/*
* __wt_btcur_next_random --
* Move to a random record in the tree.
*/
int
__wt_btcur_next_random(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_DECL_RET;
WT_SESSION_IMPL *session;
session = (WT_SESSION_IMPL *)cbt->iface.session;
btree = cbt->btree;
WT_DSTAT_INCR(session, cursor_next);
retry: WT_RET(__cursor_func_init(cbt, 1));
__cursor_position_clear(cbt);
/*
* Only supports row-store: applications can trivially select a random
* value from a column-store, if there were any reason to do so.
*/
WT_ERR(btree->type == BTREE_ROW ?
__wt_row_random(session, cbt) : ENOTSUP);
ret = cbt->compare == 0 ?
__wt_kv_return(session, cbt) : WT_NOTFOUND;
err: if (ret == WT_RESTART)
goto retry;
WT_TRET(__cursor_func_resolve(cbt, ret));
return (ret);
}
/*
* __wt_btcur_update --
* Update a record in the tree.
*/
int
__wt_btcur_update(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_CURSOR *cursor;
WT_SESSION_IMPL *session;
int ret;
btree = cbt->btree;
cursor = &cbt->iface;
session = (WT_SESSION_IMPL *)cursor->session;
WT_BSTAT_INCR(session, cursor_updates);
if (btree->type == BTREE_ROW)
WT_RET(__cursor_size_chk(session, &cursor->key));
WT_RET(__cursor_size_chk(session, &cursor->value));
retry: __cursor_func_init(cbt, 1);
switch (btree->type) {
case BTREE_COL_FIX:
if (cursor->value.size != 1)
WT_RET_MSG(session, EINVAL,
"item size of %" PRIu32 " does not match "
"fixed-length file requirement of 1 byte",
cursor->value.size);
/* FALLTHROUGH */
case BTREE_COL_VAR:
WT_ERR(__wt_col_search(session, cbt, 1));
/*
* Update the record if it exists. Creating a record past the
* end of the tree in a fixed-length column-store implicitly
* fills the gap with empty records. Update the record in that
* case, the record exists.
*/
if ((cbt->compare != 0 || __cursor_invalid(cbt)) &&
!__cursor_fix_implicit(btree, cbt))
ret = WT_NOTFOUND;
else if ((ret = __wt_col_modify(session, cbt, 3)) == WT_RESTART)
goto retry;
break;
case BTREE_ROW:
/* Update the record it it exists. */
WT_ERR(__wt_row_search(session, cbt, 1));
if (cbt->compare != 0 || __cursor_invalid(cbt))
ret = WT_NOTFOUND;
else if ((ret = __wt_row_modify(session, cbt, 0)) == WT_RESTART)
goto retry;
break;
WT_ILLEGAL_VALUE(session);
}
err: __cursor_func_resolve(cbt, ret);
return (ret);
}
/*
* __wt_btcur_remove --
* Remove a record from the tree.
*/
int
__wt_btcur_remove(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_CURSOR *cursor;
WT_SESSION_IMPL *session;
int ret;
btree = cbt->btree;
cursor = &cbt->iface;
session = (WT_SESSION_IMPL *)cursor->session;
WT_BSTAT_INCR(session, cursor_removes);
if (btree->type == BTREE_ROW)
WT_RET(__cursor_size_chk(session, &cursor->key));
retry: __cursor_func_init(cbt, 1);
switch (btree->type) {
case BTREE_COL_FIX:
case BTREE_COL_VAR:
WT_ERR(__wt_col_search(session, cbt, 1));
/*
* Remove the record if it exists. Creating a record past the
* end of the tree in a fixed-length column-store implicitly
* fills the gap with empty records. Return success in that
* case, the record was deleted successfully.
*/
if (cbt->compare != 0 || __cursor_invalid(cbt))
ret =
__cursor_fix_implicit(btree, cbt) ? 0 : WT_NOTFOUND;
else if ((ret = __wt_col_modify(session, cbt, 2)) == WT_RESTART)
goto retry;
break;
case BTREE_ROW:
/* Remove the record if it exists. */
WT_ERR(__wt_row_search(session, cbt, 1));
if (cbt->compare != 0 || __cursor_invalid(cbt))
ret = WT_NOTFOUND;
else if ((ret = __wt_row_modify(session, cbt, 1)) == WT_RESTART)
goto retry;
break;
WT_ILLEGAL_VALUE(session);
}
err: __cursor_func_resolve(cbt, ret);
return (ret);
}
/*
* __wt_btcur_search --
* Search for a matching record in the tree.
