- developers - lists.mariadb.org

Re: c50eed1e904: MDEV-32645 CAST(AS UNSIGNED) fails with --view-protocol
by Sergei Golubchik 06 Jan '24

06 Jan '24

Hi, Alexander, Just one comment below On Jan 06, Alexander Barkov wrote: > revision-id: c50eed1e904 (mariadb-10.4.30-318-gc50eed1e904) > parent(s): 28cdbab1637 > author: Alexander Barkov > committer: Alexander Barkov > timestamp: 2023-11-13 11:18:16 +0400 > message: > > MDEV-32645 CAST(AS UNSIGNED) fails with --view-protocol > > Item_float::neg() did not preserve the "presentation" from "this". > So > CAST(-1e0 AS UNSIGNED) -- cast from double to … [View More]

1 0

Re: 9e7c80f8006: MDEV-32803 Assertion `total == 0' failed in Event_log::write_cache_raw
by Sergei Golubchik 05 Jan '24

05 Jan '24

Hi, Nikita, First, I'd suggest this change: --- a/sql/online_alter.cc +++ b/sql/online_alter.cc @@ -257,8 +257,7 @@ int online_alter_end_trans(Online_alter_cache_list { my_error(ER_ERROR_ON_WRITE, MYF(ME_ERROR_LOG), binlog->get_name(), errno); - cleanup_cache_list(cache_list, is_ending_transaction); - DBUG_RETURN(error); + break; } } and also I'd suggest to move if (ending_trans) from cleanup_cache_list() to the caller (there's only … [View More]

1 0

Re: 1c9edc72fc2: MDEV-32830 I. refactor XA binlogging for better integration with BGC/replication/recovery
by Sergei Golubchik 05 Jan '24

05 Jan '24

Hi, Andrei, On Jan 05, Andrei wrote: > commit 1c9edc72fc2 > Author: Andrei <andrei.elkin(a)mariadb.com> > Date: Wed Nov 1 20:32:45 2023 +0200 > > MDEV-32830 I. refactor XA binlogging for better integration with BGC/replication/recovery > > This commit is the part I in the series of three that addresses MDEV-31949. > > This part improves upon MDEV-742 design of the XA binlogging. > When binlog is ON, to execute XA-prepare first … [View More]in engines to write > the xa-prepare replication event last. > MYSQL_BIN_LOG::run_ordered_commit() is made to execute XA commit and > rollback in engines, as well as > it completes with the XA transaction state, so that its > XID has become available for next use, e.g by another slave parallel > worker, that e.g could be waiting for its order to commit a being > prepared XID transaction. Is it intentional that line length in the paragraph above varies randomly from line to line? And I don't understand why MYSQL_BIN_LOG::run_ordered_commit() has to do the complete XA commit in all engines? it's not its job. And it won't make XID available sooner, it'll only make XID available in a different function. > Notable changes: > > sql/handler.cc > - ha_prepare() now is ensured to execute binlog_hton::prepare() as > the last branch to prepare; recovery concerns are addressed > separately (not by MDEV-21469, but rather along the plan of bug#76233) > - read-only cases are handled more uniformly with extending > has_binlog_hton() that registers the binlog branch at ha_prepare() > when necessary > - ha_rollback_trans() executes binlog_hton::rollback() as first > branch, as a part of the 76233 planned recovery; of <what> planned recovery? > - ditto to the external completion of XA via ha_commit_or_rollback_by_xid(). > sql/log.cc > - binlog_commit,rollback() simplified/clarified on the condition of an empty > statement/trx > - MYSQL_BIN_LOG::run_commit_ordered() is extended to cover > XA-{commit, rollback} respective actions in Engines > - XID unlogging by XA is converted to employ the standard > binlog-checkpoint mechanism (cache_mngr->using_xa= TRUE) > sql/xa.cc > - trans_xa_commit,rollback() are refactored to invoke a common function > for external (connection) completion > - adapting the above two and slave_applier_reset_xa_trans() to > possible (normal to slave) XID release, inside run_ordered_commit(). > > diff --git a/sql/handler.cc b/sql/handler.cc > --- a/sql/handler.cc > +++ b/sql/handler.cc > @@ -1472,6 +1472,48 @@ static int prepare_or_error(handlerton *ht, THD *thd, bool all) > return err; > } > > +/** > + Search in @c ha_info transaction list branches for binlog. > + When the 2nd argument is not NULL and the modified engine number > + is non-zero the binlog branch is activated. > + > + @param ha_info The first node in the transaction descriptor list. > + @param thd The client thread that executes the transaction. > + > + @return true when the binlog branch is registered or set up for transaction, > + false otherwise. > +*/ > +static bool has_binlog_hton(Ha_trx_info *ha_info, THD* thd= NULL) > +{ > + bool rc; > + uint rw_count; > + > + for (rc= false, rw_count= 0; ha_info && !rc; ha_info= ha_info->next()) > + { > + rc= ha_info->ht() == binlog_hton; > + rw_count += (ha_info->is_trx_read_write()); > + } > + > + if (!rc && thd && !thd->is_current_stmt_binlog_disabled() && > + unlikely(rw_count > 0)) > + { > + /* > + In case an engine is modified but binlog is not part of xa-prepared > + transaction (thd is non-NULL) set out binlogging of an empty xa-prepare. > + */ when can we land here? non-trans (or non-XA) table modified? > + DBUG_ASSERT(thd->transaction->xid_state.get_state_code() == XA_IDLE); > + > + THD_TRANS *ptr_trans= &thd->transaction->all; > + thd->ha_data[binlog_hton->slot].ha_info[1].register_ha(ptr_trans, > + binlog_hton); > + thd->ha_data[binlog_hton->slot].ha_info[1].set_trx_read_write(); > + thd->binlog_setup_trx_data(); > + > + rc= true; > + } > + > + return rc; > +} > > /** > @retval > @@ -1488,9 +1530,16 @@ int ha_prepare(THD *thd) > > if (ha_info) > { > + int err; > + bool has_binlog= has_binlog_hton(ha_info, thd); > + > for (; ha_info; ha_info= ha_info->next()) > { > handlerton *ht= ha_info->ht(); > + > + if (ht == binlog_hton) > + continue; Does binlog need a handlerton anymore? It was an easy way to get it registered for participating in 2pc. But if you always filter it out, then there's no need to add it to ha_info list in the first place, is there? > + > if (ht->prepare) > { > if (unlikely(prepare_or_error(ht, thd, all))) > diff --git a/sql/log.cc b/sql/log.cc > --- a/sql/log.cc > +++ b/sql/log.cc > @@ -1895,6 +1896,11 @@ binlog_rollback_flush_trx_cache(THD *thd, bool all, > if (thd->transaction->xid_state.get_state_code() == XA_PREPARED) > buflen= serialize_with_xid(thd->transaction->xid_state.get_xid(), > buf, query, q_len); > + cache_mngr->using_xa= TRUE; > +#ifndef DBUG_OFF > + if (thd->killed != NOT_KILLED) > + thd->lex->sql_command= SQLCOM_XA_ROLLBACK; // for run_commit_ord asserts 1. don't abbrev, commits are for others not for you 2. what other values can sql_command have here? > +#endif > } > Query_log_event end_evt(thd, buf, buflen, TRUE, TRUE, TRUE, 0); > > @@ -2186,16 +2200,13 @@ int binlog_commit(THD *thd, bool all, bool ro_1pc) > } > > if (cache_mngr->trx_cache.empty() && > - (thd->transaction->xid_state.get_state_code() != XA_PREPARED || > - !(thd->ha_data[binlog_hton->slot].ha_info[1].is_started() && > - thd->ha_data[binlog_hton->slot].ha_info[1].is_trx_read_write()))) > + (!(is_preparing_xa(thd) || is_prepared_xa(thd)) || > + !is_binlog_read_write(thd))) > { > /* > - This is an empty transaction commit (both the regular and xa), > - or such transaction xa-prepare or > - either one's statement having no effect on the transactional cache > - as any prior to it. > - The empty xa-prepare sinks in only when binlog is read-only. > + This is an empty transaction/statement commit (both the regular and xa), > + or the xa transaction is running xa-prepare or xa-commit while > + its binlog branch is not read-write. weird. how and why do you get into binlog_commit on XA PREPARE? > */ > cache_mngr->reset(false, true); > THD_STAGE_INFO(thd, org_stage); > @@ -7752,7 +7763,7 @@ MYSQL_BIN_LOG::write_transaction_to_binlog(THD *thd, > entry.all= all; > entry.using_stmt_cache= using_stmt_cache; > entry.using_trx_cache= using_trx_cache; > - entry.need_unlog= is_preparing_xa(thd); > + entry.need_unlog= 0; is `need_unlog` always 0 now? > ha_info= all ? thd->transaction->all.ha_list : thd->transaction->stmt.ha_list; > entry.ro_1pc= is_ro_1pc; > entry.end_event= end_ev; > @@ -9365,18 +9381,102 @@ TC_LOG::run_commit_ordered(THD *thd, bool all) > all ? thd->transaction->all.ha_list : thd->transaction->stmt.ha_list; > > mysql_mutex_assert_owner(&LOCK_commit_ordered); > + > + if (!ha_info) > + { > + // slave or master's xa-"complete" external what does it mean? > + DBUG_ASSERT(thd->lex->sql_command == SQLCOM_XA_COMMIT || > + thd->lex->sql_command == SQLCOM_XA_ROLLBACK); > + > + XID *xid= thd->lex->xid; > + bool commit= thd->lex->sql_command == SQLCOM_XA_COMMIT; > + struct xahton_st xaop= { xid, 1 }; > + > + plugin_foreach(NULL, commit ? xacommit_handlerton : xarollback_handlerton, > + MYSQL_STORAGE_ENGINE_PLUGIN, &xaop); is it for commit/rollback by xid? > + xid_cache_delete(thd); > + > + DBUG_ASSERT(!thd->transaction->xid_state.is_explicit_XA()); > + > + return; > + } > + > + // else > + if (is_preparing_xa(thd)) > + { > + for (; ha_info; ha_info= ha_info->next()) > + { > + DBUG_ASSERT(all); > + DBUG_ASSERT(thd->lex->sql_command == SQLCOM_XA_PREPARE); > + DBUG_ASSERT(thd->transaction->xid_state.is_explicit_XA()); > + > + handlerton *ht= ha_info->ht(); > + if (!ht->prepare) > + { > + push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, > + ER_GET_ERRNO, ER_THD(thd, ER_GET_ERRNO), > + HA_ERR_WRONG_COMMAND, > + ha_resolve_storage_engine_name(ht)); > + continue; > + } > + else if (ht != binlog_hton) > + { > + /* leader only */ > + //if (thd->binlog_setup_trx_data()->last_commit_pos_offset == 0) > + // ht->prepare(ht, thd, all); wait, what? why do you need a loop over all engines if you don't do anything there? > + } > + } > + if (thd->rgi_slave && thd->rgi_slave->is_parallel_exec) > + { > + thd->transaction->xid_state.set_state_code(XA_PREPARED); > + thd->transaction->xid_state.set_cache_element_to_recovered(); > + thd->transaction->xid_state.xid_cache_element= 0; when are you here? > + } > + } > + else if (thd->transaction->xid_state.is_explicit_XA()) > + { > + for (; ha_info; ha_info= ha_info->next()) > + { > + DBUG_ASSERT(all); > + DBUG_ASSERT(thd->lex->sql_command == SQLCOM_XA_COMMIT || > + thd->lex->sql_command == SQLCOM_XA_ROLLBACK); > + DBUG_ASSERT(is_prepared_xa(thd) || > + ((thd->lex->sql_command == SQLCOM_XA_ROLLBACK && > + thd->transaction->xid_state.get_state_code() == XA_IDLE) || > + thd->lex->xa_opt == XA_ONE_PHASE)); > + DBUG_ASSERT(!thd->transaction->xid_state.is_recovered()); > + > + handlerton *ht= ha_info->ht(); > + if (thd->lex->sql_command == SQLCOM_XA_COMMIT) > + { > + if (ht->commit_ordered) > + ht->commit_ordered(ht, thd, all); > + } > + else if (ht != binlog_hton) why not? > + { > + ht->rollback(ht, thd, all); > + } > + } > + xid_cache_delete(thd); // trx is completed now > + > + DBUG_ASSERT(!thd->transaction->xid_state.is_explicit_XA()); > + } > + else > + { when are you here? > for (; ha_info; ha_info= ha_info->next()) > { > handlerton *ht= ha_info->ht(); > if (!ht->commit_ordered) > continue; > ht->commit_ordered(ht, thd, all); > DBUG_EXECUTE_IF("enable_log_write_upto_crash", > { > DBUG_SET_INITIAL("+d,crash_after_log_write_upto"); > sleep(1000); > }); > DEBUG_SYNC(thd, "commit_after_run_commit_ordered"); > } > + } > } > > diff --git a/sql/xa.cc b/sql/xa.cc > --- a/sql/xa.cc > +++ b/sql/xa.cc > @@ -319,9 +342,16 @@ static void xid_cache_delete(THD *thd, XID_cache_element *&element) > > void xid_cache_delete(THD *thd, XID_STATE *xid_state) > { > - DBUG_ASSERT(xid_state->is_explicit_XA()); > - xid_cache_delete(thd, xid_state->xid_cache_element); > - xid_state->xid_cache_element= 0; > + if (xid_state->is_explicit_XA()) why it should be done here and not in the caller? > + { > + xid_cache_delete(thd, xid_state->xid_cache_element); > + xid_state->xid_cache_element= 0; > + } > +} > + > +void xid_cache_delete(THD *thd) > +{ > + xid_cache_delete(thd, &thd->transaction->xid_state); > } > > > @@ -547,7 +577,14 @@ bool trans_xa_prepare(THD *thd) > } > else > { > - thd->transaction->xid_state.xid_cache_element->xa_state= XA_PREPARED; > + if (thd->transaction->xid_state.xid_cache_element) > + { > + thd->transaction->xid_state.xid_cache_element->xa_state= XA_PREPARED; > + } > + else > + { > + DBUG_ASSERT(thd->rgi_slave->is_parallel_exec); what do you mean by that? That parallel slave doesn't add XA transactions to the xid_cache? > + } > MYSQL_SET_TRANSACTION_XA_STATE(thd->m_transaction_psi, XA_PREPARED); > res= thd->variables.pseudo_slave_mode || thd->slave_thread ? > slave_applier_reset_xa_trans(thd) : 0; Regards, Sergei Chief Architect, MariaDB Server and security(a)mariadb.org [View Less]

