andre@3: /*-
andre@3:  * Copyright (c) 1990, 1993, 1994
andre@3:  *	The Regents of the University of California.  All rights reserved.
andre@3:  *
andre@3:  * This code is derived from software contributed to Berkeley by
andre@3:  * Margo Seltzer.
andre@3:  *
andre@3:  * Redistribution and use in source and binary forms, with or without
andre@3:  * modification, are permitted provided that the following conditions
andre@3:  * are met:
andre@3:  * 1. Redistributions of source code must retain the above copyright
andre@3:  *    notice, this list of conditions and the following disclaimer.
andre@3:  * 2. Redistributions in binary form must reproduce the above copyright
andre@3:  *    notice, this list of conditions and the following disclaimer in the
andre@3:  *    documentation and/or other materials provided with the distribution.
andre@3:  * 3. ***REMOVED*** - see 
andre@3:  *    ftp://ftp.cs.berkeley.edu/pub/4bsd/README.Impt.License.Change
andre@3:  * 4. Neither the name of the University nor the names of its contributors
andre@3:  *    may be used to endorse or promote products derived from this software
andre@3:  *    without specific prior written permission.
andre@3:  *
andre@3:  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
andre@3:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
andre@3:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
andre@3:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
andre@3:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
andre@3:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
andre@3:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
andre@3:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
andre@3:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
andre@3:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
andre@3:  * SUCH DAMAGE.
andre@3:  *
andre@3:  *	@(#)hash.h	8.3 (Berkeley) 5/31/94
andre@3:  */
andre@3: 
andre@3: /* Operations */
andre@3: 
andre@3: #include <stdio.h>
andre@4: #include "cdefs.h"
andre@3: #include "mcom_db.h"
andre@3: typedef enum {
andre@3: 	HASH_GET, HASH_PUT, HASH_PUTNEW, HASH_DELETE, HASH_FIRST, HASH_NEXT
andre@3: } ACTION;
andre@3: 
andre@3: /* Buffer Management structures */
andre@3: typedef struct _bufhead BUFHEAD;
andre@3: 
andre@3: struct _bufhead {
andre@3: 	BUFHEAD		*prev;		/* LRU links */
andre@3: 	BUFHEAD		*next;		/* LRU links */
andre@3: 	BUFHEAD		*ovfl;		/* Overflow page buffer header */
andre@3: 	uint32	 	addr;		/* Address of this page */
andre@3: 	char		*page;		/* Actual page data */
andre@3: 	char     	is_disk;
andre@3: 	char	 	flags;
andre@3: #define	BUF_MOD		0x0001
andre@3: #define BUF_DISK	0x0002
andre@3: #define	BUF_BUCKET	0x0004
andre@3: #define	BUF_PIN		0x0008
andre@3: };
andre@3: 
andre@3: #define IS_BUCKET(X)	((X) & BUF_BUCKET)
andre@3: 
andre@3: typedef BUFHEAD **SEGMENT;
andre@3: 
andre@3: typedef int DBFILE_PTR;
andre@3: #define NO_FILE -1
andre@3: #ifdef macintosh
andre@3: #define DBFILE_OPEN(path, flag,mode) open((path), flag)
andre@3: #define EXISTS(path)
andre@3: #else
andre@3: #define DBFILE_OPEN(path, flag,mode) open((path), (flag), (mode))
andre@3: #endif
andre@3: /* Hash Table Information */
andre@3: typedef struct hashhdr {		/* Disk resident portion */
andre@3: 	int32		magic;		/* Magic NO for hash tables */
andre@3: 	int32		version;	/* Version ID */
andre@3: 	uint32		lorder;		/* Byte Order */
andre@3: 	int32		bsize;		/* Bucket/Page Size */
andre@3: 	int32		bshift;		/* Bucket shift */
andre@3: 	int32		dsize;		/* Directory Size */
andre@3: 	int32		ssize;		/* Segment Size */
andre@3: 	int32		sshift;		/* Segment shift */
andre@3: 	int32		ovfl_point;	/* Where overflow pages are being 
andre@3: 					 * allocated */
andre@3: 	int32		last_freed;	/* Last overflow page freed */
andre@3: 	int32		max_bucket;	/* ID of Maximum bucket in use */
andre@3: 	int32		high_mask;	/* Mask to modulo into entire table */
andre@3: 	int32		low_mask;	/* Mask to modulo into lower half of 
