用于EagleEye3.0 规则集漏报和误报测试的示例项目,项目收集于github和gitee
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

915 lines
34 KiB

5 months ago
/* Copyright (c) 2000, 2019, Oracle and/or its affiliates. All rights reserved.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License, version 2.0,
as published by the Free Software Foundation.
This program is also distributed with certain software (including
but not limited to OpenSSL) that is licensed under separate terms,
as designated in a particular file or component or in included license
documentation. The authors of MySQL hereby grant you an additional
permission to link the program and your derivative works with the
separately licensed software that they have included with MySQL.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License, version 2.0, for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */
/**
@file sql/sql_rename.cc
Atomic rename of table; RENAME TABLE t1 to t2, tmp to t1 [,...]
*/
#include "sql/sql_rename.h"
#include <string.h>
#include <set>
#include "my_dbug.h"
#include "my_inttypes.h"
#include "my_murmur3.h"
#include "my_sys.h"
#include "mysql/components/services/log_shared.h"
#include "mysqld_error.h"
#include "sql/dd/cache/dictionary_client.h" // dd::cache::Dictionary_client
#include "sql/dd/dd_table.h" // dd::table_storage_engine
#include "sql/dd/properties.h" // dd::Properties
#include "sql/dd/types/abstract_table.h" // dd::Abstract_table
#include "sql/dd/types/table.h" // dd::Table
#include "sql/dd_sql_view.h" // View_metadata_updater
#include "sql/derror.h" // ER_THD
#include "sql/handler.h"
#include "sql/log.h" // query_logger
#include "sql/mysqld.h" // lower_case_table_names
#include "sql/sp_cache.h" // sp_cache_invalidate
#include "sql/sql_base.h" // tdc_remove_table,
// lock_table_names,
#include "sql/sql_class.h" // THD
#include "sql/sql_handler.h" // mysql_ha_rm_tables
#include "sql/sql_table.h" // write_bin_log,
// build_table_filename
#include "sql/sql_trigger.h" // change_trigger_table_name
#include "sql/system_variables.h"
#include "sql/table.h"
#include "sql/thd_raii.h"
#include "sql/transaction.h" // trans_commit_stmt
namespace dd {
class Schema;
} // namespace dd
typedef std::set<handlerton *> post_ddl_htons_t;
static TABLE_LIST *rename_tables(
THD *thd, TABLE_LIST *table_list, bool *int_commit_done,
post_ddl_htons_t *post_ddl_htons,
Foreign_key_parents_invalidator *fk_invalidator);
static TABLE_LIST *reverse_table_list(TABLE_LIST *table_list);
namespace {
struct table_list_hash {
size_t operator()(const TABLE_LIST *table) const {
return static_cast<size_t>(murmur3_32(table->mdl_request.key.ptr(),
table->mdl_request.key.length(), 0));
}
};
struct table_list_equal {
bool operator()(const TABLE_LIST *a, const TABLE_LIST *b) const {
return a->mdl_request.key.is_equal(&b->mdl_request.key);
}
};
} // namespace
/**
Check if connection owns SNRW metadata lock on table or view.
Report apropriate error if not.
@note Unlike find_table_for_mdl_upgrade() this call can handle views.
*/
static bool check_if_owns_upgradable_mdl(THD *thd, const char *db,
const char *table_name) {
if (thd->mdl_context.owns_equal_or_stronger_lock(
MDL_key::TABLE, db, table_name, MDL_SHARED_NO_READ_WRITE))
return false; // Success.
if (thd->mdl_context.owns_equal_or_stronger_lock(
MDL_key::TABLE, db, table_name, MDL_SHARED_READ_ONLY))
my_error(ER_TABLE_NOT_LOCKED_FOR_WRITE, MYF(0), table_name);
else
my_error(ER_TABLE_NOT_LOCKED, MYF(0), table_name);
return true; // Failure.
}
/**
Find metadata lock request for table's schema in the set of schema
requests and set duration of corresponding lock to explicit.
@note We assume that there are no duplicate schemata in schema_reqs
array.
