/***************************************************************
Stores to the heap the row on which the node->pcur is positioned. */
static
void
row_upd_store_row(
/*==============*/
upd_node_t* node) /* in: row update node */
{
dict_index_t* clust_index;
upd_t* update;
rec_t* rec;
ut_ad(node->pcur->latch_mode != BTR_NO_LATCHES);
if (node->row != NULL) {
mem_heap_empty(node->heap);
node->row = NULL;
}
clust_index = dict_table_get_first_index(node->table);
rec = btr_pcur_get_rec(node->pcur);
node->row = row_build(ROW_COPY_DATA, clust_index, rec, node->heap);
node->ext_vec = mem_heap_alloc(node->heap, sizeof(ulint)
* rec_get_n_fields(rec));
if (node->is_delete) {
update = NULL;
} else {
update = node->update;
}
node->n_ext_vec = btr_push_update_extern_fields(node->ext_vec,
rec, update);
}
/***************************************************************//**
Searches an index record.
@return TRUE if found */
UNIV_INTERN
ibool
row_search_index_entry(
/*===================*/
dict_index_t* index, /*!< in: index */
const dtuple_t* entry, /*!< in: index entry */
ulint mode, /*!< in: BTR_MODIFY_LEAF, ... */
btr_pcur_t* pcur, /*!< in/out: persistent cursor, which must
be closed by the caller */
mtr_t* mtr) /*!< in: mtr */
{
ulint n_fields;
ulint low_match;
rec_t* rec;
ut_ad(dtuple_check_typed(entry));
btr_pcur_open(index, entry, PAGE_CUR_LE, mode, pcur, mtr);
low_match = btr_pcur_get_low_match(pcur);
rec = btr_pcur_get_rec(pcur);
n_fields = dtuple_get_n_fields(entry);
return(!page_rec_is_infimum(rec) && low_match == n_fields);
}
ibool
row_undo_search_clust_to_pcur(
/*==========================*/
/* out: TRUE if found; NOTE the node->pcur
must be closed by the caller, regardless of
the return value */
undo_node_t* node) /* in: row undo node */
{
dict_index_t* clust_index;
ibool found;
mtr_t mtr;
ibool ret;
rec_t* rec;
mem_heap_t* heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
ulint* offsets = offsets_;
*offsets_ = (sizeof offsets_) / sizeof *offsets_;
mtr_start(&mtr);
clust_index = dict_table_get_first_index(node->table);
found = row_search_on_row_ref(&(node->pcur), BTR_MODIFY_LEAF,
node->table, node->ref, &mtr);
rec = btr_pcur_get_rec(&(node->pcur));
offsets = rec_get_offsets(rec, clust_index, offsets,
ULINT_UNDEFINED, &heap);
if (!found || 0 != ut_dulint_cmp(node->roll_ptr,
row_get_rec_roll_ptr(rec, clust_index,
offsets))) {
/* We must remove the reservation on the undo log record
BEFORE releasing the latch on the clustered index page: this
is to make sure that some thread will eventually undo the
modification corresponding to node->roll_ptr. */
/* fputs("--------------------undoing a previous version\n",
stderr); */
ret = FALSE;
} else {
node->row = row_build(ROW_COPY_DATA, clust_index, rec,
offsets, node->heap);
btr_pcur_store_position(&(node->pcur), &mtr);
ret = TRUE;
}
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
if (UNIV_LIKELY_NULL(heap)) {
mem_heap_free(heap);
}
return(ret);
}
/***************************************************************//**
Creates an index tree for the index if it is not a member of a cluster.
@return DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
static
ulint
dict_create_index_tree_step(
/*========================*/
ind_node_t* node) /*!< in: index create node */
{
dict_index_t* index;
dict_table_t* sys_indexes;
dict_table_t* table;
dtuple_t* search_tuple;
ulint zip_size;
btr_pcur_t pcur;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
index = node->index;
table = node->table;
sys_indexes = dict_sys->sys_indexes;
/* Run a mini-transaction in which the index tree is allocated for
the index and its root address is written to the index entry in
sys_indexes */
mtr_start(&mtr);
search_tuple = dict_create_search_tuple(node->ind_row, node->heap);
btr_pcur_open(UT_LIST_GET_FIRST(sys_indexes->indexes),
search_tuple, PAGE_CUR_L, BTR_MODIFY_LEAF,
&pcur, &mtr);
btr_pcur_move_to_next_user_rec(&pcur, &mtr);
zip_size = dict_table_zip_size(index->table);
node->page_no = btr_create(index->type, index->space, zip_size,
index->id, index, &mtr);
/* printf("Created a new index tree in space %lu root page %lu\n",
index->space, index->page_no); */
page_rec_write_index_page_no(btr_pcur_get_rec(&pcur),
DICT_SYS_INDEXES_PAGE_NO_FIELD,
node->page_no, &mtr);
btr_pcur_close(&pcur);
mtr_commit(&mtr);
if (node->page_no == FIL_NULL) {
return(DB_OUT_OF_FILE_SPACE);
}
return(DB_SUCCESS);
}
/*************************************************************************
Checks if possible foreign key constraints hold after a delete of the record
under pcur. NOTE that this function will temporarily commit mtr and lose
pcur position! */
static
ulint
row_upd_check_references_constraints(
/*=================================*/
/* out: DB_SUCCESS, DB_LOCK_WAIT, or an error
code */
btr_pcur_t* pcur, /* in: cursor positioned on a record; NOTE: the
cursor position is lost in this function! */
dict_table_t* table, /* in: table in question */
dict_index_t* index, /* in: index of the cursor */
que_thr_t* thr, /* in: query thread */
mtr_t* mtr) /* in: mtr */
{
dict_foreign_t* foreign;
mem_heap_t* heap;
dtuple_t* entry;
rec_t* rec;
ulint err;
rec = btr_pcur_get_rec(pcur);
heap = mem_heap_create(500);
entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec, heap);
mtr_commit(mtr);
mtr_start(mtr);
rw_lock_s_lock(&dict_foreign_key_check_lock);
foreign = UT_LIST_GET_FIRST(table->referenced_list);
while (foreign) {
if (foreign->referenced_index == index) {
err = row_ins_check_foreign_constraint(FALSE, foreign,
table, index, entry, thr);
if (err != DB_SUCCESS) {
rw_lock_s_unlock(&dict_foreign_key_check_lock);
mem_heap_free(heap);
return(err);
}
}
foreign = UT_LIST_GET_NEXT(referenced_list, foreign);
}
rw_lock_s_unlock(&dict_foreign_key_check_lock);
mem_heap_free(heap);
return(DB_SUCCESS);
}
void
row_upd_in_place_in_select(
/*=======================*/
sel_node_t* sel_node, /* in: select node */
que_thr_t* thr, /* in: query thread */
mtr_t* mtr) /* in: mtr */
{
upd_node_t* node;
btr_pcur_t* pcur;
btr_cur_t* btr_cur;
ulint err;
ut_ad(sel_node->select_will_do_update);
ut_ad(sel_node->latch_mode == BTR_MODIFY_LEAF);
ut_ad(sel_node->asc);
node = que_node_get_parent(sel_node);
ut_ad(que_node_get_type(node) == QUE_NODE_UPDATE);
pcur = node->pcur;
btr_cur = btr_pcur_get_btr_cur(pcur);
/* Copy the necessary columns from clust_rec and calculate the new
values to set */
row_upd_copy_columns(btr_pcur_get_rec(pcur),
UT_LIST_GET_FIRST(node->columns));
row_upd_eval_new_vals(node->update);
ut_ad(FALSE == rec_get_deleted_flag(btr_pcur_get_rec(pcur)));
ut_ad(node->cmpl_info & UPD_NODE_NO_SIZE_CHANGE);
ut_ad(node->cmpl_info & UPD_NODE_NO_ORD_CHANGE);
ut_ad(node->select_will_do_update);
err = btr_cur_update_in_place(BTR_NO_LOCKING_FLAG, btr_cur,
node->update, node->cmpl_info,
thr, mtr);
ut_ad(err == DB_SUCCESS);
}
/***************************************************************
Delete marks a clustered index record. */
static
ulint
row_upd_del_mark_clust_rec(
/*=======================*/
/* out: DB_SUCCESS if operation successfully
completed, else error code or DB_LOCK_WAIT */
upd_node_t* node, /* in: row update node */
dict_index_t* index, /* in: clustered index */
que_thr_t* thr, /* in: query thread */
ibool check_ref,/* in: TRUE if index may be referenced in
a foreign key constraint */
mtr_t* mtr) /* in: mtr; gets committed here */
{
btr_pcur_t* pcur;
btr_cur_t* btr_cur;
ulint err;
ut_ad(node);
ut_ad(index->type & DICT_CLUSTERED);
ut_ad(node->is_delete);
pcur = node->pcur;
btr_cur = btr_pcur_get_btr_cur(pcur);
ut_ad(FALSE == rec_get_deleted_flag(btr_pcur_get_rec(pcur)));
/* Store row because we have to build also the secondary index
entries */
row_upd_store_row(node);
/* Mark the clustered index record deleted; we do not have to check
locks, because we assume that we have an x-lock on the record */
err = btr_cur_del_mark_set_clust_rec(BTR_NO_LOCKING_FLAG, btr_cur,
TRUE, thr, mtr);
if (err == DB_SUCCESS && check_ref) {
/* NOTE that the following call loses
the position of pcur ! */
err = row_upd_check_references_constraints(pcur, index->table,
index, thr, mtr);
if (err != DB_SUCCESS) {
mtr_commit(mtr);
return(err);
}
}
mtr_commit(mtr);
return(err);
}
/***************************************************************//**
Searches an index record.
