Run garbage collection after each symlink and expose --gc in tart prune (#191)

* Properly calculate remotely-retrieved manifest digests

* Implement garbage collection when deleting cached OCI VMs

.cirrus.yml

@@ -10,7 +10,7 @@ task:
   name: Build
   only_if: $CIRRUS_TAG == ''
   macos_instance:
-    image: ghcr.io/cirruslabs/macos-monterey-xcode:14
+    image: ghcr.io/cirruslabs/macos-ventura-xcode:latest
   build_script: swift build --product tart
   sign_script: codesign --sign - --entitlements Resources/tart.entitlements --force .build/debug/tart
   binary_artifacts:
@@ -20,7 +20,7 @@ task:
   name: Release
   only_if: $CIRRUS_TAG != ''
   macos_instance:
-    image: ghcr.io/cirruslabs/macos-monterey-xcode:14
+    image: ghcr.io/cirruslabs/macos-ventura-xcode:latest
   env:
     GITHUB_TOKEN: ENCRYPTED[!98ace8259c6024da912c14d5a3c5c6aac186890a8d4819fad78f3e0c41a4e0cd3a2537dd6e91493952fb056fa434be7c!]
     GORELEASER_KEY: ENCRYPTED[!9b80b6ef684ceaf40edd4c7af93014ee156c8aba7e6e5795f41c482729887b5c31f36b651491d790f1f668670888d9fd!]

.goreleaser.yml

@@ -32,7 +32,7 @@ brews:
     tap:
       owner: cirruslabs
       name: homebrew-cli
-    caveats: See the Github repository for more information
+    caveats: See the GitHub repository for more information
     homepage: https://github.com/cirruslabs/tart
     description: Run macOS VMs on Apple Silicon
     skip_upload: auto
@@ -40,3 +40,9 @@ brews:
       - "cirruslabs/cli/softnet"
     custom_block: |
       depends_on :macos => :monterey
+      
+      on_macos do
+        unless Hardware::CPU.arm?
+          odie "Tart only works on Apple Silicon!"
+        end
+      end

Package.resolved

@@ -5,8 +5,8 @@
       "kind" : "remoteSourceControl",
       "location" : "https://github.com/swift-server/async-http-client",
       "state" : {
-        "revision" : "24425989dadab6d6e4167174791a23d4e2a6d0c3",
-        "version" : "1.10.0"
+        "revision" : "df87a860fdc41a595d5ca67f74cde9adbccc099a",
+        "version" : "1.11.4"
       }
     },
     {
@@ -18,6 +18,15 @@
         "revision" : "772883073d044bc754d401cabb6574624eb3778f"
       }
     },
+    {
+      "identity" : "swift-algorithms",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/apple/swift-algorithms",
+      "state" : {
+        "revision" : "b14b7f4c528c942f121c8b860b9410b2bf57825e",
+        "version" : "1.0.0"
+      }
+    },
     {
       "identity" : "swift-argument-parser",
       "kind" : "remoteSourceControl",
@@ -27,6 +36,15 @@
         "version" : "1.1.2"
       }
     },
+    {
+      "identity" : "swift-atomics",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/apple/swift-atomics.git",
+      "state" : {
+        "revision" : "919eb1d83e02121cdb434c7bfc1f0c66ef17febe",
+        "version" : "1.0.2"
+      }
+    },
     {
       "identity" : "swift-case-paths",
       "kind" : "remoteSourceControl",
@@ -68,8 +86,8 @@
       "kind" : "remoteSourceControl",
       "location" : "https://github.com/apple/swift-nio-http2.git",
       "state" : {
-        "revision" : "72bcaf607b40d7c51044f65b0f5ed8581a911832",
-        "version" : "1.21.0"
+        "revision" : "108ac15087ea9b79abb6f6742699cf31de0e8772",
+        "version" : "1.22.0"
       }
     },
     {
@@ -90,6 +108,15 @@
         "version" : "1.12.0"
       }
     },
+    {
+      "identity" : "swift-numerics",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/apple/swift-numerics",
+      "state" : {
+        "revision" : "0a5bc04095a675662cf24757cc0640aa2204253b",
+        "version" : "1.0.2"
+      }
+    },
     {
       "identity" : "swift-parsing",
       "kind" : "remoteSourceControl",
@@ -99,6 +126,15 @@
         "version" : "0.9.2"
       }
     },
+    {
+      "identity" : "swiftdate",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/malcommac/SwiftDate",
+      "state" : {
+        "revision" : "6190d0cefff3013e77ed567e6b074f324e5c5bf5",
+        "version" : "6.3.1"
+      }
+    },
     {
       "identity" : "xctest-dynamic-overlay",
       "kind" : "remoteSourceControl",

Package.swift

@@ -1,4 +1,4 @@
-// swift-tools-version:5.6
+// swift-tools-version:5.7
 
 import PackageDescription
 let package = Package(
@@ -13,14 +13,18 @@ let package = Package(
     .package(url: "https://github.com/apple/swift-argument-parser", from: "1.1.2"),
     .package(url: "https://github.com/mhdhejazi/Dynamic", branch: "master"),
     .package(url: "https://github.com/pointfreeco/swift-parsing", from: "0.9.2"),
-    .package(url: "https://github.com/swift-server/async-http-client", from: "1.10.0"),
+    .package(url: "https://github.com/swift-server/async-http-client", from: "1.11.4"),
+    .package(url: "https://github.com/apple/swift-algorithms", from: "1.0.0"),
+    .package(url: "https://github.com/malcommac/SwiftDate", from: "6.3.1")
   ],
   targets: [
     .executableTarget(name: "tart", dependencies: [
+      .product(name: "Algorithms", package: "swift-algorithms"),
       .product(name: "ArgumentParser", package: "swift-argument-parser"),
       .product(name: "AsyncHTTPClient", package: "async-http-client"),
       .product(name: "Dynamic", package: "Dynamic"),
       .product(name: "Parsing", package: "swift-parsing"),
+      .product(name: "SwiftDate", package: "SwiftDate"),
     ]),
     .testTarget(name: "TartTests", dependencies: ["tart"])
   ]

