Initial release

Sources/KnotchyCore/Backend/ContainerCLIBackend.swift

@@ -0,0 +1,311 @@
+import Foundation
+import Synchronization
+
+public final class ContainerCLIBackend: VMBackend {
+    public static let image = "local/knotchy-vm:latest"
+    public static let staleAfterDays = 14
+
+    private let cli: CLIRunner
+    private let vmSources: URL?
+    private let runDirectory: URL
+    private let startsInFlight = Mutex(0)
+    private let route = Mutex(NetworkRoute.direct)
+    private let startedRoute = Mutex(NetworkRoute.direct)
+
+    public var lastStartedRoute: NetworkRoute { startedRoute.withLock { $0 } }
+
+    public func setNetworkRoute(_ newRoute: NetworkRoute) {
+        route.withLock { $0 = newRoute }
+    }
+
+    /// Test-only (knotchyctl --allow-local-proxy); the app never sets it.
+    private let localProxyAllowed = Mutex(false)
+
+    public func allowLocalProxy() {
+        localProxyAllowed.withLock { $0 = true }
+    }
+
+    public init(vmSources: URL?, cli: CLIRunner = CLIRunner()) {
+        self.vmSources = vmSources
+        self.cli = cli
+        runDirectory = URL(fileURLWithPath: NSTemporaryDirectory()).appendingPathComponent("knotchy", isDirectory: true)
+    }
+
+    // MARK: Preflight
+
+    public func preflight() async -> PreflightResult {
+        guard cli.executable != nil else { return .issue(.cliMissing) }
+        if case .outdated(let found) = await runtimeStatus() { return .issue(.cliOutdated(found: found)) }
+        guard let status = try? await cli.run(["system", "status"], timeout: 15), status.status == 0,
+              field("status", in: status.text) == "running"
+        else { return .issue(.serviceStopped) }
+        let appRoot = field("paths.appRoot", in: status.text)
+            ?? NSHomeDirectory() + "/Library/Application Support/com.apple.container/"
+        let kernel = URL(fileURLWithPath: appRoot).appendingPathComponent("kernels/default.kernel-arm64")
+        guard FileManager.default.fileExists(atPath: kernel.path) else { return .issue(.kernelMissing) }
+
+        guard let inspect = try? await cli.run(["image", "inspect", Self.image], timeout: 15), inspect.status == 0
+        else { return .issue(.imageMissing) }
+        let labels = Self.labels(in: inspect.text)
+        let found = labels["aculix.knotchy.protocol"].flatMap(Int.init) ?? 0
+        guard found == WireProtocol.version else {
+            return .issue(.imageOutdated(found: found, need: WireProtocol.version))
+        }
+        prepareRunDirectory()
+        return .ready(appRoot: appRoot, staleImage: Self.staleness(built: labels["aculix.knotchy.built"]))
+    }
+
+    /// Resolved on the Mac: a proxy route gives the VM no DNS, and a VPN's DNS needs the tunnel up.
+    static func resolveIPv4(_ host: String) throws -> String {
+        var address = host
+        if !WireGuardConfig.isIPv4(host) {
+            var hints = addrinfo()
+            hints.ai_family = AF_INET
+            hints.ai_socktype = SOCK_STREAM
+            var result: UnsafeMutablePointer<addrinfo>?
+            guard getaddrinfo(host, nil, &hints, &result) == 0, let first = result, let sockaddr = first.pointee.ai_addr else {
+                throw VMBackendError.failed("Couldn't find the server “\(host)”. Check Settings → Network.")
+            }
+            defer { freeaddrinfo(result) }
+            var buffer = [CChar](repeating: 0, count: Int(INET_ADDRSTRLEN))
+            var inAddr = sockaddr.withMemoryRebound(to: sockaddr_in.self, capacity: 1) { $0.pointee.sin_addr }
+            guard inet_ntop(AF_INET, &inAddr, &buffer, socklen_t(buffer.count)) != nil else {
+                throw VMBackendError.failed("Couldn't find the server “\(host)”.")