1 0

Re: d2de9a89d80: MDEV-31531 Remove my_casedn_str() and my_caseup_str()
by Sergei Golubchik 19 Dec '23

19 Dec '23

Hi, Alexander, Looks good. A couple of minor comments below On Dec 19, Alexander Barkov wrote: > commit d2de9a89d80 > Author: Alexander Barkov <bar(a)mariadb.com> > Date: Fri Jun 23 13:24:02 2023 +0400 > > diff --git a/include/m_ctype.h b/include/m_ctype.h > index 9d13989d1fe..65ccf869992 100644 > --- a/include/m_ctype.h > +++ b/include/m_ctype.h > @@ -1512,6 +1540,14 @@ size_t my_copy_fix_mb(CHARSET_INFO *cs, > /* Functions for 8bit */ > extern … [View More]size_t my_caseup_str_8bit(CHARSET_INFO *, char *); > extern size_t my_casedn_str_8bit(CHARSET_INFO *, char *); > +static inline size_t my_caseup_str_latin1(char *str) caseup variant doesn't seem to be used anywhere > +{ > + return my_caseup_str_8bit(&my_charset_latin1, str); > +} > +static inline size_t my_casedn_str_latin1(char *str) > +{ > + return my_casedn_str_8bit(&my_charset_latin1, str); > +} > extern size_t my_caseup_8bit(CHARSET_INFO *, > const char *src, size_t srclen, > char *dst, size_t dstlen); > diff --git a/sql/sql_acl.cc b/sql/sql_acl.cc > index 09af5523c76..3166c9e56cf 100644 > --- a/sql/sql_acl.cc > +++ b/sql/sql_acl.cc > @@ -2543,6 +2541,37 @@ static void push_new_user(const ACL_USER &user) > } > > > +/** > + Make a database name on mem_root from a String, > + apply lower-case conversion if lower_case_table_names says so. > + Perform database name length limit validation. > + > + @param thd - the THD, to get the warning text from > + @param mem_root - allocate the result on this memory root > + @param dbstr - the String, e.g. with Field::val_str() result > + > + @return - {NULL,0} in case of EOM or a bad database name, > + or a good database name otherwise. > +*/ > +static LEX_STRING make_and_check_db_name(MEM_ROOT *mem_root, > + const String &dbstr) > +{ > + LEX_STRING dbls= lower_case_table_names ? > + lex_string_casedn_root(mem_root, files_charset_info, > + dbstr.ptr(), dbstr.length()) : > + lex_string_strmake_root(mem_root, > + dbstr.ptr(), dbstr.length()); > + if (!dbls.str) > + return LEX_STRING{NULL, 0}; // EOM what's the difference between this and return {NULL, 0}; > + if (dbls.length > SAFE_NAME_LEN) > + { > + sql_print_warning(ER_DEFAULT(ER_WRONG_DB_NAME), dbls.str); > + return LEX_STRING{NULL, 0}; // Bad name > + } > + return dbls; // Good name > +} > + > + > /* > Initialize structures responsible for user/db-level privilege checking > and load information about grants from open privilege tables. > diff --git a/sql/sql_lex.cc b/sql/sql_lex.cc > index a84fd29affc..f61e5085e47 100644 > --- a/sql/sql_lex.cc > +++ b/sql/sql_lex.cc > @@ -4186,6 +4187,18 @@ bool LEX::copy_db_to(LEX_CSTRING *to) > return thd->copy_db_to(to); > } > > + > +Lex_ident_db_normalized LEX::copy_db_normalized() > +{ > + if (sphead && sphead->m_name.str) > + { > + DBUG_ASSERT(sphead->m_db.str && sphead->m_db.length); you missed that DBUG_ASSERT(sphead->m_db.str); DBUG_ASSERT(sphead->m_db.length); > + return thd->to_ident_db_normalized_with_error(sphead->m_db); > + } > + return thd->copy_db_normalized(); > +} > + > + > /** > Initialize offset and limit counters. > > diff --git a/storage/perfschema/pfs_program.cc b/storage/perfschema/pfs_program.cc > index de456610519..f3734cb383b 100644 > --- a/storage/perfschema/pfs_program.cc > +++ b/storage/perfschema/pfs_program.cc > @@ -118,31 +118,21 @@ static void set_program_key(PFS_program_key *key, > */ > > char *ptr= &key->m_hash_key[0]; > + const char *end= ptr + sizeof(key->m_hash_key) - 1; > > ptr[0]= object_type; > ptr++; > > if (object_name_length > 0) > - { > - char tmp_object_name[COL_OBJECT_NAME_SIZE + 1]; > - memcpy(tmp_object_name, object_name, object_name_length); > - tmp_object_name[object_name_length]= '\0'; > - my_casedn_str(system_charset_info, tmp_object_name); > - memcpy(ptr, tmp_object_name, object_name_length); > - ptr+= object_name_length; > - } > + ptr+= system_charset_info->casedn(object_name, object_name_length, > + ptr, end - ptr); > ptr[0]= 0; > ptr++; > > if (schema_name_length > 0) > - { > - char tmp_schema_name[COL_OBJECT_SCHEMA_SIZE + 1]; > - memcpy(tmp_schema_name, schema_name, schema_name_length); > - tmp_schema_name[schema_name_length]='\0'; > - my_casedn_str(system_charset_info, tmp_schema_name); > - memcpy(ptr, tmp_schema_name, schema_name_length); > - ptr+= schema_name_length; > - } > + ptr+= system_charset_info->opt_casedn(schema_name, schema_name_length, > + ptr, end - ptr, > + lower_case_table_names); why system_charset_info and not files_charset_info ? > ptr[0]= 0; > ptr++; > Regards, Sergei Chief Architect, MariaDB Server and security(a)mariadb.org [View Less]

