/*
* __bt_dleaf --
* Delete a single record from a leaf page.
*
* Parameters:
* t: tree
* key: referenced key
* h: page
* idx: index on page to delete
*
* Returns:
* RET_SUCCESS, RET_ERROR.
*/
int
__bt_dleaf(BTREE *t, const DBT *key, PAGE *h, u_int idx)
{
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 == idx &&
__bt_curdel(t, key, h, idx))
return (RET_ERROR);
/* If the entry uses overflow pages, make them available for reuse. */
to = bl = GETBLEAF(h, idx);
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[idx];
for (cnt = idx, ip = &h->linp[0]; cnt--; ++ip)
if (ip[0] < offset)
ip[0] += nbytes;
for (cnt = NEXTINDEX(h) - idx; --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 > idx)
--t->bt_cursor.pg.index;
return (RET_SUCCESS);
}
/*
* __REC_DLEAF -- Delete a single record from a recno leaf page.
*
* Parameters:
* t: tree
* idx: index on current page to delete
*
* Returns:
* RET_SUCCESS, RET_ERROR.
*/
int
__rec_dleaf(BTREE *t, PAGE *h, u_int32_t idx)
{
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, idx);
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[idx];
for (cnt = &h->linp[idx] - (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);
}
/*
* __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(BTREE *t, PAGE *h)
{
BINTERNAL *bi;
PAGE *pg;
EPGNO *parent;
indx_t cnt, idx, *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);
idx = parent->index;
bi = GETBINTERNAL(pg, idx);
/* 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[idx];
for (cnt = idx, ip = &pg->linp[0]; cnt--; ++ip)
if (ip[0] < offset)
ip[0] += nksize;
for (cnt = NEXTINDEX(pg) - idx; --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));
}