Update.

[glibc.git] / db2 / db / db_dup.c

/*-
 * See the file LICENSE for redistribution information.
 *
 * Copyright (c) 1996, 1997
 *      Sleepycat Software.  All rights reserved.
 */

#include "config.h"

#ifndef lint
static const char sccsid[] = "@(#)db_dup.c      10.11 (Sleepycat) 1/8/98";
#endif /* not lint */

#ifndef NO_SYSTEM_INCLUDES
#include <sys/types.h>
#include <sys/stat.h>

#include <errno.h>
#include <fcntl.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#endif

#include "db_int.h"
#include "db_page.h"
#include "db_swap.h"
#include "btree.h"
#include "db_am.h"
#include "common_ext.h"

static int __db_addpage __P((DB *,
    PAGE **, db_indx_t *, int (*)(DB *, u_int32_t, PAGE **)));
static int __db_dsplit __P((DB *,
    PAGE **, db_indx_t *, u_int32_t, int (*)(DB *, u_int32_t, PAGE **)));

/*
 * __db_dput --
 *      Put a duplicate item onto a duplicate page at the given index.
 *
 * PUBLIC: int __db_dput __P((DB *,
 * PUBLIC:    DBT *, PAGE **, db_indx_t *, int (*)(DB *, u_int32_t, PAGE **)));
 */
int
__db_dput(dbp, dbt, pp, indxp, newfunc)
        DB *dbp;
        DBT *dbt;
        PAGE **pp;
        db_indx_t *indxp;
        int (*newfunc) __P((DB *, u_int32_t, PAGE **));
{
        BOVERFLOW bo;
        DBT *data_dbtp, hdr_dbt, *hdr_dbtp;
        PAGE *pagep;
        db_indx_t size, isize;
        db_pgno_t pgno;
        int ret;

        /*
         * We need some access method independent threshold for when we put
         * a duplicate item onto an overflow page.
         */
        if (dbt->size > 0.25 * dbp->pgsize) {
                if ((ret = __db_poff(dbp, dbt, &pgno, newfunc)) != 0)
                        return (ret);
                B_TSET(bo.type, B_OVERFLOW, 0);
                bo.tlen = dbt->size;
                bo.pgno = pgno;
                hdr_dbt.data = &bo;
                hdr_dbt.size = isize = BOVERFLOW_SIZE;
                hdr_dbtp = &hdr_dbt;
                size = BOVERFLOW_PSIZE;
                data_dbtp = NULL;
        } else {
                size = BKEYDATA_PSIZE(dbt->size);
                isize = BKEYDATA_SIZE(dbt->size);
                hdr_dbtp = NULL;
                data_dbtp = dbt;
        }

        pagep = *pp;
        if (size > P_FREESPACE(pagep)) {
                if (*indxp == NUM_ENT(*pp) && NEXT_PGNO(*pp) == PGNO_INVALID)
                        ret = __db_addpage(dbp, pp, indxp, newfunc);
                else
                        ret = __db_dsplit(dbp, pp, indxp, isize, newfunc);
                if (ret != 0)
                        /* XXX: Pages not returned to free list. */
                        return (ret);
                pagep = *pp;
        }

        /*
         * Now, pagep references the page on which to insert and indx is the
         * the location to insert.
         */
        if ((ret = __db_pitem(dbp,
            pagep, (u_int32_t)*indxp, isize, hdr_dbtp, data_dbtp)) != 0)
                return (ret);

        (void)memp_fset(dbp->mpf, pagep, DB_MPOOL_DIRTY);
        return (0);
}

/*
 * __db_drem --
 *      Remove a duplicate at the given index on the given page.
 *
 * PUBLIC: int __db_drem __P((DB *,
 * PUBLIC:    PAGE **, u_int32_t, int (*)(DB *, PAGE *)));
 */
int
__db_drem(dbp, pp, indx, freefunc)
        DB *dbp;
        PAGE **pp;
        u_int32_t indx;
        int (*freefunc) __P((DB *, PAGE *));
{
        PAGE *pagep;
        int ret;

        pagep = *pp;

