X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Finode.c;h=585d16ac2da21acd34ebf42cbf5af039ebe56351;hb=a06dee6da2babd19a392ff30a399b89b85018a36;hp=e953b78f7d9a76c6d7740a3003f30391b0266ba2;hpb=62f5e4fa67dde8255fa18a06d5354cdca02b6fc7;p=bcachefs-tools-debian

diff --git a/libbcachefs/inode.c b/libbcachefs/inode.c
index e953b78..585d16a 100644
--- a/libbcachefs/inode.c
+++ b/libbcachefs/inode.c
@@ -1,12 +1,17 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include "bcachefs.h"
+#include "btree_key_cache.h"
 #include "bkey_methods.h"
 #include "btree_update.h"
+#include "buckets.h"
 #include "error.h"
 #include "extents.h"
+#include "extent_update.h"
 #include "inode.h"
 #include "str_hash.h"
+#include "subvolume.h"
+#include "varint.h"
 
 #include <linux/random.h>
 
@@ -20,39 +25,6 @@ const char * const bch2_inode_opts[] = {
 };
 
 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
-static const u8 bits_table[8] = {
-	1  * 8 - 1,
-	2  * 8 - 2,
-	3  * 8 - 3,
-	4  * 8 - 4,
-	6  * 8 - 5,
-	8  * 8 - 6,
-	10 * 8 - 7,
-	13 * 8 - 8,
-};
-
-static int inode_encode_field(u8 *out, u8 *end, u64 hi, u64 lo)
-{
-	__be64 in[2] = { cpu_to_be64(hi), cpu_to_be64(lo), };
-	unsigned shift, bytes, bits = likely(!hi)
-		? fls64(lo)
-		: fls64(hi) + 64;
-
-	for (shift = 1; shift <= 8; shift++)
-		if (bits < bits_table[shift - 1])
-			goto got_shift;
-
-	BUG();
-got_shift:
-	bytes = byte_table[shift - 1];
-
-	BUG_ON(out + bytes > end);
-
-	memcpy(out, (u8 *) in + 16 - bytes, bytes);
-	*out |= (1 << 8) >> shift;
-
-	return bytes;
-}
 
 static int inode_decode_field(const u8 *in, const u8 *end,
 			      u64 out[2], unsigned *out_bits)
@@ -88,77 +60,102 @@ 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 inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
+					   const struct bch_inode_unpacked *inode)
 {
-	u8 *out = packed->inode.v.fields;
+	struct bkey_i_inode_v3 *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;
 
-	bkey_inode_init(&packed->inode.k_i);
-	packed->inode.k.p.inode		= inode->bi_inum;
+	bkey_inode_v3_init(&packed->inode.k_i);
+	packed->inode.k.p.offset	= inode->bi_inum;
+	packed->inode.v.bi_journal_seq	= cpu_to_le64(inode->bi_journal_seq);
 	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);
+	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
+	packed->inode.v.bi_sectors	= cpu_to_le64(inode->bi_sectors);
+	packed->inode.v.bi_size		= cpu_to_le64(inode->bi_size);
+	packed->inode.v.bi_version	= cpu_to_le64(inode->bi_version);
+	SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
+	SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
 
-#define x(_name, _bits)					\
-	out += inode_encode_field(out, end, 0, inode->_name);		\
+
+#define x(_name, _bits)							\
 	nr_fields++;							\
 									\
 	if (inode->_name) {						\
+		ret = bch2_varint_encode_fast(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()
+	BCH_INODE_FIELDS_v3()
 #undef  x
+	BUG_ON(out > end);
 
 	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(&packed->inode.v, nr_fields);
+	SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
 
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 		struct bch_inode_unpacked unpacked;
 
-		int ret = bch2_inode_unpack(inode_i_to_s_c(&packed->inode),
+		int ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i),
 					   &unpacked);
 		BUG_ON(ret);
 		BUG_ON(unpacked.bi_inum		!= inode->bi_inum);
 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
+		BUG_ON(unpacked.bi_sectors	!= inode->bi_sectors);
+		BUG_ON(unpacked.bi_size		!= inode->bi_size);
+		BUG_ON(unpacked.bi_version	!= inode->bi_version);
 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
 
