Repository Public License
1// ORIX Secure2// File: Packages/ORIXFileInspection/Tests/ORIXFileInspectionTests/MagicDetectorTests.swift3// Description: Type detection from content on minimal samples of every recognised format, and its edge cases.4//5// Author: Simon-Pierre Boucher6// Contact: contact@spboucher.ai7import Foundation8import ORIXSecureModel9import Testing10@testable import ORIXFileInspection1112@Suite("Magic-number detection")13struct MagicDetectorTests {14 @Test("Each minimal sample is recognised", arguments: Samples.all)15 func recognises(_ sample: Sample) throws {16 let core = try inspectBytes(sample.bytes, name: sample.name)17 #expect(core.type == sample.type, "\(sample.name): detected \(core.type), signature \(core.signature)")18 expectHonest(core.findings)19 }2021 @Test("Samples named after their type raise no extension finding", arguments: Samples.all)22 func noMismatchOnHonestNames(_ sample: Sample) throws {23 let core = try inspectBytes(sample.bytes, name: sample.name)24 let mismatch = core.findings.filter { $0.id.hasPrefix("file.type.extension") || $0.id == "file.type.executableDisguised" || $0.id == "file.type.signatureMissing" }25 #expect(mismatch.isEmpty, "\(sample.name): \(mismatch.map(\.title))")26 }2728 @Test("The sample set covers every detectable type")29 func coverage() {30 let covered = Set(Samples.all.map(\.type))31 let missing = DetectedFileType.allCases.filter { !covered.contains($0) }32 #expect(missing.isEmpty, "no sample for \(missing)")33 }3435 @Test("Detection through a real file gives the same answer")36 func throughFiles() throws {37 let dir = try TempDir()38 for sample in Samples.all {39 let url = try dir.write(sample.name, sample.bytes)40 let type = try MagicDetector.detect(fileAt: url)41 #expect(type == sample.type, "\(sample.name)")42 }43 }4445 @Test("MZ without a PE header at e_lfanew is not called a Windows program")46 func mzWithoutPE() throws {47 var data = Samples.pe48 data[0x80] = 0x4E49 let wrong = try MagicDetector.detect(source: DataByteSource(data))50 #expect(wrong.type == .dosExecutable)51 var beyond = Samples.pe52 beyond.replaceSubrange(0x3C..<0x40, with: le32(0xFFFF_FFF0))53 let outside = try MagicDetector.detect(source: DataByteSource(beyond))54 #expect(outside.type == .dosExecutable)55 }5657 @Test("CAFEBABE: universal Mach-O and Java class are told apart")58 func cafebabe() throws {59 let fat = try MagicDetector.detect(source: DataByteSource([0xCA, 0xFE, 0xBA, 0xBE] + be32(3) + zeros(80)))60 #expect(fat.type == .machOFat)61 let java8 = try MagicDetector.detect(source: DataByteSource([0xCA, 0xFE, 0xBA, 0xBE] + be16(0) + be16(52) + zeros(80)))62 #expect(java8.type == .javaClass)63 let java21 = try MagicDetector.detect(source: DataByteSource([0xCA, 0xFE, 0xBA, 0xBE] + be16(0) + be16(65) + zeros(80)))64 #expect(java21.type == .javaClass)65 // Claims 30 architectures but is far too short to hold them.66 let lying = try MagicDetector.detect(source: DataByteSource([0xCA, 0xFE, 0xBA, 0xBE] + be32(30) + zeros(16)))67 #expect(lying.type != .machOFat)68 }6970 @Test("A PDF header within the first 1024 bytes is still a PDF")71 func pdfHeaderOffset() throws {72 let data = bytes("junk-line\n") + Samples.pdf73 let core = try inspectBytes(data, name: "shifted.pdf")74 #expect(core.type == .pdf)75 #expect(core.pdf?.headerOffset == 10)76 #expect(core.findings.has("file.pdf.headerOffset"))77 let late = [UInt8](repeating: 0x41, count: 2000) + Samples.pdf78 let other = try MagicDetector.detect(source: DataByteSource(late))79 #expect(other.type == .text)80 }8182 @Test("Text and binary are told apart")83 func textVersusBinary() throws {84 let nul = try MagicDetector.detect(source: DataByteSource(bytes("abc\u{0}def")))85 #expect(nul.type == .unknown)86 let latin1 = try MagicDetector.detect(source: DataByteSource([0x63, 0x61, 0x66, 0xE9, 0x20, 0x21]))87 #expect(latin1.type == .unknown)88 // A sample cut in the middle of a multi-byte character stays text.89 var long = [UInt8](repeating: 0x61, count: MagicDetector.headLength - 1)90 long += bytes("é") + bytes(" suite")91 let cut = try MagicDetector.detect(source: DataByteSource(long))92 #expect(cut.type == .text)93 let script = try MagicDetector.detect(source: DataByteSource(bytes("#!/bin/zsh -e\nls\n")))94 #expect(script.interpreter == "/bin/zsh -e")95 }9697 @Test("A UTF-16 byte order mark is not mistaken for MP3 frame sync")98 func utf16NotMP3() throws {99 let detection = try MagicDetector.detect(source: DataByteSource([0xFF, 0xFE, 0x41, 0x00, 0x42, 0x00]))100 #expect(detection.type == .utf16Text)101 }102103 @Test("The ISO media brand decides between image, audio and video")104 func brands() throws {105 let cases: [(String, [String], DetectedFileType)] = [106 ("heix", [], .heic), ("msf1", ["hevc"], .heic), ("mif1", ["avif"], .avif), ("avis", [], .avif),107 ("M4B ", [], .m4a), ("mp42", ["isom"], .mp4), ("3gp5", [], .mp4), ("qt ", [], .quickTime),108 ]109 for (major, compatible, expected) in cases {110 let detection = try MagicDetector.detect(source: DataByteSource(Samples.ftyp(major, compatible)))111 #expect(detection.type == expected, "brand \(major)")112 }113 }114115 @Test("ZIP containers are refined from entry names only")116 func zipRefinement() throws {117 // OOXML markers without [Content_Types].xml stay a plain ZIP.118 let fake = buildZip([ZipEntrySpec("word/document.xml", bytes("<x/>"))])119 let core = try inspectBytes(fake, name: "fake.docx")120 #expect(core.type == .zip)121 #expect(core.findings.has("file.ooxml.contentTypes"))122 // A compressed (not stored) mimetype entry is not read; the container hint still works.123 let epub = buildZip([ZipEntrySpec("mimetype", bytes("xxxx"), method: 8), ZipEntrySpec("META-INF/container.xml", bytes("<c/>"))])124 let detected = try inspectBytes(epub, name: "book.epub")125 #expect(detected.type == .epub)126 }127128 @Test("A DMG is recognised by its trailer even when it starts like GZIP or BZIP2")129 func dmgTrailer() throws {130 let gzipStart = [0x1F, 0x8B, 0x08, 0x00] + zeros(700) + Samples.koly131 let first = try MagicDetector.detect(source: DataByteSource(gzipStart))132 #expect(first.type == .dmg)133 // A trailer with the wrong version is ignored.134 var bad = Samples.koly135 bad[7] = 9136 let second = try MagicDetector.detect(source: DataByteSource([0x78, 0x9C] + zeros(600) + bad))137 #expect(second.type == .unknown)138 }139140 @Test("Display names exist for every type and avoid verdict words")141 func displayNames() {142 for type in DetectedFileType.allCases {143 #expect(!type.displayName.isEmpty)144 for word in forbiddenWords { #expect(!type.displayName.lowercased().contains(word)) }145 }146 }147}148