1 0

MTR failures on official Debian buildd after MariaDB 1:10.11.6-1 upload
by Otto Kekäläinen 17 Dec '23

17 Dec '23

Hi! I just wanted to let you know there are a bunch of MTR failures/regressions at https://buildd.debian.org/status/package.php?p=mariadb which I need to fix or add to skip-lists. The previous upload of MariaDB 1:10.11.5-3 had more 'green' than ever before, and now there are regressions on 6 archs. [image: image.png]

3 6

Re: 2821cf816a4: MDEV-31531 Remove my_casedn_str() and my_caseup_str()
by Sergei Golubchik 16 Dec '23

16 Dec '23

Hi, Alexander, > commit 2821cf816a4 > Author: Alexander Barkov <bar(a)mariadb.com> > Date: Fri Jun 23 13:24:02 2023 +0400 > > MDEV-31531 Remove my_casedn_str() and my_caseup_str() please, explain here why > > diff --git a/include/m_ctype.h b/include/m_ctype.h > index 9d13989d1fe..d2629121cfe 100644 > --- a/include/m_ctype.h > +++ b/include/m_ctype.h > @@ -848,6 +844,26 @@ struct charset_info_st > return (cset->casedn)(this, src, … [View More]srclen, dst, dstlen); > } > > + /* Convert to a lower-cased 0-terminated string */ > + size_t casedn_z(const char *src, size_t srclen, > + char *dst, size_t dstlen) const > + { > + DBUG_ASSERT(dstlen); DBUG_ASSERT(src != dst) ? > + size_t len= casedn(src, srclen, dst, dstlen - 1); > + dst[len]= '\0'; > + return len; > + } > + > + /* Convert to a upper-cased 0-terminated string */ > + size_t caseup_z(const char *src, size_t srclen, > + char *dst, size_t dstlen) const > + { > + DBUG_ASSERT(dstlen); > + size_t len= caseup(src, srclen, dst, dstlen - 1); > + dst[len]= '\0'; > + return len; > + } > + > uint caseup_multiply() const > { > return (cset->caseup_multiply)(this); > @@ -1512,6 +1528,14 @@ size_t my_copy_fix_mb(CHARSET_INFO *cs, > /* Functions for 8bit */ > extern size_t my_caseup_str_8bit(CHARSET_INFO *, char *); > extern size_t my_casedn_str_8bit(CHARSET_INFO *, char *); > +static inline size_t my_caseup_str_latin1(char *str) > +{ > + return my_caseup_str_8bit(&my_charset_latin1, str); > +} > +static inline size_t my_casedn_str_latin1(char *str) > +{ > + return my_casedn_str_8bit(&my_charset_latin1, str); > +} may be my_charset_ascii? with an assert that it's indeed ascii? and a comment that it's used for generated temp table names and such > extern size_t my_caseup_8bit(CHARSET_INFO *, > const char *src, size_t srclen, > char *dst, size_t dstlen); > diff --git a/sql/char_buffer.h b/sql/char_buffer.h > index 7b855f5401c..1091775259d 100644 > --- a/sql/char_buffer.h > +++ b/sql/char_buffer.h > @@ -71,25 +76,92 @@ class CharBuffer > m_buff[m_length]= '\0'; > return *this; > } > + CharBuffer<buff_sz> & copy_caseup(CHARSET_INFO *cs, const LEX_CSTRING &str) > + { > + DBUG_ASSERT(!buffer_overlaps(str)); > + m_length= cs->cset->caseup(cs, str.str, str.length, m_buff, buff_sz); > + DBUG_ASSERT(is_sane()); > + m_buff[m_length]= '\0'; // See comments in copy_casedn() > + return *this; > + } > CharBuffer<buff_sz> & copy_casedn(CHARSET_INFO *cs, const LEX_CSTRING &str, > bool casedn) > { > casedn ? copy_casedn(cs, str) : copy_bin(str); > return *this; > } > + > + // Append one character > + CharBuffer<buff_sz> & append_char(char ch) > + { > + DBUG_ASSERT(is_sane()); > + if (available_size()) > + { > + m_buff[m_length++]= ch; > + m_buff[m_length]= '\0'; > + } > + DBUG_ASSERT(is_sane()); > + return *this; > + } > + > + // Append a string > + CharBuffer<buff_sz> & append_bin(const LEX_CSTRING &str) why _bin ? and why even have a suffix, and not just append() here and above? > + { > + DBUG_ASSERT(is_sane()); > + DBUG_ASSERT(!buffer_overlaps(str)); > + size_t len= MY_MIN(available_size(), str.length); > + memcpy(m_buff + m_length, str.str, len); > + m_length+= len; > + DBUG_ASSERT(is_sane()); > + m_buff[m_length]= '\0'; > + return *this; > + } > + > // Append a string with casedn conversion > CharBuffer<buff_sz> & append_casedn(CHARSET_INFO *cs, const LEX_CSTRING &str) > { > DBUG_ASSERT(is_sane()); > DBUG_ASSERT(!buffer_overlaps(str)); > size_t casedn_length= cs->casedn(str.str, str.length, > - m_buff + m_length, buff_sz - m_length); > + m_buff + m_length, available_size()); > + m_length+= casedn_length; > + DBUG_ASSERT(is_sane()); > + m_buff[m_length]= '\0'; > + return *this; > + } > + > + CharBuffer<buff_sz> & append_opt_casedn(CHARSET_INFO *cs, > + const LEX_CSTRING &str, > + bool casedn) > + { > + return casedn ? append_casedn(cs, str) : append_bin(str); > + } > + > + // Append a string with caseup conversion > + CharBuffer<buff_sz> & append_caseup(CHARSET_INFO *cs, const LEX_CSTRING &str) > + { > + DBUG_ASSERT(is_sane()); > + DBUG_ASSERT(!buffer_overlaps(str)); > + size_t casedn_length= cs->caseup(str.str, str.length, > + m_buff + m_length, available_size()); > m_length+= casedn_length; > DBUG_ASSERT(is_sane()); > m_buff[m_length]= '\0'; > return *this; > } > > + CharBuffer<buff_sz> & truncate(size_t length) > + { > + DBUG_ASSERT(is_sane()); > + if (m_length > length) > + { > + m_length= length; > + m_buff[m_length]= '\0'; > + DBUG_ASSERT(is_sane()); > + } > + return *this; > + } > + > LEX_CSTRING to_lex_cstring() const > { > return LEX_CSTRING{m_buff, m_length}; > diff --git a/sql/item_sum.cc b/sql/item_sum.cc > index bbd09a59267..824ed42ec91 100644 > --- a/sql/item_sum.cc > +++ b/sql/item_sum.cc > @@ -3769,20 +3769,15 @@ int group_concat_key_cmp_with_order_with_nulls(void *arg, const void *key1_arg, > } > > > -static void report_cut_value_error(THD *thd, uint row_count, const char *fname) > +static void report_cut_value_error(THD *thd, uint row_count, > + const LEX_CSTRING &fname) > { > - size_t fn_len= strlen(fname); > - char *fname_upper= (char *) my_alloca(fn_len + 1); > - if (!fname_upper) > - fname_upper= (char*) fname; // Out of memory > - else > - memcpy(fname_upper, fname, fn_len+1); > - my_caseup_str(&my_charset_latin1, fname_upper); > + CharBuffer<NAME_LEN> fname_upper; > + fname_upper.copy_caseup(&my_charset_latin1, fname); why bother? we don't do it in other error cases > push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, > ER_CUT_VALUE_GROUP_CONCAT, > ER_THD(thd, ER_CUT_VALUE_GROUP_CONCAT), > - row_count, fname_upper); > + row_count, fname_upper.ptr()); > - my_afree(fname_upper); > } > > > diff --git a/sql/lex_ident.h b/sql/lex_ident.h > index f39273b7da5..db96e40cab8 100644 > --- a/sql/lex_ident.h > +++ b/sql/lex_ident.h > @@ -45,6 +46,10 @@ class Lex_ident_fs: public LEX_CSTRING > bool is_in_lower_case() const; > bool ok_for_lower_case_names() const; > #endif > + bool check_db_name_quick() const please, use a name that makes it clear what the return value is Not if (str.check_db_name_quick()) but if (str.is_invalid_db_name()) > + { > + return !length || length > NAME_LEN || str[length-1] == ' '; > + } > }; > > > @@ -160,4 +171,95 @@ class DBNameBuffer: public CharBuffer<SAFE_NAME_LEN + MY_CS_MBMAXLEN> > }; > > > +class Identifier_chain2 > +{ > + LEX_CSTRING m_name[2]; > +public: > + Identifier_chain2() > + :m_name{Lex_cstring(), Lex_cstring()} > + { } > + Identifier_chain2(const LEX_CSTRING &a, const LEX_CSTRING &b) > + :m_name{a, b} > + { } > + > + const LEX_CSTRING& operator [] (size_t i) const > + { > + return m_name[i]; > + } > + > + static Identifier_chain2 split(const LEX_CSTRING &txt) > + { > + DBUG_ASSERT(txt.str[txt.length] == '\0'); // Expect 0-terminated input > + const char *dot= strchr(txt.str, '.'); > + if (!dot) > + return Identifier_chain2(Lex_cstring(), txt); > + size_t length0= dot - txt.str; > + Lex_cstring name0(txt.str, length0); > + Lex_cstring name1(txt.str + length0 + 1, txt.length - length0 - 1); > + return Identifier_chain2(name0, name1); > + } > + > + // Export as a qualified name string: 'db.name' > + size_t make_qname(char *dst, size_t dstlen) const > + { > + return my_snprintf(dst, dstlen, "%.