        /* Check if we are freeing a big item. */
        if (B_TYPE(GET_BKEYDATA(pagep, indx)->type) == B_OVERFLOW) {
                if ((ret = __db_doff(dbp,
                    GET_BOVERFLOW(pagep, indx)->pgno, freefunc)) != 0)
                        return (ret);
                ret = __db_ditem(dbp, pagep, indx, BOVERFLOW_SIZE);
        } else
                ret = __db_ditem(dbp, pagep, indx,
                    BKEYDATA_SIZE(GET_BKEYDATA(pagep, indx)->len));
        if (ret != 0)
                return (ret);

        if (NUM_ENT(pagep) == 0) {
                /*
                 * If the page is emptied, then the page is freed and the pp
                 * parameter is set to reference the next, locked page in the
                 * duplicate chain, if one exists.  If there was no such page,
                 * then it is set to NULL.
                 *
                 * !!!
                 * __db_relink will set the dirty bit for us.
                 */
                if ((ret = __db_relink(dbp, pagep, pp, 0)) != 0)
                        return (ret);
                if ((ret = freefunc(dbp, pagep)) != 0)
                        return (ret);
        } else
                (void)memp_fset(dbp->mpf, pagep, DB_MPOOL_DIRTY);

        return (0);
}

/*
 * __db_dend --
 *      Find the last page in a set of offpage duplicates.
 *
 * PUBLIC: int __db_dend __P((DB *, db_pgno_t, PAGE **));
 */
int
__db_dend(dbp, pgno, pagep)
        DB *dbp;
        db_pgno_t pgno;
        PAGE **pagep;
{
        PAGE *h;
        int ret;

        /*
         * This implements DB_KEYLAST.  The last page is returned in pp; pgno
         * should be the page number of the first page of the duplicate chain.
         */
        for (;;) {
                if ((ret = memp_fget(dbp->mpf, &pgno, 0, &h)) != 0) {
                        (void)__db_pgerr(dbp, pgno);
                        return (ret);
                }
                if ((pgno = NEXT_PGNO(h)) == PGNO_INVALID)
                        break;
                (void)memp_fput(dbp->mpf, h, 0);
        }

        *pagep = h;
        return (0);
}

/*
 * __db_dsplit --
 *      Split a page of duplicates, calculating the split point based
 *      on an element of size "size" being added at "*indxp".
 *      On entry hp contains a pointer to the page-pointer of the original
 *      page.  On exit, it returns a pointer to the page containing "*indxp"
 *      and "indxp" has been modified to reflect the index on the new page
 *      where the element should be added.  The function returns with
 *      the page on which the insert should happen, not yet put.
 */
static int
__db_dsplit(dbp, hp, indxp, size, newfunc)
        DB *dbp;
        PAGE **hp;
        db_indx_t *indxp;
        u_int32_t size;
        int (*newfunc) __P((DB *, u_int32_t, PAGE **));
{
        PAGE *h, *np, *tp;
        BKEYDATA *bk;
        DBT page_dbt;
        db_indx_t indx, nindex, oindex, sum;
        db_indx_t halfbytes, i, lastsum;
        int did_indx, ret, s;

        h = *hp;
        indx = *indxp;

        /* Create a temporary page to do compaction onto. */
        if ((tp = (PAGE *)__db_malloc(dbp->pgsize)) == NULL)
                return (ENOMEM);
#ifdef DEBUG
        memset(tp, 0xff, dbp->pgsize);
#endif
        /* Create new page for the split. */
        if ((ret = newfunc(dbp, P_DUPLICATE, &np)) != 0) {
                FREE(tp, dbp->pgsize);
                return (ret);
        }

        P_INIT(np, dbp->pgsize, PGNO(np), PGNO(h), NEXT_PGNO(h), 0,
            P_DUPLICATE);
        P_INIT(tp, dbp->pgsize, PGNO(h), PREV_PGNO(h), PGNO(np), 0,
            P_DUPLICATE);

