projects / unix-history / blob
? search:
summary | tags | clone url | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Oh GACK! src-clean doesn't quite work that easily since cleandist rebuilds the
[unix-history] / lib / libc / db / btree / split.c
/*-
* Copyright (c) 1990 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)
static char sccsid[] = "@(#)split.c 5.2 (Berkeley) 2/22/91";
#endif /* LIBC_SCCS and not lint */
#include <sys/types.h>
#include <db.h>
#include <stdlib.h>
#include <string.h>
#include "btree.h"
/*
* _BT_SPLIT -- Split a page into two pages.
*
* Splits are caused by insertions, and propogate up the tree in
* the usual way. The root page is always page 1 in the file on
* disk, so root splits are handled specially. On entry to this
* routine, t->bt_curpage is the page to be split.
*
* Parameters:
* t -- btree in which to do split.
*
* Returns:
* RET_SUCCESS, RET_ERROR.
*
* Side Effects:
* Changes the notion of the current page.
*/
int
_bt_split(t)
BTREE_P t;
{
BTHEADER *h;
BTHEADER *left, *right;
pgno_t nextpgno, parent;
int nbytes, len;
IDATUM *id;
DATUM *d;
char *src;
IDATUM *new;
pgno_t oldchain;
u_char flags;
h = (BTHEADER *) t->bt_curpage;
/* split root page specially, since it must remain page 1 */
if (h->h_pgno == P_ROOT) {
return (_bt_splitroot(t));
}
/*
* This is a little complicated. We go to some trouble to
* figure out which of the three possible cases -- in-memory tree,
* disk tree (no cache), and disk tree (cache) -- we have, in order
* to avoid unnecessary copying. If we have a disk cache, then we
* have to do some extra copying, though, since the cache code
* manages buffers externally to this code.
*/
if (ISDISK(t) && ISCACHE(t)) {
if ((left = (BTHEADER *) malloc((unsigned) t->bt_psize))
== (BTHEADER *) NULL)
return (RET_ERROR);
left->h_pgno = left->h_prevpg = left->h_nextpg = P_NONE;
left->h_flags = t->bt_curpage->h_flags;
left->h_lower = (index_t)
(((char *) &(left->h_linp[0])) - ((char *) left));
left->h_upper = t->bt_psize;
} else {
if ((left = _bt_allocpg(t)) == (BTHEADER *) NULL)
return (RET_ERROR);
}
left->h_pgno = h->h_pgno;
if ((right = _bt_allocpg(t)) == (BTHEADER *) NULL)
return (RET_ERROR);
right->h_pgno = ++(t->bt_npages);
/* now do the split */
if (_bt_dopsplit(t, left, right) == RET_ERROR)
return (RET_ERROR);
right->h_prevpg = left->h_pgno;
nextpgno = right->h_nextpg = h->h_nextpg;
left->h_nextpg = right->h_pgno;
left->h_prevpg = h->h_prevpg;
/* okay, now use the left half of the page as the new page */
if (ISDISK(t) && ISCACHE(t)) {
(void) bcopy((char *) left, (char *) t->bt_curpage,
(int) t->bt_psize);
(void) free ((char *) left);
left = t->bt_curpage;
} else {
(void) free((char *) t->bt_curpage);
t->bt_curpage = left;
}
/*
* Write the new pages out. We need them to stay where they are
* until we're done updating the parent pages.