*/
int
__wt_btcur_search(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_CURSOR *cursor;
WT_ITEM *val;
WT_SESSION_IMPL *session;
int ret;
btree = cbt->btree;
cursor = &cbt->iface;
session = (WT_SESSION_IMPL *)cursor->session;
WT_BSTAT_INCR(session, cursor_read);
if (btree->type == BTREE_ROW)
WT_RET(__cursor_size_chk(session, &cursor->key));
__cursor_func_init(cbt, 1);
WT_ERR(btree->type == BTREE_ROW ?
__wt_row_search(session, cbt, 0) :
__wt_col_search(session, cbt, 0));
if (cbt->compare != 0 || __cursor_invalid(cbt)) {
/*
* Creating a record past the end of the tree in a fixed-length
* column-store implicitly fills the gap with empty records.
*/
if (__cursor_fix_implicit(btree, cbt)) {
cbt->v = 0;
val = &cbt->iface.value;
val->data = &cbt->v;
val->size = 1;
} else
ret = WT_NOTFOUND;
} else
ret = __wt_kv_return(session, cbt, 0);
err: __cursor_func_resolve(cbt, ret);
return (ret);
}
/*
* __wt_btcur_prev --
* Move to the previous record in the tree.
*/
int
__wt_btcur_prev(WT_CURSOR_BTREE *cbt, int discard)
{
WT_DECL_RET;
WT_PAGE *page;
WT_SESSION_IMPL *session;
uint32_t flags;
int newpage;
session = (WT_SESSION_IMPL *)cbt->iface.session;
WT_DSTAT_INCR(session, cursor_prev);
flags = WT_TREE_SKIP_INTL | WT_TREE_PREV; /* Tree walk flags. */
if (discard)
LF_SET(WT_TREE_DISCARD);
retry: WT_RET(__cursor_func_init(cbt, 0));
__cursor_position_clear(cbt);
/*
* If we aren't already iterating in the right direction, there's
* some setup to do.
*/
if (!F_ISSET(cbt, WT_CBT_ITERATE_PREV))
__wt_btcur_iterate_setup(cbt, 0);
/*
* If this is a modification, we're about to read information from the
* page, save the write generation.
*/
page = cbt->page;
if (discard && page != NULL) {
WT_ERR(__wt_page_modify_init(session, page));
WT_ORDERED_READ(cbt->write_gen, page->modify->write_gen);
}
/*
* Walk any page we're holding until the underlying call returns not-
* found. Then, move to the previous page, until we reach the start
* of the file.
*/
for (newpage = 0;; newpage = 1) {
if (F_ISSET(cbt, WT_CBT_ITERATE_APPEND)) {
switch (page->type) {
case WT_PAGE_COL_FIX:
ret = __cursor_fix_append_prev(cbt, newpage);
break;
case WT_PAGE_COL_VAR:
ret = __cursor_var_append_prev(cbt, newpage);
break;
WT_ILLEGAL_VALUE_ERR(session);
}
if (ret == 0)
break;
F_CLR(cbt, WT_CBT_ITERATE_APPEND);
if (ret != WT_NOTFOUND)
break;
newpage = 1;
}
if (page != NULL) {
switch (page->type) {
case WT_PAGE_COL_FIX:
ret = __cursor_fix_prev(cbt, newpage);
break;
case WT_PAGE_COL_VAR:
ret = __cursor_var_prev(cbt, newpage);
break;
case WT_PAGE_ROW_LEAF:
ret = __cursor_row_prev(cbt, newpage);
break;
WT_ILLEGAL_VALUE_ERR(session);
}
if (ret != WT_NOTFOUND)
break;
}
cbt->page = NULL;
WT_ERR(__wt_tree_walk(session, &page, flags));
WT_ERR_TEST(page == NULL, WT_NOTFOUND);
WT_ASSERT(session,
page->type != WT_PAGE_COL_INT &&
page->type != WT_PAGE_ROW_INT);
cbt->page = page;
/* Initialize the page's modification information */
if (discard) {
WT_ERR(__wt_page_modify_init(session, page));
WT_ORDERED_READ(
cbt->write_gen, page->modify->write_gen);
}
/*
* The last page in a column-store has appended entries.
