static ham_status_t
__insert_nosplit(ham_page_t *page, ham_key_t *key,
ham_offset_t rid, ham_record_t *record,
ham_bt_cursor_t *cursor, insert_hints_t *hints)
{
ham_status_t st;
ham_u16_t count;
ham_size_t keysize;
ham_size_t new_dupe_id = 0;
int_key_t *bte = 0;
btree_node_t *node;
ham_db_t *db=page_get_owner(page);
ham_bool_t exists = HAM_FALSE;
ham_s32_t slot;
ham_assert(page_get_owner(page), (0));
ham_assert(device_get_env(page_get_device(page)) == db_get_env(page_get_owner(page)), (0));
node=ham_page_get_btree_node(page);
count=btree_node_get_count(node);
keysize=db_get_keysize(db);
if (btree_node_get_count(node)==0)
{
slot = 0;
}
else if (hints->force_append)
{
slot = count;
}
else if (hints->force_prepend)
{
/* insert at beginning; shift all up by one */
slot = 0;
}
else
{
int cmp;
hints->cost++;
st=btree_get_slot(db, page, key, &slot, &cmp);
if (st)
return (st);
/* insert the new key at the beginning? */
if (slot == -1)
{
slot = 0;
}
else
{
/*
* key exists already
*/
if (cmp == 0)
{
if (hints->flags & HAM_OVERWRITE)
{
/*
* no need to overwrite the key - it already exists!
* however, we have to overwrite the data!
*/
if (!btree_node_is_leaf(node))
return (HAM_SUCCESS);
}
else if (!(hints->flags & HAM_DUPLICATE))
return (HAM_DUPLICATE_KEY);
/* do NOT shift keys up to make room; just overwrite the current [slot] */
exists = HAM_TRUE;
}
else
{
/*
* otherwise, if the new key is > then the slot key, move to
* the next slot
*/
if (cmp > 0)
{
slot++;
}
}
}
}
/*
* in any case, uncouple the cursors and see if we must shift any elements to the
* right
*/
bte=btree_node_get_key(db, node, slot);
ham_assert(bte, (0));
if (!exists)
{
if (count > slot)
{
/* uncouple all cursors & shift any elements following [slot] */
st=bt_uncouple_all_cursors(page, slot);
if (st)
return (st);
hints->cost += stats_memmove_cost((db_get_int_key_header_size()+keysize)*(count-slot));
memmove(((char *)bte)+db_get_int_key_header_size()+keysize, bte,
(db_get_int_key_header_size()+keysize)*(count-slot));
}
/*
* if a new key is created or inserted: initialize it with zeroes
*/
memset(bte, 0, db_get_int_key_header_size()+keysize);
}
/*
* if we're in the leaf: insert, overwrite or append the blob
* (depends on the flags)
*/
if (btree_node_is_leaf(node)) {
ham_status_t st;
hints->cost++;
st=key_set_record(db, bte, record,
cursor
? bt_cursor_get_dupe_id(cursor)
: 0,
hints->flags, &new_dupe_id);
if (st)
return (st);
hints->processed_leaf_page = page;
hints->processed_slot = slot;
}
else {
key_set_ptr(bte, rid);
}
page_set_dirty(page, db_get_env(db));
key_set_size(bte, key->size);
/*
* set a flag if the key is extended, and does not fit into the
* btree
*/
if (key->size > db_get_keysize(db))
key_set_flags(bte, key_get_flags(bte)|KEY_IS_EXTENDED);
/*
* if we have a cursor: couple it to the new key
*
* the cursor always points to NIL.