*/
if (_bt_write(t, left, NORELEASE) == RET_ERROR)
return (RET_ERROR);
if (_bt_write(t, right, NORELEASE) == RET_ERROR)
return (RET_ERROR);
/* update 'prev' pointer of old neighbor of left */
if (nextpgno != P_NONE) {
if (_bt_getpage(t, nextpgno) == RET_ERROR)
return (RET_ERROR);
h = t->bt_curpage;
h->h_prevpg = right->h_pgno;
h->h_flags |= F_DIRTY;
}
if ((parent = _bt_pop(t)) != P_NONE) {
if (right->h_flags & F_LEAF) {
d = (DATUM *) GETDATUM(right, 0);
len = d->d_ksize;
if (d->d_flags & D_BIGKEY) {
bcopy(&(d->d_bytes[0]),
(char *) &oldchain,
sizeof(oldchain));
if (_bt_markchain(t, oldchain) == RET_ERROR)
return (RET_ERROR);
src = (char *) &oldchain;
flags = D_BIGKEY;
} else {
src = (char *) &(d->d_bytes[0]);
flags = 0;
}
} else {
id = (IDATUM *) GETDATUM(right, 0);
len = id->i_size;
flags = id->i_flags;
src = (char *) &(id->i_bytes[0]);
}
nbytes = len + (sizeof(IDATUM) - sizeof(char));
new = (IDATUM *) malloc((unsigned) nbytes);
if (new == (IDATUM *) NULL)
return (RET_ERROR);
new->i_size = len;
new->i_pgno = right->h_pgno;
new->i_flags = flags;
if (len > 0)
(void) bcopy(src, (char *) &(new->i_bytes[0]), len);
nbytes = LONGALIGN(nbytes) + sizeof(index_t);
if (_bt_getpage(t, parent) == RET_ERROR)
return (RET_ERROR);
h = t->bt_curpage;
/*
* Split the parent if we need to, then reposition the
* tree's current page pointer for the new datum.
*/
if ((h->h_upper - h->h_lower) < nbytes) {
if (_bt_split(t) == RET_ERROR)
return (RET_ERROR);
if (_bt_reposition(t, new, parent, right->h_prevpg)
== RET_ERROR)
return (RET_ERROR);
}
/* okay, now insert the new idatum */
if (_bt_inserti(t, new, right->h_prevpg) == RET_ERROR)
return (RET_ERROR);
}
/*
* Okay, split is done; don't need right page stapled down anymore.
* The page we call 'left' above is the new version of the old
* (split) page, so we can't release it.
*/
if (_bt_release(t, right) == RET_ERROR)
return (RET_ERROR);
if (ISDISK(t) && !ISCACHE(t))
(void) free((char *) right);
return (RET_SUCCESS);
}
/*
* _BT_REPOSITION -- Reposition the current page pointer of a btree.
*
* After splitting a node in the tree in order to make room for
* an insertion, we need to figure out which page after the split
* should get the item we want to insert. This routine positions
* the tree's current page pointer appropriately.
*
* Parameters:
* t -- tree to position
* new -- the item we want to insert
* parent -- parent of the node that we just split
* prev -- page number of item directly to the left of
* new's position in the tree.
*
* Returns:
* RET_SUCCESS, RET_ERROR.
*
* Side Effects:
* None.
*/
int
_bt_reposition(t, new, parent, prev)
BTREE_P t;
IDATUM *new;
pgno_t parent;
pgno_t prev;
{
index_t i, next;
IDATUM *idx;
if (parent == P_ROOT) {
/*
* If we just split the root page, then there are guaranteed
* to be exactly two IDATUMs on it. Look at both of them
* to see if they point to the page that we want.
*/
if (_bt_getpage(t, parent) == RET_ERROR)
return (RET_ERROR);
next = NEXTINDEX(t->bt_curpage);
for (i = 0; i < next; i++) {
idx = (IDATUM *) GETDATUM(t->bt_curpage, i);
if (_bt_getpage(t, idx->i_pgno) == RET_ERROR)
return (RET_ERROR);
if (_bt_isonpage(t, new, prev) == RET_SUCCESS)
return (RET_SUCCESS);
if (_bt_getpage(t, parent) == RET_ERROR)
return (RET_ERROR);
}
} else {
/*
* Get the parent page -- which is where the new item would
* have gone -- and figure out whether the new item now goes
* on the parent, or the page immediately to the right of
* the parent.
