Hi, We are running a VPN in a network extension using the NEPacketTunnelProvider APIs. When we re-configure the VPN adapter, such as on a roam, any active NSURL session task fails. Using a test app we've been seeing that TCP streaming using BSD sockets remains unaffected, but NSURL tasks get killed. This is what the NSURL task fails with. 021-04-05 17:18:31.202678-0400 TCPStreamer[17967:460702] Task 2B2AE33C-145B-46E5-96F0-15C1E21791B5.4006 finished with error [-1009] Error Domain=NSURLErrorDomain Code=-1009 "The Internet connection appears to be offline." UserInfo={​​​​​​​​_kCFStreamErrorCodeKey=50, NSUnderlyingError=0x1144d42a0 {​​​​​​​​Error Domain=kCFErrorDomainCFNetwork Code=-1009 "(null)" UserInfo={​​​​​​​​_kCFStreamErrorCodeKey=50, _kCFStreamErrorDomainKey=1}​​​​​​​​}​​​​​​​​, _NSURLErrorFailingURLSessionTaskErrorKey=LocalDownloadTask 2B2AE33C-145B-46E5-96F0-15C1E21791B5.4006, _NSURLErrorRelatedURLSessionTaskErrorKey=(   "LocalDownloadTask 2B2AE33C-145B-46E5-96F0-15C1E21791B5.4006" ), NSLocalizedDescription=The Internet connection appears to be offline., NSErrorFailingURLStringKey=URL, NSErrorFailingURLKey=URL, _kCFStreamErrorDomainKey=1}​​​​​​​​ 2021-04-05 17:18:31.202699-0400 TCPStreamer[17967:460702] The Internet connection appears to be offline. 2021-04-05 17:18:31.203880-0400 TCPStreamer[17967:460700] Connection 4007: received failure notification 2021-04-05 17:18:31.203897-0400 TCPStreamer[17967:460700] Connection 4007: failed to connect 1:50, reason -1 2021-04-05 17:18:31.203934-0400 TCPStreamer[17967:460700] Connection 4007: encountered error(1:50) 2021-04-05 17:18:31.204591-0400 TCPStreamer[17967:460700] Task C0AB5E94-0CD8-4FBD-804F-99CA7C75F4CC.4007 HTTP load failed, 0/0 bytes (error code: -1009 [1:50]) I am wondering if this is intended, or if this is something I should file a bug for. Running the same test on macOS does not produce this issue. Thanks

Hi, I'm writing a test app to send ICMPv6 packets. When I try to set the IPV6_DONTFRAG socket option, I get the error "operation not permitted" (EPERM). I am able to set the IPV6_DONTFRAG socket option on UDP sockets running IPv6. I am also able to set the IP_DONTFRAG socket option on ICMPv4 sockets. To summarize: UDP - IPV4, don't fragment option can be set UDP - IPV6, don't fragment option can be set ICMP - IPV4, don't fragment option can be set ICMP - IPV6, don't fragment option can't be set Is there a reason/workaround to get this to work? Another example, if the following command, I get that output language ~ % ping6 -D www.google.com ping6: IPV6_DONTFRAG: Operation not permitted If I run language sudo ping6 -D -s 1600 www.google.com the pings start to be sent, but according to a packet capture, they are fragmented. It seems like the socket option is not set/not followed? Thanks

Hi, If I am sending a UDP packet or ICMP packet through an unbound BSD socket, and I have the don't fragment bit enabled on those packets, is there a way to get the source IP address and interface MTU of whatever packet that was sent? I am specifically looking the case where the packet fails to leave the device I am using, due to the packet size being larger than the interface MTU. The way that this is indicated to me is the send operation returning EMSGSIZE. This is useful with PMTU discovery for example on a cellular interface, where the MTU can be around ~1430. Thank you.

Somewhere between iOS 16 and iOS 16.5, we've been noticing a new timer when starting up our PacketTunnelProvider. When we start the VPN session and if we take longer than 60 seconds to call the completion handler in (void)startTunnelWithOptions:(NSDictionary *)options completionHandler:(void (^)(NSError *))completionHandler We see that our VPN gets shutdown by the OS. 11:19:15.371532-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)]: Plugin NEVPNTunnelPlugin(com.netmotionwireless.Mobility[inactive]) initialized with Mach-O UUIDs ( "69923795-443E-3B0D-9D51-1DC84EB26A08" ) 11:19:15.372733-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)] in state NESMVPNSessionStateStarting: plugin NEVPNTunnelPlugin(com.netmotionwireless.Mobility[inactive]) started with PID 27315 error (null) 11:19:15.389348-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)] in state NESMVPNSessionStateStarting: plugin NEVPNTunnelPlugin(com.netmotionwireless.Mobility[inactive]) attached IPC with endpoint 0xd5a820210 . . . 11:20:15.290251-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)]: State timer (60 seconds) fired in state NESMVPNSessionStateStarting 11:20:15.290375-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)] in state NESMVPNSessionStateStarting: timed out 11:20:15.293574-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)]: Leaving state NESMVPNSessionStateStarting 11:20:15.293813-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)]: Entering state NESMVPNSessionStateStopping, timeout 20 seconds 11:20:15.294034-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)]: config request: pushing handler [(null)] (null) 11:20:15.294286-0700 nesessionmanager <NESMServer: 0xd5a904120>: Request to uninstall session: NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)] 11:20:15.294426-0700 nesessionmanager NESMVPNSession[Primary Tunnel:test:7A492A00-109B-4DC9-970F-563A7BBC65A6:(null)]: status changed to disconnecting We can see this with the log message State timer (60 seconds) fired in state NESMVPNSessionStateStarting Is there anything we can do to influence the length of this timer or change the state the VPN is in, other than calling the completion handler? Thanks