+            }
+            address = String(cString: buffer)
+        }
+        if address.hasPrefix("127.") || address == "0.0.0.0" {
+            throw VMBackendError.failed("The VM can't reach this Mac's localhost (“\(host)”). A proxy on this Mac must listen on the Mac's network address instead.")
+        }
+        return address
+    }
+
+    public func imageStatus() async -> ImageStatus {
+        guard cli.executable != nil else { return .noContainerTool }
+        guard let inspect = try? await cli.run(["image", "inspect", Self.image], timeout: 15), inspect.status == 0
+        else { return .missing }
+        let built = Self.labels(in: inspect.text)["aculix.knotchy.built"].flatMap { ISO8601DateFormatter().date(from: $0) }
+        return .built(built)
+    }
+
+    public enum ImageStatus: Equatable, Sendable {
+        case noContainerTool
+        case missing
+        case built(Date?)
+    }
+
+    public func runtimeStatus() async -> RuntimeStatus {
+        guard cli.executable != nil else { return .missing }
+        let output = try? await cli.run(["--version"], timeout: 10)
+        return ContainerRuntime.status(versionOutput: output?.text ?? "")
+    }
+
+    public func stopService() async throws {
+        let result = try await cli.run(["system", "stop"], timeout: 60)
+        guard result.status == 0 else { throw VMBackendError.failed("Couldn't stop the container service: \(result.lastLine)") }
+    }
+
+    private func field(_ name: String, in text: String) -> String? {
+        for line in text.split(whereSeparator: \.isNewline) {
+            let parts = line.split(separator: " ", maxSplits: 1, omittingEmptySubsequences: true)
+            if parts.count == 2, parts[0] == name {
+                return parts[1].trimmingCharacters(in: .whitespaces)
+            }
+        }
+        return nil
+    }
+
+    static func labels(in json: String) -> [String: String] {
+        var found: [String: String] = [:]
+        func walk(_ value: Any) {
+            if let dict = value as? [String: Any] {
+                for (key, child) in dict {
+                    if key.hasPrefix("aculix.knotchy."), let text = child as? String { found[key] = text }
+                    walk(child)
+                }
+            } else if let array = value as? [Any] {
+                array.forEach(walk)
+            }
+        }
+        if let object = try? JSONSerialization.jsonObject(with: Data(json.utf8)) { walk(object) }
+        return found
+    }
+
+    static func staleness(built: String?, now: Date = Date()) -> StaleImage? {
+        let formatter = ISO8601DateFormatter()
+        guard let built, let date = formatter.date(from: built) else { return StaleImage(builtDaysAgo: nil) }
+        let days = Int(now.timeIntervalSince(date) / 86_400)
+        return days > staleAfterDays ? StaleImage(builtDaysAgo: days) : nil
+    }
+
+    private func prepareRunDirectory() {
+        let fm = FileManager.default
+        let path = runDirectory.path
+        var isDirectory: ObjCBool = false
+        let attributes = try? fm.attributesOfItem(atPath: path)
+        let isSymlink = (attributes?[.type] as? FileAttributeType) == .typeSymbolicLink
+        let owner = (attributes?[.ownerAccountID] as? NSNumber)?.uint32Value
+        let mode = (attributes?[.posixPermissions] as? NSNumber)?.intValue
+        if isSymlink || !fm.fileExists(atPath: path, isDirectory: &isDirectory) || !isDirectory.boolValue
+            || owner != getuid() || mode != 0o700 {
+            try? fm.removeItem(atPath: path)
+            try? fm.createDirectory(atPath: path, withIntermediateDirectories: true,
+                                    attributes: [.posixPermissions: 0o700])
+        }
+        // Secrets left by a crash; a VM's init deletes its own within a second of booting.
