytes);
+
+	SET_INODE_NR_FIELDS(&k->v, nr_fields);
+}
+
+void bch2_inode_pack(struct bch_fs *c,
+		     struct bkey_inode_buf *packed,
+		     const struct bch_inode_unpacked *inode)
+{
+	bkey_inode_init(&packed->inode.k_i);
+	packed->inode.k.p.offset	= inode->bi_inum;
+	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
+	packed->inode.v.bi_flags	= cpu_to_le32(inode->bi_flags);
+	packed->inode.v.bi_mode		= cpu_to_le16(inode->bi_mode);
+
+	if (c->sb.features & (1ULL << BCH_FEATURE_new_varint)) {
+		SET_INODE_NEW_VARINT(&packed->inode.v, true);
+		bch2_inode_pack_v2(packed, inode);
+	} else {
+		bch2_inode_pack_v1(packed, inode);
+	}
 
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 		struct bch_inode_unpacked unpacked;
@@ -129,27 +193,24 @@ void bch2_inode_pack(struct bkey_inode_buf *packed,
 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
 
-#define BCH_INODE_FIELD(_name, _bits)	BUG_ON(unpacked._name != inode->_name);
+#define x(_name, _bits)	if (unpacked._name != inode->_name)		\
+			panic("unpacked %llu should be %llu",		\
+			      (u64) unpacked._name, (u64) inode->_name);
 		BCH_INODE_FIELDS()
-#undef  BCH_INODE_FIELD
+#undef  x
 	}
 }
 
-int bch2_inode_unpack(struct bkey_s_c_inode inode,
-		      struct bch_inode_unpacked *unpacked)
+static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
+				struct bch_inode_unpacked *unpacked)
 {
 	const u8 *in = inode.v->fields;
-	const u8 *end = (void *) inode.v + bkey_val_bytes(inode.k);
+	const u8 *end = bkey_val_end(inode);
 	u64 field[2];
 	unsigned fieldnr = 0, field_bits;
 	int ret;
 
-	unpacked->bi_inum	= inode.k->p.inode;
-	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
-	unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
-	unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
-
-#define BCH_INODE_FIELD(_name, _bits)					\
+#define x(_name, _bits)					\
 	if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {			\
 		memset(&unpacked->_name, 0,				\
 		       sizeof(*unpacked) -				\
@@ -168,10 +229,113 @@ int bch2_inode_unpack(struct bkey_s_c_inode inode,
 	in += ret;
 
 	BCH_INODE_FIELDS()
-#undef  BCH_INODE_FIELD
+#undef  x
+
+	/* XXX: signal if there were more fields than expected? */
+	return 0;
+}
+
+static int bch2_inode_unpack_v2(struct bkey_s_c_inode inode,
+				struct bch_inode_unpacked *unpacked)
+{
+	const u8 *in = inode.v->fields;
+	const u8 *end = bkey_val_end(inode);
+	unsigned fieldnr = 0;
+	int ret;
+	u64 v[2];
+
+#define x(_name, _bits)							\
+	if (fieldnr < INODE_NR_FIELDS(inode.v)) {			\
+		ret = bch2_varint_decode(in, end, &v[0]);		\
+		if (ret < 0)						\
+			return ret;					\
+		in += ret;						\
+									\
+		if (_bits > 64) {					\
+			ret = bch2_varint_decode(in, end, &v[1]);	\
+			if (ret < 0)					\
+				return ret;				\
+			in += ret;					\
+		} else {						\
+			v[1] = 0;					\
+		}							\
+	} else {							\
+		v[0] = v[1] = 0;					\
+	}								\
+									\
+	unpacked->_name = v[0];						\
+	if (v[1] || v[0] != unpacked->_name)				\
+		return -1;						\
+	fieldnr++;
+
+	BCH_INODE_FIELDS()
+#undef  x
 
