Add a VM-based natlab test that exercises the peer-relay feature
(feature/relayserver) end-to-end across three Tailscale nodes whose
network topology makes a direct A<->B UDP path impossible: both peers
are behind HardNAT (FreeBSD/pfSense-style endpoint-dependent NAT) with
no port-mapping services, while the relay node is behind One2OneNAT so
its STUN-discovered WAN endpoint is reachable from both peers. The
test enables the relay server via EditPrefs, then waits for an a->b
PingDisco whose PingResult.PeerRelay is set (proving magicsock chose
the peer-relay path, not DERP), and finally asserts that the relay's
DebugPeerRelaySessions LocalAPI reports the session.
The existing TestPeerRelayPing in tstest/integration runs three
tailscaled processes on the loopback interface with no NATs; this new
vmtest covers peer relay through real per-VM kernels and NATs.
To wire control-server capabilities into vmtest, also add a
PeerRelayGrants() EnvOption (sibling of AllOnline,
SameTailnetUser) that flips testcontrol.Server.PeerRelayGrants so the
wildcard packet filter grants tailcfg.PeerCapabilityRelay and
PeerCapabilityRelayTarget; without those caps magicsock won't consider
any peer a candidate relay.
Updates #13038
Change-Id: Ib3440b83ec442da0d3b89ffa48ceea9398ea9062
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
In a lot of places, we construct an error to End a step, then immediately log
it to the governing test as test fatal. Save ourselves a bit of boilerplate by
putting methods on Step for that.
There are a couple cases this doesn't cover, e.g., where we construct the Step
outside a subtest that wants to fail individually, but it helps enough to pay
for its lines.
Updates #13038
Change-Id: I71f9900942962de16609b6b198d3ba13d6958a5f
Signed-off-by: M. J. Fromberger <fromberger@tailscale.com>
Instead of having two entry points for running natlab tests, start
converting the connectivity tests to use the vmtest framework.
Grid and pair tests have yet to be moved over.
Updates #13038
Signed-off-by: Claus Lensbøl <claus@tailscale.com>
Add new clientmetric counters for establishing contact with peers while using
cached network map data. To do this, instrument the magicsock.Conn with a bit
to indicate whether its peer data came from a cached netmap. If so, there are
two conditions we will count as establishing connectivity to a peer:
- Receipt of a CallMeMaybe from a peer via disco.
- Establishing a valid endpoint address for a peer.
In vmtest, add Env.ClientMetrics to scrape metrics from the specified node.
Use this to check that counters were updated in caching tests.
Updates https://github.com/tailscale/projects/issues/13
Updates #12639
Change-Id: Ie8cf3244ac8af4f5bcfe4d0d944078da2ba08990
Signed-off-by: M. J. Fromberger <fromberger@tailscale.com>
macOS limits Unix socket paths to 104 bytes. The Go test TempDir
path (e.g. /var/folders/.../TestDirectConnection...679197086/001/)
easily exceeds that, causing "bind: invalid argument". Create a
short /tmp/vmtest* directory for all socket files (vnet, QMP,
dgram) so the paths stay well under the limit on every platform.
Updates #13038
Change-Id: I721d24561d1766aaa964692bc77f40a131aa9455
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
startCloudQEMU hardcoded -machine q35,accel=kvm and -cpu host,
which fails on any host without KVM (notably macOS). Replace
with a qemuAccelArgs helper that probes /dev/kvm and falls back
to QEMU's TCG software emulation, matching the pattern already
used by tstest/integration/nat. Also wire the helper into
startGokrazyQEMU so gokrazy VMs pick up KVM when available.
Updates #13038
Change-Id: I7745518db823279b1880957bb14ca2ffdaab4c50
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The natlab vmtest suite (tstest/natlab/vmtest) and the integration nat
tests are gated behind --run-vm-tests because they need KVM and are
slow. Until now nothing in CI exercised them apart from a single
canary TestEasyEasy run on every PR.
Add .github/workflows/natlab-test.yml that runs the full opt-in suite
on demand (workflow_dispatch), on PRs labeled "natlab", and on main
every 12 hours via cron. The workflow has two phases:
- "prepare" builds the gokrazy VM image, downloads the Ubuntu and
FreeBSD cloud images once via the new natlabprep tool, and emits
a dynamic JSON matrix of every TestX function it finds in the two
opt-in packages.
- "test" is a per-test matrix that depends on prepare. Each matrix
job restores the shared caches and runs a single test, so adding
a new TestFoo is automatically picked up on the next run without
any workflow edits.
Rename the existing natlab-integrationtest.yml to natlab-basic.yml
since it's the small smoke variant (just TestEasyEasy on every PR);
the new natlab-test.yml is the bigger suite. The job inside is
renamed to EasyEasy for the same reason.
Move the macOS arm64 host check from vmtest.Env.Start into
vmtest.Env.AddNode so a test that adds a vmtest.MacOS node skips
immediately on a non-macOS host, and add an explicit
skipIfNotMacOSArm64 helper at the top of the two macOS-only tests
so the platform requirement is obvious to readers.
Quiet the takeAgentConnOne miss log in tstest/natlab/vnet by default
(it was the overwhelming majority of bytes in CI logs, with no signal
in healthy runs) and replace it with a periodic "still waiting" line
that only fires after 10s, so a truly stuck agent connection still
surfaces.
Updates #13038
Change-Id: I4582098d8865200fd5a73a9b696942319ccf3bf0
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
When a peer is not able to connect to control after a restart and is
using a cached netmap, that nodes should be able to connect to another
peer in its tailnet (given that the home DERP of that peer has not
changed in the meantime).
Add test that starts two peers and connects them to a tailnet with
caching enabled. Then blackhole traffic to control from one peer and
restart it. Verify that the connection between the two ends up direct.
Adds facilities for expecting a certain path type between nodes.
