dict_table_t*
dict_load_table(
/*============*/
/* out: table, NULL if does not exist */
char* name) /* in: table name */
{
dict_table_t* table;
dict_table_t* sys_tables;
btr_pcur_t pcur;
dict_index_t* sys_index;
dtuple_t* tuple;
mem_heap_t* heap;
dfield_t* dfield;
rec_t* rec;
byte* field;
ulint len;
char* buf;
ulint space;
ulint n_cols;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
heap = mem_heap_create(1000);
mtr_start(&mtr);
sys_tables = dict_table_get_low("SYS_TABLES");
sys_index = UT_LIST_GET_FIRST(sys_tables->indexes);
tuple = dtuple_create(heap, 1);
dfield = dtuple_get_nth_field(tuple, 0);
dfield_set_data(dfield, name, ut_strlen(name));
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 */
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap);
return(NULL);
}
field = rec_get_nth_field(rec, 0, &len);
/* Check if the table name in record is the searched one */
if (len != ut_strlen(name) || ut_memcmp(name, field, len) != 0) {
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap);
return(NULL);
}
ut_a(0 == ut_strcmp("SPACE",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_tables), 9))->name));
field = rec_get_nth_field(rec, 9, &len);
space = mach_read_from_4(field);
ut_a(0 == ut_strcmp("N_COLS",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_tables), 4))->name));
field = rec_get_nth_field(rec, 4, &len);
n_cols = mach_read_from_4(field);
table = dict_mem_table_create(name, space, n_cols);
ut_a(0 == ut_strcmp("ID",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_tables), 3))->name));
field = rec_get_nth_field(rec, 3, &len);
table->id = mach_read_from_8(field);
field = rec_get_nth_field(rec, 5, &len);
table->type = mach_read_from_4(field);
if (table->type == DICT_TABLE_CLUSTER_MEMBER) {
ut_a(0);
field = rec_get_nth_field(rec, 6, &len);
table->mix_id = mach_read_from_8(field);
field = rec_get_nth_field(rec, 8, &len);
buf = mem_heap_alloc(heap, len);
ut_memcpy(buf, field, len);
table->cluster_name = buf;
}
if ((table->type == DICT_TABLE_CLUSTER)
|| (table->type == DICT_TABLE_CLUSTER_MEMBER)) {
field = rec_get_nth_field(rec, 7, &len);
table->mix_len = mach_read_from_4(field);
}
btr_pcur_close(&pcur);
mtr_commit(&mtr);
if (table->type == DICT_TABLE_CLUSTER_MEMBER) {
/* Load the cluster table definition if not yet in
memory cache */
dict_table_get_low(table->cluster_name);
}
dict_load_columns(table, heap);
dict_table_add_to_cache(table);
dict_load_indexes(table, heap);
ut_a(DB_SUCCESS == dict_load_foreigns(table->name));
mem_heap_free(heap);
return(table);
}
dict_table_t*
dict_load_table_on_id(
/*==================*/
/* out: table; NULL if table does not exist */
dulint table_id) /* in: table id */
{
byte id_buf[8];
btr_pcur_t pcur;
mem_heap_t* heap;
dtuple_t* tuple;
dfield_t* dfield;
dict_index_t* sys_table_ids;
dict_table_t* sys_tables;
rec_t* rec;
byte* field;
ulint len;
dict_table_t* table;
char* name;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
/* NOTE that the operation of this function is protected by
the dictionary mutex, and therefore no deadlocks can occur
with other dictionary operations. */
mtr_start(&mtr);
/*---------------------------------------------------*/
/* Get the secondary index based on ID for table SYS_TABLES */
sys_tables = dict_sys->sys_tables;
sys_table_ids = dict_table_get_next_index(
dict_table_get_first_index(sys_tables));
heap = mem_heap_create(256);
tuple = dtuple_create(heap, 1);
dfield = dtuple_get_nth_field(tuple, 0);
/* Write the table id in byte format to id_buf */
mach_write_to_8(id_buf, table_id);
dfield_set_data(dfield, id_buf, 8);
dict_index_copy_types(tuple, sys_table_ids, 1);
btr_pcur_open_on_user_rec(sys_table_ids, 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 */
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap);
return(NULL);
}
/*---------------------------------------------------*/
/* Now we have the record in the secondary index containing the
table ID and NAME */
rec = btr_pcur_get_rec(&pcur);
field = rec_get_nth_field(rec, 0, &len);
ut_ad(len == 8);
/* Check if the table id in record is the one searched for */
if (ut_dulint_cmp(table_id, mach_read_from_8(field)) != 0) {
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap);
return(NULL);
}
/* Now we get the table name from the record */
field = rec_get_nth_field(rec, 1, &len);
name = mem_heap_alloc(heap, len + 1);
ut_memcpy(name, field, len);
name[len] = '\0';
/* Load the table definition to memory */
table = dict_load_table(name);
ut_a(table);
btr_pcur_close(&pcur);
mtr_commit(&mtr);
mem_heap_free(heap);
return(table);
}
/*****************************************************************//**
Constructs the last committed version of a clustered index record,
which should be seen by a semi-consistent read.