+        let stale = Date().addingTimeInterval(-120)
+        for name in (try? fm.contentsOfDirectory(atPath: path)) ?? [] where name.hasSuffix(".secrets") || name.hasSuffix(".env") {
+            let item = runDirectory.appendingPathComponent(name)
+            if let modified = (try? fm.attributesOfItem(atPath: item.path))?[.modificationDate] as? Date, modified < stale {
+                try? fm.removeItem(at: item)
+            }
+        }
+    }
+
+    // MARK: Start and destroy
+
+    public func start(id: String, url: URL, renderSize: PixelSize, blockCIDRs: [String]) async throws -> VMSession {
+        startsInFlight.withLock { $0 += 1 }
+        defer { startsInFlight.withLock { $0 -= 1 } }
+        prepareRunDirectory()
+        let socket = runDirectory.appendingPathComponent("\(id).sock").path
+        unlink(socket)
+        let chosenRoute = route.withLock { $0 }
+        let launch = try chosenRoute.launch(resolve: Self.resolveIPv4, hostNetworks: blockCIDRs,
+                                            allowLocalProxy: localProxyAllowed.withLock { $0 })
+        var args = [
+            "run", "-d", "--rm", "--name", id,
+            "--cpus", "4", "--memory", "2g", "--shm-size", "512m",
+            "--dns", "1.1.1.1",
+            "--cap-add", "NET_ADMIN",
+            // WebKit's bwrap can't mount procfs over a masked /proc, and init writes /proc/sys.
+            "--masked-path", "NONE", "--read-only-path", "NONE",
+            "--publish-socket", "\(socket):/tmp/knotchy/knotchy.sock",
+            "-e", "KNOTCHY_URL=\(url.absoluteString)",
+            "-e", "KNOTCHY_SIZE=\(renderSize)",
+            "-e", "KNOTCHY_BLOCK_CIDRS=\(blockCIDRs.joined(separator: " "))",
+        ]
+        for setting in launch.environment { args += ["-e", setting] }
+        if !launch.secrets.isEmpty {
+            // Init reads, deletes and unmounts this (needs SYS_ADMIN), then drops all capabilities.
+            let share = secretsDirectory(id)
+            try? FileManager.default.removeItem(at: share)
+            guard (try? FileManager.default.createDirectory(at: share, withIntermediateDirectories: false,
+                                                             attributes: [.posixPermissions: 0o700])) != nil,
+                  FileManager.default.createFile(atPath: share.appendingPathComponent("route").path,
+                                                 contents: Data((launch.secrets.joined(separator: "\n") + "\n").utf8),
+                                                 attributes: [.posixPermissions: 0o600])
+            else {
+                try? FileManager.default.removeItem(at: share)
+                throw VMBackendError.failed("Couldn't pass the network settings to the VM.")
+            }
+            args += ["--cap-add", "SYS_ADMIN", "--mount", "type=bind,source=\(share.path),target=/run/knotchy-secrets"]
+        }
+        args.append(Self.image)
+        let output: CLIRunner.Output
+        do {
+            output = try await cli.run(args, timeout: 60)
+        } catch {
+            throw VMBackendError.failed("Starting the VM timed out.")
