[MDEV-34348] MariaDB is violating clang-16 -Wcast-function-type-strict - Jira

Details

Type: Bug
Status: Closed (View Workflow)
Priority: Major
Resolution: Fixed
Affects Version/s: 10.5, 10.6, 10.11, 11.4
Fix Version/s: 10.5.28, 10.6.21, 10.11.11, 11.4.5, 11.7.2
Component/s: Server
Labels:
- UBSAN

Description

D134831 introduced a compile-time check -Wcast-function-type-strict for something that had been reported by -fsanitize=undefined for a long time.

The flag is similar to the non-strict GCC check that we have for a long time enabled by -Wextra.

We should fix all occurrences of this warning and enable this warning in our build script when available, so that the mandatory MemorySanitizer builder that is based on clang-16 or later would fail already at compilation time.

Attachments

Issue Links

includes

MDEV-35530 UBSAN: runtime error: call to function get_sys_var_length on server startup or version query

Closed

MDEV-35531 UBSAN: runtime error: call to function init_simple_key_cache through pointer to incorrect function type

Closed

MDEV-35532 UBSAN: runtime error: call to function FT_STOPWORD_cmp through pointer to incorrect function

Closed

MDEV-35533 UBSAN: runtime error: call to function sort_keys(st_key*, st_key*) through pointer to incorrect function type

Closed

MDEV-35534 UBSAN: runtime error: call to function thd_decrement_pending_ops through pointer to incorrect function type

Closed

MDEV-35535 UBSAN: runtime error: call to function free_user_var through pointer to incorrect function type

Closed

MDEV-35536 UBSAN: runtime error: call to function end_simple_key_cache through pointer to incorrect function type

Closed

MDEV-35539 UBSAN: runtime error: call to function adjust_optimizer_costs() through pointer to incorrect function type

Closed

MDEV-36337 runtime error: call to function (udf_example) netaphon through pointer to incorrect function type 'char *(*)(st_udf_init *, st_udf_args *, char *, unsigned long *, unsigned char *, unsigned char *)

Stalled

is blocked by

MDEV-33161 Function pointer signature mismatch in LF_HASH

Closed

is duplicated by

MDEV-34509 UBSAN: call to function option_cmp(my_option*, my_option*) through pointer to incorrect function type

Closed

MDEV-34511 UBSAN: qsort based function argument mismatch

Closed

relates to

CONC-766 Disable clang -Wcast-function-type-strict for makecontext

Closed

MDEV-25454 Make MariaDB server UBSAN safe

Confirmed

MDEV-34408 Facilitate the addition of warnings into the build system

Closed

MDEV-34889 The parameters of qsort_cmp2 are in an inefficient order

Confirmed

MDEV-35698 UBSAN: runtime error: call to function mysql_sys_var_int through pointer to incorrect function type in spider_param_force_commit

Closed

MDEV-35726 runtime error: call to function mysql_sys_var_int through pointer to incorrect function type in thd_lock_wait_timeout on ALTER

Closed

MDEV-35735 runtime error: call to function spider_direct_sql, spider_flush_table_mon_cache, spider_copy_tables through pointer to incorrect function type in udf_handler::val_in

Closed

MDEV-35736 UBSAN: runtime error: call to function mysql_sys_var_long through pointer to incorrect function type in myrocks::get_or_create_tx and other myrocks::* functions

Closed

CONC-711 UBSAN: client_mpvio_write_packet through pointer to incorrect function

Closed

MDEV-34512 UBSAN system variable type mismatches for ha_myisam.cc and ha_maria.cc

Closed

MDEV-34770 UBSAN: runtime error: load of address 0x... with insufficient space for an object of type 'uchar' in sys_vars.inl

Closed

MDEV-35554 runtime error: call to function show_cached_thread_count()/show_binlog_space_total() through pointer to incorrect function type

Closed

MDEV-35623 UBSAN: function-type-mismatch in spider_param_conn_recycle[_mode/_strict], spider_param_internal_xa_snapshot, and spider_param_block_size

Closed

MDEV-35687 Various UBSAN function-type-mismatch errors when using MTR in maria_open, mi_open, _ma_open_datafile, mi_open_datafile and thr_multi_lock

Closed

MDEV-35822 UBSAN: UndefinedBehaviorSanitizer: function-type-mismatch in process_timers when using MAX_EXECUTION_TIME

Closed

(4 includes, 1 is blocked by, 2 is duplicated by, 15 relates to)

Activity

Ascending order - Click to sort in descending order

View 6 older comments

Brandon Nesterenko added a comment - 2024-10-02 15:19

Hi marko! I've addressed your first round of review findings (PR 3490)

Brandon Nesterenko added a comment - 2024-10-02 15:19 Hi marko ! I've addressed your first round of review findings ( PR 3490 )

Marko Mäkelä added a comment - 2024-10-14 13:01

Thank you, this looks mostly good. OK to push after addressing my final comments.

Marko Mäkelä added a comment - 2024-10-14 13:01 Thank you, this looks mostly good. OK to push after addressing my final comments.

Brandon Nesterenko added a comment - 2024-11-23 17:11

Pushed into 10.5 with head commit 773cb726a84 (and many before).