*s.%.*s", > + (int) m_name[0].length, m_name[0].str, > + (int) m_name[1].length, m_name[1].str); > + } > + > + // Export as a qualified name string, allocate on mem_root. > + bool make_qname(MEM_ROOT *mem_root, LEX_CSTRING *dst) const > + { > + const uint dot= !!m_name[0].length; > + char *tmp; > + /* format: [pkg + dot] + name + '\0' */ > + dst->length= m_name[0].length + dot + m_name[1].length; > + if (unlikely(!(dst->str= tmp= (char*) alloc_root(mem_root, > + dst->length + 1)))) > + return true; > + snprintf(tmp, dst->length + 1, "%.*s%.*s%.*s", > + (int) m_name[0].length, (m_name[0].length ? m_name[0].str : ""), > + dot, ".", > + (int) m_name[1].length, m_name[1].str); > + return false; > + } > + > + /* > + Build a separated two step name, e.g. "ident1/ident2", 0-terminated. > + with an optional lower-case conversion. may be should be inside InnoDB? > + @param [OUT] dst - the destination > + @param dst_size - number of bytes available in dst > + @param sep - the separator character > + @param casedn - whether to convert components to lower case > + @return - number of bytes written to "dst", not counting > + the trailing '\0' byte. > + */ > + size_t make_sep_name_opt_casedn(char *dst, size_t dst_size, > + int sep, bool casedn) const > + { > + /* > + Minimal possible buffer is 4 bytes: 'd/t\0' > + where 'd' is the database name, 't' is the table name. > + Callers should never pass a smaller buffer. > + */ > + DBUG_ASSERT(dst_size > 4); > + DBUG_ASSERT(m_name[0].length + m_name[1].length + 2 < FN_REFLEN - 1); > + if (casedn) > + { > + CHARSET_INFO *cs= &my_charset_utf8mb3_general_ci; > + char *ptr= dst, *end= dst + dst_size; > + ptr+= cs->casedn(m_name[0].str, m_name[0].length, ptr, end - ptr - 2); > + *ptr++= (char) sep; > + ptr+= cs->casedn_z(m_name[1].str, m_name[1].length, ptr, end - ptr); > + return ptr - dst; > + } > + memcpy(dst, m_name[0].str, m_name[0].length); > + dst[m_name[0].length] = (char) sep; > + /* Copy the name and null-byte. */ > + memcpy(dst + m_name[0].length + 1, m_name[1].str, m_name[1].length + 1); > + return m_name[0].length + 1/*slash*/ + m_name[1].length; s/slash/sep/ > + } > +}; > + > + > #endif // LEX_IDENT_INCLUDED > diff --git a/sql/log_event_server.cc b/sql/log_event_server.cc > index 6030202adc7..0fc1d4ce962 100644 > --- a/sql/log_event_server.cc > +++ b/sql/log_event_server.cc > @@ -5777,19 +5778,29 @@ int Table_map_log_event::do_apply_event(rpl_group_info *rgi) > /* Step the query id to mark what columns that are actually used. */ > thd->set_query_id(next_query_id()); > > + db_mem_alloced= tname_mem_alloced= NAME_LEN * > + (lower_case_table_names ? 1 : > + files_charset_info->casedn_multiply()) > + + 1 /* for '\0' */; 1. may be db_mem_length/tname_mem_length instead of NAME_LEN? 2. is the ?: condition right? Looks like it's inverted > if (!(memory= my_multi_malloc(PSI_INSTRUMENT_ME, MYF(MY_WME), > &table_list, (uint) sizeof(RPL_TABLE_LIST), > - &db_mem, (uint) NAME_LEN + 1, > - &tname_mem, (uint) NAME_LEN + 1, > + &db_mem, (uint) db_mem_alloced, > + &tname_mem, (uint) tname_mem_alloced, > NullS))) > DBUG_RETURN(HA_ERR_OUT_OF_MEM); > > - db_mem_length= strmov(db_mem, m_dbnam) - db_mem; > - tname_mem_length= strmov(tname_mem, m_tblnam) - tname_mem; > if (lower_case_table_names) > { > - my_casedn_str(files_charset_info, (char*)tname_mem); > - my_casedn_str(files_charset_info, (char*)db_mem); > + db_mem_length= files_charset_info->casedn_z(m_dbnam, m_dblen, > + db_mem, db_mem_alloced); > + tname_mem_length= files_charset_info->casedn_z(m_tblnam, m_tbllen, > + tname_mem, > + tname_mem_alloced); > + } > + else > + { > + db_mem_length= strmov(db_mem, m_dbnam) - db_mem; > + tname_mem_length= strmov(tname_mem, m_tblnam) - tname_mem; > } > > /* call from mysql_client_binlog_statement() will not set rli->mi */ > diff --git a/sql/sp.cc b/sql/sp.cc > index 0a8aa24789b..537d29fb67e 100644 > --- a/sql/sp.cc > +++ b/sql/sp.cc > @@ -2281,12 +2281,20 @@ Sp_handler::sp_exist_routines(THD *thd, TABLE_LIST *routines) const > for (routine= routines; routine; routine= routine->next_global) > { > sp_name *name; > - LEX_CSTRING lex_db; > + LEX_CSTRING lex_db= thd->make_ident_opt_casedn(routine->db, > + lower_case_table_names); > + if (!lex_db.str) > + DBUG_RETURN(TRUE); // EOM // EOM, error was already sent right? (also, in other places) > LEX_CSTRING lex_name; > - thd->make_lex_string(&lex_db, routine->db.str, routine->db.length); > thd->make_lex_string(&lex_name, routine->table_name.str, > routine->table_name.length); Rather inconsistent :( May be, makes sense to create LEX_CSTRING make_lex_string(const char* str, size_t length) const ? > + /* > + routine->db was earlier tested with check_db_name(). > + Now it's lower-cased according to lower_case_table_names. > + It's safe to make a Lex_ident_db_normalized. > + */ > - name= new sp_name(&lex_db, &lex_name, true); > + name= new (thd->mem_root) sp_name(Lex_ident_db_normalized(lex_db), > + lex_name, true); > sp_object_found= sp_find_routine(thd, name, false) != NULL; > thd->get_stmt_da()->clear_warning_info(thd->query_id); > if (! sp_object_found) > diff --git a/sql/sql_acl.cc b/sql/sql_acl.cc > index 09af5523c76..70dde7dc0ed 100644 > --- a/sql/sql_acl.cc > +++ b/sql/sql_acl.cc > @@ -2543,6 +2543,37 @@ static void push_new_user(const ACL_USER &user) > } > > > +/** > + Make a database name on mem_root from a String, > + apply lower-case conversion if lower_case_table_names says so. > + Perform database name length limit validation. > + > + @param thd - the THD, to get the warning text from > + @param mem_root - allocate the result on this memory root > + @param dbstr - the String, e.g. with Field::val_str() result > + > + @return - {NULL,0} in case of EOM or a bad database name, > + or a good database name otherwise. > +*/ > +static LEX_STRING make_and_check_db_name(THD *thd, MEM_ROOT *mem_root, > + const String &dbstr) > +{ > + LEX_STRING dbls= lower_case_table_names ? > + lex_string_casedn_root(mem_root, files_charset_info, > + dbstr.ptr(), dbstr.length()) : > + lex_string_strmake_root(mem_root, > + dbstr.ptr(), dbstr.length()); > + if (!dbls.str) > + return LEX_STRING{NULL, 0}; // EOM > + if (dbls.length > SAFE_NAME_LEN) > + { > + sql_print_warning(ER_THD(thd, ER_WRONG_DB_NAME), dbls.str); It's an error log, should every message be localized according to connection-specific locale? > + return LEX_STRING{NULL, 0}; // Bad name > + } > + return dbls; // Good name > +} > + > + > /* > Initialize structures responsible for user/db-level