/* Copyright (c) 2000, 2013, Oracle and/or its affiliates. Copyright (c) 2008, 2014, SkySQL Ab. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; version 2 of the License. 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 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 */ /***************************************************************************** ** ** This file implements classes defined in sql_class.h ** Especially the classes to handle a result from a select ** *****************************************************************************/ #ifdef USE_PRAGMA_IMPLEMENTATION #pragma implementation // gcc: Class implementation #endif #include /* NO_EMBEDDED_ACCESS_CHECKS */ #include "sql_priv.h" #include "unireg.h" // REQUIRED: for other includes #include "sql_class.h" #include "sql_cache.h" // query_cache_abort #include "sql_base.h" // close_thread_tables #include "sql_time.h" // date_time_format_copy #include "tztime.h" // MYSQL_TIME <-> my_time_t #include "sql_acl.h" // NO_ACCESS, // acl_getroot_no_password #include "sql_base.h" // close_temporary_tables #include "sql_handler.h" // mysql_ha_cleanup #include "rpl_rli.h" #include "rpl_filter.h" #include "rpl_record.h" #include "slave.h" #include #include "log_event.h" #include "sql_audit.h" #include #include #include #ifdef __WIN__ #include #endif #include #include #include "sp_rcontext.h" #include "sp_cache.h" #include "transaction.h" #include "sql_select.h" /* declares create_tmp_table() */ #include "debug_sync.h" #include "sql_parse.h" // is_update_query #include "sql_callback.h" #include "lock.h" #include "wsrep_mysqld.h" #include "wsrep_thd.h" #include "sql_connect.h" #include "my_atomic.h" /* The following is used to initialise Table_ident with a internal table name */ char internal_table_name[2]= "*"; char empty_c_string[1]= {0}; /* used for not defined db */ const char * const THD::DEFAULT_WHERE= "field list"; /**************************************************************************** ** User variables ****************************************************************************/ extern "C" uchar *get_var_key(user_var_entry *entry, size_t *length, my_bool not_used __attribute__((unused))) { *length= entry->name.length; return (uchar*) entry->name.str; } extern "C" void free_user_var(user_var_entry *entry) { char *pos= (char*) entry+ALIGN_SIZE(sizeof(*entry)); if (entry->value && entry->value != pos) my_free(entry->value); my_free(entry); } bool Key_part_spec::operator==(const Key_part_spec& other) const { return length == other.length && !my_strcasecmp(system_charset_info, field_name.str, other.field_name.str); } /** Construct an (almost) deep copy of this key. Only those elements that are known to never change are not copied. If out of memory, a partial copy is returned and an error is set in THD. */ Key::Key(const Key &rhs, MEM_ROOT *mem_root) :type(rhs.type), key_create_info(rhs.key_create_info), columns(rhs.columns, mem_root), name(rhs.name), option_list(rhs.option_list), generated(rhs.generated), create_if_not_exists(rhs.create_if_not_exists) { list_copy_and_replace_each_value(columns, mem_root); } /** Construct an (almost) deep copy of this foreign key. Only those elements that are known to never change are not copied. If out of memory, a partial copy is returned and an error is set in THD. */ Foreign_key::Foreign_key(const Foreign_key &rhs, MEM_ROOT *mem_root) :Key(rhs,mem_root), ref_db(rhs.ref_db), ref_table(rhs.ref_table), ref_columns(rhs.ref_columns,mem_root), delete_opt(rhs.delete_opt), update_opt(rhs.update_opt), match_opt(rhs.match_opt) { list_copy_and_replace_each_value(ref_columns, mem_root); } /* Test if a foreign key (= generated key) is a prefix of the given key (ignoring key name, key type and order of columns) NOTES: This is only used to test if an index for a FOREIGN KEY exists IMPLEMENTATION We only compare field names RETURN 0 Generated key is a prefix of other key 1 Not equal */ bool foreign_key_prefix(Key *a, Key *b) { /* Ensure that 'a' is the generated key */ if (a->generated) { if (b->generated && a->columns.elements > b->columns.elements) swap_variables(Key*, a, b); // Put shorter key in 'a' } else { if (!b->generated) return TRUE; // No foreign key swap_variables(Key*, a, b); // Put generated key in 'a' } /* Test if 'a' is a prefix of 'b' */ if (a->columns.elements > b->columns.elements) return TRUE; // Can't be prefix List_iterator col_it1(a->columns); List_iterator col_it2(b->columns); const Key_part_spec *col1, *col2; #ifdef ENABLE_WHEN_INNODB_CAN_HANDLE_SWAPED_FOREIGN_KEY_COLUMNS while ((col1= col_it1++)) { bool found= 0; col_it2.rewind(); while ((col2= col_it2++)) { if (*col1 == *col2) { found= TRUE; break; } } if (!found) return TRUE; // Error } return FALSE; // Is prefix #else while ((col1= col_it1++)) { col2= col_it2++; if (!(*col1 == *col2)) return TRUE; } return FALSE; // Is prefix #endif } /* @brief Check if the foreign key options are compatible with the specification of the columns on which the key is created @retval FALSE The foreign key options are compatible with key columns @retval TRUE Otherwise */ bool Foreign_key::validate(List &table_fields) { Create_field *sql_field; Key_part_spec *column; List_iterator cols(columns); List_iterator it(table_fields); DBUG_ENTER("Foreign_key::validate"); while ((column= cols++)) { it.rewind(); while ((sql_field= it++) && my_strcasecmp(system_charset_info, column->field_name.str, sql_field->field_name)) {} if (!sql_field) { my_error(ER_KEY_COLUMN_DOES_NOT_EXITS, MYF(0), column->field_name); DBUG_RETURN(TRUE); } if (type == Key::FOREIGN_KEY && sql_field->vcol_info) { if (delete_opt == FK_OPTION_SET_NULL) { my_error(ER_WRONG_FK_OPTION_FOR_VIRTUAL_COLUMN, MYF(0), "ON DELETE SET NULL"); DBUG_RETURN(TRUE); } if (update_opt == FK_OPTION_SET_NULL) { my_error(ER_WRONG_FK_OPTION_FOR_VIRTUAL_COLUMN, MYF(0), "ON UPDATE SET NULL"); DBUG_RETURN(TRUE); } if (update_opt == FK_OPTION_CASCADE) { my_error(ER_WRONG_FK_OPTION_FOR_VIRTUAL_COLUMN, MYF(0), "ON UPDATE CASCADE"); DBUG_RETURN(TRUE); } } } DBUG_RETURN(FALSE); } /**************************************************************************** ** Thread specific functions ****************************************************************************/ #ifdef ONLY_FOR_MYSQL_CLOSED_SOURCE_SCHEDULED /** Get reference to scheduler data object @param thd THD object @retval Scheduler data object on THD */ void *thd_get_scheduler_data(THD *thd) { return thd->scheduler.data; } /** Set reference to Scheduler data object for THD object @param thd THD object @param psi Scheduler data object to set on THD */ void thd_set_scheduler_data(THD *thd, void *data) { thd->scheduler.data= data; } /** Get reference to Performance Schema object for THD object @param thd THD object @retval Performance schema object for thread on THD */ PSI_thread *thd_get_psi(THD *thd) { return thd->scheduler.m_psi; } /** Get net_wait_timeout for THD object @param thd THD object @retval net_wait_timeout value for thread on THD */ ulong thd_get_net_wait_timeout(THD* thd) { return thd->variables.net_wait_timeout; } /** Set reference to Performance Schema object for THD object @param thd THD object @param psi Performance schema object for thread */ void thd_set_psi(THD *thd, PSI_thread *psi) { thd->scheduler.m_psi= psi; } /** Set the state on connection to killed @param thd THD object */ void thd_set_killed(THD *thd) { thd->killed= KILL_CONNECTION; } /** Clear errors from the previous THD @param thd THD object */ void thd_clear_errors(THD *thd) { my_errno= 0; thd->mysys_var->abort= 0; } /** Set thread stack in THD object @param thd Thread object @param stack_start Start of stack to set in THD object */ void thd_set_thread_stack(THD *thd, char *stack_start) { thd->thread_stack= stack_start; } /** Close the socket used by this connection @param thd THD object */ void thd_close_connection(THD *thd) { if (thd->net.vio) vio_close(thd->net.vio); } /** Get current THD object from thread local data @retval The THD object for the thread, NULL if not connection thread */ THD *thd_get_current_thd() { return current_thd; } /** Lock data that needs protection in THD object @param thd THD object */ void thd_lock_data(THD *thd) { mysql_mutex_lock(&thd->LOCK_thd_data); } /** Unlock data that needs protection in THD object @param thd THD object */ void thd_unlock_data(THD *thd) { mysql_mutex_unlock(&thd->LOCK_thd_data); } /** Support method to check if connection has already started transcaction @param client_cntx Low level client context @retval TRUE if connection already started transaction */ bool thd_is_transaction_active(THD *thd) { return thd->transaction.is_active(); } /** Check if there is buffered data on the socket representing the connection @param thd THD object */ int thd_connection_has_data(THD *thd) { Vio *vio= thd->net.vio; return vio->has_data(vio); } /** Set reading/writing on socket, used by SHOW PROCESSLIST @param thd THD object @param val Value to set it to (0 or 1) */ void thd_set_net_read_write(THD *thd, uint val) { thd->net.reading_or_writing= val; } /** Get reading/writing on socket from THD object @param thd THD object @retval net.reading_or_writing value for thread on THD. */ uint thd_get_net_read_write(THD *thd) { return thd->net.reading_or_writing; } /** Set reference to mysys variable in THD object @param thd THD object @param mysys_var Reference to set */ void thd_set_mysys_var(THD *thd, st_my_thread_var *mysys_var) { thd->set_mysys_var(mysys_var); } /** Get socket file descriptor for this connection @param thd THD object @retval Socket of the connection */ my_socket thd_get_fd(THD *thd) { return mysql_socket_getfd(thd->net.vio->mysql_socket); } #endif /** Get thread attributes for connection threads @retval Reference to thread attribute for connection threads */ pthread_attr_t *get_connection_attrib(void) { return &connection_attrib; } /** Get max number of connections @retval Max number of connections for MySQL Server */ ulong get_max_connections(void) { return max_connections; } /* The following functions form part of the C plugin API */ extern "C" int mysql_tmpfile(const char *prefix) { char filename[FN_REFLEN]; File fd = create_temp_file(filename, mysql_tmpdir, prefix, #ifdef __WIN__ O_BINARY | O_TRUNC | O_SEQUENTIAL | O_SHORT_LIVED | #endif /* __WIN__ */ O_CREAT | O_EXCL | O_RDWR | O_TEMPORARY, MYF(MY_WME)); if (fd >= 0) { #ifndef __WIN__ /* This can be removed once the following bug is fixed: Bug #28903 create_temp_file() doesn't honor O_TEMPORARY option (file not removed) (Unix) */ unlink(filename); #endif /* !__WIN__ */ } return fd; } extern "C" int thd_in_lock_tables(const THD *thd) { return MY_TEST(thd->in_lock_tables); } extern "C" int thd_tablespace_op(const THD *thd) { return MY_TEST(thd->tablespace_op); } extern "C" const char *set_thd_proc_info(THD *thd_arg, const char *info, const char *calling_function, const char *calling_file, const unsigned int calling_line) { PSI_stage_info old_stage; PSI_stage_info new_stage; old_stage.m_key= 0; old_stage.m_name= info; set_thd_stage_info(thd_arg, & old_stage, & new_stage, calling_function, calling_file, calling_line); return new_stage.m_name; } extern "C" void set_thd_stage_info(void *thd_arg, const PSI_stage_info *new_stage, PSI_stage_info *old_stage, const char *calling_func, const char *calling_file, const unsigned int calling_line) { THD *thd= (THD*) thd_arg; if (thd == NULL) thd= current_thd; thd->enter_stage(new_stage, old_stage, calling_func, calling_file, calling_line); } void THD::enter_stage(const PSI_stage_info *new_stage, PSI_stage_info *old_stage, const char *calling_func, const char *calling_file, const unsigned int calling_line) { DBUG_PRINT("THD::enter_stage", ("%s:%d", calling_file, calling_line)); if (old_stage != NULL) { old_stage->m_key= m_current_stage_key; old_stage->m_name= proc_info; } if (new_stage != NULL) { const char *msg= new_stage->m_name; #if defined(ENABLED_PROFILING) profiling.status_change(msg, calling_func, calling_file, calling_line); #endif m_current_stage_key= new_stage->m_key; proc_info= msg; #ifdef HAVE_PSI_THREAD_INTERFACE MYSQL_SET_STAGE(m_current_stage_key, calling_file, calling_line); #endif } return; } void thd_enter_cond(MYSQL_THD thd, mysql_cond_t *cond, mysql_mutex_t *mutex, const PSI_stage_info *stage, PSI_stage_info *old_stage, const char *src_function, const char *src_file, int src_line) { if (!thd) thd= current_thd; return thd->enter_cond(cond, mutex, stage, old_stage, src_function, src_file, src_line); } void thd_exit_cond(MYSQL_THD thd, const PSI_stage_info *stage, const char *src_function, const char *src_file, int src_line) { if (!thd) thd= current_thd; thd->exit_cond(stage, src_function, src_file, src_line); return; } extern "C" void **thd_ha_data(const THD *thd, const struct handlerton *hton) { return (void **) &thd->ha_data[hton->slot].ha_ptr; } extern "C" void thd_storage_lock_wait(THD *thd, long long value) { thd->utime_after_lock+= value; } /** Provide a handler data getter to simplify coding */ extern "C" void *thd_get_ha_data(const THD *thd, const struct handlerton *hton) { return *thd_ha_data(thd, hton); } /** Provide a handler data setter to simplify coding @see thd_set_ha_data() definition in plugin.h */ extern "C" void thd_set_ha_data(THD *thd, const struct handlerton *hton, const void *ha_data) { plugin_ref *lock= &thd->ha_data[hton->slot].lock; if (ha_data && !*lock) *lock= ha_lock_engine(NULL, (handlerton*) hton); else if (!ha_data && *lock) { plugin_unlock(NULL, *lock); *lock= NULL; } *thd_ha_data(thd, hton)= (void*) ha_data; } /** Allow storage engine to wakeup commits waiting in THD::wait_for_prior_commit. @see thd_wakeup_subsequent_commits() definition in plugin.h */ extern "C" void thd_wakeup_subsequent_commits(THD *thd, int wakeup_error) { thd->wakeup_subsequent_commits(wakeup_error); } extern "C" long long thd_test_options(const THD *thd, long long test_options) { return thd->variables.option_bits & test_options; } extern "C" int thd_sql_command(const THD *thd) { return (int) thd->lex->sql_command; } extern "C" int thd_tx_isolation(const THD *thd) { return (int) thd->tx_isolation; } extern "C" int thd_tx_is_read_only(const THD *thd) { return (int) thd->tx_read_only; } extern "C" { /* Functions for thd_error_context_service */ const char *thd_get_error_message(const THD *thd) { return thd->get_stmt_da()->message(); } uint thd_get_error_number(const THD *thd) { return thd->get_stmt_da()->sql_errno(); } ulong thd_get_error_row(const THD *thd) { return thd->get_stmt_da()->current_row_for_warning(); } void thd_inc_error_row(THD *thd) { thd->get_stmt_da()->inc_current_row_for_warning(); } } /** Dumps a text description of a thread, its security context (user, host) and the current query. @param thd thread context @param buffer pointer to preferred result buffer @param length length of buffer @param max_query_len how many chars of query to copy (0 for all) @req LOCK_thread_count @note LOCK_thread_count mutex is not necessary when the function is invoked on the currently running thread (current_thd) or if the caller in some other way guarantees that access to thd->query is serialized. @return Pointer to string */ extern "C" char *thd_get_error_context_description(THD *thd, char *buffer, unsigned int length, unsigned int max_query_len) { String str(buffer, length, &my_charset_latin1); const Security_context *sctx= &thd->main_security_ctx; char header[256]; int len; /* The pointers thd->query and thd->proc_info might change since they are being modified concurrently. This is acceptable for proc_info since its values doesn't have to very accurate and the memory it points to is static, but we need to attempt a snapshot on the pointer values to avoid using NULL values. The pointer to thd->query however, doesn't point to static memory and has to be protected by thd->LOCK_thd_data or risk pointing to uninitialized memory. */ const char *proc_info= thd->proc_info; len= my_snprintf(header, sizeof(header), "MySQL thread id %lu, OS thread handle 0x%lx, query id %lu", thd->thread_id, (ulong) thd->real_id, (ulong) thd->query_id); str.length(0); str.append(header, len); if (sctx->host) { str.append(' '); str.append(sctx->host); } if (sctx->ip) { str.append(' '); str.append(sctx->ip); } if (sctx->user) { str.append(' '); str.append(sctx->user); } if (proc_info) { str.append(' '); str.append(proc_info); } /* Don't wait if LOCK_thd_data is used as this could cause a deadlock */ if (!mysql_mutex_trylock(&thd->LOCK_thd_data)) { if (thd->query()) { if (max_query_len < 1) len= thd->query_length(); else len= MY_MIN(thd->query_length(), max_query_len); str.append('\n'); str.append(thd->query(), len); } mysql_mutex_unlock(&thd->LOCK_thd_data); } if (str.c_ptr_safe() == buffer) return buffer; /* We have to copy the new string to the destination buffer because the string was reallocated to a larger buffer to be able to fit. */ DBUG_ASSERT(buffer != NULL); length= MY_MIN(str.length(), length-1); memcpy(buffer, str.c_ptr_quick(), length); /* Make sure that the new string is null terminated */ buffer[length]= '\0'; return buffer; } #if MARIA_PLUGIN_INTERFACE_VERSION < 0x0200 /** TODO: This function is for API compatibility, remove it eventually. All engines should switch to use thd_get_error_context_description() plugin service function. */ extern "C" char *thd_security_context(THD *thd, char *buffer, unsigned int length, unsigned int max_query_len) { return thd_get_error_context_description(thd, buffer, length, max_query_len); } #endif /** Implementation of Drop_table_error_handler::handle_condition(). The reason in having this implementation is to silence technical low-level warnings during DROP TABLE operation. Currently we don't want to expose the following warnings during DROP TABLE: - Some of table files are missed or invalid (the table is going to be deleted anyway, so why bother that something was missed); - A trigger associated with the table does not have DEFINER (One of the MySQL specifics now is that triggers are loaded for the table being dropped. So, we may have a warning that trigger does not have DEFINER attribute during DROP TABLE operation). @return TRUE if the condition is handled. */ bool Drop_table_error_handler::handle_condition(THD *thd, uint sql_errno, const char* sqlstate, Sql_condition::enum_warning_level level, const char* msg, Sql_condition ** cond_hdl) { *cond_hdl= NULL; return ((sql_errno == EE_DELETE && my_errno == ENOENT) || sql_errno == ER_TRG_NO_DEFINER); } /** Send timeout to thread. Note that this is always safe as the thread will always remove it's timeouts at end of query (and thus before THD is destroyed) */ extern "C" void thd_kill_timeout(THD* thd) { thd->status_var.max_statement_time_exceeded++; mysql_mutex_lock(&thd->LOCK_thd_data); /* Kill queries that can't cause data corruptions */ thd->awake(KILL_TIMEOUT); mysql_mutex_unlock(&thd->LOCK_thd_data); } THD::THD(bool is_wsrep_applier) :Statement(&main_lex, &main_mem_root, STMT_CONVENTIONAL_EXECUTION, /* statement id */ 0), rli_fake(0), rgi_fake(0), rgi_slave(NULL), in_sub_stmt(0), log_all_errors(0), binlog_unsafe_warning_flags(0), binlog_table_maps(0), table_map_for_update(0), arg_of_last_insert_id_function(FALSE), first_successful_insert_id_in_prev_stmt(0), first_successful_insert_id_in_prev_stmt_for_binlog(0), first_successful_insert_id_in_cur_stmt(0), stmt_depends_on_first_successful_insert_id_in_prev_stmt(FALSE), m_examined_row_count(0), accessed_rows_and_keys(0), m_statement_psi(NULL), m_idle_psi(NULL), thread_id(0), global_disable_checkpoint(0), failed_com_change_user(0), is_fatal_error(0), transaction_rollback_request(0), is_fatal_sub_stmt_error(0), rand_used(0), time_zone_used(0), in_lock_tables(0), bootstrap(0), derived_tables_processing(FALSE), spcont(NULL), m_parser_state(NULL), #if defined(ENABLED_DEBUG_SYNC) debug_sync_control(0), #endif /* defined(ENABLED_DEBUG_SYNC) */ wait_for_commit_ptr(0), main_da(0, false, false), m_stmt_da(&main_da), tdc_hash_pins(0) #ifdef WITH_WSREP , wsrep_applier(is_wsrep_applier), wsrep_applier_closing(false), wsrep_client_thread(false), wsrep_apply_toi(false), wsrep_po_handle(WSREP_PO_INITIALIZER), wsrep_po_cnt(0), wsrep_apply_format(0) #endif { ulong tmp; mdl_context.init(this); /* We set THR_THD to temporally point to this THD to register all the variables that allocates memory for this THD */ THD *old_THR_THD= current_thd; set_current_thd(this); status_var.local_memory_used= status_var.global_memory_used= 0; main_da.init(); /* Pass nominal parameters to init_alloc_root only to ensure that the destructor works OK in case of an error. The main_mem_root will be re-initialized in init_for_queries(). */ init_sql_alloc(&main_mem_root, ALLOC_ROOT_MIN_BLOCK_SIZE, 0, MYF(MY_THREAD_SPECIFIC)); stmt_arena= this; thread_stack= 0; scheduler= thread_scheduler; // Will be fixed later event_scheduler.data= 0; event_scheduler.m_psi= 0; skip_wait_timeout= false; extra_port= 0; catalog= (char*)"std"; // the only catalog we have for now main_security_ctx.init(); security_ctx= &main_security_ctx; no_errors= 0; password= 0; query_start_used= query_start_sec_part_used= 0; count_cuted_fields= CHECK_FIELD_IGNORE; killed= NOT_KILLED; col_access=0; is_slave_error= thread_specific_used= FALSE; my_hash_clear(&handler_tables_hash); my_hash_clear(&ull_hash); tmp_table=0; cuted_fields= 0L; m_sent_row_count= 0L; limit_found_rows= 0; m_row_count_func= -1; statement_id_counter= 0UL; // Must be reset to handle error with THD's created for init of mysqld lex->current_select= 0; user_time.val= start_time= start_time_sec_part= 0; start_utime= prior_thr_create_utime= 0L; utime_after_lock= 0L; progress.arena= 0; progress.report_to_client= 0; progress.max_counter= 0; current_linfo = 0; slave_thread = 0; connection_name.str= 0; connection_name.length= 0; bzero(&variables, sizeof(variables)); file_id = 0; query_id= 0; query_name_consts= 0; db_charset= global_system_variables.collation_database; bzero(ha_data, sizeof(ha_data)); mysys_var=0; binlog_evt_union.do_union= FALSE; enable_slow_log= 0; durability_property= HA_REGULAR_DURABILITY; #ifndef DBUG_OFF dbug_sentry=THD_SENTRY_MAGIC; #endif #ifndef EMBEDDED_LIBRARY mysql_audit_init_thd(this); #endif net.vio=0; net.buff= 0; client_capabilities= 0; // minimalistic