-#define x(_name, _bits)	BUG_ON(unpacked._name != inode->_name);
-		BCH_INODE_FIELDS()
+#define x(_name, _bits)	if (unpacked._name != inode->_name)		\
+			panic("unpacked %llu should be %llu",		\
+			      (u64) unpacked._name, (u64) inode->_name);
+		BCH_INODE_FIELDS_v3()
 #undef  x
 	}
 }
 
-int bch2_inode_unpack(struct bkey_s_c_inode inode,
-		      struct bch_inode_unpacked *unpacked)
+void bch2_inode_pack(struct bkey_inode_buf *packed,
+		     const struct bch_inode_unpacked *inode)
+{
+	bch2_inode_pack_inlined(packed, inode);
+}
+
+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 x(_name, _bits)					\
 	if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {			\
-		memset(&unpacked->_name, 0,				\
-		       sizeof(*unpacked) -				\
-		       offsetof(struct bch_inode_unpacked, _name));	\
+		unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
+		memset((void *) unpacked + offset, 0,			\
+		       sizeof(*unpacked) - offset);			\
 		return 0;						\
 	}								\
 									\
@@ -172,44 +169,186 @@ int bch2_inode_unpack(struct bkey_s_c_inode inode,
 	unpacked->_name = field[1];					\
 	in += ret;
 
-	BCH_INODE_FIELDS()
+	BCH_INODE_FIELDS_v2()
 #undef  x
 
 	/* XXX: signal if there were more fields than expected? */
+	return 0;
+}
 
+static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
+				const u8 *in, const u8 *end,
+				unsigned nr_fields)
+{
+	unsigned fieldnr = 0;
+	int ret;
+	u64 v[2];
+
+#define x(_name, _bits)							\
+	if (fieldnr < nr_fields) {					\
+		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
+		if (ret < 0)						\
+			return ret;					\
+		in += ret;						\
+									\
+		if (_bits > 64) {					\
+			ret = bch2_varint_decode_fast(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_v2()
+#undef  x
+
+	/* XXX: signal if there were more fields than expected? */
+	return 0;
+}
+
+static int bch2_inode_unpack_v3(struct bkey_s_c k,
+				struct bch_inode_unpacked *unpacked)
+{
+	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
+	const u8 *in = inode.v->fields;
+	const u8 *end = bkey_val_end(inode);
+	unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
+	unsigned fieldnr = 0;
+	int ret;
+	u64 v[2];
+
+	unpacked->bi_inum	= inode.k->p.offset;
+	unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
+	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
+	unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
+	unpacked->bi_sectors	= le64_to_cpu(inode.v->bi_sectors);
+	unpacked->bi_size	= le64_to_cpu(inode.v->bi_size);
+	unpacked->bi_version	= le64_to_cpu(inode.v->bi_version);
+	unpacked->bi_mode	= INODEv3_MODE(inode.v);
+
+#define x(_name, _bits)							\
+	if (fieldnr < nr_fields) {					\
+		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
+		if (ret < 0)						\
+			return ret;					\
+		in += ret;						\
+									\
+		if (_bits > 64) {					\
+			ret = bch2_varint_decode_fast(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_v3()
+#undef  x
+
+	/* XXX: signal if there were more fields than expected? */
 	return 0;
 }
 
-struct btree_iter *bch2_inode_peek(struct btree_trans *trans,
-				   struct bch_inode_unpacked *inode,
-				   u64 inum, unsigned flags)
+static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
+					       struct bch_inode_unpacked *unpacked)
+{
+	memset(unpacked, 0, sizeof(*unpacked));
+
+	switch (k.k->type) {
+	case KEY_TYPE_inode: {
+		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
+
+		unpacked->bi_inum	= inode.k->p.offset;
+		unpacked->bi_journal_seq= 0;
+		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(unpacked, inode.v->fields,
+						    bkey_val_end(inode),
+						    INODE_NR_FIELDS(inode.v));
+		} else {
+			return bch2_inode_unpack_v1(inode, unpacked);
+		}
+		break;
+	}
+	case KEY_TYPE_inode_v2: {
+		struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
+
+		unpacked->bi_inum	= inode.k->p.offset;
+		unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
+		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
+		unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
+		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
+
+		return bch2_inode_unpack_v2(unpacked, inode.v->fields,
+					    bkey_val_end(inode),
+					    INODEv2_NR_FIELDS(inode.v));
+	}
+	default:
+		BUG();
+	}
+}
+
+int bch2_inode_unpack(struct bkey_s_c k,
+		      struct bch_inode_unpacked *unpacked)
+{
+	if (likely(k.k->type == KEY_TYPE_inode_v3))
+		return bch2_inode_unpack_v3(k, unpacked);
+	return bch2_inode_unpack_slowpath(k, unpacked);
+}
+
+int bch2_inode_peek(struct btree_trans *trans,
+		    struct btree_iter *iter,
+		    struct bch_inode_unpacked *inode,
+		    subvol_inum inum, unsigned flags)
 {
-	struct btree_iter *iter;
 	struct bkey_s_c k;
+	u32 snapshot;
 	int ret;
 
