ibool
dtuple_check_typed_no_assert(
/*=========================*/
/* out: TRUE if ok */
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
ulint i;
if (dtuple_get_n_fields(tuple) > REC_MAX_N_FIELDS) {
fprintf(stderr,
"InnoDB: Error: index entry has %lu fields\n",
(ulong) dtuple_get_n_fields(tuple));
dump:
fputs("InnoDB: Tuple contents: ", stderr);
dtuple_print(stderr, tuple);
putc('\n', stderr);
return(FALSE);
}
for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
field = dtuple_get_nth_field(tuple, i);
if (!dfield_check_typed_no_assert(field)) {
goto dump;
}
}
return(TRUE);
}
upd_t*
row_upd_build_sec_rec_difference_binary(
/*====================================*/
/* out, own: update vector of differing
fields */
dict_index_t* index, /* in: index */
dtuple_t* entry, /* in: entry to insert */
rec_t* rec, /* in: secondary index record */
mem_heap_t* heap) /* in: memory heap from which allocated */
{
upd_field_t* upd_field;
dfield_t* dfield;
byte* data;
ulint len;
upd_t* update;
ulint n_diff;
ulint i;
/* This function is used only for a secondary index */
ut_ad(0 == (index->type & DICT_CLUSTERED));
update = upd_create(dtuple_get_n_fields(entry), heap);
n_diff = 0;
for (i = 0; i < dtuple_get_n_fields(entry); i++) {
data = rec_get_nth_field(rec, i, &len);
dfield = dtuple_get_nth_field(entry, i);
ut_a(len == dfield_get_len(dfield));
/* NOTE: we compare the fields as binary strings!
(No collation) */
if (!dfield_data_is_binary_equal(dfield, len, data)) {
upd_field = upd_get_nth_field(update, n_diff);
dfield_copy(&(upd_field->new_val), dfield);
upd_field_set_field_no(upd_field, i, index);
upd_field->extern_storage = FALSE;
n_diff++;
}
}
update->n_fields = n_diff;
return(update);
}
/**************************************************************//**
Checks if a dtuple is a prefix of a record. The last field in dtuple
is allowed to be a prefix of the corresponding field in the record.
@return TRUE if prefix */
UNIV_INTERN
ibool
cmp_dtuple_is_prefix_of_rec(
/*========================*/
const dtuple_t* dtuple, /*!< in: data tuple */
const rec_t* rec, /*!< in: physical record */
const ulint* offsets)/*!< in: array returned by rec_get_offsets() */
{
ulint n_fields;
ulint matched_fields = 0;
ulint matched_bytes = 0;
ut_ad(rec_offs_validate(rec, NULL, offsets));
n_fields = dtuple_get_n_fields(dtuple);
if (n_fields > rec_offs_n_fields(offsets)) {
return(FALSE);
}
cmp_dtuple_rec_with_match(dtuple, rec, offsets,
&matched_fields, &matched_bytes);
if (matched_fields == n_fields) {
return(TRUE);
}
if (matched_fields == n_fields - 1
&& matched_bytes == dfield_get_len(
dtuple_get_nth_field(dtuple, n_fields - 1))) {
return(TRUE);
}
return(FALSE);
}
void
dtuple_print(
/*=========*/
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
ulint n_fields;
ulint i;
n_fields = dtuple_get_n_fields(tuple);
printf("DATA TUPLE: %lu fields;\n", n_fields);
for (i = 0; i < n_fields; i++) {
printf(" %lu:", i);
field = dtuple_get_nth_field(tuple, i);
if (field->len != UNIV_SQL_NULL) {
ut_print_buf(field->data, field->len);
} else {
printf(" SQL NULL");
}
printf(";");
}
