Thread: BUG #17816: Invalid memory access in translate function

The following bug has been logged on the website:

Bug reference:      17816
Logged by:          Daniil Anisimov
Email address:      anisimow.d@gmail.com
PostgreSQL version: 15.2
Operating system:   Ubuntu 22.04.2 LTS
Description:

When running this query in psql with valgrind-enabled server:
select translate('(123)', '()', '');

I get:
==00:00:00:05.576 29520== Invalid read of size 1
==00:00:00:05.576 29520==    at 0x7605BE: pg_utf_mblen (wchar.c:553)
==00:00:00:05.576 29520==    by 0x71CEDE: pg_mblen (mbutils.c:968)
==00:00:00:05.576 29520==    by 0x67B1E3: translate (oracle_compat.c:864)
==00:00:00:05.576 29520==    by 0x3FE8AB: ExecInterpExpr
(execExprInterp.c:751)
==00:00:00:05.576 29520==    by 0x3FB0B1: ExecInterpExprStillValid
(execExprInterp.c:1826)
==00:00:00:05.576 29520==    by 0x4EC87F: ExecEvalExprSwitchContext
(executor.h:341)
==00:00:00:05.576 29520==    by 0x4EC87F: evaluate_expr (clauses.c:4823)
==00:00:00:05.576 29520==    by 0x4ECA5B: evaluate_function
(clauses.c:4325)
==00:00:00:05.576 29520==    by 0x4EEF13: simplify_function
(clauses.c:3908)
==00:00:00:05.576 29520==    by 0x4ECD1F: eval_const_expressions_mutator
(clauses.c:2427)
==00:00:00:05.576 29520==    by 0x47DB6C: expression_tree_mutator
(nodeFuncs.c:3080)
==00:00:00:05.576 29520==    by 0x4EE159: eval_const_expressions_mutator
(clauses.c:3527)
==00:00:00:05.576 29520==    by 0x47DDB6: expression_tree_mutator
(nodeFuncs.c:3166)
==00:00:00:05.576 29520==  Address 0x6f541d4 is 7,156 bytes inside a block
of size 8,192 alloc'd
==00:00:00:05.576 29520==    at 0x4848899: malloc (in
/usr/libexec/valgrind/vgpreload_memcheck-amd64-linux.so)
==00:00:00:05.576 29520==    by 0x732BCF: AllocSetContextCreateInternal
(aset.c:469)
==00:00:00:05.576 29520==    by 0x51F8D2: PostmasterMain
(postmaster.c:612)
==00:00:00:05.576 29520==    by 0x464A71: main (main.c:202)
==00:00:00:05.576 29520==
{
   <insert_a_suppression_name_here>
   Memcheck:Addr1
   fun:pg_utf_mblen
   fun:pg_mblen
   fun:translate
   fun:ExecInterpExpr
   fun:ExecInterpExprStillValid
   fun:ExecEvalExprSwitchContext
   fun:evaluate_expr
   fun:evaluate_function
   fun:simplify_function
   fun:eval_const_expressions_mutator
   fun:expression_tree_mutator
   fun:eval_const_expressions_mutator
   fun:expression_tree_mutator
}
==00:00:00:05.576 29520==
==00:00:00:05.576 29520== Exit program on first error
(--exit-on-first-error=yes)

In the translate() function in src/backend/utils/adt/oracle_compat.c used
pg_mblen(p) even if p has zero length:
for (i = 0; i < from_index; i++)
{
    p += pg_mblen(p);
    if (p >= (to_ptr + tolen))
        break;
}

These changes fix the issue:
for (i = 0; i < from_index; i++)
{
    if (p >= (to_ptr + tolen))
        break;
    p += pg_mblen(p);
}

PG Bug reporting form <noreply@postgresql.org> writes:
> When running this query in psql with valgrind-enabled server:
> select translate('(123)', '()', '');

> I get:
> ==00:00:00:05.576 29520== Invalid read of size 1
> ==00:00:00:05.576 29520==    at 0x7605BE: pg_utf_mblen (wchar.c:553)
> ==00:00:00:05.576 29520==    by 0x71CEDE: pg_mblen (mbutils.c:968)
> ==00:00:00:05.576 29520==    by 0x67B1E3: translate (oracle_compat.c:864)

Thanks for the report!

> These changes fix the issue:
> for (i = 0; i < from_index; i++)
> {
>     if (p >= (to_ptr + tolen))
>         break;
>     p += pg_mblen(p);
> }

Yeah, I agree, we need to switch the order of those two statements.
Will fix.

(Might be worth pulling out the common subexpression here?
I wonder if all compilers are smart enough to evaluate to_ptr + to_len
just once.)

            regards, tom lane