@return DB_SUCCESS or DB_MISSING_HISTORY */
UNIV_INTERN
ulint
row_vers_build_for_semi_consistent_read(
/*====================================*/
const rec_t* rec, /*!< in: record in a clustered index; the
caller must have a latch on the page; this
latch locks the top of the stack of versions
of this records */
mtr_t* mtr, /*!< in: mtr holding the latch on rec */
dict_index_t* index, /*!< in: the clustered index */
ulint** offsets,/*!< in/out: offsets returned by
rec_get_offsets(rec, index) */
mem_heap_t** offset_heap,/*!< in/out: memory heap from which
the offsets are allocated */
mem_heap_t* in_heap,/*!< in: memory heap from which the memory for
*old_vers is allocated; memory for possible
intermediate versions is allocated and freed
locally within the function */
const rec_t** old_vers)/*!< out: rec, old version, or NULL if the
record does not exist in the view, that is,
it was freshly inserted afterwards */
{
const rec_t* version;
mem_heap_t* heap = NULL;
byte* buf;
ulint err;
trx_id_t rec_trx_id = 0;
ut_ad(dict_index_is_clust(index));
ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
|| mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */
ut_ad(rec_offs_validate(rec, index, *offsets));
rw_lock_s_lock(&(purge_sys->latch));
/* The S-latch on purge_sys prevents the purge view from
changing. Thus, if we have an uncommitted transaction at
this point, then purge cannot remove its undo log even if
the transaction could commit now. */
version = rec;
for (;;) {
trx_t* version_trx;
mem_heap_t* heap2;
rec_t* prev_version;
trx_id_t version_trx_id;
version_trx_id = row_get_rec_trx_id(version, index, *offsets);
if (rec == version) {
rec_trx_id = version_trx_id;
}
mutex_enter(&kernel_mutex);
version_trx = trx_get_on_id(version_trx_id);
mutex_exit(&kernel_mutex);
if (!version_trx
|| version_trx->conc_state == TRX_NOT_STARTED
|| version_trx->conc_state == TRX_COMMITTED_IN_MEMORY) {
/* We found a version that belongs to a
committed transaction: return it. */
if (rec == version) {
*old_vers = rec;
err = DB_SUCCESS;
break;
}
/* We assume that a rolled-back transaction stays in
TRX_ACTIVE state until all the changes have been
rolled back and the transaction is removed from
the global list of transactions. */
if (rec_trx_id == version_trx_id) {
/* The transaction was committed while
we searched for earlier versions.
Return the current version as a
semi-consistent read. */
version = rec;
*offsets = rec_get_offsets(version,
index, *offsets,
ULINT_UNDEFINED,
offset_heap);
}
buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets));
*old_vers = rec_copy(buf, version, *offsets);
rec_offs_make_valid(*old_vers, index, *offsets);
err = DB_SUCCESS;
break;
}
heap2 = heap;
heap = mem_heap_create(1024);
err = trx_undo_prev_version_build(rec, mtr, version, index,
*offsets, heap,
&prev_version);
if (heap2) {
mem_heap_free(heap2); /* free version */
}
if (UNIV_UNLIKELY(err != DB_SUCCESS)) {
break;
}
if (prev_version == NULL) {
/* It was a freshly inserted version */
*old_vers = NULL;
err = DB_SUCCESS;
break;
}
version = prev_version;
*offsets = rec_get_offsets(version, index, *offsets,
ULINT_UNDEFINED, offset_heap);
}/* for (;;) */
if (heap) {
mem_heap_free(heap);
}
rw_lock_s_unlock(&(purge_sys->latch));
return(err);
}
/******************************************************************//**
Add a bound literal to a symbol table.
@return symbol table node */
UNIV_INTERN
sym_node_t*
sym_tab_add_bound_lit(
/*==================*/
sym_tab_t* sym_tab, /*!< in: symbol table */
const char* name, /*!< in: name of bound literal */
ulint* lit_type) /*!< out: type of literal (PARS_*_LIT) */
{
sym_node_t* node;
pars_bound_lit_t* blit;
ulint len = 0;
blit = pars_info_get_bound_lit(sym_tab->info, name);
ut_a(blit);
node = mem_heap_alloc(sym_tab->heap, sizeof(sym_node_t));
node->common.type = QUE_NODE_SYMBOL;
node->resolved = TRUE;
node->token_type = SYM_LIT;
node->indirection = NULL;
switch (blit->type) {
case DATA_FIXBINARY:
len = blit->length;
*lit_type = PARS_FIXBINARY_LIT;
break;
case DATA_BLOB:
*lit_type = PARS_BLOB_LIT;
break;
case DATA_VARCHAR:
*lit_type = PARS_STR_LIT;
break;
case DATA_CHAR:
ut_a(blit->length > 0);
len = blit->length;
*lit_type = PARS_STR_LIT;
break;
case DATA_INT:
ut_a(blit->length > 0);
ut_a(blit->length <= 8);
len = blit->length;
*lit_type = PARS_INT_LIT;
break;
default:
ut_error;
}
dtype_set(dfield_get_type(&node->common.val),
blit->type, blit->prtype, len);
dfield_set_data(&(node->common.val), blit->address, blit->length);
node->common.val_buf_size = 0;
node->prefetch_buf = NULL;
node->cursor_def = NULL;
UT_LIST_ADD_LAST(sym_list, sym_tab->sym_list, node);
node->sym_table = sym_tab;
return(node);
}
/*****************************************************************//**
Based on a table object, this function builds the entry to be inserted
in the SYS_COLUMNS system table.
@return the tuple which should be inserted */
static
dtuple_t*
dict_create_sys_columns_tuple(
/*==========================*/
const dict_table_t* table, /*!< in: table */
ulint i, /*!< in: column number */
mem_heap_t* heap) /*!< in: memory heap from
which the memory for the built
tuple is allocated */
{
dict_table_t* sys_columns;
dtuple_t* entry;
const dict_col_t* column;
dfield_t* dfield;
byte* ptr;
const char* col_name;
ut_ad(table);
ut_ad(heap);
column = dict_table_get_nth_col(table, i);
sys_columns = dict_sys->sys_columns;
entry = dtuple_create(heap, 7 + DATA_N_SYS_COLS);
dict_table_copy_types(entry, sys_columns);
/* 0: TABLE_ID -----------------------*/
dfield = dtuple_get_nth_field(entry, 0/*TABLE_ID*/);
ptr = mem_heap_alloc(heap, 8);
mach_write_to_8(ptr, table->id);
dfield_set_data(dfield, ptr, 8);
/* 1: POS ----------------------------*/
dfield = dtuple_get_nth_field(entry, 1/*POS*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, i);
dfield_set_data(dfield, ptr, 4);
/* 4: NAME ---------------------------*/
dfield = dtuple_get_nth_field(entry, 2/*NAME*/);
col_name = dict_table_get_col_name(table, i);
dfield_set_data(dfield, col_name, ut_strlen(col_name));
/* 5: MTYPE --------------------------*/
dfield = dtuple_get_nth_field(entry, 3/*MTYPE*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, column->mtype);
dfield_set_data(dfield, ptr, 4);
/* 6: PRTYPE -------------------------*/
dfield = dtuple_get_nth_field(entry, 4/*PRTYPE*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, column->prtype);
dfield_set_data(dfield, ptr, 4);
/* 7: LEN ----------------------------*/
dfield = dtuple_get_nth_field(entry, 5/*LEN*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, column->len);
dfield_set_data(dfield, ptr, 4);
/* 8: PREC ---------------------------*/
dfield = dtuple_get_nth_field(entry, 6/*PREC*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, 0/* unused */);
dfield_set_data(dfield, ptr, 4);
/*---------------------------------*/
return(entry);
}
/*****************************************************************//**
Constructs the version of a clustered index record which a consistent
read should see. We assume that the trx id stored in rec is such that
the consistent read should not see rec in its present version.