Updates: #19597
Signed-off-by: Claus Lensbøl <claus@tailscale.com>
Re-exec the test binary as a thin wrapper that holds a pipe inherited
from the test. When the test goes away (any reason, including SIGKILL,
panic, or OOM), the kernel closes the pipe write end; the wrapper sees
EOF and SIGKILLs itself, taking QEMU and its children with it.
Updates #13038
Change-Id: Ib2151098193551396c1d7bb51b07da3bd6b2cfb4
Signed-off-by: Fernando Serboncini <fserb@tailscale.com>
Running all vmtests in tstest/natlab/vmtest locally was breaking later
tasks in the queue. The goroutine dump on timeout had goroutines hanging
around for 9 minutes, meaning that something was not getting cleaned up.
goroutine 262 [select, 9 minutes]:
gvisor.dev/gvisor/pkg/tcpip/adapters/gonet.commonRead({...})
Add a timeout of Now() to gonet TCP connections when the test ends
(inspired by ServeUnixConn()), and wait for them to shut down before
exiting the test.
Updates #13038
Signed-off-by: Claus Lensbøl <claus@tailscale.com>
The purpose of this package is to test the iOS dependency closure, but
it had drifted from the actual import list of the ipn-go-bridge package
in the corp repo (the Go side of the iOS / macOS app).
Update the imports to match ipn-go-bridge's GOOS=ios import list,
adding many missing packages including wgengine/netstack,
feature/{taildrop,syspolicy,condregister}, the util/syspolicy/*
subpackages, types/{key,lazy,logid,netmap}, tsd, safesocket,
util/{eventbus,must,set}, and several net/* and ipn/* packages.
Drop two now-stale BadDeps entries (for now!): database/sql/driver and
github.com/google/uuid are reached via wgengine/netstack ->
github.com/prometheus-community/pro-bing, which netstack imports on
darwin || ios for ICMP user-ping, so the iOS app already ships them.
But we should fix that later.
Updates #19633
Change-Id: Ic50779fdb195685a2e8ccd7c513eee91b0feeaf8
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For testing the loading of netmap cache from disk, the cache needs to
exist. The simple solution is to start two nodes and connect them to
control, with the netmap caching capability set. Then cut the connection
to control, restart the nodes, and ping between them.
This tests that we can start from a cache and get to running state, but
also that we are able to establish a connection between the nodes.
For now this is not testing how the nodes are able to talk to each other
(DERP vs direct).
Updates #19597
Signed-off-by: Claus Lensbøl <claus@tailscale.com>
The test was flaky under stress with "AddRawMapResponse N: node not
connected" failures. The root cause was in testcontrol's addDebugMessage:
it conflated "no streaming poll registered" with "wake-up channel buffer
momentarily full". The single-slot updatesCh is just a lossy wake-up
signal, but the streaming serveMap loop has fast paths
(takeRawMapMessage and the hasPendingRawMapMessage continue) that don't
drain it. A stale notification could remain buffered, causing the next
sendUpdate to fail even though msgToSend had been queued and the
streaming poll would still pick it up.
Detect the real failure case (no streaming poll) by checking
s.updates[nodeID] directly, and treat sendUpdate's buffer-full result as
benign — the message is in msgToSend, which is the source of truth.
Also plumb an optional *health.Tracker through tsp.ClientOpts to the
underlying ts2021.Client and supply one in the tests, eliminating the
"## WARNING: (non-fatal) nil health.Tracker (being strict in CI)" stack
dumps emitted by controlhttp.(*Dialer).forceNoise443 under CI.
Fixes#19583
Change-Id: Ib2334376585e8d6562f000a0b71dea0117acb0ff
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a vmtest that brings up two gokrazy nodes A and B behind two
One2OneNAT networks (so direct UDP works in both directions and any
slowness can't be blamed on NAT traversal), establishes a WireGuard
tunnel A → B with TSMP, then rotates B's disco key four times and
asserts that the data plane recovers in both directions after each
rotation. All pings are TSMP (the data-plane ping; disco pings would
not exercise the WireGuard tunnel itself).
The five pings:
1. A → B (initial; brings up the tunnel; 30s budget)
2. B → A after rotate (LocalAPI rotate-disco-key debug action)
3. A → B after rotate (LocalAPI)
4. B → A after restart (SIGKILL; gokrazy supervisor respawns)
5. A → B after restart (SIGKILL)
Each post-rotation ping gets a 15-second budget. Two unavoidable
multi-second waits dominate today:
- The rotate-then-a→b phase takes ~10s on main because of LazyWG.
After B's WantRunning bounce, B's wgengine resets its
sentActivityAt/recvActivityAt maps and trims A out of the
wireguard-go config as an "idle peer"; B only re-adds A on
inbound activity, by which point A's first few TSMP packets
have been silently dropped at B's tundev. The
bradfitz/rm_lazy_wg branch removes that trimming entirely
(verified locally: this phase drops to <100ms there).
- The restart phases take ~5s for wireguard-go's RekeyTimeout
handshake retry. After SIGKILL+respawn the first WG handshake
init from the restarted node sometimes goes into the void
(likely the brief peer-removed window in the receiver's
two-step maybeReconfigWireguardLocked reconfig during which
the peer is absent from wireguard-go), and wg-go's 5s+jitter
retransmit timer is the next opportunity to retry. That retry
succeeds and the staged TSMP packet flushes. Intrinsic to the
protocol's retransmit policy.
Once LazyWG is removed and the first-handshake-after-reconfig race
is fixed, the budget should drop to 5s.