client system_thread= NON_SYSTEM_THREAD; cleanup_done= abort_on_warning= 0; peer_port= 0; // For SHOW PROCESSLIST transaction.m_pending_rows_event= 0; transaction.on= 1; wt_thd_lazy_init(&transaction.wt, &variables.wt_deadlock_search_depth_short, &variables.wt_timeout_short, &variables.wt_deadlock_search_depth_long, &variables.wt_timeout_long); #ifdef SIGNAL_WITH_VIO_CLOSE active_vio = 0; #endif mysql_mutex_init(key_LOCK_thd_data, &LOCK_thd_data, MY_MUTEX_INIT_FAST); mysql_mutex_init(key_LOCK_wakeup_ready, &LOCK_wakeup_ready, MY_MUTEX_INIT_FAST); mysql_cond_init(key_COND_wakeup_ready, &COND_wakeup_ready, 0); /* LOCK_thread_count goes before LOCK_thd_data - the former is called around 'delete thd', the latter - in THD::~THD */ mysql_mutex_record_order(&LOCK_thread_count, &LOCK_thd_data); /* Variables with default values */ proc_info="login"; where= THD::DEFAULT_WHERE; variables.server_id = global_system_variables.server_id; slave_net = 0; m_command=COM_CONNECT; *scramble= '\0'; #ifdef WITH_WSREP mysql_mutex_init(key_LOCK_wsrep_thd, &LOCK_wsrep_thd, MY_MUTEX_INIT_FAST); wsrep_ws_handle.trx_id = WSREP_UNDEFINED_TRX_ID; wsrep_ws_handle.opaque = NULL; wsrep_retry_counter = 0; wsrep_PA_safe = true; wsrep_retry_query = NULL; wsrep_retry_query_len = 0; wsrep_retry_command = COM_CONNECT; wsrep_consistency_check = NO_CONSISTENCY_CHECK; wsrep_mysql_replicated = 0; wsrep_TOI_pre_query = NULL; wsrep_TOI_pre_query_len = 0; wsrep_info[sizeof(wsrep_info) - 1] = '\0'; /* make sure it is 0-terminated */ #endif /* Call to init() below requires fully initialized Open_tables_state. */ reset_open_tables_state(this); init(); #if defined(ENABLED_PROFILING) profiling.set_thd(this); #endif user_connect=(USER_CONN *)0; my_hash_init(&user_vars, system_charset_info, USER_VARS_HASH_SIZE, 0, 0, (my_hash_get_key) get_var_key, (my_hash_free_key) free_user_var, HASH_THREAD_SPECIFIC); sp_proc_cache= NULL; sp_func_cache= NULL; /* For user vars replication*/ if (opt_bin_log) my_init_dynamic_array(&user_var_events, sizeof(BINLOG_USER_VAR_EVENT *), 16, 16, MYF(0)); else bzero((char*) &user_var_events, sizeof(user_var_events)); /* Protocol */ protocol= &protocol_text; // Default protocol protocol_text.init(this); protocol_binary.init(this); thr_timer_init(&query_timer, (void (*)(void*)) thd_kill_timeout, this); tablespace_op=FALSE; /* Initialize the random generator. We call my_rnd() without a lock as it's not really critical if two threads modifies the structure at the same time. We ensure that we have an unique number foreach thread by adding the address of the stack. */ tmp= (ulong) (my_rnd(&sql_rand) * 0xffffffff); my_rnd_init(&rand, tmp + (ulong) &rand, tmp + (ulong) ::global_query_id); substitute_null_with_insert_id = FALSE; thr_lock_info_init(&lock_info); /* safety: will be reset after start */ lock_info.mysql_thd= (void *)this; m_internal_handler= NULL; m_binlog_invoker= INVOKER_NONE; arena_for_cached_items= 0; memset(&invoker_user, 0, sizeof(invoker_user)); memset(&invoker_host, 0, sizeof(invoker_host)); prepare_derived_at_open= FALSE; create_tmp_table_for_derived= FALSE; save_prep_leaf_list= FALSE; /* Restore THR_THD */ set_current_thd(old_THR_THD); } void THD::push_internal_handler(Internal_error_handler *handler) { DBUG_ENTER("THD::push_internal_handler"); if (m_internal_handler) { handler->m_prev_internal_handler= m_internal_handler; m_internal_handler= handler; } else { m_internal_handler= handler; } DBUG_VOID_RETURN; } bool THD::handle_condition(uint sql_errno, const char* sqlstate, Sql_condition::enum_warning_level level, const char* msg, Sql_condition ** cond_hdl) { if (!m_internal_handler) { *cond_hdl= NULL; return FALSE; } for (Internal_error_handler *error_handler= m_internal_handler; error_handler; error_handler= error_handler->m_prev_internal_handler) { if (error_handler->handle_condition(this, sql_errno, sqlstate, level, msg, cond_hdl)) { return TRUE; } } return FALSE; } Internal_error_handler *THD::pop_internal_handler() { DBUG_ENTER("THD::pop_internal_handler"); DBUG_ASSERT(m_internal_handler != NULL); Internal_error_handler *popped_handler= m_internal_handler; m_internal_handler= m_internal_handler->m_prev_internal_handler; DBUG_RETURN(popped_handler); } void THD::raise_error(uint sql_errno) { const char* msg= ER(sql_errno); (void) raise_condition(sql_errno, NULL, Sql_condition::WARN_LEVEL_ERROR, msg); } void THD::raise_error_printf(uint sql_errno, ...) { va_list args; char ebuff[MYSQL_ERRMSG_SIZE]; DBUG_ENTER("THD::raise_error_printf"); DBUG_PRINT("my", ("nr: %d errno: %d", sql_errno, errno)); const char* format= ER(sql_errno); va_start(args, sql_errno); my_vsnprintf(ebuff, sizeof(ebuff), format, args); va_end(args); (void) raise_condition(sql_errno, NULL, Sql_condition::WARN_LEVEL_ERROR, ebuff); DBUG_VOID_RETURN; } void THD::raise_warning(uint sql_errno) { const char* msg= ER(sql_errno); (void) raise_condition(sql_errno, NULL, Sql_condition::WARN_LEVEL_WARN, msg); } void THD::raise_warning_printf(uint sql_errno, ...) { va_list args; char ebuff[MYSQL_ERRMSG_SIZE]; DBUG_ENTER("THD::raise_warning_printf"); DBUG_PRINT("enter", ("warning: %u", sql_errno)); const char* format= ER(sql_errno); va_start(args, sql_errno); my_vsnprintf(ebuff, sizeof(ebuff), format, args); va_end(args); (void) raise_condition(sql_errno, NULL, Sql_condition::WARN_LEVEL_WARN, ebuff); DBUG_VOID_RETURN; } void THD::raise_note(uint sql_errno) { DBUG_ENTER("THD::raise_note"); DBUG_PRINT("enter", ("code: %d", sql_errno)); if (!(variables.option_bits & OPTION_SQL_NOTES)) DBUG_VOID_RETURN; const char* msg= ER(sql_errno); (void) raise_condition(sql_errno, NULL, Sql_condition::WARN_LEVEL_NOTE, msg); DBUG_VOID_RETURN; } void THD::raise_note_printf(uint sql_errno, ...) { va_list args; char ebuff[MYSQL_ERRMSG_SIZE]; DBUG_ENTER("THD::raise_note_printf"); DBUG_PRINT("enter",("code: %u", sql_errno)); if (!(variables.option_bits & OPTION_SQL_NOTES)) DBUG_VOID_RETURN; const char* format= ER(sql_errno); va_start(args, sql_errno); my_vsnprintf(ebuff, sizeof(ebuff), format, args); va_end(args); (void) raise_condition(sql_errno, NULL, Sql_condition::WARN_LEVEL_NOTE, ebuff); DBUG_VOID_RETURN; } Sql_condition* THD::raise_condition(uint sql_errno, const char* sqlstate, Sql_condition::enum_warning_level level, const char* msg) { Diagnostics_area *da= get_stmt_da(); Sql_condition *cond= NULL; DBUG_ENTER("THD::raise_condition"); if (!(variables.option_bits & OPTION_SQL_NOTES) && (level == Sql_condition::WARN_LEVEL_NOTE)) DBUG_RETURN(NULL); da->opt_clear_warning_info(query_id); /* TODO: replace by DBUG_ASSERT(sql_errno != 0) once all bugs similar to Bug#36768 are fixed: a SQL condition must have a real (!=0) error number so that it can be caught by handlers. */ if (sql_errno == 0) sql_errno= ER_UNKNOWN_ERROR; if (msg == NULL) msg= ER(sql_errno); if (sqlstate == NULL) sqlstate= mysql_errno_to_sqlstate(sql_errno); if ((level == Sql_condition::WARN_LEVEL_WARN) && really_abort_on_warning()) { /* FIXME: push_warning and strict SQL_MODE case. */ level= Sql_condition::WARN_LEVEL_ERROR; killed= KILL_BAD_DATA; } switch (level) { case Sql_condition::WARN_LEVEL_NOTE: case Sql_condition::WARN_LEVEL_WARN: got_warning= 1; break; case Sql_condition::WARN_LEVEL_ERROR: break; default: DBUG_ASSERT(FALSE); } if (handle_condition(sql_errno, sqlstate, level, msg, &cond)) DBUG_RETURN(cond); if (level == Sql_condition::WARN_LEVEL_ERROR) { mysql_audit_general(this, MYSQL_AUDIT_GENERAL_ERROR, sql_errno, msg); is_slave_error= 1; // needed to catch query errors during replication if (!da->is_error()) { set_row_count_func(-1); da->set_error_status(sql_errno, msg, sqlstate, cond); } } query_cache_abort(&query_cache_tls); /* Avoid pushing a condition for fatal out of memory errors as this will require memory allocation and therefore might fail. Non fatal out of memory errors can occur if raised by SIGNAL/RESIGNAL statement. */ if (!(is_fatal_error && (sql_errno == EE_OUTOFMEMORY || sql_errno == ER_OUTOFMEMORY))) { cond= da->push_warning(this, sql_errno, sqlstate, level, msg); } DBUG_RETURN(cond); } extern "C" void *thd_alloc(MYSQL_THD thd, unsigned int size) { return thd->alloc(size); } extern "C" void *thd_calloc(MYSQL_THD thd, unsigned int size) { return thd->calloc(size); } extern "C" char *thd_strdup(MYSQL_THD thd, const char *str) { return thd->strdup(str); } extern "C" char *thd_strmake(MYSQL_THD thd, const char *str, unsigned int size) { return thd->strmake(str, size); } extern "C" LEX_STRING *thd_make_lex_string(THD *thd, LEX_STRING *lex_str, const char *str, unsigned int size, int allocate_lex_string) { return allocate_lex_string ? thd->make_lex_string(str, size) : thd->make_lex_string(lex_str, str, size); } extern "C" void *thd_memdup(MYSQL_THD thd, const void* str, unsigned int size) { return thd->memdup(str, size); } extern "C" void thd_get_xid(const MYSQL_THD thd, MYSQL_XID *xid) { *xid = *(MYSQL_XID *) &thd->transaction.xid_state.xid; } extern "C" my_time_t thd_TIME_to_gmt_sec(MYSQL_THD thd, const MYSQL_TIME *ltime, unsigned int *errcode) { Time_zone *tz= thd ? thd->variables.time_zone : global_system_variables.time_zone; return tz->TIME_to_gmt_sec(ltime, errcode); } extern "C" void thd_gmt_sec_to_TIME(MYSQL_THD thd, MYSQL_TIME *ltime, my_time_t t) { Time_zone *tz= thd ? thd->variables.time_zone : global_system_variables.time_zone; tz->gmt_sec_to_TIME(ltime, t); } #ifdef _WIN32 extern "C" THD *_current_thd_noinline(void) { return my_pthread_getspecific_ptr(THD*,THR_THD); } #endif /* Init common variables that has to be reset on start and on change_user */ void THD::init(void) { DBUG_ENTER("thd::init"); mysql_mutex_lock(&LOCK_global_system_variables); plugin_thdvar_init(this); /* plugin_thd_var_init() sets variables= global_system_variables, which has reset variables.pseudo_thread_id to 0. We need to correct it here to avoid temporary tables replication failure. */ variables.pseudo_thread_id= thread_id; variables.default_master_connection.str= default_master_connection_buff; ::strmake(variables.default_master_connection.str, global_system_variables.default_master_connection.str, variables.default_master_connection.length); mysql_mutex_unlock(&LOCK_global_system_variables); server_status= SERVER_STATUS_AUTOCOMMIT; if (variables.sql_mode & MODE_NO_BACKSLASH_ESCAPES) server_status|= SERVER_STATUS_NO_BACKSLASH_ESCAPES; transaction.all.modified_non_trans_table= transaction.stmt.modified_non_trans_table= FALSE; transaction.all.m_unsafe_rollback_flags= transaction.stmt.m_unsafe_rollback_flags= 0; open_options=ha_open_options; update_lock_default= (variables.low_priority_updates ? TL_WRITE_LOW_PRIORITY : TL_WRITE); tx_isolation= (enum_tx_isolation) variables.tx_isolation; tx_read_only= variables.tx_read_only; update_charset(); reset_current_stmt_binlog_format_row(); reset_binlog_local_stmt_filter(); set_status_var_init(); bzero((char *) &org_status_var, sizeof(org_status_var)); start_bytes_received= 0; last_commit_gtid.seq_no= 0; #ifdef WITH_WSREP wsrep_exec_mode= wsrep_applier ? REPL_RECV : LOCAL_STATE; wsrep_conflict_state= NO_CONFLICT; wsrep_query_state= QUERY_IDLE; wsrep_last_query_id= 0; wsrep_trx_meta.gtid= WSREP_GTID_UNDEFINED; wsrep_trx_meta.depends_on= WSREP_SEQNO_UNDEFINED; wsrep_converted_lock_session= false; wsrep_retry_counter= 0; wsrep_rli= NULL; wsrep_rgi= NULL; wsrep_PA_safe= true; wsrep_consistency_check = NO_CONSISTENCY_CHECK; wsrep_mysql_replicated = 0; wsrep_TOI_pre_query = NULL; wsrep_TOI_pre_query_len = 0; #endif if (variables.sql_log_bin) variables.option_bits|= OPTION_BIN_LOG; else variables.option_bits&= ~OPTION_BIN_LOG; select_commands= update_commands= other_commands= 0; /* Set to handle counting of aborted connections */ userstat_running= opt_userstat_running; last_global_update_time= current_connect_time= time(NULL); #if defined(ENABLED_DEBUG_SYNC) /* Initialize the Debug Sync Facility. See debug_sync.cc. */ debug_sync_init_thread(this); #endif /* defined(ENABLED_DEBUG_SYNC) */ apc_target.init(&LOCK_thd_data); DBUG_VOID_RETURN; } /* Updates some status variables to be used by update_global_user_stats */ void THD::update_stats(void) { /* sql_command == SQLCOM_END in case of parse errors or quit */ if (lex->sql_command != SQLCOM_END) { /* A SQL query. */ if (lex->sql_command == SQLCOM_SELECT) select_commands++; else if (sql_command_flags[lex->sql_command] & CF_STATUS_COMMAND) { /* Ignore 'SHOW ' commands */ } else if (is_update_query(lex->sql_command)) update_commands++; else other_commands++; } } void THD::update_all_stats() { ulonglong end_cpu_time, end_utime; double busy_time, cpu_time; /* This is set at start of query if opt_userstat_running was set */ if (!userstat_running) return; end_cpu_time= my_getcputime(); end_utime= microsecond_interval_timer(); busy_time= (end_utime - start_utime) / 1000000.0; cpu_time= (end_cpu_time - start_cpu_time) / 10000000.0; /* In case there are bad values, 2629743 is the #seconds in a month. */ if (cpu_time > 2629743.0) cpu_time= 0; status_var_add(status_var.cpu_time, cpu_time); status_var_add(status_var.busy_time, busy_time); update_global_user_stats(this, TRUE, my_time(0)); // Has to be updated after update_global_user_stats() userstat_running= 0; } /* Init THD for query processing. This has to be called once before we call mysql_parse. See also comments in sql_class.h. */ void THD::init_for_queries() { set_time(); ha_enable_transaction(this,TRUE); reset_root_defaults(mem_root, variables.query_alloc_block_size, variables.query_prealloc_size); reset_root_defaults(&transaction.mem_root, variables.trans_alloc_block_size, variables.trans_prealloc_size); transaction.xid_state.xid.null(); transaction.xid_state.in_thd=1; } /* Do what's needed when one invokes change user SYNOPSIS change_user() IMPLEMENTATION Reset all resources that are connection specific */ void THD::change_user(void) { add_status_to_global(); cleanup(); reset_killed(); cleanup_done= 0; init(); stmt_map.reset(); my_hash_init(&user_vars, system_charset_info, USER_VARS_HASH_SIZE, 0, 0, (my_hash_get_key) get_var_key, (my_hash_free_key) free_user_var, 0); sp_cache_clear(&sp_proc_cache); sp_cache_clear(&sp_func_cache); } /* Do operations that may take a long time */ void THD::cleanup(void) { DBUG_ENTER("THD::cleanup"); DBUG_ASSERT(cleanup_done == 0); killed= KILL_CONNECTION; #ifdef ENABLE_WHEN_BINLOG_WILL_BE_ABLE_TO_PREPARE if (transaction.xid_state.xa_state == XA_PREPARED) { #error xid_state in the cache should be replaced by the allocated value } #endif mysql_ha_cleanup(this); locked_tables_list.unlock_locked_tables(this); close_temporary_tables(this); transaction.xid_state.xa_state= XA_NOTR; trans_rollback(this); xid_cache_delete(&transaction.xid_state); DBUG_ASSERT(open_tables == NULL); /* If the thread was in the middle of an ongoing transaction (rolled back a few lines above) or under LOCK TABLES (unlocked the tables and left the mode a few lines above), there will be outstanding metadata locks. Release them. */ mdl_context.release_transactional_locks(); /* Release the global read lock, if acquired. */ if (global_read_lock.is_acquired()) global_read_lock.unlock_global_read_lock(this); if (user_connect) { decrease_user_connections(user_connect); user_connect= 0; // Safety } wt_thd_destroy(&transaction.wt); #if defined(ENABLED_DEBUG_SYNC) /* End the Debug Sync Facility. See debug_sync.cc. */ debug_sync_end_thread(this); #endif /* defined(ENABLED_DEBUG_SYNC) */ delete_dynamic(&user_var_events); my_hash_free(&user_vars); sp_cache_clear(&sp_proc_cache); sp_cache_clear(&sp_func_cache); mysql_ull_cleanup(this); /* All metadata locks must have been released by now. */ DBUG_ASSERT(!mdl_context.has_locks()); apc_target.destroy(); cleanup_done=1; DBUG_VOID_RETURN; } THD::~THD() { THD *orig_thd= current_thd; THD_CHECK_SENTRY(this); DBUG_ENTER("~THD()"); /* In error cases, thd may not be current thd. We have to fix this so that memory allocation counting is done correctly */ set_current_thd(this); /* Ensure that no one is using THD */ mysql_mutex_lock(&LOCK_thd_data); mysql_mutex_unlock(&LOCK_thd_data); #ifdef WITH_WSREP mysql_mutex_lock(&LOCK_wsrep_thd); mysql_mutex_unlock(&LOCK_wsrep_thd); mysql_mutex_destroy(&LOCK_wsrep_thd); if (wsrep_rli) delete wsrep_rli; if (wsrep_rgi) delete wsrep_rgi; #endif /* Close connection */ #ifndef EMBEDDED_LIBRARY if (net.vio) vio_delete(net.vio); net_end(&net); #endif stmt_map.reset(); /* close all prepared statements */ if (!cleanup_done) cleanup(); mdl_context.destroy(); ha_close_connection(this); mysql_audit_release(this); plugin_thdvar_cleanup(this); main_security_ctx.destroy(); my_free(db); db= NULL; free_root(&transaction.mem_root,MYF(0)); mysql_cond_destroy(&COND_wakeup_ready); mysql_mutex_destroy(&LOCK_wakeup_ready); mysql_mutex_destroy(&LOCK_thd_data); #ifndef DBUG_OFF dbug_sentry= THD_SENTRY_GONE; #endif #ifndef EMBEDDED_LIBRARY if (rgi_fake) { delete rgi_fake; rgi_fake= NULL; } if (rli_fake) { delete rli_fake; rli_fake= NULL; } mysql_audit_free_thd(this); if (rgi_slave) rgi_slave->cleanup_after_session(); #endif free_root(&main_mem_root, MYF(0)); main_da.free_memory(); if (tdc_hash_pins) lf_hash_put_pins(tdc_hash_pins); /* Ensure everything is freed */ if (status_var.local_memory_used != 0) { DBUG_PRINT("error", ("memory_used: %lld", status_var.local_memory_used)); SAFEMALLOC_REPORT_MEMORY(my_thread_dbug_id()); DBUG_ASSERT(status_var.local_memory_used == 0); } set_current_thd(orig_thd == this ? 0 : orig_thd); DBUG_VOID_RETURN; } /* Add all status variables to another status variable array SYNOPSIS add_to_status() to_var add to this array from_var from this array NOTES This function assumes that all variables at start are long/ulong and other types are handled explicitely */ void add_to_status(STATUS_VAR *to_var, STATUS_VAR *from_var) { ulong *end= (ulong*) ((uchar*) to_var + offsetof(STATUS_VAR, last_system_status_var) + sizeof(ulong)); ulong *to= (ulong*) to_var, *from= (ulong*) from_var; while (to != end) *(to++)+= *(from++); /* Handle the not ulong variables. See end of system_status_var */ to_var->bytes_received+= from_var->bytes_received; to_var->bytes_sent+= from_var->bytes_sent; to_var->rows_read+= from_var->rows_read; to_var->rows_sent+= from_var->rows_sent; to_var->rows_tmp_read+= from_var->rows_tmp_read; to_var->binlog_bytes_written+= from_var->binlog_bytes_written; to_var->cpu_time+= from_var->cpu_time; to_var->busy_time+= from_var->busy_time; to_var->local_memory_used+= from_var->local_memory_used; /* Update global_memory_used. We have to do this with atomic_add as the global value can change outside of LOCK_status. */ // workaround for gcc 4.2.4-1ubuntu4 -fPIE (from DEB_BUILD_HARDENING=1) int64 volatile * volatile ptr= &to_var->global_memory_used; my_atomic_add64_explicit(ptr, from_var->global_memory_used, MY_MEMORY_ORDER_RELAXED); } /* Add the difference between two status variable arrays to another one. SYNOPSIS add_diff_to_status to_var add to this array from_var from this array dec_var minus this array NOTE This function assumes that all variables at start are long/ulong and other types are handled explicitely */ void add_diff_to_status(STATUS_VAR *to_var, STATUS_VAR *from_var, STATUS_VAR *dec_var) { ulong *end= (ulong*) ((uchar*) to_var + offsetof(STATUS_VAR, last_system_status_var) + sizeof(ulong)); ulong *to= (ulong*) to_var, *from= (ulong*) from_var, *dec= (ulong*) dec_var; while (to != end) *(to++)+= *(from++) - *(dec++); to_var->bytes_received+= from_var->bytes_received - dec_var->bytes_received; to_var->bytes_sent+= from_var->bytes_sent - dec_var->bytes_sent; to_var->rows_read+= from_var->rows_read - dec_var->rows_read; to_var->rows_sent+= from_var->rows_sent - dec_var->rows_sent; to_var->rows_tmp_read+= from_var->rows_tmp_read - dec_var->rows_tmp_read; to_var->binlog_bytes_written+= from_var->binlog_bytes_written - dec_var->binlog_bytes_written; to_var->cpu_time+= from_var->cpu_time - dec_var->cpu_time; to_var->busy_time+= from_var->busy_time - dec_var->busy_time; /* We don't need to accumulate memory_used as these are not reset or used by the calling functions. See execute_show_status(). */ } #define SECONDS_TO_WAIT_FOR_KILL 2 #if !defined(__WIN__) && defined(HAVE_SELECT) /* my_sleep() can wait for sub second times */ #define WAIT_FOR_KILL_TRY_TIMES 20 #else #define WAIT_FOR_KILL_TRY_TIMES 2 #endif /** Awake a thread. @param[in] state_to_set value for THD::killed This is normally called from another thread's THD object. @note Do always call this while holding LOCK_thd_data. NOT_KILLED is used to awake a thread for a slave */ void THD::awake(killed_state state_to_set) { DBUG_ENTER("THD::awake"); DBUG_PRINT("enter", ("this: %p current_thd: %p", this, current_thd)); THD_CHECK_SENTRY(this); mysql_mutex_assert_owner(&LOCK_thd_data); print_aborted_warning(3, "KILLED"); /* Don't degrade killed state, for example from a KILL_CONNECTION to STATEMENT TIMEOUT */ if (killed >= KILL_CONNECTION) state_to_set= killed; /* Set the 'killed' flag of 'this', which is the target THD object. */ killed= state_to_set; if (state_to_set >= KILL_CONNECTION || state_to_set == NOT_KILLED) { #ifdef SIGNAL_WITH_VIO_CLOSE if (this != current_thd) { if(active_vio) vio_shutdown(active_vio, SHUT_RDWR); } #endif /* Mark the target thread's alarm request expired, and signal alarm. */ thr_alarm_kill(thread_id); /* Send an event to the scheduler that a thread should be killed. */ if (!slave_thread) MYSQL_CALLBACK(scheduler, post_kill_notification, (this)); } /* Interrupt target waiting inside a storage engine. */ if (state_to_set != NOT_KILLED) ha_kill_query(this, thd_kill_level(this)); /* Broadcast a condition to kick the target if it is waiting on it. */ if (mysys_var) { mysql_mutex_lock(&mysys_var->mutex); if (!system_thread) // Don't abort locks mysys_var->abort=1; /* This broadcast could be up in the air if the victim thread exits the cond in the time between read and broadcast, but that is ok since all we want to do is to make the victim thread get out of waiting on current_cond. If we see a non-zero current_cond: it cannot be an old value (because then exit_cond() should have run and it can't because we have mutex); so it is the true value but maybe current_mutex is not yet non-zero (we're in the middle of enter_cond() and there is a "memory order inversion"). So we test the mutex too to not lock 