Sources/tart/Commands/Clone.swift

@@ -11,6 +11,9 @@ struct Clone: AsyncParsableCommand {
   @Argument(help: "new VM name")
   var newName: String
 
+  @Flag(help: "connect to the OCI registry via insecure HTTP protocol")
+  var insecure: Bool = false
+
   func validate() throws {
     if newName.contains("/") {
       throw ValidationError("<new-name> should be a local name")
@@ -24,7 +27,7 @@ struct Clone: AsyncParsableCommand {
 
       if let remoteName = try? RemoteName(sourceName), !ociStorage.exists(remoteName) {
         // Pull the VM in case it's OCI-based and doesn't exist locally yet
-        let registry = try Registry(host: remoteName.host, namespace: remoteName.namespace)
+        let registry = try Registry(host: remoteName.host, namespace: remoteName.namespace, insecure: insecure)
         try await ociStorage.pull(remoteName, registry: registry)
       }
 

Sources/tart/Commands/IP.swift

@@ -16,14 +16,21 @@ struct IP: AsyncParsableCommand {
     do {
       let vmDir = try VMStorageLocal().open(name)
       let vmConfig = try VMConfig.init(fromURL: vmDir.configURL)
+      let vmMACAddress = MACAddress(fromString: vmConfig.macAddress.string)!
 
-      guard let ip = try await IP.resolveIP(vmConfig, secondsToWait: wait) else {
+      guard let ipViaDHCP = try await IP.resolveIP(vmMACAddress, secondsToWait: wait) else {
         print("no IP address found, is your VM running?")
 
         Foundation.exit(1)
       }
 
-      print(ip)
+      if let ipViaARP = try ARPCache.ResolveMACAddress(macAddress: vmMACAddress), ipViaARP != ipViaDHCP {
+        fputs("WARNING: DHCP lease and ARP cache entries for MAC address \(vmMACAddress) differ: "
+          + "got \(ipViaDHCP) and \(ipViaARP) respectively, consider reporting this case to"
+          + " https://github.com/cirruslabs/tart/issues/172\n", stderr)
+      }
+
+      print(ipViaDHCP)
 
       Foundation.exit(0)
     } catch {
@@ -33,12 +40,11 @@ struct IP: AsyncParsableCommand {
     }
   }
 
-  static public func resolveIP(_ config: VMConfig, secondsToWait: UInt16) async throws -> IPv4Address? {
+  static public func resolveIP(_ vmMACAddress: MACAddress, secondsToWait: UInt16) async throws -> IPv4Address? {
     let waitUntil = Calendar.current.date(byAdding: .second, value: Int(secondsToWait), to: Date.now)!
-    let vmMacAddress = MACAddress(fromString: config.macAddress.string)!
 
     repeat {
-      if let ip = try Leases().resolveMACAddress(macAddress: vmMacAddress) {
+      if let ip = try Leases().resolveMACAddress(macAddress: vmMACAddress) {
         return ip
       }
 

Sources/tart/Commands/List.swift

@@ -10,7 +10,7 @@ struct List: AsyncParsableCommand {
       print("Source\tName")
 
       displayTable("local", try VMStorageLocal().list())
-      displayTable("oci", try VMStorageOCI().list())
+      displayTable("oci", try VMStorageOCI().list().map { (name, vmDir, _) in (name, vmDir) })
 
       Foundation.exit(0)
     } catch {

Sources/tart/Commands/Login.swift

@@ -14,6 +14,9 @@ struct Login: AsyncParsableCommand {
   @Flag(help: "password-stdin")
   var passwordStdin: Bool = false
 
+  @Flag(help: "connect to the OCI registry via insecure HTTP protocol")
+  var insecure: Bool = false
+
   func validate() throws {
     let usernameProvided = username != nil
     let passwordProvided = passwordStdin
@@ -30,7 +33,9 @@ struct Login: AsyncParsableCommand {
 
       if let username = username {
         user = username
-        password = readLine()!
+
+        let passwordData = FileHandle.standardInput.readDataToEndOfFile()
+        password = String(decoding: passwordData, as: UTF8.self)
       } else {
         (user, password) = try StdinCredentials.retrieve()
       }
@@ -39,7 +44,8 @@ struct Login: AsyncParsableCommand {
       ])
 
       do {
-        let registry = try Registry(host: host, namespace: "", credentialsProvider: credentialsProvider)
+        let registry = try Registry(host: host, namespace: "", insecure: insecure,
+          credentialsProvider: credentialsProvider)
         try await registry.ping()
       } catch {
         print("invalid credentials: \(error)")