Supporting changes:
ipn/ipnlocal: DebugRotateDiscoKey now toggles WantRunning off and
back on after rotating the disco key. magicsock.Conn.RotateDiscoKey
only resets local disco state; without also dropping wireguard-go
session keys, peers keep encrypting with their stale per-peer
session against us until their rekey timer fires (WireGuard has no
data-plane signaling to invalidate sessions). Bouncing WantRunning
runs the engine through Reconfig(empty) → authReconfig, which
drops every peer's WG session so the next packet either way
triggers a fresh handshake.
ipn/ipnlocal, ipn/localapi: add a debug-only "peer-disco-keys"
LocalAPI action ([LocalBackend.DebugPeerDiscoKeys]) that returns
a map[NodePublic]DiscoPublic from the current netmap. Tests reach
it via [local.Client.DebugResultJSON]. We do not surface disco
keys via [ipnstate.PeerStatus] because adding a non-comparable
[key.DiscoPublic] field there breaks reflect-based test helpers
(e.g. TestFilterFormatAndSortExitNodes' use of cmp.Diff), and
general LocalAPI clients have no need for disco keys. Since the
debug LocalAPI is gated behind the ts_omit_debug build tag, this
endpoint is automatically stripped from small binaries.
cmd/tta: add /restart-tailscaled handler (Linux-only, via /proc walk)
to drive the SIGKILL phase. On gokrazy the supervisor respawns
tailscaled within a second.
tstest/integration/testcontrol: add Server.AllOnline. When set,
every peer entry in MapResponses is marked Online=true. Several
disco-key handling fast paths in controlclient and wgengine
(removeUnwantedDiscoUpdates, removeUnwantedDiscoUpdatesFromFull
NetmapUpdate, the wgengine tsmpLearnedDisco fast path) only fire
for online peers; without this flag, tests exercising disco-key
rotation only hit the offline-peer code paths, which mask issues
and are several seconds slower in this scenario. Finer-grained
per-node online tracking can be added later.
tstest/natlab/vmtest: add Env.RotateDiscoKey,
Env.RestartTailscaled, Env.PeerDiscoKey, Node.Name, an
[AllOnline] EnvOption that plumbs through to
testcontrol.Server.AllOnline, and an exported
Env.Ping(from, to, type, timeout). Ping replaces the unexported
helper so callers can specify both a ping type (PingDisco for
warming peer state, PingTSMP for asserting end-to-end
connectivity) and a deadline. PeerDiscoKey returns its LocalAPI
error so callers inside tstest.WaitFor can retry transient
failures rather than fataling the test.
Updates #12639
Updates #13038
Change-Id: I3644f27fc30e52990ba25a3983498cc582ddb958
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Two cloud-platform nodes (e.g. sr-a and sr-b in TestSiteToSite) boot in
parallel via errgroup and both call ensureCompiled and the inline image
preparation block, racing to Begin() the same shared *Step (which is
deduped by name in Env.Step). The second goroutine panics:
panic: Step "Compile linux_amd64 binaries": Begin called in state running
panic: Step "Prepare ubuntu-24.04 image": Begin called in state done
ensureCompiled had a TOCTOU dedup attempt (released compileMu before
doing the work, only added to the compiled set at the end), and image
preparation had no dedup at all.
Replace the compiled set with a per-key map[string]*sync.Once for each
of compile and image preparation, so concurrent callers serialize on
the Once and only the first executes Begin/work/End.
Fixes commit 02ffe5baa8.
Updates #13038
Change-Id: If710bcc9e0aafebf0ad5b61553bae11458d976d7
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Cache a pre-booted macOS VM snapshot on disk so subsequent test runs
restore from the snapshot instead of cold-booting. The snapshot is keyed
by the Tart base image digest and a code version constant
(macOSSnapshotCodeVersion); bumping either invalidates the cache.
Snapshot preparation (one-time):
- Boot the Tart base image with a NAT NIC (--nat-nic flag)
- Wait for SSH, compile and install cmd/tta as a LaunchDaemon
- TTA polls the host via AF_VSOCK for an IP assignment; during prep
the host replies "wait"
- Disconnect NIC, save VM state via SIGINT
Test fast path (cached, ~7s to agent connected):
- APFS clone the snapshot, write test-specific config.json
- Launch Host.app with --disconnected-nic --attach-network --assign-ip
- VZ restores from SaveFile.vzvmsave (~5s with 4GB RAM)
- TTA's vsock poll gets the IP config, sets static IP via ifconfig
(bypasses DHCP entirely), switches driver addr to the IP directly
(bypasses DNS), and resets the dial context so the reverse-dial
reconnects immediately
- TTA agent connects to test driver within ~2s of IP assignment
Key optimizations:
- 4GB RAM instead of 8GB: halves SaveFile.vzvmsave (1.4GB vs 2.4GB),
halves restore time (5.5s vs 11s)
- AF_VSOCK IP assignment: bypasses macOS DHCP (~5-7s saved)
- Direct IP dial: bypasses DNS resolution for test-driver.tailscale
- Dial context reset: cancels stale in-flight dials from snapshot
- Kill instead of SIGINT for test VM cleanup (no state save needed)
- Parallel VM launches
Also:
- Add TestDriverIPv4/TestDriverPort constants to vnet
- Add --nat-nic and --assign-ip flags to Host.app
- Fix SIGINT handler: retain DispatchSource globally, use dispatchMain()
- Add vsock listener (port 51011) to Host.app for IP config protocol
- Add disconnectNetwork() to VMController for clean snapshot state
- Fix Makefile: set -o pipefail so xcodebuild failures aren't swallowed
Updates #13038
Change-Id: Icbab73b57af7df3ae96136fb49cda2536310f31b
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
When --vmtest-web is set, Host.app is launched with --screenshot-port 0
to start a localhost HTTP server that captures the VZVirtualMachineView
display. The Go test harness parses the SCREENSHOT_PORT=<port> line from
stdout, then polls every 2 seconds for JPEG thumbnails and pushes them
over WebSocket to the web dashboard.
Clicking a screenshot thumbnail opens a full-resolution image proxied
through the web UI's /screenshot/{node} endpoint.
Screenshot events are excluded from the EventBus history (they're large
and only the latest matters, stored in NodeStatus.Screenshot).
