[mirror_zfs.git] / include / sys / zap_leaf.h

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 */

#ifndef	_SYS_ZAP_LEAF_H
#define	_SYS_ZAP_LEAF_H

#include <sys/zap.h>

#ifdef	__cplusplus
extern "C" {
#endif

struct zap;
struct zap_name;
struct zap_stats;

#define	ZAP_LEAF_MAGIC 0x2AB1EAF

/* chunk size = 24 bytes */
#define	ZAP_LEAF_CHUNKSIZE 24

/*
 * The amount of space available for chunks is:
 * block size (1<<l->l_bs) - hash entry size (2) * number of hash
 * entries - header space (2*chunksize)
 */
#define	ZAP_LEAF_NUMCHUNKS(l) \
	(((1<<(l)->l_bs) - 2*ZAP_LEAF_HASH_NUMENTRIES(l)) / \
	ZAP_LEAF_CHUNKSIZE - 2)

/*
 * The amount of space within the chunk available for the array is:
 * chunk size - space for type (1) - space for next pointer (2)
 */
#define	ZAP_LEAF_ARRAY_BYTES (ZAP_LEAF_CHUNKSIZE - 3)

#define	ZAP_LEAF_ARRAY_NCHUNKS(bytes) \
	(((bytes)+ZAP_LEAF_ARRAY_BYTES-1)/ZAP_LEAF_ARRAY_BYTES)

/*
 * Low water mark:  when there are only this many chunks free, start
 * growing the ptrtbl.  Ideally, this should be larger than a
 * "reasonably-sized" entry.  20 chunks is more than enough for the
 * largest directory entry (MAXNAMELEN (256) byte name, 8-byte value),
 * while still being only around 3% for 16k blocks.
 */
#define	ZAP_LEAF_LOW_WATER (20)

/*
 * The leaf hash table has block size / 2^5 (32) number of entries,
 * which should be more than enough for the maximum number of entries,
 * which is less than block size / CHUNKSIZE (24) / minimum number of
 * chunks per entry (3).
 */
#define	ZAP_LEAF_HASH_SHIFT(l) ((l)->l_bs - 5)
#define	ZAP_LEAF_HASH_NUMENTRIES(l) (1 << ZAP_LEAF_HASH_SHIFT(l))

/*
 * The chunks start immediately after the hash table.  The end of the
 * hash table is at l_hash + HASH_NUMENTRIES, which we simply cast to a
 * chunk_t.
 */
#define	ZAP_LEAF_CHUNK(l, idx) \
	((zap_leaf_chunk_t *) \
	((l)->l_phys->l_hash + ZAP_LEAF_HASH_NUMENTRIES(l)))[idx]
#define	ZAP_LEAF_ENTRY(l, idx) (&ZAP_LEAF_CHUNK(l, idx).l_entry)

typedef enum zap_chunk_type {
	ZAP_CHUNK_FREE = 253,
	ZAP_CHUNK_ENTRY = 252,
	ZAP_CHUNK_ARRAY = 251,
	ZAP_CHUNK_TYPE_MAX = 250
} zap_chunk_type_t;

#define	ZLF_ENTRIES_CDSORTED (1<<0)

/*
 * TAKE NOTE:
 * If zap_leaf_phys_t is modified, zap_leaf_byteswap() must be modified.
 */
typedef struct zap_leaf_phys {
	struct zap_leaf_header {
		/* Public to ZAP */
		uint64_t lh_block_type;		/* ZBT_LEAF */
		uint64_t lh_pad1;
		uint64_t lh_prefix;		/* hash prefix of this leaf */
		uint32_t lh_magic;		/* ZAP_LEAF_MAGIC */
		uint16_t lh_nfree;		/* number free chunks */
		uint16_t lh_nentries;		/* number of entries */
		uint16_t lh_prefix_len;		/* num bits used to id this */

		/* Private to zap_leaf */
		uint16_t lh_freelist;		/* chunk head of free list */
		uint8_t lh_flags;		/* ZLF_* flags */
		uint8_t lh_pad2[11];
	} l_hdr; /* 2 24-byte chunks */