@return DB_SUCCESS or DB_MISSING_HISTORY */
UNIV_INTERN
ulint
row_vers_build_for_consistent_read(
/*===============================*/
const rec_t* rec, /*!< in: record in a clustered index; the
caller must have a latch on the page; this
latch locks the top of the stack of versions
of this records */
mtr_t* mtr, /*!< in: mtr holding the latch on rec */
dict_index_t* index, /*!< in: the clustered index */
ulint** offsets,/*!< in/out: offsets returned by
rec_get_offsets(rec, index) */
read_view_t* view, /*!< in: the consistent read view */
mem_heap_t** offset_heap,/*!< in/out: memory heap from which
the offsets are allocated */
mem_heap_t* in_heap,/*!< in: memory heap from which the memory for
*old_vers is allocated; memory for possible
intermediate versions is allocated and freed
locally within the function */
rec_t** old_vers)/*!< out, own: old version, or NULL if the
record does not exist in the view, that is,
it was freshly inserted afterwards */
{
const rec_t* version;
rec_t* prev_version;
trx_id_t trx_id;
mem_heap_t* heap = NULL;
byte* buf;
ulint err;
ut_ad(dict_index_is_clust(index));
ut_ad(mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_X_FIX)
|| mtr_memo_contains_page(mtr, rec, MTR_MEMO_PAGE_S_FIX));
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */
ut_ad(rec_offs_validate(rec, index, *offsets));
trx_id = row_get_rec_trx_id(rec, index, *offsets);
ut_ad(!read_view_sees_trx_id(view, trx_id));
rw_lock_s_lock(&(purge_sys->latch));
version = rec;
for (;;) {
mem_heap_t* heap2 = heap;
trx_undo_rec_t* undo_rec;
roll_ptr_t roll_ptr;
undo_no_t undo_no;
heap = mem_heap_create(1024);
/* If we have high-granularity consistent read view and
creating transaction of the view is the same as trx_id in
the record we see this record only in the case when
undo_no of the record is < undo_no in the view. */
if (view->type == VIEW_HIGH_GRANULARITY
&& view->creator_trx_id == trx_id) {
roll_ptr = row_get_rec_roll_ptr(version, index,
*offsets);
undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap);
undo_no = trx_undo_rec_get_undo_no(undo_rec);
mem_heap_empty(heap);
if (view->undo_no > undo_no) {
/* The view already sees this version: we can
copy it to in_heap and return */
buf = mem_heap_alloc(in_heap,
rec_offs_size(*offsets));
*old_vers = rec_copy(buf, version, *offsets);
rec_offs_make_valid(*old_vers, index,
*offsets);
err = DB_SUCCESS;
break;
}
}
err = trx_undo_prev_version_build(rec, mtr, version, index,
*offsets, heap,
&prev_version);
if (heap2) {
mem_heap_free(heap2); /* free version */
}
if (err != DB_SUCCESS) {
break;
}
if (prev_version == NULL) {
/* It was a freshly inserted version */
*old_vers = NULL;
err = DB_SUCCESS;
break;
}
*offsets = rec_get_offsets(prev_version, index, *offsets,
ULINT_UNDEFINED, offset_heap);
trx_id = row_get_rec_trx_id(prev_version, index, *offsets);
if (read_view_sees_trx_id(view, trx_id)) {
/* The view already sees this version: we can copy
it to in_heap and return */
buf = mem_heap_alloc(in_heap, rec_offs_size(*offsets));
*old_vers = rec_copy(buf, prev_version, *offsets);
rec_offs_make_valid(*old_vers, index, *offsets);
err = DB_SUCCESS;
break;
}
version = prev_version;
}/* for (;;) */
mem_heap_free(heap);
rw_lock_s_unlock(&(purge_sys->latch));
return(err);
}
/*************************************************************//**
Inserts an entry into a hash table. If an entry with the same fold number
is found, its node is updated to point to the new data, and no new node
is inserted. If btr_search_enabled is set to FALSE, we will only allow
updating existing nodes, but no new node is allowed to be added.