Updates #13038
Change-Id: I9bc67ddd1cc72948b33c555d4be3d8db06a41f6d
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Seamless key renewal has been the default in all clients since 1.90.
We retained the ability to disable it from the control plane as a
precaution, but we haven't seen any issues that require us to disable it.
We're now removing all the code for non-seamless key renewal, because we
don't expect to turn it on again, and indeed it's been untested in the
field for three releases so might contain latent bugs!
Updates tailscale/corp#33042
Change-Id: I4b80bf07a3a50298d1c303743484169accc8844b
Signed-off-by: Alex Chan <alexc@tailscale.com>
Add macOS VM support to the vmtest framework using Tart's pre-built
macOS images (ghcr.io/cirruslabs/macos-tahoe-base) instead of building
from IPSW. The Tart image has SIP disabled and SSH enabled.
At test time, the Tart base image's disk, NVRAM, and hardware identity
are APFS-cloned into a tailmac-compatible directory layout, and the VM
is booted headlessly via tailmac's Host.app (Virtualization.framework)
with its NIC connected to vnet's dgram socket.
New features:
- tailmac.go: ensureTartImage (auto-pull), cloneTartToTailmac (format
conversion), startTailMacVM (launch + cleanup)
- NoAgent() node option for VMs without TTA installed
- LANPing() for ICMP reachability testing via TTA's /ping endpoint
- IsMacOS field on OSImage, with GOOS/GOARCH support
- Dgram socket listener in Start() for macOS VMs
- Fix ReadFromUnix error spam on dgram socket close in vnet
TestMacOSAndLinuxCanPing verifies a macOS Tart VM and a gokrazy Linux
VM can ping each other on the same vnet LAN.
Updates #13038
Change-Id: I5e73a27878abf009f780fdf11a346fc857711cff
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a vmtest that brings up two Ubuntu nodes, each behind its own
EasyNAT, joined to the tailnet. The sender pushes a small file via
"tailscale file cp" and the receiver fetches it via "tailscale file
get --wait", asserting that the filename and contents round-trip
unchanged.
To make Taildrop work in vmtest, three small pieces were needed:
The Linux/FreeBSD cloud-init now starts tailscaled with --statedir as
well as --state=mem:, so the daemon has a VarRoot to host Taildrop's
incoming-files directory. State itself remains in-memory (so nothing
persists across reboots); only the var-root scratch space is on disk.
vmtest.New grows a variadic EnvOption parameter and a SameTailnetUser
helper. When the option is passed, Start sets AllNodesSameUser=true
on the embedded testcontrol.Server. Cross-node Taildrop requires the
sender and receiver to share a Tailnet user (or have an explicit
PeerCapabilityFileSharingTarget granted between them, which we don't
plumb here), so TestTaildrop opts in. Existing tests don't.
cmd/tta gains /taildrop-send and /taildrop-recv handlers that wrap
"tailscale file cp" and "tailscale file get --wait", plus
Env.SendTaildropFile and Env.RecvTaildropFile helpers in vmtest that
drive them.
Updates #13038
Change-Id: I8f5f70f88106e6e2ee07780dd46fe00f8efcfdf1
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a vmtest that brings up a Tailscale client, an Ubuntu VM acting
as a Mullvad-style plain-WireGuard exit node, and a non-Tailscale
webserver, each on its own NAT'd vnet network with a distinct WAN
IP. The test exercises Tailscale's IsWireGuardOnly peer code path:
the way the control plane wires Mullvad exit nodes into a client's
netmap, including the per-client SelfNodeV4MasqAddrForThisPeer
source-IP rewrite that lets a Tailscale CGNAT IP egress through a
plain-WireGuard tunnel that has no idea what Tailscale is.
The mullvad VM doesn't run wireguard-tools or kernel WireGuard;
instead, a new TTA endpoint /wg-server-up creates a real Linux TUN
named wg0, drives it with wireguard-go (already vendored), and
configures the kernel side (ip addr/up, ip_forward, iptables NAT
MASQUERADE) so decrypted traffic from the peer egresses with the
mullvad VM's WAN IP. Userspace vs kernel WireGuard makes no
difference on the wire — what's being tested is Tailscale's
plain-WireGuard exit-node code path, not the kernel module — and
this lets the test avoid downloading and installing .deb packages
inside the VM.
Adds Env.BringUpMullvadWGServer (calls /wg-server-up, returns the
generated WG public key as a key.NodePublic), Env.SetExitNodeIP
(EditPrefs ExitNodeIP directly, for exit nodes whose IPs aren't
discoverable via TTA), Env.ControlServer (exposes the underlying
testcontrol.Server so tests can UpdateNode / SetMasqueradeAddresses
to inject custom peers), and Env.Status (fetches a node's tailscale
status, used to read the client's pubkey so we can pin it as the
WG server's only allowed peer).
The test verifies that the webserver's echoed source IP is the
client's WAN with no exit node selected, the mullvad VM's WAN with
the WG-only peer selected as exit, and the client's WAN again after
clearing.
Updates #13038
Change-Id: I5bac4e0d832f05929f12cb77fa9946d7f5fb5ef1
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add an optional --vmtest-web flag that starts an HTTP server showing a
live dashboard for vmtest runs. The dashboard includes:
- Step progress tracker showing all test phases (compile, image prep,
QEMU launch, agent connect, tailscale up, test-specific steps)
with status icons and elapsed times
- Per-VM "virtual monitor" cards showing serial console output
streamed in realtime via WebSocket
- Per-NIC DHCP status (supporting multi-homed VMs like subnet routers)
- Per-node Tailscale status (hidden for non-tailnet VMs)
- Test status badge (Running/Passed/Failed) with live elapsed timer
- Event log showing all lifecycle events chronologically
Architecture follows the existing util/eventbus HTMX+WebSocket pattern:
the server pushes HTML fragments with hx-swap-oob attributes over a
WebSocket, and HTMX routes them to the correct DOM elements by ID.