0. Note that there is a small chance we fail to kill. If victim has locked current_mutex, but hasn't yet entered enter_cond() (which means that current_cond and current_mutex are 0), then the victim will not get a signal and it may wait "forever" on the cond (until we issue a second KILL or the status it's waiting for happens). It's true that we have set its thd->killed but it may not see it immediately and so may have time to reach the cond_wait(). However, where possible, we test for killed once again after enter_cond(). This should make the signaling as safe as possible. However, there is still a small chance of failure on platforms with instruction or memory write reordering. We have to do the loop with trylock, because if we would use pthread_mutex_lock(), we can cause a deadlock as we are here locking the mysys_var->mutex and mysys_var->current_mutex in a different order than in the thread we are trying to kill. We only sleep for 2 seconds as we don't want to have LOCK_thd_data locked too long time. There is a small change we may not succeed in aborting a thread that is not yet waiting for a mutex, but as this happens only for a thread that was doing something else when the kill was issued and which should detect the kill flag before it starts to wait, this should be good enough. */ if (mysys_var->current_cond && mysys_var->current_mutex) { uint i; for (i= 0; i < WAIT_FOR_KILL_TRY_TIMES * SECONDS_TO_WAIT_FOR_KILL; i++) { int ret= mysql_mutex_trylock(mysys_var->current_mutex); mysql_cond_broadcast(mysys_var->current_cond); if (!ret) { /* Signal is sure to get through */ mysql_mutex_unlock(mysys_var->current_mutex); break; } my_sleep(1000000L / WAIT_FOR_KILL_TRY_TIMES); } } mysql_mutex_unlock(&mysys_var->mutex); } DBUG_VOID_RETURN; } /** Close the Vio associated this session. @remark LOCK_thd_data is taken due to the fact that the Vio might be disassociated concurrently. */ void THD::disconnect() { Vio *vio= NULL; mysql_mutex_lock(&LOCK_thd_data); killed= KILL_CONNECTION; #ifdef SIGNAL_WITH_VIO_CLOSE /* Since a active vio might might have not been set yet, in any case save a reference to avoid closing a inexistent one or closing the vio twice if there is a active one. */ vio= active_vio; close_active_vio(); #endif /* Disconnect even if a active vio is not associated. */ if (net.vio != vio) vio_close(net.vio); mysql_mutex_unlock(&LOCK_thd_data); } bool THD::notify_shared_lock(MDL_context_owner *ctx_in_use, bool needs_thr_lock_abort) { THD *in_use= ctx_in_use->get_thd(); bool signalled= FALSE; if ((in_use->system_thread & SYSTEM_THREAD_DELAYED_INSERT) && !in_use->killed) { in_use->killed= KILL_CONNECTION; mysql_mutex_lock(&in_use->mysys_var->mutex); if (in_use->mysys_var->current_cond) mysql_cond_broadcast(in_use->mysys_var->current_cond); mysql_mutex_unlock(&in_use->mysys_var->mutex); signalled= TRUE; } if (needs_thr_lock_abort) { mysql_mutex_lock(&in_use->LOCK_thd_data); for (TABLE *thd_table= in_use->open_tables; thd_table ; thd_table= thd_table->next) { /* Check for TABLE::needs_reopen() is needed since in some places we call handler::close() for table instance (and set TABLE::db_stat to 0) and do not remove such instances from the THD::open_tables for some time, during which other thread can see those instances (e.g. see partitioning code). */ if (!thd_table->needs_reopen()) { signalled|= mysql_lock_abort_for_thread(this, thd_table); if (this && WSREP(this) && wsrep_thd_is_BF(this, FALSE)) { WSREP_DEBUG("remove_table_from_cache: %llu", (unsigned long long) this->real_id); wsrep_abort_thd((void *)this, (void *)in_use, FALSE); } } } mysql_mutex_unlock(&in_use->LOCK_thd_data); } return signalled; } /* Get error number for killed state Note that the error message can't have any parameters. See thd::kill_message() */ int killed_errno(killed_state killed) { DBUG_ENTER("killed_errno"); DBUG_PRINT("enter", ("killed: %d", killed)); switch (killed) { case NOT_KILLED: case KILL_HARD_BIT: DBUG_RETURN(0); // Probably wrong usage case KILL_BAD_DATA: case KILL_BAD_DATA_HARD: case ABORT_QUERY_HARD: case ABORT_QUERY: DBUG_RETURN(0); // Not a real error case KILL_CONNECTION: case KILL_CONNECTION_HARD: case KILL_SYSTEM_THREAD: case KILL_SYSTEM_THREAD_HARD: DBUG_RETURN(ER_CONNECTION_KILLED); case KILL_QUERY: case KILL_QUERY_HARD: DBUG_RETURN(ER_QUERY_INTERRUPTED); case KILL_TIMEOUT: case KILL_TIMEOUT_HARD: DBUG_RETURN(ER_STATEMENT_TIMEOUT); case KILL_SERVER: case KILL_SERVER_HARD: DBUG_RETURN(ER_SERVER_SHUTDOWN); } DBUG_RETURN(0); // Keep compiler happy } /* Remember the location of thread info, the structure needed for sql_alloc() and the structure for the net buffer */ bool THD::store_globals() { /* Assert that thread_stack is initialized: it's necessary to be able to track stack overrun. */ DBUG_ASSERT(thread_stack); if (set_current_thd(this) || my_pthread_setspecific_ptr(THR_MALLOC, &mem_root)) return 1; /* mysys_var is concurrently readable by a killer thread. It is protected by LOCK_thd_data, it is not needed to lock while the pointer is changing from NULL not non-NULL. If the kill thread reads NULL it doesn't refer to anything, but if it is non-NULL we need to ensure that the thread doesn't proceed to assign another thread to have the mysys_var reference (which in fact refers to the worker threads local storage with key THR_KEY_mysys. */ mysys_var=my_thread_var; /* Let mysqld define the thread id (not mysys) This allows us to move THD to different threads if needed. */ mysys_var->id= thread_id; real_id= pthread_self(); // For debugging mysys_var->stack_ends_here= thread_stack + // for consistency, see libevent_thread_proc STACK_DIRECTION * (long)my_thread_stack_size; vio_set_thread_id(net.vio, real_id); /* We have to call thr_lock_info_init() again here as THD may have been created in another thread */ thr_lock_info_init(&lock_info); return 0; } /** Untie THD from current thread Used when using --thread-handling=pool-of-threads */ void THD::reset_globals() { mysql_mutex_lock(&LOCK_thd_data); mysys_var= 0; mysql_mutex_unlock(&LOCK_thd_data); /* Undocking the thread specific data. */ set_current_thd(0); my_pthread_setspecific_ptr(THR_MALLOC, NULL); } /* Cleanup after query. SYNOPSIS THD::cleanup_after_query() DESCRIPTION This function is used to reset thread data to its default state. NOTE This function is not suitable for setting thread data to some non-default values, as there is only one replication thread, so different master threads may overwrite data of each other on slave. */ void THD::cleanup_after_query() { DBUG_ENTER("THD::cleanup_after_query"); thd_progress_end(this); /* Reset rand_used so that detection of calls to rand() will save random seeds if needed by the slave. Do not reset rand_used if inside a stored function or trigger because only the call to these operations is logged. Thus only the calling statement needs to detect rand() calls made by its substatements. These substatements must not set rand_used to 0 because it would remove the detection of rand() by the calling statement. */ if (!in_sub_stmt) /* stored functions and triggers are a special case */ { /* Forget those values, for next binlogger: */ stmt_depends_on_first_successful_insert_id_in_prev_stmt= 0; auto_inc_intervals_in_cur_stmt_for_binlog.empty(); rand_used= 0; #ifndef EMBEDDED_LIBRARY /* Clean possible unused INSERT_ID events by current statement. is_update_query() is needed to ignore SET statements: Statements that don't update anything directly and don't used stored functions. This is mostly necessary to ignore statements in binlog between SET INSERT_ID and DML statement which is intended to consume its event (there can be other SET statements between them). */ if ((rgi_slave || rli_fake) && is_update_query(lex->sql_command)) auto_inc_intervals_forced.empty(); #endif } /* Forget the binlog stmt filter for the next query. There are some code paths that: - do not call THD::decide_logging_format() - do call THD::binlog_query(), making this reset necessary. */ reset_binlog_local_stmt_filter(); if (first_successful_insert_id_in_cur_stmt > 0) { /* set what LAST_INSERT_ID() will return */ first_successful_insert_id_in_prev_stmt= first_successful_insert_id_in_cur_stmt; first_successful_insert_id_in_cur_stmt= 0; substitute_null_with_insert_id= TRUE; } arg_of_last_insert_id_function= 0; /* Free Items that were created during this execution */ free_items(); /* Reset where. */ where= THD::DEFAULT_WHERE; /* reset table map for multi-table update */ table_map_for_update= 0; m_binlog_invoker= INVOKER_NONE; #ifdef WITH_WSREP if (TOTAL_ORDER == wsrep_exec_mode) { wsrep_exec_mode = LOCAL_STATE; } #endif /* WITH_WSREP */ #ifndef EMBEDDED_LIBRARY if (rgi_slave) rgi_slave->cleanup_after_query(); #endif DBUG_VOID_RETURN; } /* Convert a string to another character set SYNOPSIS convert_string() to Store new allocated string here to_cs New character set for allocated string from String to convert from_length Length of string to convert from_cs Original character set NOTES to will be 0-terminated to make it easy to pass to system funcs RETURN 0 ok 1 End of memory. In this case to->str will point to 0 and to->length will be 0. */ bool THD::convert_string(LEX_STRING *to, CHARSET_INFO *to_cs, const char *from, uint from_length, CHARSET_INFO *from_cs) { DBUG_ENTER("convert_string"); size_t new_length= to_cs->mbmaxlen * from_length; uint dummy_errors; if (!(to->str= (char*) alloc(new_length+1))) { to->length= 0; // Safety fix DBUG_RETURN(1); // EOM } to->length= copy_and_convert((char*) to->str, new_length, to_cs, from, from_length, from_cs, &dummy_errors); to->str[to->length]=0; // Safety DBUG_RETURN(0); } /* Convert string from source character set to target character set inplace. SYNOPSIS THD::convert_string DESCRIPTION Convert string using convert_buffer - buffer for character set conversion shared between all protocols. RETURN 0 ok !0 out of memory */ bool THD::convert_string(String *s, CHARSET_INFO *from_cs, CHARSET_INFO *to_cs) { uint dummy_errors; if (convert_buffer.copy(s->ptr(), s->length(), from_cs, to_cs, &dummy_errors)) return TRUE; /* If convert_buffer >> s copying is more efficient long term */ if (convert_buffer.alloced_length() >= convert_buffer.length() * 2 || !s->is_alloced()) { return s->copy(convert_buffer); } s->swap(convert_buffer); return FALSE; } /* Update some cache variables when character set changes */ void THD::update_charset() { uint32 not_used; charset_is_system_charset= !String::needs_conversion(0, variables.character_set_client, system_charset_info, ¬_used); charset_is_collation_connection= !String::needs_conversion(0, variables.character_set_client, variables.collation_connection, ¬_used); charset_is_character_set_filesystem= !String::needs_conversion(0, variables.character_set_client, variables.character_set_filesystem, ¬_used); } /* routings to adding tables to list of changed in transaction tables */ inline static void list_include(CHANGED_TABLE_LIST** prev, CHANGED_TABLE_LIST* curr, CHANGED_TABLE_LIST* new_table) { if (new_table) { *prev = new_table; (*prev)->next = curr; } } /* add table to list of changed in transaction tables */ void THD::add_changed_table(TABLE *table) { DBUG_ENTER("THD::add_changed_table(table)"); DBUG_ASSERT(in_multi_stmt_transaction_mode() && table->file->has_transactions()); add_changed_table(table->s->table_cache_key.str, (long) table->s->table_cache_key.length); DBUG_VOID_RETURN; } void THD::add_changed_table(const char *key, long key_length) { DBUG_ENTER("THD::add_changed_table(key)"); CHANGED_TABLE_LIST **prev_changed = &transaction.changed_tables; CHANGED_TABLE_LIST *curr = transaction.changed_tables; for (; curr; prev_changed = &(curr->next), curr = curr->next) { int cmp = (long)curr->key_length - (long)key_length; if (cmp < 0) { list_include(prev_changed, curr, changed_table_dup(key, key_length)); DBUG_PRINT("info", ("key_length: %ld %u", key_length, (*prev_changed)->key_length)); DBUG_VOID_RETURN; } else if (cmp == 0) { cmp = memcmp(curr->key, key, curr->key_length); if (cmp < 0) { list_include(prev_changed, curr, changed_table_dup(key, key_length)); DBUG_PRINT("info", ("key_length: %ld %u", key_length, (*prev_changed)->key_length)); DBUG_VOID_RETURN; } else if (cmp == 0) { DBUG_PRINT("info", ("already in list")); DBUG_VOID_RETURN; } } } *prev_changed = changed_table_dup(key, key_length); DBUG_PRINT("info", ("key_length: %ld %u", key_length, (*prev_changed)->key_length)); DBUG_VOID_RETURN; } CHANGED_TABLE_LIST* THD::changed_table_dup(const char *key, long key_length) { CHANGED_TABLE_LIST* new_table = (CHANGED_TABLE_LIST*) trans_alloc(ALIGN_SIZE(sizeof(CHANGED_TABLE_LIST))+ key_length + 1); if (!new_table) { my_error(EE_OUTOFMEMORY, MYF(ME_BELL+ME_FATALERROR), ALIGN_SIZE(sizeof(TABLE_LIST)) + key_length + 1); killed= KILL_CONNECTION; return 0; } new_table->key= ((char*)new_table)+ ALIGN_SIZE(sizeof(CHANGED_TABLE_LIST)); new_table->next = 0; new_table->key_length = key_length; ::memcpy(new_table->key, key, key_length); return new_table; } int THD::send_explain_fields(select_result *result, uint8 explain_flags, bool is_analyze) { List field_list; if (lex->explain_json) make_explain_json_field_list(field_list); else make_explain_field_list(field_list, explain_flags, is_analyze); result->prepare(field_list, NULL); return (result->send_result_set_metadata(field_list, Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF)); } void THD::make_explain_json_field_list(List &field_list) { Item *item= new Item_empty_string("EXPLAIN", 78, system_charset_info); field_list.push_back(item); } /* Populate the provided field_list with EXPLAIN output columns. this->lex->describe has the EXPLAIN flags The set/order of columns must be kept in sync with Explain_query::print_explain and co. */ void THD::make_explain_field_list(List &field_list, uint8 explain_flags, bool is_analyze) { Item *item; CHARSET_INFO *cs= system_charset_info; field_list.push_back(item= new Item_return_int("id",3, MYSQL_TYPE_LONGLONG)); item->maybe_null= 1; field_list.push_back(new Item_empty_string("select_type", 19, cs)); field_list.push_back(item= new Item_empty_string("table", NAME_CHAR_LEN, cs)); item->maybe_null= 1; if (explain_flags & DESCRIBE_PARTITIONS) { /* Maximum length of string that make_used_partitions_str() can produce */ item= new Item_empty_string("partitions", MAX_PARTITIONS * (1 + FN_LEN), cs); field_list.push_back(item); item->maybe_null= 1; } field_list.push_back(item= new Item_empty_string("type", 10, cs)); item->maybe_null= 1; field_list.push_back(item=new Item_empty_string("possible_keys", NAME_CHAR_LEN*MAX_KEY, cs)); item->maybe_null=1; field_list.push_back(item=new Item_empty_string("key", NAME_CHAR_LEN, cs)); item->maybe_null=1; field_list.push_back(item=new Item_empty_string("key_len", NAME_CHAR_LEN*MAX_KEY)); item->maybe_null=1; field_list.push_back(item=new Item_empty_string("ref", NAME_CHAR_LEN*MAX_REF_PARTS, cs)); item->maybe_null=1; field_list.push_back(item= new Item_return_int("rows", 10, MYSQL_TYPE_LONGLONG)); if (is_analyze) { field_list.push_back(item= new Item_return_int("r_rows", 10, MYSQL_TYPE_LONGLONG)); item->maybe_null=1; } if (is_analyze || (explain_flags & DESCRIBE_EXTENDED)) { field_list.push_back(item= new Item_float("filtered", 0.1234, 2, 4)); item->maybe_null=1; } if (is_analyze) { field_list.push_back(item= new Item_float("r_filtered", 0.1234, 2, 4)); item->maybe_null=1; } item->maybe_null= 1; field_list.push_back(new Item_empty_string("Extra", 255, cs)); } #ifdef SIGNAL_WITH_VIO_CLOSE void THD::close_active_vio() { DBUG_ENTER("close_active_vio"); mysql_mutex_assert_owner(&LOCK_thd_data); #ifndef EMBEDDED_LIBRARY if (active_vio) { vio_close(active_vio); active_vio = 0; } #endif DBUG_VOID_RETURN; } #endif struct Item_change_record: public ilink { Item **place; Item *old_value; /* Placement new was hidden by `new' in ilink (TODO: check): */ static void *operator new(size_t size, void *mem) { return mem; } static void operator delete(void *ptr, size_t size) {} static void operator delete(void *ptr, void *mem) { /* never called */ } }; /* Register an item tree tree transformation, performed by the query optimizer. We need a pointer to runtime_memroot because it may be != thd->mem_root (due to possible set_n_backup_active_arena called for thd). */ void THD::nocheck_register_item_tree_change(Item **place, Item *old_value, MEM_ROOT *runtime_memroot) { Item_change_record *change; /* Now we use one node per change, which adds some memory overhead, but still is rather fast as we use alloc_root for allocations. A list of item tree changes of an average query should be short. */ void *change_mem= alloc_root(runtime_memroot, sizeof(*change)); if (change_mem == 0) { /* OOM, thd->fatal_error() is called by the error handler of the memroot. Just return. */ return; } change= new (change_mem) Item_change_record; change->place= place; change->old_value= old_value; change_list.append(change); } /** Check and register item change if needed @param place place where we should assign new value @param new_value place of the new value @details Let C be a reference to an item that changed the reference A at the location (occurrence) L1 and this change has been registered. If C is substituted for reference A another location (occurrence) L2 that is to be registered as well than this change has to be consistent with the first change in order the procedure that rollback changes to substitute the same reference at both locations L1 and L2. */ void THD::check_and_register_item_tree_change(Item **place, Item **new_value, MEM_ROOT *runtime_memroot) { Item_change_record *change; I_List_iterator it(change_list); while ((change= it++)) { if (change->place == new_value) break; // we need only very first value } if (change) nocheck_register_item_tree_change(place, change->old_value, runtime_memroot); } void THD::rollback_item_tree_changes() { I_List_iterator it(change_list); Item_change_record *change; DBUG_ENTER("rollback_item_tree_changes"); while ((change= it++)) *change->place= change->old_value; /* We can forget about changes memory: it's allocated in runtime memroot */ change_list.empty(); DBUG_VOID_RETURN; } /***************************************************************************** ** Functions to provide a interface to select results *****************************************************************************/ select_result::select_result() { thd=current_thd; } void select_result::cleanup() { /* do nothing */ } bool select_result::check_simple_select() const { my_error(ER_SP_BAD_CURSOR_QUERY, MYF(0)); return TRUE; } static String default_line_term("\n",default_charset_info); static String default_escaped("\\",default_charset_info); static String default_field_term("\t",default_charset_info); static String default_enclosed_and_line_start("", default_charset_info); static String default_xml_row_term("", default_charset_info); sql_exchange::sql_exchange(char *name, bool flag, enum enum_filetype filetype_arg) :file_name(name), opt_enclosed(0), dumpfile(flag), skip_lines(0) { filetype= filetype_arg; field_term= &default_field_term; enclosed= line_start= &default_enclosed_and_line_start; line_term= filetype == FILETYPE_CSV ? &default_line_term : &default_xml_row_term; escaped= &default_escaped; cs= NULL; } bool sql_exchange::escaped_given(void) { return escaped != &default_escaped; } bool select_send::send_result_set_metadata(List &list, uint flags) { bool res; #ifdef WITH_WSREP if (WSREP(thd) && thd->wsrep_retry_query) { WSREP_DEBUG("skipping select metadata"); return FALSE; } #endif /* WITH_WSREP */ if (!