andre@3: 					 * table */
andre@3: 	int32		ffactor;	/* Fill factor */
andre@3: 	int32		nkeys;		/* Number of keys in hash table */
andre@3: 	int32		hdrpages;	/* Size of table header */
andre@3: 	uint32		h_charkey;	/* value of hash(CHARKEY) */
andre@3: #define NCACHED	32			/* number of bit maps and spare 
andre@3: 					 * points */
andre@3: 	int32		spares[NCACHED];/* spare pages for overflow */
andre@3: 	uint16		bitmaps[NCACHED];	/* address of overflow page 
andre@3: 						 * bitmaps */
andre@3: } HASHHDR;
andre@3: 
andre@3: typedef struct htab	 {		/* Memory resident data structure */
andre@3: 	HASHHDR 	hdr;		/* Header */
andre@3: 	int		nsegs;		/* Number of allocated segments */
andre@3: 	int		exsegs;		/* Number of extra allocated 
andre@3: 					 * segments */
andre@3: 	uint32			/* Hash function */
andre@3: 	    (*hash)(const void *, size_t);
andre@3: 	int		flags;		/* Flag values */
andre@3: 	DBFILE_PTR	fp;		/* File pointer */
andre@3: 	char            *filename;
andre@3: 	char		*tmp_buf;	/* Temporary Buffer for BIG data */
andre@3: 	char		*tmp_key;	/* Temporary Buffer for BIG keys */
andre@3: 	BUFHEAD 	*cpage;		/* Current page */
andre@3: 	int		cbucket;	/* Current bucket */
andre@3: 	int		cndx;		/* Index of next item on cpage */
andre@3: 	int		dbmerrno;		/* Error Number -- for DBM 
andre@3: 					 * compatability */
andre@3: 	int		new_file;	/* Indicates if fd is backing store 
andre@3: 					 * or no */
andre@3: 	int		save_file;	/* Indicates whether we need to flush 
andre@3: 					 * file at
andre@3: 					 * exit */
andre@3: 	uint32		*mapp[NCACHED];	/* Pointers to page maps */
andre@3: 	int		nmaps;		/* Initial number of bitmaps */
andre@3: 	int		nbufs;		/* Number of buffers left to 
andre@3: 					 * allocate */
andre@3: 	BUFHEAD 	bufhead;	/* Header of buffer lru list */
andre@3: 	SEGMENT 	*dir;		/* Hash Bucket directory */
andre@3: 	off_t		file_size;	/* in bytes */
andre@3: 	char		is_temp;	/* unlink file on close */
andre@3: 	char		updateEOF;	/* force EOF update on flush */
andre@3: } HTAB;
andre@3: 
andre@3: /*
andre@3:  * Constants
andre@3:  */
andre@3: #define DATABASE_CORRUPTED_ERROR -999   /* big ugly abort, delete database */
andre@3: #define	OLD_MAX_BSIZE		65536		/* 2^16 */
andre@3: #define MAX_BSIZE       	32l*1024l         /* 2^15 */
andre@3: #define MIN_BUFFERS		6
andre@3: #define MINHDRSIZE		512
andre@3: #define DEF_BUFSIZE		65536l		/* 64 K */
andre@3: #define DEF_BUCKET_SIZE		4096
andre@3: #define DEF_BUCKET_SHIFT	12		/* log2(BUCKET) */
andre@3: #define DEF_SEGSIZE		256
andre@3: #define DEF_SEGSIZE_SHIFT	8		/* log2(SEGSIZE)	 */
andre@3: #define DEF_DIRSIZE		256
andre@3: #define DEF_FFACTOR		65536l
andre@3: #define MIN_FFACTOR		4
andre@3: #define SPLTMAX			8
andre@3: #define CHARKEY			"%$sniglet^&"
andre@3: #define NUMKEY			1038583l
andre@3: #define BYTE_SHIFT		3
andre@3: #define INT_TO_BYTE		2
andre@3: #define INT_BYTE_SHIFT		5
andre@3: #define ALL_SET			((uint32)0xFFFFFFFF)
andre@3: #define ALL_CLEAR		0
andre@3: 
andre@3: #define PTROF(X)	((ptrdiff_t)(X) == BUF_DISK ? 0 : (X))
andre@3: #define ISDISK(X)	((X) ? ((ptrdiff_t)(X) == BUF_DISK ? BUF_DISK \
andre@3: 				: (X)->is_disk) : 0)
andre@3: 
andre@3: #define BITS_PER_MAP	32
andre@3: 
andre@3: /* Given the address of the beginning of a big map, clear/set the nth bit */
andre@3: #define CLRBIT(A, N)	((A)[(N)/BITS_PER_MAP] &= ~(1<<((N)%BITS_PER_MAP)))
andre@3: #define SETBIT(A, N)	((A)[(N)/BITS_PER_MAP] |= (1<<((N)%BITS_PER_MAP)))
andre@3: #define ISSET(A, N)	((A)[(N)/BITS_PER_MAP] & (1<<((N)%BITS_PER_MAP)))
andre@3: 
andre@3: /* Overflow management */
andre@3: /*
andre@3:  * Overflow page numbers are allocated per split point.  At each doubling of
andre@3:  * the table, we can allocate extra pages.  So, an overflow page number has
andre@3:  * the top 5 bits indicate which split point and the lower 11 bits indicate
andre@3:  * which page at that split point is indicated (pages within split points are
andre@3:  * numberered starting with 1).
