about summary refs log tree commit diff
path: root/db/btree/bt_delete.c
diff options
context:
space:
mode:
Diffstat (limited to 'db/btree/bt_delete.c')
-rw-r--r--db/btree/bt_delete.c657
1 files changed, 657 insertions, 0 deletions
diff --git a/db/btree/bt_delete.c b/db/btree/bt_delete.c
new file mode 100644
index 0000000000..ece1ab656e
--- /dev/null
+++ b/db/btree/bt_delete.c
@@ -0,0 +1,657 @@
+/*-
+ * 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)
+static char sccsid[] = "@(#)bt_delete.c	8.13 (Berkeley) 7/28/94";
+#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 *));
+static int __bt_curdel __P((BTREE *, const DBT *, PAGE *, u_int));
+static int __bt_pdelete __P((BTREE *, PAGE *));
+static int __bt_relink __P((BTREE *, PAGE *));
+static int __bt_stkacq __P((BTREE *, PAGE **, CURSOR *));
+
+/*
+ * __bt_delete
+ *	Delete the item(s) referenced by a key.
+ *
+ * Return RET_SPECIAL if the key is not found.
+ */
+int
+__bt_delete(dbp, key, flags)
+	const DB *dbp;
+	const DBT *key;
+	u_int flags;
+{
+	BTREE *t;
+	CURSOR *c;
+	PAGE *h;
+	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;
+	}
+
+	/* Check for change to a read-only tree. */
+	if (F_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 it.
+		 */
+		c = &t->bt_cursor;
+		if (F_ISSET(c, CURS_INIT)) {
+			if (F_ISSET(c, CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
+				return (RET_SPECIAL);
+			if ((h = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL)
+				return (RET_ERROR);
+
+			/*
+			 * If the page is about to be emptied, we'll need to
+			 * delete it, which means we have to acquire a stack.
+			 */
+			if (NEXTINDEX(h) == 1)
+				if (__bt_stkacq(t, &h, &t->bt_cursor))
+					return (RET_ERROR);
+
+			status = __bt_dleaf(t, NULL, h, c->pg.index);
+
+			if (NEXTINDEX(h) == 0 && status == RET_SUCCESS) {
+				if (__bt_pdelete(t, h))
+					return (RET_ERROR);
+			} else
+				mpool_put(t->bt_mp,
+				    h, status == RET_SUCCESS ? MPOOL_DIRTY : 0);
+			break;
+		}
+		/* FALLTHROUGH */
+	default:
+		errno = EINVAL;
+		return (RET_ERROR);
+	}
+	if (status == RET_SUCCESS)
+		F_SET(t, B_MODIFIED);
+	return (status);
+}
+
+/*
+ * __bt_stkacq --
+ *	Acquire a stack so we can delete a cursor entry.
+ *
+ * Parameters:
+ *	  t:	tree
+ *	 hp:	pointer to current, pinned PAGE pointer
+ *	  c:	pointer to the cursor
+ *
+ * Returns:
+ *	0 on success, 1 on failure
+ */
+static int
+__bt_stkacq(t, hp, c)
+	BTREE *t;
+	PAGE **hp;
+	CURSOR *c;
+{
+	BINTERNAL *bi;
+	EPG *e;
+	EPGNO *parent;
+	PAGE *h;
+	indx_t index;
+	pgno_t pgno;
+	recno_t nextpg, prevpg;
+	int exact, level;
+	
+	/*
+	 * Find the first occurrence of the key in the tree.  Toss the
+	 * currently locked page so we don't hit an already-locked page.
+	 */
+	h = *hp;
+	mpool_put(t->bt_mp, h, 0);
+	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
+		return (1);
+	h = e->page;
+
+	/* See if we got it in one shot. */
+	if (h->pgno == c->pg.pgno)
+		goto ret;
+
+	/*
+	 * Move right, looking for the page.  At each move we have to move
+	 * up the stack until we don't have to move to the next page.  If
+	 * we have to change pages at an internal level, we have to fix the
+	 * stack back up.