privilege checking > and load information about grants from open privilege tables. > @@ -3680,26 +3702,27 @@ privilege_t acl_get(const char *host, const char *ip, > { > privilege_t host_access(ALL_KNOWN_ACL), db_access(NO_ACL); > uint i; > - size_t key_length; > - char key[ACL_KEY_LENGTH],*tmp_db,*end; > + const char *tmp_db; > acl_entry *entry; > DBUG_ENTER("acl_get"); > > - tmp_db= strmov(strmov(key, safe_str(ip)) + 1, user) + 1; > - end= strnmov(tmp_db, db, key + sizeof(key) - tmp_db); > - > - if (end >= key + sizeof(key)) // db name was truncated > + constexpr size_t key_data_size= ACL_KEY_LENGTH + 2; > + CharBuffer<key_data_size + MY_CS_MBMAXLEN> key; why ACL_KEY_LENGTH + 2 + MY_CS_MBMAXLEN ? > + key.append_bin(Lex_cstring_strlen(safe_str(ip))); > + key.append_char('\0'); > + key.append_bin(Lex_cstring_strlen(user)); > + key.append_char('\0'); there was strmov(strmov(...)) pattern here before. as far as I can see, you can chain too: key.append_bin().append_char() .append_bin().append_char(); let's use that > + tmp_db= key.end(); > + key.append_opt_casedn(files_charset_info, Lex_cstring_strlen(db), > + lower_case_table_names); > + db= tmp_db; > + > + if (key.length() >= key_data_size) // db name was truncated > DBUG_RETURN(NO_ACL); // no privileges for an invalid db name > > - if (lower_case_table_names) > - { > - my_casedn_str(files_charset_info, tmp_db); > - db=tmp_db; > - } > - key_length= (size_t) (end-key); > - > mysql_mutex_lock(&acl_cache->lock); > - if (!db_is_pattern && (entry=acl_cache->search((uchar*) key, key_length))) > + if (!db_is_pattern && > + (entry= acl_cache->search((uchar*) key.ptr(), key.length()))) > { > db_access=entry->access; > mysql_mutex_unlock(&acl_cache->lock); > @@ -5630,28 +5662,26 @@ static GRANT_NAME *name_hash_search(HASH *name_hash, > const char *user, const char *tname, > bool exact, bool name_tolower) > { > - char helping[SAFE_NAME_LEN*2+USERNAME_LENGTH+3]; > - char *hend = helping + sizeof(helping); > - uint len; > + constexpr size_t key_data_size= SAFE_NAME_LEN * 2 + USERNAME_LENGTH + 3; > + CharBuffer<key_data_size + MY_CS_MBMAXLEN> key; > GRANT_NAME *grant_name,*found=0; > HASH_SEARCH_STATE state; > > - char *db_ptr= strmov(helping, user) + 1; > - char *tname_ptr= strnmov(db_ptr, db, hend - db_ptr) + 1; > - if (tname_ptr > hend) > - return 0; // invalid name = not found > - char *end= strnmov(tname_ptr, tname, hend - tname_ptr) + 1; > - if (end > hend) > + key.append_bin(Lex_cstring_strlen(user)); > + key.append_char('\0'); > + key.append_bin(Lex_cstring_strlen(db)); is db already normalized? If yes, then, please, document it with DBUG_ASSERT(ok_for_lower_case_names()); > + key.append_char('\0'); > + key.append_opt_casedn(files_charset_info, Lex_cstring_strlen(tname), > + name_tolower); > + key.append_char('\0'); > + if (key.length() >= key_data_size) > return 0; // invalid name = not found > > - len = (uint) (end - helping); > - if (name_tolower) > - my_casedn_str(files_charset_info, tname_ptr); > - for (grant_name= (GRANT_NAME*) my_hash_first(name_hash, (uchar*) helping, > - len, &state); > + for (grant_name= (GRANT_NAME*) my_hash_first(name_hash, (uchar*) key.ptr(), > + key.length(), &state); > grant_name ; > - grant_name= (GRANT_NAME*) my_hash_next(name_hash,(uchar*) helping, > - len, &state)) > + grant_name= (GRANT_NAME*) my_hash_next(name_hash, (uchar*) key.ptr(), > + key.length(), &state)) > { > if (exact) > { > @@ -8826,33 +8856,28 @@ static bool check_grant_db_routine(THD *thd, const char *db, HASH *hash) > bool check_grant_db(THD *thd, const char *db) > { > Security_context *sctx= thd->security_ctx; > - char helping [SAFE_NAME_LEN + USERNAME_LENGTH+2], *end; > - char helping2 [SAFE_NAME_LEN + USERNAME_LENGTH+2], *tmp_db; > - uint len, UNINIT_VAR(len2); > + constexpr size_t key_data_size= SAFE_NAME_LEN + USERNAME_LENGTH + 2; > + CharBuffer<key_data_size + MY_CS_MBMAXLEN> key, key2; > bool error= TRUE; > > - tmp_db= strmov(helping, sctx->priv_user) + 1; > - end= strnmov(tmp_db, db, helping + sizeof(helping) - tmp_db); > - > - if (end >= helping + sizeof(helping)) // db name was truncated > - return 1; // no privileges for an invalid db name > - > - if (lower_case_table_names) > - { > - end = tmp_db + my_casedn_str(files_charset_info, tmp_db); > - db=tmp_db; > - } > + key.append_bin(Lex_cstring_strlen(sctx->priv_user)); > + key.append_char('\0'); > + key.append_opt_casedn(files_charset_info, Lex_cstring_strlen(db), > + lower_case_table_names); > + key.append_char('\0'); > > - len= (uint) (end - helping) + 1; > + if (key.length() >= key_data_size) // db name was truncated > + return 1; // no privileges for an invalid db name > > /* > If a role is set, we need to check for privileges here as well. > */ > if (sctx->priv_role[0]) > { > - end= strmov(helping2, sctx->priv_role) + 1; > - end= strnmov(end, db, helping2 + sizeof(helping2) - end); > - len2= (uint) (end - helping2) + 1; > + key2.append_bin(Lex_cstring_strlen(sctx->priv_role)); > + key2.append_char('\0'); > + key2.append_bin(Lex_cstring_strlen(db)); No append_opt_casedn() here? Looks wrong. > + key2.append_char('\0'); > } > > > diff --git a/sql/sql_class.h b/sql/sql_class.h > index 214118d8431..decb9a9064e 100644 > --- a/sql/sql_class.h > +++ b/sql/sql_class.h > @@ -4918,11 +4948,46 @@ class THD: public THD_count, /* this must be first */ > my_message(ER_NO_DB_ERROR, ER(ER_NO_DB_ERROR), MYF(0)); > return TRUE; > } > + return false; > + } > + > + /* > + Copy the current database to the argument. Use the current arena to > + allocate memory for a deep copy: current database may be freed after > + a statement is parsed but before it's executed. > + > + Can only be called by owner of thd (no mutex protection) > + */ > + bool copy_db_to(LEX_CSTRING *to) > + { > + if (check_if_current_db_is_set_with_error()) > + return true; > > to->str= strmake(db.str, db.length); > to->length= db.length; > return to->str == NULL; /* True on error */ > } > + > + /* > + Make a normalized copy of the current database. > + Raise an error if no current database is set. > + Note, in case of lower_case_table_names==2, thd->db can contain the > + name in arbitrary case typed by the user, so it must be lower-cased. > + For other lower_case_table_names values the name is already in > + its normalized case, so it's copied as is. > + */ > + Lex_ident_db_normalized copy_db_normalized() > + { > + if (check_if_current_db_is_set_with_error()) > + return Lex_ident_db_normalized(); > + LEX_STRING ident= make_ident_opt_casedn(db, lower_case_table_names == 2); LEX_STRING or LEX_CSTRING ? > + /* > + A non-empty thd->db is always known to satisfy check_db_name(). > + So after optional lower-casing above it's safe to > + make Lex_ident_db_normalized. > + */ > + return Lex_ident_db_normalized(ident); > + } > /* Get db name or "". Use for printing current db */ > const char *get_db() > { return safe_str(db.str); } > diff --git a/sql/sql_lex.cc b/sql/sql_lex.cc > index a84fd29affc..8cafad65538 100644 > --- a/sql/sql_lex.cc > +++ b/sql/sql_lex.cc > @@ -4186,6 +4186,18 @@ bool LEX::copy_db_to(LEX_CSTRING *to) > return thd->copy_db_to(to); > } > > + > +Lex_ident_db_normalized LEX::copy_db_normalized() > +{ > + if (sphead && sphead->m_name.str) > + { > + DBUG_ASSERT(sphead->m_db.str && sphead->m_db.length); Please, DBUG_ASSERT(sphead->m_db.str); DBUG_ASSERT(sphead->m_db.length); > + return thd->to_ident_db_normalized_with_error(sphead->m_db); > + } > + return thd->copy_db_normalized(); > +} > + > + > /** > Initialize offset and limit counters. > > diff --git a/sql/sql_show.cc b/sql/sql_show.cc > index 1fcafd6ec08..0f39a13632a 100644 > --- a/sql/sql_show.cc > +++ b/sql/sql_show.cc > @@ -3856,30 +3859,31 @@ static bool show_status_array(THD *thd, const char *wild, > names until lp:1306875 has been fixed. > TODO: remove once lp:1306875 has been addressed. > */ > - if (!