	/*
	 * The header is followed by a hash table with
	 * ZAP_LEAF_HASH_NUMENTRIES(zap) entries.  The hash table is
	 * followed by an array of ZAP_LEAF_NUMCHUNKS(zap)
	 * zap_leaf_chunk structures.  These structures are accessed
	 * with the ZAP_LEAF_CHUNK() macro.
	 */

	uint16_t l_hash[1];
} zap_leaf_phys_t;

typedef union zap_leaf_chunk {
	struct zap_leaf_entry {
		uint8_t le_type; 		/* always ZAP_CHUNK_ENTRY */
		uint8_t le_value_intlen;	/* size of value's ints */
		uint16_t le_next;		/* next entry in hash chain */
		uint16_t le_name_chunk;		/* first chunk of the name */
		uint16_t le_name_numints;	/* ints in name (incl null) */
		uint16_t le_value_chunk;	/* first chunk of the value */
		uint16_t le_value_numints;	/* value length in ints */
		uint32_t le_cd;			/* collision differentiator */
		uint64_t le_hash;		/* hash value of the name */
	} l_entry;
	struct zap_leaf_array {
		uint8_t la_type;		/* always ZAP_CHUNK_ARRAY */
		uint8_t la_array[ZAP_LEAF_ARRAY_BYTES];
		uint16_t la_next;		/* next blk or CHAIN_END */
	} l_array;
	struct zap_leaf_free {
		uint8_t lf_type;		/* always ZAP_CHUNK_FREE */
		uint8_t lf_pad[ZAP_LEAF_ARRAY_BYTES];
		uint16_t lf_next;	/* next in free list, or CHAIN_END */
	} l_free;
} zap_leaf_chunk_t;

typedef struct zap_leaf {
	krwlock_t l_rwlock;
	uint64_t l_blkid;		/* 1<<ZAP_BLOCK_SHIFT byte block off */
	int l_bs;			/* block size shift */
	dmu_buf_t *l_dbuf;
	zap_leaf_phys_t *l_phys;
} zap_leaf_t;


typedef struct zap_entry_handle {
	/* Set by zap_leaf and public to ZAP */
	uint64_t zeh_num_integers;
	uint64_t zeh_hash;
	uint32_t zeh_cd;
	uint8_t zeh_integer_size;

	/* Private to zap_leaf */
	uint16_t zeh_fakechunk;
	uint16_t *zeh_chunkp;
	zap_leaf_t *zeh_leaf;
} zap_entry_handle_t;

/*
 * Return a handle to the named entry, or ENOENT if not found.  The hash
 * value must equal zap_hash(name).
 */
extern int zap_leaf_lookup(zap_leaf_t *l,
    struct zap_name *zn, zap_entry_handle_t *zeh);

/*
 * Return a handle to the entry with this hash+cd, or the entry with the
 * next closest hash+cd.
 */
extern int zap_leaf_lookup_closest(zap_leaf_t *l,
    uint64_t hash, uint32_t cd, zap_entry_handle_t *zeh);

/*
 * Read the first num_integers in the attribute.  Integer size
 * conversion will be done without sign extension.  Return EINVAL if
 * integer_size is too small.  Return EOVERFLOW if there are more than
 * num_integers in the attribute.
 */
extern int zap_entry_read(const zap_entry_handle_t *zeh,
    uint8_t integer_size, uint64_t num_integers, void *buf);

extern int zap_entry_read_name(struct zap *zap, const zap_entry_handle_t *zeh,
    uint16_t buflen, char *buf);

/*
 * Replace the value of an existing entry.
 *
 * May fail if it runs out of space (ENOSPC).
 */
extern int zap_entry_update(zap_entry_handle_t *zeh,
    uint8_t integer_size, uint64_t num_integers, const void *buf);

/*
 * Remove an entry.
 */
extern void zap_entry_remove(zap_entry_handle_t *zeh);

/*
 * Create an entry. An equal entry must not exist, and this entry must
 * belong in this leaf (according to its hash value).  Fills in the
 * entry handle on success.  Returns 0 on success or ENOSPC on failure.
 */
extern int zap_entry_create(zap_leaf_t *l, struct zap_name *zn, uint32_t cd,
    uint8_t integer_size, uint64_t num_integers, const void *buf,
    zap_entry_handle_t *zeh);