-	iter = bch2_trans_get_iter(trans, BTREE_ID_INODES, POS(inum, 0),
-				   BTREE_ITER_SLOTS|flags);
-	if (IS_ERR(iter))
-		return iter;
+	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
+	if (ret)
+		return ret;
 
+	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes,
+			     SPOS(0, inum.inum, snapshot),
+			     flags|BTREE_ITER_CACHED);
 	k = bch2_btree_iter_peek_slot(iter);
 	ret = bkey_err(k);
 	if (ret)
 		goto err;
 
-	ret = k.k->type == KEY_TYPE_inode ? 0 : -EIO;
+	ret = bkey_is_inode(k.k) ? 0 : -ENOENT;
 	if (ret)
 		goto err;
 
-	ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
+	ret = bch2_inode_unpack(k, inode);
 	if (ret)
 		goto err;
 
-	return iter;
+	return 0;
 err:
-	bch2_trans_iter_put(trans, iter);
-	return ERR_PTR(ret);
+	bch2_trans_iter_exit(trans, iter);
+	return ret;
 }
 
 int bch2_inode_write(struct btree_trans *trans,
@@ -222,71 +361,195 @@ int bch2_inode_write(struct btree_trans *trans,
 	if (IS_ERR(inode_p))
 		return PTR_ERR(inode_p);
 
-	bch2_inode_pack(inode_p, inode);
-	bch2_trans_update(trans, iter, &inode_p->inode.k_i);
-	return 0;
+	bch2_inode_pack_inlined(inode_p, inode);
+	inode_p->inode.k.p.snapshot = iter->snapshot;
+	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 }
 
-const char *bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k)
+struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
 {
-		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
-		struct bch_inode_unpacked unpacked;
+	struct bch_inode_unpacked u;
+	struct bkey_inode_buf *inode_p;
+	int ret;
 
-	if (k.k->p.offset)
-		return "nonzero offset";
+	if (!bkey_is_inode(&k->k))
+		return ERR_PTR(-ENOENT);
 
-	if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
-		return "incorrect value size";
+	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
+	if (IS_ERR(inode_p))
+		return ERR_CAST(inode_p);
 
-	if (k.k->p.inode < BLOCKDEV_INODE_MAX)
-		return "fs inode in blockdev range";
+	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
+	if (ret)
+		return ERR_PTR(ret);
 
-	if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
-		return "invalid str hash type";
+	bch2_inode_pack(inode_p, &u);
+	return &inode_p->inode.k_i;
+}
 
-	if (bch2_inode_unpack(inode, &unpacked))
-		return "invalid variable length fields";
+static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
+{
+	struct bch_inode_unpacked unpacked;
 
-	if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
-		return "invalid data checksum type";
+	if (k.k->p.inode) {
+		prt_printf(err, "nonzero k.p.inode");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
-		return "invalid data checksum type";
+	if (k.k->p.offset < BLOCKDEV_INODE_MAX) {
+		prt_printf(err, "fs inode in blockdev range");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (bch2_inode_unpack(k, &unpacked)) {
+		prt_printf(err, "invalid variable length fields");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1) {
+		prt_printf(err, "invalid data checksum type (%u >= %u",
+			unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1) {
+		prt_printf(err, "invalid data checksum type (%u >= %u)",
+		       unpacked.bi_compression, BCH_COMPRESSION_OPT_NR + 1);
+		return -BCH_ERR_invalid_bkey;
+	}
 
 	if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
-	    unpacked.bi_nlink != 0)
-		return "flagged as unlinked but bi_nlink != 0";
+	    unpacked.bi_nlink != 0) {
+		prt_printf(err, "flagged as unlinked but bi_nlink != 0");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode)) {
+		prt_printf(err, "subvolume root but not a directory");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	return NULL;
+	return 0;
 }
 