@return whether the record was found or buffered */
UNIV_INTERN
enum row_search_result
row_search_index_entry(
/*===================*/
dict_index_t* index, /*!< in: index */
const dtuple_t* entry, /*!< in: index entry */
ulint mode, /*!< in: BTR_MODIFY_LEAF, ... */
btr_pcur_t* pcur, /*!< in/out: persistent cursor, which must
be closed by the caller */
mtr_t* mtr) /*!< in: mtr */
{
ulint n_fields;
ulint low_match;
rec_t* rec;
ut_ad(dtuple_check_typed(entry));
btr_pcur_open(index, entry, PAGE_CUR_LE, mode, pcur, mtr);
switch (btr_pcur_get_btr_cur(pcur)->flag) {
case BTR_CUR_DELETE_REF:
ut_a(mode & BTR_DELETE);
return(ROW_NOT_DELETED_REF);
case BTR_CUR_DEL_MARK_IBUF:
case BTR_CUR_DELETE_IBUF:
case BTR_CUR_INSERT_TO_IBUF:
return(ROW_BUFFERED);
case BTR_CUR_HASH:
case BTR_CUR_HASH_FAIL:
case BTR_CUR_BINARY:
break;
}
low_match = btr_pcur_get_low_match(pcur);
rec = btr_pcur_get_rec(pcur);
n_fields = dtuple_get_n_fields(entry);
if (page_rec_is_infimum(rec)) {
return(ROW_NOT_FOUND);
} else if (low_match != n_fields) {
return(ROW_NOT_FOUND);
}
return(ROW_FOUND);
}
ibool
row_undo_search_clust_to_pcur(
/*==========================*/
/* out: TRUE if found; NOTE the node->pcur
must be closed by the caller, regardless of
the return value */
undo_node_t* node, /* in: row undo node */
que_thr_t* thr) /* in: query thread */
{
dict_index_t* clust_index;
ibool found;
mtr_t mtr;
ibool ret;
rec_t* rec;
UT_NOT_USED(thr);
mtr_start(&mtr);
clust_index = dict_table_get_first_index(node->table);
found = row_search_on_row_ref(&(node->pcur), BTR_MODIFY_LEAF,
node->table, node->ref, &mtr);
rec = btr_pcur_get_rec(&(node->pcur));
if (!found || 0 != ut_dulint_cmp(node->roll_ptr,
row_get_rec_roll_ptr(rec, clust_index))) {
/* We must remove the reservation on the undo log record
BEFORE releasing the latch on the clustered index page: this
is to make sure that some thread will eventually undo the
modification corresponding to node->roll_ptr. */
/* printf("--------------------undoing a previous version\n");
*/
ret = FALSE;
} else {
node->row = row_build(ROW_COPY_DATA, clust_index, rec,
node->heap);
btr_pcur_store_position(&(node->pcur), &mtr);
ret = TRUE;
}
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
return(ret);
}
rec_t*
row_get_clust_rec(
/*==============*/
/* out: record or NULL, if no record found */
ulint mode, /* in: BTR_MODIFY_LEAF, ... */
rec_t* rec, /* in: record in a secondary index */
dict_index_t* index, /* in: secondary index */
dict_index_t** clust_index,/* out: clustered index */
mtr_t* mtr) /* in: mtr */
{
mem_heap_t* heap;
dtuple_t* ref;
dict_table_t* table;
btr_pcur_t pcur;
ibool found;
rec_t* clust_rec;
ut_ad((index->type & DICT_CLUSTERED) == 0);
table = index->table;
heap = mem_heap_create(256);
ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec, heap);
found = row_search_on_row_ref(&pcur, mode, table, ref, mtr);
clust_rec = btr_pcur_get_rec(&pcur);
mem_heap_free(heap);
btr_pcur_close(&pcur);
*clust_index = dict_table_get_first_index(table);
if (!found) {
return(NULL);
}
return(clust_rec);
}
ibool
row_search_on_row_ref(
/*==================*/
/* out: TRUE if found */
btr_pcur_t* pcur, /* in/out: persistent cursor, which must
be closed by the caller */
ulint mode, /* in: BTR_MODIFY_LEAF, ... */
dict_table_t* table, /* in: table */
dtuple_t* ref, /* in: row reference */
mtr_t* mtr) /* in: mtr */
{
ulint low_match;
rec_t* rec;
dict_index_t* index;
page_t* page;
ut_ad(dtuple_check_typed(ref));
index = dict_table_get_first_index(table);
ut_a(dtuple_get_n_fields(ref) == dict_index_get_n_unique(index));
btr_pcur_open(index, ref, PAGE_CUR_LE, mode, pcur, mtr);
low_match = btr_pcur_get_low_match(pcur);
rec = btr_pcur_get_rec(pcur);
page = buf_frame_align(rec);
if (rec == page_get_infimum_rec(page)) {
return(FALSE);
}
if (low_match != dtuple_get_n_fields(ref)) {
return(FALSE);
}
return(TRUE);
}
ibool
row_search_index_entry(
/*===================*/
/* out: TRUE if found */
dict_index_t* index, /* in: index */
dtuple_t* entry, /* in: index entry */
ulint mode, /* in: BTR_MODIFY_LEAF, ... */
btr_pcur_t* pcur, /* in/out: persistent cursor, which must
be closed by the caller */
mtr_t* mtr) /* in: mtr */
{
ulint n_fields;
ulint low_match;
page_t* page;
rec_t* rec;
ut_ad(dtuple_check_typed(entry));
btr_pcur_open(index, entry, PAGE_CUR_LE, mode, pcur, mtr);
low_match = btr_pcur_get_low_match(pcur);
rec = btr_pcur_get_rec(pcur);
page = buf_frame_align(rec);
n_fields = dtuple_get_n_fields(entry);
if (rec == page_get_infimum_rec(page)) {
return(FALSE);
}
if (low_match != n_fields) {
/* Not found */
return(FALSE);
}
return(TRUE);
}
/***********************************************************//**
Determines if it is possible to remove a secondary index entry.
Removal is possible if the secondary index entry does not refer to any
not delete marked version of a clustered index record where DB_TRX_ID
is newer than the purge view.
NOTE: This function should only be called by the purge thread, only
while holding a latch on the leaf page of the secondary index entry
(or keeping the buffer pool watch on the page). It is possible that
this function first returns TRUE and then FALSE, if a user transaction
inserts a record that the secondary index entry would refer to.
However, in that case, the user transaction would also re-insert the
secondary index entry after purge has removed it and released the leaf
page latch.
@return TRUE if the secondary index record can be purged */
UNIV_INTERN
ibool
row_purge_poss_sec(
/*===============*/
purge_node_t* node, /*!< in/out: row purge node */
dict_index_t* index, /*!< in: secondary index */
const dtuple_t* entry) /*!< in: secondary index entry */
{
ibool can_delete;
mtr_t mtr;
ut_ad(!dict_index_is_clust(index));
mtr_start(&mtr);
can_delete = !row_purge_reposition_pcur(BTR_SEARCH_LEAF, node, &mtr)
|| !row_vers_old_has_index_entry(TRUE,
btr_pcur_get_rec(&node->pcur),
&mtr, index, entry);
btr_pcur_commit_specify_mtr(&node->pcur, &mtr);
return(can_delete);
}
/*********************************************************************//**
Fetches the clustered index record for a secondary index record. The latches
on the secondary index record are preserved.
