[C++][Gandiva] out-of-bounds read in byte_substr when offset is past the end of the text #50380

Describe the bug, including details regarding any error messages, version, and platform.

byte_substr_binary_int32_int32 (cpp/src/gandiva/precompiled/string_ops.cc) computes startPos from the offset argument but never checks that it lands inside the input. With a positive offset greater than text_len, startPos = offset - 1 >= text_len, and the length truncation

if (startPos + length > text_len) {
  *out_len = text_len - startPos;   // negative
}
memcpy(ret, text + startPos, *out_len);

produces a negative *out_len. memcpy reads that as a very large size_t, so it reads past the end of text and writes past the text_len-sized output buffer.

Reproduced with a verbatim copy of the function under AddressSanitizer using byte_substr("TestString", 10, offset=15, length=10):

ERROR: AddressSanitizer: negative-size-param: (size=-4)
    #2 byte_substr_binary_int32_int32 string_ops.cc
0x... is located 4 bytes after 10-byte region
SUMMARY: AddressSanitizer: negative-size-param in byte_substr_binary_int32_int32

The offset comes from user SQL, so any byte_substr(col, n, m) with n beyond the row length triggers it.

Component(s)

C++, Gandiva