-void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c,
-		       struct bkey_s_c k)
+int bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k,
+		       int rw, struct printbuf *err)
 {
 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
-	struct bch_inode_unpacked unpacked;
 
-	if (bch2_inode_unpack(inode, &unpacked)) {
-		pr_buf(out, "(unpack error)");
-		return;
+	if (bkey_val_bytes(k.k) < sizeof(*inode.v)) {
+		prt_printf(err, "incorrect value size (%zu < %zu)",
+		       bkey_val_bytes(k.k), sizeof(*inode.v));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
+		prt_printf(err, "invalid str hash type (%llu >= %u)",
+		       INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	return __bch2_inode_invalid(k, err);
+}
+
+int bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
+			  int rw, struct printbuf *err)
+{
+	struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
+
+	if (bkey_val_bytes(k.k) < sizeof(*inode.v)) {
+		prt_printf(err, "incorrect value size (%zu < %zu)",
+		       bkey_val_bytes(k.k), sizeof(*inode.v));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
+		prt_printf(err, "invalid str hash type (%llu >= %u)",
+		       INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	return __bch2_inode_invalid(k, err);
+}
+
+int bch2_inode_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
+			  int rw, struct printbuf *err)
+{
+	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
+
+	if (bkey_val_bytes(k.k) < sizeof(*inode.v)) {
+		prt_printf(err, "incorrect value size (%zu < %zu)",
+		       bkey_val_bytes(k.k), sizeof(*inode.v));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
+	    INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k)) {
+		prt_printf(err, "invalid fields_start (got %llu, min %u max %zu)",
+		       INODEv3_FIELDS_START(inode.v),
+		       INODEv3_FIELDS_START_INITIAL,
+		       bkey_val_u64s(inode.k));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
+		prt_printf(err, "invalid str hash type (%llu >= %u)",
+		       INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
+		return -BCH_ERR_invalid_bkey;
 	}
 
+	return __bch2_inode_invalid(k, err);
+}
+
+static void __bch2_inode_unpacked_to_text(struct printbuf *out,
+					  struct bch_inode_unpacked *inode)
+{
+	prt_printf(out, "mode %o flags %x journal_seq %llu bi_size %llu bi_sectors %llu bi_version %llu",
+	       inode->bi_mode, inode->bi_flags,
+	       inode->bi_journal_seq,
+	       inode->bi_size,
+	       inode->bi_sectors,
+	       inode->bi_version);
+
 #define x(_name, _bits)						\
-	pr_buf(out, #_name ": %llu ", (u64) unpacked._name);
-	BCH_INODE_FIELDS()
+	prt_printf(out, " "#_name " %llu", (u64) inode->_name);
+	BCH_INODE_FIELDS_v3()
 #undef  x
 }
 
-const char *bch2_inode_generation_invalid(const struct bch_fs *c,
-					  struct bkey_s_c k)
+void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
 {
-	if (k.k->p.offset)
-		return "nonzero offset";
+	prt_printf(out, "inum: %llu ", inode->bi_inum);
+	__bch2_inode_unpacked_to_text(out, inode);
+}
+
+void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
+{
+	struct bch_inode_unpacked inode;
 
-	if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
-		return "incorrect value size";
+	if (bch2_inode_unpack(k, &inode)) {
+		prt_printf(out, "(unpack error)");
+		return;
+	}
 
-	return NULL;
+	__bch2_inode_unpacked_to_text(out, &inode);
+}
+
+int bch2_inode_generation_invalid(const struct bch_fs *c, struct bkey_s_c k,
+				  int rw, struct printbuf *err)
+{
+	if (k.k->p.inode) {
+		prt_printf(err, "nonzero k.p.inode");
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation)) {
+		prt_printf(err, "incorrect value size (%zu != %zu)",
+		       bkey_val_bytes(k.k), sizeof(struct bch_inode_generation));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	return 0;
 }
 