        /* Figure out the split point */
        halfbytes = (dbp->pgsize - HOFFSET(h)) / 2;
        did_indx = 0;
        for (sum = 0, lastsum = 0, i = 0; i < NUM_ENT(h); i++) {
                if (i == indx) {
                        sum += size;
                        if (lastsum < halfbytes && sum >= halfbytes) {
                                /* We've crossed the halfway point. */
                                if ((db_indx_t)(halfbytes - lastsum) <
                                    (db_indx_t)(sum - halfbytes)) {
                                        *hp = np;
                                        *indxp = 0;
                                        i--;
                                } else
                                        *indxp = i;
                                break;
                        }
                        *indxp = i;
                        lastsum = sum;
                        did_indx = 1;
                }
                if (B_TYPE(GET_BKEYDATA(h, i)->type) == B_KEYDATA)
                        sum += BKEYDATA_SIZE(GET_BKEYDATA(h, i)->len);
                else
                        sum += BOVERFLOW_SIZE;

                if (lastsum < halfbytes && sum >= halfbytes) {
                        /* We've crossed the halfway point. */
                        if ((db_indx_t)(halfbytes - lastsum) <
                            (db_indx_t)(sum - halfbytes))
                                i--;
                        break;
                }
        }

        /*
         * Check if we have set the return values of the index pointer and
         * page pointer.
         */
        if (!did_indx) {
                *hp = np;
                *indxp = indx - i - 1;
        }

        if (DB_LOGGING(dbp)) {
                page_dbt.size = dbp->pgsize;
                page_dbt.data = h;
                if ((ret = __db_split_log(dbp->dbenv->lg_info,
                    dbp->txn, &LSN(h), 0, DB_SPLITOLD, dbp->log_fileid,
                    PGNO(h), &page_dbt, &LSN(h))) != 0) {
                        FREE(tp, dbp->pgsize);
                        return (ret);
                }
                LSN(tp) = LSN(h);
        }

        /*
         * If it's a btree, adjust the cursors.
         *
         * i is the index of the last element to stay on the page.
         */
        if (dbp->type == DB_BTREE || dbp->type == DB_RECNO)
                __bam_ca_split(dbp, PGNO(h), PGNO(h), PGNO(np), i + 1, 0);

        for (nindex = 0, oindex = i + 1; oindex < NUM_ENT(h); oindex++) {
                bk = GET_BKEYDATA(h, oindex);
                if (B_TYPE(bk->type) == B_KEYDATA)
                        s = BKEYDATA_SIZE(bk->len);
                else
                        s = BOVERFLOW_SIZE;

                np->inp[nindex++] = HOFFSET(np) -= s;
                memcpy((u_int8_t *)np + HOFFSET(np), bk, s);
                NUM_ENT(np)++;
        }

        /*
         * Now do data compaction by copying the remaining stuff onto the
         * temporary page and then copying it back to the real page.
         */
        for (nindex = 0, oindex = 0; oindex <= i; oindex++) {
                bk = GET_BKEYDATA(h, oindex);
                if (B_TYPE(bk->type) == B_KEYDATA)
                        s = BKEYDATA_SIZE(bk->len);
                else
                        s = BOVERFLOW_SIZE;

                tp->inp[nindex++] = HOFFSET(tp) -= s;
                memcpy((u_int8_t *)tp + HOFFSET(tp), bk, s);
                NUM_ENT(tp)++;
        }

        /*
         * This page (the temporary) should be only half full, so we do two
         * memcpy's, one for the top of the page and one for the bottom of
         * the page.  This way we avoid copying the middle which should be
         * about half a page.
         */
        memcpy(h, tp, LOFFSET(tp));
        memcpy((u_int8_t *)h + HOFFSET(tp),
            (u_int8_t *)tp + HOFFSET(tp), dbp->pgsize - HOFFSET(tp));
        FREE(tp, dbp->pgsize);

        if (DB_LOGGING(dbp)) {
                page_dbt.size = dbp->pgsize;
                page_dbt.data = h;
                if ((ret = __db_split_log(dbp->dbenv->lg_info,
                    dbp->txn, &LSN(h), 0, DB_SPLITNEW, dbp->log_fileid,
                    PGNO(h), &page_dbt, &LSN(h))) != 0)
                        return (ret);

                page_dbt.size = dbp->pgsize;
                page_dbt.data = np;
                if ((ret = __db_split_log(dbp->dbenv->lg_info,
                    dbp->txn, &LSN(np), 0, DB_SPLITNEW, dbp->log_fileid,
                    PGNO(np),  &page_dbt, &LSN(np))) != 0)
                        return (ret);
        }