@return record or NULL, if no record found */
UNIV_INTERN
rec_t*
row_get_clust_rec(
/*==============*/
ulint mode, /*!< in: BTR_MODIFY_LEAF, ... */
const rec_t* rec, /*!< in: record in a secondary index */
dict_index_t* index, /*!< in: secondary index */
dict_index_t** clust_index,/*!< out: clustered index */
mtr_t* mtr) /*!< in: mtr */
{
mem_heap_t* heap;
dtuple_t* ref;
dict_table_t* table;
btr_pcur_t pcur;
ibool found;
rec_t* clust_rec;
ut_ad(!dict_index_is_clust(index));
table = index->table;
heap = mem_heap_create(256);
ref = row_build_row_ref(ROW_COPY_POINTERS, index, rec, heap);
found = row_search_on_row_ref(&pcur, mode, table, ref, mtr);
clust_rec = found ? btr_pcur_get_rec(&pcur) : NULL;
mem_heap_free(heap);
btr_pcur_close(&pcur);
*clust_index = dict_table_get_first_index(table);
return(clust_rec);
}
ibool
btr_pcur_restore_position(
/*======================*/
/* out: TRUE if the cursor position
was stored when it was on a user record
and it can be restored on a user record
whose ordering fields are identical to
the ones of the original user record */
ulint latch_mode, /* in: BTR_SEARCH_LEAF, ... */
btr_pcur_t* cursor, /* in: detached persistent cursor */
mtr_t* mtr) /* in: mtr */
{
dict_index_t* index;
page_t* page;
dtuple_t* tuple;
ulint mode;
ulint old_mode;
mem_heap_t* heap;
index = btr_cur_get_index(btr_pcur_get_btr_cur(cursor));
if (UNIV_UNLIKELY(cursor->old_stored != BTR_PCUR_OLD_STORED)
|| UNIV_UNLIKELY(cursor->pos_state != BTR_PCUR_WAS_POSITIONED
&& cursor->pos_state != BTR_PCUR_IS_POSITIONED)) {
ut_print_buf(stderr, cursor, sizeof(btr_pcur_t));
if (cursor->trx_if_known) {
trx_print(stderr, cursor->trx_if_known, 0);
}
ut_error;
}
if (UNIV_UNLIKELY(
cursor->rel_pos == BTR_PCUR_AFTER_LAST_IN_TREE
|| cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE)) {
/* In these cases we do not try an optimistic restoration,
but always do a search */
btr_cur_open_at_index_side(
cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE,
index, latch_mode, btr_pcur_get_btr_cur(cursor), mtr);
cursor->block_when_stored
= buf_block_align(btr_pcur_get_page(cursor));
return(FALSE);
}
ut_a(cursor->old_rec);
ut_a(cursor->old_n_fields);
page = btr_cur_get_page(btr_pcur_get_btr_cur(cursor));
if (UNIV_LIKELY(latch_mode == BTR_SEARCH_LEAF)
|| UNIV_LIKELY(latch_mode == BTR_MODIFY_LEAF)) {
/* Try optimistic restoration */
if (UNIV_LIKELY(buf_page_optimistic_get(
latch_mode,
cursor->block_when_stored, page,
cursor->modify_clock, mtr))) {
cursor->pos_state = BTR_PCUR_IS_POSITIONED;
#ifdef UNIV_SYNC_DEBUG
buf_page_dbg_add_level(page, SYNC_TREE_NODE);
#endif /* UNIV_SYNC_DEBUG */
if (cursor->rel_pos == BTR_PCUR_ON) {
#ifdef UNIV_DEBUG
rec_t* rec;
ulint* offsets1;
ulint* offsets2;
#endif /* UNIV_DEBUG */
cursor->latch_mode = latch_mode;
#ifdef UNIV_DEBUG
rec = btr_pcur_get_rec(cursor);
heap = mem_heap_create(256);
offsets1 = rec_get_offsets(
cursor->old_rec, index, NULL,
cursor->old_n_fields, &heap);
offsets2 = rec_get_offsets(
rec, index, NULL,
cursor->old_n_fields, &heap);
ut_ad(!cmp_rec_rec(cursor->old_rec,
rec, offsets1, offsets2,
index));
mem_heap_free(heap);
#endif /* UNIV_DEBUG */
return(TRUE);
}
return(FALSE);
}
}
/* If optimistic restoration did not succeed, open the cursor anew */
heap = mem_heap_create(256);
tuple = dict_index_build_data_tuple(index, cursor->old_rec,
cursor->old_n_fields, heap);
/* Save the old search mode of the cursor */
old_mode = cursor->search_mode;
switch (cursor->rel_pos) {
case BTR_PCUR_ON:
mode = PAGE_CUR_LE;
break;
case BTR_PCUR_AFTER:
mode = PAGE_CUR_G;
break;
case BTR_PCUR_BEFORE:
mode = PAGE_CUR_L;
break;
default:
ut_error;
mode = 0; /* silence a warning */
}
btr_pcur_open_with_no_init(index, tuple, mode, latch_mode,
cursor, 0, mtr);
/* Restore the old search mode */
cursor->search_mode = old_mode;
if (btr_pcur_is_on_user_rec(cursor, mtr)) {
switch (cursor->rel_pos) {
case BTR_PCUR_ON:
if (!cmp_dtuple_rec(
tuple, btr_pcur_get_rec(cursor),
rec_get_offsets(btr_pcur_get_rec(cursor),
index, NULL,
ULINT_UNDEFINED, &heap))) {
/* We have to store the NEW value for
the modify clock, since the cursor can
now be on a different page! But we can
retain the value of old_rec */
cursor->block_when_stored =
buf_block_align(
btr_pcur_get_page(cursor));
cursor->modify_clock =
buf_block_get_modify_clock(
cursor->block_when_stored);
cursor->old_stored = BTR_PCUR_OLD_STORED;
mem_heap_free(heap);
return(TRUE);
}
break;
case BTR_PCUR_BEFORE:
page_cur_move_to_next(btr_pcur_get_page_cur(cursor));
break;
case BTR_PCUR_AFTER:
page_cur_move_to_prev(btr_pcur_get_page_cur(cursor));
break;
#ifdef UNIV_DEBUG
default:
ut_error;
#endif /* UNIV_DEBUG */
}
}
mem_heap_free(heap);
/* We have to store new position information, modify_clock etc.,
to the cursor because it can now be on a different page, the record
under it may have been removed, etc. */
btr_pcur_store_position(cursor, mtr);
return(FALSE);
}
/*******************************************************************
Removes a clustered index record. The pcur in node was positioned on the
record, now it is detached. */
static
ulint
row_undo_ins_remove_clust_rec(
/*==========================*/
/* out: DB_SUCCESS or DB_OUT_OF_FILE_SPACE */
undo_node_t* node, /* in: undo node */
que_thr_t* thr) /* in: query thread */
{
btr_cur_t* btr_cur;
ibool success;
ulint err;
ulint n_tries = 0;
mtr_t mtr;
UT_NOT_USED(thr);
mtr_start(&mtr);
success = btr_pcur_restore_position(BTR_MODIFY_LEAF, &(node->pcur),
&mtr);
ut_a(success);
if (ut_dulint_cmp(node->table->id, DICT_INDEXES_ID) == 0) {
/* Drop the index tree associated with the row in
SYS_INDEXES table: */
dict_drop_index_tree(btr_pcur_get_rec(&(node->pcur)), &mtr);
mtr_commit(&mtr);
mtr_start(&mtr);
success = btr_pcur_restore_position(BTR_MODIFY_LEAF,
&(node->pcur), &mtr);
ut_a(success);
}
btr_cur = btr_pcur_get_btr_cur(&(node->pcur));
success = btr_cur_optimistic_delete(btr_cur, &mtr);
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
if (success) {
trx_undo_rec_release(node->trx, node->undo_no);
return(DB_SUCCESS);
}
retry:
/* If did not succeed, try pessimistic descent to tree */
mtr_start(&mtr);
success = btr_pcur_restore_position(BTR_MODIFY_TREE,
&(node->pcur), &mtr);
ut_a(success);
btr_cur_pessimistic_delete(&err, FALSE, btr_cur, TRUE, &mtr);
/* The delete operation may fail if we have little
file space left: TODO: easiest to crash the database
and restart with more file space */
if (err == DB_OUT_OF_FILE_SPACE
&& n_tries < BTR_CUR_RETRY_DELETE_N_TIMES) {
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
n_tries++;
os_thread_sleep(BTR_CUR_RETRY_SLEEP_TIME);
goto retry;
}
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
trx_undo_rec_release(node->trx, node->undo_no);
return(err);
}
/************************************************************************
Loads definitions for table indexes. Adds them to the data dictionary cache.
