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);
}
/***************************************************************
Purges an update of an existing record. Also purges an update of a delete
marked record if that record contained an externally stored field. */
static
void
row_purge_upd_exist_or_extern(
/*==========================*/
purge_node_t* node, /* in: row purge node */
que_thr_t* thr) /* in: query thread */
{
mem_heap_t* heap;
dtuple_t* entry;
dict_index_t* index;
upd_field_t* ufield;
ibool is_insert;
ulint rseg_id;
ulint page_no;
ulint offset;
ulint internal_offset;
byte* data_field;
ulint data_field_len;
ulint i;
mtr_t mtr;
ut_ad(node && thr);
if (node->rec_type == TRX_UNDO_UPD_DEL_REC) {
goto skip_secondaries;
}
heap = mem_heap_create(1024);
while (node->index != NULL) {
index = node->index;
if (row_upd_changes_ord_field_binary(NULL, node->index,
node->update)) {
/* Build the older version of the index entry */
entry = row_build_index_entry(node->row, index, heap);
row_purge_remove_sec_if_poss(node, thr, index, entry);
}
node->index = dict_table_get_next_index(node->index);
}
mem_heap_free(heap);
skip_secondaries:
/* Free possible externally stored fields */
for (i = 0; i < upd_get_n_fields(node->update); i++) {
ufield = upd_get_nth_field(node->update, i);
if (ufield->extern_storage) {
/* We use the fact that new_val points to
node->undo_rec and get thus the offset of
dfield data inside the unod record. Then we
can calculate from node->roll_ptr the file
address of the new_val data */
internal_offset = ((byte*)ufield->new_val.data)
- node->undo_rec;
ut_a(internal_offset < UNIV_PAGE_SIZE);
trx_undo_decode_roll_ptr(node->roll_ptr,
&is_insert, &rseg_id,
&page_no, &offset);
mtr_start(&mtr);
/* We have to acquire an X-latch to the clustered
index tree */
index = dict_table_get_first_index(node->table);
mtr_x_lock(dict_tree_get_lock(index->tree), &mtr);
/* We assume in purge of externally stored fields
that the space id of the undo log record is 0! */
data_field = buf_page_get(0, page_no, RW_X_LATCH, &mtr)
+ offset + internal_offset;
buf_page_dbg_add_level(buf_frame_align(data_field),
SYNC_TRX_UNDO_PAGE);
data_field_len = ufield->new_val.len;
btr_free_externally_stored_field(index, data_field,
data_field_len, FALSE, &mtr);
mtr_commit(&mtr);
}
}
}
ulint
trx_rseg_header_create(
/*===================*/
/* out: page number of the created segment,
FIL_NULL if fail */
ulint space, /* in: space id */
ulint max_size, /* in: max size in pages */
ulint* slot_no, /* out: rseg id == slot number in trx sys */
mtr_t* mtr) /* in: mtr */
{
ulint page_no;
trx_rsegf_t* rsegf;
trx_sysf_t* sys_header;
ulint i;
page_t* page;
ut_ad(mtr);
#ifdef UNIV_SYNC_DEBUG
ut_ad(mutex_own(&kernel_mutex));
#endif /* UNIV_SYNC_DEBUG */
ut_ad(mtr_memo_contains(mtr, fil_space_get_latch(space),
MTR_MEMO_X_LOCK));
sys_header = trx_sysf_get(mtr);
*slot_no = trx_sysf_rseg_find_free(mtr);
if (*slot_no == ULINT_UNDEFINED) {
return(FIL_NULL);
}
/* Allocate a new file segment for the rollback segment */
page = fseg_create(space, 0, TRX_RSEG + TRX_RSEG_FSEG_HEADER, mtr);
if (page == NULL) {
/* No space left */
return(FIL_NULL);
}
#ifdef UNIV_SYNC_DEBUG
buf_page_dbg_add_level(page, SYNC_RSEG_HEADER_NEW);
#endif /* UNIV_SYNC_DEBUG */
page_no = buf_frame_get_page_no(page);