*/
static void find_and_set_explicit_duration_for_schema_mdl(
THD *thd, TABLE_LIST *table,
Prealloced_array<MDL_request *, 1> *schema_reqs) {
auto same_db = [table](const MDL_request *mdl_request) {
return table->db_length == mdl_request->key.db_name_length() &&
memcmp(table->db, mdl_request->key.db_name(), table->db_length) == 0;
};
auto sch_req =
std::find_if(schema_reqs->begin(), schema_reqs->end(), same_db);
if (sch_req != schema_reqs->end()) {
thd->mdl_context.set_lock_duration((*sch_req)->ticket, MDL_EXPLICIT);
/*
Remove found request by replacing it with the last one.
This is necessary to avoid setting duration for the same schema ticket
twice and also to speed up further calls to this function.
*/
*sch_req = schema_reqs->back();
schema_reqs->pop_back();
} else {
// We must have handled this schema already.
}
}
/**
Rename tables from the list.
@param thd Thread context.
@param table_list Every two entries in the table_list form
a pair of original name and the new name.
@return True - on failure, false - on success.
*/
bool mysql_rename_tables(THD *thd, TABLE_LIST *table_list) {
TABLE_LIST *ren_table = 0;
DBUG_TRACE;
mysql_ha_rm_tables(thd, table_list);
/*
The below Auto_releaser allows to keep uncommitted versions of data-
dictionary objects cached in the Dictionary_client for the whole duration
of the statement.
*/
dd::cache::Dictionary_client::Auto_releaser releaser(thd->dd_client());
if (query_logger.is_log_table_enabled(QUERY_LOG_GENERAL) ||
query_logger.is_log_table_enabled(QUERY_LOG_SLOW)) {
int to_table;
const char *rename_log_table[2] = {NULL, NULL};
/*
Rules for rename of a log table:
IF 1. Log tables are enabled
AND 2. Rename operates on the log table and nothing is being
renamed to the log table.
DO 3. Throw an error message.
ELSE 4. Perform rename.
*/
for (to_table = 0, ren_table = table_list; ren_table;
to_table = 1 - to_table, ren_table = ren_table->next_local) {
int log_table_rename = 0;
if ((log_table_rename =
query_logger.check_if_log_table(ren_table, true))) {
/*
as we use log_table_rename as an array index, we need it to start
with 0, while QUERY_LOG_SLOW == 1 and QUERY_LOG_GENERAL == 2.
So, we shift the value to start with 0;
*/
log_table_rename--;
if (rename_log_table[log_table_rename]) {
if (to_table)
rename_log_table[log_table_rename] = NULL;
else {
/*
Two renames of "log_table TO" w/o rename "TO log_table" in
between.
*/
my_error(ER_CANT_RENAME_LOG_TABLE, MYF(0), ren_table->table_name,
ren_table->table_name);
return true;
}
} else {
if (to_table) {
/*
Attempt to rename a table TO log_table w/o renaming
log_table TO some table.
*/
my_error(ER_CANT_RENAME_LOG_TABLE, MYF(0), ren_table->table_name,
ren_table->table_name);
return true;
} else {
/* save the name of the log table to report an error */
rename_log_table[log_table_rename] = ren_table->table_name;
}
}
}
}
if (rename_log_table[0] || rename_log_table[1]) {
if (rename_log_table[0])
my_error(ER_CANT_RENAME_LOG_TABLE, MYF(0), rename_log_table[0],
rename_log_table[0]);
else
my_error(ER_CANT_RENAME_LOG_TABLE, MYF(0), rename_log_table[1],
rename_log_table[1]);
return true;
}
}
/*
Array in which pointers to MDL requests for acquired schema locks are
stored. Each schema can be present in this array only once.
*/
Prealloced_array<MDL_request *, 1> schema_reqs(PSI_INSTRUMENT_ME);
if (thd->locked_tables_mode) {
/*
LOCK TABLES case.
Check that tables to be renamed are locked for WRITE. Take into
account that name of table to be renamed might be result of some
previous step in multi-step RENAME TABLES.
In theory, we could disregard whether they locked or not and just try
to acquire exclusive metadata locks on them, but this is too deadlock
prone.
Most probably, there is no tables which correspond to target table
names, so similar check doesn't make sense for them.
In theory, we can reduce chance of MDL deadlocks by also checking at
this stage that all child and parent tables for FKs in which tables
to be renamed participate are locked for WRITE (as we will have to
acquire to exclusive MDLs on these tables later).
But this is, probably, too severe restriction which will make
RENAMES TABLES under LOCK TABLES hard to use in 3rd-party online
ALTER TABLE tools.
