Summary On macOS 26.4.1 (25E253), iCloud Drive file uploads can enter a silent deadlock where every upload attempt fails at the transport layer. No error is surfaced anywhere — not in Finder, not in System Settings, not in the iCloud status panel. The upload queue simply stops. Other iCloud services (Photos, Mail, App Store) continue to work normally through the same networking infrastructure at the same time. Root Cause The issue is a stale HTTP/3 (QUIC) session cached in the user-level nsurlsessiond process's BackgroundConnectionPool. The deadlock cycle: cloudd requests an upload to the GCS storage endpoint nsurlsessiond provides the cached (broken) HTTP/3 session The TLS handshake succeeds, but the body upload dies mid-transfer (err=T, requestDuration=-1.000, responseHeaderBytes=0) cloudd retries with a new connectionUUID — but nsurlsessiond still routes through the same poisoned QUIC session This repeats indefinitely Killing cloudd alone does not help — nsurlsessiond retains the poisoned pool. Only killing both the user-level cloudd and nsurlsessiond clears the pool and forces a fresh protocol negotiation. The Smoking Gun After killing both daemons, the system falls back to HTTP/1.1 for the stuck uploads — and they complete instantly: Before Kill After Kill Protocol h3 (QUIC) http/1.1 (TCP) Largest upload Failed at partial offsets 26 MB in 1.6 seconds Server response 0 bytes 596 bytes (normal) Same endpoint, same files, same network interface (en5), same power state. The only change was the protocol negotiation after a fresh nsurlsessiond. Reproduction Reproduced 3 times on April 11, 2026 using a standardized set of 8 test files (8 bytes to 20 MB) in a non-shared iCloud Drive folder. Each run showed the identical pattern: Small files (<100 KB) squeeze through before the QUIC session stalls Larger files trigger the deadlock every time 5–6 retries with fresh connectionUUIDs, all failing over protocol=h3 After kill cloudd + nsurlsessiond: immediate flush via protocol=http/1.1 An automated evidence-collection script (collect_h3_deadlock_evidence.sh) captures paired before-kill / after-kill logs. Included in the Feedback report. Symptom Check (for others hitting this) /usr/bin/log show --predicate 'process == "cloudd"' --last 5m 2>&1 \ | grep "putContainer.*err=T.*requestDuration=-1.000.*protocol=h3" | wc -l Output > 0 = this deadlock. Output = 0 = different issue. Recovery (one-liner) kill $(ps -axo user,pid,command | awk -v u="$USER" \ '($1==u && /CloudKitDaemon.framework.*cloudd/ && !/--system/) \ || ($1==u && /\/usr\/libexec\/nsurlsessiond/ && !/--privileged/) \ {print $2}') Both daemons respawn within 1–2 seconds. Do not use killall nsurlsessiond — it would also kill the privileged system instance. What was ruled out Network connectivity (Photos uploaded 8 MB through the same pool simultaneously) iCloud account (metadata operations succeeding, only body uploads failing) File type/content (random data, correlation is with size, not type) Storage quota (1.65 TB free) CFNetworkHTTP3Enabled=false (key is ineffective in 26.4.1) Suggested fixes (from the Feedback report) CFNetwork: Invalidate the QUIC session after N consecutive requestDuration=-1.000 failures CloudKit/NSURLSession: Expose a pool invalidation API like [NSURLSession invalidatePoolEntryForEndpoint:] cloudd: Self-healing retry — create a fresh NSURLSession after M consecutive deadlock-signature failures Finder: At minimum, surface the stuck state to the user instead of failing silently Filed as FB22476701 — includes full reproduction timelines, request/connection UUIDs, sysdiagnose, and a 12-page investigation PDF with architecture diagrams and protocol comparison tables. If you're experiencing the same issue, please file a duplicate referencing FB22476701 — Apple prioritizes by duplicate count. System MacBook Air, macOS 26.4.1 (25E253) iCloud Drive with Desktop & Documents sync en0 (WLAN) + en5 (USB-LAN via Studio Display)

Hello, I have an app that is using iOS 26 Network Framework APIs. It is using QUIC, TLS 1.3 and Bonjour. For TLS I am using a PKCS#12 identity. All works well and as expected if the devices (iPhone with no cellular, iPhone with cellular, and iPad no cellular) are all on the same wifi network. If I turn off my router (ie no more wifi network) and leave on the wifi toggle on the iOS devices - only the non cellular iPhone and iPad are able to discovery and connect to each other. My iPhone with cellular is not able to. By sharing my logs with Cursor AI it was determined that the connection between the two problematic peers (iPad with no cellular and iPhone with cellular) never even makes it to the TLS step because I never see the logs where I print out the certs I compare. I tried doing "builder.requiredInterfaceType(.wifi)" but doing that blocked the two non cellular devices from working. I also tried "builder.prohibitedInterfaceTypes([.cellular])" but that also did not work. Is AWDL on it's way out? Should I focus my energy on Wi-Fi Aware? Regards, Captadoh

Is it somehow possible to get the transport layer (UDP and TCP) payload amounts for TLS or QUIC connections established via the Network framework? (From within the app itself that establishes the connections.) I am currently using the ntstat.h kernel socket calls, but I hope there is a simpler solution. With ntstat, I have not yet been able to observe a specific connection. I have to search for the connection I am looking in all (userspace) connections.