andre@3:  */
andre@3: 
andre@3: #define SPLITSHIFT	11
andre@3: #define SPLITMASK	0x7FF
andre@3: #define SPLITNUM(N)	(((uint32)(N)) >> SPLITSHIFT)
andre@3: #define OPAGENUM(N)	((N) & SPLITMASK)
andre@3: #define	OADDR_OF(S,O)	((uint32)((uint32)(S) << SPLITSHIFT) + (O))
andre@3: 
andre@3: #define BUCKET_TO_PAGE(B) \
andre@3: 	(B) + hashp->HDRPAGES + ((B) ? hashp->SPARES[__log2((uint32)((B)+1))-1] : 0)
andre@3: #define OADDR_TO_PAGE(B) 	\
andre@3: 	BUCKET_TO_PAGE ( (1 << SPLITNUM((B))) -1 ) + OPAGENUM((B));
andre@3: 
andre@3: /*
andre@3:  * page.h contains a detailed description of the page format.
andre@3:  *
andre@3:  * Normally, keys and data are accessed from offset tables in the top of
andre@3:  * each page which point to the beginning of the key and data.  There are
andre@3:  * four flag values which may be stored in these offset tables which indicate
andre@3:  * the following:
andre@3:  *
andre@3:  *
andre@3:  * OVFLPAGE	Rather than a key data pair, this pair contains
andre@3:  *		the address of an overflow page.  The format of
andre@3:  *		the pair is:
andre@3:  *		    OVERFLOW_PAGE_NUMBER OVFLPAGE
andre@3:  *
andre@3:  * PARTIAL_KEY	This must be the first key/data pair on a page
andre@3:  *		and implies that page contains only a partial key.
andre@3:  *		That is, the key is too big to fit on a single page
andre@3:  *		so it starts on this page and continues on the next.
andre@3:  *		The format of the page is:
andre@3:  *		    KEY_OFF PARTIAL_KEY OVFL_PAGENO OVFLPAGE
andre@3:  *		
andre@3:  *		    KEY_OFF -- offset of the beginning of the key
andre@3:  *		    PARTIAL_KEY -- 1
andre@3:  *		    OVFL_PAGENO - page number of the next overflow page
andre@3:  *		    OVFLPAGE -- 0
andre@3:  *
andre@3:  * FULL_KEY	This must be the first key/data pair on the page.  It
andre@3:  *		is used in two cases.
andre@3:  *
andre@3:  *		Case 1:
andre@3:  *		    There is a complete key on the page but no data
andre@3:  *		    (because it wouldn't fit).  The next page contains
andre@3:  *		    the data.
andre@3:  *
andre@3:  *		    Page format it:
andre@3:  *		    KEY_OFF FULL_KEY OVFL_PAGENO OVFL_PAGE
andre@3:  *
andre@3:  *		    KEY_OFF -- offset of the beginning of the key
andre@3:  *		    FULL_KEY -- 2
andre@3:  *		    OVFL_PAGENO - page number of the next overflow page
andre@3:  *		    OVFLPAGE -- 0
andre@3:  *
andre@3:  *		Case 2:
andre@3:  *		    This page contains no key, but part of a large
andre@3:  *		    data field, which is continued on the next page.
andre@3:  *
andre@3:  *		    Page format it:
andre@3:  *		    DATA_OFF FULL_KEY OVFL_PAGENO OVFL_PAGE
andre@3:  *
andre@3:  *		    KEY_OFF -- offset of the beginning of the data on
andre@3:  *				this page
andre@3:  *		    FULL_KEY -- 2
andre@3:  *		    OVFL_PAGENO - page number of the next overflow page
andre@3:  *		    OVFLPAGE -- 0
andre@3:  *
andre@3:  * FULL_KEY_DATA 
andre@3:  *		This must be the first key/data pair on the page.
andre@3:  *		There are two cases:
andre@3:  *
andre@3:  *		Case 1:
andre@3:  *		    This page contains a key and the beginning of the
andre@3:  *		    data field, but the data field is continued on the
andre@3:  *		    next page.
andre@3:  *
andre@3:  *		    Page format is:
andre@3:  *		    KEY_OFF FULL_KEY_DATA OVFL_PAGENO DATA_OFF
andre@3:  *
andre@3:  *		    KEY_OFF -- offset of the beginning of the key
andre@3:  *		    FULL_KEY_DATA -- 3
andre@3:  *		    OVFL_PAGENO - page number of the next overflow page
andre@3:  *		    DATA_OFF -- offset of the beginning of the data
andre@3:  *
andre@3:  *		Case 2:
andre@3:  *		    This page contains the last page of a big data pair.