printf("\n");
dtuple_validate(tuple);
}
/***************************************************************//**
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);
}
/**********************************************************//**
The following function prints the contents of a tuple. */
UNIV_INTERN
void
dtuple_print(
/*=========*/
FILE* f, /*!< in: output stream */
const dtuple_t* tuple) /*!< in: tuple */
{
ulint n_fields;
ulint i;
n_fields = dtuple_get_n_fields(tuple);
fprintf(f, "DATA TUPLE: %lu fields;\n", (ulong) n_fields);
for (i = 0; i < n_fields; i++) {
fprintf(f, " %lu:", (ulong) i);
dfield_print_raw(f, dtuple_get_nth_field(tuple, i));
putc(';', f);
putc('\n', f);
}
ut_ad(dtuple_validate(tuple));
}
void
dtuple_print(
/*=========*/
FILE* f, /* in: output stream */
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
ulint n_fields;
ulint i;
n_fields = dtuple_get_n_fields(tuple);
fprintf(f, "DATA TUPLE: %lu fields;\n", (ulong) n_fields);
for (i = 0; i < n_fields; i++) {
fprintf(f, " %lu:", (ulong) i);
field = dtuple_get_nth_field(tuple, i);
if (field->len != UNIV_SQL_NULL) {
ut_print_buf(f, field->data, field->len);
} else {
fputs(" SQL NULL", f);
}
putc(';', f);
}
putc('\n', f);
ut_ad(dtuple_validate(tuple));
}
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
dtuple_datas_are_ordering_equal(
/*============================*/
/* out: TRUE if length and fieds are equal
when compared with cmp_data_data:
NOTE: in character type fields some letters
are identified with others! (collation) */
dtuple_t* tuple1, /* in: tuple 1 */
dtuple_t* tuple2) /* in: tuple 2 */
{
dfield_t* field1;
dfield_t* field2;
ulint n_fields;
ulint i;
ut_ad(tuple1 && tuple2);
ut_ad(tuple1->magic_n == DATA_TUPLE_MAGIC_N);
ut_ad(tuple2->magic_n == DATA_TUPLE_MAGIC_N);
ut_ad(dtuple_check_typed(tuple1));
ut_ad(dtuple_check_typed(tuple2));
n_fields = dtuple_get_n_fields(tuple1);
if (n_fields != dtuple_get_n_fields(tuple2)) {
return(FALSE);
}
for (i = 0; i < n_fields; i++) {
field1 = dtuple_get_nth_field(tuple1, i);
field2 = dtuple_get_nth_field(tuple2, i);
if (0 != cmp_dfield_dfield(field1, field2)) {
return(FALSE);
}
}
return(TRUE);
}
/***************************************************************//**
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
dtuple_check_typed_no_assert(
/*=========================*/
/* out: TRUE if ok */
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
ulint i;
char err_buf[1000];
if (dtuple_get_n_fields(tuple) > REC_MAX_N_FIELDS) {
fprintf(stderr,
"InnoDB: Error: index entry has %lu fields\n",
dtuple_get_n_fields(tuple));
dtuple_sprintf(err_buf, 900, tuple);
fprintf(stderr,
"InnoDB: Tuple contents: %s\n", err_buf);
return(FALSE);
}
for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
field = dtuple_get_nth_field(tuple, i);
if (!dfield_check_typed_no_assert(field)) {
dtuple_sprintf(err_buf, 900, tuple);
fprintf(stderr,
"InnoDB: Tuple contents: %s\n", err_buf);
return(FALSE);
}
}
return(TRUE);
}
/************************************************************//**
Compare two data tuples, respecting the collation of character fields.