*/
malloc_unordered_set<TABLE_LIST *, table_list_hash, table_list_equal>
new_names(PSI_INSTRUMENT_ME);
TABLE_LIST *new_table;
for (ren_table = table_list; ren_table; ren_table = new_table->next_local) {
new_table = ren_table->next_local;
auto new_name_it = new_names.find(ren_table);
if (new_name_it == new_names.end()) {
if (check_if_owns_upgradable_mdl(thd, ren_table->db,
ren_table->table_name))
return true;
} else {
new_names.erase(new_name_it);
}
new_names.insert(new_table);
}
/*
Now proceed to acquiring exclusive metadata locks on both source and
target table names as well as necessary schema, global and backup locks.
Since we already have SNRW locks on source table names, we, in fact, are
upgrading locks for them.
*/
}
if (lock_table_names(thd, table_list, 0, thd->variables.lock_wait_timeout, 0,
&schema_reqs) ||
lock_trigger_names(thd, table_list))
return true;
const dd::Table *table_def = nullptr;
TABLE_LIST *table;
for (table = table_list; table && table->next_local;
table = table->next_local) {
if (thd->dd_client()->acquire(table->db, table->table_name, &table_def)) {
return true;
}
if (table_def && table_def->hidden() == dd::Abstract_table::HT_HIDDEN_SE) {
my_error(ER_NO_SUCH_TABLE, MYF(0), table->db, table->table_name);
return true;
}
}
for (ren_table = table_list; ren_table; ren_table = ren_table->next_local) {
if (thd->locked_tables_mode)
close_all_tables_for_name(thd, ren_table->db, ren_table->table_name,
false);
else
tdc_remove_table(thd, TDC_RT_REMOVE_ALL, ren_table->db,
ren_table->table_name, false);
}
bool error = false;
bool int_commit_done = false;
/*
Indicates whether we managed fully revert non-atomic RENAME TABLES
after the failure.
*/
bool int_commit_full_revert = false;
std::set<handlerton *> post_ddl_htons;
Foreign_key_parents_invalidator fk_invalidator;
/*
An exclusive lock on table names is satisfactory to ensure
no other thread accesses this table.
*/
if ((ren_table = rename_tables(thd, table_list, &int_commit_done,
&post_ddl_htons, &fk_invalidator))) {
/* Rename didn't succeed; rename back the tables in reverse order */
TABLE_LIST *table;
if (int_commit_done) {
/* Reverse the table list */
table_list = reverse_table_list(table_list);
/* Find the last renamed table */
for (table = table_list; table->next_local != ren_table;
table = table->next_local->next_local)
;
table = table->next_local->next_local; // Skip error table
/*
Revert to old names. In 5.7 we have ignored most of errors occurring
in the process. However, this looks like a risky idea -- by ignoring
errors we are likely to end up in some awkward state and not going to
restore status quo ante.
So starting from 8.0 we chose to abort reversal on the first failure.
We will still end up in some awkward case in this case but at least
no additional damage will be done. Note that since InnoDB tables are
new default and this engine supports atomic DDL, non-atomic RENAME
TABLES, which this code deals with, is not the main use case anyway.
*/
int_commit_full_revert = !rename_tables(thd, table, &int_commit_done,
&post_ddl_htons, &fk_invalidator);
/* Revert the table list (for prepared statements) */
table_list = reverse_table_list(table_list);
}
error = true;
}
if (!error) {
error = write_bin_log(thd, true, thd->query().str, thd->query().length,
!int_commit_done);
}
if (!error) {
Uncommitted_tables_guard uncommitted_tables(thd);
for (ren_table = table_list; ren_table;
ren_table = ren_table->next_local->next_local) {
TABLE_LIST *new_table = ren_table->next_local;
DBUG_ASSERT(new_table);
uncommitted_tables.add_table(ren_table);
uncommitted_tables.add_table(new_table);
if ((error = update_referencing_views_metadata(
thd, ren_table, new_table->db, new_table->table_name,
int_commit_done, &uncommitted_tables)))
break;
}
}
if (!error && !int_commit_done) {
error = (trans_commit_stmt(thd) || trans_commit_implicit(thd));
if (!error) {
/*
Don't try to invalidate foreign key parents on error,
as we might miss necessary locks on them.
*/
fk_invalidator.invalidate(thd);
}
}
if (error) {
trans_rollback_stmt(thd);
/*
Full rollback in case we have THD::transaction_rollback_request
and to synchronize DD state in cache and on disk (as statement
rollback doesn't clear DD cache of modified uncommitted objects).