*/
static
void
dict_load_indexes(
/*==============*/
dict_table_t* table, /* in: table */
mem_heap_t* heap) /* in: memory heap for temporary storage */
{
dict_table_t* sys_indexes;
dict_index_t* sys_index;
dict_index_t* index;
btr_pcur_t pcur;
dtuple_t* tuple;
dfield_t* dfield;
rec_t* rec;
byte* field;
ulint len;
ulint name_len;
char* name_buf;
ulint type;
ulint space;
ulint page_no;
ulint n_fields;
byte* buf;
ibool is_sys_table;
dulint id;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
if ((ut_dulint_get_high(table->id) == 0)
&& (ut_dulint_get_low(table->id) < DICT_HDR_FIRST_ID)) {
is_sys_table = TRUE;
} else {
is_sys_table = FALSE;
}
mtr_start(&mtr);
sys_indexes = dict_table_get_low("SYS_INDEXES");
sys_index = UT_LIST_GET_FIRST(sys_indexes->indexes);
tuple = dtuple_create(heap, 1);
dfield = dtuple_get_nth_field(tuple, 0);
buf = mem_heap_alloc(heap, 8);
mach_write_to_8(buf, table->id);
dfield_set_data(dfield, buf, 8);
dict_index_copy_types(tuple, sys_index, 1);
btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
BTR_SEARCH_LEAF, &pcur, &mtr);
for (;;) {
if (!btr_pcur_is_on_user_rec(&pcur, &mtr)) {
break;
}
rec = btr_pcur_get_rec(&pcur);
field = rec_get_nth_field(rec, 0, &len);
ut_ad(len == 8);
if (ut_memcmp(buf, field, len) != 0) {
break;
}
ut_a(!rec_get_deleted_flag(rec));
field = rec_get_nth_field(rec, 1, &len);
ut_ad(len == 8);
id = mach_read_from_8(field);
ut_a(0 == ut_strcmp("NAME",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_indexes), 4))->name));
field = rec_get_nth_field(rec, 4, &name_len);
name_buf = mem_heap_alloc(heap, name_len + 1);
ut_memcpy(name_buf, field, name_len);
name_buf[name_len] = '\0';
field = rec_get_nth_field(rec, 5, &len);
n_fields = mach_read_from_4(field);
field = rec_get_nth_field(rec, 6, &len);
type = mach_read_from_4(field);
field = rec_get_nth_field(rec, 7, &len);
space = mach_read_from_4(field);
ut_a(0 == ut_strcmp("PAGE_NO",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_indexes), 8))->name));
field = rec_get_nth_field(rec, 8, &len);
page_no = mach_read_from_4(field);
if (is_sys_table
&& ((type & DICT_CLUSTERED)
|| ((table == dict_sys->sys_tables)
&& (name_len == ut_strlen("ID_IND"))
&& (0 == ut_memcmp(name_buf, "ID_IND",
name_len))))) {
/* The index was created in memory already in
booting */
} else {
index = dict_mem_index_create(table->name, name_buf,
space, type, n_fields);
index->page_no = page_no;
index->id = id;
dict_load_fields(table, index, heap);
dict_index_add_to_cache(table, index);
}
btr_pcur_move_to_next_user_rec(&pcur, &mtr);
}
btr_pcur_close(&pcur);
mtr_commit(&mtr);
}
/************************************************************************
Loads foreign key constraint col names (also for the referenced table). */
static
void
dict_load_foreign_cols(
/*===================*/
char* id, /* in: foreign constraint id as a null-
terminated string */
dict_foreign_t* foreign)/* in: foreign constraint object */
{
dict_table_t* sys_foreign_cols;
dict_index_t* sys_index;
btr_pcur_t pcur;
dtuple_t* tuple;
dfield_t* dfield;
char* col_name;
rec_t* rec;
byte* field;
ulint len;
ulint i;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
foreign->foreign_col_names = mem_heap_alloc(foreign->heap,
foreign->n_fields * sizeof(void*));
foreign->referenced_col_names = mem_heap_alloc(foreign->heap,
foreign->n_fields * sizeof(void*));
mtr_start(&mtr);
sys_foreign_cols = dict_table_get_low("SYS_FOREIGN_COLS");
sys_index = UT_LIST_GET_FIRST(sys_foreign_cols->indexes);
tuple = dtuple_create(foreign->heap, 1);
dfield = dtuple_get_nth_field(tuple, 0);
dfield_set_data(dfield, id, ut_strlen(id));
dict_index_copy_types(tuple, sys_index, 1);
btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
BTR_SEARCH_LEAF, &pcur, &mtr);
for (i = 0; i < foreign->n_fields; i++) {
rec = btr_pcur_get_rec(&pcur);
ut_a(btr_pcur_is_on_user_rec(&pcur, &mtr));
ut_a(!rec_get_deleted_flag(rec));
field = rec_get_nth_field(rec, 0, &len);
ut_a(len == ut_strlen(id));
ut_a(ut_memcmp(id, field, len) == 0);
field = rec_get_nth_field(rec, 1, &len);
ut_a(len == 4);
ut_a(i == mach_read_from_4(field));
field = rec_get_nth_field(rec, 4, &len);
col_name = mem_heap_alloc(foreign->heap, len + 1);
ut_memcpy(col_name, field, len);
col_name[len] = '\0';
foreign->foreign_col_names[i] = col_name;
field = rec_get_nth_field(rec, 5, &len);
col_name = mem_heap_alloc(foreign->heap, len + 1);
ut_memcpy(col_name, field, len);
col_name[len] = '\0';
foreign->referenced_col_names[i] = col_name;
btr_pcur_move_to_next_user_rec(&pcur, &mtr);
}
btr_pcur_close(&pcur);
mtr_commit(&mtr);
}
/************************************************************************
Loads definitions for index fields. */
static
void
dict_load_fields(
/*=============*/
dict_table_t* table, /* in: table */
dict_index_t* index, /* in: index whose fields to load */
mem_heap_t* heap) /* in: memory heap for temporary storage */
{
dict_table_t* sys_fields;
dict_index_t* sys_index;
btr_pcur_t pcur;
dtuple_t* tuple;
dfield_t* dfield;
char* col_name;
rec_t* rec;
byte* field;
ulint len;
byte* buf;
ulint i;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
UT_NOT_USED(table);
mtr_start(&mtr);
sys_fields = dict_table_get_low("SYS_FIELDS");
sys_index = UT_LIST_GET_FIRST(sys_fields->indexes);
tuple = dtuple_create(heap, 1);
dfield = dtuple_get_nth_field(tuple, 0);
buf = mem_heap_alloc(heap, 8);
mach_write_to_8(buf, index->id);
dfield_set_data(dfield, buf, 8);
dict_index_copy_types(tuple, sys_index, 1);
btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
BTR_SEARCH_LEAF, &pcur, &mtr);
for (i = 0; i < index->n_fields; i++) {
rec = btr_pcur_get_rec(&pcur);
ut_a(btr_pcur_is_on_user_rec(&pcur, &mtr));
ut_a(!rec_get_deleted_flag(rec));
field = rec_get_nth_field(rec, 0, &len);
ut_ad(len == 8);
ut_a(ut_memcmp(buf, field, len) == 0);
field = rec_get_nth_field(rec, 1, &len);
ut_ad(len == 4);
ut_a(i == mach_read_from_4(field));
ut_a(0 == ut_strcmp("COL_NAME",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_fields), 4))->name));
field = rec_get_nth_field(rec, 4, &len);
col_name = mem_heap_alloc(heap, len + 1);
ut_memcpy(col_name, field, len);
col_name[len] = '\0';
dict_mem_index_add_field(index, col_name, 0);
btr_pcur_move_to_next_user_rec(&pcur, &mtr);
}
btr_pcur_close(&pcur);
mtr_commit(&mtr);
}
/***************************************************************
Updates the clustered index record. */
static
ulint
row_upd_clust_step(
/*===============*/
/* out: DB_SUCCESS if operation successfully
completed, DB_LOCK_WAIT in case of a lock wait,
else error code */
upd_node_t* node, /* in: row update node */
que_thr_t* thr) /* in: query thread */
{
dict_index_t* index;
btr_pcur_t* pcur;
ibool success;
ibool check_ref;
ulint err;
mtr_t* mtr;
mtr_t mtr_buf;
index = dict_table_get_first_index(node->table);
check_ref = row_upd_index_is_referenced(index);
pcur = node->pcur;
/* We have to restore the cursor to its position */
mtr = &mtr_buf;
mtr_start(mtr);
/* If the restoration does not succeed, then the same
transaction has deleted the record on which the cursor was,
and that is an SQL error. If the restoration succeeds, it may
still be that the same transaction has successively deleted
and inserted a record with the same ordering fields, but in
that case we know that the transaction has at least an
implicit x-lock on the record. */
ut_a(pcur->rel_pos == BTR_PCUR_ON);
success = btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);
if (!success) {
err = DB_RECORD_NOT_FOUND;
mtr_commit(mtr);
return(err);
}
/* If this is a row in SYS_INDEXES table of the data dictionary,
then we have to free the file segments of the index tree associated
with the index */
if (ut_dulint_cmp(node->table->id, DICT_INDEXES_ID) == 0) {
dict_drop_index_tree(btr_pcur_get_rec(pcur), mtr);
mtr_commit(mtr);
mtr_start(mtr);
success = btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur,
mtr);
if (!success) {
err = DB_ERROR;
mtr_commit(mtr);
return(err);
}
}
if (!node->has_clust_rec_x_lock) {
err = lock_clust_rec_modify_check_and_lock(0,
btr_pcur_get_rec(pcur),
index, thr);
if (err != DB_SUCCESS) {
mtr_commit(mtr);
return(err);
}
}
/* NOTE: the following function calls will also commit mtr */
if (node->is_delete) {
err = row_upd_del_mark_clust_rec(node, index, thr, check_ref,
mtr);
if (err != DB_SUCCESS) {
return(err);
}
node->state = UPD_NODE_UPDATE_ALL_SEC;
node->index = dict_table_get_next_index(index);
return(err);
}
/* If the update is made for MySQL, we already have the update vector
ready, else we have to do some evaluation: */
if (!node->in_mysql_interface) {
/* Copy the necessary columns from clust_rec and calculate the
new values to set */
row_upd_copy_columns(btr_pcur_get_rec(pcur),
UT_LIST_GET_FIRST(node->columns));
row_upd_eval_new_vals(node->update);
}
if (node->cmpl_info & UPD_NODE_NO_ORD_CHANGE) {
err = row_upd_clust_rec(node, index, thr, mtr);
return(err);
}
row_upd_store_row(node);
if (row_upd_changes_ord_field_binary(node->row, index, node->update)) {
/* Update causes an ordering field (ordering fields within
the B-tree) of the clustered index record to change: perform
the update by delete marking and inserting.