@return 1, 0 , -1 if tuple1 is greater, equal, less, respectively,
than tuple2 */
UNIV_INTERN
int
dtuple_coll_cmp(
/*============*/
const dtuple_t* tuple1, /*!< in: tuple 1 */
const dtuple_t* tuple2) /*!< in: tuple 2 */
{
ulint n_fields;
ulint i;
ut_ad(tuple1 && tuple2);
ut_ad(tuple1->magic_n == DATA_TUPLE_MAGIC_N);
ut_ad(tuple2->magic_n == DATA_TUPLE_MAGIC_N);
ut_ad(dtuple_check_typed(tuple1));
ut_ad(dtuple_check_typed(tuple2));
n_fields = dtuple_get_n_fields(tuple1);
if (n_fields != dtuple_get_n_fields(tuple2)) {
return(n_fields < dtuple_get_n_fields(tuple2) ? -1 : 1);
}
for (i = 0; i < n_fields; i++) {
int cmp;
const dfield_t* field1 = dtuple_get_nth_field(tuple1, i);
const dfield_t* field2 = dtuple_get_nth_field(tuple2, i);
cmp = cmp_dfield_dfield(field1, field2);
if (cmp) {
return(cmp);
}
}
return(0);
}
void
row_build_row_ref_from_row(
/*=======================*/
dtuple_t* ref, /* in/out: row reference built; see the
NOTE below! ref must have the right number
of fields! */
dict_table_t* table, /* in: table */
dtuple_t* row) /* in: row
NOTE: the data fields in ref will point
directly into data of this row */
{
dict_index_t* clust_index;
dict_field_t* field;
dfield_t* dfield;
dfield_t* dfield2;
dict_col_t* col;
ulint ref_len;
ulint i;
ut_ad(ref && table && row);
clust_index = dict_table_get_first_index(table);
ref_len = dict_index_get_n_unique(clust_index);
ut_ad(ref_len == dtuple_get_n_fields(ref));
for (i = 0; i < ref_len; i++) {
dfield = dtuple_get_nth_field(ref, i);
field = dict_index_get_nth_field(clust_index, i);
col = dict_field_get_col(field);
dfield2 = dtuple_get_nth_field(row, dict_col_get_no(col));
dfield_copy(dfield, dfield2);
if (field->prefix_len > 0
&& dfield->len != UNIV_SQL_NULL
&& dfield->len > field->prefix_len) {
dfield->len = field->prefix_len;
}
}
ut_ad(dtuple_check_typed(ref));
}
/**********************************************************//**
Validates the consistency of a tuple which must be complete, i.e,
all fields must have been set.
@return TRUE if ok */
UNIV_INTERN
ibool
dtuple_validate(
/*============*/
const dtuple_t* tuple) /*!< in: tuple */
{
const dfield_t* field;
ulint n_fields;
ulint len;
ulint i;
ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
n_fields = dtuple_get_n_fields(tuple);
/* We dereference all the data of each field to test
for memory traps */
for (i = 0; i < n_fields; i++) {
field = dtuple_get_nth_field(tuple, i);
len = dfield_get_len(field);
if (!dfield_is_null(field)) {
const byte* data = dfield_get_data(field);
#ifndef UNIV_DEBUG_VALGRIND
ulint j;
for (j = 0; j < len; j++) {
data_dummy += *data; /* fool the compiler not
to optimize out this
code */
data++;
}
#endif /* !UNIV_DEBUG_VALGRIND */
UNIV_MEM_ASSERT_RW(data, len);
}
}
ut_a(dtuple_check_typed(tuple));
return(TRUE);
}
ibool
dtuple_check_typed(
/*===============*/
/* out: TRUE if ok */
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
ulint i;
for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
field = dtuple_get_nth_field(tuple, i);
ut_a(dfield_check_typed(field));
}
return(TRUE);
}
/**********************************************************//**
Checks that a data tuple is typed. Asserts an error if not.
@return TRUE if ok */
UNIV_INTERN
ibool
dtuple_check_typed(
/*===============*/
const dtuple_t* tuple) /*!< in: tuple */
{
const dfield_t* field;
ulint i;
for (i = 0; i < dtuple_get_n_fields(tuple); i++) {
field = dtuple_get_nth_field(tuple, i);
ut_a(dfield_check_typed(field));
}
return(TRUE);
}
ulint
dtuple_sprintf(
/*===========*/
/* out: printed length in bytes */
char* buf, /* in: print buffer */
ulint buf_len,/* in: buf length in bytes */
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
ulint n_fields;
ulint len;
ulint i;
len = 0;
n_fields = dtuple_get_n_fields(tuple);
for (i = 0; i < n_fields; i++) {
if (len + 30 > buf_len) {
return(len);
}
len += sprintf(buf + len, " %lu:", i);
field = dtuple_get_nth_field(tuple, i);
if (field->len != UNIV_SQL_NULL) {
if (5 * field->len + len + 30 > buf_len) {
return(len);
}
len += ut_sprintf_buf(buf + len, field->data,
field->len);
} else {
len += sprintf(buf + len, " SQL NULL");
}
len += sprintf(buf + len, ";");
}
return(len);
}
ibool
dtuple_validate(
