[Security Issue] rbs_string_strip_whitespace reads past empty ranges

Inizia in src/string.c, da rbs_string_strip_whitespace, e analizza i suoi chiamanti nel parsing di integer-literal e delle annotazioni. Esegui l’harness di AddressSanitizer fornito, quindi test/rbs/type_parsing_test.rb; il lavoro è completato quando gli intervalli vuoti non causano più una lettura out-of-bounds e i test elencati hanno esito positivo.

Summary

rbs_string_strip_whitespace dereferences the start of a string range before checking whether the range is empty. An empty range therefore produces an out-of-bounds read under AddressSanitizer. The parser calls this helper for annotation bodies, whose grammar permits an empty body (<>).

Affected current master: 5de6ecda19eaec2bd225db301674dc650dc220c3
Affected release checked: RBS 4.2.0 (8aee3b6fa6231dbd6fefe01f2ba20e3104228a6f)

Source

The current loop condition reads *new_start before evaluating new_start < self->end:

while (isspace(*new_start) && new_start < self->end) {
    new_start++;
}

The trailing-whitespace loop uses the same unsafe evaluation order. Call sites are in integer-literal and annotation parsing.

Reproduction

This small harness invokes the helper with a valid empty half-open range at the end of a one-byte allocation:

#include <stdlib.h>
#include "rbs/string.h"

int main(void) {
    char *storage = malloc(1);
    rbs_string_t empty = rbs_string_new(storage + 1, storage + 1);
    rbs_string_t stripped = rbs_string_strip_whitespace(&empty);
    free(storage);
    return stripped.start == stripped.end ? 0 : 1;
}

Compile with AddressSanitizer against src/string.c and run it. On macOS/arm64 with Apple clang 17.0.0:

ERROR: AddressSanitizer: heap-buffer-overflow
READ of size 1
    #0 rbs_string_strip_whitespace
    #1 main
0 bytes after 1-byte region

Suggested fix

Check the range before dereferencing it, and cast through unsigned char before passing bytes to isspace:

-while (isspace(*new_start) && new_start < self->end) {
+while (new_start < self->end && isspace((unsigned char) *new_start)) {
...
-while (isspace(*new_end) && new_start < new_end) {
+while (new_start < new_end && isspace((unsigned char) *new_end)) {

With that change, the ASan harness exits successfully. The native extension compiles on Ruby 4.0.6, and test/rbs/type_parsing_test.rb passes 40 tests / 525 assertions. I have not established code-execution impact; the demonstrated failure is an out-of-bounds read on an empty parser string range.