Key components:
- vmstatus.go: Step tracker (Begin/End lifecycle), EventBus (pub/sub
with history for late joiners), VMEvent types, NodeStatus tracking
- web.go: HTTP server, WebSocket handler, template loading, ANSI-to-HTML
conversion via robert-nix/ansihtml, deterministic port selection
- assets/: HTML templates, CSS, HTMX library (copied from eventbus)
- vnet/vnet.go: DHCP event callback on Server for observing DHCP lifecycle
- qemu.go: Console log file tailing with manual offset-based reading
Usage:
go test ./tstest/natlab/vmtest/ --run-vm-tests --vmtest-web=:0 -v
When using :0, a deterministic port based on the test name is tried
first so re-runs get the same URL, falling back to OS-assigned on
conflict.
Updates #13038
Change-Id: I45281347b3d7af78ed9f4ff896033984f84dcb4d
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a --test-version flag to run the natlab VM tests against
released tailscale/tailscaled binaries downloaded from
pkgs.tailscale.com instead of building from the source tree.
The value can be a concrete release like "1.97.255", or "stable" /
"unstable" which resolve to the latest TarballsVersion on that track
via pkgs.tailscale.com/<track>/?mode=json. The track for a concrete
version is derived from its minor (even=stable, odd=unstable). The
host architecture (amd64 or arm64) selects the tarball.
Tarballs are cached + extracted under
~/.cache/tailscale-vmtest/builds/<version>_<arch>/ so they are not
re-fetched per test. tta is still always built from the local tree.
Cloud VMs (Ubuntu, Debian) pick up the downloaded binaries via the
existing files.tailscale file server. Non-Linux GOOS (FreeBSD) falls
back to building from source since pkgs.tailscale.com only ships
Linux tarballs. Gokrazy nodes continue to use binaries baked into
the gokrazy image; --test-version is a no-op for them.
Updates #13038
Change-Id: I213ef7db362dd17bf69d2685cbf2ab0ec5a3fee1
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add two tests building on TestExitNode's framework:
TestSubnetRouterPublicIP brings up a client, a subnet router, and a
webserver, each on its own NAT'd network with distinct WAN IPs. The
subnet router advertises the webserver's network as a route. The test
toggles the client's --accept-routes preference and asserts that the
webserver's echoed source IP switches between the client's own WAN
(direct dial) and the subnet router's WAN (forwarded through the
router and SNAT'd).
TestSubnetRouterAndExitNode adds a fourth node, an exit node that
advertises 0.0.0.0/0 + ::/0, and uses a table-driven layout with
subtests to cover the four combinations of (exit on/off, subnet
on/off). The case where both are on confirms longest-prefix match
wins: the subnet router's /24 takes precedence over the exit node's
/0. The exit node itself is configured with --accept-routes=off so
that, in the exit-only case, it forwards directly to the simulated
internet rather than re-routing the forwarded traffic via the subnet
router (which would otherwise mask the exit node's WAN as the
observed source).
Adds an Env.SetAcceptRoutes helper for toggling the RouteAll pref via
EditPrefs, used by both tests.
Updates #13038
Change-Id: Ifc2726db1df2f039c477c222484f535bebc40445
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add NIC attachment hot-swap support to Host.app: VZNetworkDevice.attachment
is writable at runtime, so --disconnected-nic creates a NIC with no
attachment, and --attach-network hot-swaps it to a vnet dgram socket
after boot/restore. macOS detects link-up and does DHCP.
Refactor TailMacConfigHelper: extract createDgramAttachment() and
createDisconnectedNetworkDeviceConfiguration() from the monolithic
createSocketNetworkDeviceConfiguration().
Add --screenshot-port flag for headless mode. Host.app serves GET
/screenshot as JPEG via a localhost HTTP server, capturing the
VZVirtualMachineView via CGWindowListCreateImage. The Go test harness
polls these to push live thumbnails to the web dashboard.
Also: SIGINT handler in headless mode for clean VM state save.
Updates #13038
Change-Id: I42fba0ecd760371b4ec5b26a0557e3dd0ba9ecae
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a vmtest TestExitNode that brings up a client, two exit nodes, and a
non-Tailscale webserver, each on its own NAT'd vnet network with a
distinct WAN IP. The test cycles the client's exit node setting between
off, exit1, and exit2 and asserts that the webserver echoes the expected
post-NAT source IP for each.
Three pieces were needed to make this work:
vnet now forwards TCP between simulated networks at the packet level,
mirroring the existing UDP path. When a guest VM sends TCP to another
simulated network's WAN IP, the source network's gateway rewrites src
via doNATOut and routeTCPPacket hands the packet off to the destination
network, which rewrites dst via doNATIn and writes the rewritten frame
onto the destination LAN. The TCP stacks of the two guest VM kernels
talk end-to-end; vnet just NATs the IP/port headers in flight, so all
TCP semantics (handshakes, options, sequence numbers, payload) are
preserved without a gvisor TCP termination in the middle. Adds a
focused TestInterNetworkTCP that exercises this path without any
Tailscale machinery.
cmd/tta binds its outbound dial to the default route's interface using
SO_BINDTODEVICE. Without that, the moment tailscaled installs
0.0.0.0/0 → tailscale0 in response to setting an exit node, TTA's
existing TCP connection to test-driver gets rerouted through the exit
node. From the test driver's perspective the connection's packets then
arrive with the exit node's WAN IP as the source rather than the
client's, so they don't match the existing flow and the connection is
dead — manifesting in the test as a hang on EditPrefs (which had
actually completed in milliseconds on the daemon side, but whose
response never made it back). Pinning the socket to the underlying NIC
keeps TTA's agent connection on a real interface regardless of any
policy routing tailscaled installs later. We bind rather than carry the
Tailscale bypass fwmark because the fwmark approach is conditional on
tailscaled having configured SO_MARK-based policy routing, while
binding is unconditional.
vmtest grows an Env.SetExitNode helper that sets ExitNodeIP via
EditPrefs through the agent, used by the new test.