I want to know the right way/API/usage to use NWConnectionGroup to send both datagram and non-datagram stream. I am currently working on an P2P video streaming app. I want to leverage NWConnectionGroup over QUIC to handle both message channel (traditionally handled by a TCP connection) and media channel (traditionally handled by sth. over UDP) to transmit SRT packets back and forth. I created a NWConnectionGroup and it worked fine on non-datagram parts. The problems are with datagram part. I tried extracting a connection with datagram = true either from the group or from message, doesn't and in some cases it breaks other non-datagram connections. I currently send datagram directly using the NWConnectionGroup.send(content:completion). It kinda works but I keep seeing it canceled a lot of messages, which breaks SRT shortly after start. The warnings belong flooded my console. (Seems like want me to create a connection to transmit datagram, how?) nw_connection_create_with_connection [C1600] Original connection not yet connected nw_connection_group_create_connection_for_endpoint_and_parameters [G1] failed to create connection with parameters quic, local: fe80::439:68b4:6ec2:694%en0.60517, definite, attribution: developer, server I must use it in wrong way. What should I do to fix it?

I want to configure one aspect of my networking configuration (the QUIC keepalive interval). This only seems to be configurable via Network.framework’s nw_quic_set_keepalive_interval. Is there any way to apply this to a URLSession? Or do I need to implement the whole connection management myself using Network.framework?

Hello, I have an app that was using the iOS 18 Network Framework APIs. It used Peer to Peer, QUIC and Bonjour. It was all working as expected. I wanted to upgrade to the new iOS 26 Network Framework APIs (NetworkBrowser, NetworkListener, NetworkConnection...). I have things working (multiple devices can discover each other, connection to each other and send messages to each other) but my app crashes when I go to toggle of all the networking stuff. In the iOS 18 Network Framework API NWConnection had a .cancel() function I could use to tell the other side the connection was done. I dont see a cancel function for NetworkConnection. My question is - how do I properly close down a NetworkConnection and also properly tell the other side the connection is done.

Hello, I am studying the Building peer-to-peer apps codebase https://developer.apple.com/documentation/wifiaware/building-peer-to-peer-apps and am wondering why no connection is ever started? I searched the codebase and didn't find .start() be called once. Start function I'm referencing https://developer.apple.com/documentation/network/networkconnection/start() Are NetworkConnections started automatically? Note that I am using QUIC NetworkConnections (NetworkConnection) in what I'm trying to do.

Hello Apple Support Team, We are experiencing a performance issue with HTTP/3 in our iOS application during testing. Problem Description: Network requests using HTTP/3 are significantly slower than expected. This issue occurs on both Wi-Fi and 4G networks, with both IPv4 and IPv6. The same setup worked correctly in an earlier experiment. Key Observations: The slowdown disappears when the device uses: · A personal hotspot. · Network Link Conditioner (with no limitations applied). · Internet sharing from a MacBook via USB (where traffic was also inspected with Wireshark without issues). The problem is specific to HTTP/3 and does not occur with HTTP/2. The issue is reproducible on iOS 15, 18.7, and the latest iOS 26 beta. HTTP/3 is confirmed to be active (via assumeHttp3Capable and Alt-Svc header). Crucially, the same backend endpoint works with normal performance on Android devices and using curl with HTTP/3 support from the same network. I've checked the CFNetwork logs in the Console but haven't found any suspicious errors or obvious clues that explain the slowdown. We are using a standard URLSession with basic configuration. Attempted to collect qlog diagnostics by setting the QUIC_LOG_DIRECTORY=~/ tmp environment variable, but the logs were not generated. Question: What could cause HTTP/3 performance to improve only when the device is connected through a hotspot, unrestricted Network Link Conditioner, or USB-tethered connection? The fact that Android and curl work correctly points to an issue specific to the iOS network stack. Are there known conditions or policies (e.g., related to network interface handling, QoS, or specific packet processing) that could lead to this behavior? Additionally, why might the qlog environment variable fail to produce logs, and are there other ways to obtain detailed HTTP/3 diagnostic information from iOS? Any guidance on further diagnostic steps or specific system logs to examine would be greatly appreciated. Thank you for your assistance.