@return TRUE if succeed, FALSE if no more memory could be allocated */
UNIV_INTERN
ibool
ha_insert_for_fold_func(
/*====================*/
hash_table_t* table, /*!< in: hash table */
ulint fold, /*!< in: folded value of data; if a node with
the same fold value already exists, it is
updated to point to the same data, and no new
node is created! */
#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
buf_block_t* block, /*!< in: buffer block containing the data */
#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
rec_t* data) /*!< in: data, must not be NULL */
{
hash_cell_t* cell;
ha_node_t* node;
ha_node_t* prev_node;
ulint hash;
ut_ad(data);
ut_ad(table);
ut_ad(table->magic_n == HASH_TABLE_MAGIC_N);
#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
ut_a(block->frame == page_align(data));
#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
#ifdef UNIV_SYNC_DEBUG
ut_ad(rw_lock_own(&btr_search_latch, RW_LOCK_EX));
#endif /* UNIV_SYNC_DEBUG */
ASSERT_HASH_MUTEX_OWN(table, fold);
ut_ad(btr_search_enabled);
hash = hash_calc_hash(fold, table);
cell = hash_get_nth_cell(table, hash);
prev_node = cell->node;
while (prev_node != NULL) {
if (prev_node->fold == fold) {
#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
# ifndef UNIV_HOTBACKUP
if (table->adaptive) {
buf_block_t* prev_block = prev_node->block;
ut_a(prev_block->frame
== page_align(prev_node->data));
ut_a(prev_block->n_pointers > 0);
prev_block->n_pointers--;
block->n_pointers++;
}
# endif /* !UNIV_HOTBACKUP */
prev_node->block = block;
#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
prev_node->data = data;
return(TRUE);
}
prev_node = prev_node->next;
}
/* We have to allocate a new chain node */
node = mem_heap_alloc(hash_get_heap(table, fold), sizeof(ha_node_t));
if (node == NULL) {
/* It was a btr search type memory heap and at the moment
no more memory could be allocated: return */
ut_ad(hash_get_heap(table, fold)->type & MEM_HEAP_BTR_SEARCH);
return(FALSE);
}
ha_node_set_data(node, block, data);
#if defined UNIV_AHI_DEBUG || defined UNIV_DEBUG
# ifndef UNIV_HOTBACKUP
if (table->adaptive) {
block->n_pointers++;
}
# endif /* !UNIV_HOTBACKUP */
#endif /* UNIV_AHI_DEBUG || UNIV_DEBUG */
node->fold = fold;
node->next = NULL;
prev_node = cell->node;
if (prev_node == NULL) {
cell->node = node;
return(TRUE);
}
while (prev_node->next != NULL) {
prev_node = prev_node->next;
}
prev_node->next = node;
return(TRUE);
}
/************************************************************************
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);
}
byte*
trx_undo_update_rec_get_update(
/*===========================*/
/* out: remaining part of the record,
NULL if an error detected, which means that
the record is corrupted */
byte* ptr, /* in: remaining part in update undo log
record, after reading the row reference
NOTE that this copy of the undo log record must
be preserved as long as the update vector is
used, as we do NOT copy the data in the
record! */
dict_index_t* index, /* in: clustered index */
ulint type, /* in: TRX_UNDO_UPD_EXIST_REC,
TRX_UNDO_UPD_DEL_REC, or
TRX_UNDO_DEL_MARK_REC; in the last case,
only trx id and roll ptr fields are added to
the update vector */
dulint trx_id, /* in: transaction id from this undo record */
dulint roll_ptr,/* in: roll pointer from this undo record */
ulint info_bits,/* in: info bits from this undo record */
trx_t* trx, /* in: transaction */
mem_heap_t* heap, /* in: memory heap from which the memory
needed is allocated */
upd_t** upd) /* out, own: update vector */
{
upd_field_t* upd_field;
upd_t* update;
ulint n_fields;
byte* buf;
byte* field;
ulint len;
ulint field_no;
ulint i;
ut_a(index->type & DICT_CLUSTERED);
if (type != TRX_UNDO_DEL_MARK_REC) {
ptr = trx_undo_update_rec_get_n_upd_fields(ptr, &n_fields);
} else {
n_fields = 0;
}
update = upd_create(n_fields + 2, heap);
update->info_bits = info_bits;
/* Store first trx id and roll ptr to update vector */
upd_field = upd_get_nth_field(update, n_fields);
buf = mem_heap_alloc(heap, DATA_TRX_ID_LEN);
trx_write_trx_id(buf, trx_id);
upd_field_set_field_no(upd_field,
dict_index_get_sys_col_pos(index, DATA_TRX_ID),
index, trx);
dfield_set_data(&(upd_field->new_val), buf, DATA_TRX_ID_LEN);
upd_field = upd_get_nth_field(update, n_fields + 1);
buf = mem_heap_alloc(heap, DATA_ROLL_PTR_LEN);
trx_write_roll_ptr(buf, roll_ptr);
upd_field_set_field_no(upd_field,
dict_index_get_sys_col_pos(index, DATA_ROLL_PTR),
index, trx);
dfield_set_data(&(upd_field->new_val), buf, DATA_ROLL_PTR_LEN);
/* Store then the updated ordinary columns to the update vector */
for (i = 0; i < n_fields; i++) {
ptr = trx_undo_update_rec_get_field_no(ptr, &field_no);
if (field_no >= dict_index_get_n_fields(index)) {
fprintf(stderr,
"InnoDB: Error: trying to access update undo rec field %lu in ", (ulong) field_no);
dict_index_name_print(stderr, trx, index);
fprintf(stderr, "\n"
"InnoDB: but index has only %lu fields\n"
"InnoDB: Submit a detailed bug report to http://bugs.mysql.com\n"
"InnoDB: Run also CHECK TABLE ",
(ulong) dict_index_get_n_fields(index));
ut_print_name(stderr, trx, index->table_name);
fprintf(stderr, "\n"
"InnoDB: n_fields = %lu, i = %lu, ptr %p\n",
(ulong) n_fields, (ulong) i, ptr);
return(NULL);
}
ptr = trx_undo_rec_get_col_val(ptr, &field, &len);
upd_field = upd_get_nth_field(update, i);
upd_field_set_field_no(upd_field, field_no, index, trx);
if (len != UNIV_SQL_NULL && len >= UNIV_EXTERN_STORAGE_FIELD) {
upd_field->extern_storage = TRUE;
len -= UNIV_EXTERN_STORAGE_FIELD;
}
dfield_set_data(&(upd_field->new_val), field, len);
}
*upd = update;
return(ptr);
}
big_rec_t*
dtuple_convert_big_rec(
/*===================*/
/* out, own: created big record vector,
NULL if we are not able to shorten
the entry enough, i.e., if there are
too many short fields in entry */
dict_index_t* index, /* in: index */
dtuple_t* entry, /* in: index entry */
ulint* ext_vec,/* in: array of externally stored fields,
or NULL: if a field already is externally