Updates #13038
Change-Id: I9fc8f91848b7aa2297ef3eaf71fed9d96056a024
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
While working on #19444, I assumed that omitting `Start` would return a
clock that started at January 1, year 1, because that's the zero value
for a `time.Time`, but actually it uses the current UTC time instead.
This behaviour is non-obvious, so document it.
Updates #cleanup
Change-Id: Id91400778578655953ff3e1671ce470db97cfe91
Signed-off-by: Alex Chan <alexc@tailscale.com>
Add a Go benchmark that exercises a single tailnet client (a [tsnet.Server]
running in the test process) against a synthetic large initial netmap and
a stream of caller-driven peer add/remove deltas, all in-process.
The harness is split in two parts:
- tstest/largetailnet, a reusable package containing a [Streamer]
that hijacks the map long-poll on a [testcontrol.Server] via the new
AltMapStream hook, sends one initial MapResponse with N synthetic
peers, and forwards caller-supplied delta MapResponses on the same
stream. Helpers like MakePeer / AllocPeer build synthetic peers with
unique IDs and addresses derived from the Tailscale ULA range.
- tstest/largetailnet/largetailnet_test.go, BenchmarkGiantTailnet
(headless tailscaled workload, no IPN bus subscriber) and
BenchmarkGiantTailnetBusWatcher (GUI-client workload with one
Notify subscriber attached). Both are gated on
--actually-test-giant-tailnet (skipped by default), stand up an
in-process testcontrol + tsnet.Server, let Up block until the
initial N-peer netmap has been processed, then ResetTimer and run
add+remove pairs via b.Loop. Per-delta sync is via a test-only
[ipnlocal.LocalBackend.AwaitNodeKeyForTest] channel that closes
once the just-added peer key appears in the netmap (no-watcher
variant) or via bus-Notify drain (bus-watcher variant).
To support the hijack, [testcontrol.Server] grows an AltMapStream hook
and a small MapStreamWriter interface for benchmarks/stress tests that
need to drive a controlled MapResponse sequence; the normal serveMap
path is untouched when AltMapStream is nil. The streamer answers
non-streaming "lite" map polls (which controlclient issues before the
streaming long-poll to push HostInfo) with an empty MapResponse and
returns immediately, so the streaming poll that follows is the one
that gets the initial netmap.
The benchmark is intended for before/after comparisons of netmap- and
delta-handling changes targeted at large tailnets. CPU profiles on
unmodified main show the expected O(N) hotspots:
setControlClientStatusLocked / authReconfigLocked /
userspaceEngine.Reconfig / setNetMapLocked, plus JSON encoding of the
full Notify.NetMap to bus watchers (which dominates the BusWatcher
variant).
Median ms/op over 10 runs on unmodified main, by tailnet size N:
N no-watcher bus-watcher
10000 32 166
50000 222 865
100000 504 1765
250000 1551 4696
Recommended invocation:
go test ./tstest/largetailnet/ -run=^$ \
-bench='BenchmarkGiantTailnet(BusWatcher)?$' \
-benchtime=2000x -timeout=10m \
--actually-test-giant-tailnet \
--giant-tailnet-n=250000 \
-cpuprofile=/tmp/giant.cpu.pprof
Updates #12542
Change-Id: I4f5b2bb271a36ba853d5a0ffe82054ef2b15c585
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Verifies that site-to-site Tailscale subnet routing with
--snat-subnet-routes=false preserves the original source IP
end-to-end.
Topology: two sites, each with a Linux subnet router on a NATted WAN
plus an internal LAN, and a non-Tailscale backend on each LAN. Backends
are given static routes pointing to their local subnet router for the
remote site's prefix; an HTTP GET from backend-a to backend-b over
Tailscale returns a body containing backend-a's LAN IP.
Adds the supporting vmtest.SNATSubnetRoutes NodeOption and plumbs
snat-subnet-routes through TTA's /up handler. The webserver started by
vmtest.WebServer now also echoes the remote IP, for the preservation
assertion.
Adds a /add-route TTA endpoint (Linux-only for now) and a vmtest
Env.AddRoute helper so the test can install the backend static routes
through TTA rather than needing a host SSH key and debug NIC.
ensureGokrazy now always rebuilds the natlab qcow2 (once per test
process, via sync.Once) so the test picks up the new TTA and webserver
behavior.
This is pulled out of a larger pending change that adds FreeBSD
site-to-site subnet routing support; figured we should have at least
the Linux test covering what works today.
Updates #5573
Change-Id: I881c55b0f118ac9094546b5fbe68dddf179bb042
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The package updates started getting really slow yesterday. We can do
better, but attempt a band aid fix for now, as the test is failing about
a third of the time on PR CI.
Updates tailscale/corp#40465
Change-Id: Icf53292ba83dd1ed76b9bdf9fb94a8f6fb448c07
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
If we get a 429 response during node registration, use the `Retry-After`
header for backoff instead of the regular exponential backoff.
The rate limiter error is propagated to the user, just like other
registration errors are, e.g.
```
$ tailscale up
backend error: node registration rate limited; will retry after 57s
exit status 1
```
Updates tailscale/corp#39533
Signed-off-by: Anton Tolchanov <anton@tailscale.com>
On some nodes (found via natlab), the existing nodes last seen could be
unset. For these cases, we would want to accept the key and write a last
seen. This was breaking the cached netmap natlab tests.
Updates #12639
Signed-off-by: Claus Lensbøl <claus@tailscale.com>
Clients with the newly added node attribute
`"disable-linux-cgnat-drop-rule"` will not automatically drop inbound
traffic on non-Tailscale network interfaces with the source IP in the
CGNAT IP range. This is an initial proof-of-concept for enabling
connectivity with off-Tailnet CGNAT endpoints.