+	 */
+	while (h->pgno != c->pg.pgno) {
+		if ((nextpg = h->nextpg) == P_INVALID)
+			break;
+		mpool_put(t->bt_mp, h, 0);
+
+		/* Move up the stack. */
+		for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
+			/* Get the parent page. */
+			if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
+				return (1);
+
+			/* Move to the next index. */
+			if (parent->index != NEXTINDEX(h) - 1) {
+				index = parent->index + 1;
+				BT_PUSH(t, h->pgno, index);
+				break;
+			}
+			mpool_put(t->bt_mp, h, 0);
+		}
+
+		/* Restore the stack. */
+		while (level--) {
+			/* Push the next level down onto the stack. */
+			bi = GETBINTERNAL(h, index);
+			pgno = bi->pgno;
+			BT_PUSH(t, pgno, 0);
+
+			/* Lose the currently pinned page. */
+			mpool_put(t->bt_mp, h, 0);
+
+			/* Get the next level down. */
+			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
+				return (1);
+			index = 0;
+		}
+		mpool_put(t->bt_mp, h, 0);
+		if ((h = mpool_get(t->bt_mp, nextpg, 0)) == NULL)
+			return (1);
+	}
+
+	if (h->pgno == c->pg.pgno)
+		goto ret;
+
+	/* Reacquire the original stack. */
+	mpool_put(t->bt_mp, h, 0);
+	if ((e = __bt_search(t, &c->key, &exact)) == NULL)
+		return (1);
+	h = e->page;
+
+	/*
+	 * Move left, looking for the page.  At each move we have to move
+	 * up the stack until we don't have to change pages to move to the
+	 * next page.  If we have to change pages at an internal level, we
+	 * have to fix the stack back up.
+	 */
+	while (h->pgno != c->pg.pgno) {
+		if ((prevpg = h->prevpg) == P_INVALID)
+			break;
+		mpool_put(t->bt_mp, h, 0);
+
+		/* Move up the stack. */
+		for (level = 0; (parent = BT_POP(t)) != NULL; ++level) {
+			/* Get the parent page. */
+			if ((h = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
+				return (1);
+
+			/* Move to the next index. */
+			if (parent->index != 0) {
+				index = parent->index - 1;
+				BT_PUSH(t, h->pgno, index);
+				break;
+			}
+			mpool_put(t->bt_mp, h, 0);
+		}
+
+		/* Restore the stack. */
+		while (level--) {
+			/* Push the next level down onto the stack. */
+			bi = GETBINTERNAL(h, index);
+			pgno = bi->pgno;
+
+			/* Lose the currently pinned page. */
+			mpool_put(t->bt_mp, h, 0);
+
+			/* Get the next level down. */
+			if ((h = mpool_get(t->bt_mp, pgno, 0)) == NULL)
+				return (1);
+
+			index = NEXTINDEX(h) - 1;
+			BT_PUSH(t, pgno, index);
+		}
+		mpool_put(t->bt_mp, h, 0);
+		if ((h = mpool_get(t->bt_mp, prevpg, 0)) == NULL)
+			return (1);
+	}
+	
+
+ret:	mpool_put(t->bt_mp, h, 0);
+	return ((*hp = mpool_get(t->bt_mp, c->pg.pgno, 0)) == NULL);
+}
+
+/*
+ * __bt_bdelete --
+ *	Delete all key/data pairs matching the specified key.
+ *
+ * Parameters:
+ *	  t:	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;
+	PAGE *h;
+	int deleted, exact, redo;
+
+	deleted = 0;
+
+	/* Find any matching record; __bt_search pins the page. */
+loop:	if ((e = __bt_search(t, key, &exact)) == NULL)
+		return (deleted ? RET_SUCCESS : RET_ERROR);
+	if (!exact) {
+		mpool_put(t->bt_mp, e->page, 0);
+		return (deleted ? RET_SUCCESS : RET_SPECIAL);
+	}
+
+	/*
+	 * Delete forward, then delete backward, from the found key.  If
+	 * there are duplicates and we reach either side of the page, do
+	 * the key search again, so that we get them all.
+	 */
+	redo = 0;
+	h = e->page;
+	do {
+		if (__bt_dleaf(t, key, h, e->index)) {
+			mpool_put(t->bt_mp, h, 0);
+			return (RET_ERROR);
+		}
+		if (F_ISSET(t, B_NODUPS)) {
+			if (NEXTINDEX(h) == 0) {
+				if (__bt_pdelete(t, h))
+					return (RET_ERROR);
+			} else
+				mpool_put(t->bt_mp, h, MPOOL_DIRTY);
+			return (RET_SUCCESS);
+		}
+		deleted = 1;
+	} while (e->index < NEXTINDEX(h) && __bt_cmp(t, key, e) == 0);
+
+	/* Check for right-hand edge of the page. */
+	if (e->index == NEXTINDEX(h))
+		redo = 1;
+
+	/* Delete from the key to the beginning of the page. */
+	while (e->index-- > 0) {
+		if (__bt_cmp(t, key, e) != 0)
+			break;
+		if (__bt_dleaf(t, key, h, e->index) == RET_ERROR) {
+			mpool_put(t->bt_mp, h, 0);
+			return (RET_ERROR);
+		}
+		if (e->index == 0)
+			redo = 1;
+	}
+
+	/* Check for an empty page. */
+	if (NEXTINDEX(h) == 0) {
+		if (__bt_pdelete(t, h))
+			return (RET_ERROR);
+		goto loop;
+	}
+
+	/* Put the page. */
+	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
+
+	if (redo)
+		goto loop;
+	return (RET_SUCCESS);
+}
+
+/*
+ * __bt_pdelete --
+ *	Delete a single page from the tree.