(*prefix) && !strncasecmp(name_buffer, "wsrep", strlen("wsrep"))) > + if (!prefix.length && > + !strncasecmp(name_buffer.ptr(), "wsrep", strlen("wsrep"))) > { > is_wsrep_var= TRUE; > } > #endif /* WITH_WSREP */ > > if (ucase_names) > - my_caseup_str(system_charset_info, name_buffer); > + name_buffer.append_caseup(system_charset_info, var_name); > else > { > - my_casedn_str(system_charset_info, name_buffer); > - DBUG_ASSERT(name_buffer[0] >= 'a'); > - DBUG_ASSERT(name_buffer[0] <= 'z'); > - > + name_buffer.append_casedn(system_charset_info, var_name); > // WSREP_TODO: remove once lp:1306875 has been addressed. > if (IF_WSREP(is_wsrep_var == FALSE, 1) && > status_var) > - name_buffer[0]-= 'a' - 'A'; > + { > + char *ptr= (char*) name_buffer.ptr(); > + if (ptr[0] >= 'a' && ptr[0] <= 'z') > + ptr[0]-= 'a' - 'A'; > + } is it still needed? > } > > > restore_record(table, s->default_values); > - table->field[0]->store(name_buffer, strlen(name_buffer), > - system_charset_info); > + table->field[0]->store(name_buffer.to_lex_cstring(), system_charset_info); > > /* > Compare name for types that can't return arrays. We do this to not > @@ -4355,18 +4361,15 @@ bool get_lookup_field_values(THD *thd, COND *cond, bool fix_table_name_case, > > if (lower_case_table_names && !rc) > { > - /* > - We can safely do in-place upgrades here since all of the above cases > - are allocating a new memory buffer for these strings. > - */ > if (lookup_field_values->db_value.str && lookup_field_values->db_value.str[0]) > - my_casedn_str(system_charset_info, > - (char*) lookup_field_values->db_value.str); > + lookup_field_values->db_value= thd->make_ident_casedn( > + lookup_field_values->db_value); > + > if (fix_table_name_case && > lookup_field_values->table_value.str && > lookup_field_values->table_value.str[0]) > - my_casedn_str(system_charset_info, > - (char*) lookup_field_values->table_value.str); > + lookup_field_values->table_value= thd->make_ident_casedn( > + lookup_field_values->table_value); Why is it even needed? If db/table names are compared case insensitively, it shouldn't matter if they're lowercased or not, should it? > } > > return rc; > diff --git a/sql/sql_type_fixedbin.h b/sql/sql_type_fixedbin.h > index 8aaa7582cd5..e9728b78a3d 100644 > --- a/sql/sql_type_fixedbin.h > +++ b/sql/sql_type_fixedbin.h > @@ -267,8 +267,7 @@ class Type_handler_fbt: public Type_handler > void print(String *str, enum_query_type query_type) override > { > StringBuffer<FbtImpl::max_char_length()+64> tmp; > - tmp.append(singleton()->name().lex_cstring()); > - my_caseup_str(&my_charset_latin1, tmp.c_ptr()); > + tmp.copy_caseup(&my_charset_latin1, singleton()->name().lex_cstring()); why bother? > str->append(tmp); > str->append('\''); > m_value.to_string(&tmp); > diff --git a/storage/innobase/dict/dict0load.cc b/storage/innobase/dict/dict0load.cc > index bf59fed9ee4..104a8dc8780 100644 > --- a/storage/innobase/dict/dict0load.cc > +++ b/storage/innobase/dict/dict0load.cc > @@ -2955,7 +2955,7 @@ dict_load_foreign( > > foreign->foreign_table_name = mem_heap_strdupl( > foreign->heap, (char*) field, len); > - dict_mem_foreign_table_name_lookup_set(foreign, TRUE); > + foreign->foreign_table_name_lookup_set(); please, ask Marko to review InnoDB changes > > const size_t foreign_table_name_len = len; > const size_t table_name_len = strlen(table_name); > diff --git a/storage/perfschema/pfs_instr_class.h b/storage/perfschema/pfs_instr_class.h > index f353c410d4c..d083db63968 100644 > --- a/storage/perfschema/pfs_instr_class.h > +++ b/storage/perfschema/pfs_instr_class.h > @@ -265,6 +265,40 @@ struct PFS_table_share_key > char m_hash_key[PFS_TABLESHARE_HASHKEY_SIZE]; > /** Length in bytes of @c m_hash_key. */ > uint m_key_length; > + > + size_t available_length() const > + { > + return sizeof(m_hash_key) - m_key_length; > + } > + char *end() > + { > + return m_hash_key + m_key_length; > + } > + // Append and 0-terminate a string with an optional lower-case conversion > + void append_opt_casedn_z(CHARSET_INFO *cs, > + const char *str, size_t length, > + bool casedn) > + { > + DBUG_ASSERT(length <= sizeof(m_hash_key)); // Expect valid db/tbl names > + MY_STRCOPY_STATUS st; > + size_t dst_length= available_length(); > + if (dst_length > 0) > + { > + dst_length--; > + /* > + The code below uses copy_fix() instead of memcpy() to make > + sure we don't break a multi-byte character in the middle. > + */ > + m_key_length+= (uint) (casedn ? > + cs->casedn(str, length, end(), dst_length) : > + cs->copy_fix(end(), dst_length, > + str, length, length, &st)); may be let's not do copy_fix here? there are table names and in a perfschema, for PFS_table_share_key. memcpy would be enough. > + m_hash_key[m_key_length++]= '\0'; > + } > + } > + void set(bool temporary, > + const char *schema_name, size_t schema_name_length, > + const char *table_name, size_t table_name_length); > }; > > /** Table index or 'key' */ > diff --git a/storage/perfschema/pfs_program.cc b/storage/perfschema/pfs_program.cc > index de456610519..6c35a745d30 100644 > --- a/storage/perfschema/pfs_program.cc > +++ b/storage/perfschema/pfs_program.cc > @@ -118,31 +118,20 @@ static void set_program_key(PFS_program_key *key, > */ > > char *ptr= &key->m_hash_key[0]; > + const char *end= ptr + sizeof(key->m_hash_key) - 1; > > ptr[0]= object_type; > ptr++; > > if (object_name_length > 0) > - { > - char tmp_object_name[COL_OBJECT_NAME_SIZE + 1]; > - memcpy(tmp_object_name, object_name, object_name_length); > - tmp_object_name[object_name_length]= '\0'; > - my_casedn_str(system_charset_info, tmp_object_name); > - memcpy(ptr, tmp_object_name, object_name_length); > - ptr+= object_name_length; > - } > + ptr+= system_charset_info->casedn(object_name, object_name_length, > + ptr, end - ptr); > ptr[0]= 0; > ptr++; > > if (schema_name_length > 0) > - { > - char tmp_schema_name[COL_OBJECT_SCHEMA_SIZE + 1]; > - memcpy(tmp_schema_name, schema_name, schema_name_length); > - tmp_schema_name[schema_name_length]='\0'; > - my_casedn_str(system_charset_info, tmp_schema_name); > - memcpy(ptr, tmp_schema_name, schema_name_length); > - ptr+= schema_name_length; > - } > + ptr+= system_charset_info->casedn(schema_name, schema_name_length, > + ptr, end - ptr); That's strange. casedn of schema when lower_case_table_names==0 ? > ptr[0]= 0; > ptr++; > Regards, Sergei Chief Architect, MariaDB Server and security(a)mariadb.org [View Less]