 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
@@ -294,7 +557,7 @@ void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
 {
 	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));
+	prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
 }
 
 void bch2_inode_init_early(struct bch_fs *c,
@@ -350,6 +613,7 @@ static inline u32 bkey_generation(struct bkey_s_c k)
 {
 	switch (k.k->type) {
 	case KEY_TYPE_inode:
+	case KEY_TYPE_inode_v2:
 		BUG();
 	case KEY_TYPE_inode_generation:
 		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
@@ -358,83 +622,176 @@ static inline u32 bkey_generation(struct bkey_s_c k)
 	}
 }
 
+/*
+ * This just finds an empty slot:
+ */
 int bch2_inode_create(struct btree_trans *trans,
+		      struct btree_iter *iter,
 		      struct bch_inode_unpacked *inode_u,
-		      u64 min, u64 max, u64 *hint)
+		      u32 snapshot, u64 cpu)
 {
 	struct bch_fs *c = trans->c;
-	struct bkey_inode_buf *inode_p;
-	struct btree_iter *iter;
-	u64 start;
-	int ret;
-
-	if (!max)
-		max = ULLONG_MAX;
-
-	if (c->opts.inodes_32bit)
-		max = min_t(u64, max, U32_MAX);
+	struct bkey_s_c k;
+	u64 min, max, start, pos, *hint;
+	int ret = 0;
+	unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
+
+	if (c->opts.shard_inode_numbers) {
+		bits -= c->inode_shard_bits;
+
+		min = (cpu << bits);
+		max = (cpu << bits) | ~(ULLONG_MAX << bits);
+
+		min = max_t(u64, min, BLOCKDEV_INODE_MAX);
+		hint = c->unused_inode_hints + cpu;
+	} else {
+		min = BLOCKDEV_INODE_MAX;
+		max = ~(ULLONG_MAX << bits);
+		hint = c->unused_inode_hints;
+	}
 
 	start = READ_ONCE(*hint);
 
 	if (start >= max || start < min)
 		start = min;
 
-	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);
+	pos = start;
+	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
+			     BTREE_ITER_ALL_SNAPSHOTS|
+			     BTREE_ITER_INTENT);
 again:
+	while ((k = bch2_btree_iter_peek(iter)).k &&
+	       !(ret = bkey_err(k)) &&
+	       bkey_lt(k.k->p, POS(0, max))) {
+		while (pos < iter->pos.offset) {
+			if (!bch2_btree_key_cache_find(c, BTREE_ID_inodes, POS(0, pos)))
+				goto found_slot;
+
+			pos++;
+		}
+
+		if (k.k->p.snapshot == snapshot &&
+		    !bkey_is_inode(k.k) &&
+		    !bch2_btree_key_cache_find(c, BTREE_ID_inodes, SPOS(0, pos, snapshot))) {
+			bch2_btree_iter_advance(iter);
+			continue;
+		}
+
+		/*
+		 * We don't need to iterate over keys in every snapshot once
+		 * we've found just one:
+		 */
+		pos = iter->pos.offset + 1;
+		bch2_btree_iter_set_pos(iter, POS(0, pos));
+	}
+
+	while (!ret && pos < max) {
+		if (!bch2_btree_key_cache_find(c, BTREE_ID_inodes, POS(0, pos)))
+			goto found_slot;
+
+		pos++;
+	}
+
+	if (!ret && start == min)
+		ret = -BCH_ERR_ENOSPC_inode_create;
+
+	if (ret) {
+		bch2_trans_iter_exit(trans, iter);
+		return ret;
+	}
+
+	/* Retry from start */
+	pos = start = min;
+	bch2_btree_iter_set_pos(iter, POS(0, pos));
+	goto again;
+found_slot:
+	bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
+	k = bch2_btree_iter_peek_slot(iter);
+	ret = bkey_err(k);
+	if (ret) {
+		bch2_trans_iter_exit(trans, iter);
+		return ret;
+	}
+
+	/* We may have raced while the iterator wasn't pointing at pos: */
+	if (bkey_is_inode(k.k) ||
+	    bch2_btree_key_cache_find(c, BTREE_ID_inodes, k.k->p))
+		goto again;
+
+	*hint			= k.k->p.offset;
+	inode_u->bi_inum	= k.k->p.offset;
+	inode_u->bi_generation	= bkey_generation(k);
+	return 0;
+}
+
+static int bch2_inode_delete_keys(struct btree_trans *trans,
+				  subvol_inum inum, enum btree_id id)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	struct bkey_i delete;
+	u32 snapshot;
+	int ret = 0;
+
+	/*
+	 * We're never going to be deleting extents, no need to use an extent
+	 * iterator:
+	 */
+	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
+			     BTREE_ITER_INTENT);
+
 	while (1) {
-		struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
+		bch2_trans_begin(trans);
 