/* Determine whether there is another entry with the same normalized form. */
extern boolean_t zap_entry_normalization_conflict(zap_entry_handle_t *zeh,
    struct zap_name *zn, const char *name, struct zap *zap);

/*
 * Other stuff.
 */

extern void zap_leaf_init(zap_leaf_t *l, boolean_t sort);
extern void zap_leaf_byteswap(zap_leaf_phys_t *buf, int len);
extern void zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort);
extern void zap_leaf_stats(struct zap *zap, zap_leaf_t *l,
    struct zap_stats *zs);

#ifdef	__cplusplus
}
#endif

#endif /* _SYS_ZAP_LEAF_H */
Commit	Line	Data
34dc7c2f BB	1	/*
	2	* CDDL HEADER START
	3	*
	4	* The contents of this file are subject to the terms of the
	5	* Common Development and Distribution License (the "License").
	6	* You may not use this file except in compliance with the License.
	7	*
	8	* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
	9	* or http://www.opensolaris.org/os/licensing.
	10	* See the License for the specific language governing permissions
	11	* and limitations under the License.
	12	*
	13	* When distributing Covered Code, include this CDDL HEADER in each
	14	* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
	15	* If applicable, add the following below this CDDL HEADER, with the
	16	* fields enclosed by brackets "[]" replaced with your own identifying
	17	* information: Portions Copyright [yyyy] [name of copyright owner]
	18	*
	19	* CDDL HEADER END
	20	*/
	21	/*
428870ff	22	* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
34dc7c2f BB	23	*/
	24
	25	#ifndef _SYS_ZAP_LEAF_H
	26	#define _SYS_ZAP_LEAF_H
	27
428870ff	28	#include <sys/zap.h>
34dc7c2f BB	29
	30	#ifdef __cplusplus
	31	extern "C" {
	32	#endif
	33
	34	struct zap;
428870ff BB	35	struct zap_name;
428870ff BB	36	struct zap_stats;
34dc7c2f BB	37
	38	#define ZAP_LEAF_MAGIC 0x2AB1EAF
	39
	40	/* chunk size = 24 bytes */
	41	#define ZAP_LEAF_CHUNKSIZE 24
	42
	43	/*
	44	* The amount of space available for chunks is:
	45	* block size (1<<l->l_bs) - hash entry size (2) * number of hash
	46	* entries - header space (2*chunksize)
	47	*/
	48	#define ZAP_LEAF_NUMCHUNKS(l) \
	49	(((1<<(l)->l_bs) - 2*ZAP_LEAF_HASH_NUMENTRIES(l)) / \
	50	ZAP_LEAF_CHUNKSIZE - 2)
	51
	52	/*
	53	* The amount of space within the chunk available for the array is:
	54	* chunk size - space for type (1) - space for next pointer (2)
	55	*/
	56	#define ZAP_LEAF_ARRAY_BYTES (ZAP_LEAF_CHUNKSIZE - 3)
	57
	58	#define ZAP_LEAF_ARRAY_NCHUNKS(bytes) \
	59	(((bytes)+ZAP_LEAF_ARRAY_BYTES-1)/ZAP_LEAF_ARRAY_BYTES)
	60
	61	/*
	62	* Low water mark: when there are only this many chunks free, start
	63	* growing the ptrtbl. Ideally, this should be larger than a
	64	* "reasonably-sized" entry. 20 chunks is more than enough for the
	65	* largest directory entry (MAXNAMELEN (256) byte name, 8-byte value),
	66	* while still being only around 3% for 16k blocks.
	67	*/
	68	#define ZAP_LEAF_LOW_WATER (20)
	69
	70	/*
	71	* The leaf hash table has block size / 2^5 (32) number of entries,
	72	* which should be more than enough for the maximum number of entries,
	73	* which is less than block size / CHUNKSIZE (24) / minimum number of