*/
if (cursor)
{
if ((st=bt_cursor_set_to_nil(cursor)))
return (st);
ham_assert(!(bt_cursor_get_flags(cursor)&BT_CURSOR_FLAG_UNCOUPLED),
("coupling an uncoupled cursor, but need a nil-cursor"));
ham_assert(!(bt_cursor_get_flags(cursor)&BT_CURSOR_FLAG_COUPLED),
("coupling a coupled cursor, but need a nil-cursor"));
bt_cursor_set_flags(cursor,
bt_cursor_get_flags(cursor)|BT_CURSOR_FLAG_COUPLED);
bt_cursor_set_coupled_page(cursor, page);
bt_cursor_set_coupled_index(cursor, slot);
bt_cursor_set_dupe_id(cursor, new_dupe_id);
memset(bt_cursor_get_dupe_cache(cursor), 0, sizeof(dupe_entry_t));
page_add_cursor(page, (ham_cursor_t *)cursor);
}
/*
* if we've overwritten a key: no need to continue, we're done
*/
if (exists)
return (0);
/*
* we insert the extended key, if necessary
*/
key_set_key(bte, key->data,
db_get_keysize(db) < key->size ? db_get_keysize(db) : key->size);
/*
* if we need an extended key, allocate a blob and store
* the blob-id in the key
*/
if (key->size > db_get_keysize(db))
{
ham_offset_t blobid;
key_set_key(bte, key->data, db_get_keysize(db));
st=key_insert_extended(&blobid, db, page, key);
ham_assert(st ? blobid == 0 : 1, (0));
if (!blobid)
return st ? st : HAM_INTERNAL_ERROR;
key_set_extended_rid(db, bte, blobid);
}
/*
* update the btree node-header
*/
btree_node_set_count(node, count+1);
return (0);
}
static ham_status_t
__insert_split(ham_page_t *page, ham_key_t *key,
ham_offset_t rid, insert_scratchpad_t *scratchpad,
insert_hints_t *hints)
{
int cmp;
ham_status_t st;
ham_page_t *newpage, *oldsib;
int_key_t *nbte, *obte;
btree_node_t *nbtp, *obtp, *sbtp;
ham_size_t count, keysize;
ham_db_t *db=page_get_owner(page);
ham_env_t *env = db_get_env(db);
ham_key_t pivotkey, oldkey;
ham_offset_t pivotrid;
ham_u16_t pivot;
ham_bool_t pivot_at_end=HAM_FALSE;
ham_assert(page_get_owner(page), (0));
ham_assert(device_get_env(page_get_device(page))
== db_get_env(page_get_owner(page)), (0));
ham_assert(hints->force_append == HAM_FALSE, (0));
keysize=db_get_keysize(db);
/*
* allocate a new page
*/
hints->cost++;
st=db_alloc_page(&newpage, db, PAGE_TYPE_B_INDEX, 0);
ham_assert(st ? page == NULL : 1, (0));
ham_assert(!st ? page != NULL : 1, (0));
if (st)
return st;
ham_assert(page_get_owner(newpage), (""));
/* clear the node header */
memset(page_get_payload(newpage), 0, sizeof(btree_node_t));
stats_page_is_nuked(db, page, HAM_TRUE);
/*
* move half of the key/rid-tuples to the new page
*
* !! recno: keys are sorted; we do a "lazy split"
*/
nbtp=ham_page_get_btree_node(newpage);
nbte=btree_node_get_key(db, nbtp, 0);
obtp=ham_page_get_btree_node(page);
obte=btree_node_get_key(db, obtp, 0);
count=btree_node_get_count(obtp);
/*
* for databases with sequential access (this includes recno databases):
* do not split in the middle, but at the very end of the page
*
* if this page is the right-most page in the index, and this key is
* inserted at the very end, then we select the same pivot as for
* sequential access
*/
if (db_get_data_access_mode(db)&HAM_DAM_SEQUENTIAL_INSERT)
pivot_at_end=HAM_TRUE;
else if (btree_node_get_right(obtp)==0) {
cmp=key_compare_pub_to_int(db, page, key, btree_node_get_count(obtp)-1);
if (cmp>0)
pivot_at_end=HAM_TRUE;
}
/*
* internal pages set the count of the new page to count-pivot-1 (because
* the pivot element will become ptr_left of the new page).
* by using pivot=count-2 we make sure that at least 1 element will remain
* in the new node.
*/
if (pivot_at_end) {
pivot=count-2;
}
else {
pivot=count/2;
}
/*
* uncouple all cursors
*/
st=bt_uncouple_all_cursors(page, pivot);
if (st)
return (st);
/*
* if we split a leaf, we'll insert the pivot element in the leaf
* page, too. in internal nodes, we don't insert it, but propagate
* it to the parent node only.