*/
if (_bt_getpage(t, parent) == RET_ERROR)
return (RET_ERROR);
if (_bt_isonpage(t, new, prev) == RET_SUCCESS)
return (RET_SUCCESS);
if (_bt_getpage(t, t->bt_curpage->h_nextpg) == RET_ERROR)
return (RET_ERROR);
if (_bt_isonpage(t, new, prev) == RET_SUCCESS)
return (RET_SUCCESS);
}
return (RET_ERROR);
}
/*
* _BT_ISONPAGE -- Is the IDATUM for a given page number on the current page?
*
* This routine is used by _bt_reposition to decide whether the current
* page is the correct one on which to insert a new item.
*
* Parameters:
* t -- tree to check
* new -- the item that will be inserted (used for binary search)
* prev -- page number of page whose IDATUM is immediately to
* the left of new's position in the tree.
*
* Returns:
* RET_SUCCESS, RET_ERROR (corresponding to TRUE, FALSE).
*/
int
_bt_isonpage(t, new, prev)
BTREE_P t;
IDATUM *new;
pgno_t prev;
{
BTHEADER *h = (BTHEADER *) t->bt_curpage;
index_t i, next;
IDATUM *idx;
i = _bt_binsrch(t, &(new->i_bytes[0]));
while (i != 0 && _bt_cmp(t, &(new->i_bytes[0]), i) == 0)
--i;
next = NEXTINDEX(h);
idx = (IDATUM *) GETDATUM(h, i);
while (i < next && idx->i_pgno != prev) {
i++;
idx = (IDATUM *) GETDATUM(h, i);
}
if (idx->i_pgno == prev)
return (RET_SUCCESS);
else
return (RET_ERROR);
}
/*
* _BT_SPLITROOT -- Split the root of a btree.
*
* The root page for a btree is always page one. This means that in
* order to split the root, we need to do extra work.
*
* Parameters:
* t -- tree to split
*
* Returns:
* RET_SUCCESS, RET_ERROR.
*
* Side Effects:
* Splits root upward in the usual way, adding two new pages
* to the tree (rather than just one, as in usual splits).
*/
int
_bt_splitroot(t)
BTREE_P t;
{
BTHEADER *h = t->bt_curpage;
BTHEADER *left, *right;
IDATUM *id;
BTHEADER *where_h;
char *src, *dest;
int len, nbytes;
u_long was_leaf;
pgno_t oldchain;
u_char flags;
/* get two new pages for the split */
if ((left = _bt_allocpg(t)) == (BTHEADER *) NULL)
return (RET_ERROR);
left->h_pgno = ++(t->bt_npages);
if ((right = _bt_allocpg(t)) == (BTHEADER *) NULL)
return (RET_ERROR);
right->h_pgno = ++(t->bt_npages);
/* do the split */
if (_bt_dopsplit(t, left, right) == RET_ERROR)
return (RET_ERROR);
/* connect the new pages correctly */
right->h_prevpg = left->h_pgno;
left->h_nextpg = right->h_pgno;
/*
* Write the child pages out now. We need them to remain
* where they are until we finish updating parent pages,
* however.