+        }
+        guard output.status == 0 else {
+            cleanUp(id)
+            throw VMBackendError.failed("Couldn't start the VM: \(output.lastLine)")
+        }
+        startedRoute.withLock { $0 = chosenRoute }
+        return VMSession(id: id, socketPath: socket)
+    }
+
+    public func destroy(id: String) async throws {
+        _ = try? await cli.run(["stop", "-t", "2", id], timeout: 10)
+        if await waitUntilGone(id, for: 5) { return cleanUp(id) }
+        _ = try? await cli.run(["delete", "--force", id], timeout: 10)
+        if await waitUntilGone(id, for: 5) { return cleanUp(id) }
+        throw VMBackendError.destroyStuck(id: id)
+    }
+
+    public func forceDestroy(id: String) async throws {
+        _ = try? await cli.run(["delete", "--force", id], timeout: 10)
+        if await waitUntilGone(id, for: 5) { return cleanUp(id) }
+        throw VMBackendError.destroyStuck(id: id)
+    }
+
+    public func sweepOrphans(keeping: Set<String>) async {
+        guard let listed = try? await cli.run(["ls", "-a", "-q"], timeout: 10) else { return }
+        for id in Self.ids(in: listed.text) where Self.isSessionID(id) && !keeping.contains(id) {
+            _ = try? await cli.run(["delete", "--force", id], timeout: 10)
+            cleanUp(id)
+        }
+    }
+
+    static func isSessionID(_ id: String) -> Bool {
+        id.count == 16 && id.hasPrefix("knotchy-") && id.dropFirst(8).allSatisfy(\.isHexDigit)
+    }
+
+    static func ids(in text: String) -> [String] {
+        text.split(whereSeparator: \.isNewline).map { $0.trimmingCharacters(in: .whitespaces) }.filter { !$0.isEmpty }
+    }
+
+    private func isListed(_ id: String) async -> Bool? {
+        guard let listed = try? await cli.run(["ls", "-a", "-q"], timeout: 10), listed.status == 0 else { return nil }
+        return Self.ids(in: listed.text).contains(id)
+    }
+
+    private func waitUntilGone(_ id: String, for seconds: Double) async -> Bool {
+        let deadline = Date().addingTimeInterval(seconds)
+        repeat {
+            if await isListed(id) == false { return true }
+            try? await Task.sleep(for: .milliseconds(100))
+        } while Date() < deadline
+        return false
+    }
+
+    private func cleanUp(_ id: String) {
+        unlink(runDirectory.appendingPathComponent("\(id).sock").path)
+        try? FileManager.default.removeItem(at: secretsDirectory(id))
+    }
+
+    private func secretsDirectory(_ id: String) -> URL {
+        runDirectory.appendingPathComponent("\(id).secrets", isDirectory: true)
+    }
+
+    // MARK: Setup actions
+
+    public func fix(_ issue: SetupIssue, output: @escaping @Sendable (String) -> Void) async throws {
+        switch issue {
+        case .serviceStopped, .kernelMissing:
+            let result = try await cli.run(["system", "start", "--enable-kernel-install"], timeout: 120, line: output)
+            guard result.status == 0 else { throw VMBackendError.failed("Couldn't start the service: \(result.lastLine)") }
+        case .imageMissing, .imageOutdated, .imageStale:
+            guard let vmSources else { throw VMBackendError.failed("The VM sources are missing from the app.") }
+            let now = Date()
+            let day = DateFormatter()
+            day.dateFormat = "yyyy-MM-dd"
+            let args = [
+                "build", "--cpus", "4", "--memory", "4g", "--dns", "1.1.1.1", "--pull",
+                "--build-arg", "APT_REFRESH=\(day.string(from: now))",
+                "--build-arg", "BUILD_DATE=\(ISO8601DateFormatter().string(from: now))",
+                "-f", vmSources.appendingPathComponent("Containerfile").path,
+                "-t", Self.image, vmSources.path,
+            ]
+            let result = try await cli.run(args, timeout: 15 * 60, line: output)
+            guard result.status == 0 else { throw VMBackendError.failed("The image build failed: \(result.lastLine)") }
+        case .cliMissing, .cliOutdated, .lowDisk:
+            throw VMBackendError.failed("There's no automatic fix for this.")
+        }
+    }
+
+    // MARK: Quit path
+
+    public func destroyBlocking(ids: [String], deadline: Duration) {
+        let end = Date().addingTimeInterval(TimeInterval(deadline.components.seconds))
+        // Wait for in-flight starts so a VM created at the last moment is destroyed too.
+        while startsInFlight.withLock({ $0 }) > 0, Date() < end { usleep(50_000) }
+        for id in ids {
+            let left = end.timeIntervalSinceNow
+            guard left > 0.5 else { return }
+            _ = try? cli.runBlocking(["delete", "--force", id], timeout: min(10, left))
+            cleanUp(id)
+        }
+    }
+}