*/
if (btree_node_is_leaf(obtp)) {
hints->cost += stats_memmove_cost((db_get_int_key_header_size()+keysize)*(count-pivot));
memcpy((char *)nbte,
((char *)obte)+(db_get_int_key_header_size()+keysize)*pivot,
(db_get_int_key_header_size()+keysize)*(count-pivot));
}
else {
hints->cost += stats_memmove_cost((db_get_int_key_header_size()+keysize)*(count-pivot-1));
memcpy((char *)nbte,
((char *)obte)+(db_get_int_key_header_size()+keysize)*(pivot+1),
(db_get_int_key_header_size()+keysize)*(count-pivot-1));
}
/*
* store the pivot element, we'll need it later to propagate it
* to the parent page
*/
nbte=btree_node_get_key(db, obtp, pivot);
memset(&pivotkey, 0, sizeof(pivotkey));
memset(&oldkey, 0, sizeof(oldkey));
oldkey.data=key_get_key(nbte);
oldkey.size=key_get_size(nbte);
oldkey._flags=key_get_flags(nbte);
st = util_copy_key(db, &oldkey, &pivotkey);
if (st)
{
(void)db_free_page(newpage, DB_MOVE_TO_FREELIST);
goto fail_dramatically;
}
pivotrid=page_get_self(newpage);
/*
* adjust the page count
*/
if (btree_node_is_leaf(obtp)) {
btree_node_set_count(obtp, pivot);
btree_node_set_count(nbtp, count-pivot);
}
else {
btree_node_set_count(obtp, pivot);
btree_node_set_count(nbtp, count-pivot-1);
}
/*
* if we're in an internal page: fix the ptr_left of the new page
* (it points to the ptr of the pivot key)
*/
if (!btree_node_is_leaf(obtp)) {
/*
* nbte still contains the pivot key
*/
btree_node_set_ptr_left(nbtp, key_get_ptr(nbte));
}
/*
* insert the new element
*/
hints->cost++;
cmp=key_compare_pub_to_int(db, page, key, pivot);
if (cmp < -1)
{
st = (ham_status_t)cmp;
goto fail_dramatically;
}
if (cmp>=0)
st=__insert_nosplit(newpage, key, rid,
scratchpad->record, scratchpad->cursor, hints);
else
st=__insert_nosplit(page, key, rid,
scratchpad->record, scratchpad->cursor, hints);
if (st)
{
goto fail_dramatically;
}
scratchpad->cursor=0; /* don't overwrite cursor if __insert_nosplit
is called again */
/*
* fix the double-linked list of pages, and mark the pages as dirty
*/
if (btree_node_get_right(obtp))
{
st=db_fetch_page(&oldsib, db, btree_node_get_right(obtp), 0);
if (st)
goto fail_dramatically;
}
else
{
oldsib=0;
}
if (oldsib) {
st=ham_log_add_page_before(oldsib);
if (st)
goto fail_dramatically;
}
btree_node_set_left (nbtp, page_get_self(page));
btree_node_set_right(nbtp, btree_node_get_right(obtp));
btree_node_set_right(obtp, page_get_self(newpage));
if (oldsib) {
sbtp=ham_page_get_btree_node(oldsib);
btree_node_set_left(sbtp, page_get_self(newpage));
page_set_dirty(oldsib, env);
}
page_set_dirty(newpage, env);
page_set_dirty(page, env);
/*
* propagate the pivot key to the parent page
*/
ham_assert(!(scratchpad->key.flags & HAM_KEY_USER_ALLOC), (0));
if (scratchpad->key.data)
allocator_free(env_get_allocator(env), scratchpad->key.data);
scratchpad->key=pivotkey;
scratchpad->rid=pivotrid;
ham_assert(!(scratchpad->key.flags & HAM_KEY_USER_ALLOC), (0));
return (SPLIT);
fail_dramatically:
ham_assert(!(pivotkey.flags & HAM_KEY_USER_ALLOC), (0));
if (pivotkey.data)
allocator_free(env_get_allocator(env), pivotkey.data);
return st;
}
/**
* uncouple all cursors from a page
*
* @remark this is called whenever the page is deleted or
* becoming invalid
@note This is a B+-tree 'backend' method.
*/
static ham_status_t
my_fun_uncouple_all_cursors(ham_btree_t *be, ham_page_t *page, ham_size_t start)
{
return (bt_uncouple_all_cursors(page, start));
}