stored, then we cannot move it to the vector
this function returns */
ulint n_ext_vec)/* in: number of elements is ext_vec */
{
mem_heap_t* heap;
big_rec_t* vector;
dfield_t* dfield;
ulint size;
ulint n_fields;
ulint longest;
ulint longest_i = ULINT_MAX;
ibool is_externally_stored;
ulint i;
ulint j;
char err_buf[1000];
ut_a(dtuple_check_typed_no_assert(entry));
size = rec_get_converted_size(entry);
if (size > 1000000000) {
fprintf(stderr,
"InnoDB: Warning: tuple size very big: %lu\n", size);
dtuple_sprintf(err_buf, 900, entry);
fprintf(stderr,
"InnoDB: Tuple contents: %s\n", err_buf);
}
heap = mem_heap_create(size + dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t) + 1000);
vector = mem_heap_alloc(heap, sizeof(big_rec_t));
vector->heap = heap;
vector->fields = mem_heap_alloc(heap, dtuple_get_n_fields(entry)
* sizeof(big_rec_field_t));
/* Decide which fields to shorten: the algorithm is to look for
the longest field whose type is DATA_BLOB */
n_fields = 0;
while ((rec_get_converted_size(entry)
>= page_get_free_space_of_empty() / 2)
|| rec_get_converted_size(entry) >= REC_MAX_DATA_SIZE) {
longest = 0;
for (i = dict_index_get_n_unique_in_tree(index);
i < dtuple_get_n_fields(entry); i++) {
/* Skip over fields which already are externally
stored */
is_externally_stored = FALSE;
if (ext_vec) {
for (j = 0; j < n_ext_vec; j++) {
if (ext_vec[j] == i) {
is_externally_stored = TRUE;
}
}
}
if (!is_externally_stored
&& dict_index_get_nth_type(index, i)->mtype
== DATA_BLOB) {
dfield = dtuple_get_nth_field(entry, i);
if (dfield->len != UNIV_SQL_NULL &&
dfield->len > longest) {
longest = dfield->len;
longest_i = i;
}
}
}
/* We do not store externally fields which are smaller than
DICT_MAX_COL_PREFIX_LEN */
ut_a(DICT_MAX_COL_PREFIX_LEN > REC_1BYTE_OFFS_LIMIT);
if (longest < BTR_EXTERN_FIELD_REF_SIZE + 10
+ DICT_MAX_COL_PREFIX_LEN) {
/* Cannot shorten more */
mem_heap_free(heap);
return(NULL);
}
/* Move data from field longest_i to big rec vector;
we do not let data size of the remaining entry
drop below 128 which is the limit for the 2-byte
offset storage format in a physical record. This
we accomplish by storing 128 bytes of data in entry
itself, and only the remaining part to big rec vec.
We store the first bytes locally to the record. Then
we can calculate all ordering fields in all indexes
from locally stored data. */
dfield = dtuple_get_nth_field(entry, longest_i);
vector->fields[n_fields].field_no = longest_i;
ut_a(dfield->len > DICT_MAX_COL_PREFIX_LEN);
vector->fields[n_fields].len = dfield->len
- DICT_MAX_COL_PREFIX_LEN;
vector->fields[n_fields].data = mem_heap_alloc(heap,
vector->fields[n_fields].len);
/* Copy data (from the end of field) to big rec vector */
ut_memcpy(vector->fields[n_fields].data,
((byte*)dfield->data) + dfield->len
- vector->fields[n_fields].len,
vector->fields[n_fields].len);
dfield->len = dfield->len - vector->fields[n_fields].len
+ BTR_EXTERN_FIELD_REF_SIZE;
/* Set the extern field reference in dfield to zero */
memset(((byte*)dfield->data)
+ dfield->len - BTR_EXTERN_FIELD_REF_SIZE,
0, BTR_EXTERN_FIELD_REF_SIZE);
n_fields++;
}
vector->n_fields = n_fields;
return(vector);
}
/*****************************************************************//**
Based on a table object, this function builds the entry to be inserted
in the SYS_TABLES system table.
@return the tuple which should be inserted */
static
dtuple_t*
dict_create_sys_tables_tuple(
/*=========================*/
const dict_table_t* table, /*!< in: table */
mem_heap_t* heap) /*!< in: memory heap from
which the memory for the built
tuple is allocated */
{
dict_table_t* sys_tables;
dtuple_t* entry;
dfield_t* dfield;
byte* ptr;
ut_ad(table);
ut_ad(heap);
sys_tables = dict_sys->sys_tables;
entry = dtuple_create(heap, 8 + DATA_N_SYS_COLS);
dict_table_copy_types(entry, sys_tables);
/* 0: NAME -----------------------------*/
dfield = dtuple_get_nth_field(entry, 0/*NAME*/);
dfield_set_data(dfield, table->name, ut_strlen(table->name));
/* 3: ID -------------------------------*/
dfield = dtuple_get_nth_field(entry, 1/*ID*/);
ptr = mem_heap_alloc(heap, 8);
mach_write_to_8(ptr, table->id);
dfield_set_data(dfield, ptr, 8);
/* 4: N_COLS ---------------------------*/
dfield = dtuple_get_nth_field(entry, 2/*N_COLS*/);
#if DICT_TF_COMPACT != 1
#error
#endif
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, table->n_def
| ((table->flags & DICT_TF_COMPACT) << 31));
dfield_set_data(dfield, ptr, 4);
/* 5: TYPE -----------------------------*/
dfield = dtuple_get_nth_field(entry, 3/*TYPE*/);
ptr = mem_heap_alloc(heap, 4);
if (table->flags & (~DICT_TF_COMPACT & ~(~0 << DICT_TF_BITS))) {
ut_a(table->flags & DICT_TF_COMPACT);
ut_a(dict_table_get_format(table) >= DICT_TF_FORMAT_ZIP);
ut_a((table->flags & DICT_TF_ZSSIZE_MASK)
<= (DICT_TF_ZSSIZE_MAX << DICT_TF_ZSSIZE_SHIFT));
ut_a(!(table->flags & (~0 << DICT_TF2_BITS)));
mach_write_to_4(ptr, table->flags & ~(~0 << DICT_TF_BITS));
} else {
mach_write_to_4(ptr, DICT_TABLE_ORDINARY);
}
dfield_set_data(dfield, ptr, 4);
/* 6: MIX_ID (obsolete) ---------------------------*/
dfield = dtuple_get_nth_field(entry, 4/*MIX_ID*/);
ptr = mem_heap_zalloc(heap, 8);
dfield_set_data(dfield, ptr, 8);
/* 7: MIX_LEN (additional flags) --------------------------*/
dfield = dtuple_get_nth_field(entry, 5/*MIX_LEN*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, table->flags >> DICT_TF2_SHIFT);
dfield_set_data(dfield, ptr, 4);
/* 8: CLUSTER_NAME ---------------------*/
dfield = dtuple_get_nth_field(entry, 6/*CLUSTER_NAME*/);
dfield_set_null(dfield); /* not supported */
/* 9: SPACE ----------------------------*/
dfield = dtuple_get_nth_field(entry, 7/*SPACE*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, table->space);
dfield_set_data(dfield, ptr, 4);
/*----------------------------------*/
return(entry);
}
/*****************************************************************//**
Based on an index object, this function builds the entry to be inserted
in the SYS_FIELDS system table.