(res= thd->protocol->send_result_set_metadata(&list, flags))) is_result_set_started= 1; return res; } void select_send::abort_result_set() { DBUG_ENTER("select_send::abort_result_set"); if (is_result_set_started && thd->spcont) { /* We're executing a stored procedure, have an open result set and an SQL exception condition. In this situation we must abort the current statement, silence the error and start executing the continue/exit handler if one is found. Before aborting the statement, let's end the open result set, as otherwise the client will hang due to the violation of the client/server protocol. */ thd->spcont->end_partial_result_set= TRUE; } DBUG_VOID_RETURN; } /** Cleanup an instance of this class for re-use at next execution of a prepared statement/ stored procedure statement. */ void select_send::cleanup() { is_result_set_started= FALSE; } /* Send data to client. Returns 0 if ok */ int select_send::send_data(List &items) { Protocol *protocol= thd->protocol; DBUG_ENTER("select_send::send_data"); /* unit is not set when using 'delete ... returning' */ if (unit && unit->offset_limit_cnt) { // using limit offset,count unit->offset_limit_cnt--; DBUG_RETURN(FALSE); } if (thd->killed == ABORT_QUERY) DBUG_RETURN(FALSE); /* We may be passing the control from mysqld to the client: release the InnoDB adaptive hash S-latch to avoid thread deadlocks if it was reserved by thd */ ha_release_temporary_latches(thd); protocol->prepare_for_resend(); if (protocol->send_result_set_row(&items)) { protocol->remove_last_row(); DBUG_RETURN(TRUE); } thd->inc_sent_row_count(1); if (thd->vio_ok()) DBUG_RETURN(protocol->write()); DBUG_RETURN(0); } bool select_send::send_eof() { /* We may be passing the control from mysqld to the client: release the InnoDB adaptive hash S-latch to avoid thread deadlocks if it was reserved by thd */ ha_release_temporary_latches(thd); /* Don't send EOF if we're in error condition (which implies we've already sent or are sending an error) */ if (thd->is_error()) return TRUE; ::my_eof(thd); is_result_set_started= 0; return FALSE; } /************************************************************************ Handling writing to file ************************************************************************/ bool select_to_file::send_eof() { int error= MY_TEST(end_io_cache(&cache)); if (mysql_file_close(file, MYF(MY_WME)) || thd->is_error()) error= true; if (!error && !suppress_my_ok) { ::my_ok(thd,row_count); } file= -1; return error; } void select_to_file::cleanup() { /* In case of error send_eof() may be not called: close the file here. */ if (file >= 0) { (void) end_io_cache(&cache); mysql_file_close(file, MYF(0)); file= -1; } path[0]= '\0'; row_count= 0; } select_to_file::~select_to_file() { if (file >= 0) { // This only happens in case of error (void) end_io_cache(&cache); mysql_file_close(file, MYF(0)); file= -1; } } /*************************************************************************** ** Export of select to textfile ***************************************************************************/ select_export::~select_export() { thd->set_sent_row_count(row_count); } /* Create file with IO cache SYNOPSIS create_file() thd Thread handle path File name exchange Excange class cache IO cache RETURN >= 0 File handle -1 Error */ static File create_file(THD *thd, char *path, sql_exchange *exchange, IO_CACHE *cache) { File file; uint option= MY_UNPACK_FILENAME | MY_RELATIVE_PATH; #ifdef DONT_ALLOW_FULL_LOAD_DATA_PATHS option|= MY_REPLACE_DIR; // Force use of db directory #endif if (!dirname_length(exchange->file_name)) { strxnmov(path, FN_REFLEN-1, mysql_real_data_home, thd->db ? thd->db : "", NullS); (void) fn_format(path, exchange->file_name, path, "", option); } else (void) fn_format(path, exchange->file_name, mysql_real_data_home, "", option); if (!is_secure_file_path(path)) { /* Write only allowed to dir or subdir specified by secure_file_priv */ my_error(ER_OPTION_PREVENTS_STATEMENT, MYF(0), "--secure-file-priv"); return -1; } if (!access(path, F_OK)) { my_error(ER_FILE_EXISTS_ERROR, MYF(0), exchange->file_name); return -1; } /* Create the file world readable */ if ((file= mysql_file_create(key_select_to_file, path, 0666, O_WRONLY|O_EXCL, MYF(MY_WME))) < 0) return file; #ifdef HAVE_FCHMOD (void) fchmod(file, 0666); // Because of umask() #else (void) chmod(path, 0666); #endif if (init_io_cache(cache, file, 0L, WRITE_CACHE, 0L, 1, MYF(MY_WME))) { mysql_file_close(file, MYF(0)); /* Delete file on error, it was just created */ mysql_file_delete(key_select_to_file, path, MYF(0)); return -1; } return file; } int select_export::prepare(List &list, SELECT_LEX_UNIT *u) { bool blob_flag=0; bool string_results= FALSE, non_string_results= FALSE; unit= u; if ((uint) strlen(exchange->file_name) + NAME_LEN >= FN_REFLEN) strmake_buf(path,exchange->file_name); write_cs= exchange->cs ? exchange->cs : &my_charset_bin; if ((file= create_file(thd, path, exchange, &cache)) < 0) return 1; /* Check if there is any blobs in data */ { List_iterator_fast li(list); Item *item; while ((item=li++)) { if (item->max_length >= MAX_BLOB_WIDTH) { blob_flag=1; break; } if (item->result_type() == STRING_RESULT) string_results= TRUE; else non_string_results= TRUE; } } if (exchange->escaped->numchars() > 1 || exchange->enclosed->numchars() > 1) { my_error(ER_WRONG_FIELD_TERMINATORS, MYF(0)); return TRUE; } if (exchange->escaped->length() > 1 || exchange->enclosed->length() > 1 || !my_isascii(exchange->escaped->ptr()[0]) || !my_isascii(exchange->enclosed->ptr()[0]) || !exchange->field_term->is_ascii() || !exchange->line_term->is_ascii() || !exchange->line_start->is_ascii()) { /* Current LOAD DATA INFILE recognizes field/line separators "as is" without converting from client charset to data file charset. So, it is supposed, that input file of LOAD DATA INFILE consists of data in one charset and separators in other charset. For the compatibility with that [buggy] behaviour SELECT INTO OUTFILE implementation has been saved "as is" too, but the new warning message has been added: Non-ASCII separator arguments are not fully supported */ push_warning(thd, Sql_condition::WARN_LEVEL_WARN, WARN_NON_ASCII_SEPARATOR_NOT_IMPLEMENTED, ER(WARN_NON_ASCII_SEPARATOR_NOT_IMPLEMENTED)); } field_term_length=exchange->field_term->length(); field_term_char= field_term_length ? (int) (uchar) (*exchange->field_term)[0] : INT_MAX; if (!exchange->line_term->length()) exchange->line_term=exchange->field_term; // Use this if it exists field_sep_char= (exchange->enclosed->length() ? (int) (uchar) (*exchange->enclosed)[0] : field_term_char); if (exchange->escaped->length() && (exchange->escaped_given() || !(thd->variables.sql_mode & MODE_NO_BACKSLASH_ESCAPES))) escape_char= (int) (uchar) (*exchange->escaped)[0]; else escape_char= -1; is_ambiguous_field_sep= MY_TEST(strchr(ESCAPE_CHARS, field_sep_char)); is_unsafe_field_sep= MY_TEST(strchr(NUMERIC_CHARS, field_sep_char)); line_sep_char= (exchange->line_term->length() ? (int) (uchar) (*exchange->line_term)[0] : INT_MAX); if (!field_term_length) exchange->opt_enclosed=0; if (!exchange->enclosed->length()) exchange->opt_enclosed=1; // A little quicker loop fixed_row_size= (!field_term_length && !exchange->enclosed->length() && !blob_flag); if ((is_ambiguous_field_sep && exchange->enclosed->is_empty() && (string_results || is_unsafe_field_sep)) || (exchange->opt_enclosed && non_string_results && field_term_length && strchr(NUMERIC_CHARS, field_term_char))) { push_warning(thd, Sql_condition::WARN_LEVEL_WARN, ER_AMBIGUOUS_FIELD_TERM, ER(ER_AMBIGUOUS_FIELD_TERM)); is_ambiguous_field_term= TRUE; } else is_ambiguous_field_term= FALSE; return 0; } #define NEED_ESCAPING(x) ((int) (uchar) (x) == escape_char || \ (enclosed ? (int) (uchar) (x) == field_sep_char \ : (int) (uchar) (x) == field_term_char) || \ (int) (uchar) (x) == line_sep_char || \ !(x)) int select_export::send_data(List &items) { DBUG_ENTER("select_export::send_data"); char buff[MAX_FIELD_WIDTH],null_buff[2],space[MAX_FIELD_WIDTH]; char cvt_buff[MAX_FIELD_WIDTH]; String cvt_str(cvt_buff, sizeof(cvt_buff), write_cs); bool space_inited=0; String tmp(buff,sizeof(buff),&my_charset_bin),*res; tmp.length(0); if (unit->offset_limit_cnt) { // using limit offset,count unit->offset_limit_cnt--; DBUG_RETURN(0); } if (thd->killed == ABORT_QUERY) DBUG_RETURN(0); row_count++; Item *item; uint used_length=0,items_left=items.elements; List_iterator_fast li(items); if (my_b_write(&cache,(uchar*) exchange->line_start->ptr(), exchange->line_start->length())) goto err; while ((item=li++)) { Item_result result_type=item->result_type(); bool enclosed = (exchange->enclosed->length() && (!exchange->opt_enclosed || result_type == STRING_RESULT)); res=item->str_result(&tmp); if (res && !my_charset_same(write_cs, res->charset()) && !my_charset_same(write_cs, &my_charset_bin)) { const char *well_formed_error_pos; const char *cannot_convert_error_pos; const char *from_end_pos; const char *error_pos; uint32 bytes; uint64 estimated_bytes= ((uint64) res->length() / res->charset()->mbminlen + 1) * write_cs->mbmaxlen + 1; set_if_smaller(estimated_bytes, UINT_MAX32); if (cvt_str.realloc((uint32) estimated_bytes)) { my_error(ER_OUTOFMEMORY, MYF(ME_FATALERROR), (uint32) estimated_bytes); goto err; } bytes= well_formed_copy_nchars(write_cs, (char *) cvt_str.ptr(), cvt_str.alloced_length(), res->charset(), res->ptr(), res->length(), UINT_MAX32, // copy all input chars, // i.e. ignore nchars parameter &well_formed_error_pos, &cannot_convert_error_pos, &from_end_pos); error_pos= well_formed_error_pos ? well_formed_error_pos : cannot_convert_error_pos; if (error_pos) { char printable_buff[32]; convert_to_printable(printable_buff, sizeof(printable_buff), error_pos, res->ptr() + res->length() - error_pos, res->charset(), 6); push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, ER_TRUNCATED_WRONG_VALUE_FOR_FIELD, ER(ER_TRUNCATED_WRONG_VALUE_FOR_FIELD), "string", printable_buff, item->name, static_cast(row_count)); } else if (from_end_pos < res->ptr() + res->length()) { /* result is longer than UINT_MAX32 and doesn't fit into String */ push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, WARN_DATA_TRUNCATED, ER(WARN_DATA_TRUNCATED), item->full_name(), static_cast(row_count)); } cvt_str.length(bytes); res= &cvt_str; } if (res && enclosed) { if (my_b_write(&cache,(uchar*) exchange->enclosed->ptr(), exchange->enclosed->length())) goto err; } if (!res) { // NULL if (!fixed_row_size) { if (escape_char != -1) // Use \N syntax { null_buff[0]=escape_char; null_buff[1]='N'; if (my_b_write(&cache,(uchar*) null_buff,2)) goto err; } else if (my_b_write(&cache,(uchar*) "NULL",4)) goto err; } else { used_length=0; // Fill with space } } else { if (fixed_row_size) used_length=MY_MIN(res->length(),item->max_length); else used_length=res->length(); if ((result_type == STRING_RESULT || is_unsafe_field_sep) && escape_char != -1) { char *pos, *start, *end; CHARSET_INFO *res_charset= res->charset(); CHARSET_INFO *character_set_client= thd->variables. character_set_client; bool check_second_byte= (res_charset == &my_charset_bin) && character_set_client-> escape_with_backslash_is_dangerous; DBUG_ASSERT(character_set_client->mbmaxlen == 2 || !character_set_client->escape_with_backslash_is_dangerous); for (start=pos=(char*) res->ptr(),end=pos+used_length ; pos != end ; pos++) { #ifdef USE_MB if (use_mb(res_charset)) { int l; if ((l=my_ismbchar(res_charset, pos, end))) { pos += l-1; continue; } } #endif /* Special case when dumping BINARY/VARBINARY/BLOB values for the clients with character sets big5, cp932, gbk and sjis, which can have the escape character (0x5C "\" by default) as the second byte of a multi-byte sequence. If - pos[0] is a valid multi-byte head (e.g 0xEE) and - pos[1] is 0x00, which will be escaped as "\0", then we'll get "0xEE + 0x5C + 0x30" in the output file. If this file is later loaded using this sequence of commands: mysql> create table t1 (a varchar(128)) character set big5; mysql> LOAD DATA INFILE 'dump.txt' INTO TABLE t1; then 0x5C will be misinterpreted as the second byte of a multi-byte character "0xEE + 0x5C", instead of escape character for 0x00. To avoid this confusion, we'll escape the multi-byte head character too, so the sequence "0xEE + 0x00" will be dumped as "0x5C + 0xEE + 0x5C + 0x30". Note, in the condition below we only check if mbcharlen is equal to 2, because there are no character sets with mbmaxlen longer than 2 and with escape_with_backslash_is_dangerous set. DBUG_ASSERT before the loop makes that sure. */ if ((NEED_ESCAPING(*pos) || (check_second_byte && my_mbcharlen(character_set_client, (uchar) *pos) == 2 && pos + 1 < end && NEED_ESCAPING(pos[1]))) && /* Don't escape field_term_char by doubling - doubling is only valid for ENCLOSED BY characters: */ (enclosed || !is_ambiguous_field_term || (int) (uchar) *pos != field_term_char)) { char tmp_buff[2]; tmp_buff[0]= ((int) (uchar) *pos == field_sep_char && is_ambiguous_field_sep) ? field_sep_char : escape_char; tmp_buff[1]= *pos ? *pos : '0'; if (my_b_write(&cache,(uchar*) start,(uint) (pos-start)) || my_b_write(&cache,(uchar*) tmp_buff,2)) goto err; start=pos+1; } } if (my_b_write(&cache,(uchar*) start,(uint) (pos-start))) goto err; } else if (my_b_write(&cache,(uchar*) res->ptr(),used_length)) goto err; } if (fixed_row_size) { // Fill with space if (item->max_length > used_length) { if (!space_inited) { space_inited=1; bfill(space,sizeof(space),' '); } uint length=item->max_length-used_length; for (; length > sizeof(space) ; length-=sizeof(space)) { if (my_b_write(&cache,(uchar*) space,sizeof(space))) goto err; } if (my_b_write(&cache,(uchar*) space,length)) goto err; } } if (res && enclosed) { if (my_b_write(&cache, (uchar*) exchange->enclosed->ptr(), exchange->enclosed->length())) goto err; } if (--items_left) { if (my_b_write(&cache, (uchar*) exchange->field_term->ptr(), field_term_length)) goto err; } } if (my_b_write(&cache,(uchar*) exchange->line_term->ptr(), exchange->line_term->length())) goto err; DBUG_RETURN(0); err: DBUG_RETURN(1); } /*************************************************************************** ** Dump of select to a binary file ***************************************************************************/ int select_dump::prepare(List &list __attribute__((unused)), SELECT_LEX_UNIT *u) { unit= u; return (int) ((file= create_file(thd, path, exchange, &cache)) < 0); } int select_dump::send_data(List &items) { List_iterator_fast li(items); char buff[MAX_FIELD_WIDTH]; String tmp(buff,sizeof(buff),&my_charset_bin),*res; tmp.length(0); Item *item; DBUG_ENTER("select_dump::send_data"); if (unit->offset_limit_cnt) { // using limit offset,count unit->offset_limit_cnt--; DBUG_RETURN(0); } if (thd->killed == ABORT_QUERY) DBUG_RETURN(0); if (row_count++ > 1) { my_message(ER_TOO_MANY_ROWS, ER(ER_TOO_MANY_ROWS), MYF(0)); goto err; } while ((item=li++)) { res=item->str_result(&tmp); if (!res) // If NULL { if (my_b_write(&cache,(uchar*) "",1)) goto err; } else if (my_b_write(&cache,(uchar*) res->ptr(),res->length())) { my_error(ER_ERROR_ON_WRITE, MYF(0), path, my_errno); goto err; } } DBUG_RETURN(0); err: DBUG_RETURN(1); } select_subselect::select_subselect(Item_subselect *item_arg) { item= item_arg; } int select_singlerow_subselect::send_data(List &items) { DBUG_ENTER("select_singlerow_subselect::send_data"); Item_singlerow_subselect *it= (Item_singlerow_subselect *)item; if (it->assigned()) { my_message(ER_SUBQUERY_NO_1_ROW, ER(ER_SUBQUERY_NO_1_ROW), MYF(current_thd->lex->ignore ? ME_JUST_WARNING : 0)); DBUG_RETURN(1); } if (unit->offset_limit_cnt) { // Using limit offset,count unit->offset_limit_cnt--; DBUG_RETURN(0); } if (thd->killed == ABORT_QUERY) DBUG_RETURN(0); List_iterator_fast li(items); Item *val_item; for (uint i= 0; (val_item= li++); i++) it->store(i, val_item); it->assigned(1); DBUG_RETURN(0); } void select_max_min_finder_subselect::cleanup() { DBUG_ENTER("select_max_min_finder_subselect::cleanup"); cache= 0; DBUG_VOID_RETURN; } int select_max_min_finder_subselect::send_data(List &items) { DBUG_ENTER("select_max_min_finder_subselect::send_data"); Item_maxmin_subselect *it= (Item_maxmin_subselect *)item; List_iterator_fast li(items); Item *val_item= li++; it->register_value(); if (it->assigned()) { cache->store(val_item); if ((this->*op)()) it->store(0, cache); } else { if (!cache) { cache= Item_cache::get_cache(val_item); switch (val_item->result_type()) { case REAL_RESULT: op= &select_max_min_finder_subselect::cmp_real; break; case INT_RESULT: op= &select_max_min_finder_subselect::cmp_int; break; case STRING_RESULT: op= &select_max_min_finder_subselect::cmp_str; break; case DECIMAL_RESULT: op= &select_max_min_finder_subselect::cmp_decimal; break; case ROW_RESULT: case TIME_RESULT: case IMPOSSIBLE_RESULT: // This case should never be choosen DBUG_ASSERT(0); op= 0; } } cache->store(val_item); it->store(0, cache); } it->assigned(1); DBUG_RETURN(0); } bool select_max_min_finder_subselect::cmp_real() { Item *maxmin= ((Item_singlerow_subselect *)item)->element_index(0); double val1= cache->val_real(), val2= maxmin->val_real(); /* Ignore NULLs for ANY and keep them for ALL subqueries */ if (cache->null_value) return (is_all && !maxmin->null_value) || (!is_all && maxmin->null_value); if (maxmin->null_value) return !is_all; if (fmax) return(val1 > val2); return (val1 < val2); } bool select_max_min_finder_subselect::cmp_int() { Item *maxmin= ((Item_singlerow_subselect *)item)->element_index(0); longlong val1= cache->val_int(), val2= maxmin->val_int(); /* Ignore NULLs for ANY and keep them for ALL subqueries */ if (cache->null_value) return (is_all && !maxmin->null_value) || (!is_all && maxmin->null_value); if (maxmin->null_value) return !is_all; if (fmax) return(val1 > val2); return (val1 < val2); } bool select_max_min_finder_subselect::cmp_decimal() { Item *maxmin= ((Item_singlerow_subselect *)item)->element_index(0); my_decimal cval, *cvalue= cache->val_decimal(&cval); my_decimal mval, *mvalue= maxmin->val_decimal(&mval); /* Ignore NULLs for ANY and keep them for ALL subqueries */ if (cache->null_value) return (is_all && !maxmin->null_value) || (!is_all && maxmin->null_value); if (maxmin->null_value) return !is_all; if (fmax) return (my_decimal_cmp(cvalue, mvalue) > 0) ; return (my_decimal_cmp(cvalue,mvalue) < 0); } bool select_max_min_finder_subselect::cmp_str() { String *val1, *val2, buf1, buf2; Item *maxmin= ((Item_singlerow_subselect *)item)->element_index(0); /* as far as both operand is Item_cache buf1 & buf2 will not be used, but added for safety */ val1= cache->val_str(&buf1); val2= maxmin->val_str(&buf1); /* Ignore NULLs for ANY and keep them for ALL subqueries */ if (cache->null_value) return (is_all && !maxmin->null_value) || (!is_all && maxmin->null_value); if (maxmin->null_value) return !is_all; if (fmax) return (sortcmp(val1, val2, cache->collation.collation) > 0) ; return (sortcmp(val1, val2, cache->collation.collation) < 0); } int select_exists_subselect::send_data(List &items) { DBUG_ENTER("select_exists_subselect::send_data"); Item_exists_subselect *it= (Item_exists_subselect *)item; if (unit->offset_limit_cnt) { // Using limit offset,count unit->offset_limit_cnt--; DBUG_RETURN(0); } if (thd->killed == ABORT_QUERY) DBUG_RETURN(0); it->value= 1; it->assigned(1); DBUG_RETURN(0); } /*************************************************************************** Dump of select to variables ***************************************************************************/ int select_dumpvar::prepare(List &list, SELECT_LEX_UNIT *u) { unit= u; if (var_list.elements != list.elements) { my_message(ER_WRONG_NUMBER_OF_COLUMNS_IN_SELECT, ER(ER_WRONG_NUMBER_OF_COLUMNS_IN_SELECT), MYF(0)); return 1; } return 0; } bool select_dumpvar::check_simple_select() const { my_error(ER_SP_BAD_CURSOR_SELECT, MYF(0)); return TRUE; } void select_dumpvar::cleanup() { row_count= 0; } Query_arena::Type Query_arena::type() const { DBUG_ASSERT(0); /* Should never be called */ return STATEMENT; } void Query_arena::free_items() { Item *next; DBUG_ENTER("Query_arena::free_items"); /* This works because items are allocated with sql_alloc() */ for (; free_list; free_list= next) { next= free_list->next; DBUG_ASSERT(free_list != next); DBUG_PRINT("info", ("free item: 0x%lx", (ulong) free_list)); free_list->delete_self(); } /* Postcondition: free_list is 0 */ DBUG_VOID_RETURN; } void Query_arena::set_query_arena(Query_arena *set) { mem_root= set->mem_root; free_list= set->free_list; state= set->state; } void Query_arena::cleanup_stmt() { DBUG_ASSERT(! "Query_arena::cleanup_stmt() not implemented"); } /* Statement functions */ Statement::Statement(LEX *lex_arg, MEM_ROOT *mem_root_arg, enum enum_state state_arg, ulong id_arg) :Query_arena(mem_root_arg, state_arg), id(id_arg), mark_used_columns(MARK_COLUMNS_READ), lex(lex_arg), db(NULL), db_length(0) { name.str= NULL; } Query_arena::Type Statement::type() const { return STATEMENT; } void Statement::set_statement(Statement *stmt) { id= stmt->id; mark_used_columns= stmt->mark_used_columns; lex= stmt->lex; query_string= stmt->query_string; } void Statement::set_n_backup_statement(Statement *stmt, Statement *backup) { DBUG_ENTER("Statement::set_n_backup_statement"); backup->set_statement(this); set_statement(stmt); DBUG_VOID_RETURN; } void Statement::restore_backup_statement(Statement *stmt, Statement *backup) { DBUG_ENTER("Statement::restore_backup_statement"); stmt->set_statement(this); set_statement(backup); DBUG_VOID_RETURN; } void THD::end_statement() { DBUG_ENTER("THD::end_statement"); /* Cleanup SQL processing state to reuse this statement in next query. */ lex_end(lex); delete lex->result; lex->result= 0; /* Note that free_list is freed in cleanup_after_query() */ /* Don't free mem_root, as mem_root is freed in the end of dispatch_command (once for any command). */ DBUG_VOID_RETURN; } /* Start using arena specified by @set. Current arena data will be saved to *backup. */ void THD::set_n_backup_active_arena(Query_arena *set, Query_arena *backup) { DBUG_ENTER("THD::set_n_backup_active_arena"); DBUG_ASSERT(backup->is_backup_arena == FALSE); backup->set_query_arena(this); set_query_arena(set); #ifndef DBUG_OFF backup->is_backup_arena= TRUE; #endif DBUG_VOID_RETURN; } /* Stop using the temporary arena, and start again using the arena that is specified in *backup. The temporary arena is returned back into *set. */ void THD::restore_active_arena(Query_arena *set, Query_arena *backup) { DBUG_ENTER("THD::restore_active_arena"); DBUG_ASSERT(backup->is_backup_arena); set->set_query_arena(this); set_query_arena(backup); #ifndef DBUG_OFF backup->is_backup_arena= FALSE; #endif DBUG_VOID_RETURN; } Statement::~Statement() { } C_MODE_START static uchar * get_statement_id_as_hash_key(const uchar *record, size_t *key_length, my_bool not_used __attribute__((unused))) { const Statement *statement= (const Statement *) record; *key_length= sizeof(statement->id); return (uchar *) &((const Statement *) statement)->id; } static void delete_statement_as_hash_key(void *key) { delete (Statement *) key; } static uchar *get_stmt_name_hash_key(Statement *entry, size_t *length, my_bool not_used __attribute__((unused))) { *length= entry->name.length; return (uchar*) entry->name.str; } C_MODE_END Statement_map::Statement_map() : last_found_statement(0) { enum { START_STMT_HASH_SIZE = 16, START_NAME_HASH_SIZE = 16 }; my_hash_init(&st_hash, &my_charset_bin, START_STMT_HASH_SIZE, 0, 0, get_statement_id_as_hash_key, delete_statement_as_hash_key, MYF(0)); my_hash_init(&names_hash, system_charset_info, START_NAME_HASH_SIZE, 0, 0, (my_hash_get_key) get_stmt_name_hash_key, NULL,MYF(0)); } /* Insert a new statement to the thread-local statement map. DESCRIPTION If there was an old statement with the same name, replace it with the new one. Otherwise, check if max_prepared_stmt_count is not reached yet, increase prepared_stmt_count, and insert the new statement. It's okay to delete an old statement and fail to insert the new one. POSTCONDITIONS All named prepared statements are also present in names_hash. Statement names in names_hash are unique. The statement is added only if prepared_stmt_count < max_prepard_stmt_count last_found_statement always points to a valid statement or is 0 RETURN VALUE 0 success 1 error: out of resources or max_prepared_stmt_count limit has been reached. An error is sent to the client, the statement is deleted. */ int Statement_map::insert(THD *thd, Statement *statement) { if (my_hash_insert(&st_hash, (uchar*) statement)) { /* Delete is needed only in case of an insert failure. In all other cases hash_delete will also delete the statement. */ delete statement; my_error(ER_OUT_OF_RESOURCES, MYF(0)); goto err_st_hash; } if (statement->name.str && my_hash_insert(&names_hash, (uchar*) statement)) { my_error(ER_OUT_OF_RESOURCES, MYF(0)); goto err_names_hash; } mysql_mutex_lock(&LOCK_prepared_stmt_count); /* We don't check that