Performing Malware Triage With Yara

Performing Malware Triage with YARA

When to Use

• Rapidly classifying a large batch of malware samples against known family signatures
• Writing detection rules for a newly analyzed malware family based on unique byte patterns
• Scanning file shares, endpoints, or memory dumps for indicators of a specific threat
• Building automated triage pipelines that classify samples before manual analysis
• Hunting for variants of a known threat across an enterprise using YARA scans

Do not use as the sole analysis method; YARA triage identifies known patterns but does not reveal new or unknown malware behaviors.

Common Misconfigurations & Verification

• Over-fitting to the packer stub. Rules built from a UPX/Themida-packed sample match the packer, not the family — they fire on unrelated packed goodware and miss unpacked variants. Write strings/hex from the unpacked payload (mutexes, C2 templates, decryption constants), not the loader.
• Weak or missing anchors make rules slow and noisy. A condition that is just 2 of ($s*) with no uint16(0) == 0x5A4D / filesize < N gate forces YARA to scan every file fully. Always lead with a cheap discriminator ($mz at 0, pe.is_pe, filesize bound) so the expensive string matching short-circuits.
• Atom problems and regex cost. Short hex atoms ({ 00 00 }), fully-wildcarded patterns ({ ?? ?? ?? ?? }), and unbounded regex (/.*/) trigger YARA's "slowing down scanning" warning and cause timeouts. Use yara -p rule.yar sample (atom profiling) and keep ≥4 fixed contiguous bytes per pattern.
• False positives from shared library strings. Strings from OpenSSL, the VC++ runtime, Go/Rust std, or common installers will match clean files. Never use them as sole anchors.
• Verify it actually fires AND stays quiet: confirm true positive with yara -s family.yar known_sample (the -s output must show your intended strings hit at sane offsets, not a coincidental match). Then run against a goodware corpus — yara -r family.yar /clean/ | wc -l must be 0. In Python: yara.compile(...).match(path) over a labeled set to measure FP/FN before deployment.
• Test variant resilience: match against multiple samples of the family, not one hash, so the rule survives recompilation and minor string changes.

Prerequisites

• YARA 4.x installed (apt install yara or pip install yara-python)
• YARA rule repositories (YARA-Rules, awesome-yara, Malpedia rules, Florian Roth's signature-base)
• Python 3.8+ with yara-python for scripted scanning
• Sample collection organized in a directory structure for batch scanning
• Understanding of PE file format, hex patterns, and regular expressions for rule writing

Workflow

Step 1: Scan Samples with Existing Rule Sets

Apply community and commercial YARA rules to classify samples:

# Scan a single file
yara -s malware_rules.yar suspect.exe

# Scan a directory of samples
yara -r malware_rules.yar /path/to/samples/

# Scan with multiple rule files
yara -r rules/apt_rules.yar rules/ransomware_rules.yar rules/trojan_rules.yar suspect.exe

# Scan with timeout (prevent hanging on large files)
yara -t 30 malware_rules.yar suspect.exe

# Scan and show matching strings
yara -s -r malware_rules.yar suspect.exe

# Scan with compiled rules (faster for repeated scans)
yarac malware_rules.yar compiled_rules.yarc
yara compiled_rules.yarc suspect.exe

# Download community rule sets
git clone https://github.com/Yara-Rules/rules.git yara-community-rules
git clone https://github.com/Neo23x0/signature-base.git signature-base

# Scan with signature-base
yara -r signature-base/yara/*.yar suspect.exe

Step 2: Write Rules for Unique String Patterns

Create YARA rules based on strings extracted during malware analysis:

rule MalwareX_Strings {
    meta:
        description = "Detects MalwareX based on unique strings"
        author = "analyst"
        date = "2025-09-15"
        reference = "Internal Analysis Report #1547"
        hash = "e3b0c44298fc1c149afbf4c8996fb924"
        tlp = "WHITE"

    strings:
        // C2 URL pattern
        $url1 = "/gate.php?id=" ascii
        $url2 = "/panel/connect.php" ascii