*/
trans_rollback(thd);
}
for (handlerton *hton : post_ddl_htons) hton->post_ddl(thd);
if (thd->locked_tables_mode) {
if (!error) {
/*
Adjust locked tables list and reopen tables under new names.
Also calculate sets of metadata locks to release (on old table
names) and to keep until UNLOCK TABLES (on new table names).
In addition to keeping locks on tables we also do the same for
schemas in order to keep set of metadata locks consistent with
one acquired by LOCK TABLES. We don't release locks on old table
schemas as it is non-trivial to figure out which locks can be
released.
Tablespaces do not need special handling though, as metadata locks
on them are acquired at LOCK TABLES time and are unaffected by
RENAME TABLES.
*/
malloc_unordered_set<TABLE_LIST *, table_list_hash, table_list_equal>
to_release(PSI_INSTRUMENT_ME), to_keep(PSI_INSTRUMENT_ME);
TABLE_LIST *new_table;
for (ren_table = table_list; ren_table;
ren_table = new_table->next_local) {
new_table = ren_table->next_local;
thd->locked_tables_list.rename_locked_table(
ren_table, new_table->db, new_table->table_name,
new_table->mdl_request.ticket);
to_release.insert(ren_table);
to_keep.erase(ren_table);
to_keep.insert(new_table);
to_release.erase(new_table);
}
error = thd->locked_tables_list.reopen_tables(thd);
for (TABLE_LIST *t : to_release) {
// Also releases locks with EXPLICIT duration for the same name.
thd->mdl_context.release_all_locks_for_name(t->mdl_request.ticket);
}
for (TABLE_LIST *t : to_keep) {
thd->mdl_context.set_lock_duration(t->mdl_request.ticket, MDL_EXPLICIT);
t->mdl_request.ticket->downgrade_lock(MDL_SHARED_NO_READ_WRITE);
find_and_set_explicit_duration_for_schema_mdl(thd, t, &schema_reqs);
}
} else if (!int_commit_done || int_commit_full_revert) {
/*
Error happened but all (actually not quite all, see below) changes
were reverted. We just need to reopen tables.
Since changes were reverted no additional metadata locks need to
be kept after statement end. All additional locks acquired by
this statement will be released automatically at its end, since
they have transactional duration.
In case of non-atomic RENAME TABLE previously orphan foreign keys
which got new parents will keep these parents after reversal, but
this is not important in this context.
*/
thd->locked_tables_list.reopen_tables(thd);
} else {
/*
Error happened and we failed to revert all changes. We simply close
all tables involved.
*/
thd->locked_tables_list.unlink_all_closed_tables(thd, NULL, 0);
/*
We need to keep metadata locks on both old and new table names
to avoid breaking foreign key invariants for LOCK TABLES.
So we set duration of locks on new names to explicit and downgrade
them from X to SNRW metadata locks. Also keep locks for new schemas.
Prune list of duplicates first as setting explicit duration for the
same MDL ticket twice is disallowed.
*/
malloc_unordered_set<TABLE_LIST *, table_list_hash, table_list_equal>
to_keep(PSI_INSTRUMENT_ME);
TABLE_LIST *new_table;
for (ren_table = table_list; ren_table;
ren_table = new_table->next_local) {
new_table = ren_table->next_local;
to_keep.insert(new_table);
}
for (TABLE_LIST *t : to_keep) {
thd->mdl_context.set_lock_duration(t->mdl_request.ticket, MDL_EXPLICIT);
t->mdl_request.ticket->downgrade_lock(MDL_SHARED_NO_READ_WRITE);
find_and_set_explicit_duration_for_schema_mdl(thd, t, &schema_reqs);
}
}
}
if (!error) my_ok(thd);
return error;
}
/*
reverse table list
SYNOPSIS
reverse_table_list()
table_list pointer to table _list
RETURN
pointer to new (reversed) list
*/
static TABLE_LIST *reverse_table_list(TABLE_LIST *table_list) {
TABLE_LIST *prev = 0;
while (table_list) {
TABLE_LIST *next = table_list->next_local;
table_list->next_local = prev;
prev = table_list;
table_list = next;
}
return (prev);
}
/**
Rename a single table or a view.
@param[in] thd Thread handle.
@param[in] ren_table A table/view to be renamed.