/* Get the rollback segment file page */
rsegf = trx_rsegf_get_new(space, page_no, mtr);
/* Initialize max size field */
mlog_write_ulint(rsegf + TRX_RSEG_MAX_SIZE, max_size, MLOG_4BYTES, mtr);
/* Initialize the history list */
mlog_write_ulint(rsegf + TRX_RSEG_HISTORY_SIZE, 0, MLOG_4BYTES, mtr);
flst_init(rsegf + TRX_RSEG_HISTORY, mtr);
/* Reset the undo log slots */
for (i = 0; i < TRX_RSEG_N_SLOTS; i++) {
trx_rsegf_set_nth_undo(rsegf, i, FIL_NULL, mtr);
}
/* Add the rollback segment info to the free slot in the trx system
header */
trx_sysf_rseg_set_space(sys_header, *slot_no, space, mtr);
trx_sysf_rseg_set_page_no(sys_header, *slot_no, page_no, mtr);
return(page_no);
}
/***************************************************************
Purges an update of an existing record. Also purges an update of a delete
marked record if that record contained an externally stored field. */
static
void
row_purge_upd_exist_or_extern(
/*==========================*/
purge_node_t* node) /* in: row purge node */
{
mem_heap_t* heap;
dtuple_t* entry;
dict_index_t* index;
upd_field_t* ufield;
ibool is_insert;
ulint rseg_id;
ulint page_no;
ulint offset;
ulint internal_offset;
byte* data_field;
ulint data_field_len;
ulint i;
mtr_t mtr;
ut_ad(node);
if (node->rec_type == TRX_UNDO_UPD_DEL_REC) {
goto skip_secondaries;
}
heap = mem_heap_create(1024);
while (node->index != NULL) {
index = node->index;
if (row_upd_changes_ord_field_binary(NULL, node->index,
node->update)) {
/* Build the older version of the index entry */
entry = row_build_index_entry(node->row, index, heap);
row_purge_remove_sec_if_poss(node, index, entry);
}
node->index = dict_table_get_next_index(node->index);
}
mem_heap_free(heap);
skip_secondaries:
/* Free possible externally stored fields */
for (i = 0; i < upd_get_n_fields(node->update); i++) {
ufield = upd_get_nth_field(node->update, i);
if (ufield->extern_storage) {
/* We use the fact that new_val points to
node->undo_rec and get thus the offset of
dfield data inside the unod record. Then we
can calculate from node->roll_ptr the file
address of the new_val data */
internal_offset = ((byte*)ufield->new_val.data)
- node->undo_rec;
ut_a(internal_offset < UNIV_PAGE_SIZE);
trx_undo_decode_roll_ptr(node->roll_ptr,
&is_insert, &rseg_id,
&page_no, &offset);
mtr_start(&mtr);
/* We have to acquire an X-latch to the clustered
index tree */
index = dict_table_get_first_index(node->table);
mtr_x_lock(dict_index_get_lock(index), &mtr);
/* NOTE: we must also acquire an X-latch to the
root page of the tree. We will need it when we
free pages from the tree. If the tree is of height 1,
the tree X-latch does NOT protect the root page,
because it is also a leaf page. Since we will have a
latch on an undo log page, we would break the
latching order if we would only later latch the
root page of such a tree! */
btr_root_get(index, &mtr);
/* We assume in purge of externally stored fields
that the space id of the undo log record is 0! */
data_field = buf_page_get(0, page_no, RW_X_LATCH, &mtr)
+ offset + internal_offset;
#ifdef UNIV_SYNC_DEBUG
buf_page_dbg_add_level(buf_frame_align(data_field),
SYNC_TRX_UNDO_PAGE);
#endif /* UNIV_SYNC_DEBUG */
data_field_len = ufield->new_val.len;
btr_free_externally_stored_field(index, data_field,
data_field_len,
FALSE, &mtr);
mtr_commit(&mtr);
}
}
}