prepared_stmt_count is <= max_prepared_stmt_count because we would like to allow to lower the total limit of prepared statements below the current count. In that case no new statements can be added until prepared_stmt_count drops below the limit. */ if (prepared_stmt_count >= max_prepared_stmt_count) { mysql_mutex_unlock(&LOCK_prepared_stmt_count); my_error(ER_MAX_PREPARED_STMT_COUNT_REACHED, MYF(0), max_prepared_stmt_count); goto err_max; } prepared_stmt_count++; mysql_mutex_unlock(&LOCK_prepared_stmt_count); last_found_statement= statement; return 0; err_max: if (statement->name.str) my_hash_delete(&names_hash, (uchar*) statement); err_names_hash: my_hash_delete(&st_hash, (uchar*) statement); err_st_hash: return 1; } void Statement_map::close_transient_cursors() { #ifdef TO_BE_IMPLEMENTED Statement *stmt; while ((stmt= transient_cursor_list.head())) stmt->close_cursor(); /* deletes itself from the list */ #endif } void Statement_map::erase(Statement *statement) { if (statement == last_found_statement) last_found_statement= 0; if (statement->name.str) my_hash_delete(&names_hash, (uchar *) statement); my_hash_delete(&st_hash, (uchar *) statement); mysql_mutex_lock(&LOCK_prepared_stmt_count); DBUG_ASSERT(prepared_stmt_count > 0); prepared_stmt_count--; mysql_mutex_unlock(&LOCK_prepared_stmt_count); } void Statement_map::reset() { /* Must be first, hash_free will reset st_hash.records */ mysql_mutex_lock(&LOCK_prepared_stmt_count); DBUG_ASSERT(prepared_stmt_count >= st_hash.records); prepared_stmt_count-= st_hash.records; mysql_mutex_unlock(&LOCK_prepared_stmt_count); my_hash_reset(&names_hash); my_hash_reset(&st_hash); last_found_statement= 0; } Statement_map::~Statement_map() { /* Must go first, hash_free will reset st_hash.records */ mysql_mutex_lock(&LOCK_prepared_stmt_count); DBUG_ASSERT(prepared_stmt_count >= st_hash.records); prepared_stmt_count-= st_hash.records; mysql_mutex_unlock(&LOCK_prepared_stmt_count); my_hash_free(&names_hash); my_hash_free(&st_hash); } bool my_var_user::set(THD *thd, Item *item) { Item_func_set_user_var *suv= new Item_func_set_user_var(name, item); suv->save_item_result(item); return suv->fix_fields(thd, 0) || suv->update(); } bool my_var_sp::set(THD *thd, Item *item) { return thd->spcont->set_variable(thd, offset, &item); } int select_dumpvar::send_data(List &items) { List_iterator_fast var_li(var_list); List_iterator it(items); Item *item; my_var *mv; DBUG_ENTER("select_dumpvar::send_data"); if (unit->offset_limit_cnt) { // using limit offset,count unit->offset_limit_cnt--; DBUG_RETURN(0); } if (row_count++) { my_message(ER_TOO_MANY_ROWS, ER(ER_TOO_MANY_ROWS), MYF(0)); DBUG_RETURN(1); } while ((mv= var_li++) && (item= it++)) { if (mv->set(thd, item)) DBUG_RETURN(1); } DBUG_RETURN(thd->is_error()); } bool select_dumpvar::send_eof() { if (! row_count) push_warning(thd, Sql_condition::WARN_LEVEL_WARN, ER_SP_FETCH_NO_DATA, ER(ER_SP_FETCH_NO_DATA)); /* Don't send EOF if we're in error condition (which implies we've already sent or are sending an error) */ if (thd->is_error()) return true; if (!suppress_my_ok) ::my_ok(thd,row_count); return 0; } bool select_materialize_with_stats:: create_result_table(THD *thd_arg, List *column_types, bool is_union_distinct, ulonglong options, const char *table_alias, bool bit_fields_as_long, bool create_table, bool keep_row_order) { DBUG_ASSERT(table == 0); tmp_table_param.field_count= column_types->elements; tmp_table_param.bit_fields_as_long= bit_fields_as_long; if (! (table= create_tmp_table(thd_arg, &tmp_table_param, *column_types, (ORDER*) 0, is_union_distinct, 1, options, HA_POS_ERROR, (char*) table_alias, keep_row_order))) return TRUE; col_stat= (Column_statistics*) table->in_use->alloc(table->s->fields * sizeof(Column_statistics)); if (!col_stat) return TRUE; reset(); table->file->extra(HA_EXTRA_WRITE_CACHE); table->file->extra(HA_EXTRA_IGNORE_DUP_KEY); return FALSE; } void select_materialize_with_stats::reset() { memset(col_stat, 0, table->s->fields * sizeof(Column_statistics)); max_nulls_in_row= 0; count_rows= 0; } void select_materialize_with_stats::cleanup() { reset(); select_union::cleanup(); } /** Override select_union::send_data to analyze each row for NULLs and to update null_statistics before sending data to the client. @return TRUE if fatal error when sending data to the client @return FALSE on success */ int select_materialize_with_stats::send_data(List &items) { List_iterator_fast item_it(items); Item *cur_item; Column_statistics *cur_col_stat= col_stat; uint nulls_in_row= 0; int res; if ((res= select_union::send_data(items))) return res; if (table->null_catch_flags & REJECT_ROW_DUE_TO_NULL_FIELDS) { table->null_catch_flags&= ~REJECT_ROW_DUE_TO_NULL_FIELDS; return 0; } /* Skip duplicate rows. */ if (write_err == HA_ERR_FOUND_DUPP_KEY || write_err == HA_ERR_FOUND_DUPP_UNIQUE) return 0; ++count_rows; while ((cur_item= item_it++)) { if (cur_item->is_null_result()) { ++cur_col_stat->null_count; cur_col_stat->max_null_row= count_rows; if (!cur_col_stat->min_null_row) cur_col_stat->min_null_row= count_rows; ++nulls_in_row; } ++cur_col_stat; } if (nulls_in_row > max_nulls_in_row) max_nulls_in_row= nulls_in_row; return 0; } /**************************************************************************** TMP_TABLE_PARAM ****************************************************************************/ void TMP_TABLE_PARAM::init() { DBUG_ENTER("TMP_TABLE_PARAM::init"); DBUG_PRINT("enter", ("this: 0x%lx", (ulong)this)); field_count= sum_func_count= func_count= hidden_field_count= 0; group_parts= group_length= group_null_parts= 0; quick_group= 1; table_charset= 0; precomputed_group_by= 0; bit_fields_as_long= 0; materialized_subquery= 0; force_not_null_cols= 0; skip_create_table= 0; DBUG_VOID_RETURN; } void thd_increment_bytes_sent(ulong length) { THD *thd=current_thd; if (likely(thd != 0)) { /* current_thd == 0 when close_connection() calls net_send_error() */ thd->status_var.bytes_sent+= length; } } void thd_increment_bytes_received(ulong length) { current_thd->status_var.bytes_received+= length; } void thd_increment_net_big_packet_count(ulong length) { current_thd->status_var.net_big_packet_count+= length; } void THD::set_status_var_init() { bzero((char*) &status_var, offsetof(STATUS_VAR, last_cleared_system_status_var)); } void Security_context::init() { host= user= ip= external_user= 0; host_or_ip= "connecting host"; priv_user[0]= priv_host[0]= proxy_user[0]= priv_role[0]= '\0'; master_access= 0; #ifndef NO_EMBEDDED_ACCESS_CHECKS db_access= NO_ACCESS; #endif } void Security_context::destroy() { DBUG_PRINT("info", ("freeing security context")); // If not pointer to constant if (host != my_localhost) { my_free(host); host= NULL; } if (user != delayed_user) { my_free(user); user= NULL; } if (external_user) { my_free(external_user); user= NULL; } my_free(ip); ip= NULL; } void Security_context::skip_grants() { /* privileges for the user are unknown everything is allowed */ host_or_ip= (char *)""; master_access= ~NO_ACCESS; *priv_user= *priv_host= '\0'; } bool Security_context::set_user(char *user_arg) { my_free(user); user= my_strdup(user_arg, MYF(0)); return user == 0; } #ifndef NO_EMBEDDED_ACCESS_CHECKS /** Initialize this security context from the passed in credentials and activate it in the current thread. @param thd @param definer_user @param definer_host @param db @param[out] backup Save a pointer to the current security context in the thread. In case of success it points to the saved old context, otherwise it points to NULL. During execution of a statement, multiple security contexts may be needed: - the security context of the authenticated user, used as the default security context for all top-level statements - in case of a view or a stored program, possibly the security context of the definer of the routine, if the object is defined with SQL SECURITY DEFINER option. The currently "active" security context is parameterized in THD member security_ctx. By default, after a connection is established, this member points at the "main" security context - the credentials of the authenticated user. Later, if we would like to execute some sub-statement or a part of a statement under credentials of a different user, e.g. definer of a procedure, we authenticate this user in a local instance of Security_context by means of this method (and ultimately by means of acl_getroot), and make the local instance active in the thread by re-setting thd->security_ctx pointer. Note, that the life cycle and memory management of the "main" and temporary security contexts are different. For the main security context, the memory for user/host/ip is allocated on system heap, and the THD class frees this memory in its destructor. The only case when contents of the main security context may change during its life time is when someone issued CHANGE USER command. Memory management of a "temporary" security context is responsibility of the module that creates it. @retval TRUE there is no user with the given credentials. The erro is reported in the thread. @retval FALSE success */ bool Security_context:: change_security_context(THD *thd, LEX_STRING *definer_user, LEX_STRING *definer_host, LEX_STRING *db, Security_context **backup) { bool needs_change; DBUG_ENTER("Security_context::change_security_context"); DBUG_ASSERT(definer_user->str && definer_host->str); *backup= NULL; needs_change= (strcmp(definer_user->str, thd->security_ctx->priv_user) || my_strcasecmp(system_charset_info, definer_host->str, thd->security_ctx->priv_host)); if (needs_change) { if (acl_getroot(this, definer_user->str, definer_host->str, definer_host->str, db->str)) { my_error(ER_NO_SUCH_USER, MYF(0), definer_user->str, definer_host->str); DBUG_RETURN(TRUE); } *backup= thd->security_ctx; thd->security_ctx= this; } DBUG_RETURN(FALSE); } void Security_context::restore_security_context(THD *thd, Security_context *backup) { if (backup) thd->security_ctx= backup; } #endif bool Security_context::user_matches(Security_context *them) { return ((user != NULL) && (them->user != NULL) && !strcmp(user, them->user)); } /**************************************************************************** Handling of open and locked tables states. This is used when we want to open/lock (and then close) some tables when we already have a set of tables open and locked. We use these methods for access to mysql.proc table to find definitions of stored routines. ****************************************************************************/ void THD::reset_n_backup_open_tables_state(Open_tables_backup *backup) { DBUG_ENTER("reset_n_backup_open_tables_state"); backup->set_open_tables_state(this); backup->mdl_system_tables_svp= mdl_context.mdl_savepoint(); reset_open_tables_state(this); state_flags|= Open_tables_state::BACKUPS_AVAIL; DBUG_VOID_RETURN; } void THD::restore_backup_open_tables_state(Open_tables_backup *backup) { DBUG_ENTER("restore_backup_open_tables_state"); mdl_context.rollback_to_savepoint(backup->mdl_system_tables_svp); /* Before we will throw away current open tables state we want to be sure that it was properly cleaned up. */ DBUG_ASSERT(open_tables == 0 && temporary_tables == 0 && derived_tables == 0 && lock == 0 && locked_tables_mode == LTM_NONE && m_reprepare_observer == NULL); set_open_tables_state(backup); DBUG_VOID_RETURN; } #if MARIA_PLUGIN_INTERFACE_VERSION < 0x0200 /** This is a backward compatibility method, made obsolete by the thd_kill_statement service. Keep it here to avoid breaking the ABI in case some binary plugins still use it. */ #undef thd_killed extern "C" int thd_killed(const MYSQL_THD thd) { return thd_kill_level(thd) > THD_ABORT_SOFTLY; } #else #error now thd_killed() function can go away #endif /* return thd->killed status to the client, mapped to the API enum thd_kill_levels values. */ extern "C" enum thd_kill_levels thd_kill_level(const MYSQL_THD thd) { THD* current= current_thd; if (!thd) thd= current; if (thd == current) { Apc_target *apc_target= (Apc_target*)&thd->apc_target; if (apc_target->have_apc_requests()) apc_target->process_apc_requests(); } if (likely(thd->killed == NOT_KILLED)) return THD_IS_NOT_KILLED; return thd->killed & KILL_HARD_BIT ? THD_ABORT_ASAP : THD_ABORT_SOFTLY; } /** Send an out-of-band progress report to the client The report is sent every 'thd->...progress_report_time' second, however not more often than global.progress_report_time. If global.progress_report_time is 0, then don't send progress reports, but check every second if the value has changed */ static void thd_send_progress(THD *thd) { /* Check if we should send the client a progress report */ ulonglong report_time= my_interval_timer(); if (report_time > thd->progress.next_report_time) { uint seconds_to_next= MY_MAX(thd->variables.progress_report_time, global_system_variables.progress_report_time); if (seconds_to_next == 0) // Turned off seconds_to_next= 1; // Check again after 1 second thd->progress.next_report_time= (report_time + seconds_to_next * 1000000000ULL); if (global_system_variables.progress_report_time && thd->variables.progress_report_time) net_send_progress_packet(thd); } } /** Initialize progress report handling **/ extern "C" void thd_progress_init(MYSQL_THD thd, uint max_stage) { DBUG_ASSERT(thd->stmt_arena != thd->progress.arena); if (thd->progress.arena) return; // already initialized /* Send progress reports to clients that supports it, if the command is a high level command (like ALTER TABLE) and we are not in a stored procedure */ thd->progress.report= ((thd->client_capabilities & CLIENT_PROGRESS) && thd->progress.report_to_client && !thd->in_sub_stmt); thd->progress.next_report_time= 0; thd->progress.stage= 0; thd->progress.counter= thd->progress.max_counter= 0; thd->progress.max_stage= max_stage; thd->progress.arena= thd->stmt_arena; } /* Inform processlist and the client that some progress has been made */ extern "C" void thd_progress_report(MYSQL_THD thd, ulonglong progress, ulonglong max_progress) { if (thd->stmt_arena != thd->progress.arena) return; if (thd->progress.max_counter != max_progress) // Simple optimization { mysql_mutex_lock(&thd->LOCK_thd_data); thd->progress.counter= progress; thd->progress.max_counter= max_progress; mysql_mutex_unlock(&thd->LOCK_thd_data); } else thd->progress.counter= progress; if (thd->progress.report) thd_send_progress(thd); } /** Move to next stage in process list handling This will reset the timer to ensure the progress is sent to the client if client progress reports are activated. */ extern "C" void thd_progress_next_stage(MYSQL_THD thd) { if (thd->stmt_arena != thd->progress.arena) return; mysql_mutex_lock(&thd->LOCK_thd_data); thd->progress.stage++; thd->progress.counter= 0; DBUG_ASSERT(thd->progress.stage < thd->progress.max_stage); mysql_mutex_unlock(&thd->LOCK_thd_data); if (thd->progress.report) { thd->progress.next_report_time= 0; // Send new stage info thd_send_progress(thd); } } /** Disable reporting of progress in process list. @note This function is safe to call even if one has not called thd_progress_init. This function should be called by all parts that does progress reporting to ensure that progress list doesn't contain 100 % done forever. */ extern "C" void thd_progress_end(MYSQL_THD thd) { if (thd->stmt_arena != thd->progress.arena) return; /* It's enough to reset max_counter to set disable progress indicator in processlist. */ thd->progress.max_counter= 0; thd->progress.arena= 0; } /** Return the thread id of a user thread @param thd user thread @return thread id */ extern "C" unsigned long thd_get_thread_id(const MYSQL_THD thd) { return((unsigned long)thd->thread_id); } /** Check if THD socket is still connected. */ extern "C" int thd_is_connected(MYSQL_THD thd) { return thd->is_connected(); } #ifdef INNODB_COMPATIBILITY_HOOKS extern "C" const struct charset_info_st *thd_charset(MYSQL_THD thd) { return(thd->charset()); } /** OBSOLETE : there's no way to ensure the string is null terminated. Use thd_query_string instead() */ extern "C" char **thd_query(MYSQL_THD thd) { return (&thd->query_string.string.str); } /** Get the current query string for the thread. @param The MySQL internal thread pointer @return query string and length. May be non-null-terminated. */ extern "C" LEX_STRING * thd_query_string (MYSQL_THD thd) { return(&thd->query_string.string); } extern "C" int thd_slave_thread(const MYSQL_THD thd) { return(thd->slave_thread); } /* Returns true for a worker thread in parallel replication. */ extern "C" int thd_rpl_is_parallel(const MYSQL_THD thd) { return thd->rgi_slave && thd->rgi_slave->is_parallel_exec; } /* This function can optionally be called to check if thd_report_wait_for() needs to be called for waits done by a given transaction. If this function returns false for a given thd, there is no need to do any calls to thd_report_wait_for() on that thd. This call is optional; it is safe to call thd_report_wait_for() in any case. This call can be used to save some redundant calls to thd_report_wait_for() if desired. (This is unlikely to matter much unless there are _lots_ of waits to report, as the overhead of thd_report_wait_for() is small). */ extern "C" int thd_need_wait_for(const MYSQL_THD thd) { rpl_group_info *rgi; if (mysql_bin_log.is_open()) return true; if (!thd) return false; rgi= thd->rgi_slave; if (!rgi) return false; return rgi->is_parallel_exec; } /* Used by InnoDB/XtraDB to report that one transaction THD is about to go to wait for a transactional lock held by another transactions OTHER_THD. This is used for parallel replication, where transactions are required to commit in the same order on the slave as they did on the master. If the transactions on the slave encounters lock conflicts on the slave that did not exist on the master, this can cause deadlocks. Normally, such conflicts will not occur, because the same conflict would have prevented the two transactions from committing in parallel on the master, thus preventing them from running in parallel on the slave in the first place. However, it is possible in case when the optimizer chooses a different plan on the slave than on the master (eg. table scan instead of index scan). InnoDB/XtraDB reports lock waits using this call. If a lock wait causes a deadlock with the pre-determined commit order, we kill the later transaction, and later re-try it, to resolve the deadlock. This call need only receive reports about waits for locks that will remain until the holding transaction commits. InnoDB/XtraDB auto-increment locks are released earlier, and so need not be reported. (Such false positives are not harmful, but could lead to unnecessary kill and retry, so best avoided). */ extern "C" void thd_report_wait_for(MYSQL_THD thd, MYSQL_THD other_thd) { rpl_group_info *rgi; rpl_group_info *other_rgi; if (!thd) return; DEBUG_SYNC(thd, "thd_report_wait_for"); thd->transaction.stmt.mark_trans_did_wait(); if (!other_thd) return; rgi= thd->rgi_slave; other_rgi= other_thd->rgi_slave; if (!rgi || !other_rgi) return; if (!rgi->is_parallel_exec) return; if (rgi->rli != other_rgi->rli) return; if (!rgi->gtid_sub_id || !other_rgi->gtid_sub_id) return; if (rgi->current_gtid.domain_id != other_rgi->current_gtid.domain_id) return; if (rgi->gtid_sub_id > other_rgi->gtid_sub_id) return; /* This transaction is about to wait for another transaction that is required by replication binlog order to commit after. This would cause a deadlock. So send a kill to the other transaction, with a temporary error; this will cause replication to rollback (and later re-try) the other transaction, releasing the lock for this transaction so replication can proceed. */ other_rgi->killed_for_retry= true; mysql_mutex_lock(&other_thd->LOCK_thd_data); other_thd->awake(KILL_CONNECTION); mysql_mutex_unlock(&other_thd->LOCK_thd_data); } /* This function is called from InnoDB/XtraDB to check if the commit order of two transactions has already been decided by the upper layer. This happens in parallel replication, where the commit order is forced to be the same on the slave as it was originally on the master. If this function returns false, it means that such commit order will be enforced. This allows the storage engine to optionally omit gap lock waits or similar measures that would otherwise be needed to ensure that transactions would be serialised in a way that would cause a commit order that is correct for binlogging for statement-based replication. Since transactions are only run in parallel on the slave if they ran without lock conflicts on the master, normally no lock conflicts on the slave happen during parallel replication. However, there are a couple of corner cases where it can happen, like these secondary-index operations: T1: INSERT INTO t1 VALUES (7, NULL); T2: DELETE FROM t1 WHERE b <= 3; T1: UPDATE t1 SET secondary=NULL WHERE primary=1 T2: DELETE t1 WHERE secondary <= 3 The DELETE takes a gap lock that can block the INSERT/UPDATE, but the row locks set by INSERT/UPDATE do not block the DELETE. Thus, the execution order of the transactions determine whether a lock conflict occurs or not. Thus a lock conflict can occur on the slave where it did not on the master. If this function returns true, normal locking should be done as required by the binlogging and transaction isolation level in effect. But if it returns false, the correct order will be enforced anyway, and InnoDB/XtraDB can avoid taking the gap lock, preventing the lock conflict. Calling this function is just an optimisation to avoid unnecessary deadlocks. If it was not used, a gap lock would be set that could eventually cause a deadlock; the deadlock would be caught by thd_report_wait_for() and the transaction T2 killed and rolled back (and later re-tried). */ extern "C" int thd_need_ordering_with(const MYSQL_THD thd, const MYSQL_THD other_thd) { rpl_group_info *rgi, *other_rgi; DBUG_EXECUTE_IF("disable_thd_need_ordering_with", return 1;); if (!thd || !other_thd) return 1; rgi= thd->rgi_slave; other_rgi= other_thd->rgi_slave; if (!rgi || !other_rgi) return 1; if (!rgi->is_parallel_exec) return 1; if (rgi->rli != other_rgi->rli) return 1; if (rgi->current_gtid.domain_id != other_rgi->current_gtid.domain_id) return 1; if (!rgi->commit_id || rgi->commit_id != other_rgi->commit_id) return 1; DBUG_EXECUTE_IF("thd_need_ordering_with_force", return 1;); /* Otherwise, these two threads are doing parallel replication within the same replication domain. Their commit order is already fixed, so we do not need gap locks or similar to otherwise enforce ordering (and in fact such locks could lead to unnecessary deadlocks and transaction retry). */ return 0; } /* If the storage engine detects a deadlock, and needs to choose a victim transaction to roll back, it can call this function to ask the upper server layer for which of two possible transactions is prefered to be aborted and rolled back. In parallel replication, if two transactions are running in parallel and one is fixed to commit before the other, then the one that commits later will be prefered as the victim - chosing the early transaction as a victim will not resolve the deadlock anyway, as the later transaction still needs to wait for the earlier to commit. Otherwise, a transaction that uses only transactional tables, and can thus be safely rolled back, will be prefered as a deadlock victim over a transaction that also modified non-transactional (eg. MyISAM) tables. The return value is -1 if the first transaction is prefered as a deadlock victim, 1 if the second transaction is prefered, or 0 for no preference (in which case the storage engine can make the choice as it prefers). */ extern "C" int thd_deadlock_victim_preference(const MYSQL_THD thd1, const MYSQL_THD thd2) { rpl_group_info *rgi1, *rgi2; bool nontrans1, nontrans2; if (!thd1 || !thd2) return 0; /* If the transactions are participating in the same replication domain in parallel replication, then request to select the one that will commit later (in the fixed commit order from the master) as the deadlock victim. */ rgi1= thd1->rgi_slave; rgi2= thd2->rgi_slave; if (rgi1 && rgi2 && rgi1->is_parallel_exec && rgi1->rli == rgi2->rli && rgi1->current_gtid.domain_id == rgi2->current_gtid.domain_id) return rgi1->gtid_sub_id < rgi2->gtid_sub_id ? 