* We handle it separately from the usual cursor code:
* it's only that one page and it's in a simple format.
*/
if (page->type != WT_PAGE_ROW_LEAF &&
(cbt->ins_head = WT_COL_APPEND(page)) != NULL)
F_SET(cbt, WT_CBT_ITERATE_APPEND);
}
err: if (ret == WT_RESTART)
goto retry;
WT_TRET(__cursor_func_resolve(cbt, ret));
return (ret);
}
/*
* __wt_btcur_prev --
* Move to the previous record in the tree.
*/
int
__wt_btcur_prev(WT_CURSOR_BTREE *cbt)
{
WT_DECL_RET;
WT_SESSION_IMPL *session;
int newpage;
session = (WT_SESSION_IMPL *)cbt->iface.session;
WT_BSTAT_INCR(session, cursor_read_prev);
__cursor_func_init(cbt, 0);
/*
* If we aren't already iterating in the right direction, there's
* some setup to do.
*/
if (!F_ISSET(cbt, WT_CBT_ITERATE_PREV))
__wt_btcur_iterate_setup(cbt, 0);
/*
* Walk any page we're holding until the underlying call returns not-
* found. Then, move to the previous page, until we reach the start
* of the file.
*/
for (newpage = 0;; newpage = 1) {
if (F_ISSET(cbt, WT_CBT_ITERATE_APPEND)) {
switch (cbt->page->type) {
case WT_PAGE_COL_FIX:
ret = __cursor_fix_append_prev(cbt, newpage);
break;
case WT_PAGE_COL_VAR:
ret = __cursor_var_append_prev(cbt, newpage);
break;
WT_ILLEGAL_VALUE_ERR(session);
}
if (ret == 0)
break;
F_CLR(cbt, WT_CBT_ITERATE_APPEND);
if (ret != WT_NOTFOUND)
break;
newpage = 1;
}
if (cbt->page != NULL) {
switch (cbt->page->type) {
case WT_PAGE_COL_FIX:
ret = __cursor_fix_prev(cbt, newpage);
break;
case WT_PAGE_COL_VAR:
ret = __cursor_var_prev(cbt, newpage);
break;
case WT_PAGE_ROW_LEAF:
ret = __cursor_row_prev(cbt, newpage);
break;
WT_ILLEGAL_VALUE_ERR(session);
}
if (ret != WT_NOTFOUND)
break;
}
do {
WT_ERR(__wt_tree_np(session, &cbt->page, 0, 0));
WT_ERR_TEST(cbt->page == NULL, WT_NOTFOUND);
} while (
cbt->page->type == WT_PAGE_COL_INT ||
cbt->page->type == WT_PAGE_ROW_INT);
/*
* The last page in a column-store has appended entries.
* We handle it separately from the usual cursor code:
* it's only that one page and it's in a simple format.
*/
if (cbt->page->type != WT_PAGE_ROW_LEAF &&
(cbt->ins_head = WT_COL_APPEND(cbt->page)) != NULL)
F_SET(cbt, WT_CBT_ITERATE_APPEND);
}
err: __cursor_func_resolve(cbt, ret);
return (ret);
}
/*
* __wt_btcur_insert --
* Insert a record into the tree.
*/
int
__wt_btcur_insert(WT_CURSOR_BTREE *cbt)
{
WT_BTREE *btree;
WT_CURSOR *cursor;
WT_SESSION_IMPL *session;
int ret;
btree = cbt->btree;
cursor = &cbt->iface;
session = (WT_SESSION_IMPL *)cursor->session;
WT_BSTAT_INCR(session, cursor_inserts);
if (btree->type == BTREE_ROW)
WT_RET(__cursor_size_chk(session, &cursor->key));
WT_RET(__cursor_size_chk(session, &cursor->value));
retry: __cursor_func_init(cbt, 1);
switch (btree->type) {
case BTREE_COL_FIX:
case BTREE_COL_VAR:
/*
* If WT_CURSTD_APPEND is set, insert a new record (ignoring
* the application's record number). First we search for the
* maximum possible record number so the search ends on the
* last page. The real record number is assigned by the
* serialized append operation.