/*============*/
/* out: TRUE if ok */
dtuple_t* tuple) /* in: tuple */
{
dfield_t* field;
byte* data;
ulint n_fields;
ulint len;
ulint i;
ulint j;
ut_ad(tuple->magic_n == DATA_TUPLE_MAGIC_N);
n_fields = dtuple_get_n_fields(tuple);
/* We dereference all the data of each field to test
for memory traps */
for (i = 0; i < n_fields; i++) {
field = dtuple_get_nth_field(tuple, i);
len = dfield_get_len(field);
if (len != UNIV_SQL_NULL) {
data = field->data;
for (j = 0; j < len; j++) {
data_dummy += *data; /* fool the compiler not
to optimize out this
code */
data++;
}
}
}
ut_a(dtuple_check_typed(tuple));
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);
}
/*******************************************************************//**
Builds from a secondary index record a row reference with which we can
search the clustered index record. */
UNIV_INTERN
void
row_build_row_ref_in_tuple(
/*=======================*/
dtuple_t* ref, /*!< in/out: row reference built;
see the NOTE below! */
const rec_t* rec, /*!< in: record in the index;
NOTE: the data fields in ref
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! */
const dict_index_t* index, /*!< in: secondary index */
ulint* offsets,/*!< in: rec_get_offsets(rec, index)
or NULL */
trx_t* trx) /*!< in: transaction */
{
const dict_index_t* clust_index;
dfield_t* dfield;
const byte* field;
ulint len;
ulint ref_len;
ulint pos;
ulint clust_col_prefix_len;
ulint i;
mem_heap_t* heap = NULL;
ulint offsets_[REC_OFFS_NORMAL_SIZE];
rec_offs_init(offsets_);
ut_a(ref);
ut_a(index);
ut_a(rec);
ut_ad(!dict_index_is_clust(index));
if (UNIV_UNLIKELY(!index->table)) {
fputs("InnoDB: table ", stderr);
notfound:
ut_print_name(stderr, trx, TRUE, index->table_name);
fputs(" for index ", stderr);
ut_print_name(stderr, trx, FALSE, index->name);
fputs(" not found\n", stderr);
ut_error;
}
clust_index = dict_table_get_first_index(index->table);
if (UNIV_UNLIKELY(!clust_index)) {
fputs("InnoDB: clust index for table ", stderr);
goto notfound;
}
if (!offsets) {
offsets = rec_get_offsets(rec, index, offsets_,
ULINT_UNDEFINED, &heap);
} else {
ut_ad(rec_offs_validate(rec, index, offsets));
}
/* Secondary indexes must not contain externally stored columns. */
ut_ad(!rec_offs_any_extern(offsets));
ref_len = dict_index_get_n_unique(clust_index);
ut_ad(ref_len == dtuple_get_n_fields(ref));
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 (UNIV_LIKELY_NULL(heap)) {
mem_heap_free(heap);
}
}
ulint
dtuple_get_n_fields_noninline(
dtuple_t* tuple) /* in: tuple */
{
return(dtuple_get_n_fields(tuple));
}
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;
ut_a(dtuple_check_typed_no_assert(entry));
size = rec_get_converted_size(index, entry);
if (UNIV_UNLIKELY(size > 1000000000)) {
fprintf(stderr,
"InnoDB: Warning: tuple size very big: %lu\n", (ulong) size);
fputs("InnoDB: Tuple contents: ", stderr);
dtuple_print(stderr, entry);
putc('\n', stderr);
}
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(index, entry)
>= ut_min(page_get_free_space_of_empty(
index->table->comp) / 2,
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) {
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_INDEX_COL_LEN */
ut_a(DICT_MAX_INDEX_COL_LEN > REC_1BYTE_OFFS_LIMIT);
if (longest < BTR_EXTERN_FIELD_REF_SIZE + 10
+ DICT_MAX_INDEX_COL_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_INDEX_COL_LEN);
vector->fields[n_fields].len = dfield->len
- DICT_MAX_INDEX_COL_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);
}
rec_t*
rec_convert_dtuple_to_rec_low(
/*==========================*/
/* out: pointer to the origin of physical
record */
byte* destination, /* in: start address of the physical record */
dtuple_t* dtuple, /* in: data tuple */
ulint data_size) /* in: data size of dtuple */
{
dfield_t* field;
ulint n_fields;
rec_t* rec;
ulint end_offset;
ulint ored_offset;
byte* data;
ulint len;
ulint i;
ut_ad(destination && dtuple);
ut_ad(dtuple_validate(dtuple));
ut_ad(dtuple_check_typed(dtuple));
ut_ad(dtuple_get_data_size(dtuple) == data_size);
n_fields = dtuple_get_n_fields(dtuple);
ut_ad(n_fields > 0);
/* Calculate the offset of the origin in the physical record */
rec = destination + rec_get_converted_extra_size(data_size, n_fields);
