diff options
author | Theo de Raadt <deraadt@cvs.openbsd.org> | 1995-10-18 08:53:40 +0000 |
---|---|---|
committer | Theo de Raadt <deraadt@cvs.openbsd.org> | 1995-10-18 08:53:40 +0000 |
commit | d6583bb2a13f329cf0332ef2570eb8bb8fc0e39c (patch) | |
tree | ece253b876159b39c620e62b6c9b1174642e070e /lib/libc/db |
initial import of NetBSD tree
Diffstat (limited to 'lib/libc/db')
56 files changed, 12622 insertions, 0 deletions
diff --git a/lib/libc/db/Makefile.inc b/lib/libc/db/Makefile.inc new file mode 100644 index 00000000000..6cdb4cf3013 --- /dev/null +++ b/lib/libc/db/Makefile.inc @@ -0,0 +1,11 @@ +# $NetBSD: Makefile.inc,v 1.5 1995/02/27 13:19:23 cgd Exp $ +# @(#)Makefile.inc 8.2 (Berkeley) 2/21/94 +# +CFLAGS+=-D__DBINTERFACE_PRIVATE + +.include "${.CURDIR}/db/btree/Makefile.inc" +.include "${.CURDIR}/db/db/Makefile.inc" +.include "${.CURDIR}/db/hash/Makefile.inc" +.include "${.CURDIR}/db/man/Makefile.inc" +.include "${.CURDIR}/db/mpool/Makefile.inc" +.include "${.CURDIR}/db/recno/Makefile.inc" diff --git a/lib/libc/db/README b/lib/libc/db/README new file mode 100644 index 00000000000..df1be282063 --- /dev/null +++ b/lib/libc/db/README @@ -0,0 +1,2 @@ +$NetBSD: README,v 1.2 1995/02/27 13:19:30 cgd Exp $ +(was symlink to VERSION. -- cgd) diff --git a/lib/libc/db/VERSION b/lib/libc/db/VERSION new file mode 100644 index 00000000000..440b71bad09 --- /dev/null +++ b/lib/libc/db/VERSION @@ -0,0 +1,113 @@ +# $NetBSD: VERSION,v 1.7 1995/02/27 13:19:33 cgd Exp $ +# @(#)VERSION 8.17 (Berkeley) 6/20/94 + +This is version 1.79 of the Berkeley DB code. + +For information on compiling and installing this software, see the file +PORT/README. + +If your version of the DB code doesn't have a copy of this version file, +it's really old, please update it! + +Newer versions of this software will periodically be made available by +anonymous ftp from ftp.cs.berkeley.edu. An archive in compressed format +is in ucb/4bsd/db.tar.Z, or in gzip format in ucb/4bsd/db.tar.gz. If +you'd like to receive announcements of future releases of this software, +send email to the contact address below. + +Email questions may be addressed to Keith Bostic at bostic@cs.berkeley.edu. + +============================================ +Distribution contents: + +Makefile.inc Ignore this, it's the 4.4BSD subsystem Makefile. +PORT The per OS/architecture directories to use to build + libdb.a, if you're not running 4.4BSD. See the file + PORT/README for more information. +README This file. +VERSION This file. +btree B+tree routines. +db Dbopen(3) interface routine. +doc USENIX papers, formatted manual pages. +hash Extended linear hashing routines. +man Unformatted manual pages. +mpool Memory pool routines. +recno Fixed/variable length record routines. +test Test package. + +============================================ +1.78 -> 1.79 Mon Jun 20 17:36:47 EDT 1994 + all: Minor cleanups of 1.78 for porting reasons; only + major change was inlining check of NULL pointer + so that __fix_realloc goes away. + +1.77 -> 1.78 Thu Jun 16 19:06:43 EDT 1994 + all: Move "standard" size typedef's into db.h. + +1.76 -> 1.77 Thu Jun 16 16:48:38 EDT 1994 + hash: Delete __init_ routine, has special meaning to OSF 2.0. + +1.74 -> 1.76 + all: Finish up the port to the Alpha. + +1.73 -> 1.74 + recno: Don't put the record if rec_search fails, in rec_rdelete. + Create fixed-length intermediate records past "end" of DB + correctly. + Realloc bug when reading in fixed records. + all: First cut at port to Alpha (64-bit architecture) using + 4.4BSD basic integral types typedef's. + Cast allocation pointers to shut up old compilers. + Rework PORT directory into OS/machine directories. + +1.72 -> 1.73 + btree: If enough duplicate records were inserted and then deleted + that internal pages had references to empty pages of the + duplicate keys, the search function ended up on the wrong + page. + +1.7 -> 1.72 12 Oct 1993 + hash: Support NET/2 hash formats. + +1.7 -> 1.71 16 Sep 1993 + btree/recno: + Fix bug in internal search routines that caused + return of invalid pointers. + +1.6 -> 1.7 07 Sep 1993 + hash: Fixed big key overflow bugs. + test: Portability hacks, rewrite test script, Makefile. + btree/recno: + Stop copying non-overflow key/data pairs. + PORT: Break PORT directory up into per architecture/OS + subdirectories. + +1.5 -> 1.6 06 Jun 1993 + hash: In PAIRFITS, the first comparison should look at (P)[2]. + The hash_realloc function was walking off the end of memory. + The overflow page number was wrong when bumping splitpoint. + +1.4 -> 1.5 23 May 1993 + hash: Set hash default fill factor dynamically. + recno: Fixed bug in sorted page splits. + Add page size parameter support. + Allow recno to specify the name of the underlying btree; + used for vi recovery. + btree/recno: + Support 64K pages. + btree/hash/recno: + Provide access to an underlying file descriptor. + Change sync routines to take a flag argument, recno + uses this to sync out the underlying btree. + +1.3 -> 1.4 10 May 1993 + recno: Delete the R_CURSORLOG flag from the recno interface. + Zero-length record fix for non-mmap reads. + Try and make SIZE_T_MAX test in open portable. + +1.2 -> 1.3 01 May 1993 + btree: Ignore user byte-order setting when reading already + existing database. Fixes to byte-order conversions. + +1.1 -> 1.2 15 Apr 1993 + No bug fixes, only compatibility hacks. diff --git a/lib/libc/db/btree/Makefile.inc b/lib/libc/db/btree/Makefile.inc new file mode 100644 index 00000000000..d187103e2d3 --- /dev/null +++ b/lib/libc/db/btree/Makefile.inc @@ -0,0 +1,8 @@ +# $NetBSD: Makefile.inc,v 1.5 1995/02/27 13:19:53 cgd Exp $ +# @(#)Makefile.inc 8.1 (Berkeley) 6/4/93 + +.PATH: ${.CURDIR}/db/btree + +SRCS+= bt_close.c bt_conv.c bt_debug.c bt_delete.c bt_get.c bt_open.c \ + bt_overflow.c bt_page.c bt_put.c bt_search.c bt_seq.c bt_split.c \ + bt_stack.c bt_utils.c diff --git a/lib/libc/db/btree/bt_close.c b/lib/libc/db/btree/bt_close.c new file mode 100644 index 00000000000..2b27327f7ed --- /dev/null +++ b/lib/libc/db/btree/bt_close.c @@ -0,0 +1,210 @@ +/* $NetBSD: bt_close.c,v 1.6 1995/02/27 13:19:59 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_close.c 8.3 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: bt_close.c,v 1.6 1995/02/27 13:19:59 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#include <db.h> +#include "btree.h" + +static int bt_meta __P((BTREE *)); + +/* + * BT_CLOSE -- Close a btree. + * + * Parameters: + * dbp: pointer to access method + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__bt_close(dbp) + DB *dbp; +{ + BTREE *t; + int fd; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* + * Delete any already deleted record that we've been saving + * because the cursor pointed to it. + */ + if (ISSET(t, B_DELCRSR) && __bt_crsrdel(t, &t->bt_bcursor)) + return (RET_ERROR); + + if (__bt_sync(dbp, 0) == RET_ERROR) + return (RET_ERROR); + + if (mpool_close(t->bt_mp) == RET_ERROR) + return (RET_ERROR); + + if (t->bt_stack) + free(t->bt_stack); + if (t->bt_kbuf) + free(t->bt_kbuf); + if (t->bt_dbuf) + free(t->bt_dbuf); + + fd = t->bt_fd; + free(t); + free(dbp); + return (close(fd) ? RET_ERROR : RET_SUCCESS); +} + +/* + * BT_SYNC -- sync the btree to disk. + * + * Parameters: + * dbp: pointer to access method + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +int +__bt_sync(dbp, flags) + const DB *dbp; + u_int flags; +{ + BTREE *t; + int status; + PAGE *h; + void *p; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* Sync doesn't currently take any flags. */ + if (flags != 0) { + errno = EINVAL; + return (RET_ERROR); + } + + if (ISSET(t, B_INMEM | B_RDONLY) || !ISSET(t, B_MODIFIED)) + return (RET_SUCCESS); + + if (ISSET(t, B_METADIRTY) && bt_meta(t) == RET_ERROR) + return (RET_ERROR); + + /* + * Nastiness. If the cursor has been marked for deletion, but not + * actually deleted, we have to make a copy of the page, delete the + * key/data item, sync the file, and then restore the original page + * contents. + */ + if (ISSET(t, B_DELCRSR)) { + if ((p = (void *)malloc(t->bt_psize)) == NULL) + return (RET_ERROR); + if ((h = mpool_get(t->bt_mp, t->bt_bcursor.pgno, 0)) == NULL) + return (RET_ERROR); + memmove(p, h, t->bt_psize); + if ((status = + __bt_dleaf(t, h, t->bt_bcursor.index)) == RET_ERROR) + goto ecrsr; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + } + + if ((status = mpool_sync(t->bt_mp)) == RET_SUCCESS) + CLR(t, B_MODIFIED); + +ecrsr: if (ISSET(t, B_DELCRSR)) { + if ((h = mpool_get(t->bt_mp, t->bt_bcursor.pgno, 0)) == NULL) + return (RET_ERROR); + memmove(h, p, t->bt_psize); + free(p); + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + } + return (status); +} + +/* + * BT_META -- write the tree meta data to disk. + * + * Parameters: + * t: tree + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +static int +bt_meta(t) + BTREE *t; +{ + BTMETA m; + void *p; + + if ((p = mpool_get(t->bt_mp, P_META, 0)) == NULL) + return (RET_ERROR); + + /* Fill in metadata. */ + m.m_magic = BTREEMAGIC; + m.m_version = BTREEVERSION; + m.m_psize = t->bt_psize; + m.m_free = t->bt_free; + m.m_nrecs = t->bt_nrecs; + m.m_flags = t->bt_flags & SAVEMETA; + + memmove(p, &m, sizeof(BTMETA)); + mpool_put(t->bt_mp, p, MPOOL_DIRTY); + return (RET_SUCCESS); +} diff --git a/lib/libc/db/btree/bt_conv.c b/lib/libc/db/btree/bt_conv.c new file mode 100644 index 00000000000..1d513865455 --- /dev/null +++ b/lib/libc/db/btree/bt_conv.c @@ -0,0 +1,227 @@ +/* $NetBSD: bt_conv.c,v 1.5 1995/02/27 13:20:04 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_conv.c 8.3 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: bt_conv.c,v 1.5 1995/02/27 13:20:04 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> + +#include <stdio.h> + +#include <db.h> +#include "btree.h" + +static void mswap __P((PAGE *)); + +/* + * __BT_BPGIN, __BT_BPGOUT -- + * Convert host-specific number layout to/from the host-independent + * format stored on disk. + * + * Parameters: + * t: tree + * pg: page number + * h: page to convert + */ +void +__bt_pgin(t, pg, pp) + void *t; + pgno_t pg; + void *pp; +{ + PAGE *h; + indx_t i, top; + u_char flags; + char *p; + + if (!ISSET(((BTREE *)t), B_NEEDSWAP)) + return; + if (pg == P_META) { + mswap(pp); + return; + } + + h = pp; + M_32_SWAP(h->pgno); + M_32_SWAP(h->prevpg); + M_32_SWAP(h->nextpg); + M_32_SWAP(h->flags); + M_16_SWAP(h->lower); + M_16_SWAP(h->upper); + + top = NEXTINDEX(h); + if ((h->flags & P_TYPE) == P_BINTERNAL) + for (i = 0; i < top; i++) { + M_16_SWAP(h->linp[i]); + p = (char *)GETBINTERNAL(h, i); + P_32_SWAP(p); + p += sizeof(u_int32_t); + P_32_SWAP(p); + p += sizeof(pgno_t); + if (*(u_char *)p & P_BIGKEY) { + p += sizeof(u_char); + P_32_SWAP(p); + p += sizeof(pgno_t); + P_32_SWAP(p); + } + } + else if ((h->flags & P_TYPE) == P_BLEAF) + for (i = 0; i < top; i++) { + M_16_SWAP(h->linp[i]); + p = (char *)GETBLEAF(h, i); + P_32_SWAP(p); + p += sizeof(u_int32_t); + P_32_SWAP(p); + p += sizeof(u_int32_t); + flags = *(u_char *)p; + if (flags & (P_BIGKEY | P_BIGDATA)) { + p += sizeof(u_char); + if (flags & P_BIGKEY) { + P_32_SWAP(p); + p += sizeof(pgno_t); + P_32_SWAP(p); + } + if (flags & P_BIGDATA) { + p += sizeof(u_int32_t); + P_32_SWAP(p); + p += sizeof(pgno_t); + P_32_SWAP(p); + } + } + } +} + +void +__bt_pgout(t, pg, pp) + void *t; + pgno_t pg; + void *pp; +{ + PAGE *h; + indx_t i, top; + u_char flags; + char *p; + + if (!ISSET(((BTREE *)t), B_NEEDSWAP)) + return; + if (pg == P_META) { + mswap(pp); + return; + } + + h = pp; + top = NEXTINDEX(h); + if ((h->flags & P_TYPE) == P_BINTERNAL) + for (i = 0; i < top; i++) { + p = (char *)GETBINTERNAL(h, i); + P_32_SWAP(p); + p += sizeof(u_int32_t); + P_32_SWAP(p); + p += sizeof(pgno_t); + if (*(u_char *)p & P_BIGKEY) { + p += sizeof(u_char); + P_32_SWAP(p); + p += sizeof(pgno_t); + P_32_SWAP(p); + } + M_16_SWAP(h->linp[i]); + } + else if ((h->flags & P_TYPE) == P_BLEAF) + for (i = 0; i < top; i++) { + p = (char *)GETBLEAF(h, i); + P_32_SWAP(p); + p += sizeof(u_int32_t); + P_32_SWAP(p); + p += sizeof(u_int32_t); + flags = *(u_char *)p; + if (flags & (P_BIGKEY | P_BIGDATA)) { + p += sizeof(u_char); + if (flags & P_BIGKEY) { + P_32_SWAP(p); + p += sizeof(pgno_t); + P_32_SWAP(p); + } + if (flags & P_BIGDATA) { + p += sizeof(u_int32_t); + P_32_SWAP(p); + p += sizeof(pgno_t); + P_32_SWAP(p); + } + } + M_16_SWAP(h->linp[i]); + } + + M_32_SWAP(h->pgno); + M_32_SWAP(h->prevpg); + M_32_SWAP(h->nextpg); + M_32_SWAP(h->flags); + M_16_SWAP(h->lower); + M_16_SWAP(h->upper); +} + +/* + * MSWAP -- Actually swap the bytes on the meta page. + * + * Parameters: + * p: page to convert + */ +static void +mswap(pg) + PAGE *pg; +{ + char *p; + + p = (char *)pg; + P_32_SWAP(p); /* m_magic */ + p += sizeof(u_int32_t); + P_32_SWAP(p); /* m_version */ + p += sizeof(u_int32_t); + P_32_SWAP(p); /* m_psize */ + p += sizeof(u_int32_t); + P_32_SWAP(p); /* m_free */ + p += sizeof(u_int32_t); + P_32_SWAP(p); /* m_nrecs */ + p += sizeof(u_int32_t); + P_32_SWAP(p); /* m_flags */ + p += sizeof(u_int32_t); +} diff --git a/lib/libc/db/btree/bt_debug.c b/lib/libc/db/btree/bt_debug.c new file mode 100644 index 00000000000..fb967122aaa --- /dev/null +++ b/lib/libc/db/btree/bt_debug.c @@ -0,0 +1,337 @@ +/* $NetBSD: bt_debug.c,v 1.5 1995/02/27 13:20:12 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_debug.c 8.3 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: bt_debug.c,v 1.5 1995/02/27 13:20:12 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "btree.h" + +#ifdef DEBUG +/* + * BT_DUMP -- Dump the tree + * + * Parameters: + * dbp: pointer to the DB + */ +void +__bt_dump(dbp) + DB *dbp; +{ + BTREE *t; + PAGE *h; + pgno_t i; + char *sep; + + t = dbp->internal; + (void)fprintf(stderr, "%s: pgsz %d", + ISSET(t, B_INMEM) ? "memory" : "disk", t->bt_psize); + if (ISSET(t, R_RECNO)) + (void)fprintf(stderr, " keys %lu", t->bt_nrecs); +#undef X +#define X(flag, name) \ + if (ISSET(t, flag)) { \ + (void)fprintf(stderr, "%s%s", sep, name); \ + sep = ", "; \ + } + if (t->bt_flags) { + sep = " flags ("; + X(B_DELCRSR, "DELCRSR"); + X(R_FIXLEN, "FIXLEN"); + X(B_INMEM, "INMEM"); + X(B_NODUPS, "NODUPS"); + X(B_RDONLY, "RDONLY"); + X(R_RECNO, "RECNO"); + X(B_SEQINIT, "SEQINIT"); + X(B_METADIRTY,"METADIRTY"); + (void)fprintf(stderr, ")\n"); + } +#undef X + + for (i = P_ROOT; (h = mpool_get(t->bt_mp, i, 0)) != NULL; ++i) { + __bt_dpage(h); + (void)mpool_put(t->bt_mp, h, 0); + } +} + +/* + * BT_DMPAGE -- Dump the meta page + * + * Parameters: + * h: pointer to the PAGE + */ +void +__bt_dmpage(h) + PAGE *h; +{ + BTMETA *m; + char *sep; + + m = (BTMETA *)h; + (void)fprintf(stderr, "magic %lx\n", m->m_magic); + (void)fprintf(stderr, "version %lu\n", m->m_version); + (void)fprintf(stderr, "psize %lu\n", m->m_psize); + (void)fprintf(stderr, "free %lu\n", m->m_free); + (void)fprintf(stderr, "nrecs %lu\n", m->m_nrecs); + (void)fprintf(stderr, "flags %lu", m->m_flags); +#undef X +#define X(flag, name) \ + if (m->m_flags & flag) { \ + (void)fprintf(stderr, "%s%s", sep, name); \ + sep = ", "; \ + } + if (m->m_flags) { + sep = " ("; + X(B_NODUPS, "NODUPS"); + X(R_RECNO, "RECNO"); + (void)fprintf(stderr, ")"); + } +} + +/* + * BT_DNPAGE -- Dump the page + * + * Parameters: + * n: page number to dump. + */ +void +__bt_dnpage(dbp, pgno) + DB *dbp; + pgno_t pgno; +{ + BTREE *t; + PAGE *h; + + t = dbp->internal; + if ((h = mpool_get(t->bt_mp, pgno, 0)) != NULL) { + __bt_dpage(h); + (void)mpool_put(t->bt_mp, h, 0); + } +} + +/* + * BT_DPAGE -- Dump the page + * + * Parameters: + * h: pointer to the PAGE + */ +void +__bt_dpage(h) + PAGE *h; +{ + BINTERNAL *bi; + BLEAF *bl; + RINTERNAL *ri; + RLEAF *rl; + indx_t cur, top; + char *sep; + + (void)fprintf(stderr, " page %d: (", h->pgno); +#undef X +#define X(flag, name) \ + if (h->flags & flag) { \ + (void)fprintf(stderr, "%s%s", sep, name); \ + sep = ", "; \ + } + sep = ""; + X(P_BINTERNAL, "BINTERNAL") /* types */ + X(P_BLEAF, "BLEAF") + X(P_RINTERNAL, "RINTERNAL") /* types */ + X(P_RLEAF, "RLEAF") + X(P_OVERFLOW, "OVERFLOW") + X(P_PRESERVE, "PRESERVE"); + (void)fprintf(stderr, ")\n"); +#undef X + + (void)fprintf(stderr, "\tprev %2d next %2d", h->prevpg, h->nextpg); + if (h->flags & P_OVERFLOW) + return; + + top = NEXTINDEX(h); + (void)fprintf(stderr, " lower %3d upper %3d nextind %d\n", + h->lower, h->upper, top); + for (cur = 0; cur < top; cur++) { + (void)fprintf(stderr, "\t[%03d] %4d ", cur, h->linp[cur]); + switch(h->flags & P_TYPE) { + case P_BINTERNAL: + bi = GETBINTERNAL(h, cur); + (void)fprintf(stderr, + "size %03d pgno %03d", bi->ksize, bi->pgno); + if (bi->flags & P_BIGKEY) + (void)fprintf(stderr, " (indirect)"); + else if (bi->ksize) + (void)fprintf(stderr, + " {%.*s}", (int)bi->ksize, bi->bytes); + break; + case P_RINTERNAL: + ri = GETRINTERNAL(h, cur); + (void)fprintf(stderr, "entries %03d pgno %03d", + ri->nrecs, ri->pgno); + break; + case P_BLEAF: + bl = GETBLEAF(h, cur); + if (bl->flags & P_BIGKEY) + (void)fprintf(stderr, + "big key page %lu size %u/", + *(pgno_t *)bl->bytes, + *(u_int32_t *)(bl->bytes + sizeof(pgno_t))); + else if (bl->ksize) + (void)fprintf(stderr, "%s/", bl->bytes); + if (bl->flags & P_BIGDATA) + (void)fprintf(stderr, + "big data page %lu size %u", + *(pgno_t *)(bl->bytes + bl->ksize), + *(u_int32_t *)(bl->bytes + bl->ksize + + sizeof(pgno_t))); + else if (bl->dsize) + (void)fprintf(stderr, "%.*s", + (int)bl->dsize, bl->bytes + bl->ksize); + break; + case P_RLEAF: + rl = GETRLEAF(h, cur); + if (rl->flags & P_BIGDATA) + (void)fprintf(stderr, + "big data page %lu size %u", + *(pgno_t *)rl->bytes, + *(u_int32_t *)(rl->bytes + sizeof(pgno_t))); + else if (rl->dsize) + (void)fprintf(stderr, + "%.*s", (int)rl->dsize, rl->bytes); + break; + } + (void)fprintf(stderr, "\n"); + } +} +#endif + +#ifdef STATISTICS +/* + * BT_STAT -- Gather/print the tree statistics + * + * Parameters: + * dbp: pointer to the DB + */ +void +__bt_stat(dbp) + DB *dbp; +{ + extern u_long bt_cache_hit, bt_cache_miss, bt_pfxsaved, bt_rootsplit; + extern u_long bt_sortsplit, bt_split; + BTREE *t; + PAGE *h; + pgno_t i, pcont, pinternal, pleaf; + u_long ifree, lfree, nkeys; + int levels; + + t = dbp->internal; + pcont = pinternal = pleaf = 0; + nkeys = ifree = lfree = 0; + for (i = P_ROOT; (h = mpool_get(t->bt_mp, i, 0)) != NULL; ++i) { + switch(h->flags & P_TYPE) { + case P_BINTERNAL: + case P_RINTERNAL: + ++pinternal; + ifree += h->upper - h->lower; + break; + case P_BLEAF: + case P_RLEAF: + ++pleaf; + lfree += h->upper - h->lower; + nkeys += NEXTINDEX(h); + break; + case P_OVERFLOW: + ++pcont; + break; + } + (void)mpool_put(t->bt_mp, h, 0); + } + + /* Count the levels of the tree. */ + for (i = P_ROOT, levels = 0 ;; ++levels) { + h = mpool_get(t->bt_mp, i, 0); + if (h->flags & (P_BLEAF|P_RLEAF)) { + if (levels == 0) + levels = 1; + (void)mpool_put(t->bt_mp, h, 0); + break; + } + i = ISSET(t, R_RECNO) ? + GETRINTERNAL(h, 0)->pgno : + GETBINTERNAL(h, 0)->pgno; + (void)mpool_put(t->bt_mp, h, 0); + } + + (void)fprintf(stderr, "%d level%s with %ld keys", + levels, levels == 1 ? "" : "s", nkeys); + if (ISSET(t, R_RECNO)) + (void)fprintf(stderr, " (%ld header count)", t->bt_nrecs); + (void)fprintf(stderr, + "\n%lu pages (leaf %ld, internal %ld, overflow %ld)\n", + pinternal + pleaf + pcont, pleaf, pinternal, pcont); + (void)fprintf(stderr, "%ld cache hits, %ld cache misses\n", + bt_cache_hit, bt_cache_miss); + (void)fprintf(stderr, "%ld splits (%ld root splits, %ld sort splits)\n", + bt_split, bt_rootsplit, bt_sortsplit); + pleaf *= t->bt_psize - BTDATAOFF; + if (pleaf) + (void)fprintf(stderr, + "%.0f%% leaf fill (%ld bytes used, %ld bytes free)\n", + ((double)(pleaf - lfree) / pleaf) * 100, + pleaf - lfree, lfree); + pinternal *= t->bt_psize - BTDATAOFF; + if (pinternal) + (void)fprintf(stderr, + "%.0f%% internal fill (%ld bytes used, %ld bytes free\n", + ((double)(pinternal - ifree) / pinternal) * 100, + pinternal - ifree, ifree); + if (bt_pfxsaved) + (void)fprintf(stderr, "prefix checking removed %lu bytes.\n", + bt_pfxsaved); +} +#endif diff --git a/lib/libc/db/btree/bt_delete.c b/lib/libc/db/btree/bt_delete.c new file mode 100644 index 00000000000..e5ced2bc5c9 --- /dev/null +++ b/lib/libc/db/btree/bt_delete.c @@ -0,0 +1,330 @@ +/* $NetBSD: bt_delete.c,v 1.6 1995/02/27 13:20:16 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_delete.c 8.4 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: bt_delete.c,v 1.6 1995/02/27 13:20:16 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stdio.h> +#include <string.h> + +#include <db.h> +#include "btree.h" + +static int bt_bdelete __P((BTREE *, const DBT *)); + +/* + * __BT_DELETE -- Delete the item(s) referenced by a key. + * + * Parameters: + * dbp: pointer to access method + * key: key to delete + * flags: R_CURSOR if deleting what the cursor references + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found. + */ +int +__bt_delete(dbp, key, flags) + const DB *dbp; + const DBT *key; + u_int flags; +{ + BTREE *t; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + if (ISSET(t, B_RDONLY)) { + errno = EPERM; + return (RET_ERROR); + } + + switch(flags) { + case 0: + status = bt_bdelete(t, key); + break; + case R_CURSOR: + /* + * If flags is R_CURSOR, delete the cursor; must already have + * started a scan and not have already deleted the record. For + * the delete cursor bit to have been set requires that the + * scan be initialized, so no reason to check. + */ + if (!ISSET(t, B_SEQINIT)) + goto einval; + status = ISSET(t, B_DELCRSR) ? + RET_SPECIAL : __bt_crsrdel(t, &t->bt_bcursor); + break; + default: +einval: errno = EINVAL; + return (RET_ERROR); + } + if (status == RET_SUCCESS) + SET(t, B_MODIFIED); + return (status); +} + +/* + * BT_BDELETE -- Delete all key/data pairs matching the specified key. + * + * Parameters: + * tree: tree + * key: key to delete + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found. + */ +static int +bt_bdelete(t, key) + BTREE *t; + const DBT *key; +{ + EPG *e, save; + PAGE *h; + pgno_t cpgno, pg; + indx_t cindex; + int deleted, dirty1, dirty2, exact; + + /* Find any matching record; __bt_search pins the page. */ + if ((e = __bt_search(t, key, &exact)) == NULL) + return (RET_ERROR); + if (!exact) { + mpool_put(t->bt_mp, e->page, 0); + return (RET_SPECIAL); + } + + /* + * Delete forward, then delete backward, from the found key. The + * ordering is so that the deletions don't mess up the page refs. + * The first loop deletes the key from the original page, the second + * unpins the original page. In the first loop, dirty1 is set if + * the original page is modified, and dirty2 is set if any subsequent + * pages are modified. In the second loop, dirty1 starts off set if + * the original page has been modified, and is set if any subsequent + * pages are modified. + * + * If find the key referenced by the cursor, don't delete it, just + * flag it for future deletion. The cursor page number is P_INVALID + * unless the sequential scan is initialized, so no reason to check. + * A special case is when the already deleted cursor record was the + * only record found. If so, then the delete opertion fails as no + * records were deleted. + * + * Cycle in place in the current page until the current record doesn't + * match the key or the page is empty. If the latter, walk forward, + * skipping empty pages and repeating until a record doesn't match + * the key or the end of the tree is reached. + */ + cpgno = t->bt_bcursor.pgno; + cindex = t->bt_bcursor.index; + save = *e; + dirty1 = 0; + for (h = e->page, deleted = 0;;) { + dirty2 = 0; + do { + if (h->pgno == cpgno && e->index == cindex) { + if (!ISSET(t, B_DELCRSR)) { + SET(t, B_DELCRSR); + deleted = 1; + } + ++e->index; + } else { + if (__bt_dleaf(t, h, e->index)) { + if (h->pgno != save.page->pgno) + mpool_put(t->bt_mp, h, dirty2); + mpool_put(t->bt_mp, save.page, dirty1); + return (RET_ERROR); + } + if (h->pgno == save.page->pgno) + dirty1 = MPOOL_DIRTY; + else + dirty2 = MPOOL_DIRTY; + deleted = 1; + } + } while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0); + + /* + * Quit if didn't find a match, no next page, or first key on + * the next page doesn't match. Don't unpin the original page + * unless an error occurs. + */ + if (e->index < NEXTINDEX(h)) + break; + for (;;) { + if ((pg = h->nextpg) == P_INVALID) + goto done1; + if (h->pgno != save.page->pgno) + mpool_put(t->bt_mp, h, dirty2); + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) { + mpool_put(t->bt_mp, save.page, dirty1); + return (RET_ERROR); + } + if (NEXTINDEX(h) != 0) { + e->page = h; + e->index = 0; + break; + } + } + + if (__bt_cmp(t, key, e) != 0) + break; + } + + /* + * Reach here with the original page and the last page referenced + * pinned (they may be the same). Release it if not the original. + */ +done1: if (h->pgno != save.page->pgno) + mpool_put(t->bt_mp, h, dirty2); + + /* + * Walk backwards from the record previous to the record returned by + * __bt_search, skipping empty pages, until a record doesn't match + * the key or reach the beginning of the tree. + */ + *e = save; + for (;;) { + if (e->index) + --e->index; + for (h = e->page; e->index; --e->index) { + if (__bt_cmp(t, key, e) != 0) + goto done2; + if (h->pgno == cpgno && e->index == cindex) { + if (!ISSET(t, B_DELCRSR)) { + SET(t, B_DELCRSR); + deleted = 1; + } + } else { + if (__bt_dleaf(t, h, e->index) == RET_ERROR) { + mpool_put(t->bt_mp, h, dirty1); + return (RET_ERROR); + } + if (h->pgno == save.page->pgno) + dirty1 = MPOOL_DIRTY; + deleted = 1; + } + } + + if ((pg = h->prevpg) == P_INVALID) + goto done2; + mpool_put(t->bt_mp, h, dirty1); + dirty1 = 0; + if ((e->page = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + e->index = NEXTINDEX(e->page); + } + + /* + * Reach here with the last page that was looked at pinned. Release + * it. + */ +done2: mpool_put(t->bt_mp, h, dirty1); + return (deleted ? RET_SUCCESS : RET_SPECIAL); +} + +/* + * __BT_DLEAF -- Delete a single record from a leaf page. + * + * Parameters: + * t: tree + * index: index on current page to delete + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +int +__bt_dleaf(t, h, index) + BTREE *t; + PAGE *h; + indx_t index; +{ + register BLEAF *bl; + register indx_t cnt, *ip, offset; + register u_int32_t nbytes; + char *from; + void *to; + + /* + * Delete a record from a btree leaf page. Internal records are never + * deleted from internal pages, regardless of the records that caused + * them to be added being deleted. Pages made empty by deletion are + * not reclaimed. They are, however, made available for reuse. + * + * Pack the remaining entries at the end of the page, shift the indices + * down, overwriting the deleted record and its index. If the record + * uses overflow pages, make them available for reuse. + */ + to = bl = GETBLEAF(h, index); + if (bl->flags & P_BIGKEY && __ovfl_delete(t, bl->bytes) == RET_ERROR) + return (RET_ERROR); + if (bl->flags & P_BIGDATA && + __ovfl_delete(t, bl->bytes + bl->ksize) == RET_ERROR) + return (RET_ERROR); + nbytes = NBLEAF(bl); + + /* + * Compress the key/data pairs. Compress and adjust the [BR]LEAF + * offsets. Reset the headers. + */ + from = (char *)h + h->upper; + memmove(from + nbytes, from, (char *)to - from); + h->upper += nbytes; + + offset = h->linp[index]; + for (cnt = index, ip = &h->linp[0]; cnt--; ++ip) + if (ip[0] < offset) + ip[0] += nbytes; + for (cnt = NEXTINDEX(h) - index; --cnt; ++ip) + ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1]; + h->lower -= sizeof(indx_t); + return (RET_SUCCESS); +} diff --git a/lib/libc/db/btree/bt_get.c b/lib/libc/db/btree/bt_get.c new file mode 100644 index 00000000000..2adc73feb7a --- /dev/null +++ b/lib/libc/db/btree/bt_get.c @@ -0,0 +1,244 @@ +/* $NetBSD: bt_get.c,v 1.6 1995/02/27 13:20:22 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_get.c 8.2 (Berkeley) 9/7/93"; +#else +static char rcsid[] = "$NetBSD: bt_get.c,v 1.6 1995/02/27 13:20:22 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stddef.h> +#include <stdio.h> + +#include <db.h> +#include "btree.h" + +/* + * __BT_GET -- Get a record from the btree. + * + * Parameters: + * dbp: pointer to access method + * key: key to find + * data: data to return + * flag: currently unused + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found. + */ +int +__bt_get(dbp, key, data, flags) + const DB *dbp; + const DBT *key; + DBT *data; + u_int flags; +{ + BTREE *t; + EPG *e; + int exact, status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* Get currently doesn't take any flags. */ + if (flags) { + errno = EINVAL; + return (RET_ERROR); + } + + if ((e = __bt_search(t, key, &exact)) == NULL) + return (RET_ERROR); + if (!exact) { + mpool_put(t->bt_mp, e->page, 0); + return (RET_SPECIAL); + } + + /* + * A special case is if we found the record but it's flagged for + * deletion. In this case, we want to find another record with the + * same key, if it exists. Rather than look around the tree we call + * __bt_first and have it redo the search, as __bt_first will not + * return keys marked for deletion. Slow, but should never happen. + */ + if (ISSET(t, B_DELCRSR) && e->page->pgno == t->bt_bcursor.pgno && + e->index == t->bt_bcursor.index) { + mpool_put(t->bt_mp, e->page, 0); + if ((e = __bt_first(t, key, &exact)) == NULL) + return (RET_ERROR); + if (!exact) + return (RET_SPECIAL); + } + + status = __bt_ret(t, e, NULL, data); + /* + * If the user is doing concurrent access, we copied the + * key/data, toss the page. + */ + if (ISSET(t, B_DB_LOCK)) + mpool_put(t->bt_mp, e->page, 0); + else + t->bt_pinned = e->page; + return (status); +} + +/* + * __BT_FIRST -- Find the first entry. + * + * Parameters: + * t: the tree + * key: the key + * + * Returns: + * The first entry in the tree greater than or equal to key. + */ +EPG * +__bt_first(t, key, exactp) + BTREE *t; + const DBT *key; + int *exactp; +{ + register PAGE *h; + register EPG *e; + EPG save; + pgno_t cpgno, pg; + indx_t cindex; + int found; + + /* + * Find any matching record; __bt_search pins the page. Only exact + * matches are tricky, otherwise just return the location of the key + * if it were to be inserted into the tree. + */ + if ((e = __bt_search(t, key, exactp)) == NULL) + return (NULL); + if (!*exactp) + return (e); + + if (ISSET(t, B_DELCRSR)) { + cpgno = t->bt_bcursor.pgno; + cindex = t->bt_bcursor.index; + } else { + cpgno = P_INVALID; + cindex = 0; /* GCC thinks it's uninitialized. */ + } + + /* + * Walk backwards, skipping empty pages, as long as the entry matches + * and there are keys left in the tree. Save a copy of each match in + * case we go too far. A special case is that we don't return a match + * on records that the cursor references that have already been flagged + * for deletion. + */ + save = *e; + h = e->page; + found = 0; + do { + if (cpgno != h->pgno || cindex != e->index) { + if (save.page->pgno != e->page->pgno) { + mpool_put(t->bt_mp, save.page, 0); + save = *e; + } else + save.index = e->index; + found = 1; + } + /* + * Make a special effort not to unpin the page the last (or + * original) match was on, but also make sure it's unpinned + * if an error occurs. + */ + while (e->index == 0) { + if (h->prevpg == P_INVALID) + goto done1; + if (h->pgno != save.page->pgno) + mpool_put(t->bt_mp, h, 0); + if ((h = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL) { + if (h->pgno == save.page->pgno) + mpool_put(t->bt_mp, save.page, 0); + return (NULL); + } + e->page = h; + e->index = NEXTINDEX(h); + } + --e->index; + } while (__bt_cmp(t, key, e) == 0); + + /* + * Reach here with the last page that was looked at pinned, which may + * or may not be the same as the last (or original) match page. If + * it's not useful, release it. + */ +done1: if (h->pgno != save.page->pgno) + mpool_put(t->bt_mp, h, 0); + + /* + * If still haven't found a record, the only possibility left is the + * next one. Move forward one slot, skipping empty pages and check. + */ + if (!found) { + h = save.page; + if (++save.index == NEXTINDEX(h)) { + do { + pg = h->nextpg; + mpool_put(t->bt_mp, h, 0); + if (pg == P_INVALID) { + *exactp = 0; + return (e); + } + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (NULL); + } while ((save.index = NEXTINDEX(h)) == 0); + save.page = h; + } + if (__bt_cmp(t, key, &save) != 0) { + *exactp = 0; + return (e); + } + } + *e = save; + *exactp = 1; + return (e); +} diff --git a/lib/libc/db/btree/bt_open.c b/lib/libc/db/btree/bt_open.c new file mode 100644 index 00000000000..63162702051 --- /dev/null +++ b/lib/libc/db/btree/bt_open.c @@ -0,0 +1,446 @@ +/* $NetBSD: bt_open.c,v 1.7 1995/02/27 13:20:27 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_open.c 8.7 (Berkeley) 6/16/94"; +#else +static char rcsid[] = "$NetBSD: bt_open.c,v 1.7 1995/02/27 13:20:27 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +/* + * Implementation of btree access method for 4.4BSD. + * + * The design here was originally based on that of the btree access method + * used in the Postgres database system at UC Berkeley. This implementation + * is wholly independent of the Postgres code. + */ + +#include <sys/param.h> +#include <sys/stat.h> + +#include <errno.h> +#include <fcntl.h> +#include <limits.h> +#include <signal.