+ *
+ * Parameters:
+ *	t:	tree
+ *	h:	leaf page
+ *
+ * Returns:
+ *	RET_SUCCESS, RET_ERROR.
+ *
+ * Side-effects:
+ *	mpool_put's the page
+ */
+static int
+__bt_pdelete(t, h)
+	BTREE *t;
+	PAGE *h;
+{
+	BINTERNAL *bi;
+	PAGE *pg;
+	EPGNO *parent;
+	indx_t cnt, index, *ip, offset;
+	u_int32_t nksize;
+	char *from;
+
+	/*
+	 * Walk the parent page stack -- a LIFO stack of the pages that were
+	 * traversed when we searched for the page where the delete occurred.
+	 * 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 deleted
+	 * a page, so we have to delete the key from the parent page.
+	 *
+	 * If the delete from the parent page makes it empty, this process may
+	 * continue all the way up the tree.  We stop if we reach the root page
+	 * (which is never deleted, it's just not worth the effort) or if the
+	 * delete does not empty the page.
+	 */
+	while ((parent = BT_POP(t)) != NULL) {
+		/* Get the parent page. */
+		if ((pg = mpool_get(t->bt_mp, parent->pgno, 0)) == NULL)
+			return (RET_ERROR);
+		
+		index = parent->index;
+		bi = GETBINTERNAL(pg, index);
+
+		/* Free any overflow pages. */
+		if (bi->flags & P_BIGKEY &&
+		    __ovfl_delete(t, bi->bytes) == RET_ERROR) {
+			mpool_put(t->bt_mp, pg, 0);
+			return (RET_ERROR);
+		}
+
+		/*
+		 * Free the parent if it has only the one key and it's not the
+		 * root page. If it's the rootpage, turn it back into an empty
+		 * leaf page.
+		 */
+		if (NEXTINDEX(pg) == 1)
+			if (pg->pgno == P_ROOT) {
+				pg->lower = BTDATAOFF;
+				pg->upper = t->bt_psize;
+				pg->flags = P_BLEAF;
+			} else {
+				if (__bt_relink(t, pg) || __bt_free(t, pg))
+					return (RET_ERROR);
+				continue;
+			}
+		else {
+			/* Pack remaining key items at the end of the page. */
+			nksize = NBINTERNAL(bi->ksize);
+			from = (char *)pg + pg->upper;
+			memmove(from + nksize, from, (char *)bi - from);
+			pg->upper += nksize;
+
+			/* Adjust indices' offsets, shift the indices down. */
+			offset = pg->linp[index];
+			for (cnt = index, ip = &pg->linp[0]; cnt--; ++ip)
+				if (ip[0] < offset)
+					ip[0] += nksize;
+			for (cnt = NEXTINDEX(pg) - index; --cnt; ++ip)
+				ip[0] = ip[1] < offset ? ip[1] + nksize : ip[1];
+			pg->lower -= sizeof(indx_t);
+		}
+
+		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
+		break;
+	}
+
+	/* Free the leaf page, as long as it wasn't the root. */
+	if (h->pgno == P_ROOT) {
+		mpool_put(t->bt_mp, h, MPOOL_DIRTY);
+		return (RET_SUCCESS);
+	}
+	return (__bt_relink(t, h) || __bt_free(t, h));
+}
+
+/*
+ * __bt_dleaf --
+ *	Delete a single record from a leaf page.
+ *
+ * Parameters:
+ *	t:	tree
+ *    key:	referenced key
+ *	h:	page
+ *	index:	index on page to delete
+ *
+ * Returns:
+ *	RET_SUCCESS, RET_ERROR.
+ */
+int
+__bt_dleaf(t, key, h, index)
+	BTREE *t;
+	const DBT *key;
+	PAGE *h;
+	u_int index;
+{
+	BLEAF *bl;
+	indx_t cnt, *ip, offset;
+	u_int32_t nbytes;
+	void *to;
+	char *from;
+
+	/* If this record is referenced by the cursor, delete the cursor. */
+	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
+	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
+	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index == index &&
+	    __bt_curdel(t, key, h, index))
+		return (RET_ERROR);
+
+	/* If the entry 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);
+
+	/* Pack the remaining key/data items at the end of the page. */
+	nbytes = NBLEAF(bl);
+	from = (char *)h + h->upper;
+	memmove(from + nbytes, from, (char *)to - from);
+	h->upper += nbytes;
+
+	/* Adjust the indices' offsets, shift the indices down. */
+	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);
+
+	/* If the cursor is on this page, adjust it as necessary. */
+	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
+	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
+	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index > index)
+		--t->bt_cursor.pg.index;
+
+	return (RET_SUCCESS);
+}
+
+/*
+ * __bt_curdel --
+ *	Delete the cursor.