Many merge conflicts in each forward branch. Each main branch has a dedicated *-merge-fix branch, organized with multiple incremental commits that fix relevant merge conflicts (as well as additional required fixes that are organized in the same area). In-between each version branch, the incremental commits have been squashed into the prior commit of the same domain. In other words, the incremental commits for each branch are only for that branch, and the incremental commits of the prior branches were squashed in. Branches are as follows:

bb-10.6-MDEV-34348-merge-fix
bb-10.11-MDEV-34348-merge-fix
bb-11.4-MDEV-34348-merge-fix
11.7 is to-come (still needs one last minor decision to be made before push)

Brandon Nesterenko added a comment - 2024-11-23 17:11 Pushed into 10.5 with head commit 773cb726a84 (and many before). Many merge conflicts in each forward branch. Each main branch has a dedicated *-merge-fix branch, organized with multiple incremental commits that fix relevant merge conflicts (as well as additional required fixes that are organized in the same area). In-between each version branch, the incremental commits have been squashed into the prior commit of the same domain. In other words, the incremental commits for each branch are only for that branch, and the incremental commits of the prior branches were squashed in. Branches are as follows: bb-10.6-MDEV-34348-merge-fix bb-10.11-MDEV-34348-merge-fix bb-11.4-MDEV-34348-merge-fix 11.7 is to-come (still needs one last minor decision to be made before push)

Marko Mäkelä added a comment - 2024-12-13 16:46

Note: In the PARSEC authentication plugin in MariaDB Connector/C version 3.4 (~~CONC-743~~) we are unable to fix this warning. I just mentioned the problem to the author of Nettle today, and he is reluctant to fix that. So, we may need to suppress -fsanitize=undefined in the function compute_derived_key() when not using this implementation:

int compute_derived_key(const char* password, size_t pass_len,

                        const struct Passwd_in_memory *params,

                        uchar *derived_key)

#if defined(HAVE_OPENSSL)

  return !PKCS5_PBKDF2_HMAC(password, (int)pass_len, params->salt,

                            CHALLENGE_SALT_LENGTH,

                            1 << (params->iterations + 10),

                            EVP_sha512(), PBKDF2_HASH_LENGTH, derived_key);

#elif defined(HAVE_GNUTLS) /* HAVE_GNUTLS */

  struct hmac_sha512_ctx ctx;

  hmac_sha512_set_key(&ctx, pass_len, (const uint8_t *)password);

/*

    pbkdf2/nettle functions are third party, so ignore the bad function casts

*/

# if defined __clang_major__ && __clang_major__ >= 16

#  pragma clang diagnostic push

#  pragma clang diagnostic ignored "-Wcast-function-type-strict"

# endif

  pbkdf2(&ctx, (nettle_hash_update_func *)hmac_sha512_update,

         (nettle_hash_digest_func *)hmac_sha512_digest, SHA512_DIGEST_SIZE,

         1024 << params->iterations, CHALLENGE_SALT_LENGTH, params->salt,

         PBKDF2_HASH_LENGTH, derived_key);

# ifdef __clang__

#  pragma clang diagnostic pop

# endif

Actually, I realize that we could write our own wrapper functions when we’re running under clang. For other compilers that we are aware of (GCC, MSVC), the function pointer cast seems to work fine.

Marko Mäkelä added a comment - 2024-12-13 16:46 Note: In the PARSEC authentication plugin in MariaDB Connector/C version 3.4 ( CONC-743 ) we are unable to fix this warning. I just mentioned the problem to the author of Nettle today, and he is reluctant to fix that. So, we may need to suppress -fsanitize=undefined in the function compute_derived_key() when not using this implementation: int compute_derived_key( const char * password, size_t pass_len, const struct Passwd_in_memory *params, uchar *derived_key) { #if defined(HAVE_OPENSSL) return !PKCS5_PBKDF2_HMAC(password, ( int )pass_len, params->salt, CHALLENGE_SALT_LENGTH, 1 << (params->iterations + 10), EVP_sha512(), PBKDF2_HASH_LENGTH, derived_key); #elif defined(HAVE_GNUTLS) /* HAVE_GNUTLS */ struct hmac_sha512_ctx ctx; hmac_sha512_set_key(&ctx, pass_len, ( const uint8_t *)password); /* pbkdf2/nettle functions are third party, so ignore the bad function casts */ # if defined __clang_major__ && __clang_major__ >= 16 # pragma clang diagnostic push # pragma clang diagnostic ignored "-Wcast-function-type-strict" # endif pbkdf2(&ctx, (nettle_hash_update_func *)hmac_sha512_update, (nettle_hash_digest_func *)hmac_sha512_digest, SHA512_DIGEST_SIZE, 1024 << params->iterations, CHALLENGE_SALT_LENGTH, params->salt, PBKDF2_HASH_LENGTH, derived_key); # ifdef __clang__ # pragma clang diagnostic pop # endif Actually, I realize that we could write our own wrapper functions when we’re running under clang. For other compilers that we are aware of (GCC, MSVC), the function pointer cast seems to work fine.

Daniel Black added a comment - 2025-04-03 03:27

C/C makecontext masking - https://github.com/mariadb-corporation/mariadb-connector-c/pull/274

Daniel Black added a comment - 2025-04-03 03:27 C/C makecontext masking - https://github.com/mariadb-corporation/mariadb-connector-c/pull/274

People

Assignee:: Brandon Nesterenko

Reporter:: Marko Mäkelä

Votes:: 0 Vote for this issue

Watchers:: 4 Start watching this issue

Dates

Created:: 2024-06-10 09:41

Updated:: 2025-04-10 05:51

Resolved:: 2024-11-23 17:11

Git Integration

Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.

MariaDB Server