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#include <db.h> +#include "btree.h" + +static int byteorder __P((void)); +static int nroot __P((BTREE *)); +static int tmp __P((void)); + +/* + * __BT_OPEN -- Open a btree. + * + * Creates and fills a DB struct, and calls the routine that actually + * opens the btree. + * + * Parameters: + * fname: filename (NULL for in-memory trees) + * flags: open flag bits + * mode: open permission bits + * b: BTREEINFO pointer + * + * Returns: + * NULL on failure, pointer to DB on success. + * + */ +DB * +__bt_open(fname, flags, mode, openinfo, dflags) + const char *fname; + int flags, mode, dflags; + const BTREEINFO *openinfo; +{ + struct stat sb; + BTMETA m; + BTREE *t; + BTREEINFO b; + DB *dbp; + pgno_t ncache; + ssize_t nr; + int machine_lorder; + + t = NULL; + + /* + * Intention is to make sure all of the user's selections are okay + * here and then use them without checking. Can't be complete, since + * we don't know the right page size, lorder or flags until the backing + * file is opened. Also, the file's page size can cause the cachesize + * to change. + */ + machine_lorder = byteorder(); + if (openinfo) { + b = *openinfo; + + /* Flags: R_DUP. */ + if (b.flags & ~(R_DUP)) + goto einval; + + /* + * Page size must be indx_t aligned and >= MINPSIZE. Default + * page size is set farther on, based on the underlying file + * transfer size. + */ + if (b.psize && + (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || + b.psize & sizeof(indx_t) - 1)) + goto einval; + + /* Minimum number of keys per page; absolute minimum is 2. */ + if (b.minkeypage) { + if (b.minkeypage < 2) + goto einval; + } else + b.minkeypage = DEFMINKEYPAGE; + + /* If no comparison, use default comparison and prefix. */ + if (b.compare == NULL) { + b.compare = __bt_defcmp; + if (b.prefix == NULL) + b.prefix = __bt_defpfx; + } + + if (b.lorder == 0) + b.lorder = machine_lorder; + } else { + b.compare = __bt_defcmp; + b.cachesize = 0; + b.flags = 0; + b.lorder = machine_lorder; + b.minkeypage = DEFMINKEYPAGE; + b.prefix = __bt_defpfx; + b.psize = 0; + } + + /* Check for the ubiquitous PDP-11. */ + if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) + goto einval; + + /* Allocate and initialize DB and BTREE structures. */ + if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) + goto err; + memset(t, 0, sizeof(BTREE)); + t->bt_bcursor.pgno = P_INVALID; + t->bt_fd = -1; /* Don't close unopened fd on error. */ + t->bt_lorder = b.lorder; + t->bt_order = NOT; + t->bt_cmp = b.compare; + t->bt_pfx = b.prefix; + t->bt_rfd = -1; + + if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) + goto err; + t->bt_flags = 0; + if (t->bt_lorder != machine_lorder) + SET(t, B_NEEDSWAP); + + dbp->type = DB_BTREE; + dbp->internal = t; + dbp->close = __bt_close; + dbp->del = __bt_delete; + dbp->fd = __bt_fd; + dbp->get = __bt_get; + dbp->put = __bt_put; + dbp->seq = __bt_seq; + dbp->sync = __bt_sync; + + /* + * If no file name was supplied, this is an in-memory btree and we + * open a backing temporary file. Otherwise, it's a disk-based tree. + */ + if (fname) { + switch(flags & O_ACCMODE) { + case O_RDONLY: + SET(t, B_RDONLY); + break; + case O_RDWR: + break; + case O_WRONLY: + default: + goto einval; + } + + if ((t->bt_fd = open(fname, flags, mode)) < 0) + goto err; + + } else { + if ((flags & O_ACCMODE) != O_RDWR) + goto einval; + if ((t->bt_fd = tmp()) == -1) + goto err; + SET(t, B_INMEM); + } + + if (fcntl(t->bt_fd, F_SETFD, 1) == -1) + goto err; + + if (fstat(t->bt_fd, &sb)) + goto err; + if (sb.st_size) { + if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0) + goto err; + if (nr != sizeof(BTMETA)) + goto eftype; + + /* + * Read in the meta-data. This can change the notion of what + * the lorder, page size and flags are, and, when the page size + * changes, the cachesize value can change too. If the user + * specified the wrong byte order for an existing database, we + * don't bother to return an error, we just clear the NEEDSWAP + * bit. + */ + if (m.m_magic == BTREEMAGIC) + CLR(t, B_NEEDSWAP); + else { + SET(t, B_NEEDSWAP); + M_32_SWAP(m.m_magic); + M_32_SWAP(m.m_version); + M_32_SWAP(m.m_psize); + M_32_SWAP(m.m_free); + M_32_SWAP(m.m_nrecs); + M_32_SWAP(m.m_flags); + } + if (m.m_magic != BTREEMAGIC || m.m_version != BTREEVERSION) + goto eftype; + if (m.m_psize < MINPSIZE || m.m_psize > MAX_PAGE_OFFSET + 1 || + m.m_psize & sizeof(indx_t) - 1) + goto eftype; + if (m.m_flags & ~SAVEMETA) + goto eftype; + b.psize = m.m_psize; + t->bt_flags |= m.m_flags; + t->bt_free = m.m_free; + t->bt_nrecs = m.m_nrecs; + } else { + /* + * Set the page size to the best value for I/O to this file. + * Don't overflow the page offset type. + */ + if (b.psize == 0) { + b.psize = sb.st_blksize; + if (b.psize < MINPSIZE) + b.psize = MINPSIZE; + if (b.psize > MAX_PAGE_OFFSET + 1) + b.psize = MAX_PAGE_OFFSET + 1; + } + + /* Set flag if duplicates permitted. */ + if (!(b.flags & R_DUP)) + SET(t, B_NODUPS); + + t->bt_free = P_INVALID; + t->bt_nrecs = 0; + SET(t, B_METADIRTY); + } + + t->bt_psize = b.psize; + + /* Set the cache size; must be a multiple of the page size. */ + if (b.cachesize && b.cachesize & b.psize - 1) + b.cachesize += (~b.cachesize & b.psize - 1) + 1; + if (b.cachesize < b.psize * MINCACHE) + b.cachesize = b.psize * MINCACHE; + + /* Calculate number of pages to cache. */ + ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; + + /* + * The btree data structure requires that at least two keys can fit on + * a page, but other than that there's no fixed requirement. The user + * specified a minimum number per page, and we translated that into the + * number of bytes a key/data pair can use before being placed on an + * overflow page. This calculation includes the page header, the size + * of the index referencing the leaf item and the size of the leaf item + * structure. Also, don't let the user specify a minkeypage such that + * a key/data pair won't fit even if both key and data are on overflow + * pages. + */ + t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - + (sizeof(indx_t) + NBLEAFDBT(0, 0)); + if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) + t->bt_ovflsize = + NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); + + /* Initialize the buffer pool. */ + if ((t->bt_mp = + mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) + goto err; + if (!ISSET(t, B_INMEM)) + mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); + + /* Create a root page if new tree. */ + if (nroot(t) == RET_ERROR) + goto err; + + /* Global flags. */ + if (dflags & DB_LOCK) + SET(t, B_DB_LOCK); + if (dflags & DB_SHMEM) + SET(t, B_DB_SHMEM); + if (dflags & DB_TXN) + SET(t, B_DB_TXN); + + return (dbp); + +einval: errno = EINVAL; + goto err; + +eftype: errno = EFTYPE; + goto err; + +err: if (t) { + if (t->bt_dbp) + free(t->bt_dbp); + if (t->bt_fd != -1) + (void)close(t->bt_fd); + free(t); + } + return (NULL); +} + +/* + * NROOT -- Create the root of a new tree. + * + * Parameters: + * t: tree + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +static int +nroot(t) + BTREE *t; +{ + PAGE *meta, *root; + pgno_t npg; + + if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { + mpool_put(t->bt_mp, meta, 0); + return (RET_SUCCESS); + } + if (errno != EINVAL) /* It's OK to not exist. */ + return (RET_ERROR); + errno = 0; + + if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) + return (RET_ERROR); + + if ((root = mpool_new(t->bt_mp, &npg)) == NULL) + return (RET_ERROR); + + if (npg != P_ROOT) + return (RET_ERROR); + root->pgno = npg; + root->prevpg = root->nextpg = P_INVALID; + root->lower = BTDATAOFF; + root->upper = t->bt_psize; + root->flags = P_BLEAF; + memset(meta, 0, t->bt_psize); + mpool_put(t->bt_mp, meta, MPOOL_DIRTY); + mpool_put(t->bt_mp, root, MPOOL_DIRTY); + return (RET_SUCCESS); +} + +static int +tmp() +{ + sigset_t set, oset; + int fd; + char *envtmp; + char path[MAXPATHLEN]; + + envtmp = getenv("TMPDIR"); + (void)snprintf(path, + sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp"); + + (void)sigfillset(&set); + (void)sigprocmask(SIG_BLOCK, &set, &oset); + if ((fd = mkstemp(path)) != -1) + (void)unlink(path); + (void)sigprocmask(SIG_SETMASK, &oset, NULL); + return(fd); +} + +static int +byteorder() +{ + u_int32_t x; + u_char *p; + + x = 0x01020304; + p = (u_char *)&x; + switch (*p) { + case 1: + return (BIG_ENDIAN); + case 4: + return (LITTLE_ENDIAN); + default: + return (0); + } +} + +int +__bt_fd(dbp) + const DB *dbp; +{ + BTREE *t; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* In-memory database can't have a file descriptor. */ + if (ISSET(t, B_INMEM)) { + errno = ENOENT; + return (-1); + } + return (t->bt_fd); +} diff --git a/lib/libc/db/btree/bt_overflow.c b/lib/libc/db/btree/bt_overflow.c new file mode 100644 index 00000000000..b5c43c60435 --- /dev/null +++ b/lib/libc/db/btree/bt_overflow.c @@ -0,0 +1,234 @@ +/* $NetBSD: bt_overflow.c,v 1.5 1995/02/27 13:20:33 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_overflow.c 8.4 (Berkeley) 6/20/94"; +#else +static char rcsid[] = "$NetBSD: bt_overflow.c,v 1.5 1995/02/27 13:20:33 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "btree.h" + +/* + * Big key/data code. + * + * Big key and data entries are stored on linked lists of pages. The initial + * reference is byte string stored with the key or data and is the page number + * and size. The actual record is stored in a chain of pages linked by the + * nextpg field of the PAGE header. + * + * The first page of the chain has a special property. If the record is used + * by an internal page, it cannot be deleted and the P_PRESERVE bit will be set + * in the header. + * + * XXX + * A single DBT is written to each chain, so a lot of space on the last page + * is wasted. This is a fairly major bug for some data sets. + */ + +/* + * __OVFL_GET -- Get an overflow key/data item. + * + * Parameters: + * t: tree + * p: pointer to { pgno_t, u_int32_t } + * buf: storage address + * bufsz: storage size + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__ovfl_get(t, p, ssz, buf, bufsz) + BTREE *t; + void *p; + size_t *ssz; + char **buf; + size_t *bufsz; +{ + PAGE *h; + pgno_t pg; + size_t nb, plen; + u_int32_t sz; + + memmove(&pg, p, sizeof(pgno_t)); + memmove(&sz, (char *)p + sizeof(pgno_t), sizeof(u_int32_t)); + *ssz = sz; + +#ifdef DEBUG + if (pg == P_INVALID || sz == 0) + abort(); +#endif + /* Make the buffer bigger as necessary. */ + if (*bufsz < sz) { + *buf = (char *)(*buf == NULL ? malloc(sz) : realloc(*buf, sz)); + if (*buf == NULL) + return (RET_ERROR); + *bufsz = sz; + } + + /* + * Step through the linked list of pages, copying the data on each one + * into the buffer. Never copy more than the data's length. + */ + plen = t->bt_psize - BTDATAOFF; + for (p = *buf;; p = (char *)p + nb, pg = h->nextpg) { + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + + nb = MIN(sz, plen); + memmove(p, (char *)h + BTDATAOFF, nb); + mpool_put(t->bt_mp, h, 0); + + if ((sz -= nb) == 0) + break; + } + return (RET_SUCCESS); +} + +/* + * __OVFL_PUT -- Store an overflow key/data item. + * + * Parameters: + * t: tree + * data: DBT to store + * pgno: storage page number + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__ovfl_put(t, dbt, pg) + BTREE *t; + const DBT *dbt; + pgno_t *pg; +{ + PAGE *h, *last; + void *p; + pgno_t npg; + size_t nb, plen; + u_int32_t sz; + + /* + * Allocate pages and copy the key/data record into them. Store the + * number of the first page in the chain. + */ + plen = t->bt_psize - BTDATAOFF; + for (last = NULL, p = dbt->data, sz = dbt->size;; + p = (char *)p + plen, last = h) { + if ((h = __bt_new(t, &npg)) == NULL) + return (RET_ERROR); + + h->pgno = npg; + h->nextpg = h->prevpg = P_INVALID; + h->flags = P_OVERFLOW; + h->lower = h->upper = 0; + + nb = MIN(sz, plen); + memmove((char *)h + BTDATAOFF, p, nb); + + if (last) { + last->nextpg = h->pgno; + mpool_put(t->bt_mp, last, MPOOL_DIRTY); + } else + *pg = h->pgno; + + if ((sz -= nb) == 0) { + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + break; + } + } + return (RET_SUCCESS); +} + +/* + * __OVFL_DELETE -- Delete an overflow chain. + * + * Parameters: + * t: tree + * p: pointer to { pgno_t, u_int32_t } + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__ovfl_delete(t, p) + BTREE *t; + void *p; +{ + PAGE *h; + pgno_t pg; + size_t plen; + u_int32_t sz; + + memmove(&pg, p, sizeof(pgno_t)); + memmove(&sz, (char *)p + sizeof(pgno_t), sizeof(u_int32_t)); + +#ifdef DEBUG + if (pg == P_INVALID || sz == 0) + abort(); +#endif + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + + /* Don't delete chains used by internal pages. */ + if (h->flags & P_PRESERVE) { + mpool_put(t->bt_mp, h, 0); + return (RET_SUCCESS); + } + + /* Step through the chain, calling the free routine for each page. */ + for (plen = t->bt_psize - BTDATAOFF;; sz -= plen) { + pg = h->nextpg; + __bt_free(t, h); + if (sz <= plen) + break; + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + } + return (RET_SUCCESS); +} diff --git a/lib/libc/db/btree/bt_page.c b/lib/libc/db/btree/bt_page.c new file mode 100644 index 00000000000..d6a44941854 --- /dev/null +++ b/lib/libc/db/btree/bt_page.c @@ -0,0 +1,99 @@ +/* $NetBSD: bt_page.c,v 1.5 1995/02/27 13:20:36 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_page.c 8.2 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: bt_page.c,v 1.5 1995/02/27 13:20:36 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <stdio.h> + +#include <db.h> +#include "btree.h" + +/* + * __BT_FREE -- Put a page on the freelist. + * + * Parameters: + * t: tree + * h: page to free + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__bt_free(t, h) + BTREE *t; + PAGE *h; +{ + /* Insert the page at the start of the free list. */ + h->prevpg = P_INVALID; + h->nextpg = t->bt_free; + t->bt_free = h->pgno; + + /* Make sure the page gets written back. */ + return (mpool_put(t->bt_mp, h, MPOOL_DIRTY)); +} + +/* + * __BT_NEW -- Get a new page, preferably from the freelist. + * + * Parameters: + * t: tree + * npg: storage for page number. + * + * Returns: + * Pointer to a page, NULL on error. + */ +PAGE * +__bt_new(t, npg) + BTREE *t; + pgno_t *npg; +{ + PAGE *h; + + if (t->bt_free != P_INVALID && + (h = mpool_get(t->bt_mp, t->bt_free, 0)) != NULL) { + *npg = t->bt_free; + t->bt_free = h->nextpg; + return (h); + } + return (mpool_new(t->bt_mp, npg)); +} diff --git a/lib/libc/db/btree/bt_put.c b/lib/libc/db/btree/bt_put.c new file mode 100644 index 00000000000..9ddc98d8ec1 --- /dev/null +++ b/lib/libc/db/btree/bt_put.c @@ -0,0 +1,324 @@ +/* $NetBSD: bt_put.c,v 1.7 1995/02/27 13:20:40 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_put.c 8.4 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: bt_put.c,v 1.7 1995/02/27 13:20:40 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "btree.h" + +static EPG *bt_fast __P((BTREE *, const DBT *, const DBT *, int *)); + +/* + * __BT_PUT -- Add a btree item to the tree. + * + * Parameters: + * dbp: pointer to access method + * key: key + * data: data + * flag: R_NOOVERWRITE + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the + * tree and R_NOOVERWRITE specified. + */ +int +__bt_put(dbp, key, data, flags) + const DB *dbp; + DBT *key; + const DBT *data; + u_int flags; +{ + BTREE *t; + DBT tkey, tdata; + EPG *e; + PAGE *h; + indx_t index, nxtindex; + pgno_t pg; + u_int32_t nbytes; + int dflags, exact, status; + char *dest, db[NOVFLSIZE], kb[NOVFLSIZE]; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + switch (flags) { + case R_CURSOR: + if (!ISSET(t, B_SEQINIT)) + goto einval; + if (ISSET(t, B_DELCRSR)) + goto einval; + break; + case 0: + case R_NOOVERWRITE: + break; + default: +einval: errno = EINVAL; + return (RET_ERROR); + } + + if (ISSET(t, B_RDONLY)) { + errno = EPERM; + return (RET_ERROR); + } + + /* + * If the key/data won't fit on a page, store it on indirect pages. + * Only store the key on the overflow page if it's too big after the + * data is on an overflow page. + * + * XXX + * If the insert fails later on, these pages aren't recovered. + */ + dflags = 0; + if (key->size + data->size > t->bt_ovflsize) { + if (key->size > t->bt_ovflsize) { +storekey: if (__ovfl_put(t, key, &pg) == RET_ERROR) + return (RET_ERROR); + tkey.data = kb; + tkey.size = NOVFLSIZE; + memmove(kb, &pg, sizeof(pgno_t)); + memmove(kb + sizeof(pgno_t), + &key->size, sizeof(u_int32_t)); + dflags |= P_BIGKEY; + key = &tkey; + } + if (key->size + data->size > t->bt_ovflsize) { + if (__ovfl_put(t, data, &pg) == RET_ERROR) + return (RET_ERROR); + tdata.data = db; + tdata.size = NOVFLSIZE; + memmove(db, &pg, sizeof(pgno_t)); + memmove(db + sizeof(pgno_t), + &data->size, sizeof(u_int32_t)); + dflags |= P_BIGDATA; + data = &tdata; + } + if (key->size + data->size > t->bt_ovflsize) + goto storekey; + } + + /* Replace the cursor. */ + if (flags == R_CURSOR) { + if ((h = mpool_get(t->bt_mp, t->bt_bcursor.pgno, 0)) == NULL) + return (RET_ERROR); + index = t->bt_bcursor.index; + goto delete; + } + + /* + * Find the key to delete, or, the location at which to insert. Bt_fast + * and __bt_search pin the returned page. + */ + if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL) + if ((e = __bt_search(t, key, &exact)) == NULL) + return (RET_ERROR); + h = e->page; + index = e->index; + + /* + * Add the specified key/data pair to the tree. If an identical key + * is already in the tree, and R_NOOVERWRITE is set, an error is + * returned. If R_NOOVERWRITE is not set, the key is either added (if + * duplicates are permitted) or an error is returned. + * + * Pages are split as required. + */ + switch (flags) { + case R_NOOVERWRITE: + if (!exact) + break; + /* + * One special case is if the cursor references the record and + * it's been flagged for deletion. Then, we delete the record, + * leaving the cursor there -- this means that the inserted + * record will not be seen in a cursor scan. + */ + if (ISSET(t, B_DELCRSR) && t->bt_bcursor.pgno == h->pgno && + t->bt_bcursor.index == index) { + CLR(t, B_DELCRSR); + goto delete; + } + mpool_put(t->bt_mp, h, 0); + return (RET_SPECIAL); + default: + if (!exact || !ISSET(t, B_NODUPS)) + break; +delete: if (__bt_dleaf(t, h, index) == RET_ERROR) { + mpool_put(t->bt_mp, h, 0); + return (RET_ERROR); + } + break; + } + + /* + * If not enough room, or the user has put a ceiling on the number of + * keys permitted in the page, split the page. The split code will + * insert the key and data and unpin the current page. If inserting + * into the offset array, shift the pointers up. + */ + nbytes = NBLEAFDBT(key->size, data->size); + if (h->upper - h->lower < nbytes + sizeof(indx_t)) { + if ((status = __bt_split(t, h, key, + data, dflags, nbytes, index)) != RET_SUCCESS) + return (status); + goto success; + } + + if (index < (nxtindex = NEXTINDEX(h))) + memmove(h->linp + index + 1, h->linp + index, + (nxtindex - index) * sizeof(indx_t)); + h->lower += sizeof(indx_t); + + h->linp[index] = h->upper -= nbytes; + dest = (char *)h + h->upper; + WR_BLEAF(dest, key, data, dflags); + + if (t->bt_order == NOT) + if (h->nextpg == P_INVALID) { + if (index == NEXTINDEX(h) - 1) { + t->bt_order = FORWARD; + t->bt_last.index = index; + t->bt_last.pgno = h->pgno; + } + } else if (h->prevpg == P_INVALID) { + if (index == 0) { + t->bt_order = BACK; + t->bt_last.index = 0; + t->bt_last.pgno = h->pgno; + } + } + + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + +success: + if (flags == R_SETCURSOR) { + t->bt_bcursor.pgno = e->page->pgno; + t->bt_bcursor.index = e->index; + } + SET(t, B_MODIFIED); + return (RET_SUCCESS); +} + +#ifdef STATISTICS +u_long bt_cache_hit, bt_cache_miss; +#endif + +/* + * BT_FAST -- Do a quick check for sorted data. + * + * Parameters: + * t: tree + * key: key to insert + * + * Returns: + * EPG for new record or NULL if not found. + */ +static EPG * +bt_fast(t, key, data, exactp) + BTREE *t; + const DBT *key, *data; + int *exactp; +{ + PAGE *h; + u_int32_t nbytes; + int cmp; + + if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) { + t->bt_order = NOT; + return (NULL); + } + t->bt_cur.page = h; + t->bt_cur.index = t->bt_last.index; + + /* + * If won't fit in this page or have too many keys in this page, have + * to search to get split stack. + */ + nbytes = NBLEAFDBT(key->size, data->size); + if (h->upper - h->lower < nbytes + sizeof(indx_t)) + goto miss; + + if (t->bt_order == FORWARD) { + if (t->bt_cur.page->nextpg != P_INVALID) + goto miss; + if (t->bt_cur.index != NEXTINDEX(h) - 1) + goto miss; + if ((cmp = __bt_cmp(t, key, &t->bt_cur)) < 0) + goto miss; + t->bt_last.index = cmp ? ++t->bt_cur.index : t->bt_cur.index; + } else { + if (t->bt_cur.page->prevpg != P_INVALID) + goto miss; + if (t->bt_cur.index != 0) + goto miss; + if ((cmp = __bt_cmp(t, key, &t->bt_cur)) > 0) + goto miss; + t->bt_last.index = 0; + } + *exactp = cmp == 0; +#ifdef STATISTICS + ++bt_cache_hit; +#endif + return (&t->bt_cur); + +miss: +#ifdef STATISTICS + ++bt_cache_miss; +#endif + t->bt_order = NOT; + mpool_put(t->bt_mp, h, 0); + return (NULL); +} diff --git a/lib/libc/db/btree/bt_search.c b/lib/libc/db/btree/bt_search.c new file mode 100644 index 00000000000..c98fe7346ea --- /dev/null +++ b/lib/libc/db/btree/bt_search.c @@ -0,0 +1,241 @@ +/* $NetBSD: bt_search.c,v 1.7 1995/02/27 13:20:44 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_search.c 8.6 (Berkeley) 3/15/94"; +#else +static char rcsid[] = "$NetBSD: bt_search.c,v 1.7 1995/02/27 13:20:44 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <stdio.h> + +#include <db.h> +#include "btree.h" + +static int bt_snext __P((BTREE *, PAGE *, const DBT *, int *)); +static int bt_sprev __P((BTREE *, PAGE *, const DBT *, int *)); + +/* + * __BT_SEARCH -- Search a btree for a key. + * + * Parameters: + * t: tree to search + * key: key to find + * exactp: pointer to exact match flag + * + * Returns: + * The EPG for matching record, if any, or the EPG for the location + * of the key, if it were inserted into the tree, is entered into + * the bt_cur field of the tree. A pointer to the field is returned. + */ +EPG * +__bt_search(t, key, exactp) + BTREE *t; + const DBT *key; + int *exactp; +{ + PAGE *h; + indx_t base, index, lim; + pgno_t pg; + int cmp; + + BT_CLR(t); + for (pg = P_ROOT;;) { + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (NULL); + + /* Do a binary search on the current page. */ + t->bt_cur.page = h; + for (base = 0, lim = NEXTINDEX(h); lim; lim >>= 1) { + t->bt_cur.index = index = base + (lim >> 1); + if ((cmp = __bt_cmp(t, key, &t->bt_cur)) == 0) { + if (h->flags & P_BLEAF) { + *exactp = 1; + return (&t->bt_cur); + } + goto next; + } + if (cmp > 0) { + base = index + 1; + --lim; + } + } + + /* + * If it's a leaf page, and duplicates aren't allowed, we're + * done. If duplicates are allowed, it's possible that there + * were duplicate keys on duplicate pages, and they were later + * deleted, so we could be on a page with no matches while + * there are matches on other pages. If we're at the start or + * end of a page, check on both sides. + */ + if (h->flags & P_BLEAF) { + t->bt_cur.index = base; + *exactp = 0; + if (!ISSET(t, B_NODUPS)) { + if (base == 0 && + bt_sprev(t, h, key, exactp)) + return (&t->bt_cur); + if (base == NEXTINDEX(h) && + bt_snext(t, h, key, exactp)) + return (&t->bt_cur); + } + return (&t->bt_cur); + } + + /* + * No match found. Base is the smallest index greater than + * key and may be zero or a last + 1 index. If it's non-zero, + * decrement by one, and record the internal page which should + * be a parent page for the key. If a split later occurs, the + * inserted page will be to the right of the saved page. + */ + index = base ? base - 1 : base; + +next: if (__bt_push(t, h->pgno, index) == RET_ERROR) + return (NULL); + pg = GETBINTERNAL(h, index)->pgno; + mpool_put(t->bt_mp, h, 0); + } +} + +/* + * BT_SNEXT -- Check for an exact match after the key. + * + * Parameters: + * t: tree to search + * h: current page. + * key: key to find + * exactp: pointer to exact match flag + * + * Returns: + * If an exact match found. + */ +static int +bt_snext(t, h, key, exactp) + BTREE *t; + PAGE *h; + const DBT *key; + int *exactp; +{ + EPG e; + PAGE *tp; + pgno_t pg; + + /* Skip until reach the end of the tree or a key. */ + for (pg = h->nextpg; pg != P_INVALID;) { + if ((tp = mpool_get(t->bt_mp, pg, 0)) == NULL) { + mpool_put(t->bt_mp, h, 0); + return (NULL); + } + if (NEXTINDEX(tp) != 0) + break; + pg = tp->prevpg; + mpool_put(t->bt_mp, tp, 0); + } + /* + * The key is either an exact match, or not as good as + * the one we already have. + */ + if (pg != P_INVALID) { + e.page = tp; + e.index = NEXTINDEX(tp) - 1; + if (__bt_cmp(t, key, &e) == 0) { + mpool_put(t->bt_mp, h, 0); + t->bt_cur = e; + *exactp = 1; + return (1); + } + } + return (0); +} + +/* + * BT_SPREV -- Check for an exact match before the key. + * + * Parameters: + * t: tree to search + * h: current page. + * key: key to find + * exactp: pointer to exact match flag + * + * Returns: + * If an exact match found. + */ +static int +bt_sprev(t, h, key, exactp) + BTREE *t; + PAGE *h; + const DBT *key; + int *exactp; +{ + EPG e; + PAGE *tp; + pgno_t pg; + + /* Skip until reach the beginning of the tree or a key. */ + for (pg = h->prevpg; pg != P_INVALID;) { + if ((tp = mpool_get(t->bt_mp, pg, 0)) == NULL) { + mpool_put(t->bt_mp, h, 0); + return (NULL); + } + if (NEXTINDEX(tp) != 0) + break; + pg = tp->prevpg; + mpool_put(t->bt_mp, tp, 0); + } + /* + * The key is either an exact match, or not as good as + * the one we already have. + */ + if (pg != P_INVALID) { + e.page = tp; + e.index = NEXTINDEX(tp) - 1; + if (__bt_cmp(t, key, &e) == 0) { + mpool_put(t->bt_mp, h, 0); + t->bt_cur = e; + *exactp = 1; + return (1); + } + } + return (0); +} diff --git a/lib/libc/db/btree/bt_seq.c b/lib/libc/db/btree/bt_seq.c new file mode 100644 index 00000000000..f029cc3e174 --- /dev/null +++ b/lib/libc/db/btree/bt_seq.c @@ -0,0 +1,386 @@ +/* $NetBSD: bt_seq.c,v 1.6 1995/02/27 13:20:51 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_seq.c 8.2 (Berkeley) 9/7/93"; +#else +static char rcsid[] = "$NetBSD: bt_seq.c,v 1.6 1995/02/27 13:20:51 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stddef.h> +#include <stdio.h> +#include <stdlib.h> + +#include <db.h> +#include "btree.h" + +static int bt_seqadv __P((BTREE *, EPG *, int)); +static int bt_seqset __P((BTREE *, EPG *, DBT *, int)); + +/* + * Sequential scan support. + * + * The tree can be scanned sequentially, starting from either end of the tree + * or from any specific key. A scan request before any scanning is done is + * initialized as starting from the least node. + * + * Each tree has an EPGNO which has the current position of the cursor. The + * cursor has to survive deletions/insertions in the tree without losing its + * position. This is done by noting deletions without doing them, and then + * doing them when the cursor moves (or the tree is closed). + */ + +/* + * __BT_SEQ -- Btree sequential scan interface. + * + * Parameters: + * dbp: pointer to access method + * key: key for positioning and return value + * data: data return value + * flags: R_CURSOR, R_FIRST, R_LAST, R_NEXT, R_PREV. + * + * Returns: + * RET_ERROR, RET_SUCCESS or RET_SPECIAL if there's no next key. + */ +int +__bt_seq(dbp, key, data, flags) + const DB *dbp; + DBT *key, *data; + u_int flags; +{ + BTREE *t; + EPG e; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* + * If scan unitialized as yet, or starting at a specific record, set + * the scan to a specific key. Both bt_seqset and bt_seqadv pin the + * page the cursor references if they're successful. + */ + switch(flags) { + case R_NEXT: + case R_PREV: + if (ISSET(t, B_SEQINIT)) { + status = bt_seqadv(t, &e, flags); + break; + } + /* FALLTHROUGH */ + case R_CURSOR: + case R_FIRST: + case R_LAST: + status = bt_seqset(t, &e, key, flags); + break; + default: + errno = EINVAL; + return (RET_ERROR); + } + + if (status == RET_SUCCESS) { + status = __bt_ret(t, &e, key, data); + + /* Update the actual cursor. */ + t->bt_bcursor.pgno = e.page->pgno; + t->bt_bcursor.index = e.index; + + /* + * If the user is doing concurrent access, we copied the + * key/data, toss the page. + */ + if (ISSET(t, B_DB_LOCK)) + mpool_put(t->bt_mp, e.page, 0); + else + t->bt_pinned = e.page; + SET(t, B_SEQINIT); + } + return (status); +} + +/* + * BT_SEQSET -- Set the sequential scan to a specific key. + * + * Parameters: + * t: tree + * ep: storage for returned key + * key: key for initial scan position + * flags: R_CURSOR, R_FIRST, R_LAST, R_NEXT, R_PREV + * + * Side effects: + * Pins the page the cursor references. + * + * Returns: + * RET_ERROR, RET_SUCCESS or RET_SPECIAL if there's no next key. + */ +static int +bt_seqset(t, ep, key, flags) + BTREE *t; + EPG *ep; + DBT *key; + int flags; +{ + EPG *e; + PAGE *h; + pgno_t pg; + int exact; + + /* + * Delete any already deleted record that we've been saving because + * the cursor pointed to it. Since going to a specific key, should + * delete any logically deleted records so they aren't found. + */ + if (ISSET(t, B_DELCRSR) && __bt_crsrdel(t, &t->bt_bcursor)) + return (RET_ERROR); + + /* + * Find the first, last or specific key in the tree and point the cursor + * at it. The cursor may not be moved until a new key has been found. + */ + switch(flags) { + case R_CURSOR: /* Keyed scan. */ + /* + * Find the first instance of the key or the smallest key which + * is greater than or equal to the specified key. If run out + * of keys, return RET_SPECIAL. + */ + if (key->data == NULL || key->size == 0) { + errno = EINVAL; + return (RET_ERROR); + } + e = __bt_first(t, key, &exact); /* Returns pinned page. */ + if (e == NULL) + return (RET_ERROR); + /* + * If at the end of a page, skip any empty pages and find the + * next entry. + */ + if (e->index == NEXTINDEX(e->page)) { + h = e->page; + do { + pg = h->nextpg; + mpool_put(t->bt_mp, h, 0); + if (pg == P_INVALID) + return (RET_SPECIAL); + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + } while (NEXTINDEX(h) == 0); + e->index = 0; + e->page = h; + } + *ep = *e; + break; + case R_FIRST: /* First record. */ + case R_NEXT: + /* Walk down the left-hand side of the tree. */ + for (pg = P_ROOT;;) { + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + if (h->flags & (P_BLEAF | P_RLEAF)) + break; + pg = GETBINTERNAL(h, 0)->pgno; + mpool_put(t->bt_mp, h, 0); + } + + /* Skip any empty pages. */ + while (NEXTINDEX(h) == 0 && h->nextpg != P_INVALID) { + pg = h->nextpg; + mpool_put(t->bt_mp, h, 0); + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + } + + if (NEXTINDEX(h) == 0) { + mpool_put(t->bt_mp, h, 0); + return (RET_SPECIAL); + } + + ep->page = h; + ep->index = 0; + break; + case R_LAST: /* Last record. */ + case R_PREV: + /* Walk down the right-hand side of the tree. */ + for (pg = P_ROOT;;) { + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + if (h->flags & (P_BLEAF | P_RLEAF)) + break; + pg = GETBINTERNAL(h, NEXTINDEX(h) - 1)->pgno; + mpool_put(t->bt_mp, h, 0); + } + + /* Skip any empty pages. */ + while (NEXTINDEX(h) == 0 && h->prevpg != P_INVALID) { + pg = h->prevpg; + mpool_put(t->bt_mp, h, 0); + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + } + + if (NEXTINDEX(h) == 0) { + mpool_put(t->bt_mp, h, 0); + return (RET_SPECIAL); + } + + ep->page = h; + ep->index = NEXTINDEX(h) - 1; + break; + } + return (RET_SUCCESS); +} + +/* + * BT_SEQADVANCE -- Advance the sequential scan. + * + * Parameters: + * t: tree + * flags: R_NEXT, R_PREV + * + * Side effects: + * Pins the page the new key/data record is on. + * + * Returns: + * RET_ERROR, RET_SUCCESS or RET_SPECIAL if there's no next key. + */ +static int +bt_seqadv(t, e, flags) + BTREE *t; + EPG *e; + int flags; +{ + EPGNO *c, delc; + PAGE *h; + indx_t index; + pgno_t pg; + + /* Save the current cursor if going to delete it. */ + c = &t->bt_bcursor; + if (ISSET(t, B_DELCRSR)) + delc = *c; + + if ((h = mpool_get(t->bt_mp, c->pgno, 0)) == NULL) + return (RET_ERROR); + + /* + * Find the next/previous record in the tree and point the cursor at it. + * The cursor may not be moved until a new key has been found. + */ + index = c->index; + switch(flags) { + case R_NEXT: /* Next record. */ + if (++index == NEXTINDEX(h)) { + do { + pg = h->nextpg; + mpool_put(t->bt_mp, h, 0); + if (pg == P_INVALID) + return (RET_SPECIAL); + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + } while (NEXTINDEX(h) == 0); + index = 0; + } + break; + case R_PREV: /* Previous record. */ + if (index-- == 0) { + do { + pg = h->prevpg; + mpool_put(t->bt_mp, h, 0); + if (pg == P_INVALID) + return (RET_SPECIAL); + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + } while (NEXTINDEX(h) == 0); + index = NEXTINDEX(h) - 1; + } + break; + } + + e->page = h; + e->index = index; + + /* + * Delete any already deleted record that we've been saving because the + * cursor pointed to it. This could cause the new index to be shifted + * down by one if the record we're deleting is on the same page and has + * a larger index. + */ + if (ISSET(t, B_DELCRSR)) { + CLR(t, B_DELCRSR); /* Don't try twice. */ + if (c->pgno == delc.pgno && c->index > delc.index) + --c->index; + if (__bt_crsrdel(t, &delc)) + return (RET_ERROR); + } + return (RET_SUCCESS); +} + +/* + * __BT_CRSRDEL -- Delete the record referenced by the cursor. + * + * Parameters: + * t: tree + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__bt_crsrdel(t, c) + BTREE *t; + EPGNO *c; +{ + PAGE *h; + int status; + + CLR(t, B_DELCRSR); /* Don't try twice. */ + if ((h = mpool_get(t->bt_mp, c->pgno, 0)) == NULL) + return (RET_ERROR); + status = __bt_dleaf(t, h, c->index); + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + return (status); +} diff --git a/lib/libc/db/btree/bt_split.c b/lib/libc/db/btree/bt_split.c new file mode 100644 index 00000000000..22334c1e430 --- /dev/null +++ b/lib/libc/db/btree/bt_split.c @@ -0,0 +1,833 @@ +/* $NetBSD: bt_split.c,v 1.5 1995/02/27 13:20:55 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_split.c 8.6 (Berkeley) 6/16/94"; +#else +static char rcsid[] = "$NetBSD: bt_split.c,v 1.5 1995/02/27 13:20:55 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <limits.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "btree.h" + +static int bt_broot __P((BTREE *, PAGE *, PAGE *, PAGE *)); +static PAGE *bt_page + __P((BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t)); +static int bt_preserve __P((BTREE *, pgno_t)); +static PAGE *bt_psplit + __P((BTREE *, PAGE *, PAGE *, PAGE *, indx_t *, size_t)); +static PAGE *bt_root + __P((BTREE *, PAGE *, PAGE **, PAGE **, indx_t *, size_t)); +static int bt_rroot __P((BTREE *, PAGE *, PAGE *, PAGE *)); +static recno_t rec_total __P((PAGE *)); + +#ifdef STATISTICS +u_long bt_rootsplit, bt_split, bt_sortsplit, bt_pfxsaved; +#endif + +/* + * __BT_SPLIT -- Split the tree. + * + * Parameters: + * t: tree + * sp: page to split + * key: key to insert + * data: data to insert + * flags: BIGKEY/BIGDATA flags + * ilen: insert length + * skip: index to leave open + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__bt_split(t, sp, key, data, flags, ilen, argskip) + BTREE *t; + PAGE *sp; + const DBT *key, *data; + int flags; + size_t ilen; + u_int32_t argskip; +{ + BINTERNAL *bi; + BLEAF *bl, *tbl; + DBT a, b; + EPGNO *parent; + PAGE *h, *l, *r, *lchild, *rchild; + indx_t nxtindex; + u_int16_t skip; + u_int32_t n, nbytes, nksize; + int parentsplit; + char *dest; + + /* + * Split the page into two pages, l and r. The split routines return + * a pointer to the page into which the key should be inserted and with + * skip set to the offset which should be used. Additionally, l and r + * are pinned. + */ + skip = argskip; + h = sp->pgno == P_ROOT ? + bt_root(t, sp, &l, &r, &skip, ilen) : + bt_page(t, sp, &l, &r, &skip, ilen); + if (h == NULL) + return (RET_ERROR); + + /* + * Insert the new key/data pair into the leaf page. (Key inserts + * always cause a leaf page to split first.) + */ + h->linp[skip] = h->upper -= ilen; + dest = (char *)h + h->upper; + if (ISSET(t, R_RECNO)) + WR_RLEAF(dest, data, flags) + else + WR_BLEAF(dest, key, data, flags) + + /* If the root page was split, make it look right. */ + if (sp->pgno == P_ROOT && + (ISSET(t, R_RECNO) ? + bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR) + goto err2; + + /* + * Now we walk the parent page stack -- a LIFO stack of the pages that + * were traversed when we searched for the page that split. Each stack + * entry is a page number and a page index offset. The offset is for + * the page traversed on the search. We've just split a page, so we + * have to insert a new key into the parent page. + * + * If the insert into the parent page causes it to split, may have to + * continue splitting all the way up the tree. We stop if the root + * splits or the page inserted into didn't have to split to hold the + * new key. Some algorithms replace the key for the old page as well + * as the new page. We don't, as there's no reason to believe that the + * first key on the old page is any better than the key we have, and, + * in the case of a key being placed at index 0 causing the split, the + * key is unavailable. + * + * There are a maximum of 5 pages pinned at any time. We keep the left + * and right pages pinned while working on the parent. The 5 are the + * two children, left parent and right parent (when the parent splits) + * and the root page or the overflow key page when calling bt_preserve. + * This code must make sure that all pins are released other than the + * root page or overflow page which is unlocked elsewhere. + */ + while ((parent = BT_POP(t)) != NULL) { + lchild = l; + rchild = r; + + /* Get the parent page. */ + if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL) + goto err2; + + /* + * The new key goes ONE AFTER the index, because the split + * was to the right. + */ + skip = parent->index + 1; + + /* + * Calculate the space needed on the parent page. + * + * Prefix trees: space hack when inserting into BINTERNAL + * pages. Retain only what's needed to distinguish between + * the new entry and the LAST entry on the page to its left. + * If the keys compare equal, retain the entire key. Note, + * we don't touch overflow keys, and the entire key must be + * retained for the next-to-left most key on the leftmost + * page of each level, or the search will fail. Applicable + * ONLY to internal pages that have leaf pages as children. + * Further reduction of the key between pairs of internal + * pages loses too much information. + */ + switch (rchild->flags & P_TYPE) { + case P_BINTERNAL: + bi = GETBINTERNAL(rchild, 0); + nbytes = NBINTERNAL(bi->ksize); + break; + case P_BLEAF: + bl = GETBLEAF(rchild, 0); + nbytes = NBINTERNAL(bl->ksize); + if (t->bt_pfx && !