MDEV-23210 Assertion `(length % 4) == 0' failed in my_lengthsp_utf32 on ALTER TABLE, SELECT and INSERT

Problem:
Parse-time conversion from binary to tricky character sets like utf32
produced ill-formed strings. So, later a chash happened in debug builds,
or a wrong SHOW CREATE TABLE was returned in release builds.

Fix:

1. Backporting a few methods from 10.3:
  - THD::check_string_for_wellformedness()
  - THD::convert_string() overloads
  - THD::make_text_string_connection()

2. Adding a new method THD::reinterpret_string_from_binary(),
   which makes sure to either returns a well-formed string
   (optionally prepending with zero bytes), or returns an error.

mysql-test/r/ctype_utf32.result

@@ -2890,5 +2890,28 @@ HEX(c1)
 0000006100000063
 DROP TABLE t1;
 #
+# MDEV-23210 Assertion `(length % 4) == 0' failed in my_lengthsp_utf32 on ALTER TABLE, SELECT and INSERT
+#
+CREATE TABLE t1 (a CHAR(1));
+SET COLLATION_CONNECTION=utf32_general_ci, CHARACTER_SET_CLIENT=binary;
+ALTER TABLE t1 CHANGE a a ENUM('a','a') CHARACTER SET utf32;
+ERROR HY000: Column 'a' has duplicated value 'a' in ENUM
+ALTER TABLE t1 CHANGE a a ENUM('aaa') CHARACTER SET utf32;
+ERROR HY000: Invalid utf32 character string: '\x00aaa'
+ALTER TABLE t1 CHANGE a a ENUM('aa') CHARACTER SET utf32;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` enum('慡') CHARACTER SET utf32 DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+ALTER TABLE t1 CHANGE a a ENUM('a','b') CHARACTER SET utf32;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` enum('a','b') CHARACTER SET utf32 DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+SET NAMES utf8;
+#
 # End of 10.2 tests
 #

mysql-test/r/ctype_utf32_uca.result

@@ -7941,5 +7941,20 @@ EXECUTE s;
 DEALLOCATE PREPARE s;
 SET NAMES utf8;
 #
+# MDEV-23210 Assertion `(length % 4) == 0' failed in my_lengthsp_utf32 on ALTER TABLE, SELECT and INSERT
+#
+CREATE TABLE t1 (a CHAR(1));
+SET COLLATION_CONNECTION=utf32_myanmar_ci, CHARACTER_SET_CLIENT=binary;
+ALTER TABLE t1 CHANGE a a ENUM('a','a') CHARACTER SET utf32;
+ERROR HY000: Column 'a' has duplicated value 'a' in ENUM
+ALTER TABLE t1 CHANGE a a ENUM('a','b') CHARACTER SET utf32;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` enum('a','b') CHARACTER SET utf32 DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+SET NAMES utf8;
+#
 # End of 10.2 tests
 #

mysql-test/t/ctype_utf32.test

@@ -1048,6 +1048,25 @@ INSERT INTO t1 (c1) VALUES (1),(2),(3);
 SELECT HEX(c1) FROM t1 ORDER BY c1;
 DROP TABLE t1;
 
+
+--echo #
+--echo # MDEV-23210 Assertion `(length % 4) == 0' failed in my_lengthsp_utf32 on ALTER TABLE, SELECT and INSERT
+--echo #
+
+CREATE TABLE t1 (a CHAR(1));
+SET COLLATION_CONNECTION=utf32_general_ci, CHARACTER_SET_CLIENT=binary;
+--error ER_DUPLICATED_VALUE_IN_TYPE
+ALTER TABLE t1 CHANGE a a ENUM('a','a') CHARACTER SET utf32;
+--error ER_INVALID_CHARACTER_STRING
+ALTER TABLE t1 CHANGE a a ENUM('aaa') CHARACTER SET utf32;
+ALTER TABLE t1 CHANGE a a ENUM('aa') CHARACTER SET utf32;
+SHOW CREATE TABLE t1;
+ALTER TABLE t1 CHANGE a a ENUM('a','b') CHARACTER SET utf32;
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+SET NAMES utf8;
+
+
 --echo #
 --echo # End of 10.2 tests
 --echo #

mysql-test/t/ctype_utf32_uca.test

@@ -290,6 +290,19 @@ EXECUTE s;
 DEALLOCATE PREPARE s;
 SET NAMES utf8;
 