+ *
+ * Parameters:
+ *	t:	tree
+ *    key:	referenced key (or NULL)
+ *	h:	page
+ *  index:	index on page to delete
+ *
+ * Returns:
+ *	RET_SUCCESS, RET_ERROR.
+ */
+static int
+__bt_curdel(t, key, h, index)
+	BTREE *t;
+	const DBT *key;
+	PAGE *h;
+	u_int index;
+{
+	CURSOR *c;
+	EPG e;
+	PAGE *pg;
+	int curcopy, status;
+
+	/*
+	 * If there are duplicates, move forward or backward to one.
+	 * Otherwise, copy the key into the cursor area.
+	 */
+	c = &t->bt_cursor;
+	F_CLR(c, CURS_AFTER | CURS_BEFORE | CURS_ACQUIRE);
+
+	curcopy = 0;
+	if (!F_ISSET(t, B_NODUPS)) {
+		/*
+		 * We're going to have to do comparisons.  If we weren't
+		 * provided a copy of the key, i.e. the user is deleting
+		 * the current cursor position, get one.
+		 */
+		if (key == NULL) {
+			e.page = h;
+			e.index = index;
+			if ((status = __bt_ret(t, &e,
+			    &c->key, &c->key, NULL, NULL, 1)) != RET_SUCCESS)
+				return (status);
+			curcopy = 1;
+			key = &c->key;
+		}
+		/* Check previous key, if not at the beginning of the page. */
+		if (index > 0) { 
+			e.page = h;
+			e.index = index - 1;
+			if (__bt_cmp(t, key, &e) == 0) {
+				F_SET(c, CURS_BEFORE);
+				goto dup2;
+			}
+		}
+		/* Check next key, if not at the end of the page. */
+		if (index < NEXTINDEX(h) - 1) {
+			e.page = h;
+			e.index = index + 1;
+			if (__bt_cmp(t, key, &e) == 0) {
+				F_SET(c, CURS_AFTER);
+				goto dup2;
+			}
+		}
+		/* Check previous key if at the beginning of the page. */
+		if (index == 0 && h->prevpg != P_INVALID) {
+			if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
+				return (RET_ERROR);
+			e.page = pg;
+			e.index = NEXTINDEX(pg) - 1;
+			if (__bt_cmp(t, key, &e) == 0) {
+				F_SET(c, CURS_BEFORE);
+				goto dup1;
+			}
+			mpool_put(t->bt_mp, pg, 0);
+		}
+		/* Check next key if at the end of the page. */
+		if (index == NEXTINDEX(h) - 1 && h->nextpg != P_INVALID) {
+			if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
+				return (RET_ERROR);
+			e.page = pg;
+			e.index = 0;
+			if (__bt_cmp(t, key, &e) == 0) {
+				F_SET(c, CURS_AFTER);
+dup1:				mpool_put(t->bt_mp, pg, 0);
+dup2:				c->pg.pgno = e.page->pgno;
+				c->pg.index = e.index;
+				return (RET_SUCCESS);
+			}
+			mpool_put(t->bt_mp, pg, 0);
+		}
+	}
+	e.page = h;
+	e.index = index;
+	if (curcopy || (status =
+	    __bt_ret(t, &e, &c->key, &c->key, NULL, NULL, 1)) == RET_SUCCESS) {
+		F_SET(c, CURS_ACQUIRE);
+		return (RET_SUCCESS);
+	}
+	return (status);
+}
+
+/*
+ * __bt_relink --
+ *	Link around a deleted page.
+ *
+ * Parameters:
+ *	t:	tree
+ *	h:	page to be deleted
+ */
+static int
+__bt_relink(t, h)
+	BTREE *t;
+	PAGE *h;
+{
+	PAGE *pg;
+
+	if (h->nextpg != P_INVALID) {
+		if ((pg = mpool_get(t->bt_mp, h->nextpg, 0)) == NULL)
+			return (RET_ERROR);
+		pg->prevpg = h->prevpg;
+		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
+	}
+	if (h->prevpg != P_INVALID) {
+		if ((pg = mpool_get(t->bt_mp, h->prevpg, 0)) == NULL)
+			return (RET_ERROR);
+		pg->nextpg = h->nextpg;
+		mpool_put(t->bt_mp, pg, MPOOL_DIRTY);
+	}
+	return (0);
+}