@param[in] new_db The database to which the
table to be moved to.
@param[in] new_table_name The new table/view name.
@param[in] new_table_alias The new table/view alias.
@param[in,out] int_commit_done Whether intermediate commits
were done.
@param[in,out] post_ddl_htons Set of SEs supporting atomic DDL
for which post-DDL hooks needs
to be called.
@param[in,out] fk_invalidator Object keeping track of which
dd::Table objects to invalidate.
@note Unless int_commit_done is true failure of this call requires
rollback of transaction before doing anything else.
@return False on success, True if rename failed.
*/
static bool do_rename(THD *thd, TABLE_LIST *ren_table, const char *new_db,
const char *new_table_name, const char *new_table_alias,
bool *int_commit_done,
std::set<handlerton *> *post_ddl_htons,
Foreign_key_parents_invalidator *fk_invalidator) {
const char *new_alias = new_table_name;
const char *old_alias = ren_table->table_name;
DBUG_TRACE;
if (lower_case_table_names == 2) {
old_alias = ren_table->alias;
new_alias = new_table_alias;
}
DBUG_ASSERT(new_alias);
// Fail if the target table already exists
dd::cache::Dictionary_client::Auto_releaser releaser(thd->dd_client());
const dd::Schema *from_schema = nullptr;
const dd::Schema *to_schema = nullptr;
dd::Abstract_table *from_at = nullptr;
const dd::Abstract_table *to_table = nullptr;
if (thd->dd_client()->acquire(ren_table->db, &from_schema) ||
thd->dd_client()->acquire(new_db, &to_schema) ||
thd->dd_client()->acquire(new_db, new_alias, &to_table) ||
thd->dd_client()->acquire_for_modification(
ren_table->db, ren_table->table_name, &from_at))
return true;
if (to_table != nullptr) {
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), new_alias);
return true;
}
if (from_schema == nullptr) {
my_error(ER_BAD_DB_ERROR, MYF(0), ren_table->db);
return true;
}
if (to_schema == nullptr) {
my_error(ER_BAD_DB_ERROR, MYF(0), new_db);
return true;
}
if (from_at == nullptr) {
my_error(ER_NO_SUCH_TABLE, MYF(0), ren_table->db, old_alias);
return true;
}
// So here we know the source table exists and the target table does
// not exist. Next is to act based on the table type.
switch (from_at->type()) {
case dd::enum_table_type::BASE_TABLE: {
handlerton *hton = NULL;
dd::Table *from_table = dynamic_cast<dd::Table *>(from_at);
// If the engine is not found, my_error() has already been called
if (dd::table_storage_engine(thd, from_table, &hton)) return true;
if ((hton->flags & HTON_SUPPORTS_ATOMIC_DDL) && (hton->post_ddl))
post_ddl_htons->insert(hton);
if (check_table_triggers_are_not_in_the_same_schema(ren_table->db,
*from_table, new_db))
return true;
// The below code assumes that only SE capable of atomic DDL support FK.
DBUG_ASSERT(!(hton->flags & HTON_SUPPORTS_FOREIGN_KEYS) ||
(hton->flags & HTON_SUPPORTS_ATOMIC_DDL));
/*
If we are performing rename with intermediate commits then
invalidation of foreign key parents should have happened
already, right after commit. Code below dealing with failure to
acquire locks on parent and child tables relies on this
invariant.
*/
DBUG_ASSERT(!(*int_commit_done) || fk_invalidator->is_empty());
// Find if table uses general tablespace and is it encrypted.
bool is_general_tablespace = false;
bool is_table_encrypted = false;
dd::Encrypt_result new_er =
dd::is_tablespace_encrypted(thd, *from_table, &is_general_tablespace);
if (new_er.error) {
return true;
}
is_table_encrypted = new_er.value;
if (!is_general_tablespace &&
from_table->options().exists("encrypt_type")) {
dd::String_type et;
(void)from_table->options().get("encrypt_type", &et);
DBUG_ASSERT(et.empty() == false);
is_table_encrypted = is_encrypted(et);
}
/*
Check if we are allowed to move the table, if destination schema
is changed and its default encryption differs from tables
encryption type.