+--echo #
+--echo # MDEV-23210 Assertion `(length % 4) == 0' failed in my_lengthsp_utf32 on ALTER TABLE, SELECT and INSERT
+--echo #
+
+CREATE TABLE t1 (a CHAR(1));
+SET COLLATION_CONNECTION=utf32_myanmar_ci, CHARACTER_SET_CLIENT=binary;
+--error ER_DUPLICATED_VALUE_IN_TYPE
+ALTER TABLE t1 CHANGE a a ENUM('a','a') CHARACTER SET utf32;
+ALTER TABLE t1 CHANGE a a ENUM('a','b') CHARACTER SET utf32;
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+SET NAMES utf8;
+
 
 --echo #
 --echo # End of 10.2 tests

sql/sql_class.cc

@@ -2148,7 +2148,7 @@ void THD::cleanup_after_query()
 */
 
 bool THD::convert_string(LEX_STRING *to, CHARSET_INFO *to_cs,
-			 const char *from, uint from_length,
+			 const char *from, size_t from_length,
 			 CHARSET_INFO *from_cs)
 {
   DBUG_ENTER("THD::convert_string");
@@ -2170,6 +2170,58 @@ bool THD::convert_string(LEX_STRING *to, CHARSET_INFO *to_cs,
 }
 
 
+/*
+  Reinterpret a binary string to a character string
+
+  @param[OUT] to    The result will be written here,
+                    either the original string as is,
+                    or a newly alloced fixed string with
+                    some zero bytes prepended.
+  @param cs         The destination character set
+  @param str        The binary string
+  @param length     The length of the binary string
+
+  @return           false on success
+  @return           true on error
+*/
+
+bool THD::reinterpret_string_from_binary(LEX_CSTRING *to, CHARSET_INFO *cs,
+                                         const char *str, size_t length)
+{
+  /*
+    When reinterpreting from binary to tricky character sets like
+    UCS2, UTF16, UTF32, we may need to prepend some zero bytes.
+    This is possible in scenarios like this:
+      SET COLLATION_CONNECTION=utf32_general_ci, CHARACTER_SET_CLIENT=binary;
+    This code is similar to String::copy_aligned().
+  */
+  size_t incomplete= length % cs->mbminlen; // Bytes in an incomplete character
+  if (incomplete)
+  {
+    size_t zeros= cs->mbminlen - incomplete;
+    size_t aligned_length= zeros + length;
+    char *dst= (char*) alloc(aligned_length + 1);
+    if (!dst)
+    {
+      to->str= NULL; // Safety
+      to->length= 0;
+      return true;
+    }
+    bzero(dst, zeros);
+    memcpy(dst + zeros, str, length);
+    dst[aligned_length]= '\0';
+    to->str= dst;
+    to->length= aligned_length;
+  }
+  else
+  {
+    to->str= str;
+    to->length= length;
+  }
+  return check_string_for_wellformedness(to->str, to->length, cs);
+}
+
+
 /*
   Convert a string between two character sets.
   dstcs and srccs cannot be &my_charset_bin.
@@ -2274,6 +2326,21 @@ bool THD::convert_string(String *s, CHARSET_INFO *from_cs, CHARSET_INFO *to_cs)
 }
 