-		ret = bkey_err(k);
+		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
 		if (ret)
-			return ret;
+			goto err;
 
-		switch (k.k->type) {
-		case KEY_TYPE_inode:
-			/* slot used */
-			if (iter->pos.inode >= max)
-				goto out;
+		bch2_btree_iter_set_snapshot(&iter, snapshot);
 
-			bch2_btree_iter_next_slot(iter);
+		k = bch2_btree_iter_peek_upto(&iter, POS(inum.inum, U64_MAX));
+		ret = bkey_err(k);
+		if (ret)
+			goto err;
+
+		if (!k.k)
 			break;
 
-		default:
-			*hint			= k.k->p.inode;
-			inode_u->bi_inum	= k.k->p.inode;
-			inode_u->bi_generation	= bkey_generation(k);
+		bkey_init(&delete.k);
+		delete.k.p = iter.pos;
+
+		if (iter.flags & BTREE_ITER_IS_EXTENTS) {
+			bch2_key_resize(&delete.k, k.k->p.offset - iter.pos.offset);
 
-			bch2_inode_pack(inode_p, inode_u);
-			bch2_trans_update(trans, iter, &inode_p->inode.k_i);
-			return 0;
+			ret = bch2_extent_trim_atomic(trans, &iter, &delete);
+			if (ret)
+				goto err;
 		}
-	}
-out:
-	if (start != min) {
-		/* Retry from start */
-		start = min;
-		bch2_btree_iter_set_pos(iter, POS(start, 0));
-		goto again;
+
+		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
+		      bch2_trans_commit(trans, NULL, NULL,
+					BTREE_INSERT_NOFAIL);
+err:
+		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
+			break;
 	}
 
-	return -ENOSPC;
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
 }
 
-int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
+int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
 {
 	struct btree_trans trans;
-	struct btree_iter *iter;
+	struct btree_iter iter = { NULL };
 	struct bkey_i_inode_generation delete;
-	struct bpos start = POS(inode_nr, 0);
-	struct bpos end = POS(inode_nr + 1, 0);
+	struct bch_inode_unpacked inode_u;
+	struct bkey_s_c k;
+	u32 snapshot;
 	int ret;
 
+	bch2_trans_init(&trans, c, 0, 1024);
+
 	/*
 	 * If this was a directory, there shouldn't be any real dirents left -
 	 * but there could be whiteouts (from hash collisions) that we should
@@ -443,120 +800,127 @@ int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
 	 * XXX: the dirent could ideally would delete whiteouts when they're no
 	 * longer needed
 	 */
-	ret   = bch2_btree_delete_range(c, BTREE_ID_EXTENTS,
-					start, end, NULL) ?:
-		bch2_btree_delete_range(c, BTREE_ID_XATTRS,
-					start, end, NULL) ?:
-		bch2_btree_delete_range(c, BTREE_ID_DIRENTS,
-					start, end, NULL);
+	ret   = bch2_inode_delete_keys(&trans, inum, BTREE_ID_extents) ?:
+		bch2_inode_delete_keys(&trans, inum, BTREE_ID_xattrs) ?:
+		bch2_inode_delete_keys(&trans, inum, BTREE_ID_dirents);
 	if (ret)
-		return ret;
-
-	bch2_trans_init(&trans, c, 0, 0);
+		goto err;
+retry:
+	bch2_trans_begin(&trans);
 
-	iter = bch2_trans_get_iter(&trans, 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;
+	ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
+	if (ret)
+		goto err;
 
-		ret = bkey_err(k);
-		if (ret)
-			break;
+	bch2_trans_iter_init(&trans, &iter, BTREE_ID_inodes,
+			     SPOS(0, inum.inum, snapshot),
+			     BTREE_ITER_INTENT|BTREE_ITER_CACHED);
+	k = bch2_btree_iter_peek_slot(&iter);
 