2 4

review for MDEV-32839 LONG UNIQUE gives error when used with REPLACE
by Alexander Barkov 12 Dec '23

12 Dec '23

Hello Sergei, - This patch: https://github.com/MariaDB/server/commit/4f2d0541b501a7a27bf44aa16821d09395… cleanup: remove innodb-specific code around update_auto_increment() is OK to push. Please write in the commit comments that it was a dead code and that InnoDB team does know why this code was there. - This patch https://github.com/MariaDB/server/commit/c8d0f3370b7d77ecb07f8f1357da6bb1f0… MDEV-32839 LONG UNIQUE gives error when used with REPLACE is OK. I have only a proposal about … [View More]tests. You added tests above the "# End of 10.5 tests" line, but the patch is for 10.6. Also please add tests for at lease these engines: - MyISAM - InnoDB - Spider as well as partitioned tables for the same engines, because your patch for ha_partition.cc changes the code related to need_info_for_auto_inc(), which has a special implementation in ha_partition and ha_spider. - This patch: https://github.com/MariaDB/server/commit/94f26086663b7c8dbcd6b26930b64de36a… cleanup: remove partition-specific code around update_auto_increment() > diff --git a/sql/ha_partition.cc b/sql/ha_partition.cc > index 25bbcfdd75ce0..e174ef8d21940 100644 > --- a/sql/ha_partition.cc > +++ b/sql/ha_partition.cc > @@ -4308,6 +4308,7 @@ int ha_partition::external_lock(THD *thd, int lock_type) > m_part_info->part_expr->walk(&Item::register_field_in_read_map, 1, 0); > if ((error= m_part_info->vers_set_hist_part(thd))) > goto err_handler; > + need_info_for_auto_inc(); auto-increment code is often done under a condition like this: if (table->next_number_field && buf == table->record[0]) { some_auto_increment code; } This includes the code that you are removing from ha_partition::write_row() Is it ok to call need_info_for_auto_inc() unconditionally here? > } > DBUG_RETURN(0); > > @@ -4623,33 +4624,8 @@ int ha_partition::write_row(const uchar * buf) > */ > if (have_auto_increment) > { > - if (!table_share->next_number_keypart) > - if (unlikely(error= update_next_auto_inc_val())) > - goto exit; Here there was an extra condition !table_share->next_number_keypart before calling update_next_auto_inc_val(). Why was it needed? Why it's not needed in the new version of external_lock() before the call for need_info_for_auto_inc()? Also, the old code checked the return value. The new code int external_code() does not check the result of need_info_for_auto_inc(). Can't the behavior change because of this change? > - > - /* > - If we have failed to set the auto-increment value for this row, > - it is highly likely that we will not be able to insert it into > - the correct partition. We must check and fail if necessary. > - */ > if (unlikely(error= update_auto_increment())) > goto exit; > - > - /* > - Don't allow generation of auto_increment value the partitions handler. > - If a partitions handler would change the value, then it might not > - match the partition any longer. > - This can occur if 'SET INSERT_ID = 0; INSERT (NULL)', > - So allow this by adding 'MODE_NO_AUTO_VALUE_ON_ZERO' to sql_mode. > - The partitions handler::next_insert_id must always be 0. Otherwise > - we need to forward release_auto_increment, or reset it for all > - partitions. > - */ > - if (table->next_number_field->val_int() == 0) > - { > - table->auto_increment_field_not_null= TRUE; > - thd->variables.sql_mode|= MODE_NO_AUTO_VALUE_ON_ZERO; > - } Why is it safe to remove this code? > } > old_map= dbug_tmp_use_all_columns(table, &table->read_set); > error= m_part_info->get_partition_id(m_part_info, &part_id, &func_value); > @@ -11012,10 +10988,7 @@ void ha_partition::get_auto_increment(ulonglong offset, ulonglong increment, > else > { > THD *thd= ha_thd(); > - /* > - This is initialized in the beginning of the first write_row call. > - */ > - DBUG_ASSERT(part_share->auto_inc_initialized); > + update_next_auto_inc_val(); > /* > Get a lock for handling the auto_increment in part_share > for avoiding two concurrent statements getting the same number. > diff --git a/sql/ha_partition.h b/sql/ha_partition.h > index d75e38ec272a5..0ac8f696fcf55 100644 > --- a/sql/ha_partition.h > +++ b/sql/ha_partition.h > @@ -1408,9 +1408,8 @@ class ha_partition final :public handler > { > ulonglong nr= (((Field_num*) field)->unsigned_flag || > field->val_int() > 0) ? field->val_int() : 0; > + update_next_auto_inc_val(); > lock_auto_increment(); > - DBUG_ASSERT(part_share->auto_inc_initialized || > - !can_use_for_auto_inc_init()); > /* must check when the mutex is taken */ > if (nr >= part_share->next_auto_inc_val) > part_share->next_auto_inc_val= nr + 1; [View Less]

2 2

MDEV-31404 Review
by Kristian Nielsen 11 Dec '23

11 Dec '23

Hi Monty, here is my review of the patch for MDEV-31404: > commit d51086a9f1b1b6a66c7ba8eb6a6876b433fa118f (origin/bb-11.4-MDEV-31404) > Author: Monty <monty(a)mariadb.org> > Date: Sun Dec 3 21:42:44 2023 +0200 > > MDEV-31404 Implement binlog_space_limit > > binlog_space_limit is a variable in Percona server used to limit the total > size of all binary logs. > > This implementation is based on code from Percona server 5.7. … [View More]General comment: The --slave-connections-needed-for-purge defaults to 1. I think this means that --log-expire-days will have no effect on systems that have binlog enabled but no slaves configured. This is something to be aware of for backwards compatibility. What I do not like about this is the consequence it will have for all systems that have binlog enabled with some --expire-log-days set but no slaves configured. After upgrade, these systems will silently start accumulating binlogs without limit, possibly filling up disk eventually. On the other hand, it seems good to have --slave-connections-needed-for-purge enabled for the new --max-binlog-total-size option; as you pointed out, otherwise a single large transaction could immediately cause deletion of all binlogs when no slaves are connected. And having --slave-connections-needed-for-purge affect purge from --max-binlog-total-size but not for --expire-log-days seems inconsistent. In the end, maybe it isn't too serious. Many small and default-configured systems will probably be able to run for ages without binlogs getting excessive due to low traffic; and it will be easy for DBAs to just delete the binlogs if needed, or google the --slave-connections-needed-for-purge if they care enough. Two suggestions to make it easier for the DBAs to know how to handle this: 1. How about emitting a note or warning in the error log, if purge from --log-expire-days=N gets delayed for more than 2*N because no slaves are connected and --slave-connections-needed-for-purge=1 ? Note: Purge of binlogs prevented for <seconds> because no slaves are connected. To allow purge of binlogs with no slaves connected, set --slave-connections-needed-for-purge=0 2. Mention in the changelogs for 11.4 that binlogs will by default no longer be purged when no slaves are configured, and point to some documentation about how to handle this for the DBA. I think with those two suggestions it should be fine. Detailed comments on the patch below. - Kristian. > diff --git a/mysql-test/suite/binlog_encryption/binlog_index-master.opt b/mysql-test/suite/binlog_encryption/binlog_index-master.opt > new file mode 100644 > index 00000000000..f2673d08020 > --- /dev/null > +++ b/mysql-test/suite/binlog_encryption/binlog_index-master.opt > @@ -0,0 +1 @@ > +--slave_connections_needed_for_purge=0 Is this needed? If so, would it be better to put this in a general my.cnf for the entire binlog_encryption suite, like you did for mysql-test/suite/binlog/my.cnf ? So that a similar .opt will not need to be added manually for each new binlog_encryption test that will need it. > --- a/mysql-test/suite/perfschema/t/show_sanity.test > +++ b/mysql-test/suite/perfschema/t/show_sanity.test > @@ -533,6 +533,7 @@ insert into test.sanity values > ("JUNK: GLOBAL-ONLY", "I_S.SESSION_VARIABLES", "MAX_BINLOG_CACHE_SIZE"), > ("JUNK: GLOBAL-ONLY", "I_S.SESSION_VARIABLES", "MAX_BINLOG_SIZE"), > ("JUNK: GLOBAL-ONLY", "I_S.SESSION_VARIABLES", "MAX_BINLOG_STMT_CACHE_SIZE"), > + ("JUNK: GLOBAL-ONLY", "I_S.SESSION_VARIABLES", "MAX_BINLOGS_SIZE"), I think this is a left-over from previous naming, should probably be "MAX_BINLOG_TOTAL_SIZE ? > +FLUSH LOGS; > +FLUSH LOGS; > +FLUSH LOGS; > +source include/show_binary_logs.inc; > +show status like "binlog_disk_use"; I think you will need --source include/wait_for_binlog_checkpoint.inc after each of these FLUSH LOGS. Otherwise random delay of binlog checkpoint could cause the binlog_disk_use value to fluctuate and get rare random test failures. (Or alternatively consider if you could omit the select of binlog_disk_use completely, since it will probably differ between versions as event formats change.) Adding a small sleep in binlog_background_thread() just before calling mysql_bin_log.mark_xid_done() can usually reproduce the random failures from delayed binlog checkpoint. > +INSERT INTO t1 VALUES (0,repeat("a",3000)); > +show status like "binlog_disk_use"; > +Variable_name Value > +Binlog_disk_use 3848 > +# First binary should be binary.000004 > +show binary logs; > +Log_name File_size > +binary.000004 # > +INSERT INTO t1 VALUES (2,repeat("b",10)); > +# First binary should be binary.000004 > +show binary logs; > +Log_name File_size > +binary.000004 # > +binary.000005 # Here, I don't understand why binary.000004 is not purged. The binary.000004 contains the large transaction 'repeat("a", 3000)', so the binlog total size is too large. And the transaction 'repeat("n",10)' caused a binlog rotation. So why does that not cause binary.000004 to be purged due to --max-binlog-total-size=1500 ? > +show status like "binlog_disk_use"; > +Variable_name Value > +Binlog_disk_use 3647 > +INSERT INTO t1 VALUES (7,repeat("g",3000)); > +# binary.000001 should be deleted now > +show binary logs; > +Log_name File_size > +binary.000002 # > +binary.000003 # Similarly here, why binary.000002 is not purged? > +INSERT INTO t1 VALUES (4,repeat("d",3000)); > +--echo # First binary should be binary.000011 > +source include/show_binary_logs.inc; > + > +RESET MASTER; Idea: Here you could (or ask one of the other devs) add a test that the SHOW STATUS LIKE "binlog_disk_use" is equal to the size of the files on disk, to check that the running calculation of disk use doesn't miss the odd byte or two. > + if ((longlong) (binlog_space_total + binlog_pos - found_space <= > + binlog_space_limit)) > + break; // Found enough The (longlong) cast here looks misplaced. It's casting the result of the (X <= Y) comparison. I think the cast can just be removed, all the variables seem to be ulonglong type. > +static bool waiting_for_slave_to_change_binlog= 0; > +static ulonglong purge_sending_new_binlog_file= 0; I think a short comment here on these variables would be good, explaining their purpose. And same on the declaration of Atomic_counter<ulonglong> sending_new_binlog_file. > + DBUG_ASSERT(!is_relay_log || binlog_xid_count_list.is_empty()); > + if (!is_relay_log) > { > - I_List_iterator<xid_count_per_binlog> it(binlog_xid_count_list); > - while ((b= it++) && > - 0 != strncmp(log_file_name_arg+dirname_length(log_file_name_arg), > - b->binlog_name, b->binlog_name_len)) > - ; > + xid_count_per_binlog *b; > + mysql_mutex_lock(&LOCK_xid_list); Good catch that we were uselessly doing xid_count operations on the relay log, thanks! > +int MYSQL_BIN_LOG::count_binlog_space() > +{ > + int error; > + LOG_INFO log_info; > + DBUG_ENTER("count_binlog_space"); > + if (is_relay_log || !binlog_space_limit) > + DBUG_RETURN(0); Maybe this if (is_relay_log || !binlog_space_limit) could be an assert? We probably should not be calling this function at all in that case. It looks like that binlog_space_limit is already checked in all callers, not sure about relay log. (A small point, just something to consider as you like). > @@ -7608,7 +7807,8 @@ int MYSQL_BIN_LOG::rotate(bool force_rotate, bool* check_purge) > @retval > nonzero - error in rotating routine. > */ > -void MYSQL_BIN_LOG::purge() > + > +void MYSQL_BIN_LOG::purge(bool all) Would be good with a small comment here explaining the meaning of the new parameter `all`. Minor stuff: Commit message: > Binlog file file sizes are checked on startup Typo (s/file file/file/). > when updating the binary log on on binary log rotation Typo (s/on on/and on/ ?) > Without this option max-binlog-total-size would potentially deleted > binlogs needed by slaves on server startup or when a slave disconnects Typo s/deleted/delete/. > +static Sys_var_ulonglong Sys_max_binlog_total_size( > + "max_binlog_total_size", > + "Maximum space to use for all binary logs. Extra logs are deleted on " > + "server start, log rotation, FLUSH LOGGS or when writing to binlog. " > + "Default is 0, which means no size restrictions." > + "See also slave_connections_needed_for_purge", Missing space at end of line (s/restrictions."/restrictions. "/). > +static Sys_var_ulong Sys_slave_connections_needed_for_purge( > + "slave_connections_needed_for_purge", > + "Number of minimum connected slaves needed for automatic binary " > + "log purge with max_binlog_total_size, binlog_expire_logs_seconds " > + "or binlog_expire_logs_days.", Suggestion for better wording: "Minimum number of connected slaves required for automatic binary " "log purge with max_binlog_total_size, binlog_expire_logs_seconds " "or binlog_expire_logs_days." [View Less]