(bl->flags & P_BIGKEY) && + (h->prevpg != P_INVALID || skip > 1)) { + tbl = GETBLEAF(lchild, NEXTINDEX(lchild) - 1); + a.size = tbl->ksize; + a.data = tbl->bytes; + b.size = bl->ksize; + b.data = bl->bytes; + nksize = t->bt_pfx(&a, &b); + n = NBINTERNAL(nksize); + if (n < nbytes) { +#ifdef STATISTICS + bt_pfxsaved += nbytes - n; +#endif + nbytes = n; + } else + nksize = 0; + } else + nksize = 0; + break; + case P_RINTERNAL: + case P_RLEAF: + nbytes = NRINTERNAL; + break; + default: + abort(); + } + + /* Split the parent page if necessary or shift the indices. */ + if (h->upper - h->lower < nbytes + sizeof(indx_t)) { + sp = h; + h = h->pgno == P_ROOT ? + bt_root(t, h, &l, &r, &skip, nbytes) : + bt_page(t, h, &l, &r, &skip, nbytes); + if (h == NULL) + goto err1; + parentsplit = 1; + } else { + if (skip < (nxtindex = NEXTINDEX(h))) + memmove(h->linp + skip + 1, h->linp + skip, + (nxtindex - skip) * sizeof(indx_t)); + h->lower += sizeof(indx_t); + parentsplit = 0; + } + + /* Insert the key into the parent page. */ + switch(rchild->flags & P_TYPE) { + case P_BINTERNAL: + h->linp[skip] = h->upper -= nbytes; + dest = (char *)h + h->linp[skip]; + memmove(dest, bi, nbytes); + ((BINTERNAL *)dest)->pgno = rchild->pgno; + break; + case P_BLEAF: + h->linp[skip] = h->upper -= nbytes; + dest = (char *)h + h->linp[skip]; + WR_BINTERNAL(dest, nksize ? nksize : bl->ksize, + rchild->pgno, bl->flags & P_BIGKEY); + memmove(dest, bl->bytes, nksize ? nksize : bl->ksize); + if (bl->flags & P_BIGKEY && + bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR) + goto err1; + break; + case P_RINTERNAL: + /* + * Update the left page count. If split + * added at index 0, fix the correct page. + */ + if (skip > 0) + dest = (char *)h + h->linp[skip - 1]; + else + dest = (char *)l + l->linp[NEXTINDEX(l) - 1]; + ((RINTERNAL *)dest)->nrecs = rec_total(lchild); + ((RINTERNAL *)dest)->pgno = lchild->pgno; + + /* Update the right page count. */ + h->linp[skip] = h->upper -= nbytes; + dest = (char *)h + h->linp[skip]; + ((RINTERNAL *)dest)->nrecs = rec_total(rchild); + ((RINTERNAL *)dest)->pgno = rchild->pgno; + break; + case P_RLEAF: + /* + * Update the left page count. If split + * added at index 0, fix the correct page. + */ + if (skip > 0) + dest = (char *)h + h->linp[skip - 1]; + else + dest = (char *)l + l->linp[NEXTINDEX(l) - 1]; + ((RINTERNAL *)dest)->nrecs = NEXTINDEX(lchild); + ((RINTERNAL *)dest)->pgno = lchild->pgno; + + /* Update the right page count. */ + h->linp[skip] = h->upper -= nbytes; + dest = (char *)h + h->linp[skip]; + ((RINTERNAL *)dest)->nrecs = NEXTINDEX(rchild); + ((RINTERNAL *)dest)->pgno = rchild->pgno; + break; + default: + abort(); + } + + /* Unpin the held pages. */ + if (!parentsplit) { + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + break; + } + + /* If the root page was split, make it look right. */ + if (sp->pgno == P_ROOT && + (ISSET(t, R_RECNO) ? + bt_rroot(t, sp, l, r) : bt_broot(t, sp, l, r)) == RET_ERROR) + goto err1; + + mpool_put(t->bt_mp, lchild, MPOOL_DIRTY); + mpool_put(t->bt_mp, rchild, MPOOL_DIRTY); + } + + /* Unpin the held pages. */ + mpool_put(t->bt_mp, l, MPOOL_DIRTY); + mpool_put(t->bt_mp, r, MPOOL_DIRTY); + + /* Clear any pages left on the stack. */ + return (RET_SUCCESS); + + /* + * If something fails in the above loop we were already walking back + * up the tree and the tree is now inconsistent. Nothing much we can + * do about it but release any memory we're holding. + */ +err1: mpool_put(t->bt_mp, lchild, MPOOL_DIRTY); + mpool_put(t->bt_mp, rchild, MPOOL_DIRTY); + +err2: mpool_put(t->bt_mp, l, 0); + mpool_put(t->bt_mp, r, 0); + __dbpanic(t->bt_dbp); + return (RET_ERROR); +} + +/* + * BT_PAGE -- Split a non-root page of a btree. + * + * Parameters: + * t: tree + * h: root page + * lp: pointer to left page pointer + * rp: pointer to right page pointer + * skip: pointer to index to leave open + * ilen: insert length + * + * Returns: + * Pointer to page in which to insert or NULL on error. + */ +static PAGE * +bt_page(t, h, lp, rp, skip, ilen) + BTREE *t; + PAGE *h, **lp, **rp; + indx_t *skip; + size_t ilen; +{ + PAGE *l, *r, *tp; + pgno_t npg; + +#ifdef STATISTICS + ++bt_split; +#endif + /* Put the new right page for the split into place. */ + if ((r = __bt_new(t, &npg)) == NULL) + return (NULL); + r->pgno = npg; + r->lower = BTDATAOFF; + r->upper = t->bt_psize; + r->nextpg = h->nextpg; + r->prevpg = h->pgno; + r->flags = h->flags & P_TYPE; + + /* + * If we're splitting the last page on a level because we're appending + * a key to it (skip is NEXTINDEX()), it's likely that the data is + * sorted. Adding an empty page on the side of the level is less work + * and can push the fill factor much higher than normal. If we're + * wrong it's no big deal, we'll just do the split the right way next + * time. It may look like it's equally easy to do a similar hack for + * reverse sorted data, that is, split the tree left, but it's not. + * Don't even try. + */ + if (h->nextpg == P_INVALID && *skip == NEXTINDEX(h)) { +#ifdef STATISTICS + ++bt_sortsplit; +#endif + h->nextpg = r->pgno; + r->lower = BTDATAOFF + sizeof(indx_t); + *skip = 0; + *lp = h; + *rp = r; + return (r); + } + + /* Put the new left page for the split into place. */ + if ((l = (PAGE *)malloc(t->bt_psize)) == NULL) { + mpool_put(t->bt_mp, r, 0); + return (NULL); + } + l->pgno = h->pgno; + l->nextpg = r->pgno; + l->prevpg = h->prevpg; + l->lower = BTDATAOFF; + l->upper = t->bt_psize; + l->flags = h->flags & P_TYPE; + + /* Fix up the previous pointer of the page after the split page. */ + if (h->nextpg != P_INVALID) { + if ((tp = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL) { + free(l); + /* XXX mpool_free(t->bt_mp, r->pgno); */ + return (NULL); + } + tp->prevpg = r->pgno; + mpool_put(t->bt_mp, tp, 0); + } + + /* + * Split right. The key/data pairs aren't sorted in the btree page so + * it's simpler to copy the data from the split page onto two new pages + * instead of copying half the data to the right page and compacting + * the left page in place. Since the left page can't change, we have + * to swap the original and the allocated left page after the split. + */ + tp = bt_psplit(t, h, l, r, skip, ilen); + + /* Move the new left page onto the old left page. */ + memmove(h, l, t->bt_psize); + if (tp == l) + tp = h; + free(l); + + *lp = h; + *rp = r; + return (tp); +} + +/* + * BT_ROOT -- Split the root page of a btree. + * + * Parameters: + * t: tree + * h: root page + * lp: pointer to left page pointer + * rp: pointer to right page pointer + * skip: pointer to index to leave open + * ilen: insert length + * + * Returns: + * Pointer to page in which to insert or NULL on error. + */ +static PAGE * +bt_root(t, h, lp, rp, skip, ilen) + BTREE *t; + PAGE *h, **lp, **rp; + indx_t *skip; + size_t ilen; +{ + PAGE *l, *r, *tp; + pgno_t lnpg, rnpg; + +#ifdef STATISTICS + ++bt_split; + ++bt_rootsplit; +#endif + /* Put the new left and right pages for the split into place. */ + if ((l = __bt_new(t, &lnpg)) == NULL || + (r = __bt_new(t, &rnpg)) == NULL) + return (NULL); + l->pgno = lnpg; + r->pgno = rnpg; + l->nextpg = r->pgno; + r->prevpg = l->pgno; + l->prevpg = r->nextpg = P_INVALID; + l->lower = r->lower = BTDATAOFF; + l->upper = r->upper = t->bt_psize; + l->flags = r->flags = h->flags & P_TYPE; + + /* Split the root page. */ + tp = bt_psplit(t, h, l, r, skip, ilen); + + *lp = l; + *rp = r; + return (tp); +} + +/* + * BT_RROOT -- Fix up the recno root page after it has been split. + * + * Parameters: + * t: tree + * h: root page + * l: left page + * r: right page + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +static int +bt_rroot(t, h, l, r) + BTREE *t; + PAGE *h, *l, *r; +{ + char *dest; + + /* Insert the left and right keys, set the header information. */ + h->linp[0] = h->upper = t->bt_psize - NRINTERNAL; + dest = (char *)h + h->upper; + WR_RINTERNAL(dest, + l->flags & P_RLEAF ? NEXTINDEX(l) : rec_total(l), l->pgno); + + h->linp[1] = h->upper -= NRINTERNAL; + dest = (char *)h + h->upper; + WR_RINTERNAL(dest, + r->flags & P_RLEAF ? NEXTINDEX(r) : rec_total(r), r->pgno); + + h->lower = BTDATAOFF + 2 * sizeof(indx_t); + + /* Unpin the root page, set to recno internal page. */ + h->flags &= ~P_TYPE; + h->flags |= P_RINTERNAL; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + + return (RET_SUCCESS); +} + +/* + * BT_BROOT -- Fix up the btree root page after it has been split. + * + * Parameters: + * t: tree + * h: root page + * l: left page + * r: right page + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +static int +bt_broot(t, h, l, r) + BTREE *t; + PAGE *h, *l, *r; +{ + BINTERNAL *bi; + BLEAF *bl; + u_int32_t nbytes; + char *dest; + + /* + * If the root page was a leaf page, change it into an internal page. + * We copy the key we split on (but not the key's data, in the case of + * a leaf page) to the new root page. + * + * The btree comparison code guarantees that the left-most key on any + * level of the tree is never used, so it doesn't need to be filled in. + */ + nbytes = NBINTERNAL(0); + h->linp[0] = h->upper = t->bt_psize - nbytes; + dest = (char *)h + h->upper; + WR_BINTERNAL(dest, 0, l->pgno, 0); + + switch(h->flags & P_TYPE) { + case P_BLEAF: + bl = GETBLEAF(r, 0); + nbytes = NBINTERNAL(bl->ksize); + h->linp[1] = h->upper -= nbytes; + dest = (char *)h + h->upper; + WR_BINTERNAL(dest, bl->ksize, r->pgno, 0); + memmove(dest, bl->bytes, bl->ksize); + + /* + * If the key is on an overflow page, mark the overflow chain + * so it isn't deleted when the leaf copy of the key is deleted. + */ + if (bl->flags & P_BIGKEY && + bt_preserve(t, *(pgno_t *)bl->bytes) == RET_ERROR) + return (RET_ERROR); + break; + case P_BINTERNAL: + bi = GETBINTERNAL(r, 0); + nbytes = NBINTERNAL(bi->ksize); + h->linp[1] = h->upper -= nbytes; + dest = (char *)h + h->upper; + memmove(dest, bi, nbytes); + ((BINTERNAL *)dest)->pgno = r->pgno; + break; + default: + abort(); + } + + /* There are two keys on the page. */ + h->lower = BTDATAOFF + 2 * sizeof(indx_t); + + /* Unpin the root page, set to btree internal page. */ + h->flags &= ~P_TYPE; + h->flags |= P_BINTERNAL; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + + return (RET_SUCCESS); +} + +/* + * BT_PSPLIT -- Do the real work of splitting the page. + * + * Parameters: + * t: tree + * h: page to be split + * l: page to put lower half of data + * r: page to put upper half of data + * pskip: pointer to index to leave open + * ilen: insert length + * + * Returns: + * Pointer to page in which to insert. + */ +static PAGE * +bt_psplit(t, h, l, r, pskip, ilen) + BTREE *t; + PAGE *h, *l, *r; + indx_t *pskip; + size_t ilen; +{ + BINTERNAL *bi; + BLEAF *bl; + RLEAF *rl; + EPGNO *c; + PAGE *rval; + void *src; + indx_t full, half, nxt, off, skip, top, used; + u_int32_t nbytes; + int bigkeycnt, isbigkey; + + /* + * Split the data to the left and right pages. Leave the skip index + * open. Additionally, make some effort not to split on an overflow + * key. This makes internal page processing faster and can save + * space as overflow keys used by internal pages are never deleted. + */ + bigkeycnt = 0; + skip = *pskip; + full = t->bt_psize - BTDATAOFF; + half = full / 2; + used = 0; + for (nxt = off = 0, top = NEXTINDEX(h); nxt < top; ++off) { + if (skip == off) { + nbytes = ilen; + isbigkey = 0; /* XXX: not really known. */ + } else + switch (h->flags & P_TYPE) { + case P_BINTERNAL: + src = bi = GETBINTERNAL(h, nxt); + nbytes = NBINTERNAL(bi->ksize); + isbigkey = bi->flags & P_BIGKEY; + break; + case P_BLEAF: + src = bl = GETBLEAF(h, nxt); + nbytes = NBLEAF(bl); + isbigkey = bl->flags & P_BIGKEY; + break; + case P_RINTERNAL: + src = GETRINTERNAL(h, nxt); + nbytes = NRINTERNAL; + isbigkey = 0; + break; + case P_RLEAF: + src = rl = GETRLEAF(h, nxt); + nbytes = NRLEAF(rl); + isbigkey = 0; + break; + default: + abort(); + } + + /* + * If the key/data pairs are substantial fractions of the max + * possible size for the page, it's possible to get situations + * where we decide to try and copy too much onto the left page. + * Make sure that doesn't happen. + */ + if (skip <= off && used + nbytes >= full) { + --off; + break; + } + + /* Copy the key/data pair, if not the skipped index. */ + if (skip != off) { + ++nxt; + + l->linp[off] = l->upper -= nbytes; + memmove((char *)l + l->upper, src, nbytes); + } + + used += nbytes; + if (used >= half) { + if (!isbigkey || bigkeycnt == 3) + break; + else + ++bigkeycnt; + } + } + + /* + * Off is the last offset that's valid for the left page. + * Nxt is the first offset to be placed on the right page. + */ + l->lower += (off + 1) * sizeof(indx_t); + + /* + * If splitting the page that the cursor was on, the cursor has to be + * adjusted to point to the same record as before the split. If the + * cursor is at or past the skipped slot, the cursor is incremented by + * one. If the cursor is on the right page, it is decremented by the + * number of records split to the left page. + * + * Don't bother checking for the B_SEQINIT flag, the page number will + * be P_INVALID. + */ + c = &t->bt_bcursor; + if (c->pgno == h->pgno) { + if (c->index >= skip) + ++c->index; + if (c->index < nxt) /* Left page. */ + c->pgno = l->pgno; + else { /* Right page. */ + c->pgno = r->pgno; + c->index -= nxt; + } + } + + /* + * If the skipped index was on the left page, just return that page. + * Otherwise, adjust the skip index to reflect the new position on + * the right page. + */ + if (skip <= off) { + skip = 0; + rval = l; + } else { + rval = r; + *pskip -= nxt; + } + + for (off = 0; nxt < top; ++off) { + if (skip == nxt) { + ++off; + skip = 0; + } + switch (h->flags & P_TYPE) { + case P_BINTERNAL: + src = bi = GETBINTERNAL(h, nxt); + nbytes = NBINTERNAL(bi->ksize); + break; + case P_BLEAF: + src = bl = GETBLEAF(h, nxt); + nbytes = NBLEAF(bl); + break; + case P_RINTERNAL: + src = GETRINTERNAL(h, nxt); + nbytes = NRINTERNAL; + break; + case P_RLEAF: + src = rl = GETRLEAF(h, nxt); + nbytes = NRLEAF(rl); + break; + default: + abort(); + } + ++nxt; + r->linp[off] = r->upper -= nbytes; + memmove((char *)r + r->upper, src, nbytes); + } + r->lower += off * sizeof(indx_t); + + /* If the key is being appended to the page, adjust the index. */ + if (skip == top) + r->lower += sizeof(indx_t); + + return (rval); +} + +/* + * BT_PRESERVE -- Mark a chain of pages as used by an internal node. + * + * Chains of indirect blocks pointed to by leaf nodes get reclaimed when the + * record that references them gets deleted. Chains pointed to by internal + * pages never get deleted. This routine marks a chain as pointed to by an + * internal page. + * + * Parameters: + * t: tree + * pg: page number of first page in the chain. + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +static int +bt_preserve(t, pg) + BTREE *t; + pgno_t pg; +{ + PAGE *h; + + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + return (RET_ERROR); + h->flags |= P_PRESERVE; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + return (RET_SUCCESS); +} + +/* + * REC_TOTAL -- Return the number of recno entries below a page. + * + * Parameters: + * h: page + * + * Returns: + * The number of recno entries below a page. + * + * XXX + * These values could be set by the bt_psplit routine. The problem is that the + * entry has to be popped off of the stack etc. or the values have to be passed + * all the way back to bt_split/bt_rroot and it's not very clean. + */ +static recno_t +rec_total(h) + PAGE *h; +{ + recno_t recs; + indx_t nxt, top; + + for (recs = 0, nxt = 0, top = NEXTINDEX(h); nxt < top; ++nxt) + recs += GETRINTERNAL(h, nxt)->nrecs; + return (recs); +} diff --git a/lib/libc/db/btree/bt_stack.c b/lib/libc/db/btree/bt_stack.c new file mode 100644 index 00000000000..7199ae9b090 --- /dev/null +++ b/lib/libc/db/btree/bt_stack.c @@ -0,0 +1,102 @@ +/* $NetBSD: bt_stack.c,v 1.5 1995/02/27 13:21:01 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_stack.c 8.4 (Berkeley) 6/20/94"; +#else +static char rcsid[] = "$NetBSD: bt_stack.c,v 1.5 1995/02/27 13:21:01 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> + +#include <db.h> +#include "btree.h" + +/* + * When a page splits, a new record has to be inserted into its parent page. + * This page may have to split as well, all the way up to the root. Since + * parent pointers in each page would be expensive, we maintain a stack of + * parent pages as we descend the tree. + * + * XXX + * This is a concurrency problem -- if user a builds a stack, then user b + * splits the tree, then user a tries to split the tree, there's a new level + * in the tree that user a doesn't know about. + */ + +/* + * __BT_PUSH -- Push parent page info onto the stack (LIFO). + * + * Parameters: + * t: tree + * pgno: page + * index: page index + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__bt_push(t, pgno, index) + BTREE *t; + pgno_t pgno; + indx_t index; +{ + size_t sz; + + if (t->bt_sp == t->bt_maxstack) { + t->bt_maxstack += 50; + sz = t->bt_maxstack * sizeof(EPGNO); + t->bt_stack = (EPGNO *)(t->bt_stack == NULL ? + malloc(sz) : realloc(t->bt_stack, sz)); + if (t->bt_stack == NULL) { + t->bt_maxstack -= 50; + return (RET_ERROR); + } + } + + t->bt_stack[t->bt_sp].pgno = pgno; + t->bt_stack[t->bt_sp].index = index; + ++t->bt_sp; + return (RET_SUCCESS); +} diff --git a/lib/libc/db/btree/bt_utils.c b/lib/libc/db/btree/bt_utils.c new file mode 100644 index 00000000000..f9acf0ab8a1 --- /dev/null +++ b/lib/libc/db/btree/bt_utils.c @@ -0,0 +1,256 @@ +/* $NetBSD: bt_utils.c,v 1.6 1995/02/27 13:21:04 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)bt_utils.c 8.5 (Berkeley) 6/20/94"; +#else +static char rcsid[] = "$NetBSD: bt_utils.c,v 1.6 1995/02/27 13:21:04 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "btree.h" + +/* + * __BT_RET -- Build return key/data pair as a result of search or scan. + * + * Parameters: + * t: tree + * d: LEAF to be returned to the user. + * key: user's key structure (NULL if not to be filled in) + * data: user's data structure + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +int +__bt_ret(t, e, key, data) + BTREE *t; + EPG *e; + DBT *key, *data; +{ + register BLEAF *bl; + register void *p; + + bl = GETBLEAF(e->page, e->index); + + /* + * We always copy big keys/data to make them contigous. Otherwise, + * we leave the page pinned and don't copy unless the user specified + * concurrent access. + */ + if (bl->flags & P_BIGDATA) { + if (__ovfl_get(t, bl->bytes + bl->ksize, + &data->size, &t->bt_dbuf, &t->bt_dbufsz)) + return (RET_ERROR); + data->data = t->bt_dbuf; + } else if (ISSET(t, B_DB_LOCK)) { + /* Use +1 in case the first record retrieved is 0 length. */ + if (bl->dsize + 1 > t->bt_dbufsz) { + p = (void *)(t->bt_dbuf == NULL ? + malloc(bl->dsize + 1) : + realloc(t->bt_dbuf, bl->dsize + 1)); + if (p == NULL) + return (RET_ERROR); + t->bt_dbuf = p; + t->bt_dbufsz = bl->dsize + 1; + } + memmove(t->bt_dbuf, bl->bytes + bl->ksize, bl->dsize); + data->size = bl->dsize; + data->data = t->bt_dbuf; + } else { + data->size = bl->dsize; + data->data = bl->bytes + bl->ksize; + } + + if (key == NULL) + return (RET_SUCCESS); + + if (bl->flags & P_BIGKEY) { + if (__ovfl_get(t, bl->bytes, + &key->size, &t->bt_kbuf, &t->bt_kbufsz)) + return (RET_ERROR); + key->data = t->bt_kbuf; + } else if (ISSET(t, B_DB_LOCK)) { + if (bl->ksize > t->bt_kbufsz) { + p = (void *)(t->bt_kbuf == NULL ? + malloc(bl->ksize) : realloc(t->bt_kbuf, bl->ksize)); + if (p == NULL) + return (RET_ERROR); + t->bt_kbuf = p; + t->bt_kbufsz = bl->ksize; + } + memmove(t->bt_kbuf, bl->bytes, bl->ksize); + key->size = bl->ksize; + key->data = t->bt_kbuf; + } else { + key->size = bl->ksize; + key->data = bl->bytes; + } + return (RET_SUCCESS); +} + +/* + * __BT_CMP -- Compare a key to a given record. + * + * Parameters: + * t: tree + * k1: DBT pointer of first arg to comparison + * e: pointer to EPG for comparison + * + * Returns: + * < 0 if k1 is < record + * = 0 if k1 is = record + * > 0 if k1 is > record + */ +int +__bt_cmp(t, k1, e) + BTREE *t; + const DBT *k1; + EPG *e; +{ + BINTERNAL *bi; + BLEAF *bl; + DBT k2; + PAGE *h; + void *bigkey; + + /* + * The left-most key on internal pages, at any level of the tree, is + * guaranteed by the following code to be less than any user key. + * This saves us from having to update the leftmost key on an internal + * page when the user inserts a new key in the tree smaller than + * anything we've yet seen. + */ + h = e->page; + if (e->index == 0 && h->prevpg == P_INVALID && !(h->flags & P_BLEAF)) + return (1); + + bigkey = NULL; + if (h->flags & P_BLEAF) { + bl = GETBLEAF(h, e->index); + if (bl->flags & P_BIGKEY) + bigkey = bl->bytes; + else { + k2.data = bl->bytes; + k2.size = bl->ksize; + } + } else { + bi = GETBINTERNAL(h, e->index); + if (bi->flags & P_BIGKEY) + bigkey = bi->bytes; + else { + k2.data = bi->bytes; + k2.size = bi->ksize; + } + } + + if (bigkey) { + if (__ovfl_get(t, bigkey, + &k2.size, &t->bt_dbuf, &t->bt_dbufsz)) + return (RET_ERROR); + k2.data = t->bt_dbuf; + } + return ((*t->bt_cmp)(k1, &k2)); +} + +/* + * __BT_DEFCMP -- Default comparison routine. + * + * Parameters: + * a: DBT #1 + * b: DBT #2 + * + * Returns: + * < 0 if a is < b + * = 0 if a is = b + * > 0 if a is > b + */ +int +__bt_defcmp(a, b) + const DBT *a, *b; +{ + register size_t len; + register u_char *p1, *p2; + + /* + * XXX + * If a size_t doesn't fit in an int, this routine can lose. + * What we need is a integral type which is guaranteed to be + * larger than a size_t, and there is no such thing. + */ + len = MIN(a->size, b->size); + for (p1 = a->data, p2 = b->data; len--; ++p1, ++p2) + if (*p1 != *p2) + return ((int)*p1 - (int)*p2); + return ((int)a->size - (int)b->size); +} + +/* + * __BT_DEFPFX -- Default prefix routine. + * + * Parameters: + * a: DBT #1 + * b: DBT #2 + * + * Returns: + * Number of bytes needed to distinguish b from a. + */ +size_t +__bt_defpfx(a, b) + const DBT *a, *b; +{ + register u_char *p1, *p2; + register size_t cnt, len; + + cnt = 1; + len = MIN(a->size, b->size); + for (p1 = a->data, p2 = b->data; len--; ++p1, ++p2, ++cnt) + if (*p1 != *p2) + return (cnt); + + /* a->size must be <= b->size, or they wouldn't be in this order. */ + return (a->size < b->size ? a->size + 1 : a->size); +} diff --git a/lib/libc/db/btree/btree.h b/lib/libc/db/btree/btree.h new file mode 100644 index 00000000000..747ed93638e --- /dev/null +++ b/lib/libc/db/btree/btree.h @@ -0,0 +1,355 @@ +/* $NetBSD: btree.h,v 1.8 1995/02/27 13:21:08 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)btree.h 8.6 (Berkeley) 5/31/94 + */ + +#include <mpool.h> + +#define DEFMINKEYPAGE (2) /* Minimum keys per page */ +#define MINCACHE (5) /* Minimum cached pages */ +#define MINPSIZE (512) /* Minimum page size */ + +/* + * Page 0 of a btree file contains a copy of the meta-data. This page is also + * used as an out-of-band page, i.e. page pointers that point to nowhere point + * to page 0. Page 1 is the root of the btree. + */ +#define P_INVALID 0 /* Invalid tree page number. */ +#define P_META 0 /* Tree metadata page number. */ +#define P_ROOT 1 /* Tree root page number. */ + +/* + * There are five page layouts in the btree: btree internal pages (BINTERNAL), + * btree leaf pages (BLEAF), recno internal pages (RINTERNAL), recno leaf pages + * (RLEAF) and overflow pages. All five page types have a page header (PAGE). + * This implementation requires that values within structures NOT be padded. + * (ANSI C permits random padding.) If your compiler pads randomly you'll have + * to do some work to get this package to run. + */ +typedef struct _page { + pgno_t pgno; /* this page's page number */ + pgno_t prevpg; /* left sibling */ + pgno_t nextpg; /* right sibling */ + +#define P_BINTERNAL 0x01 /* btree internal page */ +#define P_BLEAF 0x02 /* leaf page */ +#define P_OVERFLOW 0x04 /* overflow page */ +#define P_RINTERNAL 0x08 /* recno internal page */ +#define P_RLEAF 0x10 /* leaf page */ +#define P_TYPE 0x1f /* type mask */ +#define P_PRESERVE 0x20 /* never delete this chain of pages */ + u_int32_t flags; + + indx_t lower; /* lower bound of free space on page */ + indx_t upper; /* upper bound of free space on page */ + indx_t linp[1]; /* indx_t-aligned VAR. LENGTH DATA */ +} PAGE; + +/* First and next index. */ +#define BTDATAOFF (sizeof(pgno_t) + sizeof(pgno_t) + sizeof(pgno_t) + \ + sizeof(u_int32_t) + sizeof(indx_t) + sizeof(indx_t)) +#define NEXTINDEX(p) (((p)->lower - BTDATAOFF) / sizeof(indx_t)) + +/* + * For pages other than overflow pages, there is an array of offsets into the + * rest of the page immediately following the page header. Each offset is to + * an item which is unique to the type of page. The h_lower offset is just + * past the last filled-in index. The h_upper offset is the first item on the + * page. Offsets are from the beginning of the page. + * + * If an item is too big to store on a single page, a flag is set and the item + * is a { page, size } pair such that the page is the first page of an overflow + * chain with size bytes of item. Overflow pages are simply bytes without any + * external structure. + * + * The page number and size fields in the items are pgno_t-aligned so they can + * be manipulated without copying. (This presumes that 32 bit items can be + * manipulated on this system.) + */ +#define LALIGN(n) \ + (((n) + sizeof(pgno_t) - 1) & ~(sizeof(pgno_t) - 1)) +#define NOVFLSIZE (sizeof(pgno_t) + sizeof(u_int32_t)) + +/* + * For the btree internal pages, the item is a key. BINTERNALs are {key, pgno} + * pairs, such that the key compares less than or equal to all of the records + * on that page. For a tree without duplicate keys, an internal page with two + * consecutive keys, a and b, will have all records greater than or equal to a + * and less than b stored on the page associated with a. Duplicate keys are + * somewhat special and can cause duplicate internal and leaf page records and + * some minor modifications of the above rule. + */ +typedef struct _binternal { + u_int32_t ksize; /* key size */ + pgno_t pgno; /* page number stored on */ +#define P_BIGDATA 0x01 /* overflow data */ +#define P_BIGKEY 0x02 /* overflow key */ + u_char flags; + char bytes[1]; /* data */ +} BINTERNAL; + +/* Get the page's BINTERNAL structure at index indx. */ +#define GETBINTERNAL(pg, indx) \ + ((BINTERNAL *)((char *)(pg) + (pg)->linp[indx])) + +/* Get the number of bytes in the entry. */ +#define NBINTERNAL(len) \ + LALIGN(sizeof(u_int32_t) + sizeof(pgno_t) + sizeof(u_char) + (len)) + +/* Copy a BINTERNAL entry to the page. */ +#define WR_BINTERNAL(p, size, pgno, flags) { \ + *(u_int32_t *)p = size; \ + p += sizeof(u_int32_t); \ + *(pgno_t *)p = pgno; \ + p += sizeof(pgno_t); \ + *(u_char *)p = flags; \ + p += sizeof(u_char); \ +} + +/* + * For the recno internal pages, the item is a page number with the number of + * keys found on that page and below. + */ +typedef struct _rinternal { + recno_t nrecs; /* number of records */ + pgno_t pgno; /* page number stored below */ +} RINTERNAL; + +/* Get the page's RINTERNAL structure at index indx. */ +#define GETRINTERNAL(pg, indx) \ + ((RINTERNAL *)((char *)(pg) + (pg)->linp[indx])) + +/* Get the number of bytes in the entry. */ +#define NRINTERNAL \ + LALIGN(sizeof(recno_t) + sizeof(pgno_t)) + +/* Copy a RINTERAL entry to the page. */ +#define WR_RINTERNAL(p, nrecs, pgno) { \ + *(recno_t *)p = nrecs; \ + p += sizeof(recno_t); \ + *(pgno_t *)p = pgno; \ +} + +/* For the btree leaf pages, the item is a key and data pair. */ +typedef struct _bleaf { + u_int32_t ksize; /* size of key */ + u_int32_t dsize; /* size of data */ + u_char flags; /* P_BIGDATA, P_BIGKEY */ + char bytes[1]; /* data */ +} BLEAF; + +/* Get the page's BLEAF structure at index indx. */ +#define GETBLEAF(pg, indx) \ + ((BLEAF *)((char *)(pg) + (pg)->linp[indx])) + +/* Get the number of bytes in the entry. */ +#define NBLEAF(p) NBLEAFDBT((p)->ksize, (p)->dsize) + +/* Get the number of bytes in the user's key/data pair. */ +#define NBLEAFDBT(ksize, dsize) \ + LALIGN(sizeof(u_int32_t) + sizeof(u_int32_t) + sizeof(u_char) + \ + (ksize) + (dsize)) + +/* Copy a BLEAF entry to the page. */ +#define WR_BLEAF(p, key, data, flags) { \ + *(u_int32_t *)p = key->size; \ + p += sizeof(u_int32_t); \ + *(u_int32_t *)p = data->size; \ + p += sizeof(u_int32_t); \ + *(u_char *)p = flags; \ + p += sizeof(u_char); \ + memmove(p, key->data, key->size); \ + p += key->size; \ + memmove(p, data->data, data->size); \ +} + +/* For the recno leaf pages, the item is a data entry. */ +typedef struct _rleaf { + u_int32_t dsize; /* size of data */ + u_char flags; /* P_BIGDATA */ + char bytes[1]; +} RLEAF; + +/* Get the page's RLEAF structure at index indx. */ +#define GETRLEAF(pg, indx) \ + ((RLEAF *)((char *)(pg) + (pg)->linp[indx])) + +/* Get the number of bytes in the entry. */ +#define NRLEAF(p) NRLEAFDBT((p)->dsize) + +/* Get the number of bytes from the user's data. */ +#define NRLEAFDBT(dsize) \ + LALIGN(sizeof(u_int32_t) + sizeof(u_char) + (dsize)) + +/* Copy a RLEAF entry to the page. */ +#define WR_RLEAF(p, data, flags) { \ + *(u_int32_t *)p = data->size; \ + p += sizeof(u_int32_t); \ + *(u_char *)p = flags; \ + p += sizeof(u_char); \ + memmove(p, data->data, data->size); \ +} + +/* + * A record in the tree is either a pointer to a page and an index in the page + * or a page number and an index. These structures are used as a cursor, stack + * entry and search returns as well as to pass records to other routines. + * + * One comment about searches. Internal page searches must find the largest + * record less than key in the tree so that descents work. Leaf page searches + * must find the smallest record greater than key so that the returned index + * is the record's correct position for insertion. + * + * One comment about cursors. The cursor key is never removed from the tree, + * even if deleted. This is because it is quite difficult to decide where the + * cursor should be when other keys have been inserted/deleted in the tree; + * duplicate keys make it impossible. This scheme does require extra work + * though, to make sure that we don't perform an operation on a deleted key. + */ +typedef struct _epgno { + pgno_t pgno; /* the page number */ + indx_t index; /* the index on the page */ +} EPGNO; + +typedef struct _epg { + PAGE *page; /* the (pinned) page */ + indx_t index; /* the index on the page */ +} EPG; + +/* + * The metadata of the tree. The m_nrecs field is used only by the RECNO code. + * This is because the btree doesn't really need it and it requires that every + * put or delete call modify the metadata. + */ +typedef struct _btmeta { + u_int32_t m_magic; /* magic number */ + u_int32_t m_version; /* version */ + u_int32_t m_psize; /* page size */ + u_int32_t m_free; /* page number of first free page */ + u_int32_t m_nrecs; /* R: number of records */ +#define SAVEMETA (B_NODUPS | R_RECNO) + u_int32_t m_flags; /* bt_flags & SAVEMETA */ + u_int32_t m_unused; /* unused */ +} BTMETA; + +/* The in-memory btree/recno data structure. */ +typedef struct _btree { + MPOOL *bt_mp; /* memory pool cookie */ + + DB *bt_dbp; /* pointer to enclosing DB */ + + EPG bt_cur; /* current (pinned) page */ + PAGE *bt_pinned; /* page pinned across calls */ + + EPGNO bt_bcursor; /* B: btree cursor */ + recno_t bt_rcursor; /* R: recno cursor (1-based) */ + +#define BT_POP(t) (t->bt_sp ? t->bt_stack + --t->bt_sp : NULL) +#define BT_CLR(t) (t->bt_sp = 0) + EPGNO *bt_stack; /* stack of parent pages */ + u_int bt_sp; /* current stack pointer */ + u_int bt_maxstack; /* largest stack */ + + char *bt_kbuf; /* key buffer */ + size_t bt_kbufsz; /* key buffer size */ + char *bt_dbuf; /* data buffer */ + size_t bt_dbufsz; /* data buffer size */ + + int bt_fd; /* tree file descriptor */ + + pgno_t bt_free; /* next free page */ + u_int32_t bt_psize; /* page size */ + indx_t bt_ovflsize; /* cut-off for key/data overflow */ + int bt_lorder; /* byte order */ + /* sorted order */ + enum { NOT, BACK, FORWARD } bt_order; + EPGNO bt_last; /* last insert */ + + /* B: key comparison function */ + int (*bt_cmp) __P((const DBT *, const DBT *)); + /* B: prefix comparison function */ + size_t (*bt_pfx) __P((const DBT *, const DBT *)); + /* R: recno input function */ + int (*bt_irec) __P((struct _btree *, recno_t)); + + FILE *bt_rfp; /* R: record FILE pointer */ + int bt_rfd; /* R: record file descriptor */ + + caddr_t bt_cmap; /* R: current point in mapped space */ + caddr_t bt_smap; /* R: start of mapped space */ + caddr_t bt_emap; /* R: end of mapped space */ + size_t bt_msize; /* R: size of mapped region. */ + + recno_t bt_nrecs; /* R: number of records */ + size_t bt_reclen; /* R: fixed record length */ + u_char bt_bval; /* R: delimiting byte/pad character */ + +/* + * NB: + * B_NODUPS and R_RECNO are stored on disk, and may not be changed. + */ +#define B_DELCRSR 0x00001 /* cursor has been deleted */ +#define B_INMEM 0x00002 /* in-memory tree */ +#define B_METADIRTY 0x00004 /* need to write metadata */ +#define B_MODIFIED 0x00008 /* tree modified */ +#define B_NEEDSWAP 0x00010 /* if byte order requires swapping */ +#define B_NODUPS 0x00020 /* no duplicate keys permitted */ +#define B_RDONLY 0x00040 /* read-only tree */ +#define R_RECNO 0x00080 /* record oriented tree */ +#define B_SEQINIT 0x00100 /* sequential scan initialized */ + +#define R_CLOSEFP 0x00200 /* opened a file pointer */ +#define R_EOF 0x00400 /* end of input file reached. */ +#define R_FIXLEN 0x00800 /* fixed length records */ +#define R_MEMMAPPED 0x01000 /* memory mapped file. */ +#define R_INMEM 0x02000 /* in-memory file */ +#define R_MODIFIED 0x04000 /* modified file */ +#define R_RDONLY 0x08000 /* read-only file */ + +#define B_DB_LOCK 0x10000 /* DB_LOCK specified. */ +#define B_DB_SHMEM 0x20000 /* DB_SHMEM specified. */ +#define B_DB_TXN 0x40000 /* DB_TXN specified. */ + + u_int32_t bt_flags; /* btree state */ +} BTREE; + +#define SET(t, f) ((t)->bt_flags |= (f)) +#define CLR(t, f) ((t)->bt_flags &= ~(f)) +#define ISSET(t, f) ((t)->bt_flags & (f)) + +#include "extern.h" diff --git a/lib/libc/db/btree/extern.h b/lib/libc/db/btree/extern.h new file mode 100644 index 00000000000..6d09c2eb204 --- /dev/null +++ b/lib/libc/db/btree/extern.h @@ -0,0 +1,72 @@ +/* $NetBSD: extern.h,v 1.5 1995/02/27 13:21:12 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)extern.h 8.4 (Berkeley) 6/4/94 + */ + +int __bt_close __P((DB *)); +int __bt_cmp __P((BTREE *, const DBT *, EPG *)); +int __bt_crsrdel __P((BTREE *, EPGNO *)); +int __bt_defcmp __P((const DBT *, const DBT *)); +size_t __bt_defpfx __P((const DBT *, const DBT *)); +int __bt_delete __P((const DB *, const DBT *, u_int)); +int __bt_dleaf __P((BTREE *, PAGE *, int)); +int __bt_fd __P((const DB *)); +EPG *__bt_first __P((BTREE *, const DBT *, int *)); +int __bt_free __P((BTREE *, PAGE *)); +int __bt_get __P((const DB *, const DBT *, DBT *, u_int)); +PAGE *__bt_new __P((BTREE *, pgno_t *)); +void __bt_pgin __P((void *, pgno_t, void *)); +void __bt_pgout __P((void *, pgno_t, void *)); +int __bt_push __P((BTREE *, pgno_t, int)); +int __bt_put __P((const DB *dbp, DBT *, const DBT *, u_int)); +int __bt_ret __P((BTREE *, EPG *, DBT *, DBT *)); +EPG *__bt_search __P((BTREE *, const DBT *, int *)); +int __bt_seq __P((const DB *, DBT *, DBT *, u_int)); +int __bt_split __P((BTREE *, PAGE *, + const DBT *, const DBT *, int, size_t, u_int32_t)); +int __bt_sync __P((const DB *, u_int)); + +int __ovfl_delete __P((BTREE *, void *)); +int __ovfl_get __P((BTREE *, void *, size_t *, char **, size_t *)); +int __ovfl_put __P((BTREE *, const DBT *, pgno_t *)); + +#ifdef DEBUG +void __bt_dnpage __P((DB *, pgno_t)); +void __bt_dpage __P((PAGE *)); +void __bt_dump __P((DB *)); +#endif +#ifdef STATISTICS +void __bt_stat __P((DB *)); +#endif diff --git a/lib/libc/db/db/Makefile.inc b/lib/libc/db/db/Makefile.inc new file mode 100644 index 00000000000..9755e8f2019 --- /dev/null +++ b/lib/libc/db/db/Makefile.inc @@ -0,0 +1,6 @@ +# $NetBSD: Makefile.inc,v 1.4 1995/02/27 13:21:22 cgd Exp $ +# @(#)Makefile.inc 8.1 (Berkeley) 6/4/93 + +.PATH: ${.CURDIR}/db/db + +SRCS+= db.c diff --git a/lib/libc/db/db/db.c b/lib/libc/db/db/db.c new file mode 100644 index 00000000000..791c2552234 --- /dev/null +++ b/lib/libc/db/db/db.c @@ -0,0 +1,105 @@ +/* $NetBSD: db.c,v 1.7 1995/02/27 13:21:27 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)db.c 8.4 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: db.c,v 1.7 1995/02/27 13:21:27 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <fcntl.h> +#include <stddef.h> +#include <stdio.h> + +#include <db.h> + +DB * +dbopen(fname, flags, mode, type, openinfo) + const char *fname; + int flags, mode; + DBTYPE type; + const void *openinfo; +{ + +#define DB_FLAGS (DB_LOCK | DB_SHMEM | DB_TXN) +#define USE_OPEN_FLAGS \ + (O_CREAT | O_EXCL | O_EXLOCK | O_NONBLOCK | O_RDONLY | \ + O_RDWR | O_SHLOCK | O_TRUNC) + + if ((flags & ~(USE_OPEN_FLAGS | DB_FLAGS)) == 0) + switch (type) { + case DB_BTREE: + return (__bt_open(fname, flags & USE_OPEN_FLAGS, + mode, openinfo, flags & DB_FLAGS)); + case DB_HASH: + return (__hash_open(fname, flags & USE_OPEN_FLAGS, + mode, openinfo, flags & DB_FLAGS)); + case DB_RECNO: + return (__rec_open(fname, flags & USE_OPEN_FLAGS, + mode, openinfo, flags & DB_FLAGS)); + } + errno = EINVAL; + return (NULL); +} + +static int +__dberr() +{ + return (RET_ERROR); +} + +/* + * __DBPANIC -- Stop. + * + * Parameters: + * dbp: pointer to the DB structure. + */ +void +__dbpanic(dbp) + DB *dbp; +{ + /* The only thing that can succeed is a close. */ + dbp->del = (int (*)())__dberr; + dbp->fd = (int (*)())__dberr; + dbp->get = (int (*)())__dberr; + dbp->put = (int (*)())__dberr; + dbp->seq = (int (*)())__dberr; + dbp->sync = (int (*)())__dberr; +} diff --git a/lib/libc/db/hash/Makefile.inc b/lib/libc/db/hash/Makefile.inc new file mode 100644 index 00000000000..00f339ac2ed --- /dev/null +++ b/lib/libc/db/hash/Makefile.inc @@ -0,0 +1,7 @@ +# $NetBSD: Makefile.inc,v 1.5 1995/02/27 13:21:44 cgd Exp $ +# @(#)Makefile.inc 8.1 (Berkeley) 6/4/93 + +.PATH: ${.CURDIR}/db/hash + +SRCS+= hash.c hash_bigkey.c hash_buf.c hash_func.c hash_log2.c \ + hash_page.c hsearch.c ndbm.c diff --git a/lib/libc/db/hash/README b/lib/libc/db/hash/README new file mode 100644 index 00000000000..fd1fc653a92 --- /dev/null +++ b/lib/libc/db/hash/README @@ -0,0 +1,73 @@ +# $NetBSD: README,v 1.3 1995/02/27 13:21:52 cgd Exp $ +# @(#)README 8.1 (Berkeley) 6/4/93 + +This package implements a superset of the hsearch and dbm/ndbm libraries. + +Test Programs: + All test programs which need key/data pairs expect them entered + with key and data on separate lines + + tcreat3.c + Takes + bucketsize (bsize), + fill factor (ffactor), and + initial number of elements (nelem). + Creates a hash table named hashtest containing the + keys/data pairs entered from standard in. + thash4.c + Takes + bucketsize (bsize), + fill factor (ffactor), + initial number of elements (nelem) + bytes of cache (ncached), and + file from which to read data (fname) + Creates a table from the key/data pairs on standard in and + then does a read of each key/data in fname + tdel.c + Takes + bucketsize (bsize), and + fill factor (ffactor). + file from which to read data (fname) + Reads each key/data pair from fname and deletes the + key from the hash table hashtest + tseq.c + Reads the key/data pairs in the file hashtest and writes them + to standard out. + tread2.c + Takes + butes of cache (ncached). + Reads key/data pairs from standard in and looks them up + in the file hashtest. + tverify.c + Reads key/data pairs from standard in, looks them up + in the file hashtest, and verifies that the data is + correct. + +NOTES: + +The file search.h is provided for using the hsearch compatible interface +on BSD systems. On System V derived systems, search.h should appear in +/usr/include. + +The man page ../man/db.3 explains the interface to the hashing system. +The file hash.ps is a postscript copy of a paper explaining +the history, implementation, and performance of the hash package. + +"bugs" or idiosyncracies + +If you have a lot of overflows, it is possible to run out of overflow +pages. Currently, this will cause a message to be printed on stderr. +Eventually, this will be indicated by a return error code. + +If you are using the ndbm interface and exit without flushing or closing the +file, you may lose updates since the package buffers all writes. Also, +the db interface only creates a single database file. To avoid overwriting +the user's original file, the suffix ".db" is appended to the file name +passed to dbm_open. Additionally, if your code "knows" about the historic +.dir and .pag files, it will break. + +There is a fundamental difference between this package and the old hsearch. +Hsearch requires the user to maintain the keys and data in the application's +allocated memory while hash takes care of all storage management. The down +side is that the byte strings passed in the ENTRY structure must be null +terminated (both the keys and the data). diff --git a/lib/libc/db/hash/extern.h b/lib/libc/db/hash/extern.h new file mode 100644 index 00000000000..af98921ef19 --- /dev/null +++ b/lib/libc/db/hash/extern.h @@ -0,0 +1,67 @@ +/* $NetBSD: extern.h,v 1.4 1995/02/27 13:21:55 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)extern.h 8.4 (Berkeley) 6/16/94 + */ + +BUFHEAD *__add_ovflpage __P((HTAB *, BUFHEAD *)); +int __addel __P((HTAB *, BUFHEAD *, const DBT *, const DBT *)); +int __big_delete __P((HTAB *, BUFHEAD *)); +int __big_insert __P((HTAB *, BUFHEAD *, const DBT *, const DBT *)); +int __big_keydata __P((HTAB *, BUFHEAD *, DBT *, DBT *, int)); +int __big_return __P((HTAB *, BUFHEAD *, int, DBT *, int)); +int __big_split __P((HTAB *, BUFHEAD *, BUFHEAD *, BUFHEAD *, + int, u_int32_t, SPLIT_RETURN *)); +int __buf_free __P((HTAB *, int, int)); +void __buf_init __P((HTAB *, int)); +u_int32_t __call_hash __P((HTAB *, char *, int)); +int __delpair __P((HTAB *, BUFHEAD *, int)); +int __expand_table __P((HTAB *)); +int __find_bigpair __P((HTAB *, BUFHEAD *, int, char *, int)); +u_int16_t __find_last_page __P((HTAB *, BUFHEAD **)); +void __free_ovflpage __P((HTAB *, BUFHEAD *)); +BUFHEAD *__get_buf __P((HTAB *, u_int32_t, BUFHEAD *, int)); +int __get_page __P((HTAB *, char *, u_int32_t, int, int, int)); +int __ibitmap __P((HTAB *, int, int, int)); +u_int32_t __log2 __P((u_int32_t)); +int __put_page __P((HTAB *, char *, u_int32_t, int, int)); +void __reclaim_buf __P((HTAB *, BUFHEAD *)); +int __split_page __P((HTAB *, u_int32_t, u_int32_t)); + +/* Default hash routine. */ +extern u_int32_t (*__default_hash) __P((const void *, size_t)); + +#ifdef HASH_STATISTICS +extern int hash_accesses, hash_collisions, hash_expansions, hash_overflows; +#endif diff --git a/lib/libc/db/hash/hash.c b/lib/libc/db/hash/hash.c new file mode 100644 index 00000000000..2fc3281cb9f --- /dev/null +++ b/lib/libc/db/hash/hash.c @@ -0,0 +1,1000 @@ +/* $NetBSD: hash.c,v 1.8 1995/02/27 13:21:59 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hash.c 8.9 (Berkeley) 6/16/94"; +#else +static char rcsid[] = "$NetBSD: hash.c,v 1.8 1995/02/27 13:21:59 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> +#include <sys/stat.h> + +#include <errno.h> +#include <fcntl.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#ifdef DEBUG +#include <assert.h> +#endif + +#include <db.h> +#include "hash.h" +#include "page.h" +#include "extern.h" + +static int alloc_segs __P((HTAB *, int)); +static int flush_meta __P((HTAB *)); +static int hash_access __P((HTAB *, ACTION, DBT *, DBT *)); +static int hash_close __P((DB *)); +static int hash_delete __P((const DB *, const DBT *, u_int32_t)); +static int hash_fd __P((const DB *)); +static int hash_get __P((const DB *, const DBT *, DBT *, u_int32_t)); +static int hash_put __P((const DB *, DBT *, const DBT *, u_int32_t)); +static void *hash_realloc __P((SEGMENT **, int, int)); +static int hash_seq __P((const DB *, DBT *, DBT *, u_int32_t)); +static int hash_sync __P((const DB *, u_int32_t)); +static int hdestroy __P((HTAB *)); +static HTAB *init_hash __P((HTAB *, const char *, HASHINFO *)); +static int init_htab __P((HTAB *, int)); +#if BYTE_ORDER == LITTLE_ENDIAN +static void swap_header __P((HTAB *)); +static void swap_header_copy __P((HASHHDR *, HASHHDR *)); +#endif + +/* Fast arithmetic, relying on powers of 2, */ +#define MOD(x, y) ((x) & ((y) - 1)) + +#define RETURN_ERROR(ERR, LOC) { save_errno = ERR; goto LOC; } + +/* Return values */ +#define SUCCESS (0) +#define ERROR (-1) +#define ABNORMAL (1) + +#ifdef HASH_STATISTICS +int hash_accesses, hash_collisions, hash_expansions, hash_overflows; +#endif + +/************************** INTERFACE ROUTINES ***************************/ +/* OPEN/CLOSE */ + +extern DB * +__hash_open(file, flags, mode, info, dflags) + const char *file; + int flags, mode, dflags; + const HASHINFO *info; /* Special directives for create */ +{ + HTAB *hashp; + struct stat statbuf; + DB *dbp; + int bpages, hdrsize, new_table, nsegs, save_errno; + + if ((flags & O_ACCMODE) == O_WRONLY) { + errno = EINVAL; + return (NULL); + } + + if (!(hashp = (HTAB *)calloc(1, sizeof(HTAB)))) + return (NULL); + hashp->fp = -1; + + /* + * Even if user wants write only, we need to be able to read + * the actual file, so we need to open it read/write. But, the + * field in the hashp structure needs to be accurate so that + * we can check accesses. + */ + hashp->flags = flags; + + new_table = 0; + if (!file || (flags & O_TRUNC) || + (stat(file, &statbuf) && (errno == ENOENT))) { + if (errno == ENOENT) + errno = 0; /* Just in case someone looks at errno */ + new_table = 1; + } + if (file) { + if ((hashp->fp = open(file, flags, mode)) == -1) + RETURN_ERROR(errno, error0); + (void)fcntl(hashp->fp, F_SETFD, 1); + } + if (new_table) { + if (!(hashp = init_hash(hashp, file, (HASHINFO *)info))) + RETURN_ERROR(errno, error1); + } else { + /* Table already exists */ + if (info && info->hash) + hashp->hash = info->hash; + else + hashp->hash = __default_hash; + + hdrsize = read(hashp->fp, &hashp->hdr, sizeof(HASHHDR)); +#if BYTE_ORDER == LITTLE_ENDIAN + swap_header(hashp); +#endif + if (hdrsize == -1) + RETURN_ERROR(errno, error1); + if (hdrsize != sizeof(HASHHDR)) + RETURN_ERROR(EFTYPE, error1); + /* Verify file type, versions and hash function */ + if (hashp->MAGIC != HASHMAGIC) + RETURN_ERROR(EFTYPE, error1); +#define OLDHASHVERSION 1 + if (hashp->VERSION != HASHVERSION && + hashp->VERSION != OLDHASHVERSION) + RETURN_ERROR(EFTYPE, error1); + if (hashp->hash(CHARKEY, sizeof(CHARKEY)) != hashp->H_CHARKEY) + RETURN_ERROR(EFTYPE, error1); + /* + * Figure out how many segments we need. Max_Bucket is the + * maximum bucket number, so the number of buckets is + * max_bucket + 1. + */ + nsegs = (hashp->MAX_BUCKET + 1 + hashp->SGSIZE - 1) / + hashp->SGSIZE; + hashp->nsegs = 0; + if (alloc_segs(hashp, nsegs)) + /* + * If alloc_segs fails, table will have been destroyed + * and errno will have been set. + */ + return (NULL); + /* Read in bitmaps */ + bpages = (hashp->SPARES[hashp->OVFL_POINT] + + (hashp->BSIZE << BYTE_SHIFT) - 1) >> + (hashp->BSHIFT + BYTE_SHIFT); + + hashp->nmaps = bpages; + (void)memset(&hashp->mapp[0], 0, bpages * sizeof(u_int32_t *)); + } + + /* Initialize Buffer Manager */ + if (info && info->cachesize) + __buf_init(hashp, info->cachesize); + else + __buf_init(hashp, DEF_BUFSIZE); + + hashp->new_file = new_table; + hashp->save_file = file && (hashp->flags & O_RDWR); + hashp->cbucket = -1; + if (!(dbp = (DB *)malloc(sizeof(DB)))) { + save_errno = errno; + hdestroy(hashp); + errno = save_errno; + return (NULL); + } + dbp->internal = hashp; + dbp->close = hash_close; + dbp->del = hash_delete; + dbp->fd = hash_fd; + dbp->get = hash_get; + dbp->put = hash_put; + dbp->seq = hash_seq; + dbp->sync = hash_sync; + dbp->type = DB_HASH; + +#ifdef DEBUG + (void)fprintf(stderr, +"%s\n%s%x\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%d\n%s%x\n%s%x\n%s%d\n%s%d\n", + "init_htab:", + "TABLE POINTER ", hashp, + "BUCKET SIZE ", hashp->BSIZE, + "BUCKET SHIFT ", hashp->BSHIFT, + "DIRECTORY SIZE ", hashp->DSIZE, + "SEGMENT SIZE ", hashp->SGSIZE, + "SEGMENT SHIFT ", hashp->SSHIFT, + "FILL FACTOR ", hashp->FFACTOR, + "MAX BUCKET ", hashp->MAX_BUCKET, + "OVFL POINT ", hashp->OVFL_POINT, + "LAST FREED ", hashp->LAST_FREED, + "HIGH MASK ", hashp->HIGH_MASK, + "LOW MASK ", hashp->LOW_MASK, + "NSEGS ", hashp->nsegs, + "NKEYS ", hashp->NKEYS); +#endif +#ifdef HASH_STATISTICS + hash_overflows = hash_accesses = hash_collisions = hash_expansions = 0; +#endif + return (dbp); + +error1: + if (hashp != NULL) + (void)close(hashp->fp); + +error0: + free(hashp); + errno = save_errno; + return (NULL); +} + +static int +hash_close(dbp) + DB *dbp; +{ + HTAB *hashp; + int retval; + + if (!dbp) + return (ERROR); + + hashp = (HTAB *)dbp->internal; + retval = hdestroy(hashp); + free(dbp); + return (retval); +} + +static int +hash_fd(dbp) + const DB *dbp; +{ + HTAB *hashp; + + if (!dbp) + return (ERROR); + + hashp = (HTAB *)dbp->internal; + if (hashp->fp == -1) { + errno = ENOENT; + return (-1); + } + return (hashp->fp); +} + +/************************** LOCAL CREATION ROUTINES **********************/ +static HTAB * +init_hash(hashp, file, info) + HTAB *hashp; + const char *file; + HASHINFO *info; +{ + struct stat statbuf; + int nelem; + + nelem = 1; + hashp->NKEYS = 0; + hashp->LORDER = BYTE_ORDER; + hashp->BSIZE = DEF_BUCKET_SIZE; + hashp->BSHIFT = DEF_BUCKET_SHIFT; + hashp->SGSIZE = DEF_SEGSIZE; + hashp->SSHIFT = DEF_SEGSIZE_SHIFT; + hashp->DSIZE = DEF_DIRSIZE; + hashp->FFACTOR = DEF_FFACTOR; + hashp->hash = __default_hash; + memset(hashp->SPARES, 0, sizeof(hashp->SPARES)); + memset(hashp->BITMAPS, 0, sizeof (hashp->BITMAPS)); + + /* Fix bucket size to be optimal for file system */ + if (file != NULL) { + if (stat(file, &statbuf)) + return (NULL); + hashp->BSIZE = statbuf.st_blksize; + hashp->BSHIFT = __log2(hashp->BSIZE); + } + + if (info) { + if (info->bsize) { + /* Round pagesize up to power of 2 */ + hashp->BSHIFT = __log2(info->bsize); + hashp->BSIZE = 1 << hashp->BSHIFT; + if (hashp->BSIZE > MAX_BSIZE) { + errno = EINVAL; + return (NULL); + } + } + if (info->ffactor) + hashp->FFACTOR = info->ffactor; + if (info->hash) + hashp->hash = info->hash; + if (info->nelem) + nelem = info->nelem; + if (info->lorder) { + if (info->lorder != BIG_ENDIAN && + info->lorder != LITTLE_ENDIAN) { + errno = EINVAL; + return (NULL); + } + hashp->LORDER = info->lorder; + } + } + /* init_htab should destroy the table and set errno if it fails */ + if (init_htab(hashp, nelem)) + return (NULL); + else + return (hashp); +} +/* + * This calls alloc_segs which may run out of memory. Alloc_segs will destroy + * the table and set errno, so we just pass the error information along. + * + * Returns 0 on No Error + */ +static int +init_htab(hashp, nelem) + HTAB *hashp; + int nelem; +{ + register int nbuckets, nsegs; + int l2; + + /* + * Divide number of elements by the fill factor and determine a + * desired number of buckets. Allocate space for the next greater + * power of two number of buckets. + */ + nelem = (nelem - 1) / hashp->FFACTOR + 1; + + l2 = __log2(MAX(nelem, 2)); + nbuckets = 1 << l2; + + hashp->SPARES[l2] = l2 + 1; + hashp->SPARES[l2 + 1] = l2 + 1; + hashp->OVFL_POINT = l2; + hashp->LAST_FREED = 2; + + /* First bitmap page is at: splitpoint l2 page offset 1 */ + if (__ibitmap(hashp, OADDR_OF(l2, 1), l2 + 1, 0)) + return (-1); + + hashp->MAX_BUCKET = hashp->LOW_MASK = nbuckets - 1; + hashp->HIGH_MASK = (nbuckets << 1) - 1; + hashp->HDRPAGES = ((MAX(sizeof(HASHHDR), MINHDRSIZE) - 1) >> + hashp->BSHIFT) + 1; + + nsegs = (nbuckets - 1) / hashp->SGSIZE + 1; + nsegs = 1 << __log2(nsegs); + + if (nsegs > hashp->DSIZE) + hashp->DSIZE = nsegs; + return (alloc_segs(hashp, nsegs)); +} + +/********************** DESTROY/CLOSE ROUTINES ************************/ + +/* + * Flushes any changes to the file if necessary and destroys the hashp + * structure, freeing all allocated space. + */ +static int +hdestroy(hashp) + HTAB *hashp; +{ + int i, save_errno; + + save_errno = 0; + +#ifdef HASH_STATISTICS + (void)fprintf(stderr, "hdestroy: accesses %ld collisions %ld\n", + hash_accesses, hash_collisions); + (void)fprintf(stderr, "hdestroy: expansions %ld\n", + hash_expansions); + (void)fprintf(stderr, "hdestroy: overflows %ld\n", + hash_overflows); + (void)fprintf(stderr, "keys %ld maxp %d segmentcount %d\n", + hashp->NKEYS, hashp->MAX_BUCKET, hashp->nsegs); + + for (i = 0; i < NCACHED; i++) + (void)fprintf(stderr, + "spares[%d] = %d\n", i, hashp->SPARES[i]); +#endif + /* + * Call on buffer manager to free buffers, and if required, + * write them to disk. + */ + if (__buf_free(hashp, 1, hashp->save_file)) + save_errno = errno; + if (hashp->dir) { + free(*hashp->dir); /* Free initial segments */ + /* Free extra segments */ + while (hashp->exsegs--) + free(hashp->dir[--hashp->nsegs]); + free(hashp->dir); + } + if (flush_meta(hashp) && !save_errno) + save_errno = errno; + /* Free Bigmaps */ + for (i = 0; i < hashp->nmaps; i++) + if (hashp->mapp[i]) + free(hashp->mapp[i]); + + if (hashp->fp != -1) + (void)close(hashp->fp); + + free(hashp); + + if (save_errno) { + errno = save_errno; + return (ERROR); + } + return (SUCCESS); +} +/* + * Write modified pages to disk + * + * Returns: + * 0 == OK + * -1 ERROR + */ +static int +hash_sync(dbp, flags) + const DB *dbp; + u_int32_t flags; +{ + HTAB *hashp; + + if (flags != 0) { + errno = EINVAL; + return (ERROR); + } + + if (!dbp) + return (ERROR); + + hashp = (HTAB *)dbp->internal; + if (!hashp->save_file) + return (0); + if (__buf_free(hashp, 0, 1) || flush_meta(hashp)) + return (ERROR); + hashp->new_file = 0; + return (0); +} + +/* + * Returns: + * 0 == OK + * -1 indicates that errno should be set + */ +static int +flush_meta(hashp) + HTAB *hashp; +{ + HASHHDR *whdrp; +#if BYTE_ORDER == LITTLE_ENDIAN + HASHHDR whdr; +#endif + int fp, i, wsize; + + if (!hashp->save_file) + return (0); + hashp->MAGIC = HASHMAGIC; + hashp->VERSION = HASHVERSION; + hashp->H_CHARKEY = hashp->hash(CHARKEY, sizeof(CHARKEY)); + + fp = hashp->fp; + whdrp = &hashp->hdr; +#if BYTE_ORDER == LITTLE_ENDIAN + whdrp = &whdr; + swap_header_copy(&hashp->hdr, whdrp); +#endif + if ((lseek(fp, (off_t)0, SEEK_SET) == -1) || + ((wsize = write(fp, whdrp, sizeof(HASHHDR))) == -1)) + return (-1); + else + if (wsize != sizeof(HASHHDR)) { + errno = EFTYPE; + hashp->errno = errno; + return (-1); + } + for (i = 0; i < NCACHED; i++) + if (hashp->mapp[i]) + if (__put_page(hashp, (char *)hashp->mapp[i], + hashp->BITMAPS[i], 0, 1)) + return (-1); + return (0); +} + +/*******************************SEARCH ROUTINES *****************************/ +/* + * All the access routines return + * + * Returns: + * 0 on SUCCESS + * 1 to indicate an external ERROR (i.e. key not found, etc) + * -1 to indicate an internal ERROR (i.e. out of memory, etc) + */ +static int +hash_get(dbp, key, data, flag) + const DB *dbp; + const DBT *key; + DBT *data; + u_int32_t flag; +{ + HTAB *hashp; + + hashp = (HTAB *)dbp->internal; + if (flag) { + hashp->errno = errno = EINVAL; + return (ERROR); + } + return (hash_access(hashp, HASH_GET, (DBT *)key, data)); +} + +static int +hash_put(dbp, key, data, flag) + const DB *dbp; + DBT *key; + const DBT *data; + u_int32_t flag; +{ + HTAB *hashp; + + hashp = (HTAB *)dbp->internal; + if (flag && flag != R_NOOVERWRITE) { + hashp->errno = errno = EINVAL; + return (ERROR); + } + if ((hashp->flags & O_ACCMODE) == O_RDONLY) { + hashp->errno = errno = EPERM; + return (ERROR); + } + return (hash_access(hashp, flag == R_NOOVERWRITE ? + HASH_PUTNEW : HASH_PUT, (DBT *)key, (DBT *)data)); +} + +static int +hash_delete(dbp, key, flag) + const DB *dbp; + const DBT *key; + u_int32_t flag; /* Ignored */ +{ + HTAB *hashp; + + hashp = (HTAB *)dbp->internal; + if (flag && flag != R_CURSOR) { + hashp->errno = errno = EINVAL; + return (ERROR); + } + if ((hashp->flags & O_ACCMODE) == O_RDONLY) { + hashp->errno = errno = EPERM; + return (ERROR); + } + return (hash_access(hashp, HASH_DELETE, (DBT *)key, NULL)); +} + +/* + * Assume that hashp has been set in wrapper routine. + */ +static int +hash_access(hashp, action, key, val) + HTAB *hashp; + ACTION action; + DBT *key, *val; +{ + register BUFHEAD *rbufp; + BUFHEAD *bufp, *save_bufp; + register u_int16_t *bp; + register int n, ndx, off, size; + register char *kp; + u_int16_t pageno; + +#ifdef HASH_STATISTICS + hash_accesses++; +#endif + + off = hashp->BSIZE; + size = key->size; + kp = (char *)key->data; + rbufp = __get_buf(hashp, __call_hash(hashp, kp, size), NULL, 0); + if (!rbufp) + return (ERROR); + save_bufp = rbufp; + + /* Pin the bucket chain */ + rbufp->flags |= BUF_PIN; + for (bp = (u_int16_t *)rbufp->page, n = *bp++, ndx = 1; ndx < n;) + if (bp[1] >= REAL_KEY) { + /* Real key/data pair */ + if (size == off - *bp && + memcmp(kp, rbufp->page + *bp, size) == 0) + goto found; + off = bp[1]; +#ifdef HASH_STATISTICS + hash_collisions++; +#endif + bp += 2; + ndx += 2; + } else if (bp[1] == OVFLPAGE) { + rbufp = __get_buf(hashp, *bp, rbufp, 0); + if (!rbufp) { + save_bufp->flags &= ~BUF_PIN; + return (ERROR); + } + /* FOR LOOP INIT */ + bp = (u_int16_t *)rbufp->page; + n = *bp++; + ndx = 1; + off = hashp->BSIZE; + } else if (bp[1] < REAL_KEY) { + if ((ndx = + __find_bigpair(hashp, rbufp, ndx, kp, size)) > 0) + goto found; + if (ndx == -2) { + bufp = rbufp; + if (!(pageno = + __find_last_page(hashp, &bufp))) { + ndx = 0; + rbufp = bufp; + break; /* FOR */ + } + rbufp = __get_buf(hashp, pageno, bufp, 0); + if (!rbufp) { + save_bufp->flags &= ~BUF_PIN; + return (ERROR); + } + /* FOR LOOP INIT */ + bp = (u_int16_t *)rbufp->page; + n = *bp++; + ndx = 1; + off = hashp->BSIZE; + } else { + save_bufp->flags &= ~BUF_PIN; + return (ERROR); + } + } + + /* Not found */ + switch (action) { + case HASH_PUT: + case HASH_PUTNEW: + if (__addel(hashp, rbufp, key, val)) { + save_bufp->flags &= ~BUF_PIN; + return (ERROR); + } else { + save_bufp->flags &= ~BUF_PIN; + return (SUCCESS); + } + case HASH_GET: + case HASH_DELETE: + default: + save_bufp->flags &= ~BUF_PIN; + return (ABNORMAL); + } + +found: + switch (action) { + case HASH_PUTNEW: + save_bufp->flags &= ~BUF_PIN; + return (ABNORMAL); + case HASH_GET: + bp = (u_int16_t *)rbufp->page; + if (bp[ndx + 1] < REAL_KEY) { + if (__big_return(hashp, rbufp, ndx, val, 0)) + return (ERROR); + } else { + val->data = (u_char *)rbufp->page + (int)bp[ndx + 1]; + val->size = bp[ndx] - bp[ndx + 1]; + } + break; + case HASH_PUT: + if ((__delpair(hashp, rbufp, ndx)) || + (__addel(hashp, rbufp, key, val))) { + save_bufp->flags &= ~BUF_PIN; + return (ERROR); + } + break; + case HASH_DELETE: + if (__delpair(hashp, rbufp, ndx)) + return (ERROR); + break; + default: + abort(); + } + save_bufp->flags &= ~BUF_PIN; + return (SUCCESS); +} + +static int +hash_seq(dbp, key, data, flag) + const DB *dbp; + DBT *key, *data; + u_int32_t flag; +{ + register u_int32_t bucket; + register BUFHEAD *bufp; + HTAB *hashp; + u_int16_t *bp, ndx; + + hashp = (HTAB *)dbp->internal; + if (flag && flag != R_FIRST && flag != R_NEXT) { + hashp->errno = errno = EINVAL; + return (ERROR); + } +#ifdef HASH_STATISTICS + hash_accesses++; +#endif + if ((hashp->cbucket < 0) || (flag == R_FIRST)) { + hashp->cbucket = 0; + hashp->cndx = 1; + hashp->cpage = NULL; + } + + for (bp = NULL; !bp || !bp[0]; ) { + if (!(bufp = hashp->cpage)) { + for (bucket = hashp->cbucket; + bucket <= hashp->MAX_BUCKET; + bucket++, hashp->cndx = 1) { + bufp = __get_buf(hashp, bucket, NULL, 0); + if (!bufp) + return (ERROR); + hashp->cpage = bufp; + bp = (u_int16_t *)bufp->page; + if (bp[0]) + break; + } + hashp->cbucket = bucket; + if (hashp->cbucket > hashp->MAX_BUCKET) { + hashp->cbucket = -1; + return (ABNORMAL); + } + } else + bp = (u_int16_t *)hashp->cpage->page; + +#ifdef DEBUG + assert(bp); + assert(bufp); +#endif + while (bp[hashp->cndx + 1] == OVFLPAGE) { + bufp = hashp->cpage = + __get_buf(hashp, bp[hashp->cndx], bufp, 0); + if (!bufp) + return (ERROR); + bp = (u_int16_t *)(bufp->page); + hashp->cndx = 1; + } + if (!bp[0]) { + hashp->cpage = NULL; + ++hashp->cbucket; + } + } + ndx = hashp->cndx; + if (bp[ndx + 1] < REAL_KEY) { + if (__big_keydata(hashp, bufp, key, data, 1)) + return (ERROR); + } else { + key->data = (u_char *)hashp->cpage->page + bp[ndx]; + key->size = (ndx > 1 ? bp[ndx - 1] : hashp->BSIZE) - bp[ndx]; + data->data = (u_char *)hashp->cpage->page + bp[ndx + 1]; + data->size = bp[ndx] - bp[ndx + 1]; + ndx += 2; + if (ndx > bp[0]) { + hashp->cpage = NULL; + hashp->cbucket++; + hashp->cndx = 1; + } else + hashp->cndx = ndx; + } + return (SUCCESS); +} + +/********************************* UTILITIES ************************/ + +/* + * Returns: + * 0 ==> OK + * -1 ==> Error + */ +extern int +__expand_table(hashp) + HTAB *hashp; +{ + u_int32_t old_bucket, new_bucket; + int dirsize, new_segnum, spare_ndx; + +#ifdef HASH_STATISTICS + hash_expansions++; +#endif + new_bucket = ++hashp->MAX_BUCKET; + old_bucket = (hashp->MAX_BUCKET & hashp->LOW_MASK); + + new_segnum = new_bucket >> hashp->SSHIFT; + + /* Check if we need a new segment */ + if (new_segnum >= hashp->nsegs) { + /* Check if we need to expand directory */ + if (new_segnum >= hashp->DSIZE) { + /* Reallocate directory */ + dirsize = hashp->DSIZE * sizeof(SEGMENT *); + if (!hash_realloc(&hashp->dir, dirsize, dirsize << 1)) + return (-1); + hashp->DSIZE = dirsize << 1; + } + if ((hashp->dir[new_segnum] = + (SEGMENT)calloc(hashp->SGSIZE, sizeof(SEGMENT))) == NULL) + return (-1); + hashp->exsegs++; + hashp->nsegs++; + } + /* + * If the split point is increasing (MAX_BUCKET's log base 2 + * * increases), we need to copy the current contents of the spare + * split bucket to the next bucket. + */ + spare_ndx = __log2(hashp->MAX_BUCKET + 1); + if (spare_ndx > hashp->OVFL_POINT) { + hashp->SPARES[spare_ndx] = hashp->SPARES[hashp->OVFL_POINT]; + hashp->OVFL_POINT = spare_ndx; + } + + if (new_bucket > hashp->HIGH_MASK) { + /* Starting a new doubling */ + hashp->LOW_MASK = hashp->HIGH_MASK; + hashp->HIGH_MASK = new_bucket | hashp->LOW_MASK; + } + /* Relocate records to the new bucket */ + return (__split_page(hashp, old_bucket, new_bucket)); +} + +/* + * If realloc guarantees that the pointer is not destroyed if the realloc + * fails, then this routine can go away. + */ +static void * +hash_realloc(p_ptr, oldsize, newsize) + SEGMENT **p_ptr; + int oldsize, newsize; +{ + register void *p; + + if (p = malloc(newsize)) { + memmove(p, *p_ptr, oldsize); + memset((char *)p + oldsize, 0, newsize - oldsize); + free(*p_ptr); + *p_ptr = p; + } + return (p); +} + +extern u_int32_t +__call_hash(hashp, k, len) + HTAB *hashp; + char *k; + int len; +{ + int n, bucket; + + n = hashp->hash(k, len); + bucket = n & hashp->HIGH_MASK; + if (bucket > hashp->MAX_BUCKET) + bucket = bucket & hashp->LOW_MASK; + return (bucket); +} + +/* + * Allocate segment table. On error, destroy the table and set errno. + * + * Returns 0 on success + */ +static int +alloc_segs(hashp, nsegs) + HTAB *hashp; + int nsegs; +{ + register int i; + register SEGMENT store; + + int save_errno; + + if ((hashp->dir = + (SEGMENT *)calloc(hashp->DSIZE, sizeof(SEGMENT *))) == NULL) { + save_errno = errno; + (void)hdestroy(hashp); + errno = save_errno; + return (-1); + } + /* Allocate segments */ + if ((store = + (SEGMENT)calloc(nsegs << hashp->SSHIFT, sizeof(SEGMENT))) == NULL) { + save_errno = errno; + (void)hdestroy(hashp); + errno = save_errno; + return (-1); + } + for (i = 0; i < nsegs; i++, hashp->nsegs++) + hashp->dir[i] = &store[i << hashp->SSHIFT]; + return (0); +} + +#if BYTE_ORDER == LITTLE_ENDIAN +/* + * Hashp->hdr needs to be byteswapped. + */ +static void +swap_header_copy(srcp, destp) + HASHHDR *srcp, *destp; +{ + int i; + + P_32_COPY(srcp->magic, destp->magic); + P_32_COPY(srcp->version, destp->version); + P_32_COPY(srcp->lorder, destp->lorder); + P_32_COPY(srcp->bsize, destp->bsize); + P_32_COPY(srcp->bshift, destp->bshift); + P_32_COPY(srcp->dsize, destp->dsize); + P_32_COPY(srcp->ssize, destp->ssize); + P_32_COPY(srcp->sshift, destp->sshift); + P_32_COPY(srcp->ovfl_point, destp->ovfl_point); + P_32_COPY(srcp->last_freed, destp->last_freed); + P_32_COPY(srcp->max_bucket, destp->max_bucket); + P_32_COPY(srcp->high_mask, destp->high_mask); + P_32_COPY(srcp->low_mask, destp->low_mask); + P_32_COPY(srcp->ffactor, destp->ffactor); + P_32_COPY(srcp->nkeys, destp->nkeys); + P_32_COPY(srcp->hdrpages, destp->hdrpages); + P_32_COPY(srcp->h_charkey, destp->h_charkey); + for (i = 0; i < NCACHED; i++) { + P_32_COPY(srcp->spares[i], destp->spares[i]); + P_16_COPY(srcp->bitmaps[i], destp->bitmaps[i]); + } +} + +static void +swap_header(hashp) + HTAB *hashp; +{ + HASHHDR *hdrp; + int i; + + hdrp = &hashp->hdr; + + M_32_SWAP(hdrp->magic); + M_32_SWAP(hdrp->version); + M_32_SWAP(hdrp->lorder); + M_32_SWAP(hdrp->bsize); + M_32_SWAP(hdrp->bshift); + M_32_SWAP(hdrp->dsize); + M_32_SWAP(hdrp->ssize); + M_32_SWAP(hdrp->sshift); + M_32_SWAP(hdrp->ovfl_point); + M_32_SWAP(hdrp->last_freed); + M_32_SWAP(hdrp->max_bucket); + M_32_SWAP(hdrp->high_mask); + M_32_SWAP(hdrp->low_mask); + M_32_SWAP(hdrp->ffactor); + M_32_SWAP(hdrp->nkeys); + M_32_SWAP(hdrp->hdrpages); + M_32_SWAP(hdrp->h_charkey); + for (i = 0; i < NCACHED; i++) { + M_32_SWAP(hdrp->spares[i]); + M_16_SWAP(hdrp->bitmaps[i]); + } +} +#endif diff --git a/lib/libc/db/hash/hash.h b/lib/libc/db/hash/hash.h new file mode 100644 index 00000000000..1ad93b74d9f --- /dev/null +++ b/lib/libc/db/hash/hash.h @@ -0,0 +1,295 @@ +/* $NetBSD: hash.h,v 1.5 1995/02/27 13:22:08 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)hash.h 8.3 (Berkeley) 5/31/94 + */ + +/* Operations */ +typedef enum { + HASH_GET, HASH_PUT, HASH_PUTNEW, HASH_DELETE, HASH_FIRST, HASH_NEXT +} ACTION; + +/* Buffer Management structures */ +typedef struct _bufhead BUFHEAD; + +struct _bufhead { + BUFHEAD *prev; /* LRU links */ + BUFHEAD *next; /* LRU links */ + BUFHEAD *ovfl; /* Overflow page buffer header */ + u_int32_t addr; /* Address of this page */ + char *page; /* Actual page data */ + char flags; +#define BUF_MOD 0x0001 +#define BUF_DISK 0x0002 +#define BUF_BUCKET 0x0004 +#define BUF_PIN 0x0008 +}; + +#define IS_BUCKET(X) ((X) & BUF_BUCKET) + +typedef BUFHEAD **SEGMENT; + +/* Hash Table Information */ +typedef struct hashhdr { /* Disk resident portion */ + int magic; /* Magic NO for hash tables */ + int version; /* Version ID */ + u_int32_t lorder; /* Byte Order */ + int bsize; /* Bucket/Page Size */ + int bshift; /* Bucket shift */ + int dsize; /* Directory Size */ + int ssize; /* Segment Size */ + int sshift; /* Segment shift */ + int ovfl_point; /* Where overflow pages are being + * allocated */ + int last_freed; /* Last overflow page freed */ + int max_bucket; /* ID of Maximum bucket in use */ + int high_mask; /* Mask to modulo into entire table */ + int low_mask; /* Mask to modulo into lower half of + * table */ + int ffactor; /* Fill factor */ + int nkeys; /* Number of keys in hash table */ + int hdrpages; /* Size of table header */ + int h_charkey; /* value of hash(CHARKEY) */ +#define NCACHED 32 /* number of bit maps and spare + * points */ + int spares[NCACHED];/* spare pages for overflow */ + u_int16_t bitmaps[NCACHED]; /* address of overflow page + * bitmaps */ +} HASHHDR; + +typedef struct htab { /* Memory resident data structure */ + HASHHDR hdr; /* Header */ + int nsegs; /* Number of allocated segments */ + int exsegs; /* Number of extra allocated + * segments */ + u_int32_t /* Hash function */ + (*hash)__P((const void *, size_t)); + int flags; /* Flag values */ + int fp; /* File pointer */ + char *tmp_buf; /* Temporary Buffer for BIG data */ + char *tmp_key; /* Temporary Buffer for BIG keys */ + BUFHEAD *cpage; /* Current page */ + int cbucket; /* Current bucket */ + int cndx; /* Index of next item on cpage */ + int errno; /* Error Number -- for DBM + * compatability */ + int new_file; /* Indicates if fd is backing store + * or no */ + int save_file; /* Indicates whether we need to flush + * file at + * exit */ + u_int32_t *mapp[NCACHED]; /* Pointers to page maps */ + int nmaps; /* Initial number of bitmaps */ + int