/* Store the number of fields */
rec_set_n_fields(rec, n_fields);
/* Set the info bits of the record */
rec_set_info_bits(rec, dtuple_get_info_bits(dtuple));
/* Store the data and the offsets */
end_offset = 0;
if (data_size <= REC_1BYTE_OFFS_LIMIT) {
rec_set_1byte_offs_flag(rec, TRUE);
for (i = 0; i < n_fields; i++) {
field = dtuple_get_nth_field(dtuple, i);
data = dfield_get_data(field);
len = dfield_get_len(field);
if (len == UNIV_SQL_NULL) {
len = dtype_get_sql_null_size(dfield_get_type(field));
data_write_sql_null(rec + end_offset, len);
end_offset += len;
ored_offset = end_offset | REC_1BYTE_SQL_NULL_MASK;
} else {
/* If the data is not SQL null, store it */
ut_memcpy(rec + end_offset, data, len);
end_offset += len;
ored_offset = end_offset;
}
rec_1_set_field_end_info(rec, i, ored_offset);
}
} else {
rec_set_1byte_offs_flag(rec, FALSE);
for (i = 0; i < n_fields; i++) {
field = dtuple_get_nth_field(dtuple, i);
data = dfield_get_data(field);
len = dfield_get_len(field);
if (len == UNIV_SQL_NULL) {
len = dtype_get_sql_null_size(dfield_get_type(field));
data_write_sql_null(rec + end_offset, len);
end_offset += len;
ored_offset = end_offset | REC_2BYTE_SQL_NULL_MASK;
} else {
/* If the data is not SQL null, store it */
ut_memcpy(rec + end_offset, data, len);
end_offset += len;
ored_offset = end_offset;
}
rec_2_set_field_end_info(rec, i, ored_offset);
}
}
ut_ad(rec_validate(rec));
return(rec);
}
/**************************************************************//**
Moves parts of long fields in entry to the big record vector so that
the size of tuple drops below the maximum record size allowed in the
database. Moves data only from those fields which are not necessary
to determine uniquely the insertion place of the tuple in the index.
@return own: created big record vector, NULL if we are not able to
shorten the entry enough, i.e., if there are too many fixed-length or
short fields in entry or the index is clustered */
UNIV_INTERN
big_rec_t*
dtuple_convert_big_rec(
/*===================*/
dict_index_t* index, /*!< in: index */
dtuple_t* entry, /*!< in/out: index entry */
ulint* n_ext) /*!< in/out: number of
externally stored columns */
{
mem_heap_t* heap;
big_rec_t* vector;
dfield_t* dfield;
dict_field_t* ifield;
ulint size;
ulint n_fields;
ulint local_len;
ulint local_prefix_len;
if (UNIV_UNLIKELY(!dict_index_is_clust(index))) {
return(NULL);
}
if (dict_table_get_format(index->table) < DICT_TF_FORMAT_ZIP) {
/* up to MySQL 5.1: store a 768-byte prefix locally */
local_len = BTR_EXTERN_FIELD_REF_SIZE + DICT_MAX_INDEX_COL_LEN;
} else {
/* new-format table: do not store any BLOB prefix locally */
local_len = BTR_EXTERN_FIELD_REF_SIZE;
}
ut_a(dtuple_check_typed_no_assert(entry));
size = rec_get_converted_size(index, entry, *n_ext);
if (UNIV_UNLIKELY(size > 1000000000)) {
fprintf(stderr,
"InnoDB: Warning: tuple size very big: %lu\n",
(ulong) size);
fputs("InnoDB: Tuple contents: ", stderr);
dtuple_print(stderr, entry);
putc('\n', stderr);
}
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
a variable-length field that yields the biggest savings when
stored externally */
n_fields = 0;
while (page_zip_rec_needs_ext(rec_get_converted_size(index, entry,
*n_ext),
dict_table_is_comp(index->table),
dict_index_get_n_fields(index),
dict_table_zip_size(index->table))) {
ulint i;
ulint longest = 0;
ulint longest_i = ULINT_MAX;
byte* data;
big_rec_field_t* b;
for (i = dict_index_get_n_unique_in_tree(index);
i < dtuple_get_n_fields(entry); i++) {
ulint savings;
dfield = dtuple_get_nth_field(entry, i);
ifield = dict_index_get_nth_field(index, i);
/* Skip fixed-length, NULL, externally stored,
or short columns */
if (ifield->fixed_len
|| dfield_is_null(dfield)
|| dfield_is_ext(dfield)
|| dfield_get_len(dfield) <= local_len
|| dfield_get_len(dfield)
<= BTR_EXTERN_FIELD_REF_SIZE * 2) {
goto skip_field;
}
savings = dfield_get_len(dfield) - local_len;
/* Check that there would be savings */
if (longest >= savings) {
goto skip_field;
}
longest_i = i;
longest = savings;
skip_field:
continue;
}
if (!longest) {
/* Cannot shorten more */
mem_heap_free(heap);
return(NULL);
}
/* Move data from field longest_i to big rec vector.