	74	* chunks per entry (3).
	75	*/
	76	#define ZAP_LEAF_HASH_SHIFT(l) ((l)->l_bs - 5)
	77	#define ZAP_LEAF_HASH_NUMENTRIES(l) (1 << ZAP_LEAF_HASH_SHIFT(l))
	78
	79	/*
	80	* The chunks start immediately after the hash table. The end of the
	81	* hash table is at l_hash + HASH_NUMENTRIES, which we simply cast to a
	82	* chunk_t.
	83	*/
	84	#define ZAP_LEAF_CHUNK(l, idx) \
	85	((zap_leaf_chunk_t *) \
	86	((l)->l_phys->l_hash + ZAP_LEAF_HASH_NUMENTRIES(l)))[idx]
	87	#define ZAP_LEAF_ENTRY(l, idx) (&ZAP_LEAF_CHUNK(l, idx).l_entry)
	88
	89	typedef enum zap_chunk_type {
	90	ZAP_CHUNK_FREE = 253,
	91	ZAP_CHUNK_ENTRY = 252,
	92	ZAP_CHUNK_ARRAY = 251,
	93	ZAP_CHUNK_TYPE_MAX = 250
	94	} zap_chunk_type_t;
	95
	96	#define ZLF_ENTRIES_CDSORTED (1<<0)
	97
	98	/*
	99	* TAKE NOTE:
	100	* If zap_leaf_phys_t is modified, zap_leaf_byteswap() must be modified.
101	*/
102	typedef struct zap_leaf_phys {
103	struct zap_leaf_header {
d3cc8b15	104	/* Public to ZAP */
34dc7c2f BB	105	uint64_t lh_block_type; /* ZBT_LEAF */
	106	uint64_t lh_pad1;
	107	uint64_t lh_prefix; /* hash prefix of this leaf */
	108	uint32_t lh_magic; /* ZAP_LEAF_MAGIC */
	109	uint16_t lh_nfree; /* number free chunks */
	110	uint16_t lh_nentries; /* number of entries */
	111	uint16_t lh_prefix_len; /* num bits used to id this */
	112
d3cc8b15	113	/* Private to zap_leaf */
34dc7c2f BB	114	uint16_t lh_freelist; /* chunk head of free list */
	115	uint8_t lh_flags; /* ZLF_* flags */
	116	uint8_t lh_pad2[11];
	117	} l_hdr; /* 2 24-byte chunks */
	118
	119	/*
	120	* The header is followed by a hash table with
	121	* ZAP_LEAF_HASH_NUMENTRIES(zap) entries. The hash table is
	122	* followed by an array of ZAP_LEAF_NUMCHUNKS(zap)
	123	* zap_leaf_chunk structures. These structures are accessed
	124	* with the ZAP_LEAF_CHUNK() macro.
	125	*/
	126
	127	uint16_t l_hash[1];
	128	} zap_leaf_phys_t;
	129
	130	typedef union zap_leaf_chunk {
	131	struct zap_leaf_entry {
	132	uint8_t le_type; /* always ZAP_CHUNK_ENTRY */
428870ff	133	uint8_t le_value_intlen; /* size of value's ints */
34dc7c2f BB	134	uint16_t le_next; /* next entry in hash chain */
34dc7c2f BB	135	uint16_t le_name_chunk; /* first chunk of the name */
428870ff	136	uint16_t le_name_numints; /* ints in name (incl null) */
34dc7c2f	137	uint16_t le_value_chunk; /* first chunk of the value */
428870ff	138	uint16_t le_value_numints; /* value length in ints */
34dc7c2f BB	139	uint32_t le_cd; /* collision differentiator */
	140	uint64_t le_hash; /* hash value of the name */
	141	} l_entry;
	142	struct zap_leaf_array {
	143	uint8_t la_type; /* always ZAP_CHUNK_ARRAY */
	144	uint8_t la_array[ZAP_LEAF_ARRAY_BYTES];
	145	uint16_t la_next; /* next blk or CHAIN_END */
	146	} l_array;
	147	struct zap_leaf_free {
	148	uint8_t lf_type; /* always ZAP_CHUNK_FREE */
	149	uint8_t lf_pad[ZAP_LEAF_ARRAY_BYTES];
	150	uint16_t lf_next; /* next in free list, or CHAIN_END */
	151	} l_free;
	152	} zap_leaf_chunk_t;
	153
	154	typedef struct zap_leaf {
b128c09f	155	krwlock_t l_rwlock;
34dc7c2f BB	156	uint64_t l_blkid; /* 1<<ZAP_BLOCK_SHIFT byte block off */