-		bch2_fs_inconsistent_on(k.k->type != KEY_TYPE_inode, c,
-					"inode %llu not found when deleting",
-					inode_nr);
+	ret = bkey_err(k);
+	if (ret)
+		goto err;
 
-		switch (k.k->type) {
-		case KEY_TYPE_inode: {
-			struct bch_inode_unpacked inode_u;
+	if (!bkey_is_inode(k.k)) {
+		bch2_fs_inconsistent(trans.c,
+				     "inode %llu not found when deleting",
+				     inum.inum);
+		ret = -EIO;
+		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;
-		}
-		}
+	bch2_inode_unpack(k, &inode_u);
 
-		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);
-		}
+	/* Subvolume root? */
+	BUG_ON(inode_u.bi_subvol);
 
-		bch2_trans_update(&trans, iter, &delete.k_i);
+	bkey_inode_generation_init(&delete.k_i);
+	delete.k.p = iter.pos;
+	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
 
-		ret = bch2_trans_commit(&trans, NULL, NULL,
-					BTREE_INSERT_ATOMIC|
-					BTREE_INSERT_NOFAIL);
-	} while (ret == -EINTR);
+	ret   = bch2_trans_update(&trans, &iter, &delete.k_i, 0) ?:
+		bch2_trans_commit(&trans, NULL, NULL,
+				BTREE_INSERT_NOFAIL);
+err:
+	bch2_trans_iter_exit(&trans, &iter);
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+		goto retry;
 
 	bch2_trans_exit(&trans);
 	return ret;
 }
 
-int bch2_inode_find_by_inum_trans(struct btree_trans *trans, u64 inode_nr,
+int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
+				  subvol_inum inum,
 				  struct bch_inode_unpacked *inode)
 {
-	struct btree_iter *iter;
-	struct bkey_s_c k;
-	int ret = -ENOENT;
-
-	iter = bch2_trans_get_iter(trans, BTREE_ID_INODES,
-			POS(inode_nr, 0), BTREE_ITER_SLOTS);
-	if (IS_ERR(iter))
-		return PTR_ERR(iter);
-
-	k = bch2_btree_iter_peek_slot(iter);
-	if (k.k->type == KEY_TYPE_inode)
-		ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
-
-	bch2_trans_iter_put(trans, iter);
+	struct btree_iter iter;
+	int ret;
 
+	ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
+	if (!ret)
+		bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
 
-int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
+int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
 			    struct bch_inode_unpacked *inode)
 {
-	return bch2_trans_do(c, NULL, 0,
-		bch2_inode_find_by_inum_trans(&trans, inode_nr, inode));
+	return bch2_trans_do(c, NULL, NULL, 0,
+		bch2_inode_find_by_inum_trans(&trans, inum, inode));
 }
 
-#ifdef CONFIG_BCACHEFS_DEBUG
-void bch2_inode_pack_test(void)
+int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
 {
-	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,
-		},
-	};
+	if (bi->bi_flags & BCH_INODE_UNLINKED)
+		bi->bi_flags &= ~BCH_INODE_UNLINKED;
+	else {
+		if (bi->bi_nlink == U32_MAX)
+			return -EINVAL;
 
-	for (u = test_inodes;
-	     u < test_inodes + ARRAY_SIZE(test_inodes);
-	     u++) {
-		struct bkey_inode_buf p;
+		bi->bi_nlink++;
+	}
+
+	return 0;
+}
+
+void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
+{
+	if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_UNLINKED)) {
+		bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
+					bi->bi_inum);
+		return;
+	}
 
-		bch2_inode_pack(&p, u);
+	if (bi->bi_flags & BCH_INODE_UNLINKED) {
+		bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
+		return;
 	}
+
+	if (bi->bi_nlink)
+		bi->bi_nlink--;
+	else
+		bi->bi_flags |= BCH_INODE_UNLINKED;
+}
+
+struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
+{
+	struct bch_opts ret = { 0 };
+#define x(_name, _bits)							\
+	if (inode->bi_##_name)						\
+		opt_set(ret, _name, inode->bi_##_name - 1);
+	BCH_INODE_OPTS()
+#undef x
+	return ret;
+}
+
+void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
+			 struct bch_inode_unpacked *inode)
+{
+#define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
+	BCH_INODE_OPTS()
+#undef x
+
+	if (opts->nocow)
+		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
 }
-#endif