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);
ifield = dict_index_get_nth_field(index, longest_i);
local_prefix_len = local_len - BTR_EXTERN_FIELD_REF_SIZE;
b = &vector->fields[n_fields];
b->field_no = longest_i;
b->len = dfield_get_len(dfield) - local_prefix_len;
b->data = (char*) dfield_get_data(dfield) + local_prefix_len;
/* Allocate the locally stored part of the column. */
data = mem_heap_alloc(heap, local_len);
/* Copy the local prefix. */
memcpy(data, dfield_get_data(dfield), local_prefix_len);
/* Clear the extern field reference (BLOB pointer). */
memset(data + local_prefix_len, 0, BTR_EXTERN_FIELD_REF_SIZE);
#if 0
/* The following would fail the Valgrind checks in
page_cur_insert_rec_low() and page_cur_insert_rec_zip().
The BLOB pointers in the record will be initialized after
the record and the BLOBs have been written. */
UNIV_MEM_ALLOC(data + local_prefix_len,
BTR_EXTERN_FIELD_REF_SIZE);
#endif
dfield_set_data(dfield, data, local_len);
dfield_set_ext(dfield);
n_fields++;
(*n_ext)++;
ut_ad(n_fields < dtuple_get_n_fields(entry));
}
vector->n_fields = n_fields;
return(vector);
}
void
row_build_row_ref_in_tuple(
/*=======================*/
dtuple_t* ref, /* in/out: row reference built; see the
NOTE below! */
dict_index_t* index, /* in: index */
rec_t* rec) /* in: record in the index;
NOTE: the data fields in ref 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! */
{
dict_table_t* table;
dict_index_t* clust_index;
dfield_t* dfield;
byte* field;
ulint len;
ulint ref_len;
ulint pos;
ulint i;
ut_a(ref && index && rec);
table = index->table;
if (!table) {
fprintf(stderr, "InnoDB: table %s for index %s not found\n",
index->table_name, index->name);
ut_a(0);
}
clust_index = dict_table_get_first_index(table);
if (!clust_index) {
fprintf(stderr,
"InnoDB: clust index for table %s for index %s not found\n",
index->table_name, index->name);
ut_a(0);
}
ref_len = dict_index_get_n_unique(clust_index);
ut_ad(ref_len == dtuple_get_n_fields(ref));
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, pos, &len);
dfield_set_data(dfield, field, len);
}
ut_ad(dtuple_check_typed(ref));
}
/*****************************************************************//**
Finds out if an active transaction has inserted or modified a secondary
index record. NOTE: the kernel mutex is temporarily released in this
function!
