Revamp of format validation in 'string.format'

When calling 'sprintf', not all conversion specifiers accept all
flags; some combinations are undefined behavior.
This commit is contained in:
Roberto Ierusalimschy
2021-09-03 13:14:56 -03:00
parent 91673a8ec0
commit 9db4bfed6b
3 changed files with 118 additions and 36 deletions

112
lstrlib.c
View File

@@ -1090,13 +1090,31 @@ static int lua_number2strx (lua_State *L, char *buff, int sz,
/* valid flags in a format specification */
#if !defined(L_FMTFLAGS)
#define L_FMTFLAGS "-+ #0"
#if !defined(L_FMTFLAGSF)
/* valid flags for a, A, e, E, f, F, g, and G conversions */
#define L_FMTFLAGSF "-+#0 "
/* valid flags for o, x, and X conversions */
#define L_FMTFLAGSX "-#0"
/* valid flags for d and i conversions */
#define L_FMTFLAGSI "-+0 "
/* valid flags for u conversions */
#define L_FMTFLAGSU "-0"
/* valid flags for c, p, and s conversions */
#define L_FMTFLAGSC "-"
#endif
/*
** maximum size of each format specification (such as "%-099.99d")
** Maximum size of each format specification (such as "%-099.99d"):
** Initial '%', flags (up to 5), width (2), period, precision (2),
** length modifier (8), conversion specifier, and final '\0', plus some
** extra.
*/
#define MAX_FORMAT 32
@@ -1189,25 +1207,53 @@ static void addliteral (lua_State *L, luaL_Buffer *b, int arg) {
}
static const char *scanformat (lua_State *L, const char *strfrmt, char *form) {
const char *p = strfrmt;
while (*p != '\0' && strchr(L_FMTFLAGS, *p) != NULL) p++; /* skip flags */
if ((size_t)(p - strfrmt) >= sizeof(L_FMTFLAGS)/sizeof(char))
luaL_error(L, "invalid format (repeated flags)");
if (isdigit(uchar(*p))) p++; /* skip width */
if (isdigit(uchar(*p))) p++; /* (2 digits at most) */
if (*p == '.') {
p++;
if (isdigit(uchar(*p))) p++; /* skip precision */
if (isdigit(uchar(*p))) p++; /* (2 digits at most) */
static const char *get2digits (const char *s) {
if (isdigit(uchar(*s))) {
s++;
if (isdigit(uchar(*s))) s++; /* (2 digits at most) */
}
if (isdigit(uchar(*p)))
luaL_error(L, "invalid format (width or precision too long)");
return s;
}
/*
** Chech whether a conversion specification is valid. When called,
** first character in 'form' must be '%' and last character must
** be a valid conversion specifier. 'flags' are the accepted flags;
** 'precision' signals whether to accept a precision.
*/
static void checkformat (lua_State *L, const char *form, const char *flags,
int precision) {
const char *spec = form + 1; /* skip '%' */
spec += strspn(spec, flags); /* skip flags */
if (*spec != '0') { /* a width cannot start with '0' */
spec = get2digits(spec); /* skip width */
if (*spec == '.' && precision) {
spec++;
spec = get2digits(spec); /* skip precision */
}
}
if (!isalpha(uchar(*spec))) /* did not go to the end? */
luaL_error(L, "invalid conversion specification: '%s'", form);
}
/*
** Get a conversion specification and copy it to 'form'.
** Return the address of its last character.
*/
static const char *getformat (lua_State *L, const char *strfrmt,
char *form) {
/* spans flags, width, and precision ('0' is included as a flag) */
size_t len = strspn(strfrmt, L_FMTFLAGSF "123456789.");
len++; /* adds following character (should be the specifier) */
/* still needs space for '%', '\0', plus a length modifier */
if (len >= MAX_FORMAT - 10)
luaL_error(L, "invalid format (too long)");
*(form++) = '%';
memcpy(form, strfrmt, ((p - strfrmt) + 1) * sizeof(char));
form += (p - strfrmt) + 1;
*form = '\0';
return p;
memcpy(form, strfrmt, len * sizeof(char));
*(form + len) = '\0';
return strfrmt + len - 1;
}
@@ -1230,6 +1276,7 @@ static int str_format (lua_State *L) {
size_t sfl;
const char *strfrmt = luaL_checklstring(L, arg, &sfl);
const char *strfrmt_end = strfrmt+sfl;
const char *flags;
luaL_Buffer b;
luaL_buffinit(L, &b);
while (strfrmt < strfrmt_end) {
@@ -1239,25 +1286,35 @@ static int str_format (lua_State *L) {
luaL_addchar(&b, *strfrmt++); /* %% */
else { /* format item */
char form[MAX_FORMAT]; /* to store the format ('%...') */
int maxitem = MAX_ITEM;
char *buff = luaL_prepbuffsize(&b, maxitem); /* to put formatted item */
int nb = 0; /* number of bytes in added item */
int maxitem = MAX_ITEM; /* maximum length for the result */
char *buff = luaL_prepbuffsize(&b, maxitem); /* to put result */
int nb = 0; /* number of bytes in result */
if (++arg > top)
return luaL_argerror(L, arg, "no value");
strfrmt = scanformat(L, strfrmt, form);
strfrmt = getformat(L, strfrmt, form);
switch (*strfrmt++) {
case 'c': {
checkformat(L, form, L_FMTFLAGSC, 0);
nb = l_sprintf(buff, maxitem, form, (int)luaL_checkinteger(L, arg));
break;
}
case 'd': case 'i':
case 'o': case 'u': case 'x': case 'X': {
flags = L_FMTFLAGSI;
goto intcase;
case 'u':
flags = L_FMTFLAGSU;
goto intcase;
case 'o': case 'x': case 'X':
flags = L_FMTFLAGSX;
intcase: {
lua_Integer n = luaL_checkinteger(L, arg);
checkformat(L, form, flags, 1);
addlenmod(form, LUA_INTEGER_FRMLEN);
nb = l_sprintf(buff, maxitem, form, (LUAI_UACINT)n);
break;
}
case 'a': case 'A':
checkformat(L, form, L_FMTFLAGSF, 1);
addlenmod(form, LUA_NUMBER_FRMLEN);
nb = lua_number2strx(L, buff, maxitem, form,
luaL_checknumber(L, arg));
@@ -1268,12 +1325,14 @@ static int str_format (lua_State *L) {
/* FALLTHROUGH */
case 'e': case 'E': case 'g': case 'G': {
lua_Number n = luaL_checknumber(L, arg);
checkformat(L, form, L_FMTFLAGSF, 1);
addlenmod(form, LUA_NUMBER_FRMLEN);
nb = l_sprintf(buff, maxitem, form, (LUAI_UACNUMBER)n);
break;
}
case 'p': {
const void *p = lua_topointer(L, arg);
checkformat(L, form, L_FMTFLAGSC, 0);
if (p == NULL) { /* avoid calling 'printf' with argument NULL */
p = "(null)"; /* result */
form[strlen(form) - 1] = 's'; /* format it as a string */
@@ -1294,7 +1353,8 @@ static int str_format (lua_State *L) {
luaL_addvalue(&b); /* keep entire string */
else {
luaL_argcheck(L, l == strlen(s), arg, "string contains zeros");
if (!strchr(form, '.') && l >= 100) {
checkformat(L, form, L_FMTFLAGSC, 1);
if (strchr(form, '.') == NULL && l >= 100) {
/* no precision and string is too long to be formatted */
luaL_addvalue(&b); /* keep entire string */
}