Core: kept JSON escaping within valid UTF-8

ngx_escape_json() escaped the JSON metacharacters and the control
bytes but copied every byte above 0x7f through unchanged, so a value
carrying malformed UTF-8 produced a quoted string that a strict
consumer rejects.  Such bytes are now replaced with U+FFFD, one per
offending byte.
This commit is contained in:
Dmitry Plotnikov 2026-10-29 15:34:07 +00:00
parent 67290bfea6
commit b2f6f09497

View file

@ -17,6 +17,7 @@ static void ngx_encode_base64_internal(ngx_str_t *dst, ngx_str_t *src,
const u_char *basis, ngx_uint_t padding);
static ngx_int_t ngx_decode_base64_internal(ngx_str_t *dst, ngx_str_t *src,
const u_char *basis);
static size_t ngx_json_utf8_sequence(u_char *src, size_t size);
void
@ -1901,22 +1902,119 @@ ngx_escape_html(u_char *dst, u_char *src, size_t size)
}
static size_t
ngx_json_utf8_sequence(u_char *src, size_t size)
{
u_char c;
size_t len;
uint32_t u, min;
c = src[0];
if (c < 0x80) {
return 1;
}
if (c < 0xc2) {
/* a continuation byte, or an overlong two-byte lead (0xc0, 0xc1) */
return 0;
}
if (c < 0xe0) {
u = c & 0x1f;
min = 0x80;
len = 2;
} else if (c < 0xf0) {
u = c & 0x0f;
min = 0x800;
len = 3;
} else if (c < 0xf5) {
u = c & 0x07;
min = 0x10000;
len = 4;
} else {
/* 0xf5..0xff would encode above U+10FFFF */
return 0;
}
if (size < len) {
return 0;
}
for (size = 1; size < len; size++) {
c = src[size];
if ((c & 0xc0) != 0x80) {
return 0;
}
u = (u << 6) | (c & 0x3f);
}
if (u < min) {
/* overlong encoding */
return 0;
}
if (u >= 0xd800 && u <= 0xdfff) {
/* surrogate halves are not Unicode scalar values */
return 0;
}
if (u > 0x10ffff) {
return 0;
}
return len;
}
uintptr_t
ngx_escape_json(u_char *dst, u_char *src, size_t size)
{
u_char *run;
u_char ch;
size_t ulen;
ngx_uint_t len;
if (dst == NULL) {
len = 0;
while (size) {
ch = *src++;
ch = *src;
if (ch == '\\' || ch == '"') {
len++;
/* skip the run of bytes that need no escaping */
} else if (ch <= 0x1f) {
while (ch > 0x1f && ch < 0x80 && ch != '\\' && ch != '"') {
src++;
if (--size == 0) {
return (uintptr_t) len;
}
ch = *src;
}
if (ch > 0x1f) {
if (ch < 0x80) {
len++;
ulen = 1;
} else {
ulen = ngx_json_utf8_sequence(src, size);
if (ulen == 0) {
/* malformed: one U+FFFD per offending byte */
len += sizeof("\xEF\xBF\xBD") - 2;
ulen = 1;
}
}
} else {
switch (ch) {
case '\n':
@ -1930,26 +2028,80 @@ ngx_escape_json(u_char *dst, u_char *src, size_t size)
default:
len += sizeof("\\u001F") - 2;
}
ulen = 1;
}
size--;
src += ulen;
size -= ulen;
}
return (uintptr_t) len;
}
while (size) {
ch = *src++;
ch = *src;
if (ch > 0x1f) {
if (ch == '\\' || ch == '"') {
if (ch < 0x80) {
if (ch != '\\' && ch != '"') {
/* copy the run with one memcpy, not byte by byte */
run = src;
do {
src++;
size--;
if (size == 0) {
break;
}
ch = *src;
} while (ch > 0x1f && ch < 0x80
&& ch != '\\' && ch != '"');
dst = ngx_cpymem(dst, run, src - run);
continue;
}
*dst++ = '\\';
*dst++ = ch;
src++;
size--;
continue;
}
*dst++ = ch;
ulen = ngx_json_utf8_sequence(src, size);
if (ulen == 0) {
/* malformed: replace the offending byte with U+FFFD */
*dst++ = 0xef; *dst++ = 0xbf; *dst++ = 0xbd;
src++;
size--;
continue;
}
size -= ulen;
while (ulen--) {
*dst++ = *src++;
}
continue;
} else {
src++;
*dst++ = '\\';
switch (ch) {