Hi, We run a PacketTunnelProvider VPN on our macbooks. When we updated the Mac to MacOS Sonoma, we see strange behavior with regards to the Wi-Fi menu in system settings. After we connect a VPN, the Wi-Fi switch gets blanked out, and it doesn't detect any other networks. This is in contrast to the wifi task-bar, where we can see it functioning normally. Wi-Fi and the VPN also function normally. Switching Wi-Fi networks does not work, and after, the Mac will not connect to any Wi-Fi networks until it restarts. After disconnecting the VPN, this problem remains. The only way to fix this is to restart the Mac. I've filed FB13205010. I'm wondering if anyone has seen anything like this, because it makes having a VPN unusable on this version of macOS. All of the following screenshots were taken at the same time.

I am running into an issue with running a PacketTunnelProvider as a system extension, and fast user switching. Since the VPN is running as root, it runs outside any user context, and is accessible to all users on the machine. I can see the following behavior: Login to User A Connect VPN Fast switch to user B Open Safari, observe that traffic is directed and sent out the VPN. (As noted by the ability to browse otherwise un-reachable websites). Is there a way to detect that a fast user switch is happening? And disconnect the VPN when user B takes control of the screen. I know NSWorkspace is not a daemon-safe API. So NSWorkspaceSessionDidResignActiveNotification and like will not be useful here. One alternative would be to use NSWorkspace notifications in the GUI and alert the system extension via XPC. But that wouldn't work in the case where there isn't a GUI, such as with ConnectOnDemand. Is there a daemon-safe API to use or some other alternative? Thanks

We are using Manged App Configurations to dynamically push values to our app. We eventually want these values to reach our Network Extension process (specifically PacketTunnelProvider). However, there's some problems here: MDM providers only allow us to send configurations to app, not the extension. There's not really a way for us to reach the app configuration from the extension (even if the extension and app are in the same app group), because the app config is placed in [NSUserDefaults standardUserDefaults] A workaround would then be for the app to monitor for any AppConfig changes using NSUserDefaultsDidChangeNotification, and then write the app config settings to a shared NSUserDefaults instance. But when the app is in the background (most of the time for network extension apps), those notifications don't fire. I've attempted to use KVO to notify on any changes such as below: [[NSUserDefaults standardUserDefaults] addObserver:self forKeyPath:@"com.apple.configuration.managed" options:NSKeyValueObservingOptionInitial | NSKeyValueObservingOptionNew context:NULL]; - (void)observeValueForKeyPath:(NSString *)keyPath ofObject:(id)object change:(NSDictionary<NSKeyValueChangeKey,id> *)change context:(void *)context { NSLog(@"%@", [change description]); } But I am not seeing any KVO notifications here, even when NSUserDefaultsDidChangeNotification fires. This would be a workaround, but if the app is not running (due to connect-on-demand) or some other reason, this still would not work. Is there any possible workarounds or things that we can do here? Any help would be appreciated. Thanks

Hi, I'm struggling to understand using Swift-C++ in the same project. I have an existing code-base that makes heavy use of Swift-Objective-C interoperability. We make use of swift classes in our project. When I enable swift-objective c interoperability I am running into numerous build errors in the generated bridging header. I'm trying to understand why these errors exist and what to do to get around them. I have a project that I've set up with some test code, and I'm running into an error here: public class Foo { let name: String public init(name: String) { self.name = name } } public class Bar { let name: String public init(name : String) { self.name = name; } public func getFoo() -> Foo { return Foo(name: self.name); } } In the header file: Unknown type name 'Foo' SWIFT_INLINE_THUNK Foo getFoo() SWIFT_SYMBOL("s:13ForestBuilder3BarC6getFooAA0E0CyF"); This error goes away if I use structs, but for the purposes of porting my codebase, I'd prefer to use classes. Do classes not play nice here? Or am I misunderstanding something. Thanks.

Hi, We are running a PacketTunnelProvider, and are testing the effects of Encrypted DNS (DoH or DoT). According to WWDC20-10047, "resolution within the VPN tunnel will use the VPN's DNS settings and not your system-wide settings". We've been testing by enabling encrypted DNS system-wide using a configuration profile. However, we can see that in System Preferences -&gt; Network, both the VPN configuration, and the encrypted DNS configuration are active and running. And that DNS is encrypted, interfering with the proper function of our VPN application. Is this expected behavior? How can the behavior of encrypted DNS in conjunction with a VPN be characterized?