@return NULL if committed, else the active transaction */
UNIV_INTERN
trx_t*
row_vers_impl_x_locked_off_kernel(
/*==============================*/
const rec_t* rec, /*!< in: record in a secondary index */
dict_index_t* index, /*!< in: the secondary index */
const ulint* offsets)/*!< in: rec_get_offsets(rec, index) */
{
dict_index_t* clust_index;
rec_t* clust_rec;
ulint* clust_offsets;
rec_t* version;
trx_id_t trx_id;
mem_heap_t* heap;
mem_heap_t* heap2;
dtuple_t* row;
dtuple_t* entry = NULL; /* assignment to eliminate compiler
warning */
trx_t* trx;
ulint rec_del;
#ifdef UNIV_DEBUG
ulint err;
#endif /* UNIV_DEBUG */
mtr_t mtr;
ulint comp;
ut_ad(mutex_own(&kernel_mutex));
#ifdef UNIV_SYNC_DEBUG
ut_ad(!rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */
mutex_exit(&kernel_mutex);
mtr_start(&mtr);
/* Search for the clustered index record: this is a time-consuming
operation: therefore we release the kernel mutex; also, the release
is required by the latching order convention. The latch on the
clustered index locks the top of the stack of versions. We also
reserve purge_latch to lock the bottom of the version stack. */
clust_rec = row_get_clust_rec(BTR_SEARCH_LEAF, rec, index,
&clust_index, &mtr);
if (!clust_rec) {
/* In a rare case it is possible that no clust rec is found
for a secondary index record: if in row0umod.c
row_undo_mod_remove_clust_low() we have already removed the
clust rec, while purge is still cleaning and removing
secondary index records associated with earlier versions of
the clustered index record. In that case there cannot be
any implicit lock on the secondary index record, because
an active transaction which has modified the secondary index
record has also modified the clustered index record. And in
a rollback we always undo the modifications to secondary index
records before the clustered index record. */
mutex_enter(&kernel_mutex);
mtr_commit(&mtr);
return(NULL);
}
heap = mem_heap_create(1024);
clust_offsets = rec_get_offsets(clust_rec, clust_index, NULL,
ULINT_UNDEFINED, &heap);
trx_id = row_get_rec_trx_id(clust_rec, clust_index, clust_offsets);
mtr_s_lock(&(purge_sys->latch), &mtr);
mutex_enter(&kernel_mutex);
trx = NULL;
if (!trx_is_active(trx_id)) {
/* The transaction that modified or inserted clust_rec is no
longer active: no implicit lock on rec */
goto exit_func;
}
if (!lock_check_trx_id_sanity(trx_id, clust_rec, clust_index,
clust_offsets, TRUE)) {
/* Corruption noticed: try to avoid a crash by returning */
goto exit_func;
}
comp = page_rec_is_comp(rec);
ut_ad(index->table == clust_index->table);
ut_ad(!!comp == dict_table_is_comp(index->table));
ut_ad(!comp == !page_rec_is_comp(clust_rec));
/* We look up if some earlier version, which was modified by the trx_id
transaction, of the clustered index record would require rec to be in
a different state (delete marked or unmarked, or have different field
values, or not existing). If there is such a version, then rec was
modified by the trx_id transaction, and it has an implicit x-lock on
rec. Note that if clust_rec itself would require rec to be in a
different state, then the trx_id transaction has not yet had time to
modify rec, and does not necessarily have an implicit x-lock on rec. */
rec_del = rec_get_deleted_flag(rec, comp);
trx = NULL;
version = clust_rec;
for (;;) {
rec_t* prev_version;
ulint vers_del;
row_ext_t* ext;
trx_id_t prev_trx_id;
mutex_exit(&kernel_mutex);
/* While we retrieve an earlier version of clust_rec, we
release the kernel mutex, because it may take time to access
the disk. After the release, we have to check if the trx_id
transaction is still active. We keep the semaphore in mtr on
the clust_rec page, so that no other transaction can update
it and get an implicit x-lock on rec. */
heap2 = heap;
heap = mem_heap_create(1024);
#ifdef UNIV_DEBUG
err =
#endif /* UNIV_DEBUG */
trx_undo_prev_version_build(clust_rec, &mtr, version,
clust_index, clust_offsets,
heap, &prev_version);
mem_heap_free(heap2); /* free version and clust_offsets */
if (prev_version == NULL) {
mutex_enter(&kernel_mutex);
if (!trx_is_active(trx_id)) {
/* Transaction no longer active: no
implicit x-lock */
break;
}