 	/* XXX: signal if there were more fields than expected? */
+	return 0;
+}
+
+int bch2_inode_unpack(struct bkey_s_c_inode inode,
+		      struct bch_inode_unpacked *unpacked)
+{
+	unpacked->bi_inum	= inode.k->p.offset;
+	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
+	unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
+	unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
+
+	if (INODE_NEW_VARINT(inode.v)) {
+		return bch2_inode_unpack_v2(inode, unpacked);
+	} else {
+		return bch2_inode_unpack_v1(inode, unpacked);
+	}
+
+	return 0;
+}
+
+struct btree_iter *bch2_inode_peek(struct btree_trans *trans,
+				   struct bch_inode_unpacked *inode,
+				   u64 inum, unsigned flags)
+{
+	struct btree_iter *iter;
+	struct bkey_s_c k;
+	int ret;
+
+	iter = bch2_trans_get_iter(trans, BTREE_ID_INODES, POS(0, inum),
+				   BTREE_ITER_CACHED|flags);
+	if (IS_ERR(iter))
+		return iter;
+
+	k = bch2_btree_iter_peek_cached(iter);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
 
+	ret = k.k->type == KEY_TYPE_inode ? 0 : -EIO;
+	if (ret)
+		goto err;
+
+	ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
+	if (ret)
+		goto err;
+
+	return iter;
+err:
+	bch2_trans_iter_put(trans, iter);
+	return ERR_PTR(ret);
+}
+
+int bch2_inode_write(struct btree_trans *trans,
+		     struct btree_iter *iter,
+		     struct bch_inode_unpacked *inode)
+{
+	struct bkey_inode_buf *inode_p;
+
+	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
+	if (IS_ERR(inode_p))
+		return PTR_ERR(inode_p);
+
+	bch2_inode_pack(trans->c, inode_p, inode);
+	bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 	return 0;
 }
 
@@ -180,13 +344,13 @@ const char *bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k)
 		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
 		struct bch_inode_unpacked unpacked;
 
-	if (k.k->p.offset)
-		return "nonzero offset";
+	if (k.k->p.inode)
+		return "nonzero k.p.inode";
 
 	if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
 		return "incorrect value size";
 
-	if (k.k->p.inode < BLOCKDEV_INODE_MAX)
+	if (k.k->p.offset < BLOCKDEV_INODE_MAX)
 		return "fs inode in blockdev range";
 
 	if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
@@ -219,17 +383,19 @@ void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c,
 		return;
 	}
 
-#define BCH_INODE_FIELD(_name, _bits)						\
+	pr_buf(out, "mode: %o ", unpacked.bi_mode);
+
+#define x(_name, _bits)						\
 	pr_buf(out, #_name ": %llu ", (u64) unpacked._name);
 	BCH_INODE_FIELDS()
-#undef  BCH_INODE_FIELD
+#undef  x
 }
 
 const char *bch2_inode_generation_invalid(const struct bch_fs *c,
 					  struct bkey_s_c k)
 {
-	if (k.k->p.offset)
-		return "nonzero offset";
+	if (k.k->p.inode)
+		return "nonzero k.p.inode";
 
 	if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
 		return "incorrect value size";
@@ -240,22 +406,29 @@ const char *bch2_inode_generation_invalid(const struct bch_fs *c,
 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
 				   struct bkey_s_c k)
 {
+	struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
+
+	pr_buf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
 }
 
-void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
-		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
-		     struct bch_inode_unpacked *parent)
+void bch2_inode_init_early(struct bch_fs *c,
+			   struct bch_inode_unpacked *inode_u)
 {
-	s64 now = timespec_to_bch2_time(c,
-		timespec64_trunc(current_kernel_time64(),
-				 c->sb.time_precision));
+	enum bch_str_hash_type str_hash =
+		bch2_str_hash_opt_to_type(c, c->opts.str_hash);
 
 	memset(inode_u, 0, sizeof(*inode_u));
 
 	/* ick */
-	inode_u->bi_flags |= c->opts.str_hash << INODE_STR_HASH_OFFSET;
-	get_random_bytes(&inode_u->bi_hash_seed, sizeof(inode_u->bi_hash_seed));
+	inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
+	get_random_bytes(&inode_u->bi_hash_seed,
+			 sizeof(inode_u->bi_hash_seed));
+}
 