TODO! What to do to the 'Halloween problem', where an update
moves the record forward in index so that it is again
updated when the cursor arrives there? Solution: the
read operation must check the undo record undo number when
choosing records to update. MySQL solves now the problem
externally! */
err = row_upd_clust_rec_by_insert(node, index, thr, check_ref,
mtr);
if (err != DB_SUCCESS) {
return(err);
}
node->state = UPD_NODE_UPDATE_ALL_SEC;
} else {
err = row_upd_clust_rec(node, index, thr, mtr);
if (err != DB_SUCCESS) {
return(err);
}
node->state = UPD_NODE_UPDATE_SOME_SEC;
}
node->index = dict_table_get_next_index(index);
return(err);
}
/***************************************************************************
Loads a foreign key constraint to the dictionary cache. */
static
ulint
dict_load_foreign(
/*==============*/
/* out: DB_SUCCESS or error code */
char* id) /* in: foreign constraint id as a null-terminated
string */
{
dict_foreign_t* foreign;
dict_table_t* sys_foreign;
btr_pcur_t pcur;
dict_index_t* sys_index;
dtuple_t* tuple;
mem_heap_t* heap2;
dfield_t* dfield;
rec_t* rec;
byte* field;
ulint len;
ulint err;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
heap2 = mem_heap_create(1000);
mtr_start(&mtr);
sys_foreign = dict_table_get_low("SYS_FOREIGN");
sys_index = UT_LIST_GET_FIRST(sys_foreign->indexes);
tuple = dtuple_create(heap2, 1);
dfield = dtuple_get_nth_field(tuple, 0);
dfield_set_data(dfield, id, ut_strlen(id));
dict_index_copy_types(tuple, sys_index, 1);
btr_pcur_open_on_user_rec(sys_index, tuple, PAGE_CUR_GE,
BTR_SEARCH_LEAF, &pcur, &mtr);
rec = btr_pcur_get_rec(&pcur);
if (!btr_pcur_is_on_user_rec(&pcur, &mtr)
|| rec_get_deleted_flag(rec)) {
/* Not found */
fprintf(stderr,
"InnoDB: Error A: cannot load foreign constraint %s\n", id);
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap2);
return(DB_ERROR);
}
field = rec_get_nth_field(rec, 0, &len);
/* Check if the id in record is the searched one */
if (len != ut_strlen(id) || ut_memcmp(id, field, len) != 0) {
fprintf(stderr,
"InnoDB: Error B: cannot load foreign constraint %s\n", id);
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap2);
return(DB_ERROR);
}
/* Read the table names and the number of columns associated
with the constraint */
mem_heap_free(heap2);
foreign = dict_mem_foreign_create();
foreign->n_fields = mach_read_from_4(rec_get_nth_field(rec, 5, &len));
ut_a(len == 4);
foreign->id = mem_heap_alloc(foreign->heap, ut_strlen(id) + 1);
ut_memcpy(foreign->id, id, ut_strlen(id) + 1);
field = rec_get_nth_field(rec, 3, &len);
foreign->foreign_table_name = mem_heap_alloc(foreign->heap, 1 + len);
ut_memcpy(foreign->foreign_table_name, field, len);
foreign->foreign_table_name[len] = '\0';
field = rec_get_nth_field(rec, 4, &len);
foreign->referenced_table_name = mem_heap_alloc(foreign->heap, 1 + len);
ut_memcpy(foreign->referenced_table_name, field, len);
foreign->referenced_table_name[len] = '\0';
btr_pcur_close(&pcur);
mtr_commit(&mtr);
dict_load_foreign_cols(id, foreign);
/* Note that there may already be a foreign constraint object in
the dictionary cache for this constraint: then the following
call only sets the pointers in it to point to the appropriate table
and index objects and frees the newly created object foreign. */
err = dict_foreign_add_to_cache(foreign);
return(err);
}
/***************************************************************
Updates a clustered index record of a row when the ordering fields do
not change. */
static
ulint
row_upd_clust_rec(
/*==============*/
/* out: DB_SUCCESS if operation successfully
completed, else error code or DB_LOCK_WAIT */
upd_node_t* node, /* in: row update node */
dict_index_t* index, /* in: clustered index */
que_thr_t* thr, /* in: query thread */
mtr_t* mtr) /* in: mtr; gets committed here */
{
big_rec_t* big_rec = NULL;
btr_pcur_t* pcur;
btr_cur_t* btr_cur;
ulint err;
ut_ad(node);
ut_ad(index->type & DICT_CLUSTERED);
pcur = node->pcur;
btr_cur = btr_pcur_get_btr_cur(pcur);
ut_ad(FALSE == rec_get_deleted_flag(btr_pcur_get_rec(pcur)));
/* Try optimistic updating of the record, keeping changes within
the page; we do not check locks because we assume the x-lock on the
record to update */
if (node->cmpl_info & UPD_NODE_NO_SIZE_CHANGE) {
err = btr_cur_update_in_place(BTR_NO_LOCKING_FLAG,
btr_cur, node->update,
node->cmpl_info, thr, mtr);
} else {
err = btr_cur_optimistic_update(BTR_NO_LOCKING_FLAG,
btr_cur, node->update,
node->cmpl_info, thr, mtr);
}
mtr_commit(mtr);
if (err == DB_SUCCESS) {
return(err);
}
/* We may have to modify the tree structure: do a pessimistic descent
down the index tree */
mtr_start(mtr);
/* NOTE: this transaction has an s-lock or x-lock on the record and
therefore other transactions cannot modify the record when we have no
latch on the page. In addition, we assume that other query threads of
the same transaction do not modify the record in the meantime.