nbufs; /* Number of buffers left to + * allocate */ + BUFHEAD bufhead; /* Header of buffer lru list */ + SEGMENT *dir; /* Hash Bucket directory */ +} HTAB; + +/* + * Constants + */ +#define MAX_BSIZE 65536 /* 2^16 */ +#define MIN_BUFFERS 6 +#define MINHDRSIZE 512 +#define DEF_BUFSIZE 65536 /* 64 K */ +#define DEF_BUCKET_SIZE 4096 +#define DEF_BUCKET_SHIFT 12 /* log2(BUCKET) */ +#define DEF_SEGSIZE 256 +#define DEF_SEGSIZE_SHIFT 8 /* log2(SEGSIZE) */ +#define DEF_DIRSIZE 256 +#define DEF_FFACTOR 65536 +#define MIN_FFACTOR 4 +#define SPLTMAX 8 +#define CHARKEY "%$sniglet^&" +#define NUMKEY 1038583 +#define BYTE_SHIFT 3 +#define INT_TO_BYTE 2 +#define INT_BYTE_SHIFT 5 +#define ALL_SET ((u_int32_t)0xFFFFFFFF) +#define ALL_CLEAR 0 + +#define PTROF(X) ((BUFHEAD *)((ptrdiff_t)(X)&~0x3)) +#define ISMOD(X) ((u_int32_t)(ptrdiff_t)(X)&0x1) +#define DOMOD(X) ((X) = (char *)((ptrdiff_t)(X)|0x1)) +#define ISDISK(X) ((u_int32_t)(ptrdiff_t)(X)&0x2) +#define DODISK(X) ((X) = (char *)((ptrdiff_t)(X)|0x2)) + +#define BITS_PER_MAP 32 + +/* Given the address of the beginning of a big map, clear/set the nth bit */ +#define CLRBIT(A, N) ((A)[(N)/BITS_PER_MAP] &= ~(1<<((N)%BITS_PER_MAP))) +#define SETBIT(A, N) ((A)[(N)/BITS_PER_MAP] |= (1<<((N)%BITS_PER_MAP))) +#define ISSET(A, N) ((A)[(N)/BITS_PER_MAP] & (1<<((N)%BITS_PER_MAP))) + +/* Overflow management */ +/* + * Overflow page numbers are allocated per split point. At each doubling of + * the table, we can allocate extra pages. So, an overflow page number has + * the top 5 bits indicate which split point and the lower 11 bits indicate + * which page at that split point is indicated (pages within split points are + * numberered starting with 1). + */ + +#define SPLITSHIFT 11 +#define SPLITMASK 0x7FF +#define SPLITNUM(N) (((u_int32_t)(N)) >> SPLITSHIFT) +#define OPAGENUM(N) ((N) & SPLITMASK) +#define OADDR_OF(S,O) ((u_int32_t)((u_int32_t)(S) << SPLITSHIFT) + (O)) + +#define BUCKET_TO_PAGE(B) \ + (B) + hashp->HDRPAGES + ((B) ? hashp->SPARES[__log2((B)+1)-1] : 0) +#define OADDR_TO_PAGE(B) \ + BUCKET_TO_PAGE ( (1 << SPLITNUM((B))) -1 ) + OPAGENUM((B)); + +/* + * page.h contains a detailed description of the page format. + * + * Normally, keys and data are accessed from offset tables in the top of + * each page which point to the beginning of the key and data. There are + * four flag values which may be stored in these offset tables which indicate + * the following: + * + * + * OVFLPAGE Rather than a key data pair, this pair contains + * the address of an overflow page. The format of + * the pair is: + * OVERFLOW_PAGE_NUMBER OVFLPAGE + * + * PARTIAL_KEY This must be the first key/data pair on a page + * and implies that page contains only a partial key. + * That is, the key is too big to fit on a single page + * so it starts on this page and continues on the next. + * The format of the page is: + * KEY_OFF PARTIAL_KEY OVFL_PAGENO OVFLPAGE + * + * KEY_OFF -- offset of the beginning of the key + * PARTIAL_KEY -- 1 + * OVFL_PAGENO - page number of the next overflow page + * OVFLPAGE -- 0 + * + * FULL_KEY This must be the first key/data pair on the page. It + * is used in two cases. + * + * Case 1: + * There is a complete key on the page but no data + * (because it wouldn't fit). The next page contains + * the data. + * + * Page format it: + * KEY_OFF FULL_KEY OVFL_PAGENO OVFL_PAGE + * + * KEY_OFF -- offset of the beginning of the key + * FULL_KEY -- 2 + * OVFL_PAGENO - page number of the next overflow page + * OVFLPAGE -- 0 + * + * Case 2: + * This page contains no key, but part of a large + * data field, which is continued on the next page. + * + * Page format it: + * DATA_OFF FULL_KEY OVFL_PAGENO OVFL_PAGE + * + * KEY_OFF -- offset of the beginning of the data on + * this page + * FULL_KEY -- 2 + * OVFL_PAGENO - page number of the next overflow page + * OVFLPAGE -- 0 + * + * FULL_KEY_DATA + * This must be the first key/data pair on the page. + * There are two cases: + * + * Case 1: + * This page contains a key and the beginning of the + * data field, but the data field is continued on the + * next page. + * + * Page format is: + * KEY_OFF FULL_KEY_DATA OVFL_PAGENO DATA_OFF + * + * KEY_OFF -- offset of the beginning of the key + * FULL_KEY_DATA -- 3 + * OVFL_PAGENO - page number of the next overflow page + * DATA_OFF -- offset of the beginning of the data + * + * Case 2: + * This page contains the last page of a big data pair. + * There is no key, only the tail end of the data + * on this page. + * + * Page format is: + * DATA_OFF FULL_KEY_DATA <OVFL_PAGENO> <OVFLPAGE> + * + * DATA_OFF -- offset of the beginning of the data on + * this page + * FULL_KEY_DATA -- 3 + * OVFL_PAGENO - page number of the next overflow page + * OVFLPAGE -- 0 + * + * OVFL_PAGENO and OVFLPAGE are optional (they are + * not present if there is no next page). + */ + +#define OVFLPAGE 0 +#define PARTIAL_KEY 1 +#define FULL_KEY 2 +#define FULL_KEY_DATA 3 +#define REAL_KEY 4 + +/* Short hands for accessing structure */ +#define BSIZE hdr.bsize +#define BSHIFT hdr.bshift +#define DSIZE hdr.dsize +#define SGSIZE hdr.ssize +#define SSHIFT hdr.sshift +#define LORDER hdr.lorder +#define OVFL_POINT hdr.ovfl_point +#define LAST_FREED hdr.last_freed +#define MAX_BUCKET hdr.max_bucket +#define FFACTOR hdr.ffactor +#define HIGH_MASK hdr.high_mask +#define LOW_MASK hdr.low_mask +#define NKEYS hdr.nkeys +#define HDRPAGES hdr.hdrpages +#define SPARES hdr.spares +#define BITMAPS hdr.bitmaps +#define VERSION hdr.version +#define MAGIC hdr.magic +#define NEXT_FREE hdr.next_free +#define H_CHARKEY hdr.h_charkey diff --git a/lib/libc/db/hash/hash_bigkey.c b/lib/libc/db/hash/hash_bigkey.c new file mode 100644 index 00000000000..993839d9f79 --- /dev/null +++ b/lib/libc/db/hash/hash_bigkey.c @@ -0,0 +1,673 @@ +/* $NetBSD: hash_bigkey.c,v 1.5 1995/02/27 13:22:16 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hash_bigkey.c 8.3 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: hash_bigkey.c,v 1.5 1995/02/27 13:22:16 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +/* + * PACKAGE: hash + * DESCRIPTION: + * Big key/data handling for the hashing package. + * + * ROUTINES: + * External + * __big_keydata + * __big_split + * __big_insert + * __big_return + * __big_delete + * __find_last_page + * Internal + * collect_key + * collect_data + */ + +#include <sys/param.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#ifdef DEBUG +#include <assert.h> +#endif + +#include <db.h> +#include "hash.h" +#include "page.h" +#include "extern.h" + +static int collect_key __P((HTAB *, BUFHEAD *, int, DBT *, int)); +static int collect_data __P((HTAB *, BUFHEAD *, int, int)); + +/* + * Big_insert + * + * You need to do an insert and the key/data pair is too big + * + * Returns: + * 0 ==> OK + *-1 ==> ERROR + */ +extern int +__big_insert(hashp, bufp, key, val) + HTAB *hashp; + BUFHEAD *bufp; + const DBT *key, *val; +{ + register u_int16_t *p; + int key_size, n, val_size; + u_int16_t space, move_bytes, off; + char *cp, *key_data, *val_data; + + cp = bufp->page; /* Character pointer of p. */ + p = (u_int16_t *)cp; + + key_data = (char *)key->data; + key_size = key->size; + val_data = (char *)val->data; + val_size = val->size; + + /* First move the Key */ + for (space = FREESPACE(p) - BIGOVERHEAD; key_size; + space = FREESPACE(p) - BIGOVERHEAD) { + move_bytes = MIN(space, key_size); + off = OFFSET(p) - move_bytes; + memmove(cp + off, key_data, move_bytes); + key_size -= move_bytes; + key_data += move_bytes; + n = p[0]; + p[++n] = off; + p[0] = ++n; + FREESPACE(p) = off - PAGE_META(n); + OFFSET(p) = off; + p[n] = PARTIAL_KEY; + bufp = __add_ovflpage(hashp, bufp); + if (!bufp) + return (-1); + n = p[0]; + if (!key_size) + if (FREESPACE(p)) { + move_bytes = MIN(FREESPACE(p), val_size); + off = OFFSET(p) - move_bytes; + p[n] = off; + memmove(cp + off, val_data, move_bytes); + val_data += move_bytes; + val_size -= move_bytes; + p[n - 2] = FULL_KEY_DATA; + FREESPACE(p) = FREESPACE(p) - move_bytes; + OFFSET(p) = off; + } else + p[n - 2] = FULL_KEY; + p = (u_int16_t *)bufp->page; + cp = bufp->page; + bufp->flags |= BUF_MOD; + } + + /* Now move the data */ + for (space = FREESPACE(p) - BIGOVERHEAD; val_size; + space = FREESPACE(p) - BIGOVERHEAD) { + move_bytes = MIN(space, val_size); + /* + * Here's the hack to make sure that if the data ends on the + * same page as the key ends, FREESPACE is at least one. + */ + if (space == val_size && val_size == val->size) + move_bytes--; + off = OFFSET(p) - move_bytes; + memmove(cp + off, val_data, move_bytes); + val_size -= move_bytes; + val_data += move_bytes; + n = p[0]; + p[++n] = off; + p[0] = ++n; + FREESPACE(p) = off - PAGE_META(n); + OFFSET(p) = off; + if (val_size) { + p[n] = FULL_KEY; + bufp = __add_ovflpage(hashp, bufp); + if (!bufp) + return (-1); + cp = bufp->page; + p = (u_int16_t *)cp; + } else + p[n] = FULL_KEY_DATA; + bufp->flags |= BUF_MOD; + } + return (0); +} + +/* + * Called when bufp's page contains a partial key (index should be 1) + * + * All pages in the big key/data pair except bufp are freed. We cannot + * free bufp because the page pointing to it is lost and we can't get rid + * of its pointer. + * + * Returns: + * 0 => OK + *-1 => ERROR + */ +extern int +__big_delete(hashp, bufp) + HTAB *hashp; + BUFHEAD *bufp; +{ + register BUFHEAD *last_bfp, *rbufp; + u_int16_t *bp, pageno; + int key_done, n; + + rbufp = bufp; + last_bfp = NULL; + bp = (u_int16_t *)bufp->page; + pageno = 0; + key_done = 0; + + while (!key_done || (bp[2] != FULL_KEY_DATA)) { + if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) + key_done = 1; + + /* + * If there is freespace left on a FULL_KEY_DATA page, then + * the data is short and fits entirely on this page, and this + * is the last page. + */ + if (bp[2] == FULL_KEY_DATA && FREESPACE(bp)) + break; + pageno = bp[bp[0] - 1]; + rbufp->flags |= BUF_MOD; + rbufp = __get_buf(hashp, pageno, rbufp, 0); + if (last_bfp) + __free_ovflpage(hashp, last_bfp); + last_bfp = rbufp; + if (!rbufp) + return (-1); /* Error. */ + bp = (u_int16_t *)rbufp->page; + } + + /* + * If we get here then rbufp points to the last page of the big + * key/data pair. Bufp points to the first one -- it should now be + * empty pointing to the next page after this pair. Can't free it + * because we don't have the page pointing to it. + */ + + /* This is information from the last page of the pair. */ + n = bp[0]; + pageno = bp[n - 1]; + + /* Now, bp is the first page of the pair. */ + bp = (u_int16_t *)bufp->page; + if (n > 2) { + /* There is an overflow page. */ + bp[1] = pageno; + bp[2] = OVFLPAGE; + bufp->ovfl = rbufp->ovfl; + } else + /* This is the last page. */ + bufp->ovfl = NULL; + n -= 2; + bp[0] = n; + FREESPACE(bp) = hashp->BSIZE - PAGE_META(n); + OFFSET(bp) = hashp->BSIZE - 1; + + bufp->flags |= BUF_MOD; + if (rbufp) + __free_ovflpage(hashp, rbufp); + if (last_bfp != rbufp) + __free_ovflpage(hashp, last_bfp); + + hashp->NKEYS--; + return (0); +} +/* + * Returns: + * 0 = key not found + * -1 = get next overflow page + * -2 means key not found and this is big key/data + * -3 error + */ +extern int +__find_bigpair(hashp, bufp, ndx, key, size) + HTAB *hashp; + BUFHEAD *bufp; + int ndx; + char *key; + int size; +{ + register u_int16_t *bp; + register char *p; + int ksize; + u_int16_t bytes; + char *kkey; + + bp = (u_int16_t *)bufp->page; + p = bufp->page; + ksize = size; + kkey = key; + + for (bytes = hashp->BSIZE - bp[ndx]; + bytes <= size && bp[ndx + 1] == PARTIAL_KEY; + bytes = hashp->BSIZE - bp[ndx]) { + if (memcmp(p + bp[ndx], kkey, bytes)) + return (-2); + kkey += bytes; + ksize -= bytes; + bufp = __get_buf(hashp, bp[ndx + 2], bufp, 0); + if (!bufp) + return (-3); + p = bufp->page; + bp = (u_int16_t *)p; + ndx = 1; + } + + if (bytes != ksize || memcmp(p + bp[ndx], kkey, bytes)) { +#ifdef HASH_STATISTICS + ++hash_collisions; +#endif + return (-2); + } else + return (ndx); +} + +/* + * Given the buffer pointer of the first overflow page of a big pair, + * find the end of the big pair + * + * This will set bpp to the buffer header of the last page of the big pair. + * It will return the pageno of the overflow page following the last page + * of the pair; 0 if there isn't any (i.e. big pair is the last key in the + * bucket) + */ +extern u_int16_t +__find_last_page(hashp, bpp) + HTAB *hashp; + BUFHEAD **bpp; +{ + BUFHEAD *bufp; + u_int16_t *bp, pageno; + int n; + + bufp = *bpp; + bp = (u_int16_t *)bufp->page; + for (;;) { + n = bp[0]; + + /* + * This is the last page if: the tag is FULL_KEY_DATA and + * either only 2 entries OVFLPAGE marker is explicit there + * is freespace on the page. + */ + if (bp[2] == FULL_KEY_DATA && + ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp)))) + break; + + pageno = bp[n - 1]; + bufp = __get_buf(hashp, pageno, bufp, 0); + if (!bufp) + return (0); /* Need to indicate an error! */ + bp = (u_int16_t *)bufp->page; + } + + *bpp = bufp; + if (bp[0] > 2) + return (bp[3]); + else + return (0); +} + +/* + * Return the data for the key/data pair that begins on this page at this + * index (index should always be 1). + */ +extern int +__big_return(hashp, bufp, ndx, val, set_current) + HTAB *hashp; + BUFHEAD *bufp; + int ndx; + DBT *val; + int set_current; +{ + BUFHEAD *save_p; + u_int16_t *bp, len, off, save_addr; + char *tp; + + bp = (u_int16_t *)bufp->page; + while (bp[ndx + 1] == PARTIAL_KEY) { + bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!bufp) + return (-1); + bp = (u_int16_t *)bufp->page; + ndx = 1; + } + + if (bp[ndx + 1] == FULL_KEY) { + bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!bufp) + return (-1); + bp = (u_int16_t *)bufp->page; + save_p = bufp; + save_addr = save_p->addr; + off = bp[1]; + len = 0; + } else + if (!FREESPACE(bp)) { + /* + * This is a hack. We can't distinguish between + * FULL_KEY_DATA that contains complete data or + * incomplete data, so we require that if the data + * is complete, there is at least 1 byte of free + * space left. + */ + off = bp[bp[0]]; + len = bp[1] - off; + save_p = bufp; + save_addr = bufp->addr; + bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!bufp) + return (-1); + bp = (u_int16_t *)bufp->page; + } else { + /* The data is all on one page. */ + tp = (char *)bp; + off = bp[bp[0]]; + val->data = (u_char *)tp + off; + val->size = bp[1] - off; + if (set_current) { + if (bp[0] == 2) { /* No more buckets in + * chain */ + hashp->cpage = NULL; + hashp->cbucket++; + hashp->cndx = 1; + } else { + hashp->cpage = __get_buf(hashp, + bp[bp[0] - 1], bufp, 0); + if (!hashp->cpage) + return (-1); + hashp->cndx = 1; + if (!((u_int16_t *) + hashp->cpage->page)[0]) { + hashp->cbucket++; + hashp->cpage = NULL; + } + } + } + return (0); + } + + val->size = collect_data(hashp, bufp, (int)len, set_current); + if (val->size == -1) + return (-1); + if (save_p->addr != save_addr) { + /* We are pretty short on buffers. */ + errno = EINVAL; /* OUT OF BUFFERS */ + return (-1); + } + memmove(hashp->tmp_buf, (save_p->page) + off, len); + val->data = (u_char *)hashp->tmp_buf; + return (0); +} +/* + * Count how big the total datasize is by recursing through the pages. Then + * allocate a buffer and copy the data as you recurse up. + */ +static int +collect_data(hashp, bufp, len, set) + HTAB *hashp; + BUFHEAD *bufp; + int len, set; +{ + register u_int16_t *bp; + register char *p; + BUFHEAD *xbp; + u_int16_t save_addr; + int mylen, totlen; + + p = bufp->page; + bp = (u_int16_t *)p; + mylen = hashp->BSIZE - bp[1]; + save_addr = bufp->addr; + + if (bp[2] == FULL_KEY_DATA) { /* End of Data */ + totlen = len + mylen; + if (hashp->tmp_buf) + free(hashp->tmp_buf); + if ((hashp->tmp_buf = (char *)malloc(totlen)) == NULL) + return (-1); + if (set) { + hashp->cndx = 1; + if (bp[0] == 2) { /* No more buckets in chain */ + hashp->cpage = NULL; + hashp->cbucket++; + } else { + hashp->cpage = + __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!hashp->cpage) + return (-1); + else if (!((u_int16_t *)hashp->cpage->page)[0]) { + hashp->cbucket++; + hashp->cpage = NULL; + } + } + } + } else { + xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!xbp || ((totlen = + collect_data(hashp, xbp, len + mylen, set)) < 1)) + return (-1); + } + if (bufp->addr != save_addr) { + errno = EINVAL; /* Out of buffers. */ + return (-1); + } + memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], mylen); + return (totlen); +} + +/* + * Fill in the key and data for this big pair. + */ +extern int +__big_keydata(hashp, bufp, key, val, set) + HTAB *hashp; + BUFHEAD *bufp; + DBT *key, *val; + int set; +{ + key->size = collect_key(hashp, bufp, 0, val, set); + if (key->size == -1) + return (-1); + key->data = (u_char *)hashp->tmp_key; + return (0); +} + +/* + * Count how big the total key size is by recursing through the pages. Then + * collect the data, allocate a buffer and copy the key as you recurse up. + */ +static int +collect_key(hashp, bufp, len, val, set) + HTAB *hashp; + BUFHEAD *bufp; + int len; + DBT *val; + int set; +{ + BUFHEAD *xbp; + char *p; + int mylen, totlen; + u_int16_t *bp, save_addr; + + p = bufp->page; + bp = (u_int16_t *)p; + mylen = hashp->BSIZE - bp[1]; + + save_addr = bufp->addr; + totlen = len + mylen; + if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) { /* End of Key. */ + if (hashp->tmp_key != NULL) + free(hashp->tmp_key); + if ((hashp->tmp_key = (char *)malloc(totlen)) == NULL) + return (-1); + if (__big_return(hashp, bufp, 1, val, set)) + return (-1); + } else { + xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!xbp || ((totlen = + collect_key(hashp, xbp, totlen, val, set)) < 1)) + return (-1); + } + if (bufp->addr != save_addr) { + errno = EINVAL; /* MIS -- OUT OF BUFFERS */ + return (-1); + } + memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], mylen); + return (totlen); +} + +/* + * Returns: + * 0 => OK + * -1 => error + */ +extern int +__big_split(hashp, op, np, big_keyp, addr, obucket, ret) + HTAB *hashp; + BUFHEAD *op; /* Pointer to where to put keys that go in old bucket */ + BUFHEAD *np; /* Pointer to new bucket page */ + /* Pointer to first page containing the big key/data */ + BUFHEAD *big_keyp; + int addr; /* Address of big_keyp */ + u_int32_t obucket;/* Old Bucket */ + SPLIT_RETURN *ret; +{ + register BUFHEAD *tmpp; + register u_int16_t *tp; + BUFHEAD *bp; + DBT key, val; + u_int32_t change; + u_int16_t free_space, n, off; + + bp = big_keyp; + + /* Now figure out where the big key/data goes */ + if (__big_keydata(hashp, big_keyp, &key, &val, 0)) + return (-1); + change = (__call_hash(hashp, key.data, key.size) != obucket); + + if (ret->next_addr = __find_last_page(hashp, &big_keyp)) { + if (!(ret->nextp = + __get_buf(hashp, ret->next_addr, big_keyp, 0))) + return (-1);; + } else + ret->nextp = NULL; + + /* Now make one of np/op point to the big key/data pair */ +#ifdef DEBUG + assert(np->ovfl == NULL); +#endif + if (change) + tmpp = np; + else + tmpp = op; + + tmpp->flags |= BUF_MOD; +#ifdef DEBUG1 + (void)fprintf(stderr, + "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr, + (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0)); +#endif + tmpp->ovfl = bp; /* one of op/np point to big_keyp */ + tp = (u_int16_t *)tmpp->page; +#ifdef DEBUG + assert(FREESPACE(tp) >= OVFLSIZE); +#endif + n = tp[0]; + off = OFFSET(tp); + free_space = FREESPACE(tp); + tp[++n] = (u_int16_t)addr; + tp[++n] = OVFLPAGE; + tp[0] = n; + OFFSET(tp) = off; + FREESPACE(tp) = free_space - OVFLSIZE; + + /* + * Finally, set the new and old return values. BIG_KEYP contains a + * pointer to the last page of the big key_data pair. Make sure that + * big_keyp has no following page (2 elements) or create an empty + * following page. + */ + + ret->newp = np; + ret->oldp = op; + + tp = (u_int16_t *)big_keyp->page; + big_keyp->flags |= BUF_MOD; + if (tp[0] > 2) { + /* + * There may be either one or two offsets on this page. If + * there is one, then the overflow page is linked on normally + * and tp[4] is OVFLPAGE. If there are two, tp[4] contains + * the second offset and needs to get stuffed in after the + * next overflow page is added. + */ + n = tp[4]; + free_space = FREESPACE(tp); + off = OFFSET(tp); + tp[0] -= 2; + FREESPACE(tp) = free_space + OVFLSIZE; + OFFSET(tp) = off; + tmpp = __add_ovflpage(hashp, big_keyp); + if (!tmpp) + return (-1); + tp[4] = n; + } else + tmpp = big_keyp; + + if (change) + ret->newp = tmpp; + else + ret->oldp = tmpp; + return (0); +} diff --git a/lib/libc/db/hash/hash_buf.c b/lib/libc/db/hash/hash_buf.c new file mode 100644 index 00000000000..7747331688a --- /dev/null +++ b/lib/libc/db/hash/hash_buf.c @@ -0,0 +1,355 @@ +/* $NetBSD: hash_buf.c,v 1.5 1995/02/27 13:22:23 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hash_buf.c 8.4 (Berkeley) 6/4/94"; +#else +static char rcsid[] = "$NetBSD: hash_buf.c,v 1.5 1995/02/27 13:22:23 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +/* + * PACKAGE: hash + * + * DESCRIPTION: + * Contains buffer management + * + * ROUTINES: + * External + * __buf_init + * __get_buf + * __buf_free + * __reclaim_buf + * Internal + * newbuf + */ + +#include <sys/param.h> + +#include <errno.h> +#include <stddef.h> +#include <stdio.h> +#include <stdlib.h> + +#ifdef DEBUG +#include <assert.h> +#endif + +#include <db.h> +#include "hash.h" +#include "page.h" +#include "extern.h" + +static BUFHEAD *newbuf __P((HTAB *, u_int32_t, BUFHEAD *)); + +/* Unlink B from its place in the lru */ +#define BUF_REMOVE(B) { \ + (B)->prev->next = (B)->next; \ + (B)->next->prev = (B)->prev; \ +} + +/* Insert B after P */ +#define BUF_INSERT(B, P) { \ + (B)->next = (P)->next; \ + (B)->prev = (P); \ + (P)->next = (B); \ + (B)->next->prev = (B); \ +} + +#define MRU hashp->bufhead.next +#define LRU hashp->bufhead.prev + +#define MRU_INSERT(B) BUF_INSERT((B), &hashp->bufhead) +#define LRU_INSERT(B) BUF_INSERT((B), LRU) + +/* + * We are looking for a buffer with address "addr". If prev_bp is NULL, then + * address is a bucket index. If prev_bp is not NULL, then it points to the + * page previous to an overflow page that we are trying to find. + * + * CAVEAT: The buffer header accessed via prev_bp's ovfl field may no longer + * be valid. Therefore, you must always verify that its address matches the + * address you are seeking. + */ +extern BUFHEAD * +__get_buf(hashp, addr, prev_bp, newpage) + HTAB *hashp; + u_int32_t addr; + BUFHEAD *prev_bp; + int newpage; /* If prev_bp set, indicates a new overflow page. */ +{ + register BUFHEAD *bp; + register u_int32_t is_disk_mask; + register int is_disk, segment_ndx; + SEGMENT segp; + + is_disk = 0; + is_disk_mask = 0; + if (prev_bp) { + bp = prev_bp->ovfl; + if (!bp || (bp->addr != addr)) + bp = NULL; + if (!newpage) + is_disk = BUF_DISK; + } else { + /* Grab buffer out of directory */ + segment_ndx = addr & (hashp->SGSIZE - 1); + + /* valid segment ensured by __call_hash() */ + segp = hashp->dir[addr >> hashp->SSHIFT]; +#ifdef DEBUG + assert(segp != NULL); +#endif + bp = PTROF(segp[segment_ndx]); + is_disk_mask = ISDISK(segp[segment_ndx]); + is_disk = is_disk_mask || !hashp->new_file; + } + + if (!bp) { + bp = newbuf(hashp, addr, prev_bp); + if (!bp || + __get_page(hashp, bp->page, addr, !prev_bp, is_disk, 0)) + return (NULL); + if (!prev_bp) + segp[segment_ndx] = + (BUFHEAD *)((ptrdiff_t)bp | is_disk_mask); + } else { + BUF_REMOVE(bp); + MRU_INSERT(bp); + } + return (bp); +} + +/* + * We need a buffer for this page. Either allocate one, or evict a resident + * one (if we have as many buffers as we're allowed) and put this one in. + * + * If newbuf finds an error (returning NULL), it also sets errno. + */ +static BUFHEAD * +newbuf(hashp, addr, prev_bp) + HTAB *hashp; + u_int32_t addr; + BUFHEAD *prev_bp; +{ + register BUFHEAD *bp; /* The buffer we're going to use */ + register BUFHEAD *xbp; /* Temp pointer */ + register BUFHEAD *next_xbp; + SEGMENT segp; + int segment_ndx; + u_int16_t oaddr, *shortp; + + oaddr = 0; + bp = LRU; + /* + * If LRU buffer is pinned, the buffer pool is too small. We need to + * allocate more buffers. + */ + if (hashp->nbufs || (bp->flags & BUF_PIN)) { + /* Allocate a new one */ + if ((bp = (BUFHEAD *)malloc(sizeof(BUFHEAD))) == NULL) + return (NULL); + if ((bp->page = (char *)malloc(hashp->BSIZE)) == NULL) { + free(bp); + return (NULL); + } + if (hashp->nbufs) + hashp->nbufs--; + } else { + /* Kick someone out */ + BUF_REMOVE(bp); + /* + * If this is an overflow page with addr 0, it's already been + * flushed back in an overflow chain and initialized. + */ + if ((bp->addr != 0) || (bp->flags & BUF_BUCKET)) { + /* + * Set oaddr before __put_page so that you get it + * before bytes are swapped. + */ + shortp = (u_int16_t *)bp->page; + if (shortp[0]) + oaddr = shortp[shortp[0] - 1]; + if ((bp->flags & BUF_MOD) && __put_page(hashp, bp->page, + bp->addr, (int)IS_BUCKET(bp->flags), 0)) + return (NULL); + /* + * Update the pointer to this page (i.e. invalidate it). + * + * If this is a new file (i.e. we created it at open + * time), make sure that we mark pages which have been + * written to disk so we retrieve them from disk later, + * rather than allocating new pages. + */ + if (IS_BUCKET(bp->flags)) { + segment_ndx = bp->addr & (hashp->SGSIZE - 1); + segp = hashp->dir[bp->addr >> hashp->SSHIFT]; +#ifdef DEBUG + assert(segp != NULL); +#endif + + if (hashp->new_file && + ((bp->flags & BUF_MOD) || + ISDISK(segp[segment_ndx]))) + segp[segment_ndx] = (BUFHEAD *)BUF_DISK; + else + segp[segment_ndx] = NULL; + } + /* + * Since overflow pages can only be access by means of + * their bucket, free overflow pages associated with + * this bucket. + */ + for (xbp = bp; xbp->ovfl;) { + next_xbp = xbp->ovfl; + xbp->ovfl = 0; + xbp = next_xbp; + + /* Check that ovfl pointer is up date. */ + if (IS_BUCKET(xbp->flags) || + (oaddr != xbp->addr)) + break; + + shortp = (u_int16_t *)xbp->page; + if (shortp[0]) + /* set before __put_page */ + oaddr = shortp[shortp[0] - 1]; + if ((xbp->flags & BUF_MOD) && __put_page(hashp, + xbp->page, xbp->addr, 0, 0)) + return (NULL); + xbp->addr = 0; + xbp->flags = 0; + BUF_REMOVE(xbp); + LRU_INSERT(xbp); + } + } + } + + /* Now assign this buffer */ + bp->addr = addr; +#ifdef DEBUG1 + (void)fprintf(stderr, "NEWBUF1: %d->ovfl was %d is now %d\n", + bp->addr, (bp->ovfl ? bp->ovfl->addr : 0), 0); +#endif + bp->ovfl = NULL; + if (prev_bp) { + /* + * If prev_bp is set, this is an overflow page, hook it in to + * the buffer overflow links. + */ +#ifdef DEBUG1 + (void)fprintf(stderr, "NEWBUF2: %d->ovfl was %d is now %d\n", + prev_bp->addr, (prev_bp->ovfl ? bp->ovfl->addr : 0), + (bp ? bp->addr : 0)); +#endif + prev_bp->ovfl = bp; + bp->flags = 0; + } else + bp->flags = BUF_BUCKET; + MRU_INSERT(bp); + return (bp); +} + +extern void +__buf_init(hashp, nbytes) + HTAB *hashp; + int nbytes; +{ + BUFHEAD *bfp; + int npages; + + bfp = &(hashp->bufhead); + npages = (nbytes + hashp->BSIZE - 1) >> hashp->BSHIFT; + npages = MAX(npages, MIN_BUFFERS); + + hashp->nbufs = npages; + bfp->next = bfp; + bfp->prev = bfp; + /* + * This space is calloc'd so these are already null. + * + * bfp->ovfl = NULL; + * bfp->flags = 0; + * bfp->page = NULL; + * bfp->addr = 0; + */ +} + +extern int +__buf_free(hashp, do_free, to_disk) + HTAB *hashp; + int do_free, to_disk; +{ + BUFHEAD *bp; + + /* Need to make sure that buffer manager has been initialized */ + if (!LRU) + return (0); + for (bp = LRU; bp != &hashp->bufhead;) { + /* Check that the buffer is valid */ + if (bp->addr || IS_BUCKET(bp->flags)) { + if (to_disk && (bp->flags & BUF_MOD) && + __put_page(hashp, bp->page, + bp->addr, IS_BUCKET(bp->flags), 0)) + return (-1); + } + /* Check if we are freeing stuff */ + if (do_free) { + if (bp->page) + free(bp->page); + BUF_REMOVE(bp); + free(bp); + bp = LRU; + } else + bp = bp->prev; + } + return (0); +} + +extern void +__reclaim_buf(hashp, bp) + HTAB *hashp; + BUFHEAD *bp; +{ + bp->ovfl = 0; + bp->addr = 0; + bp->flags = 0; + BUF_REMOVE(bp); + LRU_INSERT(bp); +} diff --git a/lib/libc/db/hash/hash_func.c b/lib/libc/db/hash/hash_func.c new file mode 100644 index 00000000000..a978b99d2fa --- /dev/null +++ b/lib/libc/db/hash/hash_func.c @@ -0,0 +1,218 @@ +/* $NetBSD: hash_func.c,v 1.5 1995/02/27 13:22:27 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hash_func.c 8.2 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: hash_func.c,v 1.5 1995/02/27 13:22:27 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <db.h> +#include "hash.h" +#include "page.h" +#include "extern.h" + +static u_int32_t hash1 __P((const void *, size_t)); +static u_int32_t hash2 __P((const void *, size_t)); +static u_int32_t hash3 __P((const void *, size_t)); +static u_int32_t hash4 __P((const void *, size_t)); + +/* Global default hash function */ +u_int32_t (*__default_hash) __P((const void *, size_t)) = hash4; + +/* + * HASH FUNCTIONS + * + * Assume that we've already split the bucket to which this key hashes, + * calculate that bucket, and check that in fact we did already split it. + * + * This came from ejb's hsearch. + */ + +#define PRIME1 37 +#define PRIME2 1048583 + +static u_int32_t +hash1(keyarg, len) + const void *keyarg; + register size_t len; +{ + register const u_char *key; + register u_int32_t h; + + /* Convert string to integer */ + for (key = keyarg, h = 0; len--;) + h = h * PRIME1 ^ (*key++ - ' '); + h %= PRIME2; + return (h); +} + +/* + * Phong's linear congruential hash + */ +#define dcharhash(h, c) ((h) = 0x63c63cd9*(h) + 0x9c39c33d + (c)) + +static u_int32_t +hash2(keyarg, len) + const void *keyarg; + size_t len; +{ + register const u_char *e, *key; + register u_int32_t h; + register u_char c; + + key = keyarg; + e = key + len; + for (h = 0; key != e;) { + c = *key++; + if (!c && key > e) + break; + dcharhash(h, c); + } + return (h); +} + +/* + * This is INCREDIBLY ugly, but fast. We break the string up into 8 byte + * units. On the first time through the loop we get the "leftover bytes" + * (strlen % 8). On every other iteration, we perform 8 HASHC's so we handle + * all 8 bytes. Essentially, this saves us 7 cmp & branch instructions. If + * this routine is heavily used enough, it's worth the ugly coding. + * + * OZ's original sdbm hash + */ +static u_int32_t +hash3(keyarg, len) + const void *keyarg; + register size_t len; +{ + register const u_char *key; + register size_t loop; + register u_int32_t h; + +#define HASHC h = *key++ + 65599 * h + + h = 0; + key = keyarg; + if (len > 0) { + loop = (len + 8 - 1) >> 3; + + switch (len & (8 - 1)) { + case 0: + do { + HASHC; + /* FALLTHROUGH */ + case 7: + HASHC; + /* FALLTHROUGH */ + case 6: + HASHC; + /* FALLTHROUGH */ + case 5: + HASHC; + /* FALLTHROUGH */ + case 4: + HASHC; + /* FALLTHROUGH */ + case 3: + HASHC; + /* FALLTHROUGH */ + case 2: + HASHC; + /* FALLTHROUGH */ + case 1: + HASHC; + } while (--loop); + } + } + return (h); +} + +/* Hash function from Chris Torek. */ +static u_int32_t +hash4(keyarg, len) + const void *keyarg; + register size_t len; +{ + register const u_char *key; + register size_t loop; + register u_int32_t h; + +#define HASH4a h = (h << 5) - h + *key++; +#define HASH4b h = (h << 5) + h + *key++; +#define HASH4 HASH4b + + h = 0; + key = keyarg; + if (len > 0) { + loop = (len + 8 - 1) >> 3; + + switch (len & (8 - 1)) { + case 0: + do { + HASH4; + /* FALLTHROUGH */ + case 7: + HASH4; + /* FALLTHROUGH */ + case 6: + HASH4; + /* FALLTHROUGH */ + case 5: + HASH4; + /* FALLTHROUGH */ + case 4: + HASH4; + /* FALLTHROUGH */ + case 3: + HASH4; + /* FALLTHROUGH */ + case 2: + HASH4; + /* FALLTHROUGH */ + case 1: + HASH4; + } while (--loop); + } + } + return (h); +} diff --git a/lib/libc/db/hash/hash_log2.c b/lib/libc/db/hash/hash_log2.c new file mode 100644 index 00000000000..9332cbac887 --- /dev/null +++ b/lib/libc/db/hash/hash_log2.c @@ -0,0 +1,60 @@ +/* $NetBSD: hash_log2.c,v 1.5 1995/02/27 13:22:30 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hash_log2.c 8.2 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: hash_log2.c,v 1.5 1995/02/27 13:22:30 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <db.h> + +u_int32_t +__log2(num) + u_int32_t num; +{ + register u_int32_t i, limit; + + limit = 1; + for (i = 0; limit < num; limit = limit << 1, i++); + return (i); +} diff --git a/lib/libc/db/hash/hash_page.c b/lib/libc/db/hash/hash_page.c new file mode 100644 index 00000000000..49ee8904a14 --- /dev/null +++ b/lib/libc/db/hash/hash_page.c @@ -0,0 +1,950 @@ +/* $NetBSD: hash_page.c,v 1.7 1995/02/27 13:22:34 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hash_page.c 8.6 (Berkeley) 6/16/94"; +#else +static char rcsid[] = "$NetBSD: hash_page.c,v 1.7 1995/02/27 13:22:34 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +/* + * PACKAGE: hashing + * + * DESCRIPTION: + * Page manipulation for hashing package. + * + * ROUTINES: + * + * External + * __get_page + * __add_ovflpage + * Internal + * overflow_page + * open_temp + */ + +#include <sys/types.h> + +#include <errno.h> +#include <fcntl.h> +#include <signal.