        // Unique mutex name
        $mutex = "Global\\CryptLocker_2025" ascii wide

        // User-Agent string
        $ua = "Mozilla/5.0 (compatible; MSIE 10.0)" ascii

        // Registry persistence path
        $reg = "Software\\Microsoft\\Windows\\CurrentVersion\\Run\\WindowsUpdate" ascii

        // Campaign identifier
        $campaign = "campaign_2025_q3" ascii

    condition:
        uint16(0) == 0x5A4D and      // PE file (MZ header)
        filesize < 500KB and          // Size constraint
        ($url1 or $url2) and          // At least one C2 URL
        ($mutex or $campaign) and     // Campaign identifier
        $ua                           // Specific User-Agent
}

Step 3: Write Rules for Byte Patterns

Create rules matching specific code sequences:

rule MalwareX_Decryptor {
    meta:
        description = "Detects MalwareX XOR decryption routine"
        author = "analyst"
        date = "2025-09-15"

    strings:
        // XOR decryption loop (x86 assembly)
        // mov al, [esi+ecx]
        // xor al, [edi+ecx]
        // mov [esi+ecx], al
        // inc ecx
        // cmp ecx, edx
        // jl loop
        $xor_loop = { 8A 04 0E 32 04 0F 88 04 0E 41 3B CA 7C F3 }

        // RC4 KSA initialization (256-byte loop)
        $rc4_ksa = { 33 C0 88 04 ?8 40 3D 00 01 00 00 7? }

        // Embedded RSA public key marker
        $rsa_key = { 06 02 00 00 00 A4 00 00 52 53 41 31 }  // PUBLICKEYBLOB

    condition:
        uint16(0) == 0x5A4D and
        ($xor_loop or $rc4_ksa) and
        $rsa_key
}

Step 4: Write Rules with PE Module

Leverage YARA's PE module for structural detection:

import "pe"
import "hash"
import "math"

rule MalwareX_PE_Characteristics {
    meta:
        description = "Detects MalwareX by PE structure and imports"
        author = "analyst"

    condition:
        pe.is_pe and

        // Compiled within specific timeframe
        pe.timestamp > 1693526400 and   // After 2023-09-01
        pe.timestamp < 1727740800 and   // Before 2024-10-01

        // Specific import hash
        pe.imphash() == "a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6" or

        // Suspicious import combination
        (
            pe.imports("kernel32.dll", "VirtualAllocEx") and
            pe.imports("kernel32.dll", "WriteProcessMemory") and
            pe.imports("kernel32.dll", "CreateRemoteThread") and
            pe.imports("wininet.dll", "InternetOpenA")
        ) or

        // High entropy .text section (packed)
        (
            for any section in pe.sections : (
                section.name == ".text" and
                math.entropy(section.raw_data_offset, section.raw_data_size) > 7.0
            )
        )
}

rule MalwareX_Rich_Header {
    meta:
        description = "Detects MalwareX by Rich header hash"

    condition:
        pe.is_pe and
        hash.md5(pe.rich_signature.clear_data) == "abc123def456abc123def456abc123de"
}

Step 5: Batch Triage with Python

Automate scanning of sample collections:

import yara
import os
import json
import hashlib
from datetime import datetime

# Compile all rule files
rule_files = {
    "apt": "rules/apt_rules.yar",
    "ransomware": "rules/ransomware_rules.yar",
    "trojan": "rules/trojan_rules.yar",
    "custom": "rules/custom_rules.yar",
}
rules = yara.compile(filepaths=rule_files)

# Scan sample directory
results = []
sample_dir = "/path/to/samples"

for filename in os.listdir(sample_dir):
    filepath = os.path.join(sample_dir, filename)
    if not os.path.isfile(filepath):
        continue

    with open(filepath, "rb") as f:
        data = f.read()
        sha256 = hashlib.sha256(data).hexdigest()

    matches = rules.match(filepath)

    result = {
        "filename": filename,
        "sha256": sha256,
        "size": len(data),
        "matches": [],
        "classification": "UNKNOWN",
    }

    for match in matches:
        result["matches"].append({
            "rule": match.rule,
            "namespace": match.namespace,
            "tags": match.tags,
            "strings": [(hex(s[0]), s[1], s[2].decode("utf-8", errors="replace")[:100])
                       for s in match.strings] if match.strings else []
        })

    if result["matches"]:
        result["classification"] = result["matches"][0]["namespace"].upper()

    results.append(result)