/* If the transaction is still active,
clust_rec must be a fresh insert, because no
previous version was found. */
ut_ad(err == DB_SUCCESS);
/* It was a freshly inserted version: there is an
implicit x-lock on rec */
trx = trx_get_on_id(trx_id);
break;
}
clust_offsets = rec_get_offsets(prev_version, clust_index,
NULL, ULINT_UNDEFINED, &heap);
vers_del = rec_get_deleted_flag(prev_version, comp);
prev_trx_id = row_get_rec_trx_id(prev_version, clust_index,
clust_offsets);
/* The stack of versions is locked by mtr. Thus, it
is safe to fetch the prefixes for externally stored
columns. */
row = row_build(ROW_COPY_POINTERS, clust_index, prev_version,
clust_offsets, NULL, &ext, heap);
entry = row_build_index_entry(row, ext, index, heap);
/* entry may be NULL if a record was inserted in place
of a deleted record, and the BLOB pointers of the new
record were not initialized yet. But in that case,
prev_version should be NULL. */
ut_a(entry);
mutex_enter(&kernel_mutex);
if (!trx_is_active(trx_id)) {
/* Transaction no longer active: no implicit x-lock */
break;
}
/* If we get here, we know that the trx_id transaction is
still active and it has modified prev_version. Let us check
if prev_version would require rec to be in a different
state. */
/* The previous version of clust_rec must be
accessible, because the transaction is still active
and clust_rec was not a fresh insert. */
ut_ad(err == DB_SUCCESS);
/* We check if entry and rec are identified in the alphabetical
ordering */
if (0 == cmp_dtuple_rec(entry, rec, offsets)) {
/* The delete marks of rec and prev_version should be
equal for rec to be in the state required by
prev_version */
if (rec_del != vers_del) {
trx = trx_get_on_id(trx_id);
break;
}
/* It is possible that the row was updated so that the
secondary index record remained the same in
alphabetical ordering, but the field values changed
still. For example, 'abc' -> 'ABC'. Check also that. */
dtuple_set_types_binary(entry,
dtuple_get_n_fields(entry));
if (0 != cmp_dtuple_rec(entry, rec, offsets)) {
trx = trx_get_on_id(trx_id);
break;
}
} else if (!rec_del) {
/* The delete mark should be set in rec for it to be
in the state required by prev_version */
trx = trx_get_on_id(trx_id);
break;
}
if (0 != ut_dulint_cmp(trx_id, prev_trx_id)) {
/* The versions modified by the trx_id transaction end
to prev_version: no implicit x-lock */
break;
}
version = prev_version;
}/* for (;;) */
exit_func:
mtr_commit(&mtr);
mem_heap_free(heap);
return(trx);
}
upd_t*
row_upd_build_difference_binary(
/*============================*/
/* out, own: update vector of differing
fields, excluding roll ptr and trx id */
dict_index_t* index, /* in: clustered index */
dtuple_t* entry, /* in: entry to insert */
ulint* ext_vec,/* in: array containing field numbers of
externally stored fields in entry, or NULL */
ulint n_ext_vec,/* in: number of fields in ext_vec */
rec_t* rec, /* in: clustered index record */
mem_heap_t* heap) /* in: memory heap from which allocated */
{
upd_field_t* upd_field;
dfield_t* dfield;
byte* data;
ulint len;
upd_t* update;
ulint n_diff;
ulint roll_ptr_pos;
ulint trx_id_pos;
ulint i;
/* This function is used only for a clustered index */
ut_a(index->type & DICT_CLUSTERED);
update = upd_create(dtuple_get_n_fields(entry), heap);
n_diff = 0;
roll_ptr_pos = dict_index_get_sys_col_pos(index, DATA_ROLL_PTR);
trx_id_pos = dict_index_get_sys_col_pos(index, DATA_TRX_ID);
for (i = 0; i < dtuple_get_n_fields(entry); i++) {
data = rec_get_nth_field(rec, i, &len);
dfield = dtuple_get_nth_field(entry, i);
/* NOTE: we compare the fields as binary strings!
(No collation) */
if (i == trx_id_pos || i == roll_ptr_pos) {
goto skip_compare;
}
if (rec_get_nth_field_extern_bit(rec, i)
!= upd_ext_vec_contains(ext_vec, n_ext_vec, i)
|| !dfield_data_is_binary_equal(dfield, len, data)) {
upd_field = upd_get_nth_field(update, n_diff);
dfield_copy(&(upd_field->new_val), dfield);
upd_field_set_field_no(upd_field, i, index);
if (upd_ext_vec_contains(ext_vec, n_ext_vec, i)) {
upd_field->extern_storage = TRUE;
} else {
upd_field->extern_storage = FALSE;
}
n_diff++;
}
skip_compare:
;
}
update->n_fields = n_diff;
return(update);
}