File size: 2,519 Bytes
18575f4
cf8a10d
44ad462
cf8a10d
 
44ad462
cf8a10d
44ad462
 
 
 
 
cf8a10d
44ad462
18575f4
cf8a10d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44ad462
cf8a10d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44ad462
cf8a10d
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
---
language:
  - en
license: mit
task_categories:
  - text-generation
tags:
  - assembly
  - security
  - compiler-bugs
  - CISB
  - program-repair
size_categories:
  - n<1K
---

# CISB-O2: Compiler-Introduced Security Bugs at -O2

## Overview

This dataset collects the **unsafe** half of the **Compiler-Introduced Security Bugs
(CISB)** benchmark: each record is a CISB case compiled with LLVM/Clang-17 at **-O2**,
where the optimizer eliminated or bypassed a security check that is still present in
the **-O0** build.

Because the -O0 build is the safe variant and the -O2 build is the unsafe one, every
record here is unsafe by construction -- there is no `flag` column. The intended task is
**repair**: given the unsafe -O2 assembly and the verifier's failure output, produce a
corrected assembly that restores the security property without changing other behavior.

## Dataset Structure

| Column | Description |
|--------|-------------|
| `task_name` | Task identifier, e.g. `b-4-var-0` |
| `source_code` | Original C source (`code.c`) |
| `x86_64` | x86-64 assembly at -O2 (**Intel syntax**, built with `-masm=intel`) |
| `aarch64_linux` | AArch64 assembly at -O2 |
| `riscv` | RISC-V RV64 (`rv64gc`/`lp64d`) assembly at -O2 |
| `run_output` | Combined stdout+stderr of the verifier harness (`test.c`) run against the x86-64 build |
| `test_c` | C source of the verifier harness (`test.c`) -- links against the assembly under test and reports its verdict via exit code |

## Statistics

- **Total records**: 36
- **Bug families**: 6 (`b-4`, `b-9`, `b-12`, `b-14`, `b-21`, `l-30`), 6 variants each

## Security Oracle

The `test.c` harness reports its verdict through the process exit code:

- `0` -> **SAFE** (security check preserved)
- `1` -> **UNSAFE** (security check eliminated) -- the expected state for every record here
- `2` -> **HARNESS_ERROR**
- killed by signal -> **CRASH**; no exit in time -> **TIMEOUT**

The `run_output` column captures what that harness printed while returning UNSAFE, so it
can be fed to a model as the failure description for a repair task.

## Usage

```python
from datasets import load_dataset

ds = load_dataset("Akirayasha/cisb-O2", split="test")

record = ds[0]
unsafe_asm = record["x86_64"]      # assembly to repair
failure     = record["run_output"] # why the verifier says it is unsafe
harness     = record["test_c"]     # verifier source, to rebuild/rerun it against a candidate
```

## Citation

Part of the CISC-to-RISC transpilation research project at MBZUAI.