1 : -1; /* If one transaction has modified non-transactional tables (so that it cannot be safely rolled back), and the other has not, then prefer to select the purely transactional one as the victim. */ nontrans1= thd1->transaction.all.modified_non_trans_table; nontrans2= thd2->transaction.all.modified_non_trans_table; if (nontrans1 && !nontrans2) return 1; else if (!nontrans1 && nontrans2) return -1; /* No preferences, let the storage engine decide. */ return 0; } extern "C" int thd_non_transactional_update(const MYSQL_THD thd) { return(thd->transaction.all.modified_non_trans_table); } extern "C" int thd_binlog_format(const MYSQL_THD thd) { if (((WSREP(thd) && wsrep_emulate_bin_log) || mysql_bin_log.is_open()) && thd->variables.option_bits & OPTION_BIN_LOG) return (int) thd->wsrep_binlog_format(); else return BINLOG_FORMAT_UNSPEC; } extern "C" void thd_mark_transaction_to_rollback(MYSQL_THD thd, bool all) { mark_transaction_to_rollback(thd, all); } extern "C" bool thd_binlog_filter_ok(const MYSQL_THD thd) { return binlog_filter->db_ok(thd->db); } /* This is similar to sqlcom_can_generate_row_events, with the expection that we only return 1 if we are going to generate row events in a transaction. CREATE OR REPLACE is always safe to do as this will run in it's own transaction. */ extern "C" bool thd_sqlcom_can_generate_row_events(const MYSQL_THD thd) { return (sqlcom_can_generate_row_events(thd) && thd->lex->sql_command != SQLCOM_CREATE_TABLE); } extern "C" enum durability_properties thd_get_durability_property(const MYSQL_THD thd) { enum durability_properties ret= HA_REGULAR_DURABILITY; if (thd != NULL) ret= thd->durability_property; return ret; } /** Get the auto_increment_offset auto_increment_increment. Exposed by thd_autoinc_service. Needed by InnoDB. @param thd Thread object @param off auto_increment_offset @param inc auto_increment_increment */ extern "C" void thd_get_autoinc(const MYSQL_THD thd, ulong* off, ulong* inc) { *off = thd->variables.auto_increment_offset; *inc = thd->variables.auto_increment_increment; } /** Is strict sql_mode set. Needed by InnoDB. @param thd Thread object @return True if sql_mode has strict mode (all or trans). @retval true sql_mode has strict mode (all or trans). @retval false sql_mode has not strict mode (all or trans). */ extern "C" bool thd_is_strict_mode(const MYSQL_THD thd) { return thd->is_strict_mode(); } /* Interface for MySQL Server, plugins and storage engines to report when they are going to sleep/stall. SYNOPSIS thd_wait_begin() thd Thread object Can be NULL, in this case current THD is used. wait_type Type of wait 1 -- short wait (e.g. for mutex) 2 -- medium wait (e.g. for disk io) 3 -- large wait (e.g. for locked row/table) NOTES This is used by the threadpool to have better knowledge of which threads that currently are actively running on CPUs. When a thread reports that it's going to sleep/stall, the threadpool scheduler is free to start another thread in the pool most likely. The expected wait time is simply an indication of how long the wait is expected to become, the real wait time could be very different. thd_wait_end MUST be called immediately after waking up again. */ extern "C" void thd_wait_begin(MYSQL_THD thd, int wait_type) { if (!thd) { thd= current_thd; if (unlikely(!thd)) return; } MYSQL_CALLBACK(thd->scheduler, thd_wait_begin, (thd, wait_type)); } /** Interface for MySQL Server, plugins and storage engines to report when they waking up from a sleep/stall. @param thd Thread handle Can be NULL, in this case current THD is used. */ extern "C" void thd_wait_end(MYSQL_THD thd) { if (!thd) { thd= current_thd; if (unlikely(!thd)) return; } MYSQL_CALLBACK(thd->scheduler, thd_wait_end, (thd)); } #endif // INNODB_COMPATIBILITY_HOOKS */ /**************************************************************************** Handling of statement states in functions and triggers. This is used to ensure that the function/trigger gets a clean state to work with and does not cause any side effects of the calling statement. It also allows most stored functions and triggers to replicate even if they are used items that would normally be stored in the binary replication (like last_insert_id() etc...) The following things is done - Disable binary logging for the duration of the statement - Disable multi-result-sets for the duration of the statement - Value of last_insert_id() is saved and restored - Value set by 'SET INSERT_ID=#' is reset and restored - Value for found_rows() is reset and restored - examined_row_count is added to the total - cuted_fields is added to the total - new savepoint level is created and destroyed NOTES: Seed for random() is saved for the first! usage of RAND() We reset examined_row_count and cuted_fields and add these to the result to ensure that if we have a bug that would reset these within a function, we are not loosing any rows from the main statement. We do not reset value of last_insert_id(). ****************************************************************************/ void THD::reset_sub_statement_state(Sub_statement_state *backup, uint new_state) { #ifndef EMBEDDED_LIBRARY /* BUG#33029, if we are replicating from a buggy master, reset auto_inc_intervals_forced to prevent substatement (triggers/functions) from using erroneous INSERT_ID value */ if (rpl_master_erroneous_autoinc(this)) { DBUG_ASSERT(backup->auto_inc_intervals_forced.nb_elements() == 0); auto_inc_intervals_forced.swap(&backup->auto_inc_intervals_forced); } #endif backup->option_bits= variables.option_bits; backup->count_cuted_fields= count_cuted_fields; backup->in_sub_stmt= in_sub_stmt; backup->enable_slow_log= enable_slow_log; backup->query_plan_flags= query_plan_flags; backup->limit_found_rows= limit_found_rows; backup->examined_row_count= m_examined_row_count; backup->sent_row_count= m_sent_row_count; backup->cuted_fields= cuted_fields; backup->client_capabilities= client_capabilities; backup->savepoints= transaction.savepoints; backup->first_successful_insert_id_in_prev_stmt= first_successful_insert_id_in_prev_stmt; backup->first_successful_insert_id_in_cur_stmt= first_successful_insert_id_in_cur_stmt; if ((!lex->requires_prelocking() || is_update_query(lex->sql_command)) && !is_current_stmt_binlog_format_row()) { variables.option_bits&= ~OPTION_BIN_LOG; } if ((backup->option_bits & OPTION_BIN_LOG) && is_update_query(lex->sql_command) && !is_current_stmt_binlog_format_row()) mysql_bin_log.start_union_events(this, this->query_id); /* Disable result sets */ client_capabilities &= ~CLIENT_MULTI_RESULTS; in_sub_stmt|= new_state; m_examined_row_count= 0; m_sent_row_count= 0; cuted_fields= 0; transaction.savepoints= 0; first_successful_insert_id_in_cur_stmt= 0; } void THD::restore_sub_statement_state(Sub_statement_state *backup) { DBUG_ENTER("THD::restore_sub_statement_state"); #ifndef EMBEDDED_LIBRARY /* BUG#33029, if we are replicating from a buggy master, restore auto_inc_intervals_forced so that the top statement can use the INSERT_ID value set before this statement. */ if (rpl_master_erroneous_autoinc(this)) { backup->auto_inc_intervals_forced.swap(&auto_inc_intervals_forced); DBUG_ASSERT(backup->auto_inc_intervals_forced.nb_elements() == 0); } #endif /* To save resources we want to release savepoints which were created during execution of function or trigger before leaving their savepoint level. It is enough to release first savepoint set on this level since all later savepoints will be released automatically. */ if (transaction.savepoints) { SAVEPOINT *sv; for (sv= transaction.savepoints; sv->prev; sv= sv->prev) {} /* ha_release_savepoint() never returns error. */ (void)ha_release_savepoint(this, sv); } count_cuted_fields= backup->count_cuted_fields; transaction.savepoints= backup->savepoints; variables.option_bits= backup->option_bits; in_sub_stmt= backup->in_sub_stmt; enable_slow_log= backup->enable_slow_log; query_plan_flags= backup->query_plan_flags; first_successful_insert_id_in_prev_stmt= backup->first_successful_insert_id_in_prev_stmt; first_successful_insert_id_in_cur_stmt= backup->first_successful_insert_id_in_cur_stmt; limit_found_rows= backup->limit_found_rows; set_sent_row_count(backup->sent_row_count); client_capabilities= backup->client_capabilities; /* If we've left sub-statement mode, reset the fatal error flag. Otherwise keep the current value, to propagate it up the sub-statement stack. */ if (!in_sub_stmt) is_fatal_sub_stmt_error= FALSE; if ((variables.option_bits & OPTION_BIN_LOG) && is_update_query(lex->sql_command) && !is_current_stmt_binlog_format_row()) mysql_bin_log.stop_union_events(this); /* The following is added to the old values as we are interested in the total complexity of the query */ inc_examined_row_count(backup->examined_row_count); cuted_fields+= backup->cuted_fields; DBUG_VOID_RETURN; } void THD::set_statement(Statement *stmt) { mysql_mutex_lock(&LOCK_thd_data); Statement::set_statement(stmt); mysql_mutex_unlock(&LOCK_thd_data); } void THD::set_sent_row_count(ha_rows count) { m_sent_row_count= count; MYSQL_SET_STATEMENT_ROWS_SENT(m_statement_psi, m_sent_row_count); } void THD::set_examined_row_count(ha_rows count) { m_examined_row_count= count; MYSQL_SET_STATEMENT_ROWS_EXAMINED(m_statement_psi, m_examined_row_count); } void THD::inc_sent_row_count(ha_rows count) { m_sent_row_count+= count; MYSQL_SET_STATEMENT_ROWS_SENT(m_statement_psi, m_sent_row_count); } void THD::inc_examined_row_count(ha_rows count) { m_examined_row_count+= count; MYSQL_SET_STATEMENT_ROWS_EXAMINED(m_statement_psi, m_examined_row_count); } void THD::inc_status_created_tmp_disk_tables() { status_var_increment(status_var.created_tmp_disk_tables_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_created_tmp_disk_tables)(m_statement_psi, 1); #endif } void THD::inc_status_created_tmp_tables() { status_var_increment(status_var.created_tmp_tables_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_created_tmp_tables)(m_statement_psi, 1); #endif } void THD::inc_status_select_full_join() { status_var_increment(status_var.select_full_join_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_select_full_join)(m_statement_psi, 1); #endif } void THD::inc_status_select_full_range_join() { status_var_increment(status_var.select_full_range_join_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_select_full_range_join)(m_statement_psi, 1); #endif } void THD::inc_status_select_range() { status_var_increment(status_var.select_range_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_select_range)(m_statement_psi, 1); #endif } void THD::inc_status_select_range_check() { status_var_increment(status_var.select_range_check_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_select_range_check)(m_statement_psi, 1); #endif } void THD::inc_status_select_scan() { status_var_increment(status_var.select_scan_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_select_scan)(m_statement_psi, 1); #endif } void THD::inc_status_sort_merge_passes() { status_var_increment(status_var.filesort_merge_passes_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_sort_merge_passes)(m_statement_psi, 1); #endif } void THD::inc_status_sort_range() { status_var_increment(status_var.filesort_range_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_sort_range)(m_statement_psi, 1); #endif } void THD::inc_status_sort_rows(ha_rows count) { statistic_add(status_var.filesort_rows_, count, &LOCK_status); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_sort_rows)(m_statement_psi, count); #endif } void THD::inc_status_sort_scan() { status_var_increment(status_var.filesort_scan_count_); #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(inc_statement_sort_scan)(m_statement_psi, 1); #endif } void THD::set_status_no_index_used() { server_status|= SERVER_QUERY_NO_INDEX_USED; #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(set_statement_no_index_used)(m_statement_psi); #endif } void THD::set_status_no_good_index_used() { server_status|= SERVER_QUERY_NO_GOOD_INDEX_USED; #ifdef HAVE_PSI_STATEMENT_INTERFACE PSI_STATEMENT_CALL(set_statement_no_good_index_used)(m_statement_psi); #endif } void THD::set_command(enum enum_server_command command) { m_command= command; #ifdef HAVE_PSI_THREAD_INTERFACE PSI_STATEMENT_CALL(set_thread_command)(m_command); #endif } /** Assign a new value to thd->query. */ void THD::set_query(const CSET_STRING &string_arg) { mysql_mutex_lock(&LOCK_thd_data); set_query_inner(string_arg); mysql_mutex_unlock(&LOCK_thd_data); #ifdef HAVE_PSI_THREAD_INTERFACE PSI_THREAD_CALL(set_thread_info)(query(), query_length()); #endif } /** Assign a new value to thd->query and thd->query_id. */ void THD::set_query_and_id(char *query_arg, uint32 query_length_arg, CHARSET_INFO *cs, query_id_t new_query_id) { mysql_mutex_lock(&LOCK_thd_data); set_query_inner(query_arg, query_length_arg, cs); mysql_mutex_unlock(&LOCK_thd_data); query_id= new_query_id; } /** Assign a new value to thd->mysys_var. */ void THD::set_mysys_var(struct st_my_thread_var *new_mysys_var) { mysql_mutex_lock(&LOCK_thd_data); mysys_var= new_mysys_var; mysql_mutex_unlock(&LOCK_thd_data); } /** Leave explicit LOCK TABLES or prelocked mode and restore value of transaction sentinel in MDL subsystem. */ void THD::leave_locked_tables_mode() { if (locked_tables_mode == LTM_LOCK_TABLES) { /* When leaving LOCK TABLES mode we have to change the duration of most of the metadata locks being held, except for HANDLER and GRL locks, to transactional for them to be properly released at UNLOCK TABLES. */ mdl_context.set_transaction_duration_for_all_locks(); /* Make sure we don't release the global read lock and commit blocker when leaving LTM. */ global_read_lock.set_explicit_lock_duration(this); /* Also ensure that we don't release metadata locks for open HANDLERs. */ if (handler_tables_hash.records) mysql_ha_set_explicit_lock_duration(this); if (ull_hash.records) mysql_ull_set_explicit_lock_duration(this); } locked_tables_mode= LTM_NONE; } void THD::get_definer(LEX_USER *definer, bool role) { binlog_invoker(role); #if !defined(MYSQL_CLIENT) && defined(HAVE_REPLICATION) if (slave_thread && has_invoker()) { definer->user = invoker_user; definer->host= invoker_host; definer->password= null_lex_str; definer->plugin= empty_lex_str; definer->auth= empty_lex_str; } else #endif get_default_definer(this, definer, role); } /** Mark transaction to rollback and mark error as fatal to a sub-statement. @param thd Thread handle @param all TRUE <=> rollback main transaction. */ void mark_transaction_to_rollback(THD *thd, bool all) { if (thd) { thd->is_fatal_sub_stmt_error= TRUE; thd->transaction_rollback_request= all; } } /*************************************************************************** Handling of XA id cacheing ***************************************************************************/ mysql_mutex_t LOCK_xid_cache; HASH xid_cache; extern "C" uchar *xid_get_hash_key(const uchar *, size_t *, my_bool); extern "C" void xid_free_hash(void *); uchar *xid_get_hash_key(const uchar *ptr, size_t *length, my_bool not_used __attribute__((unused))) { *length=((XID_STATE*)ptr)->xid.key_length(); return ((XID_STATE*)ptr)->xid.key(); } void xid_free_hash(void *ptr) { if (!((XID_STATE*)ptr)->in_thd) my_free(ptr); } #ifdef HAVE_PSI_INTERFACE static PSI_mutex_key key_LOCK_xid_cache; static PSI_mutex_info all_xid_mutexes[]= { { &key_LOCK_xid_cache, "LOCK_xid_cache", PSI_FLAG_GLOBAL} }; static void init_xid_psi_keys(void) { const char* category= "sql"; int count; if (PSI_server == NULL) return; count= array_elements(all_xid_mutexes); PSI_server->register_mutex(category, all_xid_mutexes, count); } #endif /* HAVE_PSI_INTERFACE */ bool xid_cache_init() { #ifdef HAVE_PSI_INTERFACE init_xid_psi_keys(); #endif mysql_mutex_init(key_LOCK_xid_cache, &LOCK_xid_cache, MY_MUTEX_INIT_FAST); return my_hash_init(&xid_cache, &my_charset_bin, 100, 0, 0, xid_get_hash_key, xid_free_hash, 0) != 0; } void xid_cache_free() { if (my_hash_inited(&xid_cache)) { my_hash_free(&xid_cache); mysql_mutex_destroy(&LOCK_xid_cache); } } XID_STATE *xid_cache_search(XID *xid) { mysql_mutex_lock(&LOCK_xid_cache); XID_STATE *res=(XID_STATE *)my_hash_search(&xid_cache, xid->key(), xid->key_length()); mysql_mutex_unlock(&LOCK_xid_cache); return res; } bool xid_cache_insert(XID *xid, enum xa_states xa_state) { XID_STATE *xs; my_bool res; mysql_mutex_lock(&LOCK_xid_cache); if (my_hash_search(&xid_cache, xid->key(), xid->key_length())) res=0; else if (!(xs=(XID_STATE *)my_malloc(sizeof(*xs), MYF(MY_WME)))) res=1; else { xs->xa_state=xa_state; xs->xid.set(xid); xs->in_thd=0; xs->rm_error=0; res=my_hash_insert(&xid_cache, (uchar*)xs); } mysql_mutex_unlock(&LOCK_xid_cache); return res; } bool xid_cache_insert(XID_STATE *xid_state) { mysql_mutex_lock(&LOCK_xid_cache); if (my_hash_search(&xid_cache, xid_state->xid.key(), xid_state->xid.key_length())) { mysql_mutex_unlock(&LOCK_xid_cache); my_error(ER_XAER_DUPID, MYF(0)); return true; } bool res= my_hash_insert(&xid_cache, (uchar*)xid_state); mysql_mutex_unlock(&LOCK_xid_cache); return res; } void xid_cache_delete(XID_STATE *xid_state) { mysql_mutex_lock(&LOCK_xid_cache); my_hash_delete(&xid_cache, (uchar *)xid_state); mysql_mutex_unlock(&LOCK_xid_cache); } /** Decide on logging format to use for the statement and issue errors or warnings as needed. The decision depends on the following parameters: - The logging mode, i.e., the value of binlog_format. Can be statement, mixed, or row. - The type of statement. There are three types of statements: "normal" safe statements; unsafe statements; and row injections. An unsafe statement is one that, if logged in statement format, might produce different results when replayed on the slave (e.g., INSERT DELAYED). A row injection is either a BINLOG statement, or a row event executed by the slave's SQL thread. - The capabilities of tables modified by the statement. The *capabilities vector* for a table is a set of flags associated with the table. Currently, it only includes two flags: *row capability flag* and *statement capability flag*. The row capability flag is set if and only if the engine can handle row-based logging. The statement capability flag is set if and only if the table can handle statement-based logging. Decision table for logging format --------------------------------- The following table summarizes how the format and generated warning/error depends on the tables' capabilities, the statement type, and the current binlog_format. Row capable N NNNNNNNNN YYYYYYYYY YYYYYYYYY Statement capable N YYYYYYYYY NNNNNNNNN YYYYYYYYY Statement type * SSSUUUIII SSSUUUIII SSSUUUIII binlog_format * SMRSMRSMR SMRSMRSMR SMRSMRSMR Logged format - SS-S----- -RR-RR-RR SRRSRR-RR Warning/Error 1 --2732444 5--5--6-- ---7--6-- Legend ------ Row capable: N - Some table not row-capable, Y - All tables row-capable Stmt capable: N - Some table not stmt-capable, Y - All tables stmt-capable Statement type: (S)afe, (U)nsafe, or Row (I)njection binlog_format: (S)TATEMENT, (M)IXED, or (R)OW Logged format: (S)tatement or (R)ow Warning/Error: Warnings and error messages are as follows: 1. Error: Cannot execute statement: binlogging impossible since both row-incapable engines and statement-incapable engines are involved. 2. Error: Cannot execute statement: binlogging impossible since BINLOG_FORMAT = ROW and at least one table uses a storage engine limited to statement-logging. 3. Error: Cannot execute statement: binlogging of unsafe statement is impossible when storage engine is limited to statement-logging and BINLOG_FORMAT = MIXED. 4. Error: Cannot execute row injection: binlogging impossible since at least one table uses a storage engine limited to statement-logging. 5. Error: Cannot execute statement: binlogging impossible since BINLOG_FORMAT = STATEMENT and at least one table uses a storage engine limited to row-logging. 6. Error: Cannot execute row injection: binlogging impossible since BINLOG_FORMAT = STATEMENT. 