Sources/tart/Commands/Prune.swift

@@ -0,0 +1,108 @@
+import ArgumentParser
+import Dispatch
+import SwiftUI
+import SwiftDate
+
+struct Prune: AsyncParsableCommand {
+  static var configuration = CommandConfiguration(abstract: "Prune OCI and IPSW caches")
+
+  @Option(help: ArgumentHelp("Remove cache entries last accessed more than n days ago",
+    discussion: "For example, --older-than=7 will remove entries that weren't accessed by Tart in the last 7 days.",
+    valueName: "n"))
+  var olderThan: UInt?
+
+  @Option(help: ArgumentHelp("Remove least recently used cache entries that do not fit the specified cache size budget n, expressed in gigabytes",
+    discussion: "For example, --cache-budget=50 will effectively shrink all caches to a total size of 50 gigabytes.",
+    valueName: "n"))
+  var cacheBudget: UInt?
+
+  @Flag(help: .hidden)
+  var gc: Bool = false
+
+  func validate() throws {
+    if olderThan == nil && cacheBudget == nil {
+      throw ValidationError("at least one criteria must be specified")
+    }
+  }
+
+  func run() async throws {
+    do {
+      if gc {
+        try VMStorageOCI().gc()
+      }
+
+      // Clean up cache entries based on last accessed date
+      if let olderThan = olderThan {
+        let olderThanInterval = Int(exactly: olderThan)!.days.timeInterval
+        let olderThanDate = Date().addingTimeInterval(olderThanInterval)
+
+        try Prune.pruneOlderThan(olderThanDate: olderThanDate)
+      }
+
+      // Clean up cache entries based on imposed cache size limit and entry's last accessed date
+      if let cacheBudget = cacheBudget {
+        try Prune.pruneCacheBudget(cacheBudgetBytes: UInt64(cacheBudget) * 1024 * 1024 * 1024)
+      }
+
+      Foundation.exit(0)
+    } catch {
+      print(error)
+
+      Foundation.exit(1)
+    }
+  }
+
+  static func pruneOlderThan(olderThanDate: Date) throws {
+    let prunableStorages: [PrunableStorage] = [VMStorageOCI(), try IPSWCache()]
+    let prunables: [Prunable] = try prunableStorages.flatMap { try $0.prunables() }
+
+    try prunables.filter { try $0.accessDate() <= olderThanDate }.forEach { try $0.delete() }
+  }
+
+  static func pruneCacheBudget(cacheBudgetBytes: UInt64) throws {
+    let prunableStorages: [PrunableStorage] = [VMStorageOCI(), try IPSWCache()]
+    let prunables: [Prunable] = try prunableStorages
+            .flatMap { try $0.prunables() }
+            .sorted { try $0.accessDate() < $1.accessDate() }
+
+    let cacheUsedBytes = try prunables.map { try $0.sizeBytes() }.reduce(0, +)
+    var cacheReclaimedBytes: Int = 0
+
+    var it = prunables.makeIterator()
+
+    while (cacheUsedBytes - cacheReclaimedBytes) > cacheBudgetBytes {
+      guard let prunable = it.next() else {
+        break
+      }
+
+      cacheReclaimedBytes -= try prunable.sizeBytes()
+      try prunable.delete()
+    }
+  }
+
+  static func pruneReclaim(reclaimBytes: UInt64) throws {
+    let prunableStorages: [PrunableStorage] = [VMStorageOCI(), try IPSWCache()]
+    let prunables: [Prunable] = try prunableStorages
+            .flatMap { try $0.prunables() }
+            .sorted { try $0.accessDate() < $1.accessDate() }
+
+    // Does it even make sense to start?
+    let cacheUsedBytes = try prunables.map { try $0.sizeBytes() }.reduce(0, +)
+    if cacheUsedBytes < reclaimBytes {
+      return
+    }
+
+    var cacheReclaimedBytes: Int = 0
+
+    var it = prunables.makeIterator()
+
+    while cacheReclaimedBytes <= reclaimBytes {
+      guard let prunable = it.next() else {
+        break
+      }
+
+      cacheReclaimedBytes -= try prunable.sizeBytes()
+      try prunable.delete()
+    }
+  }
+}

Sources/tart/Commands/Pull.swift

@@ -8,6 +8,9 @@ struct Pull: AsyncParsableCommand {
   @Argument(help: "remote VM name")
   var remoteName: String
 
+  @Flag(help: "connect to the OCI registry via insecure HTTP protocol")
+  var insecure: Bool = false
+
   func run() async throws {
     do {
       // Be more liberal when accepting local image as argument,
@@ -19,7 +22,7 @@ struct Pull: AsyncParsableCommand {
       }
 
       let remoteName = try RemoteName(remoteName)
-      let registry = try Registry(host: remoteName.host, namespace: remoteName.namespace)
+      let registry = try Registry(host: remoteName.host, namespace: remoteName.namespace, insecure: insecure)
 
       defaultLogger.appendNewLine("pulling \(remoteName)...")
 

Sources/tart/Commands/Push.swift

@@ -12,6 +12,17 @@ struct Push: AsyncParsableCommand {
   @Argument(help: "remote VM name(s)")
   var remoteNames: [String]
 
+  @Flag(help: "connect to the OCI registry via insecure HTTP protocol")
+  var insecure: Bool = false
+
+  @Option(help: ArgumentHelp("chunk size in MB if registry supports chunked uploads",
+    discussion: """
+                By default monolithic method is used for uploading blobs to the registry but some registries support a more efficient chunked method.
+                For example, AWS Elastic Container Registry supports only chunks larger than 5MB but GitHub Container Registry supports only chunks smaller than 4MB. Google Container Registry on the other hand doesn't support chunked uploads at all.
+                Please refer to the documentation of your particular registry in order to see if this option is suitable for you and what's the recommended chunk size.
+                """))
+  var chunkSize: Int = 0
+
   @Flag(help: ArgumentHelp("cache pushed images locally",
           discussion: "Increases disk usage, but saves time if you're going to pull the pushed images later."))
   var populateCache: Bool = false
@@ -37,12 +48,17 @@ struct Push: AsyncParsableCommand {
 
       // Push VM
       for (registryIdentifier, remoteNamesForRegistry) in registryGroups {
-        let registry = try Registry(host: registryIdentifier.host, namespace: registryIdentifier.namespace)
+        let registry = try Registry(host: registryIdentifier.host, namespace: registryIdentifier.namespace,
+          insecure: insecure)
 
         defaultLogger.appendNewLine("pushing \(localName) to "
           + "\(registryIdentifier.host)/\(registryIdentifier.namespace)\(remoteNamesForRegistry.referenceNames())...")
 