Therefore we can assert that the restoration of the cursor succeeds. */
ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
ut_ad(FALSE == rec_get_deleted_flag(btr_pcur_get_rec(pcur)));
err = btr_cur_pessimistic_update(BTR_NO_LOCKING_FLAG, btr_cur,
&big_rec, node->update,
node->cmpl_info, thr, mtr);
mtr_commit(mtr);
if (err == DB_SUCCESS && big_rec) {
mtr_start(mtr);
ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
err = btr_store_big_rec_extern_fields(index,
btr_cur_get_rec(btr_cur),
big_rec, mtr);
mtr_commit(mtr);
}
if (big_rec) {
dtuple_big_rec_free(big_rec);
}
return(err);
}
ulint
dict_load_foreigns(
/*===============*/
/* out: DB_SUCCESS or error code */
char* table_name) /* in: table name */
{
btr_pcur_t pcur;
mem_heap_t* heap;
dtuple_t* tuple;
dfield_t* dfield;
dict_index_t* sec_index;
dict_table_t* sys_foreign;
rec_t* rec;
byte* field;
ulint len;
char* id ;
ulint err;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
sys_foreign = dict_table_get_low("SYS_FOREIGN");
if (sys_foreign == NULL) {
/* No foreign keys defined yet in this database */
fprintf(stderr,
"InnoDB: Error: no foreign key system tables in the database\n");
return(DB_ERROR);
}
mtr_start(&mtr);
/* Get the secondary index based on FOR_NAME from table
SYS_FOREIGN */
sec_index = dict_table_get_next_index(
dict_table_get_first_index(sys_foreign));
start_load:
heap = mem_heap_create(256);
tuple = dtuple_create(heap, 1);
dfield = dtuple_get_nth_field(tuple, 0);
dfield_set_data(dfield, table_name, ut_strlen(table_name));
dict_index_copy_types(tuple, sec_index, 1);
btr_pcur_open_on_user_rec(sec_index, tuple, PAGE_CUR_GE,
BTR_SEARCH_LEAF, &pcur, &mtr);
loop:
rec = btr_pcur_get_rec(&pcur);
if (!btr_pcur_is_on_user_rec(&pcur, &mtr)) {
/* End of index */
goto load_next_index;
}
/* Now we have the record in the secondary index containing a table
name and a foreign constraint ID */
rec = btr_pcur_get_rec(&pcur);
field = rec_get_nth_field(rec, 0, &len);
/* Check if the table name in record is the one searched for */
if (len != ut_strlen(table_name)
|| 0 != ut_memcmp(field, table_name, len)) {
goto load_next_index;
}
if (rec_get_deleted_flag(rec)) {
goto next_rec;
}
/* Now we get a foreign key constraint id */
field = rec_get_nth_field(rec, 1, &len);
id = mem_heap_alloc(heap, len + 1);
ut_memcpy(id, field, len);
id[len] = '\0';
btr_pcur_store_position(&pcur, &mtr);
mtr_commit(&mtr);
/* Load the foreign constraint definition to the dictionary cache */
err = dict_load_foreign(id);
if (err != DB_SUCCESS) {
btr_pcur_close(&pcur);
mem_heap_free(heap);
return(err);
}
mtr_start(&mtr);
btr_pcur_restore_position(BTR_SEARCH_LEAF, &pcur, &mtr);
next_rec:
btr_pcur_move_to_next_user_rec(&pcur, &mtr);
goto loop;
load_next_index:
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap);
sec_index = dict_table_get_next_index(sec_index);
if (sec_index != NULL) {
mtr_start(&mtr);
goto start_load;
}
return(DB_SUCCESS);
}
/***********************************************************//**
Looks for the clustered index record when node has the row reference.
The pcur in node is used in the search. If found, stores the row to node,
and stores the position of pcur, and detaches it. The pcur must be closed
by the caller in any case.
@return TRUE if found; NOTE the node->pcur must be closed by the
caller, regardless of the return value */
UNIV_INTERN
ibool
row_undo_search_clust_to_pcur(
/*==========================*/
undo_node_t* node) /*!< in: row undo node */
{
dict_index_t* clust_index;
ibool found;
mtr_t mtr;
ibool ret;
rec_t* rec;
mem_heap_t* heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
ulint* offsets = offsets_;
rec_offs_init(offsets_);
mtr_start(&mtr);
clust_index = dict_table_get_first_index(node->table);
found = row_search_on_row_ref(&(node->pcur), BTR_MODIFY_LEAF,
node->table, node->ref, &mtr);
rec = btr_pcur_get_rec(&(node->pcur));
offsets = rec_get_offsets(rec, clust_index, offsets,
ULINT_UNDEFINED, &heap);
if (!found || 0 != ut_dulint_cmp(node->roll_ptr,
row_get_rec_roll_ptr(rec, clust_index,
offsets))) {
/* We must remove the reservation on the undo log record
BEFORE releasing the latch on the clustered index page: this
is to make sure that some thread will eventually undo the
modification corresponding to node->roll_ptr. */
/* fputs("--------------------undoing a previous version\n",
stderr); */
ret = FALSE;
} else {
row_ext_t** ext;
if (dict_table_get_format(node->table) >= DICT_TF_FORMAT_ZIP) {
/* In DYNAMIC or COMPRESSED format, there is
no prefix of externally stored columns in the
clustered index record. Build a cache of
column prefixes. */
ext = &node->ext;
} else {
/* REDUNDANT and COMPACT formats store a local
768-byte prefix of each externally stored
column. No cache is needed. */
ext = NULL;
node->ext = NULL;
}
node->row = row_build(ROW_COPY_DATA, clust_index, rec,
offsets, NULL, ext, node->heap);
if (node->update) {
node->undo_row = dtuple_copy(node->row, node->heap);
row_upd_replace(node->undo_row, &node->undo_ext,
clust_index, node->update, node->heap);
} else {
node->undo_row = NULL;
node->undo_ext = NULL;
}
btr_pcur_store_position(&(node->pcur), &mtr);
ret = TRUE;
}
btr_pcur_commit_specify_mtr(&(node->pcur), &mtr);
if (UNIV_LIKELY_NULL(heap)) {
mem_heap_free(heap);
}
return(ret);
}
/***************************************************************
Removes a secondary index entry if possible. */
static
ibool
row_purge_remove_sec_if_poss_low(
/*=============================*/
/* out: TRUE if success or if not found */
purge_node_t* node, /* in: row purge node */
que_thr_t* thr, /* in: query thread */
dict_index_t* index, /* in: index */
dtuple_t* entry, /* in: index entry */
ulint mode) /* in: latch mode BTR_MODIFY_LEAF or
BTR_MODIFY_TREE */
{
btr_pcur_t pcur;
btr_cur_t* btr_cur;
ibool success;
ibool old_has;
ibool found;
ulint err;
mtr_t mtr;
mtr_t* mtr_vers;
UT_NOT_USED(thr);
log_free_check();
mtr_start(&mtr);
found = row_search_index_entry(index, entry, mode, &pcur, &mtr);
if (!found) {
/* Not found */
/* printf("PURGE:........sec entry not found\n"); */
/* dtuple_print(entry); */
btr_pcur_close(&pcur);
mtr_commit(&mtr);
return(TRUE);
}
btr_cur = btr_pcur_get_btr_cur(&pcur);
/* We should remove the index record if no later version of the row,
which cannot be purged yet, requires its existence. If some requires,
we should do nothing. */
mtr_vers = mem_alloc(sizeof(mtr_t));
mtr_start(mtr_vers);
success = row_purge_reposition_pcur(BTR_SEARCH_LEAF, node, mtr_vers);
if (success) {
old_has = row_vers_old_has_index_entry(TRUE,
btr_pcur_get_rec(&(node->pcur)),
mtr_vers, index, entry);
}
btr_pcur_commit_specify_mtr(&(node->pcur), mtr_vers);
mem_free(mtr_vers);
if (!success || !old_has) {
/* Remove the index record */
if (mode == BTR_MODIFY_LEAF) {
success = btr_cur_optimistic_delete(btr_cur, &mtr);
} else {
ut_ad(mode == BTR_MODIFY_TREE);
btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
FALSE, &mtr);
if (err == DB_SUCCESS) {
success = TRUE;
} else if (err == DB_OUT_OF_FILE_SPACE) {
success = FALSE;
} else {
ut_a(0);
}
}
}
btr_pcur_close(&pcur);
mtr_commit(&mtr);
return(success);
}
/***************************************************************
Removes a delete marked clustered index record if possible. */