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> +#ifdef DEBUG +#include <assert.h> +#endif + +#include <db.h> +#include "hash.h" +#include "page.h" +#include "extern.h" + +static u_int32_t *fetch_bitmap __P((HTAB *, int)); +static u_int32_t first_free __P((u_int32_t)); +static int open_temp __P((HTAB *)); +static u_int16_t overflow_page __P((HTAB *)); +static void putpair __P((char *, const DBT *, const DBT *)); +static void squeeze_key __P((u_int16_t *, const DBT *, const DBT *)); +static int ugly_split + __P((HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int)); + +#define PAGE_INIT(P) { \ + ((u_int16_t *)(P))[0] = 0; \ + ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \ + ((u_int16_t *)(P))[2] = hashp->BSIZE; \ +} + +/* + * This is called AFTER we have verified that there is room on the page for + * the pair (PAIRFITS has returned true) so we go right ahead and start moving + * stuff on. + */ +static void +putpair(p, key, val) + char *p; + const DBT *key, *val; +{ + register u_int16_t *bp, n, off; + + bp = (u_int16_t *)p; + + /* Enter the key first. */ + n = bp[0]; + + off = OFFSET(bp) - key->size; + memmove(p + off, key->data, key->size); + bp[++n] = off; + + /* Now the data. */ + off -= val->size; + memmove(p + off, val->data, val->size); + bp[++n] = off; + + /* Adjust page info. */ + bp[0] = n; + bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t)); + bp[n + 2] = off; +} + +/* + * Returns: + * 0 OK + * -1 error + */ +extern int +__delpair(hashp, bufp, ndx) + HTAB *hashp; + BUFHEAD *bufp; + register int ndx; +{ + register u_int16_t *bp, newoff; + register int n; + u_int16_t pairlen; + + bp = (u_int16_t *)bufp->page; + n = bp[0]; + + if (bp[ndx + 1] < REAL_KEY) + return (__big_delete(hashp, bufp)); + if (ndx != 1) + newoff = bp[ndx - 1]; + else + newoff = hashp->BSIZE; + pairlen = newoff - bp[ndx + 1]; + + if (ndx != (n - 1)) { + /* Hard Case -- need to shuffle keys */ + register int i; + register char *src = bufp->page + (int)OFFSET(bp); + register char *dst = src + (int)pairlen; + memmove(dst, src, bp[ndx + 1] - OFFSET(bp)); + + /* Now adjust the pointers */ + for (i = ndx + 2; i <= n; i += 2) { + if (bp[i + 1] == OVFLPAGE) { + bp[i - 2] = bp[i]; + bp[i - 1] = bp[i + 1]; + } else { + bp[i - 2] = bp[i] + pairlen; + bp[i - 1] = bp[i + 1] + pairlen; + } + } + } + /* Finally adjust the page data */ + bp[n] = OFFSET(bp) + pairlen; + bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t); + bp[0] = n - 2; + hashp->NKEYS--; + + bufp->flags |= BUF_MOD; + return (0); +} +/* + * Returns: + * 0 ==> OK + * -1 ==> Error + */ +extern int +__split_page(hashp, obucket, nbucket) + HTAB *hashp; + u_int32_t obucket, nbucket; +{ + register BUFHEAD *new_bufp, *old_bufp; + register u_int16_t *ino; + register char *np; + DBT key, val; + int n, ndx, retval; + u_int16_t copyto, diff, off, moved; + char *op; + + copyto = (u_int16_t)hashp->BSIZE; + off = (u_int16_t)hashp->BSIZE; + old_bufp = __get_buf(hashp, obucket, NULL, 0); + if (old_bufp == NULL) + return (-1); + new_bufp = __get_buf(hashp, nbucket, NULL, 0); + if (new_bufp == NULL) + return (-1); + + old_bufp->flags |= (BUF_MOD | BUF_PIN); + new_bufp->flags |= (BUF_MOD | BUF_PIN); + + ino = (u_int16_t *)(op = old_bufp->page); + np = new_bufp->page; + + moved = 0; + + for (n = 1, ndx = 1; n < ino[0]; n += 2) { + if (ino[n + 1] < REAL_KEY) { + retval = ugly_split(hashp, obucket, old_bufp, new_bufp, + (int)copyto, (int)moved); + old_bufp->flags &= ~BUF_PIN; + new_bufp->flags &= ~BUF_PIN; + return (retval); + + } + key.data = (u_char *)op + ino[n]; + key.size = off - ino[n]; + + if (__call_hash(hashp, key.data, key.size) == obucket) { + /* Don't switch page */ + diff = copyto - off; + if (diff) { + copyto = ino[n + 1] + diff; + memmove(op + copyto, op + ino[n + 1], + off - ino[n + 1]); + ino[ndx] = copyto + ino[n] - ino[n + 1]; + ino[ndx + 1] = copyto; + } else + copyto = ino[n + 1]; + ndx += 2; + } else { + /* Switch page */ + val.data = (u_char *)op + ino[n + 1]; + val.size = ino[n] - ino[n + 1]; + putpair(np, &key, &val); + moved += 2; + } + + off = ino[n + 1]; + } + + /* Now clean up the page */ + ino[0] -= moved; + FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3); + OFFSET(ino) = copyto; + +#ifdef DEBUG3 + (void)fprintf(stderr, "split %d/%d\n", + ((u_int16_t *)np)[0] / 2, + ((u_int16_t *)op)[0] / 2); +#endif + /* unpin both pages */ + old_bufp->flags &= ~BUF_PIN; + new_bufp->flags &= ~BUF_PIN; + return (0); +} + +/* + * Called when we encounter an overflow or big key/data page during split + * handling. This is special cased since we have to begin checking whether + * the key/data pairs fit on their respective pages and because we may need + * overflow pages for both the old and new pages. + * + * The first page might be a page with regular key/data pairs in which case + * we have a regular overflow condition and just need to go on to the next + * page or it might be a big key/data pair in which case we need to fix the + * big key/data pair. + * + * Returns: + * 0 ==> success + * -1 ==> failure + */ +static int +ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved) + HTAB *hashp; + u_int32_t obucket; /* Same as __split_page. */ + BUFHEAD *old_bufp, *new_bufp; + int copyto; /* First byte on page which contains key/data values. */ + int moved; /* Number of pairs moved to new page. */ +{ + register BUFHEAD *bufp; /* Buffer header for ino */ + register u_int16_t *ino; /* Page keys come off of */ + register u_int16_t *np; /* New page */ + register u_int16_t *op; /* Page keys go on to if they aren't moving */ + + BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */ + DBT key, val; + SPLIT_RETURN ret; + u_int16_t n, off, ov_addr, scopyto; + char *cino; /* Character value of ino */ + + bufp = old_bufp; + ino = (u_int16_t *)old_bufp->page; + np = (u_int16_t *)new_bufp->page; + op = (u_int16_t *)old_bufp->page; + last_bfp = NULL; + scopyto = (u_int16_t)copyto; /* ANSI */ + + n = ino[0] - 1; + while (n < ino[0]) { + if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) { + if (__big_split(hashp, old_bufp, + new_bufp, bufp, bufp->addr, obucket, &ret)) + return (-1); + old_bufp = ret.oldp; + if (!old_bufp) + return (-1); + op = (u_int16_t *)old_bufp->page; + new_bufp = ret.newp; + if (!new_bufp) + return (-1); + np = (u_int16_t *)new_bufp->page; + bufp = ret.nextp; + if (!bufp) + return (0); + cino = (char *)bufp->page; + ino = (u_int16_t *)cino; + last_bfp = ret.nextp; + } else if (ino[n + 1] == OVFLPAGE) { + ov_addr = ino[n]; + /* + * Fix up the old page -- the extra 2 are the fields + * which contained the overflow information. + */ + ino[0] -= (moved + 2); + FREESPACE(ino) = + scopyto - sizeof(u_int16_t) * (ino[0] + 3); + OFFSET(ino) = scopyto; + + bufp = __get_buf(hashp, ov_addr, bufp, 0); + if (!bufp) + return (-1); + + ino = (u_int16_t *)bufp->page; + n = 1; + scopyto = hashp->BSIZE; + moved = 0; + + if (last_bfp) + __free_ovflpage(hashp, last_bfp); + last_bfp = bufp; + } + /* Move regular sized pairs of there are any */ + off = hashp->BSIZE; + for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) { + cino = (char *)ino; + key.data = (u_char *)cino + ino[n]; + key.size = off - ino[n]; + val.data = (u_char *)cino + ino[n + 1]; + val.size = ino[n] - ino[n + 1]; + off = ino[n + 1]; + + if (__call_hash(hashp, key.data, key.size) == obucket) { + /* Keep on old page */ + if (PAIRFITS(op, (&key), (&val))) + putpair((char *)op, &key, &val); + else { + old_bufp = + __add_ovflpage(hashp, old_bufp); + if (!old_bufp) + return (-1); + op = (u_int16_t *)old_bufp->page; + putpair((char *)op, &key, &val); + } + old_bufp->flags |= BUF_MOD; + } else { + /* Move to new page */ + if (PAIRFITS(np, (&key), (&val))) + putpair((char *)np, &key, &val); + else { + new_bufp = + __add_ovflpage(hashp, new_bufp); + if (!new_bufp) + return (-1); + np = (u_int16_t *)new_bufp->page; + putpair((char *)np, &key, &val); + } + new_bufp->flags |= BUF_MOD; + } + } + } + if (last_bfp) + __free_ovflpage(hashp, last_bfp); + return (0); +} + +/* + * Add the given pair to the page + * + * Returns: + * 0 ==> OK + * 1 ==> failure + */ +extern int +__addel(hashp, bufp, key, val) + HTAB *hashp; + BUFHEAD *bufp; + const DBT *key, *val; +{ + register u_int16_t *bp, *sop; + int do_expand; + + bp = (u_int16_t *)bufp->page; + do_expand = 0; + while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY)) + /* Exception case */ + if (bp[2] == FULL_KEY_DATA && bp[0] == 2) + /* This is the last page of a big key/data pair + and we need to add another page */ + break; + else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) { + bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!bufp) + return (-1); + bp = (u_int16_t *)bufp->page; + } else + /* Try to squeeze key on this page */ + if (FREESPACE(bp) > PAIRSIZE(key, val)) { + squeeze_key(bp, key, val); + return (0); + } else { + bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0); + if (!bufp) + return (-1); + bp = (u_int16_t *)bufp->page; + } + + if (PAIRFITS(bp, key, val)) + putpair(bufp->page, key, val); + else { + do_expand = 1; + bufp = __add_ovflpage(hashp, bufp); + if (!bufp) + return (-1); + sop = (u_int16_t *)bufp->page; + + if (PAIRFITS(sop, key, val)) + putpair((char *)sop, key, val); + else + if (__big_insert(hashp, bufp, key, val)) + return (-1); + } + bufp->flags |= BUF_MOD; + /* + * If the average number of keys per bucket exceeds the fill factor, + * expand the table. + */ + hashp->NKEYS++; + if (do_expand || + (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR)) + return (__expand_table(hashp)); + return (0); +} + +/* + * + * Returns: + * pointer on success + * NULL on error + */ +extern BUFHEAD * +__add_ovflpage(hashp, bufp) + HTAB *hashp; + BUFHEAD *bufp; +{ + register u_int16_t *sp; + u_int16_t ndx, ovfl_num; +#ifdef DEBUG1 + int tmp1, tmp2; +#endif + sp = (u_int16_t *)bufp->page; + + /* Check if we are dynamically determining the fill factor */ + if (hashp->FFACTOR == DEF_FFACTOR) { + hashp->FFACTOR = sp[0] >> 1; + if (hashp->FFACTOR < MIN_FFACTOR) + hashp->FFACTOR = MIN_FFACTOR; + } + bufp->flags |= BUF_MOD; + ovfl_num = overflow_page(hashp); +#ifdef DEBUG1 + tmp1 = bufp->addr; + tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0; +#endif + if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1))) + return (NULL); + bufp->ovfl->flags |= BUF_MOD; +#ifdef DEBUG1 + (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n", + tmp1, tmp2, bufp->ovfl->addr); +#endif + ndx = sp[0]; + /* + * Since a pair is allocated on a page only if there's room to add + * an overflow page, we know that the OVFL information will fit on + * the page. + */ + sp[ndx + 4] = OFFSET(sp); + sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE; + sp[ndx + 1] = ovfl_num; + sp[ndx + 2] = OVFLPAGE; + sp[0] = ndx + 2; +#ifdef HASH_STATISTICS + hash_overflows++; +#endif + return (bufp->ovfl); +} + +/* + * Returns: + * 0 indicates SUCCESS + * -1 indicates FAILURE + */ +extern int +__get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap) + HTAB *hashp; + char *p; + u_int32_t bucket; + int is_bucket, is_disk, is_bitmap; +{ + register int fd, page, size; + int rsize; + u_int16_t *bp; + + fd = hashp->fp; + size = hashp->BSIZE; + + if ((fd == -1) || !is_disk) { + PAGE_INIT(p); + return (0); + } + if (is_bucket) + page = BUCKET_TO_PAGE(bucket); + else + page = OADDR_TO_PAGE(bucket); + if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || + ((rsize = read(fd, p, size)) == -1)) + return (-1); + bp = (u_int16_t *)p; + if (!rsize) + bp[0] = 0; /* We hit the EOF, so initialize a new page */ + else + if (rsize != size) { + errno = EFTYPE; + return (-1); + } + if (!is_bitmap && !bp[0]) { + PAGE_INIT(p); + } else + if (hashp->LORDER != BYTE_ORDER) { + register int i, max; + + if (is_bitmap) { + max = hashp->BSIZE >> 2; /* divide by 4 */ + for (i = 0; i < max; i++) + M_32_SWAP(((int *)p)[i]); + } else { + M_16_SWAP(bp[0]); + max = bp[0] + 2; + for (i = 1; i <= max; i++) + M_16_SWAP(bp[i]); + } + } + return (0); +} + +/* + * Write page p to disk + * + * Returns: + * 0 ==> OK + * -1 ==>failure + */ +extern int +__put_page(hashp, p, bucket, is_bucket, is_bitmap) + HTAB *hashp; + char *p; + u_int32_t bucket; + int is_bucket, is_bitmap; +{ + register int fd, page, size; + int wsize; + + size = hashp->BSIZE; + if ((hashp->fp == -1) && open_temp(hashp)) + return (-1); + fd = hashp->fp; + + if (hashp->LORDER != BYTE_ORDER) { + register int i; + register int max; + + if (is_bitmap) { + max = hashp->BSIZE >> 2; /* divide by 4 */ + for (i = 0; i < max; i++) + M_32_SWAP(((int *)p)[i]); + } else { + max = ((u_int16_t *)p)[0] + 2; + for (i = 0; i <= max; i++) + M_16_SWAP(((u_int16_t *)p)[i]); + } + } + if (is_bucket) + page = BUCKET_TO_PAGE(bucket); + else + page = OADDR_TO_PAGE(bucket); + if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) || + ((wsize = write(fd, p, size)) == -1)) + /* Errno is set */ + return (-1); + if (wsize != size) { + errno = EFTYPE; + return (-1); + } + return (0); +} + +#define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1) +/* + * Initialize a new bitmap page. Bitmap pages are left in memory + * once they are read in. + */ +extern int +__ibitmap(hashp, pnum, nbits, ndx) + HTAB *hashp; + int pnum, nbits, ndx; +{ + u_int32_t *ip; + int clearbytes, clearints; + + if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) + return (1); + hashp->nmaps++; + clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1; + clearbytes = clearints << INT_TO_BYTE; + (void)memset((char *)ip, 0, clearbytes); + (void)memset(((char *)ip) + clearbytes, 0xFF, + hashp->BSIZE - clearbytes); + ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK); + SETBIT(ip, 0); + hashp->BITMAPS[ndx] = (u_int16_t)pnum; + hashp->mapp[ndx] = ip; + return (0); +} + +static u_int32_t +first_free(map) + u_int32_t map; +{ + register u_int32_t i, mask; + + mask = 0x1; + for (i = 0; i < BITS_PER_MAP; i++) { + if (!(mask & map)) + return (i); + mask = mask << 1; + } + return (i); +} + +static u_int16_t +overflow_page(hashp) + HTAB *hashp; +{ + register u_int32_t *freep; + register int max_free, offset, splitnum; + u_int16_t addr; + int bit, first_page, free_bit, free_page, i, in_use_bits, j; +#ifdef DEBUG2 + int tmp1, tmp2; +#endif + splitnum = hashp->OVFL_POINT; + max_free = hashp->SPARES[splitnum]; + + free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT); + free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1); + + /* Look through all the free maps to find the first free block */ + first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT); + for ( i = first_page; i <= free_page; i++ ) { + if (!(freep = (u_int32_t *)hashp->mapp[i]) && + !(freep = fetch_bitmap(hashp, i))) + return (NULL); + if (i == free_page) + in_use_bits = free_bit; + else + in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1; + + if (i == first_page) { + bit = hashp->LAST_FREED & + ((hashp->BSIZE << BYTE_SHIFT) - 1); + j = bit / BITS_PER_MAP; + bit = bit & ~(BITS_PER_MAP - 1); + } else { + bit = 0; + j = 0; + } + for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP) + if (freep[j] != ALL_SET) + goto found; + } + + /* No Free Page Found */ + hashp->LAST_FREED = hashp->SPARES[splitnum]; + hashp->SPARES[splitnum]++; + offset = hashp->SPARES[splitnum] - + (splitnum ? hashp->SPARES[splitnum - 1] : 0); + +#define OVMSG "HASH: Out of overflow pages. Increase page size\n" + if (offset > SPLITMASK) { + if (++splitnum >= NCACHED) { + (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); + return (NULL); + } + hashp->OVFL_POINT = splitnum; + hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; + hashp->SPARES[splitnum-1]--; + offset = 1; + } + + /* Check if we need to allocate a new bitmap page */ + if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) { + free_page++; + if (free_page >= NCACHED) { + (void)write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1); + return (NULL); + } + /* + * This is tricky. The 1 indicates that you want the new page + * allocated with 1 clear bit. Actually, you are going to + * allocate 2 pages from this map. The first is going to be + * the map page, the second is the overflow page we were + * looking for. The init_bitmap routine automatically, sets + * the first bit of itself to indicate that the bitmap itself + * is in use. We would explicitly set the second bit, but + * don't have to if we tell init_bitmap not to leave it clear + * in the first place. + */ + if (__ibitmap(hashp, (int)OADDR_OF(splitnum, offset), + 1, free_page)) + return (NULL); + hashp->SPARES[splitnum]++; +#ifdef DEBUG2 + free_bit = 2; +#endif + offset++; + if (offset > SPLITMASK) { + if (++splitnum >= NCACHED) { + (void)write(STDERR_FILENO, OVMSG, + sizeof(OVMSG) - 1); + return (NULL); + } + hashp->OVFL_POINT = splitnum; + hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1]; + hashp->SPARES[splitnum-1]--; + offset = 0; + } + } else { + /* + * Free_bit addresses the last used bit. Bump it to address + * the first available bit. + */ + free_bit++; + SETBIT(freep, free_bit); + } + + /* Calculate address of the new overflow page */ + addr = OADDR_OF(splitnum, offset); +#ifdef DEBUG2 + (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", + addr, free_bit, free_page); +#endif + return (addr); + +found: + bit = bit + first_free(freep[j]); + SETBIT(freep, bit); +#ifdef DEBUG2 + tmp1 = bit; + tmp2 = i; +#endif + /* + * Bits are addressed starting with 0, but overflow pages are addressed + * beginning at 1. Bit is a bit addressnumber, so we need to increment + * it to convert it to a page number. + */ + bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT)); + if (bit >= hashp->LAST_FREED) + hashp->LAST_FREED = bit - 1; + + /* Calculate the split number for this page */ + for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++); + offset = (i ? bit - hashp->SPARES[i - 1] : bit); + if (offset >= SPLITMASK) + return (NULL); /* Out of overflow pages */ + addr = OADDR_OF(i, offset); +#ifdef DEBUG2 + (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n", + addr, tmp1, tmp2); +#endif + + /* Allocate and return the overflow page */ + return (addr); +} + +/* + * Mark this overflow page as free. + */ +extern void +__free_ovflpage(hashp, obufp) + HTAB *hashp; + BUFHEAD *obufp; +{ + register u_int16_t addr; + u_int32_t *freep; + int bit_address, free_page, free_bit; + u_int16_t ndx; + + addr = obufp->addr; +#ifdef DEBUG1 + (void)fprintf(stderr, "Freeing %d\n", addr); +#endif + ndx = (((u_int16_t)addr) >> SPLITSHIFT); + bit_address = + (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1; + if (bit_address < hashp->LAST_FREED) + hashp->LAST_FREED = bit_address; + free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT)); + free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1); + + if (!(freep = hashp->mapp[free_page])) + freep = fetch_bitmap(hashp, free_page); +#ifdef DEBUG + /* + * This had better never happen. It means we tried to read a bitmap + * that has already had overflow pages allocated off it, and we + * failed to read it from the file. + */ + if (!freep) + assert(0); +#endif + CLRBIT(freep, free_bit); +#ifdef DEBUG2 + (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n", + obufp->addr, free_bit, free_page); +#endif + __reclaim_buf(hashp, obufp); +} + +/* + * Returns: + * 0 success + * -1 failure + */ +static int +open_temp(hashp) + HTAB *hashp; +{ + sigset_t set, oset; + static char namestr[] = "_hashXXXXXX"; + + /* Block signals; make sure file goes away at process exit. */ + (void)sigfillset(&set); + (void)sigprocmask(SIG_BLOCK, &set, &oset); + if ((hashp->fp = mkstemp(namestr)) != -1) { + (void)unlink(namestr); + (void)fcntl(hashp->fp, F_SETFD, 1); + } + (void)sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL); + return (hashp->fp != -1 ? 0 : -1); +} + +/* + * We have to know that the key will fit, but the last entry on the page is + * an overflow pair, so we need to shift things. + */ +static void +squeeze_key(sp, key, val) + u_int16_t *sp; + const DBT *key, *val; +{ + register char *p; + u_int16_t free_space, n, off, pageno; + + p = (char *)sp; + n = sp[0]; + free_space = FREESPACE(sp); + off = OFFSET(sp); + + pageno = sp[n - 1]; + off -= key->size; + sp[n - 1] = off; + memmove(p + off, key->data, key->size); + off -= val->size; + sp[n] = off; + memmove(p + off, val->data, val->size); + sp[0] = n + 2; + sp[n + 1] = pageno; + sp[n + 2] = OVFLPAGE; + FREESPACE(sp) = free_space - PAIRSIZE(key, val); + OFFSET(sp) = off; +} + +static u_int32_t * +fetch_bitmap(hashp, ndx) + HTAB *hashp; + int ndx; +{ + if (ndx >= hashp->nmaps) + return (NULL); + if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL) + return (NULL); + if (__get_page(hashp, + (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) { + free(hashp->mapp[ndx]); + return (NULL); + } + return (hashp->mapp[ndx]); +} + +#ifdef DEBUG4 +int +print_chain(addr) + int addr; +{ + BUFHEAD *bufp; + short *bp, oaddr; + + (void)fprintf(stderr, "%d ", addr); + bufp = __get_buf(hashp, addr, NULL, 0); + bp = (short *)bufp->page; + while (bp[0] && ((bp[bp[0]] == OVFLPAGE) || + ((bp[0] > 2) && bp[2] < REAL_KEY))) { + oaddr = bp[bp[0] - 1]; + (void)fprintf(stderr, "%d ", (int)oaddr); + bufp = __get_buf(hashp, (int)oaddr, bufp, 0); + bp = (short *)bufp->page; + } + (void)fprintf(stderr, "\n"); +} +#endif diff --git a/lib/libc/db/hash/hsearch.c b/lib/libc/db/hash/hsearch.c new file mode 100644 index 00000000000..ef825ab7182 --- /dev/null +++ b/lib/libc/db/hash/hsearch.c @@ -0,0 +1,113 @@ +/* $NetBSD: hsearch.c,v 1.8 1995/02/27 13:22:38 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)hsearch.c 8.3 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: hsearch.c,v 1.8 1995/02/27 13:22:38 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <fcntl.h> +#include <string.h> + +#include <db.h> +#include "search.h" + +static DB *dbp = NULL; +static ENTRY retval; + +extern int +hcreate(nel) + u_int nel; +{ + HASHINFO info; + + info.nelem = nel; + info.bsize = 256; + info.ffactor = 8; + info.cachesize = NULL; + info.hash = NULL; + info.lorder = 0; + dbp = (DB *)__hash_open(NULL, O_CREAT | O_RDWR, 0600, &info, 0); + return (dbp != NULL); +} + +extern ENTRY * +hsearch(item, action) + ENTRY item; + ACTION action; +{ + DBT key, val; + int status; + + if (!dbp) + return (NULL); + key.data = (u_char *)item.key; + key.size = strlen(item.key) + 1; + + if (action == ENTER) { + val.data = (u_char *)item.data; + val.size = strlen(item.data) + 1; + status = (dbp->put)(dbp, &key, &val, R_NOOVERWRITE); + if (status) + return (NULL); + } else { + /* FIND */ + status = (dbp->get)(dbp, &key, &val, 0); + if (status) + return (NULL); + else + item.data = (char *)val.data; + } + retval.key = item.key; + retval.data = item.data; + return (&retval); +} + +extern void +hdestroy() +{ + if (dbp) { + (void)(dbp->close)(dbp); + dbp = NULL; + } +} diff --git a/lib/libc/db/hash/ndbm.c b/lib/libc/db/hash/ndbm.c new file mode 100644 index 00000000000..875b0494e59 --- /dev/null +++ b/lib/libc/db/hash/ndbm.c @@ -0,0 +1,208 @@ +/* $NetBSD: ndbm.c,v 1.7 1995/02/27 13:22:44 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)ndbm.c 8.3 (Berkeley) 5/30/94"; +#else +static char rcsid[] = "$NetBSD: ndbm.c,v 1.7 1995/02/27 13:22:44 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +/* + * This package provides a dbm compatible interface to the new hashing + * package described in db(3). + */ + +#include <sys/param.h> + +#include <stdio.h> +#include <string.h> + +#include <ndbm.h> +#include "hash.h" + +/* + * Returns: + * *DBM on success + * NULL on failure + */ +extern DBM * +dbm_open(file, flags, mode) + const char *file; + int flags, mode; +{ + HASHINFO info; + char path[MAXPATHLEN]; + + info.bsize = 4096; + info.ffactor = 40; + info.nelem = 1; + info.cachesize = NULL; + info.hash = NULL; + info.lorder = 0; + (void)strcpy(path, file); + (void)strcat(path, DBM_SUFFIX); + return ((DBM *)__hash_open(path, flags, mode, &info, 0)); +} + +extern void +dbm_close(db) + DBM *db; +{ + (void)(db->close)(db); +} + +/* + * Returns: + * DATUM on success + * NULL on failure + */ +extern datum +dbm_fetch(db, key) + DBM *db; + datum key; +{ + datum retval; + int status; + + status = (db->get)(db, (DBT *)&key, (DBT *)&retval, 0); + if (status) { + retval.dptr = NULL; + retval.dsize = 0; + } + return (retval); +} + +/* + * Returns: + * DATUM on success + * NULL on failure + */ +extern datum +dbm_firstkey(db) + DBM *db; +{ + int status; + datum retdata, retkey; + + status = (db->seq)(db, (DBT *)&retkey, (DBT *)&retdata, R_FIRST); + if (status) + retkey.dptr = NULL; + return (retkey); +} + +/* + * Returns: + * DATUM on success + * NULL on failure + */ +extern datum +dbm_nextkey(db) + DBM *db; +{ + int status; + datum retdata, retkey; + + status = (db->seq)(db, (DBT *)&retkey, (DBT *)&retdata, R_NEXT); + if (status) + retkey.dptr = NULL; + return (retkey); +} +/* + * Returns: + * 0 on success + * <0 failure + */ +extern int +dbm_delete(db, key) + DBM *db; + datum key; +{ + int status; + + status = (db->del)(db, (DBT *)&key, 0); + if (status) + return (-1); + else + return (0); +} + +/* + * Returns: + * 0 on success + * <0 failure + * 1 if DBM_INSERT and entry exists + */ +extern int +dbm_store(db, key, content, flags) + DBM *db; + datum key, content; + int flags; +{ + return ((db->put)(db, (DBT *)&key, (DBT *)&content, + (flags == DBM_INSERT) ? R_NOOVERWRITE : 0)); +} + +extern int +dbm_error(db) + DBM *db; +{ + HTAB *hp; + + hp = (HTAB *)db->internal; + return (hp->errno); +} + +extern int +dbm_clearerr(db) + DBM *db; +{ + HTAB *hp; + + hp = (HTAB *)db->internal; + hp->errno = 0; + return (0); +} + +extern int +dbm_dirfno(db) + DBM *db; +{ + return(((HTAB *)db->internal)->fp); +} diff --git a/lib/libc/db/hash/page.h b/lib/libc/db/hash/page.h new file mode 100644 index 00000000000..ed62d595dc1 --- /dev/null +++ b/lib/libc/db/hash/page.h @@ -0,0 +1,94 @@ +/* $NetBSD: page.h,v 1.5 1995/02/27 13:22:52 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)page.h 8.2 (Berkeley) 5/31/94 + */ + +/* + * Definitions for hashing page file format. + */ + +/* + * routines dealing with a data page + * + * page format: + * +------------------------------+ + * p | n | keyoff | datoff | keyoff | + * +------------+--------+--------+ + * | datoff | free | ptr | --> | + * +--------+---------------------+ + * | F R E E A R E A | + * +--------------+---------------+ + * | <---- - - - | data | + * +--------+-----+----+----------+ + * | key | data | key | + * +--------+----------+----------+ + * + * Pointer to the free space is always: p[p[0] + 2] + * Amount of free space on the page is: p[p[0] + 1] + */ + +/* + * How many bytes required for this pair? + * 2 shorts in the table at the top of the page + room for the + * key and room for the data + * + * We prohibit entering a pair on a page unless there is also room to append + * an overflow page. The reason for this it that you can get in a situation + * where a single key/data pair fits on a page, but you can't append an + * overflow page and later you'd have to split the key/data and handle like + * a big pair. + * You might as well do this up front. + */ + +#define PAIRSIZE(K,D) (2*sizeof(u_int16_t) + (K)->size + (D)->size) +#define BIGOVERHEAD (4*sizeof(u_int16_t)) +#define KEYSIZE(K) (4*sizeof(u_int16_t) + (K)->size); +#define OVFLSIZE (2*sizeof(u_int16_t)) +#define FREESPACE(P) ((P)[(P)[0]+1]) +#define OFFSET(P) ((P)[(P)[0]+2]) +#define PAIRFITS(P,K,D) \ + (((P)[2] >= REAL_KEY) && \ + (PAIRSIZE((K),(D)) + OVFLSIZE) <= FREESPACE((P))) +#define PAGE_META(N) (((N)+3) * sizeof(u_int16_t)) + +typedef struct { + BUFHEAD *newp; + BUFHEAD *oldp; + BUFHEAD *nextp; + u_int16_t next_addr; +} SPLIT_RETURN; diff --git a/lib/libc/db/hash/search.h b/lib/libc/db/hash/search.h new file mode 100644 index 00000000000..d11ac14eefb --- /dev/null +++ b/lib/libc/db/hash/search.h @@ -0,0 +1,53 @@ +/* $NetBSD: search.h,v 1.5 1995/02/27 13:22:58 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Margo Seltzer. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)search.h 8.1 (Berkeley) 6/4/93 + */ + +/* Backward compatibility to hsearch interface. */ +typedef struct entry { + char *key; + char *data; +} ENTRY; + +typedef enum { + FIND, ENTER +} ACTION; + +int hcreate __P((unsigned int)); +void hdestroy __P((void)); +ENTRY *hsearch __P((ENTRY, ACTION)); diff --git a/lib/libc/db/man/Makefile.inc b/lib/libc/db/man/Makefile.inc new file mode 100644 index 00000000000..6ccbb21d9d1 --- /dev/null +++ b/lib/libc/db/man/Makefile.inc @@ -0,0 +1,8 @@ +# $NetBSD: Makefile.inc,v 1.5 1995/02/27 13:23:12 cgd Exp $ +# @(#)Makefile.inc 8.1 (Berkeley) 6/4/93 + +.PATH: ${.CURDIR}/db/man + +# mpool.3 +MAN+= btree.3 dbopen.3 hash.3 recno.3 +MLINKS+= dbopen.3 db.3 diff --git a/lib/libc/db/man/btree.3 b/lib/libc/db/man/btree.3 new file mode 100644 index 00000000000..f0cd4eba8d6 --- /dev/null +++ b/lib/libc/db/man/btree.3 @@ -0,0 +1,228 @@ +.\" $NetBSD: btree.3,v 1.5 1995/02/27 13:23:18 cgd Exp $ +.\" +.\" Copyright (c) 1990, 1993 +.\" The Regents of the University of California. All rights reserved. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" 3. All advertising materials mentioning features or use of this software +.\" must display the following acknowledgement: +.\" This product includes software developed by the University of +.\" California, Berkeley and its contributors. +.\" 4. Neither the name of the University nor the names of its contributors +.\" may be used to endorse or promote products derived from this software +.\" without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" @(#)btree.3 8.3 (Berkeley) 2/21/94 +.\" +.TH BTREE 3 "February 21, 1994" +.\".UC 7 +.SH NAME +btree \- btree database access method +.SH SYNOPSIS +.nf +.ft B +#include <sys/types.h> +#include <db.h> +.ft R +.fi +.SH DESCRIPTION +The routine +.IR dbopen +is the library interface to database files. +One of the supported file formats is btree files. +The general description of the database access methods is in +.IR dbopen (3), +this manual page describes only the btree specific information. +.PP +The btree data structure is a sorted, balanced tree structure storing +associated key/data pairs. +.PP +The btree access method specific data structure provided to +.I dbopen +is defined in the <db.h> include file as follows: +.PP +typedef struct { +.RS +u_long flags; +.br +u_int cachesize; +.br +int maxkeypage; +.br +int minkeypage; +.br +u_int psize; +.br +int (*compare)(const DBT *key1, const DBT *key2); +.br +size_t (*prefix)(const DBT *key1, const DBT *key2); +.br +int lorder; +.RE +} BTREEINFO; +.PP +The elements of this structure are as follows: +.TP +flags +The flag value is specified by +.IR or 'ing +any of the following values: +.RS +.TP +R_DUP +Permit duplicate keys in the tree, i.e. permit insertion if the key to be +inserted already exists in the tree. +The default behavior, as described in +.IR dbopen (3), +is to overwrite a matching key when inserting a new key or to fail if +the R_NOOVERWRITE flag is specified. +The R_DUP flag is overridden by the R_NOOVERWRITE flag, and if the +R_NOOVERWRITE flag is specified, attempts to insert duplicate keys into +the tree will fail. +.IP +If the database contains duplicate keys, the order of retrieval of +key/data pairs is undefined if the +.I get +routine is used, however, +.I seq +routine calls with the R_CURSOR flag set will always return the logical +``first'' of any group of duplicate keys. +.RE +.TP +cachesize +A suggested maximum size (in bytes) of the memory cache. +This value is +.B only +advisory, and the access method will allocate more memory rather than fail. +Since every search examines the root page of the tree, caching the most +recently used pages substantially improves access time. +In addition, physical writes are delayed as long as possible, so a moderate +cache can reduce the number of I/O operations significantly. +Obviously, using a cache increases (but only increases) the likelihood of +corruption or lost data if the system crashes while a tree is being modified. +If +.I cachesize +is 0 (no size is specified) a default cache is used. +.TP +maxkeypage +The maximum number of keys which will be stored on any single page. +Not currently implemented. +.\" The maximum number of keys which will be stored on any single page. +.\" Because of the way the btree data structure works, +.\" .I maxkeypage +.\" must always be greater than or equal to 2. +.\" If +.\" .I maxkeypage +.\" is 0 (no maximum number of keys is specified) the page fill factor is +.\" made as large as possible (which is almost invariably what is wanted). +.TP +minkeypage +The minimum number of keys which will be stored on any single page. +This value is used to determine which keys will be stored on overflow +pages, i.e. if a key or data item is longer than the pagesize divided +by the minkeypage value, it will be stored on overflow pages instead +of in the page itself. +If +.I minkeypage +is 0 (no minimum number of keys is specified) a value of 2 is used. +.TP +psize +Page size is the size (in bytes) of the pages used for nodes in the tree. +The minimum page size is 512 bytes and the maximum page size is 64K. +If +.I psize +is 0 (no page size is specified) a page size is chosen based on the +underlying file system I/O block size. +.TP +compare +Compare is the key comparison function. +It must return an integer less than, equal to, or greater than zero if the +first key argument is considered to be respectively less than, equal to, +or greater than the second key argument. +The same comparison function must be used on a given tree every time it +is opened. +If +.I compare +is NULL (no comparison function is specified), the keys are compared +lexically, with shorter keys considered less than longer keys. +.TP +prefix +Prefix is the prefix comparison function. +If specified, this routine must return the number of bytes of the second key +argument which are necessary to determine that it is greater than the first +key argument. +If the keys are equal, the key length should be returned. +Note, the usefulness of this routine is very data dependent, but, in some +data sets can produce significantly reduced tree sizes and search times. +If +.I prefix +is NULL (no prefix function is specified), +.B and +no comparison function is specified, a default lexical comparison routine +is used. +If +.I prefix +is NULL and a comparison routine is specified, no prefix comparison is +done. +.TP +lorder +The byte order for integers in the stored database metadata. +The number should represent the order as an integer; for example, +big endian order would be the number 4,321. +If +.I lorder +is 0 (no order is specified) the current host order is used. +.PP +If the file already exists (and the O_TRUNC flag is not specified), the +values specified for the parameters flags, lorder and psize are ignored +in favor of the values used when the tree was created. +.PP +Forward sequential scans of a tree are from the least key to the greatest. +.PP +Space freed up by deleting key/data pairs from the tree is never reclaimed, +although it is normally made available for reuse. +This means that the btree storage structure is grow-only. +The only solutions are to avoid excessive deletions, or to create a fresh +tree periodically from a scan of an existing one. +.PP +Searches, insertions, and deletions in a btree will all complete in +O lg base N where base is the average fill factor. +Often, inserting ordered data into btrees results in a low fill factor. +This implementation has been modified to make ordered insertion the best +case, resulting in a much better than normal page fill factor. +.SH "SEE ALSO" +.IR dbopen (3), +.IR hash (3), +.IR mpool (3), +.IR recno (3) +.sp +.IR "The Ubiquitous B-tree" , +Douglas Comer, ACM Comput. Surv. 11, 2 (June 1979), 121-138. +.sp +.IR "Prefix B-trees" , +Bayer and Unterauer, ACM Transactions on Database Systems, Vol. 2, 1 +(March 1977), 11-26. +.sp +.IR "The Art of Computer Programming Vol. 3: Sorting and Searching" , +D.E. Knuth, 1968, pp 471-480. +.SH BUGS +Only big and little endian byte order is supported. diff --git a/lib/libc/db/man/dbopen.3 b/lib/libc/db/man/dbopen.3 new file mode 100644 index 00000000000..b842160b19f --- /dev/null +++ b/lib/libc/db/man/dbopen.3 @@ -0,0 +1,478 @@ +.\" $NetBSD: dbopen.3,v 1.6 1995/02/27 13:23:25 cgd Exp $ +.\" +.\" Copyright (c) 1990, 1993 +.\" The Regents of the University of California. All rights reserved. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" 3. All advertising materials mentioning features or use of this software +.\" must display the following acknowledgement: +.\" This product includes software developed by the University of +.\" California, Berkeley and its contributors. +.\" 4. Neither the name of the University nor the names of its contributors +.\" may be used to endorse or promote products derived from this software +.\" without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" @(#)dbopen.3 8.5 (Berkeley) 1/2/94 +.