        /*
         * Figure out if the location we're interested in is on the new
         * page, and if so, reset the callers' pointer.  Push the other
         * page back to the store.
         */
        if (*hp == h)
                ret = memp_fput(dbp->mpf, np, DB_MPOOL_DIRTY);
        else
                ret = memp_fput(dbp->mpf, h, DB_MPOOL_DIRTY);

        return (ret);
}

/*
 * __db_ditem --
 *      Remove an item from a page.
 *
 * PUBLIC:  int __db_ditem __P((DB *, PAGE *, u_int32_t, u_int32_t));
 */
int
__db_ditem(dbp, pagep, indx, nbytes)
        DB *dbp;
        PAGE *pagep;
        u_int32_t indx, nbytes;
{
        DBT ldbt;
        db_indx_t cnt, offset;
        int ret;
        u_int8_t *from;

        if (DB_LOGGING(dbp)) {
                ldbt.data = P_ENTRY(pagep, indx);
                ldbt.size = nbytes;
                if ((ret = __db_addrem_log(dbp->dbenv->lg_info, dbp->txn,
                    &LSN(pagep), 0, DB_REM_DUP, dbp->log_fileid, PGNO(pagep),
                    (u_int32_t)indx, nbytes, &ldbt, NULL, &LSN(pagep))) != 0)
                        return (ret);
        }

        /*
         * If there's only a single item on the page, we don't have to
         * work hard.
         */
        if (NUM_ENT(pagep) == 1) {
                NUM_ENT(pagep) = 0;
                HOFFSET(pagep) = dbp->pgsize;
                return (0);
        }

        /*
         * Pack the remaining key/data items at the end of the page.  Use
         * memmove(3), the regions may overlap.
         */
        from = (u_int8_t *)pagep + HOFFSET(pagep);
        memmove(from + nbytes, from, pagep->inp[indx] - HOFFSET(pagep));
        HOFFSET(pagep) += nbytes;

        /* Adjust the indices' offsets. */
        offset = pagep->inp[indx];
        for (cnt = 0; cnt < NUM_ENT(pagep); ++cnt)
                if (pagep->inp[cnt] < offset)
                        pagep->inp[cnt] += nbytes;

        /* Shift the indices down. */
        --NUM_ENT(pagep);
        if (indx != NUM_ENT(pagep))
                memmove(&pagep->inp[indx], &pagep->inp[indx + 1],
                    sizeof(db_indx_t) * (NUM_ENT(pagep) - indx));

        /* If it's a btree, adjust the cursors. */
        if (dbp->type == DB_BTREE || dbp->type == DB_RECNO)
                __bam_ca_di(dbp, PGNO(pagep), indx, -1);

        return (0);
}

/*
 * __db_pitem --
 *      Put an item on a page.
 *
 * PUBLIC: int __db_pitem
 * PUBLIC:     __P((DB *, PAGE *, u_int32_t, u_int32_t, DBT *, DBT *));
 */
int
__db_pitem(dbp, pagep, indx, nbytes, hdr, data)
        DB *dbp;
        PAGE *pagep;
        u_int32_t indx;
        u_int32_t nbytes;
        DBT *hdr, *data;
{
        BKEYDATA bk;
        DBT thdr;
        int ret;
        u_int8_t *p;