-        let pushedRemoteName = try await localVMDir.pushToRegistry(registry: registry, references: remoteNamesForRegistry.map{ $0.reference.value })
+        let pushedRemoteName = try await localVMDir.pushToRegistry(
+          registry: registry, 
+          references: remoteNamesForRegistry.map{ $0.reference.value }, 
+          chunkSizeMb: chunkSize
+        )
 
         // Populate the local cache (if requested)
         if populateCache {

Sources/tart/IPSWCache.swift

@@ -0,0 +1,20 @@
+import Foundation
+import Virtualization
+
+class IPSWCache: PrunableStorage {
+  let baseURL: URL
+
+  init() throws {
+    baseURL = Config().tartCacheDir.appendingPathComponent("IPSWs", isDirectory: true)
+    try FileManager.default.createDirectory(at: baseURL, withIntermediateDirectories: true)
+  }
+
+  func locationFor(image: VZMacOSRestoreImage) -> URL {
+    baseURL.appendingPathComponent("\(image.buildVersion).ipsw", isDirectory: false)
+  }
+
+  func prunables() throws -> [Prunable] {
+    try FileManager.default.contentsOfDirectory(at: baseURL, includingPropertiesForKeys: nil)
+            .filter { $0.lastPathComponent.hasSuffix(".ipsw")}
+  }
+}

Sources/tart/MACAddressResolver/ARPCache.swift

@@ -0,0 +1,119 @@
+import Foundation
+import Network
+import Virtualization
+
+struct ARPCommandFailedError: Error, CustomStringConvertible {
+    var terminationReason: Process.TerminationReason
+    var terminationStatus: Int32
+
+    var description: String {
+        var reason: String
+
+        switch terminationReason {
+        case .exit:
+            reason = "exit code \(terminationStatus)"
+        case .uncaughtSignal:
+            reason = "uncaught signal"
+        default:
+            reason = "unknown reason"
+        }
+
+        return "arp command failed: \(reason)"
+    }
+}
+
+struct ARPCommandYieldedInvalidOutputError: Error, CustomStringConvertible {
+    var explanation: String
+
+    var description: String {
+        "arp command yielded invalid output: \(explanation)"
+    }
+}
+
+struct ARPCacheInternalError: Error, CustomStringConvertible {
+    var explanation: String
+
+    var description: String {
+        "ARPCache internal error: \(explanation)"
+    }
+}
+
+struct ARPCache {
+    static func ResolveMACAddress(macAddress: MACAddress, bridgeOnly: Bool = true) throws -> IPv4Address? {
+        let process = Process.init()
+        process.executableURL = URL.init(fileURLWithPath: "/usr/sbin/arp")
+        process.arguments = ["-an"]
+
+        let pipe = Pipe()
+        process.standardOutput = pipe
+        process.standardError = pipe
+        process.standardInput = FileHandle.nullDevice
+
+        try process.run()
+        process.waitUntilExit()
+
+        if !(process.terminationReason == .exit && process.terminationStatus == 0) {
+            throw ARPCommandFailedError(
+                    terminationReason: process.terminationReason,
+                    terminationStatus: process.terminationStatus)
+        }
+
+        guard let rawLines = try pipe.fileHandleForReading.readToEnd() else {
+            throw ARPCommandYieldedInvalidOutputError(explanation: "empty output")
+        }
+        let lines = String(decoding: rawLines, as: UTF8.self)
+                .trimmingCharacters(in: .whitespacesAndNewlines)
+                .components(separatedBy: "\n")
+
+        // Based on https://opensource.apple.com/source/network_cmds/network_cmds-606.40.2/arp.tproj/arp.c.auto.html
+        let regex = try NSRegularExpression(pattern: #"^.* \((?<ip>.*)\) at (?<mac>.*) on (?<interface>.*) .*$"#)
+
+        for line in lines {
+            let nsLineRange = NSRange(line.startIndex..<line.endIndex, in: line)
+
+            guard let match = regex.firstMatch(in: line, range: nsLineRange) else {
+                throw ARPCommandYieldedInvalidOutputError(explanation: "unparseable entry \"\(line)\"")
+            }
+
+            let rawIP = try match.getCaptureGroup(name: "ip", for: line)
+            guard let ip = IPv4Address(rawIP) else {
+                throw ARPCommandYieldedInvalidOutputError(explanation: "failed to parse IPv4 address \(rawIP)")
+            }
+
+            let rawMAC = try match.getCaptureGroup(name: "mac", for: line)
+            if rawMAC == "(incomplete)" {
+                continue
+            }
+            guard let mac = MACAddress(fromString: rawMAC) else {
+                throw ARPCommandYieldedInvalidOutputError(explanation: "failed to parse MAC address \(rawMAC)")
+            }
+
+            let interface = try match.getCaptureGroup(name: "interface", for: line)
+            if bridgeOnly && !interface.starts(with: "bridge") {
+                continue
+            }
+
+            if macAddress == mac {
+                return ip
+            }
+        }
+
+        return nil
+    }
+}
+
+extension NSTextCheckingResult {
+    func getCaptureGroup(name: String, for string: String) throws -> String {
+        let nsRange = self.range(withName: name)
+
+        if nsRange.location == NSNotFound {
+            throw ARPCacheInternalError(explanation: "attempted to retrieve non-existent named capture group \(name)")
+        }
+
+        guard let range = Range.init(nsRange, in: string) else {
+            throw ARPCacheInternalError(explanation: "failed to convert NSRange to Range")
+        }
+
+        return String(string[range])
+    }
+}

Sources/tart/MACAddressResolver/MACAddress.swift

@@ -16,6 +16,6 @@ struct MACAddress: Equatable, Hashable, CustomStringConvertible {
   }
 
   var description: String {
-    return String(format: "%02x:%02x:%02x:%02x:%02x:%02x", mac[0], mac[1], mac[2], mac[3], mac[4], mac[5])
+    String(format: "%02x:%02x:%02x:%02x:%02x:%02x", mac[0], mac[1], mac[2], mac[3], mac[4], mac[5])
   }
 }

