Bug (kind of) in 'isinstack'

The function 'isinstack' tried to work around the undefined behavior
of subtracting two pointers that do not point to the same object,
but the compiler killed to trick. (It optimizes out the safety check,
because in a correct execution it will be always true.)
This commit is contained in:
Roberto Ierusalimschy 2021-02-25 13:39:36 -03:00
parent 5205f073c5
commit e0260eb2d4
2 changed files with 17 additions and 6 deletions

View File

@ -638,14 +638,18 @@ static const char *funcnamefromcode (lua_State *L, CallInfo *ci,
/*
** The subtraction of two potentially unrelated pointers is
** not ISO C, but it should not crash a program; the subsequent
** checks are ISO C and ensure a correct result.
** Check whether pointer 'o' points to some value in the stack
** frame of the current function. Because 'o' may not point to a
** value in this stack, we cannot compare it with the region
** boundaries (undefined behaviour in ISO C).
*/
static int isinstack (CallInfo *ci, const TValue *o) {
StkId base = ci->func + 1;
ptrdiff_t i = cast(StkId, o) - base;
return (0 <= i && i < (ci->top - base) && s2v(base + i) == o);
StkId pos;
for (pos = ci->func + 1; pos < ci->top; pos++) {
if (o == s2v(pos))
return 1;
}
return 0; /* not found */
}

View File

@ -191,6 +191,13 @@ checkmessage("a = 24 // 0", "divide by zero")
checkmessage("a = 1 % 0", "'n%0'")
-- type error for an object which is neither in an upvalue nor a register.
-- The following code will try to index the value 10 that is stored in
-- the metatable, without moving it to a register.
checkmessage("local a = setmetatable({}, {__index = 10}).x",
"attempt to index a number value")
-- numeric for loops
checkmessage("for i = {}, 10 do end", "table")
checkmessage("for i = io.stdin, 10 do end", "FILE")