7. Warning: Unsafe statement binlogged in statement format since BINLOG_FORMAT = STATEMENT. In addition, we can produce the following error (not depending on the variables of the decision diagram): 8. Error: Cannot execute statement: binlogging impossible since more than one engine is involved and at least one engine is self-logging. For each error case above, the statement is prevented from being logged, we report an error, and roll back the statement. For warnings, we set the thd->binlog_flags variable: the warning will be printed only if the statement is successfully logged. @see THD::binlog_query @param[in] thd Client thread @param[in] tables Tables involved in the query @retval 0 No error; statement can be logged. @retval -1 One of the error conditions above applies (1, 2, 4, 5, or 6). */ int THD::decide_logging_format(TABLE_LIST *tables) { DBUG_ENTER("THD::decide_logging_format"); DBUG_PRINT("info", ("Query: %s", query())); DBUG_PRINT("info", ("variables.binlog_format: %lu", variables.binlog_format)); DBUG_PRINT("info", ("lex->get_stmt_unsafe_flags(): 0x%x", lex->get_stmt_unsafe_flags())); reset_binlog_local_stmt_filter(); /* We should not decide logging format if the binlog is closed or binlogging is off, or if the statement is filtered out from the binlog by filtering rules. */ if (mysql_bin_log.is_open() && (variables.option_bits & OPTION_BIN_LOG) && !(wsrep_binlog_format() == BINLOG_FORMAT_STMT && !binlog_filter->db_ok(db))) { /* Compute one bit field with the union of all the engine capabilities, and one with the intersection of all the engine capabilities. */ handler::Table_flags flags_write_some_set= 0; handler::Table_flags flags_access_some_set= 0; handler::Table_flags flags_write_all_set= HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE; /* If different types of engines are about to be updated. For example: Innodb and Falcon; Innodb and MyIsam. */ my_bool multi_write_engine= FALSE; /* If different types of engines are about to be accessed and any of them is about to be updated. For example: Innodb and Falcon; Innodb and MyIsam. */ my_bool multi_access_engine= FALSE; /* Identifies if a table is changed. */ my_bool is_write= FALSE; /* A pointer to a previous table that was changed. */ TABLE* prev_write_table= NULL; /* A pointer to a previous table that was accessed. */ TABLE* prev_access_table= NULL; /** The number of tables used in the current statement, that should be replicated. */ uint replicated_tables_count= 0; /** The number of tables written to in the current statement, that should not be replicated. A table should not be replicated when it is considered 'local' to a MySQL instance. Currently, these tables are: - mysql.slow_log - mysql.general_log - mysql.slave_relay_log_info - mysql.slave_master_info - mysql.slave_worker_info - performance_schema.* - TODO: information_schema.* In practice, from this list, only performance_schema.* tables are written to by user queries. */ uint non_replicated_tables_count= 0; #ifndef DBUG_OFF { static const char *prelocked_mode_name[] = { "NON_PRELOCKED", "PRELOCKED", "PRELOCKED_UNDER_LOCK_TABLES", }; DBUG_PRINT("debug", ("prelocked_mode: %s", prelocked_mode_name[locked_tables_mode])); } #endif /* Get the capabilities vector for all involved storage engines and mask out the flags for the binary log. */ for (TABLE_LIST *table= tables; table; table= table->next_global) { if (table->placeholder()) continue; handler::Table_flags const flags= table->table->file->ha_table_flags(); DBUG_PRINT("info", ("table: %s; ha_table_flags: 0x%llx", table->table_name, flags)); if (table->table->no_replicate) { /* The statement uses a table that is not replicated. The following properties about the table: - persistent / transient - transactional / non transactional - temporary / permanent - read or write - multiple engines involved because of this table are not relevant, as this table is completely ignored. Because the statement uses a non replicated table, using STATEMENT format in the binlog is impossible. Either this statement will be discarded entirely, or it will be logged (possibly partially) in ROW format. */ lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_SYSTEM_TABLE); if (table->lock_type >= TL_WRITE_ALLOW_WRITE) { non_replicated_tables_count++; continue; } } replicated_tables_count++; if (table->lock_type >= TL_WRITE_ALLOW_WRITE) { if (prev_write_table && prev_write_table->file->ht != table->table->file->ht) multi_write_engine= TRUE; my_bool trans= table->table->file->has_transactions(); if (table->table->s->tmp_table) lex->set_stmt_accessed_table(trans ? LEX::STMT_WRITES_TEMP_TRANS_TABLE : LEX::STMT_WRITES_TEMP_NON_TRANS_TABLE); else lex->set_stmt_accessed_table(trans ? LEX::STMT_WRITES_TRANS_TABLE : LEX::STMT_WRITES_NON_TRANS_TABLE); flags_write_all_set &= flags; flags_write_some_set |= flags; is_write= TRUE; prev_write_table= table->table; } flags_access_some_set |= flags; if (lex->sql_command != SQLCOM_CREATE_TABLE || (lex->sql_command == SQLCOM_CREATE_TABLE && lex->tmp_table())) { my_bool trans= table->table->file->has_transactions(); if (table->table->s->tmp_table) lex->set_stmt_accessed_table(trans ? LEX::STMT_READS_TEMP_TRANS_TABLE : LEX::STMT_READS_TEMP_NON_TRANS_TABLE); else lex->set_stmt_accessed_table(trans ? LEX::STMT_READS_TRANS_TABLE : LEX::STMT_READS_NON_TRANS_TABLE); } if (prev_access_table && prev_access_table->file->ht != table->table->file->ht) multi_access_engine= TRUE; prev_access_table= table->table; } DBUG_PRINT("info", ("flags_write_all_set: 0x%llx", flags_write_all_set)); DBUG_PRINT("info", ("flags_write_some_set: 0x%llx", flags_write_some_set)); DBUG_PRINT("info", ("flags_access_some_set: 0x%llx", flags_access_some_set)); DBUG_PRINT("info", ("multi_write_engine: %d", multi_write_engine)); DBUG_PRINT("info", ("multi_access_engine: %d", multi_access_engine)); int error= 0; int unsafe_flags; bool multi_stmt_trans= in_multi_stmt_transaction_mode(); bool trans_table= trans_has_updated_trans_table(this); bool binlog_direct= variables.binlog_direct_non_trans_update; if (lex->is_mixed_stmt_unsafe(multi_stmt_trans, binlog_direct, trans_table, tx_isolation)) lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_MIXED_STATEMENT); else if (multi_stmt_trans && trans_table && !binlog_direct && lex->stmt_accessed_table(LEX::STMT_WRITES_NON_TRANS_TABLE)) lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_NONTRANS_AFTER_TRANS); /* If more than one engine is involved in the statement and at least one is doing it's own logging (is *self-logging*), the statement cannot be logged atomically, so we generate an error rather than allowing the binlog to become corrupt. */ if (multi_write_engine && (flags_write_some_set & HA_HAS_OWN_BINLOGGING)) my_error((error= ER_BINLOG_MULTIPLE_ENGINES_AND_SELF_LOGGING_ENGINE), MYF(0)); else if (multi_access_engine && flags_access_some_set & HA_HAS_OWN_BINLOGGING) lex->set_stmt_unsafe(LEX::BINLOG_STMT_UNSAFE_MULTIPLE_ENGINES_AND_SELF_LOGGING_ENGINE); /* both statement-only and row-only engines involved */ if ((flags_write_all_set & (HA_BINLOG_STMT_CAPABLE | HA_BINLOG_ROW_CAPABLE)) == 0) { /* 1. Error: Binary logging impossible since both row-incapable engines and statement-incapable engines are involved */ my_error((error= ER_BINLOG_ROW_ENGINE_AND_STMT_ENGINE), MYF(0)); } /* statement-only engines involved */ else if ((flags_write_all_set & HA_BINLOG_ROW_CAPABLE) == 0) { if (lex->is_stmt_row_injection()) { /* 4. Error: Cannot execute row injection since table uses storage engine limited to statement-logging */ my_error((error= ER_BINLOG_ROW_INJECTION_AND_STMT_ENGINE), MYF(0)); } else if (wsrep_binlog_format() == BINLOG_FORMAT_ROW && sqlcom_can_generate_row_events(this)) { /* 2. Error: Cannot modify table that uses a storage engine limited to statement-logging when BINLOG_FORMAT = ROW */ my_error((error= ER_BINLOG_ROW_MODE_AND_STMT_ENGINE), MYF(0)); } else if ((unsafe_flags= lex->get_stmt_unsafe_flags()) != 0) { /* 3. Error: Cannot execute statement: binlogging of unsafe statement is impossible when storage engine is limited to statement-logging and BINLOG_FORMAT = MIXED. */ for (int unsafe_type= 0; unsafe_type < LEX::BINLOG_STMT_UNSAFE_COUNT; unsafe_type++) if (unsafe_flags & (1 << unsafe_type)) my_error((error= ER_BINLOG_UNSAFE_AND_STMT_ENGINE), MYF(0), ER(LEX::binlog_stmt_unsafe_errcode[unsafe_type])); } /* log in statement format! */ } /* no statement-only engines */ else { /* binlog_format = STATEMENT */ if (wsrep_binlog_format() == BINLOG_FORMAT_STMT) { if (lex->is_stmt_row_injection()) { /* 6. Error: Cannot execute row injection since BINLOG_FORMAT = STATEMENT */ my_error((error= ER_BINLOG_ROW_INJECTION_AND_STMT_MODE), MYF(0)); } else if ((flags_write_all_set & HA_BINLOG_STMT_CAPABLE) == 0 && sqlcom_can_generate_row_events(this)) { /* 5. Error: Cannot modify table that uses a storage engine limited to row-logging when binlog_format = STATEMENT */ if (IF_WSREP((!WSREP(this) || wsrep_exec_mode == LOCAL_STATE),1)) { my_error((error= ER_BINLOG_STMT_MODE_AND_ROW_ENGINE), MYF(0), ""); } } else if (is_write && (unsafe_flags= lex->get_stmt_unsafe_flags()) != 0) { /* 7. Warning: Unsafe statement logged as statement due to binlog_format = STATEMENT */ binlog_unsafe_warning_flags|= unsafe_flags; DBUG_PRINT("info", ("Scheduling warning to be issued by " "binlog_query: '%s'", ER(ER_BINLOG_UNSAFE_STATEMENT))); DBUG_PRINT("info", ("binlog_unsafe_warning_flags: 0x%x", binlog_unsafe_warning_flags)); } /* log in statement format! */ } /* No statement-only engines and binlog_format != STATEMENT. I.e., nothing prevents us from row logging if needed. */ else { if (lex->is_stmt_unsafe() || lex->is_stmt_row_injection() || (flags_write_all_set & HA_BINLOG_STMT_CAPABLE) == 0) { /* log in row format! */ set_current_stmt_binlog_format_row_if_mixed(); } } } if (non_replicated_tables_count > 0) { if ((replicated_tables_count == 0) || ! is_write) { DBUG_PRINT("info", ("decision: no logging, no replicated table affected")); set_binlog_local_stmt_filter(); } else { if (! is_current_stmt_binlog_format_row()) { my_error((error= ER_BINLOG_STMT_MODE_AND_NO_REPL_TABLES), MYF(0)); } else { clear_binlog_local_stmt_filter(); } } } else { clear_binlog_local_stmt_filter(); } if (error) { DBUG_PRINT("info", ("decision: no logging since an error was generated")); DBUG_RETURN(-1); } DBUG_PRINT("info", ("decision: logging in %s format", is_current_stmt_binlog_format_row() ? "ROW" : "STATEMENT")); if (variables.binlog_format == BINLOG_FORMAT_ROW && (lex->sql_command == SQLCOM_UPDATE || lex->sql_command == SQLCOM_UPDATE_MULTI || lex->sql_command == SQLCOM_DELETE || lex->sql_command == SQLCOM_DELETE_MULTI)) { String table_names; /* Generate a warning for UPDATE/DELETE statements that modify a BLACKHOLE table, as row events are not logged in row format. */ for (TABLE_LIST *table= tables; table; table= table->next_global) { if (table->placeholder()) continue; if (table->table->file->ht->db_type == DB_TYPE_BLACKHOLE_DB && table->lock_type >= TL_WRITE_ALLOW_WRITE) { table_names.append(table->table_name); table_names.append(","); } } if (!table_names.is_empty()) { bool is_update= (lex->sql_command == SQLCOM_UPDATE || lex->sql_command == SQLCOM_UPDATE_MULTI); /* Replace the last ',' with '.' for table_names */ table_names.replace(table_names.length()-1, 1, ".", 1); push_warning_printf(this, Sql_condition::WARN_LEVEL_WARN, ER_UNKNOWN_ERROR, "Row events are not logged for %s statements " "that modify BLACKHOLE tables in row format. " "Table(s): '%-.192s'", is_update ? "UPDATE" : "DELETE", table_names.c_ptr()); } } } #ifndef DBUG_OFF else DBUG_PRINT("info", ("decision: no logging since " "mysql_bin_log.is_open() = %d " "and (options & OPTION_BIN_LOG) = 0x%llx " "and binlog_format = %u " "and binlog_filter->db_ok(db) = %d", mysql_bin_log.is_open(), (variables.option_bits & OPTION_BIN_LOG), (uint) wsrep_binlog_format(), binlog_filter->db_ok(db))); #endif DBUG_RETURN(0); } /* Implementation of interface to write rows to the binary log through the thread. The thread is responsible for writing the rows it has inserted/updated/deleted. */ #ifndef MYSQL_CLIENT /* Template member function for ensuring that there is an rows log event of the apropriate type before proceeding. PRE CONDITION: - Events of type 'RowEventT' have the type code 'type_code'. POST CONDITION: If a non-NULL pointer is returned, the pending event for thread 'thd' will be an event of type 'RowEventT' (which have the type code 'type_code') will either empty or have enough space to hold 'needed' bytes. In addition, the columns bitmap will be correct for the row, meaning that the pending event will be flushed if the columns in the event differ from the columns suppled to the function. RETURNS If no error, a non-NULL pending event (either one which already existed or the newly created one). If error, NULL. */ template Rows_log_event* THD::binlog_prepare_pending_rows_event(TABLE* table, uint32 serv_id, MY_BITMAP const* cols, size_t colcnt, size_t needed, bool is_transactional, RowsEventT *hint __attribute__((unused))) { DBUG_ENTER("binlog_prepare_pending_rows_event"); /* Pre-conditions */ DBUG_ASSERT(table->s->table_map_id != ~0UL); /* Fetch the type code for the RowsEventT template parameter */ int const general_type_code= RowsEventT::TYPE_CODE; /* Ensure that all events in a GTID group are in the same cache */ if (variables.option_bits & OPTION_GTID_BEGIN) is_transactional= 1; /* There is no good place to set up the transactional data, so we have to do it here. */ if (binlog_setup_trx_data() == NULL) DBUG_RETURN(NULL); Rows_log_event* pending= binlog_get_pending_rows_event(is_transactional); if (unlikely(pending && !pending->is_valid())) DBUG_RETURN(NULL); /* Check if the current event is non-NULL and a write-rows event. Also check if the table provided is mapped: if it is not, then we have switched to writing to a new table. If there is no pending event, we need to create one. If there is a pending event, but it's not about the same table id, or not of the same type (between Write, Update and Delete), or not the same affected columns, or going to be too big, flush this event to disk and create a new pending event. */ if (!pending || pending->server_id != serv_id || pending->get_table_id() != table->s->table_map_id || pending->get_general_type_code() != general_type_code || pending->get_data_size() + needed > opt_binlog_rows_event_max_size || pending->get_width() != colcnt || !bitmap_cmp(pending->get_cols(), cols)) { /* Create a new RowsEventT... */ Rows_log_event* const ev= new RowsEventT(this, table, table->s->table_map_id, cols, is_transactional); if (unlikely(!ev)) DBUG_RETURN(NULL); ev->server_id= serv_id; // I don't like this, it's too easy to forget. /* flush the pending event and replace it with the newly created event... */ if (unlikely( mysql_bin_log.flush_and_set_pending_rows_event(this, ev, is_transactional))) { delete ev; DBUG_RETURN(NULL); } DBUG_RETURN(ev); /* This is the new pending event */ } DBUG_RETURN(pending); /* This is the current pending event */ } /* Declare in unnamed namespace. */ CPP_UNNAMED_NS_START /** Class to handle temporary allocation of memory for row data. The responsibilities of the class is to provide memory for packing one or two rows of packed data (depending on what constructor is called). In order to make the allocation more efficient for "simple" rows, i.e., rows that do not contain any blobs, a pointer to the allocated memory is of memory is stored in the table structure for simple rows. If memory for a table containing a blob field is requested, only memory for that is allocated, and subsequently released when the object is destroyed. */ class Row_data_memory { public: /** Build an object to keep track of a block-local piece of memory for storing a row of data. @param table Table where the pre-allocated memory is stored. @param length Length of data that is needed, if the record contain blobs. */ Row_data_memory(TABLE *table, size_t const len1) : m_memory(0) { #ifndef DBUG_OFF m_alloc_checked= FALSE; #endif allocate_memory(table, len1); m_ptr[0]= has_memory() ? m_memory : 0; m_ptr[1]= 0; } Row_data_memory(TABLE *table, size_t const len1, size_t const len2) : m_memory(0) { #ifndef DBUG_OFF m_alloc_checked= FALSE; #endif allocate_memory(table, len1 + len2); m_ptr[0]= has_memory() ? m_memory : 0; m_ptr[1]= has_memory() ? m_memory + len1 : 0; } ~Row_data_memory() { if (m_memory != 0 && m_release_memory_on_destruction) my_free(m_memory); } /** Is there memory allocated? @retval true There is memory allocated @retval false Memory allocation failed */ bool has_memory() const { #ifndef DBUG_OFF m_alloc_checked= TRUE; #endif return m_memory != 0; } uchar *slot(uint s) { DBUG_ASSERT(s < sizeof(m_ptr)/sizeof(*m_ptr)); DBUG_ASSERT(m_ptr[s] != 0); DBUG_ASSERT(m_alloc_checked == TRUE); return m_ptr[s]; } private: void allocate_memory(TABLE *const table, size_t const total_length) { if (table->s->blob_fields == 0) { /* The maximum length of a packed record is less than this length. We use this value instead of the supplied length when allocating memory for records, since we don't know how the memory will be used in future allocations. Since table->s->reclength is for unpacked records, we have to add two bytes for each field, which can potentially be added to hold the length of a packed field. */ size_t const maxlen= table->s->reclength + 2 * table->s->fields; /* Allocate memory for two records if memory hasn't been allocated. We allocate memory for two records so that it can be used when processing update rows as well. */ if (table->write_row_record == 0) table->write_row_record= (uchar *) alloc_root(&table->mem_root, 2 * maxlen); m_memory= table->write_row_record; m_release_memory_on_destruction= FALSE; } else { m_memory= (uchar *) my_malloc(total_length, MYF(MY_WME)); m_release_memory_on_destruction= TRUE; } } #ifndef DBUG_OFF mutable bool m_alloc_checked; #endif bool m_release_memory_on_destruction; uchar *m_memory; uchar *m_ptr[2]; }; CPP_UNNAMED_NS_END int THD::binlog_write_row(TABLE* table, bool is_trans, MY_BITMAP const* cols, size_t colcnt, uchar const *record) { DBUG_ASSERT(is_current_stmt_binlog_format_row() && ((WSREP(this) && wsrep_emulate_bin_log) || mysql_bin_log.is_open())); /* Pack records into format for transfer. We are allocating more memory than needed, but that doesn't matter. */ Row_data_memory memory(table, max_row_length(table, record)); if (!memory.has_memory()) return HA_ERR_OUT_OF_MEM; uchar *row_data= memory.slot(0); size_t const len= pack_row(table, cols, row_data, record); /* Ensure that all events in a GTID group are in the same cache */ if (variables.option_bits & OPTION_GTID_BEGIN) is_trans= 1; Rows_log_event* const ev= binlog_prepare_pending_rows_event(table, variables.server_id, cols, colcnt, len, is_trans, static_cast(0)); if (unlikely(ev == 0)) return HA_ERR_OUT_OF_MEM; return ev->add_row_data(row_data, len); } int THD::binlog_update_row(TABLE* table, bool is_trans, MY_BITMAP const* cols, size_t colcnt, const uchar *before_record, const uchar *after_record) { DBUG_ASSERT(is_current_stmt_binlog_format_row() && ((WSREP(this) && wsrep_emulate_bin_log) || mysql_bin_log.is_open())); size_t const before_maxlen = max_row_length(table, before_record); size_t const after_maxlen = max_row_length(table, after_record); Row_data_memory row_data(table, before_maxlen, after_maxlen); if (!row_data.has_memory()) return HA_ERR_OUT_OF_MEM; uchar *before_row= row_data.slot(0); uchar *after_row= row_data.slot(1); size_t const before_size= pack_row(table, cols, before_row, before_record); size_t const after_size= pack_row(table, cols, after_row, after_record); /* Ensure that all events in a GTID group are in the same cache */ if (variables.option_bits & OPTION_GTID_BEGIN) is_trans= 1; /* Don't print debug messages when running valgrind since they can trigger false warnings. */ #ifndef HAVE_valgrind DBUG_DUMP("before_record", before_record, table->s->reclength); DBUG_DUMP("after_record", after_record, table->s->reclength); DBUG_DUMP("before_row", before_row, before_size); DBUG_DUMP("after_row", after_row, after_size); #endif Rows_log_event* const ev= binlog_prepare_pending_rows_event(table, variables.server_id, cols, colcnt, before_size + after_size, is_trans, static_cast(0)); if (unlikely(ev == 0)) return HA_ERR_OUT_OF_MEM; return ev->add_row_data(before_row, before_size) || ev->add_row_data(after_row, after_size); } int THD::binlog_delete_row(TABLE* table, bool is_trans, MY_BITMAP const* cols, size_t colcnt, uchar const *record) { DBUG_ASSERT(is_current_stmt_binlog_format_row() && ((WSREP(this) && wsrep_emulate_bin_log) || mysql_bin_log.is_open())); /* Pack records into format for transfer. We are allocating more memory than needed, but that doesn't matter. */ Row_data_memory memory(table, max_row_length(table, record)); if (unlikely(!memory.has_memory())) return HA_ERR_OUT_OF_MEM; uchar *row_data= memory.slot(0); size_t const len= pack_row(table, cols, row_data, record); /* Ensure that all events in a GTID group are in the same cache */ if (variables.option_bits & OPTION_GTID_BEGIN) is_trans= 1; Rows_log_event* const ev= binlog_prepare_pending_rows_event(table, variables.server_id, cols, colcnt, len, is_trans, static_cast(0)); if (unlikely(ev == 0)) return HA_ERR_OUT_OF_MEM; return ev->add_row_data(row_data, len); } int THD::binlog_remove_pending_rows_event(bool clear_maps, bool is_transactional) { DBUG_ENTER("THD::binlog_remove_pending_rows_event"); if(!WSREP_EMULATE_BINLOG(this) && !mysql_bin_log.is_open()) DBUG_RETURN(0); /* Ensure that all events in a GTID group are in the same cache */ if (variables.option_bits & OPTION_GTID_BEGIN) is_transactional= 1; mysql_bin_log.remove_pending_rows_event(this, is_transactional); if (clear_maps) binlog_table_maps= 0; DBUG_RETURN(0); } int THD::binlog_flush_pending_rows_event(bool stmt_end, bool is_transactional) { DBUG_ENTER("THD::binlog_flush_pending_rows_event"); /* We shall flush the pending event even if we are not in row-based mode: it might be the case that we left row-based mode before flushing anything (e.g., if we have explicitly locked tables). */ if(!WSREP_EMULATE_BINLOG(this) && !mysql_bin_log.is_open()) DBUG_RETURN(0); /* Ensure that all events in a GTID group are in the same cache */ if (variables.option_bits & OPTION_GTID_BEGIN) is_transactional= 1; /* Mark the event as the last event of a statement if the stmt_end flag is set. */ int error= 0; if (Rows_log_event *pending= binlog_get_pending_rows_event(is_transactional)) { if (stmt_end) { pending->set_flags(Rows_log_event::STMT_END_F); binlog_table_maps= 0; } error= mysql_bin_log.flush_and_set_pending_rows_event(this, 0, is_transactional); } DBUG_RETURN(error); } #if !defined(DBUG_OFF) && !defined(_lint) static const char * show_query_type(THD::enum_binlog_query_type qtype) { switch (qtype) { case THD::ROW_QUERY_TYPE: return "ROW"; case THD::STMT_QUERY_TYPE: return "STMT"; case THD::QUERY_TYPE_COUNT: default: DBUG_ASSERT(0 <= qtype && qtype < THD::QUERY_TYPE_COUNT); } static char buf[64]; sprintf(buf, "UNKNOWN#%d", qtype); return buf; } #endif /* Constants required for the limit unsafe warnings suppression */ //seconds after which the limit unsafe warnings suppression will be activated #define LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT 5*60 //number of limit unsafe warnings after which the suppression will be activated #define LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT 10 static ulonglong unsafe_suppression_start_time= 0; static bool unsafe_warning_suppression_active[LEX::BINLOG_STMT_UNSAFE_COUNT]; static ulong unsafe_warnings_count[LEX::BINLOG_STMT_UNSAFE_COUNT]; static ulong total_unsafe_warnings_count; /** Auxiliary function to reset the limit unsafety warning suppression. This is done without mutex protection, but this should be good enough as it doesn't matter if we loose a couple of suppressed messages or if this is called multiple times. */ static void reset_binlog_unsafe_suppression(ulonglong now) { uint i; DBUG_ENTER("reset_binlog_unsafe_suppression"); unsafe_suppression_start_time= now; total_unsafe_warnings_count= 0; for (i= 0 ; i < LEX::BINLOG_STMT_UNSAFE_COUNT ; i++) { unsafe_warnings_count[i]= 0; unsafe_warning_suppression_active[i]= 0; } DBUG_VOID_RETURN; } /** Auxiliary function to print warning in the error log. */ static void print_unsafe_warning_to_log(int unsafe_type, char* buf, char* query) { DBUG_ENTER("print_unsafe_warning_in_log"); sprintf(buf, ER(ER_BINLOG_UNSAFE_STATEMENT), ER(LEX::binlog_stmt_unsafe_errcode[unsafe_type])); sql_print_warning(ER(ER_MESSAGE_AND_STATEMENT), buf, query); DBUG_VOID_RETURN; } /** Auxiliary function to check if the warning for unsafe repliction statements should be thrown or suppressed. Logic is: - If we get more than LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT errors of one type, that type of errors will be suppressed for LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT. - When the time limit has been reached, all suppression is reset. This means that if one gets many different types of errors, some of them may be reset less than LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT. However at least one error is disable for this time. SYNOPSIS: @params unsafe_type - The type of unsafety. RETURN: 0 0k to log 1 Message suppressed */ static bool protect_against_unsafe_warning_flood(int unsafe_type) { ulong count; ulonglong now= my_interval_timer()/1000000000ULL; DBUG_ENTER("protect_against_unsafe_warning_flood"); count= ++unsafe_warnings_count[unsafe_type]; total_unsafe_warnings_count++; /* INITIALIZING: If this is the first time this function is called with log warning enabled, the monitoring the unsafe warnings should start. */ if (unsafe_suppression_start_time == 0) { reset_binlog_unsafe_suppression(now); DBUG_RETURN(0); } /* The following is true if we got too many errors or if the error was already suppressed */ if (count >= LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT) { ulonglong diff_time= (now - unsafe_suppression_start_time); if (!unsafe_warning_suppression_active[unsafe_type]) { /* ACTIVATION: We got LIMIT_UNSAFE_WARNING_ACTIVATION_THRESHOLD_COUNT warnings in less than LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT we activate the suppression. */ if (diff_time <= LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT) { unsafe_warning_suppression_active[unsafe_type]= 1; sql_print_information("Suppressing warnings of type '%s' for up to %d seconds because of flooding", ER(LEX::binlog_stmt_unsafe_errcode[unsafe_type]), LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT); } else { /* There is no flooding till now, therefore we restart the monitoring */ reset_binlog_unsafe_suppression(now); } } else { /* This type of warnings was suppressed */ if (diff_time > LIMIT_UNSAFE_WARNING_ACTIVATION_TIMEOUT) { ulong save_count= total_unsafe_warnings_count; /* Print a suppression note and remove the suppression */ reset_binlog_unsafe_suppression(now); sql_print_information("Suppressed %lu unsafe warnings during " "the last %d seconds", save_count, (int) diff_time); } } } DBUG_RETURN(unsafe_warning_suppression_active[unsafe_type]); } /** Auxiliary method used by @c binlog_query() to raise warnings. The type of warning and the type of unsafeness is stored in THD::binlog_unsafe_warning_flags. */ void THD::issue_unsafe_warnings() { char buf[MYSQL_ERRMSG_SIZE * 2]; uint32 unsafe_type_flags; DBUG_ENTER("issue_unsafe_warnings"); /* Ensure that binlog_unsafe_warning_flags is big enough to hold all bits. This is actually a constant expression. */ DBUG_ASSERT(LEX::BINLOG_STMT_UNSAFE_COUNT <= sizeof(binlog_unsafe_warning_flags) * CHAR_BIT); if (!(unsafe_type_flags= binlog_unsafe_warning_flags)) DBUG_VOID_RETURN; // Nothing to do /* For each unsafe_type, check if the statement is unsafe in this way and issue a warning. */ for (int unsafe_type=0; unsafe_type < LEX::BINLOG_STMT_UNSAFE_COUNT; unsafe_type++) { if ((unsafe_type_flags & (1 << unsafe_type)) != 0) { push_warning_printf(this, Sql_condition::WARN_LEVEL_NOTE, ER_BINLOG_UNSAFE_STATEMENT, ER(ER_BINLOG_UNSAFE_STATEMENT), ER(LEX::binlog_stmt_unsafe_errcode[unsafe_type])); if (global_system_variables.log_warnings > 0 && !protect_against_unsafe_warning_flood(unsafe_type)) print_unsafe_warning_to_log(unsafe_type, buf, query()); } } DBUG_VOID_RETURN; } /** Log the current query. The query will be logged in either row format or statement format depending on the value of @c current_stmt_binlog_format_row field and the value of the @c qtype parameter. This function must be called: - After the all calls to ha_*_row() functions have been issued. - After any writes to system tables. Rationale: if system tables were written after a call to this function, and the master crashes after the call to this function and before writing the system tables, then the master and slave get out of sync. - Before tables are unlocked and closed. @see decide_logging_format @retval 0 Success @retval nonzero If there is a failure when writing the query (e.g., write failure), then the error code is returned. */ int THD::binlog_query(THD::enum_binlog_query_type qtype, char const *query_arg, ulong query_len, bool is_trans, bool direct, bool suppress_use, int errcode) { DBUG_ENTER("THD::binlog_query"); DBUG_PRINT("enter", ("qtype: %s query: '%-.*s'", show_query_type(qtype), (int) query_len, query_arg)); DBUG_ASSERT(query_arg); DBUG_ASSERT(WSREP_EMULATE_BINLOG(this) || mysql_bin_log.is_open()); /* If this is withing a BEGIN ... COMMIT group, don't log it */ if (variables.option_bits & OPTION_GTID_BEGIN) { direct= 0; is_trans= 1; } DBUG_PRINT("info", ("is_trans: %d direct: %d", is_trans, direct)); if (get_binlog_local_stmt_filter() == BINLOG_FILTER_SET) { /* The current statement is to be ignored, and not written to the binlog. Do not call issue_unsafe_warnings(). */ DBUG_RETURN(0); } /* If we are not in prelocked mode, mysql_unlock_tables() will be called after this binlog_query(), so we have to flush the pending rows event with the STMT_END_F set to unlock all tables at the slave side as well. If we are in prelocked mode, the flushing will be done inside the top-most close_thread_tables(). */ if (this->locked_tables_mode <= LTM_LOCK_TABLES) if (int error= binlog_flush_pending_rows_event(TRUE, is_trans)) DBUG_RETURN(error); /* Warnings for unsafe statements logged in statement format are printed in three places instead of in decide_logging_format(). This is because the warnings should be printed only if the statement is actually logged. When executing decide_logging_format(), we cannot know for sure if the statement will be logged: 1 - sp_head::execute_procedure which prints out warnings for calls to stored procedures. 2 - sp_head::execute_function which prints out warnings for calls involving functions. 3 - THD::binlog_query (here) which prints warning for top level statements not covered by the two cases above: i.e., if not insided a procedure and a function. Besides, we should not try to print these warnings if it is not possible to write statements to the binary log as it happens when the execution is inside a function, or generaly speaking, when the variables.option_bits & OPTION_BIN_LOG is false. */ if ((variables.option_bits & OPTION_BIN_LOG) && spcont == NULL && !binlog_evt_union.do_union) issue_unsafe_warnings(); switch (qtype) { /* ROW_QUERY_TYPE means that the statement may be logged either in row format or in statement format. If current_stmt_binlog_format is row, it means that the statement has already been logged in row format and hence shall not be logged again. */ case THD::ROW_QUERY_TYPE: DBUG_PRINT("debug", ("is_current_stmt_binlog_format_row: %d", is_current_stmt_binlog_format_row())); if (is_current_stmt_binlog_format_row()) DBUG_RETURN(0); /* Fall through */ /* STMT_QUERY_TYPE means that the query must be logged in statement format; it cannot be logged in row format. This is typically used by DDL statements. It is an error to use this query type if current_stmt_binlog_format_row is row. @todo Currently there are places that call this method with STMT_QUERY_TYPE and current_stmt_binlog_format is row. Fix those places and add assert to ensure correct behavior. /Sven */ case THD::STMT_QUERY_TYPE: /* The MYSQL_LOG::write() function will set the STMT_END_F flag and flush the pending rows event if necessary. */ /* Even though wsrep only supports ROW binary log format, a user can set binlog format to STATEMENT (wsrep_forced_binlog_format). In which case the control might reach here even when binary logging (--log-bin) is not enabled. This is possible because wsrep patch partially enables binary logging by setting wsrep_emulate_binlog. */ if (mysql_bin_log.is_open()) { Query_log_event qinfo(this, query_arg, query_len, is_trans, direct, suppress_use, errcode); /* Binlog table maps will be irrelevant after a Query_log_event (they are just removed on the slave side) so after the query log event is written to the binary log, we pretend that no table maps were written. */ int error= mysql_bin_log.write(&qinfo); binlog_table_maps= 0; DBUG_RETURN(error); } case THD::QUERY_TYPE_COUNT: default: DBUG_ASSERT(qtype < QUERY_TYPE_COUNT); } DBUG_RETURN(0); } void THD::wait_for_wakeup_ready() { mysql_mutex_lock(&LOCK_wakeup_ready); while (!wakeup_ready) mysql_cond_wait(&COND_wakeup_ready, &LOCK_wakeup_ready); mysql_mutex_unlock(&LOCK_wakeup_ready); } void THD::signal_wakeup_ready() { mysql_mutex_lock(&LOCK_wakeup_ready); wakeup_ready= true; mysql_mutex_unlock(&LOCK_wakeup_ready); mysql_cond_signal(&COND_wakeup_ready); } void THD::rgi_lock_temporary_tables() { mysql_mutex_lock(&rgi_slave->rli->data_lock); temporary_tables= rgi_slave->rli->save_temporary_tables; } void THD::rgi_unlock_temporary_tables() { rgi_slave->rli->save_temporary_tables= temporary_tables; mysql_mutex_unlock(&rgi_slave->rli->data_lock); } bool THD::rgi_have_temporary_tables() { return rgi_slave->rli->save_temporary_tables != 0; } void wait_for_commit::reinit() { subsequent_commits_list= NULL; next_subsequent_commit= NULL; waitee= NULL; opaque_pointer= NULL; wakeup_error= 0; wakeup_subsequent_commits_running= false; commit_started= false; #ifdef SAFE_MUTEX /* When using SAFE_MUTEX, the ordering between taking the LOCK_wait_commit mutexes is checked. This causes a problem when we re-use a mutex, as then the expected locking order may change. So in this case, do a re-init of the mutex. In release builds, we want to avoid the overhead of a re-init though. */ mysql_mutex_destroy(&LOCK_wait_commit); mysql_mutex_init(key_LOCK_wait_commit, &LOCK_wait_commit, MY_MUTEX_INIT_FAST); #endif } wait_for_commit::wait_for_commit() { mysql_mutex_init(key_LOCK_wait_commit, &LOCK_wait_commit, MY_MUTEX_INIT_FAST); mysql_cond_init(key_COND_wait_commit, &COND_wait_commit, 0); reinit(); } wait_for_commit::~wait_for_commit() { /* Since we do a dirty read of the waiting_for_commit flag in wait_for_prior_commit() and in unregister_wait_for_prior_commit(), we need to take extra care before freeing the wait_for_commit object. It is possible for the waitee to be pre-empted inside wakeup(), just after it has cleared the waiting_for_commit flag and before it has released the LOCK_wait_commit mutex. And then it is possible for the waiter to find the flag cleared in wait_for_prior_commit() and go finish up things and de-allocate the LOCK_wait_commit and COND_wait_commit objects before the waitee has time to be re-scheduled and finish unlocking the mutex and signalling the condition. This would lead to the waitee accessing no longer valid memory. To prevent this, we do an extra lock/unlock of the mutex here before deallocation; this makes certain that any waitee has completed wakeup() first. */ mysql_mutex_lock(&LOCK_wait_commit); mysql_mutex_unlock(&LOCK_wait_commit); mysql_mutex_destroy(&LOCK_wait_commit); mysql_cond_destroy(&COND_wait_commit); } void wait_for_commit::wakeup(int wakeup_error) { /* We signal each waiter on their own condition and mutex (rather than using pthread_cond_broadcast() or something like that). Otherwise we would need to somehow ensure that they were done waking up before we could allow this THD to be destroyed, which would be annoying and unnecessary. Note that wakeup_subsequent_commits2() depends on this function being a full memory barrier (it is, because it takes a mutex lock). */ mysql_mutex_lock(&LOCK_wait_commit); waitee= NULL; this->wakeup_error= wakeup_error; /* Note that it is critical that the mysql_cond_signal() here is done while still holding the mutex. As soon as we release the mutex, the waiter might deallocate the condition object. */ mysql_cond_signal(&COND_wait_commit); mysql_mutex_unlock(&LOCK_wait_commit); } /* Register that the next commit of this THD should wait to complete until commit in another THD (the waitee) has completed. The wait may occur explicitly, with the waiter sitting in wait_for_prior_commit() until the waitee calls wakeup_subsequent_commits(). Alternatively, the TC (eg. binlog) may do the commits of both waitee and waiter at once during group commit, resolving both of them in the right order. Only one waitee can be registered for a waiter; it must be removed by wait_for_prior_commit() or unregister_wait_for_prior_commit() before a new one is registered. But it is ok for several waiters to register a wait for the same waitee. It is also permissible for one THD to be both a waiter and a waitee at the same time. */ void wait_for_commit::register_wait_for_prior_commit(wait_for_commit *waitee) { DBUG_ASSERT(!this->waitee /* No prior registration allowed */); wakeup_error= 0; this->waitee= waitee; mysql_mutex_lock(&waitee->LOCK_wait_commit); /* If waitee is in the middle of wakeup, then there is nothing to wait for, so we need not register. This is necessary to avoid a race in unregister, see comments on wakeup_subsequent_commits2() for details. */ if (waitee->wakeup_subsequent_commits_running) this->waitee= NULL; else { /* Put ourself at the head of the waitee's list of transactions that must wait for it to commit first. */ this->next_subsequent_commit= waitee->subsequent_commits_list; waitee->subsequent_commits_list= this; } mysql_mutex_unlock(&waitee->LOCK_wait_commit); } /* Wait for commit of another transaction to complete, as already registered with register_wait_for_prior_commit(). If the commit already completed, returns immediately. */ int wait_for_commit::wait_for_prior_commit2(THD *thd) { PSI_stage_info old_stage; wait_for_commit *loc_waitee; mysql_mutex_lock(&LOCK_wait_commit); DEBUG_SYNC(thd, "wait_for_prior_commit_waiting"); thd->ENTER_COND(&COND_wait_commit, &LOCK_wait_commit, &stage_waiting_for_prior_transaction_to_commit, &old_stage); while ((loc_waitee= this->waitee) && !thd->check_killed()) mysql_cond_wait(&COND_wait_commit, &LOCK_wait_commit); if (!loc_waitee) { if (wakeup_error) my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); goto end; } /* Wait was interrupted by kill. We need to unregister our wait and give the error. But if a wakeup is already in progress, then we must ignore the kill and not give error, otherwise we get inconsistency between waitee and waiter as to whether we succeed or fail (eg. we may roll back but waitee might attempt to commit both us and any subsequent commits waiting for us). */ mysql_mutex_lock(&loc_waitee->LOCK_wait_commit); if (loc_waitee->wakeup_subsequent_commits_running) { /* We are being woken up; ignore the kill and just wait. */ mysql_mutex_unlock(&loc_waitee->LOCK_wait_commit); do { mysql_cond_wait(&COND_wait_commit, &LOCK_wait_commit); } while (this->waitee); if (wakeup_error) my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); goto end; } remove_from_list(&loc_waitee->subsequent_commits_list); mysql_mutex_unlock(&loc_waitee->LOCK_wait_commit); this->waitee= NULL; wakeup_error= thd->killed_errno(); if (!wakeup_error) wakeup_error= ER_QUERY_INTERRUPTED; my_message(wakeup_error, ER(wakeup_error), MYF(0)); thd->EXIT_COND(&old_stage); /* Must do the DEBUG_SYNC() _after_ exit_cond(), as DEBUG_SYNC is not safe to use within enter_cond/exit_cond. */ DEBUG_SYNC(thd, "wait_for_prior_commit_killed"); return wakeup_error; end: thd->EXIT_COND(&old_stage); return wakeup_error; } /* Wakeup anyone waiting for us to have committed. Note about locking: We have a potential race or deadlock between wakeup_subsequent_commits() in the waitee and unregister_wait_for_prior_commit() in the waiter. Both waiter and waitee needs to take their own lock before it is safe to take a lock on the other party - else the other party might disappear and invalid memory data could be accessed. But if we take the two locks in different order, we may end up in a deadlock. The waiter needs to lock the waitee to delete itself from the list in unregister_wait_for_prior_commit(). Thus wakeup_subsequent_commits() can not hold its own lock while locking waiters, as this could lead to deadlock. So we need to prevent unregister_wait_for_prior_commit() running while wakeup is in progress - otherwise the unregister could complete before the wakeup, leading to incorrect spurious wakeup or accessing invalid memory. However, if we are in the middle of running wakeup_subsequent_commits(), then there is no need for unregister_wait_for_prior_commit() in the first place - the waiter can just do a normal wait_for_prior_commit(), as it will be immediately woken up. So the solution to the potential race/deadlock is to set a flag in the waitee that wakeup_subsequent_commits() is in progress. When this flag is set, unregister_wait_for_prior_commit() becomes just wait_for_prior_commit(). Then also register_wait_for_prior_commit() needs to check if wakeup_subsequent_commits() is running, and skip the registration if so. This is needed in case a new waiter manages to register itself and immediately try to unregister while wakeup_subsequent_commits() is running. Else the new waiter would also wait rather than unregister, but it would not be woken up until next wakeup, which could be potentially much later than necessary. */ void wait_for_commit::wakeup_subsequent_commits2(int wakeup_error) { wait_for_commit *waiter; mysql_mutex_lock(&LOCK_wait_commit); wakeup_subsequent_commits_running= true; waiter= subsequent_commits_list; subsequent_commits_list= NULL; mysql_mutex_unlock(&LOCK_wait_commit); while (waiter) { /* Important: we must grab the next pointer before waking up the waiter; once the wakeup is done, the field could be invalidated at any time. */ wait_for_commit *next= waiter->next_subsequent_commit; waiter->wakeup(wakeup_error); waiter= next; } /* We need a full memory barrier between walking the list above, and clearing the flag wakeup_subsequent_commits_running below. This barrier is needed to ensure that no other thread will start to modify the list pointers before we are done traversing the list. But wait_for_commit::wakeup() does a full memory barrier already (it locks a mutex), so no extra explicit barrier is needed here. */ wakeup_subsequent_commits_running= false; } /* Cancel a previously registered wait for another THD to commit before us. */ void wait_for_commit::unregister_wait_for_prior_commit2() { wait_for_commit *loc_waitee; mysql_mutex_lock(&LOCK_wait_commit); if ((loc_waitee= this->waitee)) { mysql_mutex_lock(&loc_waitee->LOCK_wait_commit); if (loc_waitee->wakeup_subsequent_commits_running) { /* When a wakeup is running, we cannot safely remove ourselves from the list without corrupting it. Instead we can just wait, as wakeup is already in progress and will thus be immediate. See comments on wakeup_subsequent_commits2() for more details. */ mysql_mutex_unlock(&loc_waitee->LOCK_wait_commit); while (this->waitee) mysql_cond_wait(&COND_wait_commit, &LOCK_wait_commit); } else { /* Remove ourselves from the list in the waitee. */ remove_from_list(&loc_waitee->subsequent_commits_list); mysql_mutex_unlock(&loc_waitee->LOCK_wait_commit); this->waitee= NULL; } } wakeup_error= 0; mysql_mutex_unlock(&LOCK_wait_commit); } bool Discrete_intervals_list::append(ulonglong start, ulonglong val, ulonglong incr) { DBUG_ENTER("Discrete_intervals_list::append"); /* first, see if this can be merged with previous */ if ((head == NULL) || tail->merge_if_contiguous(start, val, incr)) { /* it cannot, so need to add a new interval */ Discrete_interval *new_interval= new Discrete_interval(start, val, incr); DBUG_RETURN(append(new_interval)); } DBUG_RETURN(0); } bool Discrete_intervals_list::append(Discrete_interval *new_interval) { DBUG_ENTER("Discrete_intervals_list::append"); if (unlikely(new_interval == NULL)) DBUG_RETURN(1); DBUG_PRINT("info",("adding new auto_increment interval")); if (head == NULL) head= current= new_interval; else tail->next= new_interval; tail= new_interval; elements++; DBUG_RETURN(0); } #endif /* !defined(MYSQL_CLIENT) */