*/
if (_bt_write(t, left, NORELEASE) == RET_ERROR)
return (RET_ERROR);
if (_bt_write(t, right, NORELEASE) == RET_ERROR)
return (RET_ERROR);
/* now change the root page into an internal page */
was_leaf = (h->h_flags & F_LEAF);
h->h_flags &= ~F_LEAF;
h->h_lower = (index_t) (((char *) (&(h->h_linp[0]))) - ((char *) h));
h->h_upper = (index_t) t->bt_psize;
(void) bzero((char *) &(h->h_linp[0]), (int) (h->h_upper - h->h_lower));
/* put two new keys on root page */
where_h = left;
while (where_h) {
DATUM *data;
IDATUM *idata;
if (was_leaf) {
data = (DATUM *) GETDATUM(where_h, 0);
if (where_h == left) {
len = 0; /* first key in tree is null */
} else {
if (data->d_flags & D_BIGKEY) {
bcopy(&(data->d_bytes[0]),
(char *) &oldchain,
sizeof(oldchain));
if (_bt_markchain(t, oldchain) == RET_ERROR)
return (RET_ERROR);
src = (char *) &oldchain;
flags = D_BIGKEY;
} else {
src = (char *) &(data->d_bytes[0]);
flags = 0;
}
len = data->d_ksize;
}
} else {
idata = (IDATUM *) GETDATUM(where_h, 0);
len = idata->i_size;
flags = idata->i_flags;
src = &(idata->i_bytes[0]);
}
dest = ((char *) h) + h->h_upper;
nbytes = len + (sizeof (IDATUM) - sizeof(char));
dest -= LONGALIGN(nbytes);
id = (IDATUM *) dest;
id->i_size = len;
id->i_pgno = where_h->h_pgno;
id->i_flags = flags;
if (len > 0)
(void) bcopy((char *) src, (char *) &(id->i_bytes[0]), len);
dest -= ((int) h);
h->h_linp[NEXTINDEX(h)] = (index_t) dest;
h->h_upper = (index_t) dest;
h->h_lower += sizeof(index_t);
/* next page */
if (where_h == left)
where_h = right;
else
where_h = (BTHEADER *) NULL;
}
if (_bt_release(t, left) == RET_ERROR)
return (RET_ERROR);
if (_bt_release(t, right) == RET_ERROR)
return (RET_ERROR);
/*
* That's it, split is done. If we're doing a non-cached disk
* btree, we can free up the pages we allocated, as they're on
* disk, now.
*/
if (ISDISK(t) && !ISCACHE(t)) {
(void) free ((char *) left);
(void) free ((char *) right);
}
h->h_flags |= F_DIRTY;
return (RET_SUCCESS);
}
/*
* _BT_DOPSPLIT -- Do the work of splitting a page
*
* This routine takes two page pointers and splits the data on the
* current page of the btree between them.
*
* We do a lot of work here to handle duplicate keys on a page; we
* have to place these keys carefully to guarantee that later searches
* will find them correctly. See comments in the code below for details.
*
* Parameters:
* t -- tree to split
* left -- pointer to page to get left half of the data
* right -- pointer to page to get right half of the data
*
* Returns:
* None.
*/
int
_bt_dopsplit(t, left, right)
BTREE_P t;
BTHEADER *left;
BTHEADER *right;
{
BTHEADER *h = t->bt_curpage;
size_t psize;
char *where;
BTHEADER *where_h;
index_t where_i;
int nbytes, dsize, fixedsize, freespc;
index_t i;
index_t save_lower, save_upper, save_i;
index_t switch_i;
char *save_key;
DATUM *d;
CURSOR *c;
index_t top;
int free_save;
pgno_t chain;
int ignore;
/*
* Our strategy is to put half the bytes on each page. We figure
* out how many bytes we have total, and then move items until
* the last item moved put at least 50% of the data on the left
* page.
*/
save_key = (char *) NULL;
psize = (int) t->bt_psize;
where = ((char *) left) + psize;
where_h = left;
where_i = 0;
nbytes = psize - (int) ((char *) &(h->h_linp[0]) - ((char *) h));
freespc = nbytes;
top = NEXTINDEX(h);
if (h->h_flags & F_LEAF)
fixedsize = (sizeof(DATUM) - sizeof(char));
else
fixedsize = (sizeof(IDATUM) - sizeof(char));
save_key = (char *) NULL;
/* move data */
for (i = 0; i < top; i++) {
/*
* Internal and leaf pages have different layouts for
* data items, but in both cases the first entry in the
* data item is a size_t.