Sources/tart/OCI/Registry.swift

@@ -2,6 +2,7 @@ import Foundation
 import NIOCore
 import NIOHTTP1
 import AsyncHTTPClient
+import Algorithms
 import NIOPosix
 
 enum RegistryError: Error {
@@ -84,7 +85,7 @@ class Registry {
   deinit {
     try! httpClient.syncShutdown()
   }
-  
+
   let baseURL: URL
   let namespace: String
   let credentialsProvider: CredentialsProvider
@@ -101,15 +102,13 @@ class Registry {
   }
 
   convenience init(
-          host: String,
-          namespace: String,
-          credentialsProvider: CredentialsProvider = KeychainCredentialsProvider()
+    host: String,
+    namespace: String,
+    insecure: Bool = false,
+    credentialsProvider: CredentialsProvider = KeychainCredentialsProvider()
   ) throws {
-    var baseURLComponents = URLComponents()
-
-    baseURLComponents.scheme = "https"
-    baseURLComponents.host = host
-    baseURLComponents.path = "/v2/"
+    let proto = insecure ? "http" : "https"
+    let baseURLComponents = URLComponents(string: proto + "://" + host + "/v2/")!
 
     try self.init(urlComponents: baseURLComponents, namespace: namespace, credentialsProvider: credentialsProvider)
   }
@@ -162,7 +161,7 @@ class Registry {
     return URLComponents(url: uploadLocation.absolutize(baseURL), resolvingAgainstBaseURL: true)!
   }
 
-  public func pushBlob(fromData: Data, chunkSize: Int = 5 * 1024 * 1024) async throws -> String {
+  public func pushBlob(fromData: Data, chunkSizeMb: Int = 0) async throws -> String {
     // Initiate a blob upload
     let postResponse = try await endpointRequest(.POST, "\(namespace)/blobs/uploads/",
       headers: ["Content-Length": "0"])
@@ -173,27 +172,55 @@ class Registry {
     }
 
     // Figure out where to upload the blob
-    let uploadLocation = try uploadLocationFromResponse(postResponse)
-
-    // Upload the blob
-    let headers = [
-      "Content-Length": "\(fromData.count)",
-      "Content-Type": "application/octet-stream",
-    ]
+    var uploadLocation = try uploadLocationFromResponse(postResponse)
 
     let digest = Digest.hash(fromData)
-    let parameters = [
-      "digest": digest,
-    ]
-
-    let putResponse = try await rawRequest(.PUT, uploadLocation, headers: headers, parameters: parameters,
-      body: fromData)
-    if putResponse.status != .created {
-      let body = try await postResponse.body.readTextResponse()
-      throw RegistryError.UnexpectedHTTPStatusCode(when: "pushing blob (PUT) to \(uploadLocation)",
-              code: putResponse.status.code, details: body ?? "")
+    
+    if chunkSizeMb == 0 {
+      // monolithic upload
+      let response = try await rawRequest(
+        .PUT,
+        uploadLocation,
+        headers: [
+          "Content-Type": "application/octet-stream",
+        ],
+        parameters: ["digest": digest],
+        body: fromData
+      )
+      if response.status != .created {
+        let body = try await response.body.readTextResponse()
+        throw RegistryError.UnexpectedHTTPStatusCode(when: "pushing blob (PUT) to \(uploadLocation)",
+          code: response.status.code, details: body ?? "")
+      }
+      return digest
     }
 
+    // chunked upload
+    var uploadedBytes = 0
+    let chunks = fromData.chunks(ofCount: chunkSizeMb == 0 ? fromData.count : chunkSizeMb * 1_000_000)
+    for (index, chunk) in chunks.enumerated() {
+      let lastChunk = index == (chunks.count - 1)
+      let response = try await rawRequest(
+        lastChunk ? .PUT : .PATCH, 
+        uploadLocation,
+        headers: [
+          "Content-Type": "application/octet-stream",
+          "Content-Range": "\(uploadedBytes)-\(uploadedBytes + chunk.count - 1)",
+        ],
+        parameters: lastChunk ? ["digest": digest] : [:],
+        body: chunk
+      )
+      let expectedStatus: HTTPResponseStatus = lastChunk ? .created : .accepted
+      if response.status != expectedStatus {
+        let body = try await response.body.readTextResponse()
+        throw RegistryError.UnexpectedHTTPStatusCode(when: "streaming blob to \(uploadLocation)",
+          code: response.status.code, details: body ?? "")
+      }
+      uploadedBytes += chunk.count
+      // Update location for the next chunk
+      uploadLocation = try uploadLocationFromResponse(response)
+    }
+    
     return digest
   }
 
@@ -248,6 +275,7 @@ class Registry {
       request.headers.add(name: key, value: value)
     }
     if body != nil {
+      request.headers.add(name: "Content-Length", value: "\(body!.count)")
       request.body = HTTPClientRequest.Body.bytes(body!)
     }
 
@@ -258,7 +286,7 @@ class Registry {
 
     var response = try await authAwareRequest(request: request)
 
-    if response.status == .unauthorized {
+    if doAuth && response.status == .unauthorized {
       try await auth(response: response)
       response = try await authAwareRequest(request: request)
     }

Sources/tart/Prunable.swift

@@ -0,0 +1,11 @@
+import Foundation
+
+protocol PrunableStorage {
+  func prunables() throws -> [Prunable]
+}
+
+protocol Prunable {
+  func delete() throws
+  func accessDate() throws -> Date
+  func sizeBytes() throws -> Int
+}