Hello, I have a peer to peer networking setup in my app that uses Network Framework with Bonjour and QUIC via NWBrowser, NWListener, NWConnection, and NWEndpoint and all works as expected. I watched the videos about the new iOS 26 Networking stuff (NetworkBrowser, NetworkListener, NetworkConnection) and wanted to try and migrate all my code to use the the new APIs (still use Bonjour and NOT use Wi-Fi Aware) but hit some issues. I was following how the Wi-Fi Aware example app was receiving messages for try await messageData in connection.messages { but when I got things setup with QUIC in a similar fashion I got the following compile error Requirement from conditional conformance of '(content: QUIC.ContentType, metadata: QUIC.Metadata)' to 'Copyable' Requirement from conditional conformance of '(content: QUIC.ContentType, metadata: QUIC.Metadata)' to 'Escapable' Requirement from conditional conformance of '(content: QUIC.ContentType, metadata: QUIC.Metadata)' to 'Copyable' Requirement from conditional conformance of '(content: QUIC.ContentType, metadata: QUIC.Metadata)' to 'Escapable' When I asked Cursor about what I was facing its response was as follows: "The connection.messages stream changed in the new Network APIs: it now yields typed (content, metadata) tuples. Iterating with for try await incoming in connection.messages asks the compiler to conform that tuple to Copyable/Escapable; for QUIC the tuple isn’t copyable, so you hit the conditional-conformance error." I am curious if you've been able to use the new iOS 26 network APIs with QUIC? Thank you, Captadoh

I am seeking assistance with how to properly handle / save / reuse NWConnections when it comes to the NWBrowser vs NWListener. Let me give some context surrounding why I am trying to do what I am. I am building an iOS app that has peer to peer functionality. The design is for a user (for our example the user is Bob) to have N number of devices that have my app installed on it. All these devices are near each other or on the same wifi network. As such I want all the devices to be able to discover each other and automatically connect to each other. For example if Bob had three devices (A, B, C) then A discovers B and C and has a connection to each, B discovers B and C and has a connection to each and finally C discovers A and B and has a connection to each. In the app there is a concept of a leader and a follower. A leader device issues commands to the follower devices. A follower device just waits for commands. For our example device A is the leader and devices B and C are followers. Any follower device can opt to become a leader. So if Bob taps the “become leader” button on device B - device B sends out a message to all the devices it’s connected to telling them it is becoming the new leader. Device B doesn’t need to do anything but device A needs to set itself as a follower. This detail is to show my need to have everyone connected to everyone. Please note that I am using .includePeerToPeer = true in my NWParameters. I am using http/3 and QUIC. I am using P12 identity for TLS1.3. I am successfully able to verify certs in sec_protocal_options_set_verify_block. I am able to establish connections - both from the NWBrowser and from NWListener. My issue is that it’s flaky. I found that I have to put a 3 second delay prior to establishing a connection to a peer found by the NWBrowser. I also opted to not save the incoming connection from NWListener. I only save the connection I created from the peer I found in NWBrowser. For this example there is Device X and Device Y. Device X discovers device Y and connects to it and saves the connection. Device Y discovers device X and connects to it and saves the connection. When things work they work great - I am able to send messages back and forth. Device X uses the saved connection to send a message to device Y and device Y uses the saved connection to send a message to device X. Now here come the questions. Do I save the connection I create from the peer I discovered from the NWBrowser? Do I save the connection I get from my NWListener via newConnectionHandler? And when I save a connection (be it from NWBrowser or NWListener) am I able to reuse it to send data over (ie “i am the new leader command”)? When my NWBrowser discovers a peer, should I be able to build a connection and connect to it immediately? I know if I save the connection I create from the peer I discover I am able to send messages with it. I know if I save the connection from NWListener - I am NOT able to send messages with it — but should I be able to? I have a deterministic algorithm for who makes a connection to who. Each device has an ID - it is a UUID I generate when the app loads - I store it in UserDefaults and the next time I try and fetch it so I’m not generating new UUIDs all the time. I set this deviceID as the name of the NWListener.Service I create. As a result the peer a NWBrowser discovers has the deviceID set as its name. Due to this the NWBrowser is able to determine if it should try and connect to the peer or if it should not because the discovered peer is going to try and connect to it. So the algorithm above would be great if I could save and use the connection from NWListener to send messages over.

Getting -10985 error from urlSession while attempting to make a connection. Not sure why this is happening if anyone is aware please help

Good morning, I have been playing with he new Networking framework released in beta, and i think its amazing how powerful and simple it is. However i have been tackling some issues with it, it seems that the NetworkListener does not allow us to configure a specific endpoint for any of the protocols, UDP, TCP (QUIC, TLS) Is this intended or just not missing features as of the Beta ? I figured out how to use bonjour to get a port (as i am brand new to using Networking on macOS and Swift) I get that the use of this is mainly as a client to connect to servers, but it would make more sense to have a high level abstraction of what already exist, wouldn't it be more intuitive to configure a NetworkEndpoint that contains either a Bonjour Service or an endpoint with configured port that we can then configure on the Listener, instead of doing .service(...) ?