	157	int l_bs; /* block size shift */
	158	dmu_buf_t *l_dbuf;
	159	zap_leaf_phys_t *l_phys;
	160	} zap_leaf_t;
	161
	162
	163	typedef struct zap_entry_handle {
d3cc8b15	164	/* Set by zap_leaf and public to ZAP */
34dc7c2f BB	165	uint64_t zeh_num_integers;
	166	uint64_t zeh_hash;
	167	uint32_t zeh_cd;
	168	uint8_t zeh_integer_size;
	169
d3cc8b15	170	/* Private to zap_leaf */
34dc7c2f BB	171	uint16_t zeh_fakechunk;
	172	uint16_t *zeh_chunkp;
	173	zap_leaf_t *zeh_leaf;
	174	} zap_entry_handle_t;
	175
	176	/*
	177	* Return a handle to the named entry, or ENOENT if not found. The hash
	178	* value must equal zap_hash(name).
	179	*/
	180	extern int zap_leaf_lookup(zap_leaf_t *l,
428870ff	181	struct zap_name zn, zap_entry_handle_t zeh);
34dc7c2f BB	182
	183	/*
	184	* Return a handle to the entry with this hash+cd, or the entry with the
	185	* next closest hash+cd.
	186	*/
	187	extern int zap_leaf_lookup_closest(zap_leaf_t *l,
	188	uint64_t hash, uint32_t cd, zap_entry_handle_t *zeh);
	189
	190	/*
	191	* Read the first num_integers in the attribute. Integer size
	192	* conversion will be done without sign extension. Return EINVAL if
	193	* integer_size is too small. Return EOVERFLOW if there are more than
	194	* num_integers in the attribute.
	195	*/
	196	extern int zap_entry_read(const zap_entry_handle_t *zeh,
428870ff	197	uint8_t integer_size, uint64_t num_integers, void *buf);
34dc7c2f	198
428870ff BB	199	extern int zap_entry_read_name(struct zap zap, const zap_entry_handle_t zeh,
428870ff BB	200	uint16_t buflen, char *buf);
34dc7c2f BB	201
	202	/*
	203	* Replace the value of an existing entry.
	204	*
d3cc8b15	205	* May fail if it runs out of space (ENOSPC).
34dc7c2f BB	206	*/
34dc7c2f BB	207	extern int zap_entry_update(zap_entry_handle_t *zeh,
428870ff	208	uint8_t integer_size, uint64_t num_integers, const void *buf);
34dc7c2f BB	209
	210	/*
	211	* Remove an entry.
	212	*/
	213	extern void zap_entry_remove(zap_entry_handle_t *zeh);
	214
	215	/*
	216	* Create an entry. An equal entry must not exist, and this entry must
	217	* belong in this leaf (according to its hash value). Fills in the
	218	* entry handle on success. Returns 0 on success or ENOSPC on failure.
	219	*/
428870ff BB	220	extern int zap_entry_create(zap_leaf_t l, struct zap_name zn, uint32_t cd,
	221	uint8_t integer_size, uint64_t num_integers, const void *buf,
	222	zap_entry_handle_t *zeh);
34dc7c2f	223
d3cc8b15	224	/* Determine whether there is another entry with the same normalized form. */
34dc7c2f	225	extern boolean_t zap_entry_normalization_conflict(zap_entry_handle_t *zeh,
428870ff	226	struct zap_name zn, const char name, struct zap *zap);
34dc7c2f BB	227
	228	/*
	229	* Other stuff.
	230	*/
	231
	232	extern void zap_leaf_init(zap_leaf_t *l, boolean_t sort);
	233	extern void zap_leaf_byteswap(zap_leaf_phys_t *buf, int len);
	234	extern void zap_leaf_split(zap_leaf_t l, zap_leaf_t nl, boolean_t sort);
428870ff BB	235	extern void zap_leaf_stats(struct zap zap, zap_leaf_t l,
428870ff BB	236	struct zap_stats *zs);
34dc7c2f BB	237
	238	#ifdef __cplusplus
	239	}
	240	#endif
	241
	242	#endif /* _SYS_ZAP_LEAF_H */