Sources/tart/Root.swift

@@ -16,6 +16,7 @@ struct Root: AsyncParsableCommand {
       IP.self,
       Pull.self,
       Push.self,
+      Prune.self,
       Delete.self,
     ])
 

Sources/tart/URL+AccessDate.swift

@@ -0,0 +1,25 @@
+import Foundation
+
+extension URL {
+  func accessDate() throws -> Date {
+    let attrs = try resourceValues(forKeys: [.contentAccessDateKey])
+    return attrs.contentAccessDate!
+  }
+
+  func updateAccessDate(_ accessDate: Date = Date()) throws {
+    let attrs = try resourceValues(forKeys: [.contentAccessDateKey])
+    let modificationDate = attrs.contentAccessDate!
+
+    let times = [accessDate.asTimeval(), modificationDate.asTimeval()]
+    let ret = utimes(path, times)
+    if ret != 0 {
+      throw RuntimeError("utimes(2) failed: \(ret.explanation())")
+    }
+  }
+}
+
+extension Date {
+  func asTimeval() -> timeval {
+    timeval(tv_sec: Int(timeIntervalSince1970), tv_usec: 0)
+  }
+}

Sources/tart/URL+Prunable.swift

@@ -0,0 +1,11 @@
+import Foundation
+
+extension URL: Prunable {
+  func delete() throws {
+    try FileManager.default.removeItem(at: self)
+  }
+
+  func sizeBytes() throws -> Int {
+    try resourceValues(forKeys: [.totalFileAllocatedSizeKey]).totalFileAllocatedSize!
+  }
+}

Sources/tart/VM+Recovery.swift

@@ -5,6 +5,7 @@ import Dynamic
 // Kudos to @saagarjha's VirtualApple for finding about _VZVirtualMachineStartOptions
 
 extension VZVirtualMachine {
+  @available(macOS 12, *)
   func start(_ recovery: Bool) async throws {
     if !recovery {
       // just use the regular API

Sources/tart/VM.swift

@@ -55,14 +55,11 @@ class VM: NSObject, VZVirtualMachineDelegate, ObservableObject {
       }
     }
 
-
-    let ipswCacheFolder = Config().tartCacheDir.appendingPathComponent("IPSWs", isDirectory: true)
-    try FileManager.default.createDirectory(at: ipswCacheFolder, withIntermediateDirectories: true)
-
-    let expectedIPSWLocation = ipswCacheFolder.appendingPathComponent("\(image.buildVersion).ipsw", isDirectory: false)
+    let expectedIPSWLocation = try IPSWCache().locationFor(image: image)
 
     if FileManager.default.fileExists(atPath: expectedIPSWLocation.path) {
       defaultLogger.appendNewLine("Using cached *.ipsw file...")
+      try expectedIPSWLocation.updateAccessDate()
       return expectedIPSWLocation
     }
 
@@ -167,7 +164,19 @@ class VM: NSObject, VZVirtualMachineDelegate, ObservableObject {
       try softnet.run()
     }
 
-    try await virtualMachine.start(recovery)
+    DispatchQueue.main.sync {
+      Task {
+        if #available(macOS 13, *) {
+          // new API introduced in Ventura
+          let startOptions = VZMacOSVirtualMachineStartOptions()
+          startOptions.startUpFromMacOSRecovery = recovery
+          try await virtualMachine.start(options: startOptions)
+        } else {
+          // use method that also available on Monterey
+          try await virtualMachine.start(recovery)
+        }
+      }
+    }
 
     await withTaskCancellationHandler(operation: {
       sema.wait()

Sources/tart/VMDirectory+OCI.swift

@@ -98,13 +98,14 @@ extension VMDirectory {
     try nvram.close()
   }
 
-  func pushToRegistry(registry: Registry, references: [String]) async throws -> RemoteName {
+  func pushToRegistry(registry: Registry, references: [String], chunkSizeMb: Int) async throws -> RemoteName {
     var layers = Array<OCIManifestLayer>()
 
     // Read VM's config and push it as blob
     let config = try VMConfig(fromURL: configURL)
     let configJSON = try JSONEncoder().encode(config)
-    let configDigest = try await registry.pushBlob(fromData: configJSON)
+    defaultLogger.appendNewLine("pushing config...")
+    let configDigest = try await registry.pushBlob(fromData: configJSON, chunkSizeMb: chunkSizeMb)
     layers.append(OCIManifestLayer(mediaType: Self.configMediaType, size: configJSON.count, digest: configDigest))
 
     // Progress
@@ -116,35 +117,37 @@ extension VMDirectory {
 
     // Read VM's compressed disk as chunks
     // and sequentially upload them as blobs
-    let disk = try FileHandle(forReadingFrom: diskURL)
-    var diskReadBytes: UInt64 = 0
-    let compressingFilter = try InputFilter<Data>(.compress, using: .lz4, bufferCapacity: Self.bufferSizeBytes) { _ in
-      let data = try disk.read(upToCount: Self.bufferSizeBytes)
-      diskReadBytes += UInt64(data?.count ?? 0)
+    let mappedDisk = try Data(contentsOf: diskURL, options: [.alwaysMapped])
+    let mappedDiskSize = mappedDisk.count
+    var mappedDiskReadOffset = 0
+    let compressingFilter = try InputFilter(.compress, using: .lz4, bufferCapacity: Self.bufferSizeBytes) { (length: Int) -> Data? in
+      let bytesRead = min(length, mappedDiskSize - mappedDiskReadOffset)
+      let data = mappedDisk.subdata(in: mappedDiskReadOffset ..< mappedDiskReadOffset + bytesRead)
+      mappedDiskReadOffset += bytesRead
 