Fixestailscale/corp#36270.
Signed-off-by: Naman Sood <mail@nsood.in>
Use linkat via /proc/self/fd with AT_SYMLINK_FOLLOW to create a
hardlink of the test binary instead of copying it. This avoids
copying ~50MB+ binaries into each test's temp directory, making
test setup faster and reducing disk I/O.
The simpler os.Link(b.Path, ret.Path) can't be used here because
the source binary lives in the first test's TempDir, which may be
cleaned up before later tests call CopyTo. The open FD keeps the
inode alive after the path is deleted, but os.Link needs a valid
path. (See also b9f468240f which tried os.Link but is racy for
this reason.)
The /proc/self/fd approach works without elevated privileges,
unlike AT_EMPTY_PATH which requires CAP_DAC_READ_SEARCH. If the
linkat fails for any reason (e.g. cross-filesystem temp dirs), it
falls back to the existing full-copy path.
Fixes#19397
Change-Id: I4b1f97f7e63a9ae9e09dce36dfbdd1f6cff92320
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The kernel version parser used strings.Cut with "-" to handle versions
like "5.4.0-76-generic", but Debian uses "+" in versions like
"6.12.41+deb13-amd64".
Use strings.IndexAny to find the first "-" or "+" and truncate there.
Fixes TestKernelVersion on Debian systems.
Fixes#19395
Change-Id: I70e5f95682d54baf908e51f9f4b51c130b00aaaa
Co-Authored-By: Brad Fitzpatrick <bradfitz@tailscale.com>
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
When running integration tests over SSH (e.g., in remote development
environments), the SSH_CLIENT environment variable is set. This causes
isSSHOverTailscale() to incorrectly detect an SSH session and change
behavior.
Clear SSH_CLIENT in the test node environment to prevent these false
positives.
Fixes#19393
Change-Id: I1411abf0be9704cce37051476efb04d59beed386
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
The natlab-integrationtest CI job frequently flakes by exhausting its
3m go test timeout. The root cause is that the QEMU VMs run under
pure software emulation (TCG) with no KVM. Under TCG, the guest
kernel's timer calibration busy-loops are at the mercy of host CPU
scheduling. When two VMs boot simultaneously on a 2-core CI runner,
one VM's calibration gets starved and produces wrong results, leaving
the kernel with broken timers that prevent it from ever completing
boot — even after the other VM finishes and frees up CPU.
Additionally, the microvm machine type doesn't provide HPET hardware,
but the kernel command line specified clocksource=hpet. And the VM
image build (make natlab) ran inside the test itself, consuming most
of the 3m timeout budget before the actual test started.
Fix by:
- Enabling KVM when /dev/kvm is available, so timer calibration
uses real hardware timers unaffected by host CPU scheduling.
- Adding a CI step to set /dev/kvm permissions on the GitHub
Actions runner (ubuntu-latest provides KVM but needs a udev rule).
- Pre-building the VM image in a separate CI step so it doesn't
cut into the go test -timeout budget.
- Replacing the hardcoded 60s context timeout with one derived from
t.Deadline(), so the test uses the full -timeout budget.
- Adding VM boot progress detection (AwaitFirstPacket) and QMP
diagnostics, so boot failures produce clear errors instead of
opaque "context deadline exceeded" messages.
With KVM enabled, the test passes reliably even on a single CPU core
with 3 parallel workers — a scenario that was 100% broken under TCG.
Fixes#18906
Change-Id: I4c87631a9c9678d185b9f30cb05c0f7bfa9f5c62
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For tests to loudly declare (and panic on violation) when they're doing
something that's not safe in a parallel test.
Fixes#19385
Change-Id: If79693b0c235c146871a05ed74fa9ea75bb500f9
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a --headless flag to the Host.app Run subcommand for running
macOS VMs without a GUI, enabling use from test frameworks.
Key changes:
- HostCli.swift: When --headless is set, run the VM via VMController
+ RunLoop.main.run() instead of NSApplicationMain. Using the
RunLoop (not dispatchMain) is required because VZ framework
callbacks depend on RunLoop sources.
- VMController.swift: Add headless parameter to createVirtualMachine
that configures a single socket-based NIC (no NAT NIC). This
matches the NIC configuration used when creating/saving VMs, so
saved state restoration works correctly. A NIC count mismatch
causes VZ to silently fail to execute guest code.
- TailMacConfigHelper.swift: Clean up socket network device logging.
- Config.swift: Move VM storage from ~/VM.bundle to
~/.cache/tailscale/vmtest/macos/.
- TailMac.swift: Fix dispatchMain→RunLoop.main.run() in the create
command (same VZ RunLoop requirement).
Updates #13038
Change-Id: Iea51c043aa92e8fc6257139b9f0e2e7677072fa2
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Start using a common helper for tests to declare that they require root.
This is step 1. A later step will then make this helper track which tests were
skipped so a subsequent pass will run these test as root.
Updates tailscale/corp#40007
Change-Id: I4979e1def0fa3691d38c83f48c89aaa443e7f62e
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
As a warm-up to making natlab support multiple operating systems,
start with an easy one (in that it's also Unixy and open source like
Linux) and add FreeBSD 15.0 as a VM OS option for the vmtest
integration test framework, and add TestSubnetRouterFreeBSD which
tests subnet routing through a FreeBSD VM (Gokrazy → FreeBSD →
Gokrazy).