@return the tuple which should be inserted */
static
dtuple_t*
dict_create_sys_fields_tuple(
/*=========================*/
const dict_index_t* index, /*!< in: index */
ulint i, /*!< in: field number */
mem_heap_t* heap) /*!< in: memory heap from
which the memory for the built
tuple is allocated */
{
dict_table_t* sys_fields;
dtuple_t* entry;
dict_field_t* field;
dfield_t* dfield;
byte* ptr;
ibool index_contains_column_prefix_field = FALSE;
ulint j;
ut_ad(index);
ut_ad(heap);
for (j = 0; j < index->n_fields; j++) {
if (dict_index_get_nth_field(index, j)->prefix_len > 0) {
index_contains_column_prefix_field = TRUE;
break;
}
}
field = dict_index_get_nth_field(index, i);
sys_fields = dict_sys->sys_fields;
entry = dtuple_create(heap, 3 + DATA_N_SYS_COLS);
dict_table_copy_types(entry, sys_fields);
/* 0: INDEX_ID -----------------------*/
dfield = dtuple_get_nth_field(entry, 0/*INDEX_ID*/);
ptr = mem_heap_alloc(heap, 8);
mach_write_to_8(ptr, index->id);
dfield_set_data(dfield, ptr, 8);
/* 1: POS + PREFIX LENGTH ----------------------------*/
dfield = dtuple_get_nth_field(entry, 1/*POS*/);
ptr = mem_heap_alloc(heap, 4);
if (index_contains_column_prefix_field) {
/* If there are column prefix fields in the index, then
we store the number of the field to the 2 HIGH bytes
and the prefix length to the 2 low bytes, */
mach_write_to_4(ptr, (i << 16) + field->prefix_len);
} else {
/* Else we store the number of the field to the 2 LOW bytes.
This is to keep the storage format compatible with
InnoDB versions < 4.0.14. */
mach_write_to_4(ptr, i);
}
dfield_set_data(dfield, ptr, 4);
/* 4: COL_NAME -------------------------*/
dfield = dtuple_get_nth_field(entry, 2/*COL_NAME*/);
dfield_set_data(dfield, field->name,
ut_strlen(field->name));
/*---------------------------------*/
return(entry);
}
/*****************************************************************//**
Based on an index object, this function builds the entry to be inserted
in the SYS_INDEXES system table.
@return the tuple which should be inserted */
static
dtuple_t*
dict_create_sys_indexes_tuple(
/*==========================*/
const dict_index_t* index, /*!< in: index */
mem_heap_t* heap) /*!< in: memory heap from
which the memory for the built
tuple is allocated */
{
dict_table_t* sys_indexes;
dict_table_t* table;
dtuple_t* entry;
dfield_t* dfield;
byte* ptr;
ut_ad(mutex_own(&(dict_sys->mutex)));
ut_ad(index);
ut_ad(heap);
sys_indexes = dict_sys->sys_indexes;
table = dict_table_get_low(index->table_name);
entry = dtuple_create(heap, 7 + DATA_N_SYS_COLS);
dict_table_copy_types(entry, sys_indexes);
/* 0: TABLE_ID -----------------------*/
dfield = dtuple_get_nth_field(entry, 0/*TABLE_ID*/);
ptr = mem_heap_alloc(heap, 8);
mach_write_to_8(ptr, table->id);
dfield_set_data(dfield, ptr, 8);
/* 1: ID ----------------------------*/
dfield = dtuple_get_nth_field(entry, 1/*ID*/);
ptr = mem_heap_alloc(heap, 8);
mach_write_to_8(ptr, index->id);
dfield_set_data(dfield, ptr, 8);
/* 4: NAME --------------------------*/
dfield = dtuple_get_nth_field(entry, 2/*NAME*/);
dfield_set_data(dfield, index->name, ut_strlen(index->name));
/* 5: N_FIELDS ----------------------*/
dfield = dtuple_get_nth_field(entry, 3/*N_FIELDS*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, index->n_fields);
dfield_set_data(dfield, ptr, 4);
/* 6: TYPE --------------------------*/
dfield = dtuple_get_nth_field(entry, 4/*TYPE*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, index->type);
dfield_set_data(dfield, ptr, 4);
/* 7: SPACE --------------------------*/
#if DICT_SYS_INDEXES_SPACE_NO_FIELD != 7
#error "DICT_SYS_INDEXES_SPACE_NO_FIELD != 7"
#endif
dfield = dtuple_get_nth_field(entry, 5/*SPACE*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, index->space);
dfield_set_data(dfield, ptr, 4);
/* 8: PAGE_NO --------------------------*/
#if DICT_SYS_INDEXES_PAGE_NO_FIELD != 8
#error "DICT_SYS_INDEXES_PAGE_NO_FIELD != 8"
#endif
dfield = dtuple_get_nth_field(entry, 6/*PAGE_NO*/);
ptr = mem_heap_alloc(heap, 4);
mach_write_to_4(ptr, FIL_NULL);
dfield_set_data(dfield, ptr, 4);
/*--------------------------------*/
return(entry);
}
/************************************************************************
Loads definitions for table columns. */
static
void
dict_load_columns(
/*==============*/
dict_table_t* table, /* in: table */
mem_heap_t* heap) /* in: memory heap for temporary storage */
{
dict_table_t* sys_columns;
dict_index_t* sys_index;
btr_pcur_t pcur;
dtuple_t* tuple;
dfield_t* dfield;
rec_t* rec;
byte* field;
ulint len;
byte* buf;
char* name_buf;
char* name;
ulint mtype;
ulint prtype;
ulint col_len;
ulint prec;
ulint i;
mtr_t mtr;
ut_ad(mutex_own(&(dict_sys->mutex)));
mtr_start(&mtr);
sys_columns = dict_table_get_low("SYS_COLUMNS");
sys_index = UT_LIST_GET_FIRST(sys_columns->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 (i = 0; i < table->n_cols - DATA_N_SYS_COLS; 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_dulint_cmp(table->id, mach_read_from_8(field)) == 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("NAME",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_columns), 4))->name));