-      progress.completedUnitCount += Int64(data?.count ?? 0)
+      progress.completedUnitCount = Int64(mappedDiskReadOffset)
 
       return data
     }
-    while let chunk = try compressingFilter.readData(ofLength: Self.layerLimitBytes) {
-      let chunkDigest = try await registry.pushBlob(fromData: chunk)
-      layers.append(OCIManifestLayer(mediaType: Self.diskMediaType, size: chunk.count, digest: chunkDigest))
+    while let compressedLayerData = try compressingFilter.readData(ofLength: Self.layerLimitBytes) {
+      let layerDigest = try await registry.pushBlob(fromData: compressedLayerData, chunkSizeMb: chunkSizeMb)
+      layers.append(OCIManifestLayer(mediaType: Self.diskMediaType, size: compressedLayerData.count, digest: layerDigest))
     }
 
     // Read VM's NVRAM and push it as blob
     defaultLogger.appendNewLine("pushing NVRAM...")
 
     let nvram = try FileHandle(forReadingFrom: nvramURL).readToEnd()!
-    let nvramDigest = try await registry.pushBlob(fromData: nvram)
+    let nvramDigest = try await registry.pushBlob(fromData: nvram, chunkSizeMb: chunkSizeMb)
     layers.append(OCIManifestLayer(mediaType: Self.nvramMediaType, size: nvram.count, digest: nvramDigest))
 
     // Craft a stub OCI config for Docker Hub compatibility
     let ociConfigJSON = try OCIConfig().toJSON()
-    let ociConfigDigest = try await registry.pushBlob(fromData: ociConfigJSON)
+    let ociConfigDigest = try await registry.pushBlob(fromData: ociConfigJSON, chunkSizeMb: chunkSizeMb)
     let manifest = OCIManifest(
             config: OCIManifestConfig(size: ociConfigJSON.count, digest: ociConfigDigest),
             layers: layers,
-            uncompressedDiskSize: diskReadBytes
+            uncompressedDiskSize: UInt64(mappedDiskReadOffset)
     )
 
     // Manifest

Sources/tart/VMDirectory.swift

@@ -7,7 +7,7 @@ struct UninitializedVMDirectoryError: Error {
 struct AlreadyInitializedVMDirectoryError: Error {
 }
 
-struct VMDirectory {
+struct VMDirectory: Prunable {
   var baseURL: URL
 
   var configURL: URL {
@@ -20,6 +20,10 @@ struct VMDirectory {
     baseURL.appendingPathComponent("nvram.bin")
   }
 
+  var explicitlyPulledMark: URL {
+    baseURL.appendingPathComponent(".explicitly-pulled")
+  }
+
   var name: String {
     baseURL.lastPathComponent
   }
@@ -77,4 +81,24 @@ struct VMDirectory {
     try diskFileHandle.truncate(atOffset: UInt64(sizeGB) * 1000 * 1000 * 1000)
     try diskFileHandle.close()
   }
+
+  func delete() throws {
+    try FileManager.default.removeItem(at: baseURL)
+  }
+
+  func accessDate() throws -> Date {
+    try baseURL.accessDate()
+  }
+
+  func sizeBytes() throws -> Int {
+    try configURL.sizeBytes() + diskURL.sizeBytes() + nvramURL.sizeBytes()
+  }
+
+  func markExplicitlyPulled() {
+    FileManager.default.createFile(atPath: explicitlyPulledMark.path, contents: nil)
+  }
+
+  func isExplicitlyPulled() -> Bool {
+    FileManager.default.fileExists(atPath: explicitlyPulledMark.path)
+  }
 }

Sources/tart/VMStorageOCI.swift

@@ -1,6 +1,6 @@
 import Foundation
 
-class VMStorageOCI {
+class VMStorageOCI: PrunableStorage {
   let baseURL = Config().tartCacheDir.appendingPathComponent("OCIs", isDirectory: true)
 
   private func vmURL(_ name: RemoteName) -> URL {
@@ -16,6 +16,8 @@ class VMStorageOCI {
 
     try vmDir.validate()
 
+    try vmDir.baseURL.updateAccessDate()
+
     return vmDir
   }
 
@@ -40,10 +42,48 @@ class VMStorageOCI {
 
   func delete(_ name: RemoteName) throws {
     try FileManager.default.removeItem(at: vmURL(name))
+    try gc()
   }
 
-  func list() throws -> [(String, VMDirectory)] {
-    var result: [(String, VMDirectory)] = Array()
+  func gc() throws {
+    var refCounts = Dictionary<URL, UInt>()
+
+    guard let enumerator = FileManager.default.enumerator(at: baseURL,
+            includingPropertiesForKeys: [.isSymbolicLinkKey]) else {
+      return
+    }
+
+    for case let foundURL as URL in enumerator {
+      let isSymlink = try foundURL.resourceValues(forKeys: [.isSymbolicLinkKey]).isSymbolicLink!
+
+      // Perform garbage collection for tag-based images
+      // with broken outgoing references
+      if isSymlink && foundURL == foundURL.resolvingSymlinksInPath() {
+        try FileManager.default.removeItem(at: foundURL)
+        continue
+      }
+
+      let vmDir = VMDirectory(baseURL: foundURL.resolvingSymlinksInPath())
+      if !vmDir.initialized {
+        continue
+      }
+
+      refCounts[vmDir.baseURL] = (refCounts[vmDir.baseURL] ?? 0) + (isSymlink ? 1 : 0)
+    }
+
+    // Perform garbage collection for digest-based images
+    // with no incoming references
+    for (baseURL, incRefCount) in refCounts {
+      let vmDir = VMDirectory(baseURL: baseURL)
+
+      if !vmDir.isExplicitlyPulled() && incRefCount == 0 {
+        try FileManager.default.removeItem(at: baseURL)
+      }
+    }
+  }
+
+  func list() throws -> [(String, VMDirectory, Bool)] {
+    var result: [(String, VMDirectory, Bool)] = Array()
 
     guard let enumerator = FileManager.default.enumerator(at: baseURL,
       includingPropertiesForKeys: [.isSymbolicLinkKey], options: [.producesRelativePathURLs]) else {
@@ -60,28 +100,46 @@ class VMStorageOCI {
       let parts = [foundURL.deletingLastPathComponent().relativePath, foundURL.lastPathComponent]
       var name: String
 