andre@3:  *		    There is no key, only the  tail end of the data
andre@3:  *		    on this page.
andre@3:  *
andre@3:  *		    Page format is:
andre@3:  *		    DATA_OFF FULL_KEY_DATA <OVFL_PAGENO> <OVFLPAGE>
andre@3:  *
andre@3:  *		    DATA_OFF -- offset of the beginning of the data on
andre@3:  *				this page
andre@3:  *		    FULL_KEY_DATA -- 3
andre@3:  *		    OVFL_PAGENO - page number of the next overflow page
andre@3:  *		    OVFLPAGE -- 0
andre@3:  *
andre@3:  *		    OVFL_PAGENO and OVFLPAGE are optional (they are
andre@3:  *		    not present if there is no next page).
andre@3:  */
andre@3: 
andre@3: #define OVFLPAGE	0
andre@3: #define PARTIAL_KEY	1
andre@3: #define FULL_KEY	2
andre@3: #define FULL_KEY_DATA	3
andre@3: #define	REAL_KEY	4
andre@3: 
andre@3: /* Short hands for accessing structure */
andre@3: #undef BSIZE
andre@3: #define BSIZE		hdr.bsize
andre@3: #undef BSHIFT
andre@3: #define BSHIFT		hdr.bshift
andre@3: #define DSIZE		hdr.dsize
andre@3: #define SGSIZE		hdr.ssize
andre@3: #define SSHIFT		hdr.sshift
andre@3: #define LORDER		hdr.lorder
andre@3: #define OVFL_POINT	hdr.ovfl_point
andre@3: #define	LAST_FREED	hdr.last_freed
andre@3: #define MAX_BUCKET	hdr.max_bucket
andre@3: #define FFACTOR		hdr.ffactor
andre@3: #define HIGH_MASK	hdr.high_mask
andre@3: #define LOW_MASK	hdr.low_mask
andre@3: #define NKEYS		hdr.nkeys
andre@3: #define HDRPAGES	hdr.hdrpages
andre@3: #define SPARES		hdr.spares
andre@3: #define BITMAPS		hdr.bitmaps
andre@3: #define VERSION		hdr.version
andre@3: #define MAGIC		hdr.magic
andre@3: #define NEXT_FREE	hdr.next_free
andre@3: #define H_CHARKEY	hdr.h_charkey
andre@3: 
andre@3: extern uint32 (*__default_hash) (const void *, size_t);
andre@3: void __buf_init(HTAB *hashp, int32 nbytes);
andre@3: int __big_delete(HTAB *hashp, BUFHEAD *bufp);
andre@3: BUFHEAD * __get_buf(HTAB *hashp, uint32 addr, BUFHEAD *prev_bp, int newpage);
andre@3: uint32 __call_hash(HTAB *hashp, char *k, size_t len);
andre@3: #include "page.h"
andre@3: extern int __big_split(HTAB *hashp, BUFHEAD *op,BUFHEAD *np,
andre@3: BUFHEAD *big_keyp,uint32 addr,uint32   obucket, SPLIT_RETURN *ret);
andre@3: void __free_ovflpage(HTAB *hashp, BUFHEAD *obufp);
andre@3: BUFHEAD * __add_ovflpage(HTAB *hashp, BUFHEAD *bufp);
andre@3: int __big_insert(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val);
andre@3: int __expand_table(HTAB *hashp);
andre@3: uint32 __log2(uint32 num);
andre@3: void __reclaim_buf(HTAB *hashp, BUFHEAD *bp);
andre@3: int __get_page(HTAB *hashp, char * p, uint32 bucket, int is_bucket, int is_disk, int is_bitmap);
andre@3: int __put_page(HTAB *hashp, char *p, uint32 bucket, int is_bucket, int is_bitmap);
andre@3: int __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx);
andre@3: int __buf_free(HTAB *hashp, int do_free, int to_disk);
andre@3: int __find_bigpair(HTAB *hashp, BUFHEAD *bufp, int ndx, char *key, int size);
andre@3: uint16 __find_last_page(HTAB *hashp, BUFHEAD **bpp);
andre@3: int __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT * val);
andre@3: int __big_return(HTAB *hashp, BUFHEAD *bufp, int ndx, DBT *val, int set_current);
andre@3: int __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx);
andre@3: int __big_keydata(HTAB *hashp, BUFHEAD *bufp, DBT *key, DBT *val, int set);
andre@3: int __split_page(HTAB *hashp, uint32 obucket, uint32 nbucket);