* __wt_col_append_serial_func
*/
if (F_ISSET(cursor, WT_CURSTD_APPEND))
cbt->iface.recno = UINT64_MAX;
WT_ERR(__wt_col_search(session, cbt, 1));
if (F_ISSET(cursor, WT_CURSTD_APPEND))
cbt->iface.recno = 0;
/*
* If WT_CURSTD_OVERWRITE set, insert/update the key/value pair.
*
* If WT_CURSTD_OVERWRITE not set, fail if the key exists, else
* insert the key/value pair. Creating a record past the end
* of the tree in a fixed-length column-store implicitly fills
* the gap with empty records. Fail in that case, the record
* exists.
*/
if (!F_ISSET(cursor, WT_CURSTD_OVERWRITE) &&
((cbt->compare == 0 && !__cursor_invalid(cbt)) ||
(cbt->compare != 0 && __cursor_fix_implicit(btree, cbt)))) {
ret = WT_DUPLICATE_KEY;
break;
}
if ((ret = __wt_col_modify(session, cbt, 3)) == WT_RESTART)
goto retry;
if (F_ISSET(cursor, WT_CURSTD_APPEND) && ret == 0)
cbt->iface.recno = cbt->recno;
break;
case BTREE_ROW:
/*
* If WT_CURSTD_OVERWRITE not set, fail if the key exists, else
* insert the key/value pair.
*
* If WT_CURSTD_OVERWRITE set, insert/update the key/value pair.
*/
WT_ERR(__wt_row_search(session, cbt, 1));
if (cbt->compare == 0 &&
!__cursor_invalid(cbt) &&
!F_ISSET(cursor, WT_CURSTD_OVERWRITE)) {
ret = WT_DUPLICATE_KEY;
break;
}
if ((ret = __wt_row_modify(session, cbt, 0)) == WT_RESTART)
goto retry;
break;
WT_ILLEGAL_VALUE(session);
}
err: __cursor_func_resolve(cbt, ret);
return (ret);
}
/*
* __wt_btcur_search_near --
* Search for a record in the tree.
*/
int
__wt_btcur_search_near(WT_CURSOR_BTREE *cbt, int *exact)
{
WT_BTREE *btree;
WT_ITEM *val;
WT_CURSOR *cursor;
WT_SESSION_IMPL *session;
int ret;
btree = cbt->btree;
cursor = &cbt->iface;
session = (WT_SESSION_IMPL *)cursor->session;
WT_BSTAT_INCR(session, cursor_read_near);
if (btree->type == BTREE_ROW)
WT_RET(__cursor_size_chk(session, &cursor->key));
__cursor_func_init(cbt, 1);
WT_ERR(btree->type == BTREE_ROW ?
__wt_row_search(session, cbt, 0) :
__wt_col_search(session, cbt, 0));
/*
* Creating a record past the end of the tree in a fixed-length column-
* store implicitly fills the gap with empty records. In this case, we
* instantiate the empty record, it's an exact match.
*
* Else, if we find a valid key (one that wasn't deleted), return it.
*
* Else, if we found a deleted key, try to move to the next key in the
* tree (bias for prefix searches). Cursor next skips deleted records,
* so we don't have to test for them again.
*
* Else if there's no larger tree key, redo the search and try and find
* an earlier record. If that fails, quit, there's no record to return.
*/
if (cbt->compare != 0 && __cursor_fix_implicit(btree, cbt)) {
cbt->v = 0;
val = &cbt->iface.value;
val->data = &cbt->v;
val->size = 1;
*exact = 0;
} else if (!__cursor_invalid(cbt)) {
*exact = cbt->compare;
ret = __wt_kv_return(session, cbt, cbt->compare == 0 ? 0 : 1);
} else if ((ret = __wt_btcur_next(cbt)) != WT_NOTFOUND)
*exact = 1;
else {
WT_ERR(btree->type == BTREE_ROW ?
__wt_row_search(session, cbt, 0) :
__wt_col_search(session, cbt, 0));
if (!__cursor_invalid(cbt)) {
*exact = cbt->compare;
ret = __wt_kv_return(
session, cbt, cbt->compare == 0 ? 0 : 1);
} else if ((ret = __wt_btcur_prev(cbt)) != WT_NOTFOUND)
*exact = -1;
}
err: __cursor_func_resolve(cbt, ret);
return (ret);
}