*/
if (from_schema->id() != to_schema->id() &&
to_schema->default_encryption() != is_table_encrypted) {
if (opt_table_encryption_privilege_check) {
if (check_table_encryption_admin_access(thd)) {
my_error(ER_CANNOT_SET_TABLE_ENCRYPTION, MYF(0));
return true;
}
} else if (to_schema->default_encryption() && !is_table_encrypted) {
push_warning(thd, Sql_condition::SL_WARNING,
WARN_UNENCRYPTED_TABLE_IN_ENCRYPTED_DB,
ER_THD(thd, WARN_UNENCRYPTED_TABLE_IN_ENCRYPTED_DB));
}
}
/*
Obtain exclusive metadata lock on all tables being referenced by the
old table, since these tables must be invalidated to force a cache miss
on next acquisition, in order to refresh their FK information.
Also lock all tables referencing the old table. The FK information in
these tables must be updated to refer to the new table name.
And also lock all tables referencing the new table. The FK information
in these tables must be updated to refer to the (possibly) new
unique index name.
TODO: Long-term we should consider acquiring these locks in
mysql_rename_tables() together with locks on other tables.
This should decrease probability of deadlock and improve
crash-safety for RENAME TABLES which mix InnoDB and non-InnoDB
tables (as all waiting will happen before any changes to SEs).
*/
if (hton->flags & HTON_SUPPORTS_FOREIGN_KEYS) {
/*
If we are under LOCK TABLES check that all previously orphan tables
which reference new table name through foreign keys are locked for
write. Otherwise this ALTER will leave after itself parent table
locked for WRITE without child tables locked for WRITE. This will
break FK LOCK TABLES invariants if some of previously orphan FKs
have referential actions which update child table.
Note that doing this check at earlier phase is possible but seems
to be tricky since determining orphans can be non-trivial task in
case of multi-step RENAME TABLES statement.
*/
if (thd->locked_tables_mode == LTM_LOCK_TABLES ||
thd->locked_tables_mode == LTM_PRELOCKED_UNDER_LOCK_TABLES) {
MDL_request_list orphans_mdl_requests;
if (collect_fk_children(thd, new_db, new_alias, hton, MDL_EXCLUSIVE,
&orphans_mdl_requests)) {
// See explanation for clearing foreign key invalidator below.
fk_invalidator->clear();
return true;
}
MDL_request_list::Iterator it(orphans_mdl_requests);
MDL_request *mdl_request;
while ((mdl_request = it++) != nullptr) {
if (mdl_request->key.mdl_namespace() != MDL_key::TABLE) continue;
if (check_if_owns_upgradable_mdl(thd, mdl_request->key.db_name(),
mdl_request->key.name())) {
// See explanation for clearing foreign key invalidator below.
fk_invalidator->clear();
return true;
}
}
}
if (collect_and_lock_fk_tables_for_rename_table(
thd, ren_table->db, old_alias, from_table, new_db, new_alias,
hton, fk_invalidator)) {
/*
If we are performing RENAME TABLES with intermediate commits
FK invalidator was empty before the above call. So at this
point it only contains entries on which we might miss locks.
We need to clear invalidator before starting process of
reverse renames.
If we are performing RENAME TABLES without intermediate commits
the whole statement will be rolled back and invalidation won't
happen. So it is safe to clear invalidator.
*/
fk_invalidator->clear();
return true;
}
}
if (lock_check_constraint_names_for_rename(thd, ren_table->db, old_alias,
from_table, new_db, new_alias))
return true;
/*
We commit changes to data-dictionary immediately after renaming
table in storage engine if SE doesn't support atomic DDL or
there were intermediate commits already. In the latter case
the whole statement is not crash-safe anyway and clean-up is
simpler this way.
The FKs of the renamed table must be changed to reflect the new table.
The tables referencing the old and new table names must have their FK
information updated to reflec the correct table- and unique index name.
The parents of the old FKs must be invalidated to make sure they
update the cached FK parent information upon next acquisition.
If renaming fails, my_error() has already been called
*/
if (mysql_rename_table(
thd, hton, ren_table->db, old_alias, ren_table->db, old_alias,
*to_schema, new_db, new_alias,
((hton->flags & HTON_SUPPORTS_ATOMIC_DDL) ? NO_DD_COMMIT : 0)) ||
((hton->flags & HTON_SUPPORTS_FOREIGN_KEYS) &&
adjust_fks_for_rename_table(thd, ren_table->db, old_alias, new_db,
new_alias, hton))) {
/*
If RENAME TABLE is non-atomic as whole but we didn't try to commit
the above changes we need to clean-up them before returning.