        /*
         * Put a single item onto a page.  The logic figuring out where to
         * insert and whether it fits is handled in the caller.  All we do
         * here is manage the page shuffling.  We cheat a little bit in that
         * we don't want to copy the dbt on a normal put twice.  If hdr is
         * NULL, we create a BKEYDATA structure on the page, otherwise, just
         * copy the caller's information onto the page.
         *
         * This routine is also used to put entries onto the page where the
         * entry is pre-built, e.g., during recovery.  In this case, the hdr
         * will point to the entry, and the data argument will be NULL.
         *
         * !!!
         * There's a tremendous potential for off-by-one errors here, since
         * the passed in header sizes must be adjusted for the structure's
         * placeholder for the trailing variable-length data field.
         */
        if (DB_LOGGING(dbp))
                if ((ret = __db_addrem_log(dbp->dbenv->lg_info, dbp->txn,
                    &LSN(pagep), 0, DB_ADD_DUP, dbp->log_fileid, PGNO(pagep),
                    (u_int32_t)indx, nbytes, hdr, data, &LSN(pagep))) != 0)
                        return (ret);

        if (hdr == NULL) {
                B_TSET(bk.type, B_KEYDATA, 0);
                bk.len = data == NULL ? 0 : data->size;

                thdr.data = &bk;
                thdr.size = SSZA(BKEYDATA, data);
                hdr = &thdr;
        }

        /* Adjust the index table, then put the item on the page. */
        if (indx != NUM_ENT(pagep))
                memmove(&pagep->inp[indx + 1], &pagep->inp[indx],
                    sizeof(db_indx_t) * (NUM_ENT(pagep) - indx));
        HOFFSET(pagep) -= nbytes;
        pagep->inp[indx] = HOFFSET(pagep);
        ++NUM_ENT(pagep);

        p = P_ENTRY(pagep, indx);
        memcpy(p, hdr->data, hdr->size);
        if (data != NULL)
                memcpy(p + hdr->size, data->data, data->size);

        /* If it's a btree, adjust the cursors. */
        if (dbp->type == DB_BTREE || dbp->type == DB_RECNO)
                __bam_ca_di(dbp, PGNO(pagep), indx, 1);

        return (0);
}

/*
 * __db_relink --
 *      Relink around a deleted page.
 *
 * PUBLIC: int __db_relink __P((DB *, PAGE *, PAGE **, int));
 */
int
__db_relink(dbp, pagep, new_next, needlock)
        DB *dbp;
        PAGE *pagep, **new_next;
        int needlock;
{
        PAGE *np, *pp;
        DB_LOCK npl, ppl;
        DB_LSN *nlsnp, *plsnp;
        int ret;

        ret = 0;
        np = pp = NULL;
        npl = ppl = LOCK_INVALID;
        nlsnp = plsnp = NULL;

        /* Retrieve and lock the two pages. */
        if (pagep->next_pgno != PGNO_INVALID) {
                if (needlock && (ret = __bam_lget(dbp,
                    0, pagep->next_pgno, DB_LOCK_WRITE, &npl)) != 0)
                        goto err;
                if ((ret = memp_fget(dbp->mpf,
                    &pagep->next_pgno, 0, &np)) != 0) {
                        (void)__db_pgerr(dbp, pagep->next_pgno);
                        goto err;
                }
                nlsnp = &np->lsn;
        }
        if (pagep->prev_pgno != PGNO_INVALID) {
                if (needlock && (ret = __bam_lget(dbp,
                    0, pagep->prev_pgno, DB_LOCK_WRITE, &ppl)) != 0)
                        goto err;
                if ((ret = memp_fget(dbp->mpf,
                    &pagep->prev_pgno, 0, &pp)) != 0) {
                        (void)__db_pgerr(dbp, pagep->next_pgno);
                        goto err;
                }
                plsnp = &pp->lsn;
        }

        /* Log the change. */
        if (DB_LOGGING(dbp)) {
                if ((ret = __db_relink_log(dbp->dbenv->lg_info, dbp->txn,
                    &pagep->lsn, 0, dbp->log_fileid, pagep->pgno, &pagep->lsn,
                    pagep->prev_pgno, plsnp, pagep->next_pgno, nlsnp)) != 0)
                        goto err;
                if (np != NULL)
                        np->lsn = pagep->lsn;
                if (pp != NULL)
                        pp->lsn = pagep->lsn;
        }