+void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
+			  uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
+			  struct bch_inode_unpacked *parent)
+{
 	inode_u->bi_mode	= mode;
 	inode_u->bi_uid		= uid;
 	inode_u->bi_gid		= gid;
@@ -265,13 +438,28 @@ void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
 	inode_u->bi_ctime	= now;
 	inode_u->bi_otime	= now;
 
+	if (parent && parent->bi_mode & S_ISGID) {
+		inode_u->bi_gid = parent->bi_gid;
+		if (S_ISDIR(mode))
+			inode_u->bi_mode |= S_ISGID;
+	}
+
 	if (parent) {
-#define BCH_INODE_FIELD(_name)	inode_u->_name = parent->_name;
-		BCH_INODE_FIELDS_INHERIT()
-#undef BCH_INODE_FIELD
+#define x(_name, ...)	inode_u->bi_##_name = parent->bi_##_name;
+		BCH_INODE_OPTS()
+#undef x
 	}
 }
 
+void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
+		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
+		     struct bch_inode_unpacked *parent)
+{
+	bch2_inode_init_early(c, inode_u);
+	bch2_inode_init_late(inode_u, bch2_current_time(c),
+			     uid, gid, mode, rdev, parent);
+}
+
 static inline u32 bkey_generation(struct bkey_s_c k)
 {
 	switch (k.k->type) {
@@ -284,21 +472,25 @@ static inline u32 bkey_generation(struct bkey_s_c k)
 	}
 }
 
-int __bch2_inode_create(struct btree_trans *trans,
-			struct bch_inode_unpacked *inode_u,
-			u64 min, u64 max, u64 *hint)
+int bch2_inode_create(struct btree_trans *trans,
+		      struct bch_inode_unpacked *inode_u)
 {
 	struct bch_fs *c = trans->c;
 	struct bkey_inode_buf *inode_p;
-	struct btree_iter *iter;
-	u64 start;
+	struct btree_iter *iter = NULL;
+	struct bkey_s_c k;
+	u64 min, max, start, *hint;
 	int ret;
 
-	if (!max)
-		max = ULLONG_MAX;
+	unsigned cpu = raw_smp_processor_id();
+	unsigned bits = (c->opts.inodes_32bit
+		? 31 : 63) - c->inode_shard_bits;
+
+	min = (cpu << bits);
+	max = (cpu << bits) | ~(ULLONG_MAX << bits);
 
-	if (c->opts.inodes_32bit)
-		max = min_t(u64, max, U32_MAX);
+	min = max_t(u64, min, BLOCKDEV_INODE_MAX);
+	hint = c->unused_inode_hints + cpu;
 
 	start = READ_ONCE(*hint);
 
@@ -308,64 +500,57 @@ int __bch2_inode_create(struct btree_trans *trans,
 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
 	if (IS_ERR(inode_p))
 		return PTR_ERR(inode_p);
-
-	iter = bch2_trans_get_iter(trans,
-			BTREE_ID_INODES, POS(start, 0),
-			BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
-	if (IS_ERR(iter))
-		return PTR_ERR(iter);
 again:
-	while (1) {
-		struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
-
-		ret = btree_iter_err(k);
-		if (ret)
-			return ret;
+	for_each_btree_key(trans, iter, BTREE_ID_INODES, POS(0, start),
+			   BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
+		if (bkey_cmp(iter->pos, POS(0, max)) > 0)
+			break;
 
-		switch (k.k->type) {
-		case KEY_TYPE_inode:
-			/* slot used */
-			if (iter->pos.inode >= max)
-				goto out;
+		/*
+		 * There's a potential cache coherency issue with the btree key
+		 * cache code here - we're iterating over the btree, skipping
+		 * that cache. We should never see an empty slot that isn't
+		 * actually empty due to a pending update in the key cache
+		 * because the update that creates the inode isn't done with a
+		 * cached iterator, but - better safe than sorry, check the
+		 * cache before using a slot:
+		 */
+		if (k.k->type != KEY_TYPE_inode &&
+		    !bch2_btree_key_cache_find(c, BTREE_ID_INODES, iter->pos))
+			goto found_slot;
+	}
 