static
ibool
row_purge_remove_clust_if_poss_low(
/*===============================*/
/* out: TRUE if success, or if not found, or
if modified after the delete marking */
purge_node_t* node, /* in: row purge node */
que_thr_t* thr, /* in: query thread */
ulint mode) /* in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
{
dict_index_t* index;
btr_pcur_t* pcur;
btr_cur_t* btr_cur;
ibool success;
ulint err;
mtr_t mtr;
UT_NOT_USED(thr);
index = dict_table_get_first_index(node->table);
pcur = &(node->pcur);
btr_cur = btr_pcur_get_btr_cur(pcur);
mtr_start(&mtr);
success = row_purge_reposition_pcur(mode, node, &mtr);
if (!success) {
/* The record is already removed */
btr_pcur_commit_specify_mtr(pcur, &mtr);
return(TRUE);
}
if (0 != ut_dulint_cmp(node->roll_ptr,
row_get_rec_roll_ptr(btr_pcur_get_rec(pcur), index))) {
/* Someone else has modified the record later: do not remove */
btr_pcur_commit_specify_mtr(pcur, &mtr);
return(TRUE);
}
if (mode == BTR_MODIFY_LEAF) {
success = btr_cur_optimistic_delete(btr_cur, &mtr);
} else {
ut_ad(mode == BTR_MODIFY_TREE);
btr_cur_pessimistic_delete(&err, FALSE, btr_cur, FALSE, &mtr);
if (err == DB_SUCCESS) {
success = TRUE;
} else if (err == DB_OUT_OF_FILE_SPACE) {
success = FALSE;
} else {
ut_a(0);
}
}
btr_pcur_commit_specify_mtr(pcur, &mtr);
return(success);
}
ulint
dict_truncate_index_tree(
/*=====================*/
/* out: new root page number, or
FIL_NULL on failure */
dict_table_t* table, /* in: the table the index belongs to */
btr_pcur_t* pcur, /* in/out: persistent cursor pointing to
record in the clustered index of
SYS_INDEXES table. The cursor may be
repositioned in this call. */
mtr_t* mtr) /* in: mtr having the latch
on the record page. The mtr may be
committed and restarted in this call. */
{
ulint root_page_no;
ulint space;
ulint type;
dulint index_id;
rec_t* rec;
byte* ptr;
ulint len;
ulint comp;
dict_index_t* index;
ut_ad(mutex_own(&(dict_sys->mutex)));
ut_a(!dict_table_is_comp(dict_sys->sys_indexes));
rec = btr_pcur_get_rec(pcur);
ptr = rec_get_nth_field_old(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD, &len);
ut_ad(len == 4);
root_page_no = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
if (root_page_no == FIL_NULL) {
/* The tree has been freed. */
ut_print_timestamp(stderr);
fprintf(stderr, " InnoDB: Trying to TRUNCATE"
" a missing index of table %s!\n", table->name);
return(FIL_NULL);
}
ptr = rec_get_nth_field_old(rec,
DICT_SYS_INDEXES_SPACE_NO_FIELD, &len);
ut_ad(len == 4);
space = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
if (!fil_tablespace_exists_in_mem(space)) {
/* It is a single table tablespace and the .ibd file is
missing: do nothing */
ut_print_timestamp(stderr);
fprintf(stderr, " InnoDB: Trying to TRUNCATE"
" a missing .ibd file of table %s!\n", table->name);
return(FIL_NULL);
}
ptr = rec_get_nth_field_old(rec,
DICT_SYS_INDEXES_TYPE_FIELD, &len);
ut_ad(len == 4);
type = mach_read_from_4(ptr);
ptr = rec_get_nth_field_old(rec, 1, &len);
ut_ad(len == 8);
index_id = mach_read_from_8(ptr);
/* We free all the pages but the root page first; this operation
may span several mini-transactions */
btr_free_but_not_root(space, root_page_no);
/* Then we free the root page in the same mini-transaction where
we create the b-tree and write its new root page number to the
appropriate field in the SYS_INDEXES record: this mini-transaction
marks the B-tree totally truncated */
comp = page_is_comp(btr_page_get(space, root_page_no, RW_X_LATCH,
mtr));
btr_free_root(space, root_page_no, mtr);
/* We will temporarily write FIL_NULL to the PAGE_NO field
in SYS_INDEXES, so that the database will not get into an
inconsistent state in case it crashes between the mtr_commit()
below and the following mtr_commit() call. */
page_rec_write_index_page_no(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
FIL_NULL, mtr);
/* We will need to commit the mini-transaction in order to avoid
deadlocks in the btr_create() call, because otherwise we would
be freeing and allocating pages in the same mini-transaction. */
btr_pcur_store_position(pcur, mtr);
mtr_commit(mtr);
mtr_start(mtr);
btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);
/* Find the index corresponding to this SYS_INDEXES record. */
for (index = UT_LIST_GET_FIRST(table->indexes);
index;
index = UT_LIST_GET_NEXT(indexes, index)) {
if (!ut_dulint_cmp(index->id, index_id)) {
break;
}
}
root_page_no = btr_create(type, space, index_id, comp, mtr);
if (index) {
index->page = (unsigned int) root_page_no;
} else {
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: Index %lu %lu of table %s is missing\n"
"InnoDB: from the data dictionary during TRUNCATE!\n",
ut_dulint_get_high(index_id),
ut_dulint_get_low(index_id),
table->name);
}
return(root_page_no);
}
/***********************************************************//**
Removes a delete marked clustered index record if possible.
@return TRUE if success, or if not found, or if modified after the
delete marking */
static
ibool
row_purge_remove_clust_if_poss_low(
/*===============================*/
purge_node_t* node, /*!< in: row purge node */
ulint mode) /*!< in: BTR_MODIFY_LEAF or BTR_MODIFY_TREE */
{
dict_index_t* index;
btr_pcur_t* pcur;
btr_cur_t* btr_cur;
ibool success;
ulint err;
mtr_t mtr;
rec_t* rec;
mem_heap_t* heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
rec_offs_init(offsets_);
index = dict_table_get_first_index(node->table);
pcur = &(node->pcur);
btr_cur = btr_pcur_get_btr_cur(pcur);
log_free_check();
mtr_start(&mtr);
success = row_purge_reposition_pcur(mode, node, &mtr);
if (!success) {
/* The record is already removed */
btr_pcur_commit_specify_mtr(pcur, &mtr);
return(TRUE);
}
rec = btr_pcur_get_rec(pcur);
if (node->roll_ptr != row_get_rec_roll_ptr(
rec, index, rec_get_offsets(rec, index, offsets_,
ULINT_UNDEFINED, &heap))) {
if (UNIV_LIKELY_NULL(heap)) {
mem_heap_free(heap);
}
/* Someone else has modified the record later: do not remove */
btr_pcur_commit_specify_mtr(pcur, &mtr);
return(TRUE);
}
if (UNIV_LIKELY_NULL(heap)) {
mem_heap_free(heap);
}
if (mode == BTR_MODIFY_LEAF) {
success = btr_cur_optimistic_delete(btr_cur, &mtr);
} else {
ut_ad(mode == BTR_MODIFY_TREE);
btr_cur_pessimistic_delete(&err, FALSE, btr_cur,
RB_NONE, &mtr);
if (err == DB_SUCCESS) {
success = TRUE;
} else if (err == DB_OUT_OF_FILE_SPACE) {
success = FALSE;
} else {
ut_error;
}
}
btr_pcur_commit_specify_mtr(pcur, &mtr);
return(success);
}
/**************************************************************//**
Restores the stored position of a persistent cursor bufferfixing the page and
obtaining the specified latches. If the cursor position was saved when the
(1) cursor was positioned on a user record: this function restores the position
to the last record LESS OR EQUAL to the stored record;
(2) cursor was positioned on a page infimum record: restores the position to
the last record LESS than the user record which was the successor of the page
infimum;
(3) cursor was positioned on the page supremum: restores to the first record
GREATER than the user record which was the predecessor of the supremum.
(4) cursor was positioned before the first or after the last in an empty tree:
restores to before first or after the last in the tree.
