/*******************************************************************//**
Converts an index record to a typed data tuple.
@return index entry built; does not set info_bits, and the data fields
in the entry will point directly to rec */
UNIV_INTERN
dtuple_t*
row_rec_to_index_entry_low(
/*=======================*/
const rec_t* rec, /*!< in: record in the index */
const dict_index_t* index, /*!< in: index */
const ulint* offsets,/*!< in: rec_get_offsets(rec, index) */
ulint* n_ext, /*!< out: number of externally
stored columns */
mem_heap_t* heap) /*!< in: memory heap from which
the memory needed is allocated */
{
dtuple_t* entry;
dfield_t* dfield;
ulint i;
const byte* field;
ulint len;
ulint rec_len;
ut_ad(rec && heap && index);
/* Because this function may be invoked by row0merge.c
on a record whose header is in different format, the check
rec_offs_validate(rec, index, offsets) must be avoided here. */
ut_ad(n_ext);
*n_ext = 0;
rec_len = rec_offs_n_fields(offsets);
entry = dtuple_create(heap, rec_len);
dtuple_set_n_fields_cmp(entry,
dict_index_get_n_unique_in_tree(index));
ut_ad(rec_len == dict_index_get_n_fields(index));
dict_index_copy_types(entry, index, rec_len);
for (i = 0; i < rec_len; i++) {
dfield = dtuple_get_nth_field(entry, i);
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);
(*n_ext)++;
}
}
ut_ad(dtuple_check_typed(entry));
return(entry);
}
/*******************************************************************//**
Builds an update vector based on a remaining part of an undo log record.
@return remaining part of the record, NULL if an error detected, which
means that the record is corrupted */
UNIV_INTERN
byte*
trx_undo_update_rec_get_update(
/*===========================*/
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 */
trx_id_t trx_id, /*!< in: transaction id from this undo record */
roll_ptr_t 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;
ulint i;
ut_a(dict_index_is_clust(index));
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++) {
byte* field;
ulint len;
ulint field_no;
ulint orig_len;
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, TRUE, index->table_name);
fprintf(stderr, "\n"
"InnoDB: n_fields = %lu, i = %lu, ptr %p\n",
(ulong) n_fields, (ulong) i, ptr);
*upd = NULL;
return(NULL);
}
upd_field = upd_get_nth_field(update, i);
upd_field_set_field_no(upd_field, field_no, index, trx);
ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len);
upd_field->orig_len = orig_len;
if (len == UNIV_SQL_NULL) {
dfield_set_null(&upd_field->new_val);
} else if (len < UNIV_EXTERN_STORAGE_FIELD) {
dfield_set_data(&upd_field->new_val, field, len);
} else {
len -= UNIV_EXTERN_STORAGE_FIELD;
dfield_set_data(&upd_field->new_val, field, len);
dfield_set_ext(&upd_field->new_val);
}
}
*upd = update;
return(ptr);
}
/*******************************************************************//**
Builds a partial row from an update undo log record. It contains the
columns which occur as ordering in any index of the table.
@return pointer to remaining part of undo record */
UNIV_INTERN
byte*
trx_undo_rec_get_partial_row(
/*=========================*/
byte* ptr, /*!< in: remaining part in update undo log
record of a suitable type, at the start of
the stored index columns;
NOTE that this copy of the undo log record must
be preserved as long as the partial row is
used, as we do NOT copy the data in the
record! */
dict_index_t* index, /*!< in: clustered index */
dtuple_t** row, /*!< out, own: partial row */
ibool ignore_prefix, /*!< in: flag to indicate if we
expect blob prefixes in undo. Used
only in the assertion. */
mem_heap_t* heap) /*!< in: memory heap from which the memory
needed is allocated */
{
const byte* end_ptr;
ulint row_len;
ut_ad(index);
ut_ad(ptr);
ut_ad(row);
ut_ad(heap);
ut_ad(dict_index_is_clust(index));
row_len = dict_table_get_n_cols(index->table);
*row = dtuple_create(heap, row_len);
dict_table_copy_types(*row, index->table);
end_ptr = ptr + mach_read_from_2(ptr);
ptr += 2;
while (ptr != end_ptr) {
dfield_t* dfield;
byte* field;
ulint field_no;
const dict_col_t* col;
ulint col_no;
ulint len;
ulint orig_len;
ptr = trx_undo_update_rec_get_field_no(ptr, &field_no);
col = dict_index_get_nth_col(index, field_no);
col_no = dict_col_get_no(col);
ptr = trx_undo_rec_get_col_val(ptr, &field, &len, &orig_len);
dfield = dtuple_get_nth_field(*row, col_no);
dfield_set_data(dfield, field, len);
if (len != UNIV_SQL_NULL
&& len >= UNIV_EXTERN_STORAGE_FIELD) {
dfield_set_len(dfield,
len - UNIV_EXTERN_STORAGE_FIELD);
dfield_set_ext(dfield);
/* If the prefix of this column is indexed,
ensure that enough prefix is stored in the
undo log record. */
if (!ignore_prefix && col->ord_part) {
ut_a(dfield_get_len(dfield)
>= 2 * BTR_EXTERN_FIELD_REF_SIZE);
ut_a(dict_table_get_format(index->table)
>= DICT_TF_FORMAT_ZIP
|| dfield_get_len(dfield)
>= REC_MAX_INDEX_COL_LEN
+ BTR_EXTERN_FIELD_REF_SIZE);
}
}
}
return(ptr);
}
/**************************************************************//**
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);
}
/*******************************************************************//**
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);
}