field = rec_get_nth_field(rec, 4, &len);
name_buf = mem_heap_alloc(heap, len + 1);
ut_memcpy(name_buf, field, len);
name_buf[len] = '\0';
name = name_buf;
field = rec_get_nth_field(rec, 5, &len);
mtype = mach_read_from_4(field);
field = rec_get_nth_field(rec, 6, &len);
prtype = mach_read_from_4(field);
field = rec_get_nth_field(rec, 7, &len);
col_len = mach_read_from_4(field);
ut_a(0 == ut_strcmp("PREC",
dict_field_get_col(
dict_index_get_nth_field(
dict_table_get_first_index(sys_columns), 8))->name));
field = rec_get_nth_field(rec, 8, &len);
prec = mach_read_from_4(field);
dict_mem_table_add_col(table, name, mtype, prtype, col_len,
prec);
btr_pcur_move_to_next_user_rec(&pcur, &mtr);
}
btr_pcur_close(&pcur);
mtr_commit(&mtr);
}
/*******************************************************************//**
An inverse function to row_build_index_entry. Builds a row from a
record in a clustered index.
@return own: row built; see the NOTE below! */
UNIV_INTERN
dtuple_t*
row_build(
/*======*/
ulint type, /*!< in: ROW_COPY_POINTERS or
ROW_COPY_DATA; the latter
copies also the data fields to
heap while the first only
places pointers to data fields
on the index page, and thus is
more efficient */
const dict_index_t* index, /*!< in: clustered index */
const rec_t* rec, /*!< in: record in the clustered
index; NOTE: in the case
ROW_COPY_POINTERS the data
fields in the row will point
directly into this record,
therefore, the buffer page of
this record must be at least
s-latched and the latch held
as long as the row dtuple is used! */
const ulint* offsets,/*!< in: rec_get_offsets(rec,index)
or NULL, in which case this function
will invoke rec_get_offsets() */
const dict_table_t* col_table,
/*!< in: table, to check which
externally stored columns
occur in the ordering columns
of an index, or NULL if
index->table should be
consulted instead */
row_ext_t** ext, /*!< out, own: cache of
externally stored column
prefixes, or NULL */
mem_heap_t* heap) /*!< in: memory heap from which
the memory needed is allocated */
{
dtuple_t* row;
const dict_table_t* table;
ulint n_fields;
ulint n_ext_cols;
ulint* ext_cols = NULL; /* remove warning */
ulint len;
ulint row_len;
byte* buf;
ulint i;
ulint j;
mem_heap_t* tmp_heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
rec_offs_init(offsets_);
ut_ad(index && rec && heap);
ut_ad(dict_index_is_clust(index));
ut_ad(!mutex_own(&kernel_mutex));
if (!offsets) {
offsets = rec_get_offsets(rec, index, offsets_,
ULINT_UNDEFINED, &tmp_heap);
} else {
ut_ad(rec_offs_validate(rec, index, offsets));
}
#if defined UNIV_DEBUG || defined UNIV_BLOB_LIGHT_DEBUG
if (rec_offs_any_null_extern(rec, offsets)) {
/* This condition can occur during crash recovery
before trx_rollback_active() has completed execution,
or when a concurrently executing
row_ins_index_entry_low() has committed the B-tree
mini-transaction but has not yet managed to restore
the cursor position for writing the big_rec. */
ut_a(trx_undo_roll_ptr_is_insert(
row_get_rec_roll_ptr(rec, index, offsets)));
}
#endif /* UNIV_DEBUG || UNIV_BLOB_LIGHT_DEBUG */
if (type != ROW_COPY_POINTERS) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_offs_size(offsets));
rec = rec_copy(buf, rec, offsets);
/* Avoid a debug assertion in rec_offs_validate(). */
rec_offs_make_valid(rec, index, (ulint*) offsets);
}
table = index->table;
row_len = dict_table_get_n_cols(table);
row = dtuple_create(heap, row_len);
dict_table_copy_types(row, table);
dtuple_set_info_bits(row, rec_get_info_bits(
rec, dict_table_is_comp(table)));
n_fields = rec_offs_n_fields(offsets);
n_ext_cols = rec_offs_n_extern(offsets);
if (n_ext_cols) {
ext_cols = mem_heap_alloc(heap, n_ext_cols * sizeof *ext_cols);
}
for (i = j = 0; i < n_fields; i++) {
dict_field_t* ind_field
= dict_index_get_nth_field(index, i);
const dict_col_t* col
= dict_field_get_col(ind_field);
ulint col_no
= dict_col_get_no(col);
dfield_t* dfield
= dtuple_get_nth_field(row, col_no);
if (ind_field->prefix_len == 0) {
const byte* field = rec_get_nth_field(
rec, offsets, i, &len);
dfield_set_data(dfield, field, len);
}
if (rec_offs_nth_extern(offsets, i)) {
dfield_set_ext(dfield);
if (UNIV_LIKELY_NULL(col_table)) {
ut_a(col_no
< dict_table_get_n_cols(col_table));
col = dict_table_get_nth_col(
col_table, col_no);
}
if (col->ord_part) {
/* We will have to fetch prefixes of
externally stored columns that are
referenced by column prefixes. */
ext_cols[j++] = col_no;
}
}
}
ut_ad(dtuple_check_typed(row));
if (!ext) {
/* REDUNDANT and COMPACT formats store a local
768-byte prefix of each externally stored
column. No cache is needed. */
ut_ad(dict_table_get_format(index->table)
< DICT_TF_FORMAT_ZIP);
} else if (j) {
*ext = row_ext_create(j, ext_cols, row,
dict_table_zip_size(index->table),
heap);
} else {
*ext = NULL;
}
if (tmp_heap) {
mem_heap_free(tmp_heap);
}
return(row);
}
/************************************************************************
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);
}
/*******************************************************************//**
Builds from a secondary index record a row reference with which we can
search the clustered index record.