-			bch2_btree_iter_next_slot(iter);
-			break;
+	bch2_trans_iter_put(trans, iter);
 
-		default:
-			*hint			= k.k->p.inode;
-			inode_u->bi_inum	= k.k->p.inode;
-			inode_u->bi_generation	= bkey_generation(k);
+	if (ret)
+		return ret;
 
-			bch2_inode_pack(inode_p, inode_u);
-			bch2_trans_update(trans,
-				BTREE_INSERT_ENTRY(iter, &inode_p->inode.k_i));
-			return 0;
-		}
-	}
-out:
 	if (start != min) {
 		/* Retry from start */
 		start = min;
-		bch2_btree_iter_set_pos(iter, POS(start, 0));
 		goto again;
 	}
 
 	return -ENOSPC;
-}
+found_slot:
+	*hint			= k.k->p.offset;
+	inode_u->bi_inum	= k.k->p.offset;
+	inode_u->bi_generation	= bkey_generation(k);
 
-int bch2_inode_create(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
-		      u64 min, u64 max, u64 *hint)
-{
-	return bch2_trans_do(c, NULL, BTREE_INSERT_ATOMIC,
-			__bch2_inode_create(&trans, inode_u, min, max, hint));
+	ret = bch2_inode_write(trans, iter, inode_u);
+	bch2_trans_iter_put(trans, iter);
+	return ret;
 }
 
 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 {
-	struct btree_iter iter;
+	struct btree_trans trans;
+	struct btree_iter *iter;
 	struct bkey_i_inode_generation delete;
 	struct bpos start = POS(inode_nr, 0);
 	struct bpos end = POS(inode_nr + 1, 0);
+	struct bkey_s_c k;
+	u64 bi_generation;
 	int ret;
 
 	/*
@@ -385,107 +570,89 @@ int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 	if (ret)
 		return ret;
 
-	bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(inode_nr, 0),
-			     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
-	do {
-		struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
-		u32 bi_generation = 0;
+	bch2_trans_init(&trans, c, 0, 0);
+retry:
+	bch2_trans_begin(&trans);
 
-		ret = btree_iter_err(k);
-		if (ret) {
-			bch2_btree_iter_unlock(&iter);
-			return ret;
-		}
+	bi_generation = 0;
 
-		bch2_fs_inconsistent_on(k.k->type != KEY_TYPE_inode, c,
-					"inode %llu not found when deleting",
-					inode_nr);
+	iter = bch2_trans_get_iter(&trans, BTREE_ID_INODES, POS(0, inode_nr),
+				   BTREE_ITER_CACHED|BTREE_ITER_INTENT);
+	k = bch2_btree_iter_peek_cached(iter);
 
-		switch (k.k->type) {
-		case KEY_TYPE_inode: {
-			struct bch_inode_unpacked inode_u;
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
 