# Summary
classified = sum(1 for r in results if r["classification"] != "UNKNOWN")
print(f"Scanned: {len(results)} samples")
print(f"Classified: {classified} ({classified/len(results)*100:.1f}%)")
print(f"Unknown: {len(results)-classified}")

# Export results
with open("triage_results.json", "w") as f:
    json.dump(results, f, indent=2)

Step 6: Validate and Optimize Rules

Test rules for false positives and performance:

# Test rule syntax
yara -C custom_rules.yar

# Scan known-clean directory to check false positives
yara -r custom_rules.yar /path/to/clean_files/ > false_positives.txt
wc -l false_positives.txt

# Benchmark rule performance
time yara -r custom_rules.yar /path/to/large_sample_collection/

# Profile individual rule performance
yara -p custom_rules.yar suspect.exe

Key Concepts

Tools & Systems

• YARA: Pattern matching engine for identifying and classifying malware based on text, hex, and structural patterns
• yara-python: Python bindings for YARA enabling scripted scanning, rule compilation, and integration with analysis pipelines
• yarGen: Automatic YARA rule generator that creates rules from malware samples by identifying unique strings and opcodes
• YARA-Rules (GitHub): Community-maintained repository of YARA rules covering malware families, exploits, and suspicious indicators
• Malpedia YARA: Curated YARA rules from the Malpedia malware encyclopedia with high-quality family-specific rules

Common Scenarios

Scenario: Creating Detection Rules for a New Malware Family

Context: Reverse engineering of a new malware sample has identified unique strings, byte patterns, and PE characteristics. YARA rules are needed for enterprise-wide hunting and ongoing detection.

Approach:

1. Extract unique strings from the unpacked binary (C2 URLs, mutex names, registry paths)
2. Identify unique byte sequences from the encryption routine or C2 protocol (from Ghidra analysis)
3. Record PE characteristics (imphash, Rich header hash, section names, compilation timestamp range)
4. Write a YARA rule combining string, byte pattern, and PE module conditions
5. Test against the known malware samples to confirm true positive detection
6. Test against a clean file corpus (Windows system files, common applications) to verify zero false positives
7. Deploy to enterprise scanning infrastructure and threat intelligence platform

Pitfalls:

• Writing rules too specific to a single sample (will not detect variants with minor changes)
• Writing rules too generic (matching legitimate software, causing false positives)
• Using strings that appear in common libraries or frameworks (e.g., OpenSSL strings)
• Not testing on a sufficiently large clean corpus before deployment

Output Format

YARA TRIAGE RESULTS
=====================
Scan Date:        2025-09-15
Rule Sets:        apt_rules (847 rules), ransomware_rules (312 rules),
                  trojan_rules (1,204 rules), custom_rules (45 rules)
Samples Scanned:  2,500
Processing Time:  47 seconds

CLASSIFICATION SUMMARY
APT:              12 samples (0.5%)
Ransomware:       187 samples (7.5%)
Trojan:           423 samples (16.9%)
Unknown:          1,878 samples (75.1%)

TOP MATCHING RULES
Rule                         Matches  Family
MalwareX_C2_Beacon           45       MalwareX
LockBit3_Ransom_Note         38       LockBit 3.0
Emotet_Epoch5_Loader         32       Emotet
CobaltStrike_Beacon_Config   28       Cobalt Strike
QakBot_DLL_Loader            25       QakBot

SAMPLE DETAIL
File:    suspect.exe
SHA-256: e3b0c44298fc1c149afbf4c8996fb924...
Matches:
  [1] MalwareX_Strings (custom)
      - $url1 at 0x4A20: "/gate.php?id="
      - $mutex at 0x5100: "Global\\CryptLocker_2025"
  [2] MalwareX_Decryptor (custom)
      - $xor_loop at 0x401200: { 8A 04 0E 32 04 0F ... }
  [3] MalwareX_PE_Characteristics (custom)
      - PE import combination matched
Classification: MALWAREX (HIGH CONFIDENCE)