I was excited about the new APIs added to Network.framework in iOS 26 that offer structure concurrency support out of the box and a more modern API design in general. However I have been unable to use them to create a device-to-device QUIC connection. The blocker I ran into is that NetworkListener's run method requires the network protocol to conform to OneToOneProtocol, whereas QUIC conforms to MultiplexProtocol. And there doesn't seem to be any way to accept an incoming MultiplexProtocol connection? Nor does it seem possible to turn a UDP connection into a QUIC connection using NetworkConnection.prependProtocols() as that also only works for network protocols conforming to OneToOneProtocol. I suspect this is an accidental omission in the API design (?), and already filed a Feedback (FB18620438). But maybe I am missing something and there is a workaround or a different way to listen for incoming QUIC connections using the new NetworkListener? QUIC.TLS has methods peerAuthenticationRequired(Bool) and peerAuthenticationOptional(Bool), which makes me think that peer to peer QUIC connections are intended to be supported? I would also love to see documentation for those methods. For example I wonder what exact effect peerAuthenticationRequired(false) and peerAuthenticationOptional(false) would have and how they differ.

I'm using NERelayManager to set Relay configuration which all works perfectly fine. I then do a curl with the included domain and while I see QUIC connection succeeds with relay server and H3 request goes to the server, the connection gets abruptly closed by the client with "Software caused connection abort". Console has this information: default 09:43:04.459517-0700 curl nw_flow_connected [C1.1.1 192.168.4.197:4433 in_progress socket-flow (satisfied (Path is satisfied), viable, interface: en0[802.11], ipv4, ipv6, dns, uses wifi)] Transport protocol connected (quic) default 09:43:04.459901-0700 curl [C1.1.1 192.168.4.197:4433 in_progress socket-flow (satisfied (Path is satisfied), viable, interface: en0[802.11], ipv4, ipv6, dns, uses wifi)] event: flow:finish_transport @0.131s default 09:43:04.460745-0700 curl nw_flow_connected [C1.1.1 192.168.4.197:4433 in_progress socket-flow (satisfied (Path is satisfied), viable, interface: en0[802.11], ipv4, ipv6, dns, uses wifi)] Joined protocol connected (http3) default 09:43:04.461049-0700 curl [C1.1.1 192.168.4.197:4433 in_progress socket-flow (satisfied (Path is satisfied), viable, interface: en0[802.11], ipv4, ipv6, dns, uses wifi)] event: flow:finish_transport @0.133s default 09:43:04.465115-0700 curl [C2 E47A3A0C-7275-4F6B-AEDF-59077ABAE34B 192.168.4.197:4433 quic, multipath service: 1, tls, definite, attribution: developer] cancel default 09:43:04.465238-0700 curl [C2 E47A3A0C-7275-4F6B-AEDF-59077ABAE34B 192.168.4.197:4433 quic, multipath service: 1, tls, definite, attribution: developer] cancelled [C2 FCB1CFD1-4BF9-4E37-810E-81265D141087 192.168.4.139:53898<->192.168.4.197:4433] Connected Path: satisfied (Path is satisfied), viable, interface: en0[802.11], ipv4, ipv6, dns, uses wifi Duration: 0.121s, QUIC @0.000s took 0.000s, TLS 1.3 took 0.111s bytes in/out: 2880/4322, packets in/out: 4/8, rtt: 0.074s, retransmitted bytes: 0, out-of-order bytes: 0 ecn packets sent/acked/marked/lost: 3/1/0/0 default 09:43:04.465975-0700 curl nw_flow_disconnected [C2 192.168.4.197:4433 cancelled multipath-socket-flow ((null))] Output protocol disconnected default 09:43:04.469189-0700 curl nw_endpoint_proxy_receive_report [C1.1 IPv4#124bdc4d:80 in_progress proxy (satisfied (Path is satisfied), interface: en0[802.11], ipv4, ipv6, dns, proxy, uses wifi)] Privacy proxy failed with error 53 ([C1.1.1] masque Proxy: http://192.168.4.197:4433) default 09:43:04.469289-0700 curl [C1.1.1 192.168.4.197:4433 failed socket-flow (satisfied (Path is satisfied), viable, interface: en0[802.11], ipv4, ipv6, dns, uses wifi)] event: flow:failed_connect @0.141s, error Software caused connection abort Relay server otherwise works fine with our QUIC MASQUE clients but not with built-in macOS MASQUE client. Anything I'm missing?