Hi, Sachin! On Apr 02, Sachin Setiya wrote:

revision-id: 5469d88e7b4 (mariadb-10.4.3-104-g5469d88e7b4) parent(s): 0bc42602266 author: Sachin <sachin.setiya@mariadb.com> committer: Sachin <sachin.setiya@mariadb.com> timestamp: 2019-03-28 11:29:25 +0530 message:

MDEV-19049 Server crashes in check_duplicate_long_entry_key, ASAN stack-buffer-overflow in Field_blob::get_key_image

Long Unique keys should always be last unique key.

Why?

diff --git a/sql/sql_table.cc b/sql/sql_table.cc index ad62ecc1103..c755a74e174 100644 --- a/sql/sql_table.cc +++ b/sql/sql_table.cc @@ -2796,6 +2797,14 @@ static int sort_keys(KEY *a, KEY *b) { if (!(b_flags & HA_NOSAME)) return -1; + /* + Long Unique keys should always be last unique key. + Before this patch they used to change order wrt to partial keys (MDEV-19049)

don't write the code history "before this patch it worked like that" in the code itself. When you look at the code there is no "patch", just a comment. The concepts of "patch" and "before it used to be" belongs to the history, and logically should be in the commit comment, not in a code comment.

+ */ + if (a->algorithm == HA_KEY_ALG_LONG_HASH) + return 1; + if (b->algorithm == HA_KEY_ALG_LONG_HASH) + return -1; if ((a_flags ^ b_flags) & HA_NULL_PART_KEY) { /* Sort NOT NULL keys before other keys */

Regards, Sergei Chief Architect MariaDB and security@mariadb.org