@return TRUE if the cursor position was stored when it was on a user
record and it can be restored on a user record whose ordering fields
are identical to the ones of the original user record */
UNIV_INTERN
ibool
btr_pcur_restore_position_func(
/*===========================*/
ulint latch_mode, /*!< in: BTR_SEARCH_LEAF, ... */
btr_pcur_t* cursor, /*!< in: detached persistent cursor */
const char* file, /*!< in: file name */
ulint line, /*!< in: line where called */
mtr_t* mtr) /*!< in: mtr */
{
dict_index_t* index;
dtuple_t* tuple;
ulint mode;
ulint old_mode;
mem_heap_t* heap;
ut_ad(mtr);
ut_ad(mtr->state == MTR_ACTIVE);
index = btr_cur_get_index(btr_pcur_get_btr_cur(cursor));
if (UNIV_UNLIKELY(cursor->old_stored != BTR_PCUR_OLD_STORED)
|| UNIV_UNLIKELY(cursor->pos_state != BTR_PCUR_WAS_POSITIONED
&& cursor->pos_state != BTR_PCUR_IS_POSITIONED)) {
ut_print_buf(stderr, cursor, sizeof(btr_pcur_t));
putc('\n', stderr);
if (cursor->trx_if_known) {
trx_print(stderr, cursor->trx_if_known, 0);
}
ut_error;
}
if (UNIV_UNLIKELY
(cursor->rel_pos == BTR_PCUR_AFTER_LAST_IN_TREE
|| cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE)) {
/* In these cases we do not try an optimistic restoration,
but always do a search */
btr_cur_open_at_index_side(
cursor->rel_pos == BTR_PCUR_BEFORE_FIRST_IN_TREE,
index, latch_mode, btr_pcur_get_btr_cur(cursor), mtr);
cursor->latch_mode = latch_mode;
cursor->pos_state = BTR_PCUR_IS_POSITIONED;
cursor->block_when_stored = btr_pcur_get_block(cursor);
return(FALSE);
}
ut_a(cursor->old_rec);
ut_a(cursor->old_n_fields);
if (UNIV_LIKELY(latch_mode == BTR_SEARCH_LEAF)
|| UNIV_LIKELY(latch_mode == BTR_MODIFY_LEAF)) {
/* Try optimistic restoration */
if (UNIV_LIKELY(buf_page_optimistic_get(
latch_mode,
cursor->block_when_stored,
cursor->modify_clock,
file, line, mtr))) {
cursor->pos_state = BTR_PCUR_IS_POSITIONED;
buf_block_dbg_add_level(
btr_pcur_get_block(cursor),
dict_index_is_ibuf(index)
? SYNC_IBUF_TREE_NODE : SYNC_TREE_NODE);
if (cursor->rel_pos == BTR_PCUR_ON) {
#ifdef UNIV_DEBUG
const rec_t* rec;
const ulint* offsets1;
const ulint* offsets2;
#endif /* UNIV_DEBUG */
cursor->latch_mode = latch_mode;
#ifdef UNIV_DEBUG
rec = btr_pcur_get_rec(cursor);
heap = mem_heap_create(256);
offsets1 = rec_get_offsets(
cursor->old_rec, index, NULL,
cursor->old_n_fields, &heap);
offsets2 = rec_get_offsets(
rec, index, NULL,
cursor->old_n_fields, &heap);
ut_ad(!cmp_rec_rec(cursor->old_rec,
rec, offsets1, offsets2,
index));
mem_heap_free(heap);
#endif /* UNIV_DEBUG */
return(TRUE);
}
return(FALSE);
}
}
/* If optimistic restoration did not succeed, open the cursor anew */
heap = mem_heap_create(256);
tuple = dict_index_build_data_tuple(index, cursor->old_rec,
cursor->old_n_fields, heap);
/* Save the old search mode of the cursor */
old_mode = cursor->search_mode;
switch (cursor->rel_pos) {
case BTR_PCUR_ON:
mode = PAGE_CUR_LE;
break;
case BTR_PCUR_AFTER:
mode = PAGE_CUR_G;
break;
case BTR_PCUR_BEFORE:
mode = PAGE_CUR_L;
break;
default:
ut_error;
mode = 0;
}
btr_pcur_open_with_no_init_func(index, tuple, mode, latch_mode,
cursor, 0, file, line, mtr);
/* Restore the old search mode */
cursor->search_mode = old_mode;
switch (cursor->rel_pos) {
case BTR_PCUR_ON:
if (btr_pcur_is_on_user_rec(cursor)
&& !cmp_dtuple_rec(
tuple, btr_pcur_get_rec(cursor),
rec_get_offsets(btr_pcur_get_rec(cursor),
index, NULL,
ULINT_UNDEFINED, &heap))) {
/* We have to store the NEW value for
the modify clock, since the cursor can
now be on a different page! But we can
retain the value of old_rec */
cursor->block_when_stored =
btr_pcur_get_block(cursor);
cursor->modify_clock =
buf_block_get_modify_clock(
cursor->block_when_stored);
cursor->old_stored = BTR_PCUR_OLD_STORED;
mem_heap_free(heap);
return(TRUE);
}
#ifdef UNIV_DEBUG
/* fall through */
case BTR_PCUR_BEFORE:
case BTR_PCUR_AFTER:
break;
default:
ut_error;
#endif /* UNIV_DEBUG */
}
mem_heap_free(heap);
/* We have to store new position information, modify_clock etc.,
to the cursor because it can now be on a different page, the record
under it may have been removed, etc. */
btr_pcur_store_position(cursor, mtr);
return(FALSE);
}
/*******************************************************************//**
Truncates the index tree associated with a row in SYS_INDEXES table.
@return new root page number, or FIL_NULL on failure */
UNIV_INTERN
ulint
dict_truncate_index_tree(
/*=====================*/
dict_table_t* table, /*!< in: the table the index belongs to */
ulint space, /*!< in: 0=truncate,
nonzero=create the index tree in the
given tablespace */
btr_pcur_t* pcur, /*!< in/out: persistent cursor pointing to
record in the clustered index of
SYS_INDEXES table. The cursor may be
repositioned in this call. */
mtr_t* mtr) /*!< in: mtr having the latch
on the record page. The mtr may be
committed and restarted in this call. */
{
ulint root_page_no;
ibool drop = !space;
ulint zip_size;
ulint type;
index_id_t index_id;
rec_t* rec;
const byte* ptr;
ulint len;
dict_index_t* index;
ut_ad(mutex_own(&(dict_sys->mutex)));
ut_a(!dict_table_is_comp(dict_sys->sys_indexes));
rec = btr_pcur_get_rec(pcur);
ptr = rec_get_nth_field_old(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD, &len);
ut_ad(len == 4);
root_page_no = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
if (drop && root_page_no == FIL_NULL) {
/* The tree has been freed. */
ut_print_timestamp(stderr);
fprintf(stderr, " InnoDB: Trying to TRUNCATE"
" a missing index of table %s!\n", table->name);
drop = FALSE;
}
ptr = rec_get_nth_field_old(rec,
DICT_SYS_INDEXES_SPACE_NO_FIELD, &len);
ut_ad(len == 4);
if (drop) {
space = mtr_read_ulint(ptr, MLOG_4BYTES, mtr);
}
zip_size = fil_space_get_zip_size(space);
if (UNIV_UNLIKELY(zip_size == ULINT_UNDEFINED)) {
/* It is a single table tablespace and the .ibd file is
missing: do nothing */
ut_print_timestamp(stderr);
fprintf(stderr, " InnoDB: Trying to TRUNCATE"
" a missing .ibd file of table %s!\n", table->name);
return(FIL_NULL);
}
ptr = rec_get_nth_field_old(rec,
DICT_SYS_INDEXES_TYPE_FIELD, &len);
ut_ad(len == 4);
type = mach_read_from_4(ptr);
ptr = rec_get_nth_field_old(rec, 1, &len);
ut_ad(len == 8);
index_id = mach_read_from_8(ptr);
if (!drop) {
goto create;
}
/* We free all the pages but the root page first; this operation
may span several mini-transactions */
btr_free_but_not_root(space, zip_size, root_page_no);
/* Then we free the root page in the same mini-transaction where
we create the b-tree and write its new root page number to the
appropriate field in the SYS_INDEXES record: this mini-transaction
marks the B-tree totally truncated */
btr_block_get(space, zip_size, root_page_no, RW_X_LATCH, NULL, mtr);
btr_free_root(space, zip_size, root_page_no, mtr);
create:
/* We will temporarily write FIL_NULL to the PAGE_NO field
in SYS_INDEXES, so that the database will not get into an
inconsistent state in case it crashes between the mtr_commit()
below and the following mtr_commit() call. */
page_rec_write_field(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD,
FIL_NULL, mtr);
/* We will need to commit the mini-transaction in order to avoid
deadlocks in the btr_create() call, because otherwise we would
be freeing and allocating pages in the same mini-transaction. */
btr_pcur_store_position(pcur, mtr);
mtr_commit(mtr);
mtr_start(mtr);
btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);
/* Find the index corresponding to this SYS_INDEXES record. */
for (index = UT_LIST_GET_FIRST(table->indexes);
index;
index = UT_LIST_GET_NEXT(indexes, index)) {
if (index->id == index_id) {
root_page_no = btr_create(type, space, zip_size,
index_id, index, mtr);
index->page = (unsigned int) root_page_no;
return(root_page_no);
}
}
ut_print_timestamp(stderr);
fprintf(stderr,
" InnoDB: Index %llu of table %s is missing\n"
"InnoDB: from the data dictionary during TRUNCATE!\n",
(ullint) index_id,
table->name);
return(FIL_NULL);
}