        /*
         * Modify and release the two pages.
         *
         * !!!
         * The parameter new_next gets set to the page following the page we
         * are removing.  If there is no following page, then new_next gets
         * set to NULL.
         */
        if (np != NULL) {
                np->prev_pgno = pagep->prev_pgno;
                if (new_next == NULL)
                        ret = memp_fput(dbp->mpf, np, DB_MPOOL_DIRTY);
                else {
                        *new_next = np;
                        ret = memp_fset(dbp->mpf, np, DB_MPOOL_DIRTY);
                }
                if (ret != 0)
                        goto err;
                if (needlock)
                        (void)__bam_lput(dbp, npl);
        } else if (new_next != NULL)
                *new_next = NULL;

        if (pp != NULL) {
                pp->next_pgno = pagep->next_pgno;
                if ((ret = memp_fput(dbp->mpf, pp, DB_MPOOL_DIRTY)) != 0)
                        goto err;
                if (needlock)
                        (void)__bam_lput(dbp, ppl);
        }
        return (0);

err:    if (np != NULL)
                (void)memp_fput(dbp->mpf, np, 0);
        if (needlock && npl != LOCK_INVALID)
                (void)__bam_lput(dbp, npl);
        if (pp != NULL)
                (void)memp_fput(dbp->mpf, pp, 0);
        if (needlock && ppl != LOCK_INVALID)
                (void)__bam_lput(dbp, ppl);
        return (ret);
}

/*
 * __db_ddup --
 *      Delete an offpage chain of duplicates.
 *
 * PUBLIC: int __db_ddup __P((DB *, db_pgno_t, int (*)(DB *, PAGE *)));
 */
int
__db_ddup(dbp, pgno, freefunc)
        DB *dbp;
        db_pgno_t pgno;
        int (*freefunc) __P((DB *, PAGE *));
{
        PAGE *pagep;
        DBT tmp_dbt;
        int ret;

        do {
                if ((ret = memp_fget(dbp->mpf, &pgno, 0, &pagep)) != 0) {
                        (void)__db_pgerr(dbp, pgno);
                        return (ret);
                }

                if (DB_LOGGING(dbp)) {
                        tmp_dbt.data = pagep;
                        tmp_dbt.size = dbp->pgsize;
                        if ((ret = __db_split_log(dbp->dbenv->lg_info, dbp->txn,
                            &LSN(pagep), 0, DB_SPLITOLD, dbp->log_fileid,
                            PGNO(pagep), &tmp_dbt, &LSN(pagep))) != 0)
                                return (ret);
                }
                pgno = pagep->next_pgno;
                if ((ret = freefunc(dbp, pagep)) != 0)
                        return (ret);
        } while (pgno != PGNO_INVALID);

        return (0);
}

/*
 * __db_addpage --
 *      Create a new page and link it onto the next_pgno field of the
 *      current page.
 */
static int
__db_addpage(dbp, hp, indxp, newfunc)
        DB *dbp;
        PAGE **hp;
        db_indx_t *indxp;
        int (*newfunc) __P((DB *, u_int32_t, PAGE **));
{
        PAGE *newpage;
        int ret;

        if ((ret = newfunc(dbp, P_DUPLICATE, &newpage)) != 0)
                return (ret);

        if (DB_LOGGING(dbp)) {
                if ((ret = __db_addpage_log(dbp->dbenv->lg_info,
                    dbp->txn, &LSN(*hp), 0, dbp->log_fileid,
                    PGNO(*hp), &LSN(*hp), PGNO(newpage), &LSN(newpage))) != 0) {
                        return (ret);
                }
                LSN(newpage) = LSN(*hp);
        }

        PREV_PGNO(newpage) = PGNO(*hp);
        NEXT_PGNO(*hp) = PGNO(newpage);

        if ((ret = memp_fput(dbp->mpf, *hp, DB_MPOOL_DIRTY)) != 0)
                return (ret);
        *hp = newpage;
        *indxp = 0;
        return (0);
}