Key changes:
- Add FreeBSD150 OSImage using the official FreeBSD 15.0
BASIC-CLOUDINIT cloud image (xz-compressed qcow2)
- Add GOOS()/IsFreeBSD() methods to OSImage for cross-compilation
and OS-specific behavior
- Handle xz-compressed image downloads in ensureImage
- Refactor compileBinaries into compileBinariesForOS to support
multiple GOOS targets (linux, freebsd), with binaries registered
at <goos>/<name> paths on the file server VIP
- Add FreeBSD-specific cloud-init (nuageinit) user-data generation:
string-form runcmd (nuageinit doesn't support YAML arrays),
fetch(1) instead of curl, FreeBSD sysctl names for IP forwarding,
mkdir /usr/local/bin, PATH setup for tta
- Skip network-config in cidata ISO for FreeBSD (DHCP via rc.conf)
Updates tailscale/tailscale#13038
Change-Id: Ibeb4f7d02659d5cd8e3a7c3a66ee7b1a92a0110d
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add tstest/natlab/vmtest, a high-level framework for running multi-VM
integration tests with mixed OS types (gokrazy + Ubuntu/Debian cloud
images) connected via natlab's vnet virtual network.
The vmtest package provides:
- Env type that orchestrates vnet, QEMU processes, and agent connections
- OS image support (Gokrazy, Ubuntu2404, Debian12) with download/cache
- QEMU launch per OS type (microvm for gokrazy, q35+KVM for cloud)
- Cloud-init seed ISO generation with network-config for multi-NIC
- Cross-compilation of test binaries for cloud VMs
- Debug SSH NIC on cloud VMs for interactive debugging
- Test helpers: ApproveRoutes, HTTPGet, TailscalePing, DumpStatus,
WaitForPeerRoute, SSHExec
TTA enhancements (cmd/tta):
- Parameterize /up (accept-routes, advertise-routes, snat-subnet-routes)
- Add /set, /start-webserver, /http-get endpoints
- /http-get uses local.Client.UserDial for Tailscale-routed requests
- Fix /ping for non-gokrazy systems
TestSubnetRouter exercises a 3-VM subnet router scenario:
client (gokrazy) → subnet-router (Ubuntu, dual-NIC) → backend (gokrazy)
Verifies HTTP access to the backend webserver through the Tailscale
subnet route. Passes in ~30 seconds.
Updates tailscale/tailscale#13038
Change-Id: I165b64af241d37f5f5870e796a52502fc56146fa
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Multi-NIC support:
- Add nodeNIC type and node.extraNICs for secondary network interfaces
- Add netForMAC/macForNet to route packets to the correct network by MAC
- Update initFromConfig to allocate a MAC + LAN IP per network
- Fix handleEthernetFrameFromVM, ServeUnixConn to use netForMAC
- Fix MACOfIP, writeEth, WriteUDPPacketNoNAT, gVisor write path, and
createARPResponse to use macForNet (return the MAC actually on that
network, not the node's primary MAC)
- Fix createDHCPResponse for multi-NIC (correct client IP and subnet)
- Add nodeNICMac for secondary NIC MAC generation
- Add Node accessors: NumNICs, NICMac, Networks, LanIP
DHCP fixes:
- Include LeaseTime, SubnetMask, Router, DNS in DHCP Offer (not just
Ack). systemd-networkd requires these to accept an Offer.
- Fix DHCP response source IP: use gateway IP instead of echoing
the request's destination (which was 255.255.255.255 for discovers)
New VIPs:
- cloud-init.tailscale: serves per-node cloud-init meta-data, user-data,
and network-config for VMs booting with nocloud datasource
- files.tailscale: serves binary files (tta, tailscale, tailscaled)
registered via RegisterFile for cloud VM provisioning
- Add ControlServer() accessor for test control server
This is necessary for a three-VM natlab subnet router
integration test, coming later.
Updates #13038
Change-Id: I59f9f356bae9b5509c117265237983972dfdd5af
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
When SetSubnetRoutes is called, also send updatePeerChanged to all
other connected nodes so they re-fetch their MapResponse and learn
about the updated AllowedIPs. Without this, peers never see new
subnet routes until they happen to reconnect to the control server.
Discovered while working on a three-VM natlab subnet router
integration test, coming later.
Updates #13038
Change-Id: I20e7a2fda994a8ab0e7a24240e6eae536f4f5f15
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Move the ipn/desktop blank import from cmd/tailscaled/tailscaled_windows.go
into feature/condregister/maybe_desktop_sessions.go, consistent with how
all other modular features are registered. tailscaled already imports
feature/condregister, so it still gets ipn/desktop on Windows.
Updates #12614
Change-Id: I92418c4bf0e67f0ab40542e47584762ac0ffa2b2
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add a new vet analyzer that checks t.Run subtest names don't contain
characters requiring quoting when re-running via "go test -run". This
enforces the style guide rule: don't use spaces or punctuation in
subtest names.
The analyzer flags:
- Direct t.Run calls with string literal names containing spaces,
regex metacharacters, quotes, or other problematic characters
- Table-driven t.Run(tt.name, ...) calls where tt ranges over a
slice/map literal with bad name field values
Also fix all 978 existing violations across 81 test files, replacing
spaces with hyphens and shortening long sentence-like names to concise
hyphenated forms.
Updates #19242
Change-Id: Ib0ad96a111bd8e764582d1d4902fe2599454ab65
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Previously, running `add/remove/revoke-keys` without passing any keys
would fail with an unhelpful error:
```console
$ tailscale lock revoke-keys
generation of recovery AUM failed: sending generate-recovery-aum: 500 Internal Server Error: no provided key is currently trusted
```
or
```console
$ tailscale lock revoke-keys
generation of recovery AUM failed: sending generate-recovery-aum: 500 Internal Server Error: network-lock is not active
```
Now they fail with a more useful error:
```console
$ tailscale lock revoke-keys
missing argument, expected one or more tailnet lock keys
```
Fixes#19130
Change-Id: I9d81fe2f5b92a335854e71cbc6928e7e77e537e3
Signed-off-by: Alex Chan <alexc@tailscale.com>
Brad Fitzpatrick
M. J. Fromberger
Claus Lensbøl
Fernando Serboncini
Alex Chan
Tom Proctor
Anton Tolchanov
Naman Sood
Avery Pennarun