-      if try foundURL.resourceValues(forKeys: [.isSymbolicLinkKey]).isSymbolicLink! {
+      let isSymlink = try foundURL.resourceValues(forKeys: [.isSymbolicLinkKey]).isSymbolicLink!
+      if isSymlink {
         name = parts.joined(separator: ":")
       } else {
         name = parts.joined(separator: "@")
       }
 
-      result.append((name, vmDir))
+      result.append((name, vmDir, isSymlink))
     }
 
     return result
   }
 
+  func prunables() throws -> [Prunable] {
+    try list().filter { (_, _, isSymlink) in !isSymlink }.map { (_, vmDir, _) in vmDir }
+  }
+
   func pull(_ name: RemoteName, registry: Registry) async throws {
     defaultLogger.appendNewLine("pulling manifest...")
 
-    let (manifest, _) = try await registry.pullManifest(reference: name.reference.value)
+    let (manifest, manifestData) = try await registry.pullManifest(reference: name.reference.value)
 
-    var digestName = RemoteName(host: name.host, namespace: name.namespace,
-            reference: Reference(digest: try manifest.digest()))
+    let digestName = RemoteName(host: name.host, namespace: name.namespace,
+            reference: Reference(digest: Digest.hash(manifestData)))
 
     if !exists(digestName) {
       let tmpVMDir = try VMDirectory.temporary()
+
+      // Try to reclaim some cache space if we know the VM size in advance
+      if let uncompressedDiskSize = manifest.uncompressedDiskSize() {
+        let requiredCapacityBytes = UInt64(uncompressedDiskSize + 128 * 1024 * 1024)
+
+        let attrs = try tmpVMDir.baseURL.resourceValues(forKeys: [.volumeAvailableCapacityForImportantUsageKey])
+        let availableCapacityBytes = UInt64(attrs.volumeAvailableCapacityForImportantUsage!)
+
+        if availableCapacityBytes < requiredCapacityBytes {
+          try Prune.pruneReclaim(reclaimBytes: requiredCapacityBytes - availableCapacityBytes)
+        }
+      }
+
       try await withTaskCancellationHandler(operation: {
         try await tmpVMDir.pullFromRegistry(registry: registry, manifest: manifest)
         try move(digestName, from: tmpVMDir)
@@ -93,8 +151,12 @@ class VMStorageOCI {
     }
 
     if name != digestName {
-      // Overwrite the old symbolic link
+      // Create new or overwrite the old symbolic link
       try link(from: digestName, to: name)
+    } else {
+      // Ensure that images pulled by content digest
+      // are excluded from garbage collection
+      VMDirectory(baseURL: vmURL(name)).markExplicitlyPulled()
     }
   }
 
@@ -104,6 +166,8 @@ class VMStorageOCI {
     }
 
     try FileManager.default.createSymbolicLink(at: vmURL(to), withDestinationURL: vmURL(from))
+
+    try gc()
   }
 }
 

Sources/tart/VNC/ScreenSharingVNC.swift

@@ -10,7 +10,8 @@ class ScreenSharingVNC: VNC {
   }
 
   func waitForURL() async throws -> URL {
-    let ip = try await IP.resolveIP(vmConfig, secondsToWait: 60)
+    let vmMACAddress = MACAddress(fromString: vmConfig.macAddress.string)!
+    let ip = try await IP.resolveIP(vmMACAddress, secondsToWait: 60)
 
     if let ip = ip {
       return URL(string: "vnc://\(ip)")!

Tests/TartTests/RegistryTests.swift

@@ -42,13 +42,13 @@ final class RegistryTests: XCTestCase {
         XCTAssertEqual(pushedBlob, pulledBlob)
     }
 
-    func testPushPullBlobHuge() async throws {
+    func testPushPullBlobHugeInChunks() async throws {
         // Generate a large enough blob
         let fh = FileHandle(forReadingAtPath: "/dev/urandom")!
         let largeBlobToPush = try fh.read(upToCount: 768 * 1024 * 1024)!
 
         // Push it
-        let largeBlobDigest = try await registry.pushBlob(fromData: largeBlobToPush)
+        let largeBlobDigest = try await registry.pushBlob(fromData: largeBlobToPush, chunkSizeMb: 10)
 
         // Pull it
         var pulledLargeBlob = Data()

Tests/TartTests/URLAccessDateTests.swift

@@ -0,0 +1,22 @@
+import XCTest
+@testable import tart
+
+final class URLAccessDateTests: XCTestCase {
+  func testGetAndSetAccessTime() throws {
+    // Create a temporary file
+    let tmpDir = URL(fileURLWithPath: NSTemporaryDirectory(), isDirectory: true)
+    var tmpFile = tmpDir.appendingPathComponent(UUID().uuidString)
+    FileManager.default.createFile(atPath: tmpFile.path, contents: nil)
+
+    // Ensure it's access date is different than our desired access date
+    let arbitraryDate = Date.init(year: 2008, month: 09, day: 28, hour: 23, minute: 15)
+    XCTAssertNotEqual(arbitraryDate, try tmpFile.accessDate())
+
+    // Set our desired access date for a file
+    try tmpFile.updateAccessDate(arbitraryDate)
+
+    // Ensure the access date has changed to our value
+    tmpFile.removeCachedResourceValue(forKey: .contentAccessDateKey)
+    XCTAssertEqual(arbitraryDate, try tmpFile.accessDate())
+  }
+}