2 2

Re: d25ab9fdfa1: MDEV-31608 - Connector/NET fails to connect since 10.10
by Sergei Golubchik 30 Nov '23

30 Nov '23

Hi, Vladislav, On Nov 29, Vladislav Vaintroub wrote: > revision-id: d25ab9fdfa1 (mariadb-10.11.2-62-gd25ab9fdfa1) > parent(s): 034848c6c27 > author: Vladislav Vaintroub > committer: Vladislav Vaintroub > timestamp: 2023-10-08 19:37:20 +0200 > message: > > MDEV-31608 - Connector/NET fails to connect since 10.10 > > Connector/NET does not expect collation IDs returned by "show collations" > to be NULL, runs into an exception. > > The fix is to determine … [View More]

2 1

Re: [PATCH] MDEV-32551: "Read semi-sync reply magic number error" warnings on master
by Kristian Nielsen 23 Nov '23

23 Nov '23

Hi Monty, > From: Michael Widenius <monty(a)mariadb.com> > > commit 7abf83f86829ed94de618af4749eb96d5379bd90 (origin/bb-10.6-monty) > Author: Michael Widenius <monty(a)mariadb.com> > Date: Wed Nov 8 16:57:58 2023 -0700 > > MDEV-32551: "Read semi-sync reply magic number error" warnings on master I read through the patch, looks good. A few comments below, mostly minor things and typos that you can consider as you like. - Kristian. > - We should in 11.… [View More]3 changed the default value for > rpl-semi-sync-master-wait-no-slave from TRUE to FALSE as the TRUE > Sys_semisync_master_wait_no_slave( > "rpl_semi_sync_master_wait_no_slave", > - "Wait until timeout when no semi-synchronous replication slave " > - "available (enabled by default).", > + "Wait until timeout when no semi-synchronous replication slave is " > + "available.", > GLOBAL_VAR(rpl_semi_sync_master_wait_no_slave), > CMD_LINE(OPT_ARG), DEFAULT(TRUE), > - NO_MUTEX_GUARD, NOT_IN_BINLOG, ON_CHECK(0), > - ON_UPDATE(fix_rpl_semi_sync_master_wait_no_slave)); > + NO_MUTEX_GUARD, NOT_IN_BINLOG, ON_CHECK(0)); > -my_bool rpl_semi_sync_master_wait_no_slave = 1; > +my_bool rpl_semi_sync_master_wait_no_slave = 0; There seems to be som inconsistency here: - The commit is based on 10.6, but the commit comment says 11.3 - The option help text is changed in the commit - The "DEFAULT(TRUE)" in the sysvar definition is _not_ changed in the commit - The C initialization values of the my_bool rpl_semi_sync_master_wait_no_slave _is_ changed in the commit. Should it be DEFAULT(FALSE) also in this patch, and the commit message say "10.6"? Or should these changes be removed from the 10.6 patch and put in 11.3 instead? At least this looks inconsistent. > diff --git a/mysql-test/include/wait_for_pattern_in_file.inc b/mysql-test/include/wait_for_pattern_in_file.inc > new file mode 100644 > index 00000000000..d088cb125c2 > --- /dev/null > +++ b/mysql-test/include/wait_for_pattern_in_file.inc > @@ -0,0 +1,54 @@ > +# ==== Purpose ==== > +# > +# Waits until pattern comes into log file or until a timeout is reached. Nice addition, this will be useful for other things! > +--let $keep_include_silent= 1 Why do you set this in the .inc? (Normally this is set in the .test files that want to skip the printout). +let SEARCH_SILENT=1; Maybe the .inc should restore the old value at the end (in case a .test will at some point set it globally for its own calls to include/search_pattern_in_file.inc) > +--let $keep_include_silent= 0 Here, better restore the original value (in case some .test has set it to 1 for itself). > +# Test is binlog format indepdendent, so save resources (Typo s/indepdendent/independent/) > +# The followowing command is likely to cause the slave master is not yet setup > +# for semi-sync (Typo s/followowing/following/) > - strcpy(m_reply_file_name, ""); > - strcpy(m_wait_file_name, ""); > + m_reply_file_name[0]= m_wait_file_name[0]= 0; Interestingly, GCC actually creates identical code for these (it knows to optimize strcpy()). (I'm just mentioning this as I know we both are interested in low-level code details, not asking you to revert the change). Example code: ----------------------------------------------------------------------- #include <string.h> struct X { char m_a[100]; char m_b[10]; void init1(); void init2(); }; void X::init1() { strcpy(m_a, ""); strcpy(m_b, ""); } void X::init2() { m_a[0]= m_b[0]= 0; } ----------------------------------------------------------------------- Result from gcc -O3: ----------------------------------------------------------------------- .type _ZN1X5init1Ev, @function _ZN1X5init1Ev: movb $0, (%rdi) movb $0, 100(%rdi) ret .type _ZN1X5init2Ev, @function _ZN1X5init2Ev: movb $0, 100(%rdi) movb $0, (%rdi) ret ----------------------------------------------------------------------- Identical, except that we now initialize in the opposite order ;-) > +#ifndef DBUG_OFF > +void dbug_verify_no_duplicate_slaves(Slave_ilist *m_slaves, THD *thd) > +{ You could make this (debug-only) function "static". > + Master is not responding (gone away?) or it has turned semi sync > + off. Turning off semi-sync repsonces as there is no point in sending (Typo s/repsonces/responses/) [View Less]

2 1