 
+bool THD::check_string_for_wellformedness(const char *str,
+                                          size_t length,
+                                          CHARSET_INFO *cs) const
+{
+  size_t wlen= Well_formed_prefix(cs, str, length).length();
+  if (wlen < length)
+  {
+    ErrConvString err(str, length, &my_charset_bin);
+    my_error(ER_INVALID_CHARACTER_STRING, MYF(0), cs->csname, err.ptr());
+    return true;
+  }
+  return false;
+}
+
+
 /*
   Update some cache variables when character set changes
 */

sql/sql_class.h

@@ -3503,8 +3503,31 @@ public:
     return true;     // EOM
   }
   bool convert_string(LEX_STRING *to, CHARSET_INFO *to_cs,
-		      const char *from, uint from_length,
+		      const char *from, size_t from_length,
 		      CHARSET_INFO *from_cs);
+  bool reinterpret_string_from_binary(LEX_CSTRING *to, CHARSET_INFO *to_cs,
+                                      const char *from, size_t from_length);
+  bool convert_string(LEX_CSTRING *to, CHARSET_INFO *to_cs,
+                      const char *from, size_t from_length,
+                      CHARSET_INFO *from_cs)
+  {
+    LEX_STRING tmp;
+    bool rc= convert_string(&tmp, to_cs, from, from_length, from_cs);
+    to->str= tmp.str;
+    to->length= tmp.length;
+    return rc;
+  }
+  bool convert_string(LEX_CSTRING *to, CHARSET_INFO *tocs,
+                      const LEX_CSTRING *from, CHARSET_INFO *fromcs,
+                      bool simple_copy_is_possible)
+  {
+    if (!simple_copy_is_possible)
+      return unlikely(convert_string(to, tocs, from->str, from->length, fromcs));
+    if (fromcs == &my_charset_bin)
+      return reinterpret_string_from_binary(to, tocs, from->str, from->length);
+    *to= *from;
+    return false;
+  }
   /*
     Convert a strings between character sets.
     Uses my_convert_fix(), which uses an mb_wc .. mc_mb loop internally.
@@ -3540,6 +3563,44 @@ public:
 
   bool convert_string(String *s, CHARSET_INFO *from_cs, CHARSET_INFO *to_cs);
 
+  /*
+    Check if the string is wellformed, raise an error if not wellformed.
+    @param str    - The string to check.
+    @param length - the string length.
+  */
+  bool check_string_for_wellformedness(const char *str,
+                                       size_t length,
+                                       CHARSET_INFO *cs) const;
+
+  bool make_text_string_connection(LEX_CSTRING *to,
+                                   const LEX_CSTRING *from)
+  {
+    return convert_string(to, variables.collation_connection,
+                          from, charset(), charset_is_collation_connection);
+  }
+#if MYSQL_VERSION_ID < 100300
+  /*
+    A wrapper method for 10.2. It fixes the problem
+    that various fields in bison %union use LEX_STRING.
+    In 10.3 those fields are fixed to use LEX_CSTRING.
+    Please remove this wrapper when mering to 10.3.
+  */
+  bool make_text_string_connection(LEX_STRING *to,
+                                   const LEX_STRING *from)
+  {
+    LEX_CSTRING cto;
+    LEX_CSTRING cfrom;
+    bool rc;
+    cfrom.str= from->str;
+    cfrom.length= from->length;
+    rc= make_text_string_connection(&cto, &cfrom);
+    to->str= (char*) cto.str;
+    to->length= cto.length;
+    return rc;
+  }
+#else
+#error Remove the above wrapper
+#endif
   void add_changed_table(TABLE *table);
   void add_changed_table(const char *key, long key_length);
   CHANGED_TABLE_LIST * changed_table_dup(const char *key, long key_length);

sql/sql_yacc.yy

@@ -14571,14 +14571,8 @@ TEXT_STRING_sys:
 TEXT_STRING_literal:
           TEXT_STRING
           {
-            if (thd->charset_is_collation_connection)
+            if (thd->make_text_string_connection(&$$, &$1))
-              $$= $1;
+              MYSQL_YYABORT;
-            else
-            {
-              if (thd->convert_string(&$$, thd->variables.collation_connection,
-                                  $1.str, $1.length, thd->charset()))
-                MYSQL_YYABORT;
-            } 
           }
         ;