\" +.TH DBOPEN 3 "January 2, 1994" +.UC 7 +.SH NAME +dbopen \- database access methods +.SH SYNOPSIS +.nf +.ft B +#include <sys/types.h> +#include <limits.h> +#include <db.h> + +DB * +dbopen(const char *file, int flags, int mode, DBTYPE type, +.ti +5 +const void *openinfo); +.ft R +.fi +.SH DESCRIPTION +.IR Dbopen +is the library interface to database files. +The supported file formats are btree, hashed and UNIX file oriented. +The btree format is a representation of a sorted, balanced tree structure. +The hashed format is an extensible, dynamic hashing scheme. +The flat-file format is a byte stream file with fixed or variable length +records. +The formats and file format specific information are described in detail +in their respective manual pages +.IR btree (3), +.IR hash (3) +and +.IR recno (3). +.PP +Dbopen opens +.I file +for reading and/or writing. +Files never intended to be preserved on disk may be created by setting +the file parameter to NULL. +.PP +The +.I flags +and +.I mode arguments +are as specified to the +.IR open (2) +routine, however, only the O_CREAT, O_EXCL, O_EXLOCK, O_NONBLOCK, +O_RDONLY, O_RDWR, O_SHLOCK and O_TRUNC flags are meaningful. +(Note, opening a database file O_WRONLY is not possible.) +.\"Three additional options may be specified by +.\".IR or 'ing +.\"them into the +.\".I flags +.\"argument. +.\".TP +.\"DB_LOCK +.\"Do the necessary locking in the database to support concurrent access. +.\"If concurrent access isn't needed or the database is read-only this +.\"flag should not be set, as it tends to have an associated performance +.\"penalty. +.\".TP +.\"DB_SHMEM +.\"Place the underlying memory pool used by the database in shared +.\"memory. +.\"Necessary for concurrent access. +.\".TP +.\"DB_TXN +.\"Support transactions in the database. +.\"The DB_LOCK and DB_SHMEM flags must be set as well. +.PP +The +.I type +argument is of type DBTYPE (as defined in the <db.h> include file) and +may be set to DB_BTREE, DB_HASH or DB_RECNO. +.PP +The +.I openinfo +argument is a pointer to an access method specific structure described +in the access method's manual page. +If +.I openinfo +is NULL, each access method will use defaults appropriate for the system +and the access method. +.PP +.I Dbopen +returns a pointer to a DB structure on success and NULL on error. +The DB structure is defined in the <db.h> include file, and contains at +least the following fields: +.sp +.nf +typedef struct { +.RS +DBTYPE type; +int (*close)(const DB *db); +int (*del)(const DB *db, const DBT *key, u_int flags); +int (*fd)(const DB *db); +int (*get)(const DB *db, DBT *key, DBT *data, u_int flags); +int (*put)(const DB *db, DBT *key, const DBT *data, +.ti +5 +u_int flags); +int (*sync)(const DB *db, u_int flags); +int (*seq)(const DB *db, DBT *key, DBT *data, u_int flags); +.RE +} DB; +.fi +.PP +These elements describe a database type and a set of functions performing +various actions. +These functions take a pointer to a structure as returned by +.IR dbopen , +and sometimes one or more pointers to key/data structures and a flag value. +.TP +type +The type of the underlying access method (and file format). +.TP +close +A pointer to a routine to flush any cached information to disk, free any +allocated resources, and close the underlying file(s). +Since key/data pairs may be cached in memory, failing to sync the file +with a +.I close +or +.I sync +function may result in inconsistent or lost information. +.I Close +routines return -1 on error (setting +.IR errno ) +and 0 on success. +.TP +del +A pointer to a routine to remove key/data pairs from the database. +.IP +The parameter +.I flag +may be set to the following value: +.RS +.TP +R_CURSOR +Delete the record referenced by the cursor. +The cursor must have previously been initialized. +.RE +.IP +.I Delete +routines return -1 on error (setting +.IR errno ), +0 on success, and 1 if the specified +.I key +was not in the file. +.TP +fd +A pointer to a routine which returns a file descriptor representative +of the underlying database. +A file descriptor referencing the same file will be returned to all +processes which call +.I dbopen +with the same +.I file +name. +This file descriptor may be safely used as an argument to the +.IR fcntl (2) +and +.IR flock (2) +locking functions. +The file descriptor is not necessarily associated with any of the +underlying files used by the access method. +No file descriptor is available for in memory databases. +.I Fd +routines return -1 on error (setting +.IR errno ), +and the file descriptor on success. +.TP +get +A pointer to a routine which is the interface for keyed retrieval from +the database. +The address and length of the data associated with the specified +.I key +are returned in the structure referenced by +.IR data . +.I Get +routines return -1 on error (setting +.IR errno ), +0 on success, and 1 if the +.I key +was not in the file. +.TP +put +A pointer to a routine to store key/data pairs in the database. +.IP +The parameter +.I flag +may be set to one of the following values: +.RS +.TP +R_CURSOR +Replace the key/data pair referenced by the cursor. +The cursor must have previously been initialized. +.TP +R_IAFTER +Append the data immediately after the data referenced by +.IR key , +creating a new key/data pair. +The record number of the appended key/data pair is returned in the +.I key +structure. +(Applicable only to the DB_RECNO access method.) +.TP +R_IBEFORE +Insert the data immediately before the data referenced by +.IR key , +creating a new key/data pair. +The record number of the inserted key/data pair is returned in the +.I key +structure. +(Applicable only to the DB_RECNO access method.) +.TP +R_NOOVERWRITE +Enter the new key/data pair only if the key does not previously exist. +.TP +R_SETCURSOR +Store the key/data pair, setting or initializing the position of the +cursor to reference it. +(Applicable only to the DB_BTREE and DB_RECNO access methods.) +.RE +.IP +R_SETCURSOR is available only for the DB_BTREE and DB_RECNO access +methods because it implies that the keys have an inherent order +which does not change. +.IP +R_IAFTER and R_IBEFORE are available only for the DB_RECNO +access method because they each imply that the access method is able to +create new keys. +This is only true if the keys are ordered and independent, record numbers +for example. +.IP +The default behavior of the +.I put +routines is to enter the new key/data pair, replacing any previously +existing key. +.IP +.I Put +routines return -1 on error (setting +.IR errno ), +0 on success, and 1 if the R_NOOVERWRITE +.I flag +was set and the key already exists in the file. +.TP +seq +A pointer to a routine which is the interface for sequential +retrieval from the database. +The address and length of the key are returned in the structure +referenced by +.IR key , +and the address and length of the data are returned in the +structure referenced +by +.IR data . +.IP +Sequential key/data pair retrieval may begin at any time, and the +position of the ``cursor'' is not affected by calls to the +.IR del , +.IR get , +.IR put , +or +.I sync +routines. +Modifications to the database during a sequential scan will be reflected +in the scan, i.e. records inserted behind the cursor will not be returned +while records inserted in front of the cursor will be returned. +.IP +The flag value +.B must +be set to one of the following values: +.RS +.TP +R_CURSOR +The data associated with the specified key is returned. +This differs from the +.I get +routines in that it sets or initializes the cursor to the location of +the key as well. +(Note, for the DB_BTREE access method, the returned key is not necessarily an +exact match for the specified key. +The returned key is the smallest key greater than or equal to the specified +key, permitting partial key matches and range searches.) +.TP +R_FIRST +The first key/data pair of the database is returned, and the cursor +is set or initialized to reference it. +.TP +R_LAST +The last key/data pair of the database is returned, and the cursor +is set or initialized to reference it. +(Applicable only to the DB_BTREE and DB_RECNO access methods.) +.TP +R_NEXT +Retrieve the key/data pair immediately after the cursor. +If the cursor is not yet set, this is the same as the R_FIRST flag. +.TP +R_PREV +Retrieve the key/data pair immediately before the cursor. +If the cursor is not yet set, this is the same as the R_LAST flag. +(Applicable only to the DB_BTREE and DB_RECNO access methods.) +.RE +.IP +R_LAST and R_PREV are available only for the DB_BTREE and DB_RECNO +access methods because they each imply that the keys have an inherent +order which does not change. +.IP +.I Seq +routines return -1 on error (setting +.IR errno ), +0 on success and 1 if there are no key/data pairs less than or greater +than the specified or current key. +If the DB_RECNO access method is being used, and if the database file +is a character special file and no complete key/data pairs are currently +available, the +.I seq +routines return 2. +.TP +sync +A pointer to a routine to flush any cached information to disk. +If the database is in memory only, the +.I sync +routine has no effect and will always succeed. +.IP +The flag value may be set to the following value: +.RS +.TP +R_RECNOSYNC +If the DB_RECNO access method is being used, this flag causes +the sync routine to apply to the btree file which underlies the +recno file, not the recno file itself. +(See the +.I bfname +field of the +.IR recno (3) +manual page for more information.) +.RE +.IP +.I Sync +routines return -1 on error (setting +.IR errno ) +and 0 on success. +.SH "KEY/DATA PAIRS" +Access to all file types is based on key/data pairs. +Both keys and data are represented by the following data structure: +.PP +typedef struct { +.RS +void *data; +.br +size_t size; +.RE +} DBT; +.PP +The elements of the DBT structure are defined as follows: +.TP +data +A pointer to a byte string. +.TP +size +The length of the byte string. +.PP +Key and data byte strings may reference strings of essentially unlimited +length although any two of them must fit into available memory at the same +time. +It should be noted that the access methods provide no guarantees about +byte string alignment. +.SH ERRORS +The +.I dbopen +routine may fail and set +.I errno +for any of the errors specified for the library routines +.IR open (2) +and +.IR malloc (3) +or the following: +.TP +[EFTYPE] +A file is incorrectly formatted. +.TP +[EINVAL] +A parameter has been specified (hash function, pad byte etc.) that is +incompatible with the current file specification or which is not +meaningful for the function (for example, use of the cursor without +prior initialization) or there is a mismatch between the version +number of file and the software. +.PP +The +.I close +routines may fail and set +.I errno +for any of the errors specified for the library routines +.IR close (2), +.IR read (2), +.IR write (2), +.IR free (3), +or +.IR fsync (2). +.PP +The +.IR del , +.IR get , +.I put +and +.I seq +routines may fail and set +.I errno +for any of the errors specified for the library routines +.IR read (2), +.IR write (2), +.IR free (3) +or +.IR malloc (3). +.PP +The +.I fd +routines will fail and set +.I errno +to ENOENT for in memory databases. +.PP +The +.I sync +routines may fail and set +.I errno +for any of the errors specified for the library routine +.IR fsync (2). +.SH "SEE ALSO" +.IR btree (3), +.IR hash (3), +.IR mpool (3), +.IR recno (3) +.sp +.IR "LIBTP: Portable, Modular Transactions for UNIX" , +Margo Seltzer, Michael Olson, USENIX proceedings, Winter 1992. +.SH BUGS +The typedef DBT is a mnemonic for ``data base thang'', and was used +because noone could think of a reasonable name that wasn't already used. +.PP +The file descriptor interface is a kluge and will be deleted in a +future version of the interface. +.PP +None of the access methods provide any form of concurrent access, +locking, or transactions. diff --git a/lib/libc/db/man/hash.3 b/lib/libc/db/man/hash.3 new file mode 100644 index 00000000000..6d296a8f6d1 --- /dev/null +++ b/lib/libc/db/man/hash.3 @@ -0,0 +1,154 @@ +.\" $NetBSD: hash.3,v 1.5 1995/02/27 13:23:31 cgd Exp $ +.\" +.\" Copyright (c) 1990, 1993 +.\" The Regents of the University of California. All rights reserved. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" 3. All advertising materials mentioning features or use of this software +.\" must display the following acknowledgement: +.\" This product includes software developed by the University of +.\" California, Berkeley and its contributors. +.\" 4. Neither the name of the University nor the names of its contributors +.\" may be used to endorse or promote products derived from this software +.\" without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" @(#)hash.3 8.5 (Berkeley) 2/21/94 +.\" +.TH HASH 3 "February 21, 1994" +.UC 7 +.SH NAME +hash \- hash database access method +.SH SYNOPSIS +.nf +.ft B +#include <sys/types.h> +#include <db.h> +.ft R +.fi +.SH DESCRIPTION +The routine +.IR dbopen +is the library interface to database files. +One of the supported file formats is hash files. +The general description of the database access methods is in +.IR dbopen (3), +this manual page describes only the hash specific information. +.PP +The hash data structure is an extensible, dynamic hashing scheme. +.PP +The access method specific data structure provided to +.I dbopen +is defined in the <db.h> include file as follows: +.sp +typedef struct { +.RS +u_int bsize; +.br +u_int ffactor; +.br +u_int nelem; +.br +u_int cachesize; +.br +u_int32_t (*hash)(const void *, size_t); +.br +int lorder; +.RE +} HASHINFO; +.PP +The elements of this structure are as follows: +.TP +bsize +.I Bsize +defines the hash table bucket size, and is, by default, 256 bytes. +It may be preferable to increase the page size for disk-resident tables +and tables with large data items. +.TP +ffactor +.I Ffactor +indicates a desired density within the hash table. +It is an approximation of the number of keys allowed to accumulate in any +one bucket, determining when the hash table grows or shrinks. +The default value is 8. +.TP +nelem +.I Nelem +is an estimate of the final size of the hash table. +If not set or set too low, hash tables will expand gracefully as keys +are entered, although a slight performance degradation may be noticed. +The default value is 1. +.TP +cachesize +A suggested maximum size, in bytes, of the memory cache. +This value is +.B only +advisory, and the access method will allocate more memory rather +than fail. +.TP +hash +.I Hash +is a user defined hash function. +Since no hash function performs equally well on all possible data, the +user may find that the built-in hash function does poorly on a particular +data set. +User specified hash functions must take two arguments (a pointer to a byte +string and a length) and return a 32-bit quantity to be used as the hash +value. +.TP +lorder +The byte order for integers in the stored database metadata. +The number should represent the order as an integer; for example, +big endian order would be the number 4,321. +If +.I lorder +is 0 (no order is specified) the current host order is used. +If the file already exists, the specified value is ignored and the +value specified when the tree was created is used. +.PP +If the file already exists (and the O_TRUNC flag is not specified), the +values specified for the parameters bsize, ffactor, lorder and nelem are +ignored and the values specified when the tree was created are used. +.PP +If a hash function is specified, +.I hash_open +will attempt to determine if the hash function specified is the same as +the one with which the database was created, and will fail if it is not. +.PP +Backward compatible interfaces to the routines described in +.IR dbm (3), +and +.IR ndbm (3) +are provided, however these interfaces are not compatible with +previous file formats. +.SH "SEE ALSO" +.IR btree (3), +.IR dbopen (3), +.IR mpool (3), +.IR recno (3) +.sp +.IR "Dynamic Hash Tables" , +Per-Ake Larson, Communications of the ACM, April 1988. +.sp +.IR "A New Hash Package for UNIX" , +Margo Seltzer, USENIX Proceedings, Winter 1991. +.SH BUGS +Only big and little endian byte order is supported. diff --git a/lib/libc/db/man/mpool.3 b/lib/libc/db/man/mpool.3 new file mode 100644 index 00000000000..0ad0b6864b5 --- /dev/null +++ b/lib/libc/db/man/mpool.3 @@ -0,0 +1,221 @@ +.\" $NetBSD: mpool.3,v 1.4 1995/02/27 13:23:35 cgd Exp $ +.\" +.\" Copyright (c) 1990, 1993 +.\" The Regents of the University of California. All rights reserved. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" 3. All advertising materials mentioning features or use of this software +.\" must display the following acknowledgement: +.\" This product includes software developed by the University of +.\" California, Berkeley and its contributors. +.\" 4. Neither the name of the University nor the names of its contributors +.\" may be used to endorse or promote products derived from this software +.\" without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" @(#)mpool.3 8.1 (Berkeley) 6/4/93 +.\" +.TH MPOOL 3 "June 4, 1993" +.UC 7 +.SH NAME +mpool \- shared memory buffer pool +.SH SYNOPSIS +.nf +.ft B +#include <db.h> +#include <mpool.h> + +MPOOL * +mpool_open (DBT *key, int fd, pgno_t pagesize, pgno_t maxcache); + +void +mpool_filter (MPOOL *mp, void (*pgin)(void *, pgno_t, void *), +.ti +5 +void (*pgout)(void *, pgno_t, void *), void *pgcookie); + +void * +mpool_new (MPOOL *mp, pgno_t *pgnoaddr); + +void * +mpool_get (MPOOL *mp, pgno_t pgno, u_int flags); + +int +mpool_put (MPOOL *mp, void *pgaddr, u_int flags); + +int +mpool_sync (MPOOL *mp); + +int +mpool_close (MPOOL *mp); +.ft R +.fi +.SH DESCRIPTION +.IR Mpool +is the library interface intended to provide page oriented buffer management +of files. +The buffers may be shared between processes. +.PP +The function +.I mpool_open +initializes a memory pool. +The +.I key +argument is the byte string used to negotiate between multiple +processes wishing to share buffers. +If the file buffers are mapped in shared memory, all processes using +the same key will share the buffers. +If +.I key +is NULL, the buffers are mapped into private memory. +The +.I fd +argument is a file descriptor for the underlying file, which must be seekable. +If +.I key +is non-NULL and matches a file already being mapped, the +.I fd +argument is ignored. +.PP +The +.I pagesize +argument is the size, in bytes, of the pages into which the file is broken up. +The +.I maxcache +argument is the maximum number of pages from the underlying file to cache +at any one time. +This value is not relative to the number of processes which share a file's +buffers, but will be the largest value specified by any of the processes +sharing the file. +.PP +The +.I mpool_filter +function is intended to make transparent input and output processing of the +pages possible. +If the +.I pgin +function is specified, it is called each time a buffer is read into the memory +pool from the backing file. +If the +.I pgout +function is specified, it is called each time a buffer is written into the +backing file. +Both functions are are called with the +.I pgcookie +pointer, the page number and a pointer to the page to being read or written. +.PP +The function +.I mpool_new +takes an MPOOL pointer and an address as arguments. +If a new page can be allocated, a pointer to the page is returned and +the page number is stored into the +.I pgnoaddr +address. +Otherwise, NULL is returned and errno is set. +.PP +The function +.I mpool_get +takes a MPOOL pointer and a page number as arguments. +If the page exists, a pointer to the page is returned. +Otherwise, NULL is returned and errno is set. +The flags parameter is not currently used. +.PP +The function +.I mpool_put +unpins the page referenced by +.IR pgaddr . +.I Pgaddr +must be an address previously returned by +.I mpool_get +or +.IR mpool_new . +The flag value is specified by +.IR or 'ing +any of the following values: +.TP +MPOOL_DIRTY +The page has been modified and needs to be written to the backing file. +.PP +.I Mpool_put +returns 0 on success and -1 if an error occurs. +.PP +The function +.I mpool_sync +writes all modified pages associated with the MPOOL pointer to the +backing file. +.I Mpool_sync +returns 0 on success and -1 if an error occurs. +.PP +The +.I mpool_close +function free's up any allocated memory associated with the memory pool +cookie. +Modified pages are +.B not +written to the backing file. +.I Mpool_close +returns 0 on success and -1 if an error occurs. +.SH ERRORS +The +.I mpool_open +function may fail and set +.I errno +for any of the errors specified for the library routine +.IR malloc (3). +.PP +The +.I mpool_get +function may fail and set +.I errno +for the following: +.TP 15 +[EINVAL] +The requested record doesn't exist. +.PP +The +.I mpool_new +and +.I mpool_get +functions may fail and set +.I errno +for any of the errors specified for the library routines +.IR read (2) , +.IR write (2) , +and +.IR malloc (3). +.PP +The +.I mpool_sync +function may fail and set +.I errno +for any of the errors specified for the library routine +.IR write (2). +.PP +The +.I mpool_close +function may fail and set +.I errno +for any of the errors specified for the library routine +.IR free (3). +.SH "SEE ALSO" +.IR dbopen (3), +.IR btree (3), +.IR hash (3), +.IR recno (3) diff --git a/lib/libc/db/man/recno.3 b/lib/libc/db/man/recno.3 new file mode 100644 index 00000000000..63a366b645f --- /dev/null +++ b/lib/libc/db/man/recno.3 @@ -0,0 +1,198 @@ +.\" $NetBSD: recno.3,v 1.5 1995/06/13 00:53:40 jtc Exp $ +.\" +.\" Copyright (c) 1990, 1993 +.\" The Regents of the University of California. All rights reserved. +.\" +.\" Redistribution and use in source and binary forms, with or without +.\" modification, are permitted provided that the following conditions +.\" are met: +.\" 1. Redistributions of source code must retain the above copyright +.\" notice, this list of conditions and the following disclaimer. +.\" 2. Redistributions in binary form must reproduce the above copyright +.\" notice, this list of conditions and the following disclaimer in the +.\" documentation and/or other materials provided with the distribution. +.\" 3. All advertising materials mentioning features or use of this software +.\" must display the following acknowledgement: +.\" This product includes software developed by the University of +.\" California, Berkeley and its contributors. +.\" 4. Neither the name of the University nor the names of its contributors +.\" may be used to endorse or promote products derived from this software +.\" without specific prior written permission. +.\" +.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND +.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +.\" ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE +.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS +.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) +.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT +.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY +.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF +.\" SUCH DAMAGE. +.\" +.\" @(#)recno.3 8.3 (Berkeley) 2/21/94 +.\" +.TH RECNO 3 "February 21, 1994" +.UC 7 +.SH NAME +recno \- record number database access method +.SH SYNOPSIS +.nf +.ft B +#include <sys/types.h> +#include <db.h> +.ft R +.fi +.SH DESCRIPTION +The routine +.IR dbopen +is the library interface to database files. +One of the supported file formats is record number files. +The general description of the database access methods is in +.IR dbopen (3), +this manual page describes only the recno specific information. +.PP +The record number data structure is either variable or fixed-length +records stored in a flat-file format, accessed by the logical record +number. +The existence of record number five implies the existence of records +one through four, and the deletion of record number one causes +record number five to be renumbered to record number four, as well +as the cursor, if positioned after record number one, to shift down +one record. +.PP +The recno access method specific data structure provided to +.I dbopen +is defined in the <db.h> include file as follows: +.PP +typedef struct { +.RS +u_long flags; +.br +u_int cachesize; +.br +u_int psize; +.br +int lorder; +.br +size_t reclen; +.br +u_char bval; +.br +char *bfname; +.RE +} RECNOINFO; +.PP +The elements of this structure are defined as follows: +.TP +flags +The flag value is specified by +.IR or 'ing +any of the following values: +.RS +.TP +R_FIXEDLEN +The records are fixed-length, not byte delimited. +The structure element +.I reclen +specifies the length of the record, and the structure element +.I bval +is used as the pad character. +.TP +R_NOKEY +In the interface specified by +.IR dbopen , +the sequential record retrieval fills in both the caller's key and +data structures. +If the R_NOKEY flag is specified, the +.I cursor +routines are not required to fill in the key structure. +This permits applications to retrieve records at the end of files without +reading all of the intervening records. +.TP +R_SNAPSHOT +This flag requires that a snapshot of the file be taken when +.I dbopen +is called, instead of permitting any unmodified records to be read from +the original file. +.RE +.TP +cachesize +A suggested maximum size, in bytes, of the memory cache. +This value is +.B only +advisory, and the access method will allocate more memory rather than fail. +If +.I cachesize +is 0 (no size is specified) a default cache is used. +.TP +psize +The recno access method stores the in-memory copies of its records +in a btree. +This value is the size (in bytes) of the pages used for nodes in that tree. +If +.I psize +is 0 (no page size is specified) a page size is chosen based on the +underlying file system I/O block size. +See +.IR btree (3) +for more information. +.TP +lorder +The byte order for integers in the stored database metadata. +The number should represent the order as an integer; for example, +big endian order would be the number 4,321. +If +.I lorder +is 0 (no order is specified) the current host order is used. +.TP +reclen +The length of a fixed-length record. +.TP +bval +The delimiting byte to be used to mark the end of a record for +variable-length records, and the pad character for fixed-length +records. +If no value is specified, newlines (``\en'') are used to mark the end +of variable-length records and fixed-length records are padded with +spaces. +.TP +bfname +The recno access method stores the in-memory copies of its records +in a btree. +If bfname is non-NULL, it specifies the name of the btree file, +as if specified as the file name for a dbopen of a btree file. +.PP +The data part of the key/data pair used by the recno access method +is the same as other access methods. +The key is different. +The +.I data +field of the key should be a pointer to a memory location of type +.IR recno_t , +as defined in the <db.h> include file. +This type is normally the largest unsigned integral type available to +the implementation. +The +.I size +field of the key should be the size of that type. +.PP +In the interface specified by +.IR dbopen , +using the +.I put +interface to create a new record will cause the creation of multiple, +empty records if the record number is more than one greater than the +largest record currently in the database. +.SH "SEE ALSO" +.IR btree (3), +.IR dbopen (3), +.IR hash (3), +.IR mpool (3) +.sp +.IR "Document Processing in a Relational Database System" , +Michael Stonebraker, Heidi Stettner, Joseph Kalash, Antonin Guttman, +Nadene Lynn, Memorandum No. UCB/ERL M82/32, May 1982. +.SH BUGS +Only big and little endian byte order is supported. diff --git a/lib/libc/db/mpool/Makefile.inc b/lib/libc/db/mpool/Makefile.inc new file mode 100644 index 00000000000..c9320fb0a1e --- /dev/null +++ b/lib/libc/db/mpool/Makefile.inc @@ -0,0 +1,6 @@ +# $NetBSD: Makefile.inc,v 1.4 1995/02/27 13:23:53 cgd Exp $ +# @(#)Makefile.inc 8.1 (Berkeley) 6/4/93 + +.PATH: ${.CURDIR}/db/mpool + +SRCS+= mpool.c diff --git a/lib/libc/db/mpool/README b/lib/libc/db/mpool/README new file mode 100644 index 00000000000..13002b30fd2 --- /dev/null +++ b/lib/libc/db/mpool/README @@ -0,0 +1,8 @@ +# $NetBSD: README,v 1.2 1995/02/27 13:24:00 cgd Exp $ +# @(#)README 8.1 (Berkeley) 6/4/93 + +These are the current memory pool routines. +They aren't ready for prime time, yet, and +the interface is expected to change. + +--keith diff --git a/lib/libc/db/mpool/mpool.c b/lib/libc/db/mpool/mpool.c new file mode 100644 index 00000000000..6417c29d453 --- /dev/null +++ b/lib/libc/db/mpool/mpool.c @@ -0,0 +1,540 @@ +/* $NetBSD: mpool.c,v 1.5 1995/02/27 13:24:05 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)mpool.c 8.2 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: mpool.c,v 1.5 1995/02/27 13:24:05 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> +#include <sys/stat.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#include <db.h> +#define __MPOOLINTERFACE_PRIVATE +#include "mpool.h" + +static BKT *mpool_bkt __P((MPOOL *)); +static BKT *mpool_look __P((MPOOL *, pgno_t)); +static int mpool_write __P((MPOOL *, BKT *)); +#ifdef DEBUG +static void __mpoolerr __P((const char *fmt, ...)); +#endif + +/* + * MPOOL_OPEN -- initialize a memory pool. + * + * Parameters: + * key: Shared buffer key. + * fd: File descriptor. + * pagesize: File page size. + * maxcache: Max number of cached pages. + * + * Returns: + * MPOOL pointer, NULL on error. + */ +MPOOL * +mpool_open(key, fd, pagesize, maxcache) + DBT *key; + int fd; + pgno_t pagesize, maxcache; +{ + struct stat sb; + MPOOL *mp; + int entry; + + if (fstat(fd, &sb)) + return (NULL); + /* XXX + * We should only set st_size to 0 for pipes -- 4.4BSD has the fix so + * that stat(2) returns true for ISSOCK on pipes. Until then, this is + * fairly close. + */ + if (!S_ISREG(sb.st_mode)) { + errno = ESPIPE; + return (NULL); + } + + if ((mp = (MPOOL *)malloc(sizeof(MPOOL))) == NULL) + return (NULL); + mp->free.cnext = mp->free.cprev = (BKT *)&mp->free; + mp->lru.cnext = mp->lru.cprev = (BKT *)&mp->lru; + for (entry = 0; entry < HASHSIZE; ++entry) + mp->hashtable[entry].hnext = mp->hashtable[entry].hprev = + mp->hashtable[entry].cnext = mp->hashtable[entry].cprev = + (BKT *)&mp->hashtable[entry]; + mp->curcache = 0; + mp->maxcache = maxcache; + mp->pagesize = pagesize; + mp->npages = sb.st_size / pagesize; + mp->fd = fd; + mp->pgcookie = NULL; + mp->pgin = mp->pgout = NULL; + +#ifdef STATISTICS + mp->cachehit = mp->cachemiss = mp->pagealloc = mp->pageflush = + mp->pageget = mp->pagenew = mp->pageput = mp->pageread = + mp->pagewrite = 0; +#endif + return (mp); +} + +/* + * MPOOL_FILTER -- initialize input/output filters. + * + * Parameters: + * pgin: Page in conversion routine. + * pgout: Page out conversion routine. + * pgcookie: Cookie for page in/out routines. + */ +void +mpool_filter(mp, pgin, pgout, pgcookie) + MPOOL *mp; + void (*pgin) __P((void *, pgno_t, void *)); + void (*pgout) __P((void *, pgno_t, void *)); + void *pgcookie; +{ + mp->pgin = pgin; + mp->pgout = pgout; + mp->pgcookie = pgcookie; +} + +/* + * MPOOL_NEW -- get a new page + * + * Parameters: + * mp: mpool cookie + * pgnoadddr: place to store new page number + * Returns: + * RET_ERROR, RET_SUCCESS + */ +void * +mpool_new(mp, pgnoaddr) + MPOOL *mp; + pgno_t *pgnoaddr; +{ + BKT *b; + BKTHDR *hp; + +#ifdef STATISTICS + ++mp->pagenew; +#endif + /* + * Get a BKT from the cache. Assign a new page number, attach it to + * the hash and lru chains and return. + */ + if ((b = mpool_bkt(mp)) == NULL) + return (NULL); + *pgnoaddr = b->pgno = mp->npages++; + b->flags = MPOOL_PINNED; + inshash(b, b->pgno); + inschain(b, &mp->lru); + return (b->page); +} + +/* + * MPOOL_GET -- get a page from the pool + * + * Parameters: + * mp: mpool cookie + * pgno: page number + * flags: not used + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +void * +mpool_get(mp, pgno, flags) + MPOOL *mp; + pgno_t pgno; + u_int flags; /* XXX not used? */ +{ + BKT *b; + BKTHDR *hp; + off_t off; + int nr; + + /* + * If asking for a specific page that is already in the cache, find + * it and return it. + */ + if (b = mpool_look(mp, pgno)) { +#ifdef STATISTICS + ++mp->pageget; +#endif +#ifdef DEBUG + if (b->flags & MPOOL_PINNED) + __mpoolerr("mpool_get: page %d already pinned", + b->pgno); +#endif + rmchain(b); + inschain(b, &mp->lru); + b->flags |= MPOOL_PINNED; + return (b->page); + } + + /* Not allowed to retrieve a non-existent page. */ + if (pgno >= mp->npages) { + errno = EINVAL; + return (NULL); + } + + /* Get a page from the cache. */ + if ((b = mpool_bkt(mp)) == NULL) + return (NULL); + b->pgno = pgno; + b->flags = MPOOL_PINNED; + +#ifdef STATISTICS + ++mp->pageread; +#endif + /* Read in the contents. */ + off = mp->pagesize * pgno; + if (lseek(mp->fd, off, SEEK_SET) != off) + return (NULL); + if ((nr = read(mp->fd, b->page, mp->pagesize)) != mp->pagesize) { + if (nr >= 0) + errno = EFTYPE; + return (NULL); + } + if (mp->pgin) + (mp->pgin)(mp->pgcookie, b->pgno, b->page); + + inshash(b, b->pgno); + inschain(b, &mp->lru); +#ifdef STATISTICS + ++mp->pageget; +#endif + return (b->page); +} + +/* + * MPOOL_PUT -- return a page to the pool + * + * Parameters: + * mp: mpool cookie + * page: page pointer + * pgno: page number + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +mpool_put(mp, page, flags) + MPOOL *mp; + void *page; + u_int flags; +{ + BKT *baddr; +#ifdef DEBUG + BKT *b; +#endif + +#ifdef STATISTICS + ++mp->pageput; +#endif + baddr = (BKT *)((char *)page - sizeof(BKT)); +#ifdef DEBUG + if (!(baddr->flags & MPOOL_PINNED)) + __mpoolerr("mpool_put: page %d not pinned", b->pgno); + for (b = mp->lru.cnext; b != (BKT *)&mp->lru; b = b->cnext) { + if (b == (BKT *)&mp->lru) + __mpoolerr("mpool_put: %0x: bad address", baddr); + if (b == baddr) + break; + } +#endif + baddr->flags &= ~MPOOL_PINNED; + baddr->flags |= flags & MPOOL_DIRTY; + return (RET_SUCCESS); +} + +/* + * MPOOL_CLOSE -- close the buffer pool + * + * Parameters: + * mp: mpool cookie + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +mpool_close(mp) + MPOOL *mp; +{ + BKT *b, *next; + + /* Free up any space allocated to the lru pages. */ + for (b = mp->lru.cprev; b != (BKT *)&mp->lru; b = next) { + next = b->cprev; + free(b); + } + free(mp); + return (RET_SUCCESS); +} + +/* + * MPOOL_SYNC -- sync the file to disk. + * + * Parameters: + * mp: mpool cookie + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +mpool_sync(mp) + MPOOL *mp; +{ + BKT *b; + + for (b = mp->lru.cprev; b != (BKT *)&mp->lru; b = b->cprev) + if (b->flags & MPOOL_DIRTY && mpool_write(mp, b) == RET_ERROR) + return (RET_ERROR); + return (fsync(mp->fd) ? RET_ERROR : RET_SUCCESS); +} + +/* + * MPOOL_BKT -- get/create a BKT from the cache + * + * Parameters: + * mp: mpool cookie + * + * Returns: + * NULL on failure and a pointer to the BKT on success + */ +static BKT * +mpool_bkt(mp) + MPOOL *mp; +{ + BKT *b; + + if (mp->curcache < mp->maxcache) + goto new; + + /* + * If the cache is maxxed out, search the lru list for a buffer we + * can flush. If we find one, write it if necessary and take it off + * any lists. If we don't find anything we grow the cache anyway. + * The cache never shrinks. + */ + for (b = mp->lru.cprev; b != (BKT *)&mp->lru; b = b->cprev) + if (!