*/
if (*int_commit_done && (hton->flags & HTON_SUPPORTS_ATOMIC_DDL)) {
Disable_gtid_state_update_guard disabler(thd);
trans_rollback_stmt(thd);
// Full rollback in case we have THD::transaction_rollback_request.
trans_rollback(thd);
/*
Preserve the invariant that FK invalidator is empty after each
step of non-atomic RENAME TABLE.
*/
fk_invalidator->clear();
}
return true;
}
/*
If RENAME TABLE is non-atomic but we have not committed the above
rename and changes to FK we need to do it now.
*/
if (*int_commit_done && (hton->flags & HTON_SUPPORTS_ATOMIC_DDL)) {
Disable_gtid_state_update_guard disabler(thd);
if (trans_commit_stmt(thd) || trans_commit(thd)) {
/*
Preserve the invariant that FK invalidator is empty after each
step of non-atomic RENAME TABLE.
*/
fk_invalidator->clear();
return true;
}
}
*int_commit_done |= !(hton->flags & HTON_SUPPORTS_ATOMIC_DDL);
if (*int_commit_done) {
/*
For non-atomic RENAME TABLE we try to invalidate FK parents right
after transaction commit. This enforces invariant that invalidator
is empty after each step of such RENAME TABLE.
We perform invalidation if there was commit above to handle two
cases:
- We committed rename of table in SE supporting atomic DDL (and so
possibly supporting FKs) since this RENAME TABLE already started
doing intermediate commits.
- We committed rename of table in SE not supporting atomic DDL.
Invalidation still necessary as this might be first non-atomic
rename which follows chain of atomic renames which might have
added pending invalidation requests to invalidator.
*/
fk_invalidator->invalidate(thd);
}
break;
}
case dd::enum_table_type::SYSTEM_VIEW: // Fall through
case dd::enum_table_type::USER_VIEW: {
// Changing the schema of a view is not allowed.
if (strcmp(ren_table->db, new_db)) {
my_error(ER_FORBID_SCHEMA_CHANGE, MYF(0), ren_table->db, new_db);
return true;
}
/* Rename view in the data-dictionary. */
Disable_gtid_state_update_guard disabler(thd);
// Set schema id and view name.
from_at->set_name(new_alias);
// Do the update. Errors will be reported by the dictionary subsystem.
if (thd->dd_client()->update(from_at)) {
if (*int_commit_done) {
trans_rollback_stmt(thd);
// Full rollback in case we have THD::transaction_rollback_request.
trans_rollback(thd);
}
return true;
}
if (*int_commit_done) {
if (trans_commit_stmt(thd) || trans_commit(thd)) return true;
}
sp_cache_invalidate();
break;
}
default:
DBUG_ASSERT(false); /* purecov: deadcode */
}
// Now, we know that rename succeeded, and can log the schema access
thd->add_to_binlog_accessed_dbs(ren_table->db);
thd->add_to_binlog_accessed_dbs(new_db);
return false;
}
/*
Rename all tables in list;
Return pointer to wrong entry if something goes
wrong. Note that the table_list may be empty!
*/
/**
Rename all tables/views in the list.
@param[in] thd Thread handle.
@param[in] table_list List of tables to rename.
@param[in,out] int_commit_done Whether intermediate commits
were done.
@param[in,out] post_ddl_htons Set of SEs supporting atomic DDL
for which post-DDL hooks needs
to be called.
@param[in,out] fk_invalidator Object keeping track of which
dd::Table objects to invalidate.
@note
Take a table/view name from and odd list element and rename it to a
the name taken from list element+1. Note that the table_list may be
empty.
@note Unless int_commit_done is true failure of this call requires
rollback of transaction before doing anything else.
@return 0 - on success, pointer to problematic entry if something
goes wrong.
*/
static TABLE_LIST *rename_tables(
THD *thd, TABLE_LIST *table_list, bool *int_commit_done,
post_ddl_htons_t *post_ddl_htons,
Foreign_key_parents_invalidator *fk_invalidator)
{
TABLE_LIST *ren_table, *new_table;
DBUG_TRACE;
for (ren_table = table_list; ren_table; ren_table = new_table->next_local) {
new_table = ren_table->next_local;
if (do_rename(thd, ren_table, new_table->db, new_table->table_name,
new_table->alias, int_commit_done, post_ddl_htons,
fk_invalidator))
return ren_table;
}
return 0;
}