Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.4(EOL), 10.5, 10.6, 10.7(EOL), 10.8(EOL), 10.9(EOL), 10.10(EOL), 10.11, 11.0(EOL), 11.1(EOL), 11.2(EOL), 11.3(EOL), 11.8
Description
Now that MDEV-26272 has been fixed and a cmake -DWITH_UBSAN=ON build with clang can pass the server bootstrap, we can find more undefined behaviour that is not flagged by GCC. Here is an example:
|
10.4 832e96deb6f368591dcb35a4fa33b2fd770aa17a |
/mariadb/10.4/storage/innobase/handler/ha_innodb.cc:1821:9: runtime error: call to function mysql_sys_var_long(THD*, int) through pointer to incorrect function type 'unsigned long *(*)(THD *, int)'
|
This is flagged for the function thd_lock_wait_timeout(), which simply accesses a data member that has been defined as follows:
static MYSQL_THDVAR_ULONG(lock_wait_timeout, PLUGIN_VAR_RQCMDARG, |
"Timeout in seconds an InnoDB transaction may wait for a lock before being rolled back. Values above 100000000 disable the timeout.", |
NULL, NULL, 50, 0, 1024 * 1024 * 1024, 0);
|
// ...
|
static struct st_mysql_sys_var* innobase_system_variables[]= { |
// ...
|
MYSQL_SYSVAR(lock_wait_timeout),
|
// ...
|
NULL
|
};
|
maria_declare_plugin(innobase)
|
{
|
// ...
|
innobase_system_variables
|
// ...
|
},
|
// ...
|
maria_declare_plugin_end;
|
Attachments
Issue Links
- duplicates
-
-
- Closed
-
-
-
- Closed
-
-
-
- Closed
-
-
-
- Closed
-
- is blocked by
-
-
- Closed
-
- relates to
-
MDEV-25454 Make MariaDB server UBSAN safe
-
- Confirmed
-
-
-
- Confirmed
-
Activity
Re-observed the same after the MDEV-34348 patch using a Clang 18/libstdc++-14-dev (with LLVMgold.so from LLVM-17 symlinked as it is no longer available in Clang 18) build:
|
CS 10.5.28 773cb726a8437d911d3f7ceedf4839dab6124484 (Debug, UBASAN) |
/test/clang_new/10.5_dbg_san/storage/innobase/handler/ha_innodb.cc:2377:9: runtime error: call to function mysql_sys_var_long(THD*, int) through pointer to incorrect function type 'unsigned long *(*)(THD *, int)'
|
/test/clang_new/10.5_dbg_san/sql/sql_plugin.cc:3231: note: mysql_sys_var_long(THD*, int) defined here
|
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /test/clang_new/10.5_dbg_san/storage/innobase/handler/ha_innodb.cc:2377:9
|
It can be reproduced by a simple mariadb-install-db call:
rm -Rf /test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg/data /test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg/tmp |
mkdir /test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg/tmp |
/test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg/scripts/mariadb-install-db --no-defaults --force --auth-root-authentication-method=normal --tmpdir=/test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg/tmp --basedir=/test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg --datadir=/test/CLANG_UBASAN_MD251124-mariadb-10.5.28-linux-x86_64-dbg/data |
When fixing this please also look into MDEV-34512 at the same time. Thank you!
The following UBSAN supression filter was added (which will be removed once the bug is fixed):
## MDEV-33158
|
function:innobase_trx_init
|
As expected, also observed in 10.6 post the merge of MDEV-34348 at revision 507323abe6d0cf61abb608b87b25fb5f6141aff2 (Debug, UBASAN, Clang/LLVM 18.1.3)
Issue also present in 11.8. Tested at commit f0961301c81c7f5b009c012c076abc326b203b4a (Debug, UBASAN, Clang).
Brandon can I get a review please, this fix is appended to the back of https://github.com/MariaDB/server/pull/3737
Innodb encryption test can now pass UBSAN with these fixes:
binlog_encryption.rpl_temporal_format_mariadb53_to_mysql56 'row' w6 [ skipped ] Neither MIXED nor STATEMENT binlog format
|
binlog_encryption.rpl_temporal_format_mariadb53_to_mysql56 'stmt' w1 [ pass ] 211
|
binlog_encryption.rpl_temporal_format_mysql56_to_mariadb53 'mix' w4 [ pass ] 215
|
binlog_encryption.rpl_temporal_format_mysql56_to_mariadb53 'row' w8 [ pass ] 213
|
binlog_encryption.rpl_temporal_format_mysql56_to_mariadb53 'stmt' w2 [ pass ] 208
|
encryption.tempfiles 'cbc,crc32' w8 [ pass ] 53
|
encryption.tempfiles 'crc32,ctr' w2 [ pass ] 56
|
encryption.tempfiles 'cbc,none' w8 [ pass ] 59
|
encryption.tempfiles 'ctr,none' w2 [ pass ] 74
|
encryption.innodb_encryption 'undo0' w6 [ pass ] 2138
|
Looks good overall, thanks danblack! Note I left a couple notes on the PR for your consideration.
A similar problem affects MYSQL_THDVAR_ULONGLONG and mysql_sys_var_longlong.