(b->flags & MPOOL_PINNED)) { + if (b->flags & MPOOL_DIRTY && + mpool_write(mp, b) == RET_ERROR) + return (NULL); + rmhash(b); + rmchain(b); +#ifdef STATISTICS + ++mp->pageflush; +#endif +#ifdef DEBUG + { + void *spage; + spage = b->page; + memset(b, 0xff, sizeof(BKT) + mp->pagesize); + b->page = spage; + } +#endif + return (b); + } + +new: if ((b = (BKT *)malloc(sizeof(BKT) + mp->pagesize)) == NULL) + return (NULL); +#ifdef STATISTICS + ++mp->pagealloc; +#endif +#ifdef DEBUG + memset(b, 0xff, sizeof(BKT) + mp->pagesize); +#endif + b->page = (char *)b + sizeof(BKT); + ++mp->curcache; + return (b); +} + +/* + * MPOOL_WRITE -- sync a page to disk + * + * Parameters: + * mp: mpool cookie + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +static int +mpool_write(mp, b) + MPOOL *mp; + BKT *b; +{ + off_t off; + + if (mp->pgout) + (mp->pgout)(mp->pgcookie, b->pgno, b->page); + +#ifdef STATISTICS + ++mp->pagewrite; +#endif + off = mp->pagesize * b->pgno; + if (lseek(mp->fd, off, SEEK_SET) != off) + return (RET_ERROR); + if (write(mp->fd, b->page, mp->pagesize) != mp->pagesize) + return (RET_ERROR); + b->flags &= ~MPOOL_DIRTY; + return (RET_SUCCESS); +} + +/* + * MPOOL_LOOK -- lookup a page + * + * Parameters: + * mp: mpool cookie + * pgno: page number + * + * Returns: + * NULL on failure and a pointer to the BKT on success + */ +static BKT * +mpool_look(mp, pgno) + MPOOL *mp; + pgno_t pgno; +{ + register BKT *b; + register BKTHDR *tb; + + /* XXX + * If find the buffer, put it first on the hash chain so can + * find it again quickly. + */ + tb = &mp->hashtable[HASHKEY(pgno)]; + for (b = tb->hnext; b != (BKT *)tb; b = b->hnext) + if (b->pgno == pgno) { +#ifdef STATISTICS + ++mp->cachehit; +#endif + return (b); + } +#ifdef STATISTICS + ++mp->cachemiss; +#endif + return (NULL); +} + +#ifdef STATISTICS +/* + * MPOOL_STAT -- cache statistics + * + * Parameters: + * mp: mpool cookie + */ +void +mpool_stat(mp) + MPOOL *mp; +{ + BKT *b; + int cnt; + char *sep; + + (void)fprintf(stderr, "%lu pages in the file\n", mp->npages); + (void)fprintf(stderr, + "page size %lu, cacheing %lu pages of %lu page max cache\n", + mp->pagesize, mp->curcache, mp->maxcache); + (void)fprintf(stderr, "%lu page puts, %lu page gets, %lu page new\n", + mp->pageput, mp->pageget, mp->pagenew); + (void)fprintf(stderr, "%lu page allocs, %lu page flushes\n", + mp->pagealloc, mp->pageflush); + if (mp->cachehit + mp->cachemiss) + (void)fprintf(stderr, + "%.0f%% cache hit rate (%lu hits, %lu misses)\n", + ((double)mp->cachehit / (mp->cachehit + mp->cachemiss)) + * 100, mp->cachehit, mp->cachemiss); + (void)fprintf(stderr, "%lu page reads, %lu page writes\n", + mp->pageread, mp->pagewrite); + + sep = ""; + cnt = 0; + for (b = mp->lru.cnext; b != (BKT *)&mp->lru; b = b->cnext) { + (void)fprintf(stderr, "%s%d", sep, b->pgno); + if (b->flags & MPOOL_DIRTY) + (void)fprintf(stderr, "d"); + if (b->flags & MPOOL_PINNED) + (void)fprintf(stderr, "P"); + if (++cnt == 10) { + sep = "\n"; + cnt = 0; + } else + sep = ", "; + + } + (void)fprintf(stderr, "\n"); +} +#endif + +#ifdef DEBUG +#if __STDC__ +#include <stdarg.h> +#else +#include <varargs.h> +#endif + +static void +#if __STDC__ +__mpoolerr(const char *fmt, ...) +#else +__mpoolerr(fmt, va_alist) + char *fmt; + va_dcl +#endif +{ + va_list ap; +#if __STDC__ + va_start(ap, fmt); +#else + va_start(ap); +#endif + (void)vfprintf(stderr, fmt, ap); + va_end(ap); + (void)fprintf(stderr, "\n"); + abort(); + /* NOTREACHED */ +} +#endif diff --git a/lib/libc/db/recno/Makefile.inc b/lib/libc/db/recno/Makefile.inc new file mode 100644 index 00000000000..c0fd31125a1 --- /dev/null +++ b/lib/libc/db/recno/Makefile.inc @@ -0,0 +1,7 @@ +# $NetBSD: Makefile.inc,v 1.4 1995/02/27 13:24:22 cgd Exp $ +# @(#)Makefile.inc 8.1 (Berkeley) 6/4/93 + +.PATH: ${.CURDIR}/db/recno + +SRCS+= rec_close.c rec_delete.c rec_get.c rec_open.c rec_put.c rec_search.c \ + rec_seq.c rec_utils.c diff --git a/lib/libc/db/recno/extern.h b/lib/libc/db/recno/extern.h new file mode 100644 index 00000000000..5ec5d2f220c --- /dev/null +++ b/lib/libc/db/recno/extern.h @@ -0,0 +1,56 @@ +/* $NetBSD: extern.h,v 1.4 1995/02/27 13:24:30 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)extern.h 8.3 (Berkeley) 6/4/94 + */ + +#include "../btree/extern.h" + +int __rec_close __P((DB *)); +int __rec_delete __P((const DB *, const DBT *, u_int)); +int __rec_dleaf __P((BTREE *, PAGE *, u_int32_t)); +int __rec_fd __P((const DB *)); +int __rec_fmap __P((BTREE *, recno_t)); +int __rec_fout __P((BTREE *)); +int __rec_fpipe __P((BTREE *, recno_t)); +int __rec_get __P((const DB *, const DBT *, DBT *, u_int)); +int __rec_iput __P((BTREE *, recno_t, const DBT *, u_int)); +int __rec_put __P((const DB *dbp, DBT *, const DBT *, u_int)); +int __rec_ret __P((BTREE *, EPG *, recno_t, DBT *, DBT *)); +EPG *__rec_search __P((BTREE *, recno_t, enum SRCHOP)); +int __rec_seq __P((const DB *, DBT *, DBT *, u_int)); +int __rec_sync __P((const DB *, u_int)); +int __rec_vmap __P((BTREE *, recno_t)); +int __rec_vout __P((BTREE *)); +int __rec_vpipe __P((BTREE *, recno_t)); diff --git a/lib/libc/db/recno/rec_close.c b/lib/libc/db/recno/rec_close.c new file mode 100644 index 00000000000..7c4dc283525 --- /dev/null +++ b/lib/libc/db/recno/rec_close.c @@ -0,0 +1,169 @@ +/* $NetBSD: rec_close.c,v 1.6 1995/02/27 13:24:37 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_close.c 8.3 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: rec_close.c,v 1.6 1995/02/27 13:24:37 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> +#include <sys/uio.h> +#include <sys/mman.h> + +#include <errno.h> +#include <limits.h> +#include <stdio.h> +#include <unistd.h> + +#include <db.h> +#include "recno.h" + +/* + * __REC_CLOSE -- Close a recno tree. + * + * Parameters: + * dbp: pointer to access method + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__rec_close(dbp) + DB *dbp; +{ + BTREE *t; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + if (__rec_sync(dbp, 0) == RET_ERROR) + return (RET_ERROR); + + /* Committed to closing. */ + status = RET_SUCCESS; + if (ISSET(t, R_MEMMAPPED) && munmap(t->bt_smap, t->bt_msize)) + status = RET_ERROR; + + if (!ISSET(t, R_INMEM)) + if (ISSET(t, R_CLOSEFP)) { + if (fclose(t->bt_rfp)) + status = RET_ERROR; + } else + if (close(t->bt_rfd)) + status = RET_ERROR; + + if (__bt_close(dbp) == RET_ERROR) + status = RET_ERROR; + + return (status); +} + +/* + * __REC_SYNC -- sync the recno tree to disk. + * + * Parameters: + * dbp: pointer to access method + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +int +__rec_sync(dbp, flags) + const DB *dbp; + u_int flags; +{ + struct iovec iov[2]; + BTREE *t; + DBT data, key; + off_t off; + recno_t scursor, trec; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + if (flags == R_RECNOSYNC) + return (__bt_sync(dbp, 0)); + + if (ISSET(t, R_RDONLY | R_INMEM) || !ISSET(t, R_MODIFIED)) + return (RET_SUCCESS); + + /* Read any remaining records into the tree. */ + if (!ISSET(t, R_EOF) && t->bt_irec(t, MAX_REC_NUMBER) == RET_ERROR) + return (RET_ERROR); + + /* Rewind the file descriptor. */ + if (lseek(t->bt_rfd, (off_t)0, SEEK_SET) != 0) + return (RET_ERROR); + + iov[1].iov_base = "\n"; + iov[1].iov_len = 1; + scursor = t->bt_rcursor; + + key.size = sizeof(recno_t); + key.data = &trec; + + status = (dbp->seq)(dbp, &key, &data, R_FIRST); + while (status == RET_SUCCESS) { + iov[0].iov_base = data.data; + iov[0].iov_len = data.size; + if (writev(t->bt_rfd, iov, 2) != data.size + 1) + return (RET_ERROR); + status = (dbp->seq)(dbp, &key, &data, R_NEXT); + } + t->bt_rcursor = scursor; + if (status == RET_ERROR) + return (RET_ERROR); + if ((off = lseek(t->bt_rfd, (off_t)0, SEEK_CUR)) == -1) + return (RET_ERROR); + if (ftruncate(t->bt_rfd, off)) + return (RET_ERROR); + CLR(t, R_MODIFIED); + return (RET_SUCCESS); +} diff --git a/lib/libc/db/recno/rec_delete.c b/lib/libc/db/recno/rec_delete.c new file mode 100644 index 00000000000..6f438a35cd9 --- /dev/null +++ b/lib/libc/db/recno/rec_delete.c @@ -0,0 +1,203 @@ +/* $NetBSD: rec_delete.c,v 1.7 1995/02/27 13:24:48 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_delete.c 8.6 (Berkeley) 6/4/94"; +#else +static char rcsid[] = "$NetBSD: rec_delete.c,v 1.7 1995/02/27 13:24:48 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stdio.h> +#include <string.h> + +#include <db.h> +#include "recno.h" + +static int rec_rdelete __P((BTREE *, recno_t)); + +/* + * __REC_DELETE -- Delete the item(s) referenced by a key. + * + * Parameters: + * dbp: pointer to access method + * key: key to delete + * flags: R_CURSOR if deleting what the cursor references + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found. + */ +int +__rec_delete(dbp, key, flags) + const DB *dbp; + const DBT *key; + u_int flags; +{ + BTREE *t; + recno_t nrec; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + switch(flags) { + case 0: + if ((nrec = *(recno_t *)key->data) == 0) + goto einval; + if (nrec > t->bt_nrecs) + return (RET_SPECIAL); + --nrec; + status = rec_rdelete(t, nrec); + break; + case R_CURSOR: + if (!ISSET(t, B_SEQINIT)) + goto einval; + if (t->bt_nrecs == 0) + return (RET_SPECIAL); + status = rec_rdelete(t, t->bt_rcursor - 1); + if (status == RET_SUCCESS) + --t->bt_rcursor; + break; + default: +einval: errno = EINVAL; + return (RET_ERROR); + } + + if (status == RET_SUCCESS) + SET(t, B_MODIFIED | R_MODIFIED); + return (status); +} + +/* + * REC_RDELETE -- Delete the data matching the specified key. + * + * Parameters: + * tree: tree + * nrec: record to delete + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found. + */ +static int +rec_rdelete(t, nrec) + BTREE *t; + recno_t nrec; +{ + EPG *e; + PAGE *h; + int status; + + /* Find the record; __rec_search pins the page. */ + if ((e = __rec_search(t, nrec, SDELETE)) == NULL) + return (RET_ERROR); + + /* Delete the record. */ + h = e->page; + status = __rec_dleaf(t, h, e->index); + if (status != RET_SUCCESS) { + mpool_put(t->bt_mp, h, 0); + return (status); + } + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + return (RET_SUCCESS); +} + +/* + * __REC_DLEAF -- Delete a single record from a recno leaf page. + * + * Parameters: + * t: tree + * index: index on current page to delete + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +int +__rec_dleaf(t, h, index) + BTREE *t; + PAGE *h; + u_int32_t index; +{ + RLEAF *rl; + indx_t *ip, cnt, offset; + u_int32_t nbytes; + char *from; + void *to; + + /* + * Delete a record from a recno leaf page. Internal records are never + * deleted from internal pages, regardless of the records that caused + * them to be added being deleted. Pages made empty by deletion are + * not reclaimed. They are, however, made available for reuse. + * + * Pack the remaining entries at the end of the page, shift the indices + * down, overwriting the deleted record and its index. If the record + * uses overflow pages, make them available for reuse. + */ + to = rl = GETRLEAF(h, index); + if (rl->flags & P_BIGDATA && __ovfl_delete(t, rl->bytes) == RET_ERROR) + return (RET_ERROR); + nbytes = NRLEAF(rl); + + /* + * Compress the key/data pairs. Compress and adjust the [BR]LEAF + * offsets. Reset the headers. + */ + from = (char *)h + h->upper; + memmove(from + nbytes, from, (char *)to - from); + h->upper += nbytes; + + offset = h->linp[index]; + for (cnt = &h->linp[index] - (ip = &h->linp[0]); cnt--; ++ip) + if (ip[0] < offset) + ip[0] += nbytes; + for (cnt = &h->linp[NEXTINDEX(h)] - ip; --cnt; ++ip) + ip[0] = ip[1] < offset ? ip[1] + nbytes : ip[1]; + h->lower -= sizeof(indx_t); + --t->bt_nrecs; + return (RET_SUCCESS); +} diff --git a/lib/libc/db/recno/rec_get.c b/lib/libc/db/recno/rec_get.c new file mode 100644 index 00000000000..bc25104e552 --- /dev/null +++ b/lib/libc/db/recno/rec_get.c @@ -0,0 +1,308 @@ +/* $NetBSD: rec_get.c,v 1.7 1995/02/27 13:24:57 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_get.c 8.5 (Berkeley) 6/20/94"; +#else +static char rcsid[] = "$NetBSD: rec_get.c,v 1.7 1995/02/27 13:24:57 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stddef.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#include <db.h> +#include "recno.h" + +/* + * __REC_GET -- Get a record from the btree. + * + * Parameters: + * dbp: pointer to access method + * key: key to find + * data: data to return + * flag: currently unused + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key not found. + */ +int +__rec_get(dbp, key, data, flags) + const DB *dbp; + const DBT *key; + DBT *data; + u_int flags; +{ + BTREE *t; + EPG *e; + recno_t nrec; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* Get currently doesn't take any flags, and keys of 0 are illegal. */ + if (flags || (nrec = *(recno_t *)key->data) == 0) { + errno = EINVAL; + return (RET_ERROR); + } + + /* + * If we haven't seen this record yet, try to find it in the + * original file. + */ + if (nrec > t->bt_nrecs) { + if (ISSET(t, R_EOF | R_INMEM)) + return (RET_SPECIAL); + if ((status = t->bt_irec(t, nrec)) != RET_SUCCESS) + return (status); + } + + --nrec; + if ((e = __rec_search(t, nrec, SEARCH)) == NULL) + return (RET_ERROR); + + status = __rec_ret(t, e, 0, NULL, data); + if (ISSET(t, B_DB_LOCK)) + mpool_put(t->bt_mp, e->page, 0); + else + t->bt_pinned = e->page; + return (status); +} + +/* + * __REC_FPIPE -- Get fixed length records from a pipe. + * + * Parameters: + * t: tree + * cnt: records to read + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__rec_fpipe(t, top) + BTREE *t; + recno_t top; +{ + DBT data; + recno_t nrec; + size_t len; + int ch; + char *p; + + if (t->bt_dbufsz < t->bt_reclen) { + t->bt_dbuf = (char *)(t->bt_dbuf == NULL ? + malloc(t->bt_reclen) : realloc(t->bt_dbuf, t->bt_reclen)); + if (t->bt_dbuf == NULL) + return (RET_ERROR); + t->bt_dbufsz = t->bt_reclen; + } + data.data = t->bt_dbuf; + data.size = t->bt_reclen; + + for (nrec = t->bt_nrecs; nrec < top; ++nrec) { + len = t->bt_reclen; + for (p = t->bt_dbuf;; *p++ = ch) + if ((ch = getc(t->bt_rfp)) == EOF || !len--) { + if (__rec_iput(t, nrec, &data, 0) + != RET_SUCCESS) + return (RET_ERROR); + break; + } + if (ch == EOF) + break; + } + if (nrec < top) { + SET(t, R_EOF); + return (RET_SPECIAL); + } + return (RET_SUCCESS); +} + +/* + * __REC_VPIPE -- Get variable length records from a pipe. + * + * Parameters: + * t: tree + * cnt: records to read + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__rec_vpipe(t, top) + BTREE *t; + recno_t top; +{ + DBT data; + recno_t nrec; + indx_t len; + size_t sz; + int bval, ch; + char *p; + + bval = t->bt_bval; + for (nrec = t->bt_nrecs; nrec < top; ++nrec) { + for (p = t->bt_dbuf, sz = t->bt_dbufsz;; *p++ = ch, --sz) { + if ((ch = getc(t->bt_rfp)) == EOF || ch == bval) { + data.data = t->bt_dbuf; + data.size = p - t->bt_dbuf; + if (ch == EOF && data.size == 0) + break; + if (__rec_iput(t, nrec, &data, 0) + != RET_SUCCESS) + return (RET_ERROR); + break; + } + if (sz == 0) { + len = p - t->bt_dbuf; + t->bt_dbufsz += (sz = 256); + t->bt_dbuf = (char *)(t->bt_dbuf == NULL ? + malloc(t->bt_dbufsz) : + realloc(t->bt_dbuf, t->bt_dbufsz)); + if (t->bt_dbuf == NULL) + return (RET_ERROR); + p = t->bt_dbuf + len; + } + } + if (ch == EOF) + break; + } + if (nrec < top) { + SET(t, R_EOF); + return (RET_SPECIAL); + } + return (RET_SUCCESS); +} + +/* + * __REC_FMAP -- Get fixed length records from a file. + * + * Parameters: + * t: tree + * cnt: records to read + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__rec_fmap(t, top) + BTREE *t; + recno_t top; +{ + DBT data; + recno_t nrec; + caddr_t sp, ep; + size_t len; + char *p; + + if (t->bt_dbufsz < t->bt_reclen) { + t->bt_dbuf = (char *)(t->bt_dbuf == NULL ? + malloc(t->bt_reclen) : realloc(t->bt_dbuf, t->bt_reclen)); + if (t->bt_dbuf == NULL) + return (RET_ERROR); + t->bt_dbufsz = t->bt_reclen; + } + data.data = t->bt_dbuf; + data.size = t->bt_reclen; + + sp = t->bt_cmap; + ep = t->bt_emap; + for (nrec = t->bt_nrecs; nrec < top; ++nrec) { + if (sp >= ep) { + SET(t, R_EOF); + return (RET_SPECIAL); + } + len = t->bt_reclen; + for (p = t->bt_dbuf; sp < ep && len--; *p++ = *sp++); + memset(p, t->bt_bval, len); + if (__rec_iput(t, nrec, &data, 0) != RET_SUCCESS) + return (RET_ERROR); + } + t->bt_cmap = sp; + return (RET_SUCCESS); +} + +/* + * __REC_VMAP -- Get variable length records from a file. + * + * Parameters: + * t: tree + * cnt: records to read + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__rec_vmap(t, top) + BTREE *t; + recno_t top; +{ + DBT data; + caddr_t sp, ep; + recno_t nrec; + int bval; + + sp = t->bt_cmap; + ep = t->bt_emap; + bval = t->bt_bval; + + for (nrec = t->bt_nrecs; nrec < top; ++nrec) { + if (sp >= ep) { + SET(t, R_EOF); + return (RET_SPECIAL); + } + for (data.data = sp; sp < ep && *sp != bval; ++sp); + data.size = sp - (caddr_t)data.data; + if (__rec_iput(t, nrec, &data, 0) != RET_SUCCESS) + return (RET_ERROR); + ++sp; + } + t->bt_cmap = sp; + return (RET_SUCCESS); +} diff --git a/lib/libc/db/recno/rec_open.c b/lib/libc/db/recno/rec_open.c new file mode 100644 index 00000000000..e67a7b6f000 --- /dev/null +++ b/lib/libc/db/recno/rec_open.c @@ -0,0 +1,236 @@ +/* $NetBSD: rec_open.c,v 1.6 1995/02/27 13:25:05 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * This code is derived from software contributed to Berkeley by + * Mike Olson. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_open.c 8.6 (Berkeley) 2/22/94"; +#else +static char rcsid[] = "$NetBSD: rec_open.c,v 1.6 1995/02/27 13:25:05 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> +#include <sys/mman.h> +#include <sys/stat.h> + +#include <errno.h> +#include <fcntl.h> +#include <limits.h> +#include <stddef.h> +#include <stdio.h> +#include <unistd.h> + +#include <db.h> +#include "recno.h" + +DB * +__rec_open(fname, flags, mode, openinfo, dflags) + const char *fname; + int flags, mode, dflags; + const RECNOINFO *openinfo; +{ + BTREE *t; + BTREEINFO btopeninfo; + DB *dbp; + PAGE *h; + struct stat sb; + int rfd, sverrno; + + /* Open the user's file -- if this fails, we're done. */ + if (fname != NULL && (rfd = open(fname, flags, mode)) < 0) + return (NULL); + + /* Create a btree in memory (backed by disk). */ + dbp = NULL; + if (openinfo) { + if (openinfo->flags & ~(R_FIXEDLEN | R_NOKEY | R_SNAPSHOT)) + goto einval; + btopeninfo.flags = 0; + btopeninfo.cachesize = openinfo->cachesize; + btopeninfo.maxkeypage = 0; + btopeninfo.minkeypage = 0; + btopeninfo.psize = openinfo->psize; + btopeninfo.compare = NULL; + btopeninfo.prefix = NULL; + btopeninfo.lorder = openinfo->lorder; + dbp = __bt_open(openinfo->bfname, + O_RDWR, S_IRUSR | S_IWUSR, &btopeninfo, dflags); + } else + dbp = __bt_open(NULL, O_RDWR, S_IRUSR | S_IWUSR, NULL, dflags); + if (dbp == NULL) + goto err; + + /* + * Some fields in the tree structure are recno specific. Fill them + * in and make the btree structure look like a recno structure. We + * don't change the bt_ovflsize value, it's close enough and slightly + * bigger. + */ + t = dbp->internal; + if (openinfo) { + if (openinfo->flags & R_FIXEDLEN) { + SET(t, R_FIXLEN); + t->bt_reclen = openinfo->reclen; + if (t->bt_reclen == 0) + goto einval; + } + t->bt_bval = openinfo->bval; + } else + t->bt_bval = '\n'; + + SET(t, R_RECNO); + if (fname == NULL) + SET(t, R_EOF | R_INMEM); + else + t->bt_rfd = rfd; + t->bt_rcursor = 0; + + if (fname != NULL) { + /* + * In 4.4BSD, stat(2) returns true for ISSOCK on pipes. + * Unfortunately, that's not portable, so we use lseek + * and check the errno values. + */ + errno = 0; + if (lseek(rfd, (off_t)0, SEEK_CUR) == -1 && errno == ESPIPE) { + switch (flags & O_ACCMODE) { + case O_RDONLY: + SET(t, R_RDONLY); + break; + default: + goto einval; + } +slow: if ((t->bt_rfp = fdopen(rfd, "r")) == NULL) + goto err; + SET(t, R_CLOSEFP); + t->bt_irec = + ISSET(t, R_FIXLEN) ? __rec_fpipe : __rec_vpipe; + } else { + switch (flags & O_ACCMODE) { + case O_RDONLY: + SET(t, R_RDONLY); + break; + case O_RDWR: + break; + default: + goto einval; + } + + if (fstat(rfd, &sb)) + goto err; + /* + * Kluge -- we'd like to test to see if the file is too + * big to mmap. Since, we don't know what size or type + * off_t's or size_t's are, what the largest unsigned + * integral type is, or what random insanity the local + * C compiler will perpetrate, doing the comparison in + * a portable way is flatly impossible. Hope that mmap + * fails if the file is too large. + */ + if (sb.st_size == 0) + SET(t, R_EOF); + else { + t->bt_msize = sb.st_size; + if ((t->bt_smap = mmap(NULL, t->bt_msize, + PROT_READ, MAP_PRIVATE, rfd, + (off_t)0)) == (caddr_t)-1) + goto slow; + t->bt_cmap = t->bt_smap; + t->bt_emap = t->bt_smap + sb.st_size; + t->bt_irec = ISSET(t, R_FIXLEN) ? + __rec_fmap : __rec_vmap; + SET(t, R_MEMMAPPED); + } + } + } + + /* Use the recno routines. */ + dbp->close = __rec_close; + dbp->del = __rec_delete; + dbp->fd = __rec_fd; + dbp->get = __rec_get; + dbp->put = __rec_put; + dbp->seq = __rec_seq; + dbp->sync = __rec_sync; + + /* If the root page was created, reset the flags. */ + if ((h = mpool_get(t->bt_mp, P_ROOT, 0)) == NULL) + goto err; + if ((h->flags & P_TYPE) == P_BLEAF) { + h->flags = h->flags & ~P_TYPE | P_RLEAF; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + } else + mpool_put(t->bt_mp, h, 0); + + if (openinfo && openinfo->flags & R_SNAPSHOT && + !ISSET(t, R_EOF | R_INMEM) && + t->bt_irec(t, MAX_REC_NUMBER) == RET_ERROR) + goto err; + return (dbp); + +einval: errno = EINVAL; +err: sverrno = errno; + if (dbp != NULL) + (void)__bt_close(dbp); + if (fname != NULL) + (void)close(rfd); + errno = sverrno; + return (NULL); +} + +int +__rec_fd(dbp) + const DB *dbp; +{ + BTREE *t; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + /* In-memory database can't have a file descriptor. */ + if (ISSET(t, R_INMEM)) { + errno = ENOENT; + return (-1); + } + return (t->bt_rfd); +} diff --git a/lib/libc/db/recno/rec_put.c b/lib/libc/db/recno/rec_put.c new file mode 100644 index 00000000000..77d7079fb73 --- /dev/null +++ b/lib/libc/db/recno/rec_put.c @@ -0,0 +1,259 @@ +/* $NetBSD: rec_put.c,v 1.7 1995/02/27 13:25:13 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_put.c 8.4 (Berkeley) 5/31/94"; +#else +static char rcsid[] = "$NetBSD: rec_put.c,v 1.7 1995/02/27 13:25:13 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "recno.h" + +/* + * __REC_PUT -- Add a recno item to the tree. + * + * Parameters: + * dbp: pointer to access method + * key: key + * data: data + * flag: R_CURSOR, R_IAFTER, R_IBEFORE, R_NOOVERWRITE + * + * Returns: + * RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is + * already in the tree and R_NOOVERWRITE specified. + */ +int +__rec_put(dbp, key, data, flags) + const DB *dbp; + DBT *key; + const DBT *data; + u_int flags; +{ + BTREE *t; + DBT tdata; + recno_t nrec; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + switch (flags) { + case R_CURSOR: + if (!ISSET(t, B_SEQINIT)) + goto einval; + nrec = t->bt_rcursor; + break; + case R_SETCURSOR: + if ((nrec = *(recno_t *)key->data) == 0) + goto einval; + break; + case R_IAFTER: + if ((nrec = *(recno_t *)key->data) == 0) { + nrec = 1; + flags = R_IBEFORE; + } + break; + case 0: + case R_IBEFORE: + if ((nrec = *(recno_t *)key->data) == 0) + goto einval; + break; + case R_NOOVERWRITE: + if ((nrec = *(recno_t *)key->data) == 0) + goto einval; + if (nrec <= t->bt_nrecs) + return (RET_SPECIAL); + break; + default: +einval: errno = EINVAL; + return (RET_ERROR); + } + + /* + * Make sure that records up to and including the put record are + * already in the database. If skipping records, create empty ones. + */ + if (nrec > t->bt_nrecs) { + if (!ISSET(t, R_EOF | R_INMEM) && + t->bt_irec(t, nrec) == RET_ERROR) + return (RET_ERROR); + if (nrec > t->bt_nrecs + 1) { + if (ISSET(t, R_FIXLEN)) { + if ((tdata.data = + (void *)malloc(t->bt_reclen)) == NULL) + return (RET_ERROR); + tdata.size = t->bt_reclen; + memset(tdata.data, t->bt_bval, tdata.size); + } else { + tdata.data = NULL; + tdata.size = 0; + } + while (nrec > t->bt_nrecs + 1) + if (__rec_iput(t, + t->bt_nrecs, &tdata, 0) != RET_SUCCESS) + return (RET_ERROR); + if (ISSET(t, R_FIXLEN)) + free(tdata.data); + } + } + + if ((status = __rec_iput(t, nrec - 1, data, flags)) != RET_SUCCESS) + return (status); + + if (flags == R_SETCURSOR) + t->bt_rcursor = nrec; + + SET(t, R_MODIFIED); + return (__rec_ret(t, NULL, nrec, key, NULL)); +} + +/* + * __REC_IPUT -- Add a recno item to the tree. + * + * Parameters: + * t: tree + * nrec: record number + * data: data + * + * Returns: + * RET_ERROR, RET_SUCCESS + */ +int +__rec_iput(t, nrec, data, flags) + BTREE *t; + recno_t nrec; + const DBT *data; + u_int flags; +{ + DBT tdata; + EPG *e; + PAGE *h; + indx_t index, nxtindex; + pgno_t pg; + u_int32_t nbytes; + int dflags, status; + char *dest, db[NOVFLSIZE]; + + /* + * If the data won't fit on a page, store it on indirect pages. + * + * XXX + * If the insert fails later on, these pages aren't recovered. + */ + if (data->size > t->bt_ovflsize) { + if (__ovfl_put(t, data, &pg) == RET_ERROR) + return (RET_ERROR); + tdata.data = db; + tdata.size = NOVFLSIZE; + *(pgno_t *)db = pg; + *(u_int32_t *)(db + sizeof(pgno_t)) = data->size; + dflags = P_BIGDATA; + data = &tdata; + } else + dflags = 0; + + /* __rec_search pins the returned page. */ + if ((e = __rec_search(t, nrec, + nrec > t->bt_nrecs || flags == R_IAFTER || flags == R_IBEFORE ? + SINSERT : SEARCH)) == NULL) + return (RET_ERROR); + + h = e->page; + index = e->index; + + /* + * Add the specified key/data pair to the tree. The R_IAFTER and + * R_IBEFORE flags insert the key after/before the specified key. + * + * Pages are split as required. + */ + switch (flags) { + case R_IAFTER: + ++index; + break; + case R_IBEFORE: + break; + default: + if (nrec < t->bt_nrecs && + __rec_dleaf(t, h, index) == RET_ERROR) { + mpool_put(t->bt_mp, h, 0); + return (RET_ERROR); + } + break; + } + + /* + * If not enough room, split the page. The split code will insert + * the key and data and unpin the current page. If inserting into + * the offset array, shift the pointers up. + */ + nbytes = NRLEAFDBT(data->size); + if (h->upper - h->lower < nbytes + sizeof(indx_t)) { + status = __bt_split(t, h, NULL, data, dflags, nbytes, index); + if (status == RET_SUCCESS) + ++t->bt_nrecs; + return (status); + } + + if (index < (nxtindex = NEXTINDEX(h))) + memmove(h->linp + index + 1, h->linp + index, + (nxtindex - index) * sizeof(indx_t)); + h->lower += sizeof(indx_t); + + h->linp[index] = h->upper -= nbytes; + dest = (char *)h + h->upper; + WR_RLEAF(dest, data, dflags); + + ++t->bt_nrecs; + SET(t, B_MODIFIED); + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + + return (RET_SUCCESS); +} diff --git a/lib/libc/db/recno/rec_search.c b/lib/libc/db/recno/rec_search.c new file mode 100644 index 00000000000..208328d9ddd --- /dev/null +++ b/lib/libc/db/recno/rec_search.c @@ -0,0 +1,133 @@ +/* $NetBSD: rec_search.c,v 1.6 1995/02/27 13:25:17 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_search.c 8.3 (Berkeley) 2/21/94"; +#else +static char rcsid[] = "$NetBSD: rec_search.c,v 1.6 1995/02/27 13:25:17 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <stdio.h> + +#include <db.h> +#include "recno.h" + +/* + * __REC_SEARCH -- Search a btree for a key. + * + * Parameters: + * t: tree to search + * recno: key to find + * op: search operation + * + * Returns: + * EPG for matching record, if any, or the EPG for the location of the + * key, if it were inserted into the tree. + * + * Returns: + * The EPG for matching record, if any, or the EPG for the location + * of the key, if it were inserted into the tree, is entered into + * the bt_cur field of the tree. A pointer to the field is returned. + */ +EPG * +__rec_search(t, recno, op) + BTREE *t; + recno_t recno; + enum SRCHOP op; +{ + register indx_t index; + register PAGE *h; + EPGNO *parent; + RINTERNAL *r; + pgno_t pg; + indx_t top; + recno_t total; + int sverrno; + + BT_CLR(t); + for (pg = P_ROOT, total = 0;;) { + if ((h = mpool_get(t->bt_mp, pg, 0)) == NULL) + goto err; + if (h->flags & P_RLEAF) { + t->bt_cur.page = h; + t->bt_cur.index = recno - total; + return (&t->bt_cur); + } + for (index = 0, top = NEXTINDEX(h);;) { + r = GETRINTERNAL(h, index); + if (++index == top || total + r->nrecs > recno) + break; + total += r->nrecs; + } + + if (__bt_push(t, pg, index - 1) == RET_ERROR) + return (NULL); + + pg = r->pgno; + switch (op) { + case SDELETE: + --GETRINTERNAL(h, (index - 1))->nrecs; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + break; + case SINSERT: + ++GETRINTERNAL(h, (index - 1))->nrecs; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + break; + case SEARCH: + mpool_put(t->bt_mp, h, 0); + break; + } + + } + /* Try and recover the tree. */ +err: sverrno = errno; + if (op != SEARCH) + while ((parent = BT_POP(t)) != NULL) { + if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL) + break; + if (op == SINSERT) + --GETRINTERNAL(h, parent->index)->nrecs; + else + ++GETRINTERNAL(h, parent->index)->nrecs; + mpool_put(t->bt_mp, h, MPOOL_DIRTY); + } + errno = sverrno; + return (NULL); +} diff --git a/lib/libc/db/recno/rec_seq.c b/lib/libc/db/recno/rec_seq.c new file mode 100644 index 00000000000..b91f3a27bf6 --- /dev/null +++ b/lib/libc/db/recno/rec_seq.c @@ -0,0 +1,137 @@ +/* $NetBSD: rec_seq.c,v 1.6 1995/02/27 13:25:24 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#ifndef lint +#if 0 +static char sccsid[] = "@(#)rec_seq.c 8.2 (Berkeley) 9/7/93"; +#else +static char rcsid[] = "$NetBSD: rec_seq.c,v 1.6 1995/02/27 13:25:24 cgd Exp $"; +#endif +#endif /* not lint */ + +#include <sys/types.h> + +#include <errno.h> +#include <limits.h> +#include <stdio.h> +#include <string.h> + +#include <db.h> +#include "recno.h" + +/* + * __REC_SEQ -- Recno sequential scan interface. + * + * Parameters: + * dbp: pointer to access method + * key: key for positioning and return value + * data: data return value + * flags: R_CURSOR, R_FIRST, R_LAST, R_NEXT, R_PREV. + * + * Returns: + * RET_ERROR, RET_SUCCESS or RET_SPECIAL if there's no next key. + */ +int +__rec_seq(dbp, key, data, flags) + const DB *dbp; + DBT *key, *data; + u_int flags; +{ + BTREE *t; + EPG *e; + recno_t nrec; + int status; + + t = dbp->internal; + + /* Toss any page pinned across calls. */ + if (t->bt_pinned != NULL) { + mpool_put(t->bt_mp, t->bt_pinned, 0); + t->bt_pinned = NULL; + } + + switch(flags) { + case R_CURSOR: + if ((nrec = *(recno_t *)key->data) == 0) + goto einval; + break; + case R_NEXT: + if (ISSET(t, B_SEQINIT)) { + nrec = t->bt_rcursor + 1; + break; + } + /* FALLTHROUGH */ + case R_FIRST: + nrec = 1; + break; + case R_PREV: + if (ISSET(t, B_SEQINIT)) { + if ((nrec = t->bt_rcursor - 1) == 0) + return (RET_SPECIAL); + break; + } + /* FALLTHROUGH */ + case R_LAST: + if (!ISSET(t, R_EOF | R_INMEM) && + t->bt_irec(t, MAX_REC_NUMBER) == RET_ERROR) + return (RET_ERROR); + nrec = t->bt_nrecs; + break; + default: +einval: errno = EINVAL; + return (RET_ERROR); + } + + if (t->bt_nrecs == 0 || nrec > t->bt_nrecs) { + if (!ISSET(t, R_EOF | R_INMEM) && + (status = t->bt_irec(t, nrec)) != RET_SUCCESS) + return (status); + if (t->bt_nrecs == 0 || nrec > t->bt_nrecs) + return (RET_SPECIAL); + } + + if ((e = __rec_search(t, nrec - 1, SEARCH)) == NULL) + return (RET_ERROR); + + SET(t, B_SEQINIT); + t->bt_rcursor = nrec; + + status = __rec_ret(t, e, nrec, key, data); + if (ISSET(t, B_DB_LOCK)) + mpool_put(t->bt_mp, e->page, 0); + else + t->bt_pinned = e->page; + return (status); +} diff --git a/lib/libc/db/recno/rec_utils.c b/lib/libc/db/recno/rec_utils.c new file mode 100644 index 00000000000..7aab4ac3a91 --- /dev/null +++ b/lib/libc/db/recno/rec_utils.c @@ -0,0 +1,125 @@ +/* $NetBSD: rec_utils.c,v 1.6 1995/02/27 13:25:29 cgd Exp $ */ + +/*- + * Copyright (c) 1990, 1993, 1994 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#if defined(LIBC_SCCS) && !defined(lint) +#if 0 +static char sccsid[] = "@(#)rec_utils.c 8.4 (Berkeley) 6/20/94"; +#else +static char rcsid[] = "$NetBSD: rec_utils.c,v 1.6 1995/02/27 13:25:29 cgd Exp $"; +#endif +#endif /* LIBC_SCCS and not lint */ + +#include <sys/param.h> + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> + +#include <db.h> +#include "recno.h" + +/* + * __REC_RET -- Build return data as a result of search or scan. + * + * Parameters: + * t: tree + * d: LEAF to be returned to the user. + * data: user's data structure + * + * Returns: + * RET_SUCCESS, RET_ERROR. + */ +int +__rec_ret(t, e, nrec, key, data) + BTREE *t; + EPG *e; + recno_t nrec; + DBT *key, *data; +{ + register RLEAF *rl; + register void *p; + + if (data == NULL) + goto retkey; + + rl = GETRLEAF(e->page, e->index); + + /* + * We always copy big data to make it contigous. Otherwise, we + * leave the page pinned and don't copy unless the user specified + * concurrent access. + */ + if (rl->flags & P_BIGDATA) { + if (__ovfl_get(t, rl->bytes, + &data->size, &t->bt_dbuf, &t->bt_dbufsz)) + return (RET_ERROR); + data->data = t->bt_dbuf; + } else if (ISSET(t, B_DB_LOCK)) { + /* Use +1 in case the first record retrieved is 0 length. */ + if (rl->dsize + 1 > t->bt_dbufsz) { + p = (void *)(t->bt_dbuf == NULL ? + malloc(rl->dsize + 1) : + realloc(t->bt_dbuf, rl->dsize + 1)); + if (p == NULL) + return (RET_ERROR); + t->bt_dbuf = p; + t->bt_dbufsz = rl->dsize + 1; + } + memmove(t->bt_dbuf, rl->bytes, rl->dsize); + data->size = rl->dsize; + data->data = t->bt_dbuf; + } else { + data->size = rl->dsize; + data->data = rl->bytes; + } + +retkey: if (key == NULL) + return (RET_SUCCESS); + + /* We have to copy the key, it's not on the page. */ + if (sizeof(recno_t) > t->bt_kbufsz) { + p = (void *)(t->bt_kbuf == NULL ? + malloc(sizeof(recno_t)) : + realloc(t->bt_kbuf, sizeof(recno_t))); + if (p == NULL) + return (RET_ERROR); + t->bt_kbuf = p; + t->bt_kbufsz = sizeof(recno_t); + } + memmove(t->bt_kbuf, &nrec, sizeof(recno_t)); + key->size = sizeof(recno_t); + key->data = t->bt_kbuf; + return (RET_SUCCESS); +} diff --git a/lib/libc/db/recno/recno.h b/lib/libc/db/recno/recno.h new file mode 100644 index 00000000000..5c1c716ace3 --- /dev/null +++ b/lib/libc/db/recno/recno.h @@ -0,0 +1,41 @@ +/* $NetBSD: recno.h,v 1.4 1995/02/27 13:25:35 cgd Exp $ */ + +/*- + * Copyright (c) 1991, 1993 + * The Regents of the University of California. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the University of + * California, Berkeley and its contributors. + * 4. Neither the name of the University nor the names of its contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * @(#)recno.h 8.1 (Berkeley) 6/4/93 + */ + +enum SRCHOP { SDELETE, SINSERT, SEARCH}; /* Rec_search operation. */ + +#include "../btree/btree.h" +#include "extern.h" |