-			if (!bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u))
-				bi_generation = inode_u.bi_generation + 1;
-			break;
-		}
-		case KEY_TYPE_inode_generation: {
-			struct bkey_s_c_inode_generation g =
-				bkey_s_c_to_inode_generation(k);
-			bi_generation = le32_to_cpu(g.v->bi_generation);
-			break;
-		}
-		}
-
-		if (!bi_generation) {
-			bkey_init(&delete.k);
-			delete.k.p.inode = inode_nr;
-		} else {
-			bkey_inode_generation_init(&delete.k_i);
-			delete.k.p.inode = inode_nr;
-			delete.v.bi_generation = cpu_to_le32(bi_generation);
-		}
-
-		ret = bch2_btree_insert_at(c, NULL, NULL,
-				BTREE_INSERT_ATOMIC|
-				BTREE_INSERT_NOFAIL,
-				BTREE_INSERT_ENTRY(&iter, &delete.k_i));
-	} while (ret == -EINTR);
-
-	bch2_btree_iter_unlock(&iter);
+	bch2_fs_inconsistent_on(k.k->type != KEY_TYPE_inode, c,
+				"inode %llu not found when deleting",
+				inode_nr);
+
+	switch (k.k->type) {
+	case KEY_TYPE_inode: {
+		struct bch_inode_unpacked inode_u;
+
+		if (!bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u))
+			bi_generation = inode_u.bi_generation + 1;
+		break;
+	}
+	case KEY_TYPE_inode_generation: {
+		struct bkey_s_c_inode_generation g =
+			bkey_s_c_to_inode_generation(k);
+		bi_generation = le32_to_cpu(g.v->bi_generation);
+		break;
+	}
+	}
+
+	if (!bi_generation) {
+		bkey_init(&delete.k);
+		delete.k.p.offset = inode_nr;
+	} else {
+		bkey_inode_generation_init(&delete.k_i);
+		delete.k.p.offset = inode_nr;
+		delete.v.bi_generation = cpu_to_le32(bi_generation);
+	}
+
+	bch2_trans_update(&trans, iter, &delete.k_i, 0);
+
+	ret = bch2_trans_commit(&trans, NULL, NULL,
+				BTREE_INSERT_NOFAIL);
+err:
+	if (ret == -EINTR)
+		goto retry;
+
+	bch2_trans_exit(&trans);
 	return ret;
 }
 
-int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
-			    struct bch_inode_unpacked *inode)
+int bch2_inode_find_by_inum_trans(struct btree_trans *trans, u64 inode_nr,
+				  struct bch_inode_unpacked *inode)
 {
-	struct btree_iter iter;
+	struct btree_iter *iter;
 	struct bkey_s_c k;
-	int ret = -ENOENT;
-
-	for_each_btree_key(&iter, c, BTREE_ID_INODES,
-			   POS(inode_nr, 0),
-			   BTREE_ITER_SLOTS, k) {
-		switch (k.k->type) {
-		case KEY_TYPE_inode:
-			ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
-			break;
-		default:
-			/* hole, not found */
-			break;
-		}
+	int ret;
 
-		break;
-	}
+	iter = bch2_trans_get_iter(trans, BTREE_ID_INODES,
+			POS(0, inode_nr), BTREE_ITER_CACHED);
+	if (IS_ERR(iter))
+		return PTR_ERR(iter);
+
+	k = bch2_btree_iter_peek_cached(iter);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
 
-	return bch2_btree_iter_unlock(&iter) ?: ret;
+	ret = k.k->type == KEY_TYPE_inode
+		? bch2_inode_unpack(bkey_s_c_to_inode(k), inode)
+		: -ENOENT;
+err:
+	bch2_trans_iter_put(trans, iter);
+	return ret;
 }
 
-#ifdef CONFIG_BCACHEFS_DEBUG
-void bch2_inode_pack_test(void)
+int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
+			    struct bch_inode_unpacked *inode)
 {
-	struct bch_inode_unpacked *u, test_inodes[] = {
-		{
-			.bi_atime	= U64_MAX,
-			.bi_ctime	= U64_MAX,
-			.bi_mtime	= U64_MAX,
-			.bi_otime	= U64_MAX,
-			.bi_size	= U64_MAX,
-			.bi_sectors	= U64_MAX,
-			.bi_uid		= U32_MAX,
-			.bi_gid		= U32_MAX,
-			.bi_nlink	= U32_MAX,
-			.bi_generation	= U32_MAX,
-			.bi_dev		= U32_MAX,
-		},
-	};
-
-	for (u = test_inodes;
-	     u < test_inodes + ARRAY_SIZE(test_inodes);
-	     u++) {
-		struct bkey_inode_buf p;
-
-		bch2_inode_pack(&p, u);
-	}
+	return bch2_trans_do(c, NULL, NULL, 0,
+		bch2_inode_find_by_inum_trans(&trans, inode_nr, inode));
 }
-#endif