*/
d = (DATUM *) GETDATUM(h,i);
if (h->h_flags & F_LEAF) {
dsize = d->d_ksize + d->d_dsize + fixedsize;
} else {
dsize = d->d_ksize + fixedsize;
}
/*
* If a page contains duplicate keys, we have to be
* careful about splits. A sequence of duplicate keys
* may not begin in the middle of one page and end in
* the middle of another; it must begin on a page boundary,
* in order for searches on the internal nodes to work
* correctly.
*/
if (where_h == left) {
if (save_key == (char *) NULL) {
if (h->h_flags & F_LEAF) {
if (d->d_flags & D_BIGKEY) {
free_save = TRUE;
bcopy(&(d->d_bytes[0]),
(char *) &chain,
sizeof(chain));
if (_bt_getbig(t, chain,
&save_key,
&ignore)
== RET_ERROR)
return (RET_ERROR);
} else {
free_save = FALSE;
save_key = (char *) &(d->d_bytes[0]);
}
} else {
IDATUM *id = (IDATUM *) d;
if (id->i_flags & D_BIGKEY) {
free_save = TRUE;
bcopy(&(id->i_bytes[0]),
(char *) &chain,
sizeof(chain));
if (_bt_getbig(t, chain,
&save_key,
&ignore)
== RET_ERROR)
return (RET_ERROR);
} else {
free_save = FALSE;
save_key = (char *)
&(id->i_bytes[0]);
}
}
save_i = 0;
save_lower = where_h->h_lower;
save_upper = where_h->h_upper;
} else {
if (_bt_cmp(t, save_key, i) != 0) {
save_lower = where_h->h_lower;
save_upper = where_h->h_upper;
save_i = i;
}
if (h->h_flags & F_LEAF) {
if (free_save)
(void) free(save_key);
if (d->d_flags & D_BIGKEY) {
free_save = TRUE;
bcopy(&(d->d_bytes[0]),
(char *) &chain,
sizeof(chain));
if (_bt_getbig(t, chain,
&save_key,
&ignore)
== RET_ERROR)
return (RET_ERROR);
} else {
free_save = FALSE;
save_key = (char *) &(d->d_bytes[0]);
}
} else {
IDATUM *id = (IDATUM *) d;
if (id->i_flags & D_BIGKEY) {
free_save = TRUE;
bcopy(&(id->i_bytes[0]),
(char *) &chain,
sizeof(chain));
if (_bt_getbig(t, chain,
&save_key,
&ignore)
== RET_ERROR)
return (RET_ERROR);
} else {
free_save = FALSE;
save_key = (char *)
&(id->i_bytes[0]);
}
}
}
}
/* copy data and update page state */
where -= LONGALIGN(dsize);
(void) bcopy((char *) d, (char *) where, dsize);
where_h->h_upper = where_h->h_linp[where_i] =
(index_t) (where - (int) where_h);
where_h->h_lower += sizeof(index_t);
where_i++;
/* if we've moved half, switch to the right-hand page */
nbytes -= LONGALIGN(dsize) + sizeof(index_t);
if ((freespc - nbytes) > nbytes) {
nbytes = 2 * freespc;
/* identical keys go on the same page */
if (save_i > 0) {
/* i gets incremented at loop bottom... */
i = save_i - 1;
where_h->h_lower = save_lower;
where_h->h_upper = save_upper;
}
where = ((char *) right) + psize;
where_h = right;
switch_i = where_i;
where_i = 0;
}
}
/*
* If there was an active scan on the database, and we just
* split the page that the cursor was on, we may need to
* adjust the cursor to point to the same entry as before the
* split.
*/
c = &(t->bt_cursor);
if ((t->bt_flags & BTF_SEQINIT)
&& (c->c_pgno == h->h_pgno)
&& (c->c_index >= switch_i)) {
c->c_pgno = where_h->h_pgno;
c->c_index -= where_i;
}
return (RET_SUCCESS);
}
Continuous source code history from original PDP-7 UNIX to FreeBSD 2, the first fully unencumbered FreeBSD release.