Hi, We are currently working on porting our PacketTunnelProvider app extension to run as a system extension. Things are mostly working great, but we're now testing upgrades from the existing app extension to a system extension. We have an existing configuration that gets created and runs perfectly fine with the app extension. Then, when we go and upgrade to the system extension, and attempt to connect using the same existing configuration. We see this error in the nesessionmanager logs: 10:00:57.717694-0700 nesessionmanager Signature check failed: code failed to satisfy specified code requirement(s) error 10:00:57.717914-0700 nesessionmanager Rejecting agent com.agentBundleID.bundleID because it does not satisfy the code signature requirements error 10:00:57.717937-0700 nesessionmanager Failed to launch com.agentBundleID.bundleID If we create a new configuration profile in our upgraded app w/system extension it works fine. The problem only occurs with existing profiles. Our app doesn't even get any notification about this error, startVPNTunnelWithOptions:andReturnError: doesn't return any error that we can work with. My gut tells me this has to do with the ProviderDesignatedRequirement not being correct, but I really have no way to confirm this at all. The NETunnelProviderProtocol has no way to specify that in its API. Our providerBundleIdentifier was unchanged between the two extensions. Is there anything that we can do here? Or are we stuck re-creating the configuration profile after an upgrade?

Hi, One of our customers is seeing a crash in our Content Filter in our network system extension. We're kind of at a loss for the cause of this as only one specific person is running into this and we're not at all in the stacktrace, out of the hundreds of others deployed with our extension. It would be greatly appreciated if we could have any help in diagnosing this issue. Attached is the crash report, and below is the crashing stacktrace. If this crash log is not sufficient, I have many more from the customer that I can attatch here. crash.txt Thread 4 Crashed:: Dispatch queue: NEFilterExtensionProviderContext queue 0 libsystem_kernel.dylib 0x18cd4e600 __pthread_kill + 8 1 libsystem_pthread.dylib 0x18cd86f70 pthread_kill + 288 2 libsystem_c.dylib 0x18cc93908 abort + 128 3 libc++abi.dylib 0x18cd3d44c abort_message + 132 4 libc++abi.dylib 0x18cd2ba40 demangling_terminate_handler() + 348 5 libobjc.A.dylib 0x18c9d13e4 _objc_terminate() + 156 6 libc++abi.dylib 0x18cd3c710 std::__terminate(void (*)()) + 16 7 libc++abi.dylib 0x18cd3c6b4 std::terminate() + 108 8 libdispatch.dylib 0x18cbd466c _dispatch_client_callout + 40 9 libdispatch.dylib 0x18cbdbc60 _dispatch_lane_serial_drain + 744 10 libdispatch.dylib 0x18cbdc79c _dispatch_lane_invoke + 432 11 libdispatch.dylib 0x18cbe77e8 _dispatch_root_queue_drain_deferred_wlh + 288 12 libdispatch.dylib 0x18cbe7034 _dispatch_workloop_worker_thread + 540 13 libsystem_pthread.dylib 0x18cd833d8 _pthread_wqthread + 288 14 libsystem_pthread.dylib 0x18cd820f0 start_wqthread + 8

Hi, I run a PacketTunnelProvider embedded within a system extension. We have been having success using this; however we have problems with accessing certificates/private keys manually imported in the file-based keychain. As per this, we are explicitly targeting the file-based keychain. However when attempting to access the certificate and private key we get the following error: System error using certificate key from keychain: Error Domain=NSOSStatusErrorDomain Code=-25308 "CSSM Exception: -2147415840 CSSMERR_CSP_NO_USER_INTERACTION" (errKCInteractionNotAllowed / errSecInteractionNotAllowed: / Interaction is not allowed As per the online documentation, I would expect to be prompted for the access to the application: When an app attempts to access a keychain item for a particular purpose—like using a private key to sign a document—the system looks for an entry in the item’s ACL containing the operation. If there’s no entry that lists the operation, then the system denies access and it’s up to the calling app to try something else or to notify the user. If there is an entry that lists the operation, the system checks whether the calling app is among the entry’s trusted apps. If so, the system grants access. Otherwise, the system prompts the user for confirmation. The user may choose to Deny, Allow, or Always Allow the access. In the latter case, the system adds the app to the list of trusted apps for that entry, enabling the app to gain access in the future without prompting the user again But I do not see that prompt, and I only see the permission denied error in my program. I can work around this one of two ways Change the access control of the keychain item to Allow all applications to access this item. This is not preferable, as it essentially disables any ACLs for this item. Embed the certificate in a configuration profile that is pushed down to the device via MDM or something similar. This works at a larger scale, but if I'm trying to manually test out a certificate, I don't always want to have to set this up. Is there another way that I go about adding my application to the ACL of the keychain item? Thanks!

Hi, I am wondering if anyone knows anything about when TestFlight for macOS will officially be released? I was hoping it would be released with macOS 12, but that has not been the case. I know Apple updated the App Store Connect to be able to upload macOS builds to TestFlight, and the TestFlight beta was last updated in August. But I have not heard anything since. Thanks