@return own: row reference built; see the NOTE below! */
UNIV_INTERN
dtuple_t*
row_build_row_ref(
/*==============*/
ulint type, /*!< in: ROW_COPY_DATA, or ROW_COPY_POINTERS:
the former copies also the data fields to
heap, whereas the latter only places pointers
to data fields on the index page */
dict_index_t* index, /*!< in: secondary index */
const rec_t* rec, /*!< in: record in the index;
NOTE: in the case ROW_COPY_POINTERS
the data fields in the row will point
directly into this record, therefore,
the buffer page of this record must be
at least s-latched and the latch held
as long as the row reference is used! */
mem_heap_t* heap) /*!< in: memory heap from which the memory
needed is allocated */
{
dict_table_t* table;
dict_index_t* clust_index;
dfield_t* dfield;
dtuple_t* ref;
const byte* field;
ulint len;
ulint ref_len;
ulint pos;
byte* buf;
ulint clust_col_prefix_len;
ulint i;
mem_heap_t* tmp_heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
ulint* offsets = offsets_;
rec_offs_init(offsets_);
ut_ad(index && rec && heap);
ut_ad(!dict_index_is_clust(index));
offsets = rec_get_offsets(rec, index, offsets,
ULINT_UNDEFINED, &tmp_heap);
/* Secondary indexes must not contain externally stored columns. */
ut_ad(!rec_offs_any_extern(offsets));
if (type == ROW_COPY_DATA) {
/* Take a copy of rec to heap */
buf = mem_heap_alloc(heap, rec_offs_size(offsets));
rec = rec_copy(buf, rec, offsets);
/* Avoid a debug assertion in rec_offs_validate(). */
rec_offs_make_valid(rec, index, offsets);
}
table = index->table;
clust_index = dict_table_get_first_index(table);
ref_len = dict_index_get_n_unique(clust_index);
ref = dtuple_create(heap, ref_len);
dict_index_copy_types(ref, clust_index, ref_len);
for (i = 0; i < ref_len; i++) {
dfield = dtuple_get_nth_field(ref, i);
pos = dict_index_get_nth_field_pos(index, clust_index, i);
ut_a(pos != ULINT_UNDEFINED);
field = rec_get_nth_field(rec, offsets, pos, &len);
dfield_set_data(dfield, field, len);
/* If the primary key contains a column prefix, then the
secondary index may contain a longer prefix of the same
column, or the full column, and we must adjust the length
accordingly. */
clust_col_prefix_len = dict_index_get_nth_field(
clust_index, i)->prefix_len;
if (clust_col_prefix_len > 0) {
if (len != UNIV_SQL_NULL) {
const dtype_t* dtype
= dfield_get_type(dfield);
dfield_set_len(dfield,
dtype_get_at_most_n_mbchars(
dtype->prtype,
dtype->mbminlen,
dtype->mbmaxlen,
clust_col_prefix_len,
len, (char*) field));
}
}
}
ut_ad(dtuple_check_typed(ref));
if (tmp_heap) {
mem_heap_free(tmp_heap);
}
return(ref);
}
/************************************************************************
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);
}
/********************************************************************//**
Add a single foreign key definition to the data dictionary tables in the
database. We also generate names to constraints that were not named by the
user. A generated constraint has a name of the format
databasename/tablename_ibfk_NUMBER, where the numbers start from 1, and
are given locally for this table, that is, the number is not global, as in
the old format constraints < 4.0.18 it used to be.
@return error code or DB_SUCCESS */
static
ulint
dict_create_add_foreign_to_dictionary(
/*==================================*/
ulint* id_nr, /*!< in/out: number to use in id generation;
incremented if used */
dict_table_t* table, /*!< in: table */
dict_foreign_t* foreign,/*!< in: foreign */
trx_t* trx) /*!< in: transaction */
{
ulint error;
ulint i;
pars_info_t* info = pars_info_create();
if (foreign->id == NULL) {
/* Generate a new constraint id */
ulint namelen = strlen(table->name);
char* id = mem_heap_alloc(foreign->heap, namelen + 20);
/* no overflow if number < 1e13 */
sprintf(id, "%s_ibfk_%lu", table->name, (ulong) (*id_nr)++);
foreign->id = id;
}
pars_info_add_str_literal(info, "id", foreign->id);
pars_info_add_str_literal(info, "for_name", table->name);
pars_info_add_str_literal(info, "ref_name",
foreign->referenced_table_name);
pars_info_add_int4_literal(info, "n_cols",
foreign->n_fields + (foreign->type << 24));
error = dict_foreign_eval_sql(info,
"PROCEDURE P () IS\n"
"BEGIN\n"
"INSERT INTO SYS_FOREIGN VALUES"
"(:id, :for_name, :ref_name, :n_cols);\n"
"END;\n"
, table, foreign, trx);
if (error != DB_SUCCESS) {
return(error);
}
for (i = 0; i < foreign->n_fields; i++) {
error = dict_create_add_foreign_field_to_dictionary(
i, table, foreign, trx);
if (error != DB_SUCCESS) {
return(error);
}
}
error = dict_foreign_eval_sql(NULL,
"PROCEDURE P () IS\n"
"BEGIN\n"
"COMMIT WORK;\n"
"END;\n"
, table, foreign, trx);
return(error);
}
