Implements the controller for the new ProxyGroup CRD, designed for
running proxies in a high availability configuration. Each proxy gets
its own config and state Secret, and its own tailscale node ID.
We are currently mounting all of the config secrets into the container,
but will stop mounting them and instead read them directly from the kube
API once #13578 is implemented.
Updates #13406
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
Adds a new reconciler that reconciles ExternalName Services that define a
tailnet target that should be exposed to cluster workloads on a ProxyGroup's
proxies.
The reconciler ensures that for each such service, the config mounted to
the proxies is updated with the tailnet target definition and that
and EndpointSlice and ClusterIP Service are created for the service.
Adds a new reconciler that ensures that as proxy Pods become ready to route
traffic to a tailnet target, the EndpointSlice for the target is updated
with the Pods' endpoints.
Updates tailscale/tailscale#13406
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
The operator creates a non-reusable auth key for each of
the cluster proxies that it creates and puts in the tailscaled
configfile mounted to the proxies.
The proxies are always tagged, and their state is persisted
in a Kubernetes Secret, so their node keys are expected to never
be regenerated, so that they don't need to re-auth.
Some tailnet configurations however have seen issues where the auth
keys being left in the tailscaled configfile cause the proxies
to end up in unauthorized state after a restart at a later point
in time.
Currently, we have not found a way to reproduce this issue,
however this commit removes the auth key from the config once
the proxy can be assumed to have logged in.
If an existing, logged-in proxy is upgraded to this version,
its redundant auth key will be removed from the conffile.
If an existing, logged-in proxy is downgraded from this version
to a previous version, it will work as before without re-issuing key
as the previous code did not enforce that a key must be present.
Updates tailscale/tailscale#13451
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
The ProxyGroup CRD specifies a set of N pods which will each be a
tailnet device, and will have M different ingress or egress services
mapped onto them. It is the mechanism for specifying how highly
available proxies need to be. This commit only adds the definition, no
controller loop, and so it is not currently functional.
This commit also splits out TailnetDevice and RecorderTailnetDevice
into separate structs because the URL field is specific to recorders,
but we want a more generic struct for use in the ProxyGroup status field.
Updates #13406
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
cmd/k8s-operator,k8s-operator,kube: Add TSRecorder CRD + controller
Deploys tsrecorder images to the operator's cluster. S3 storage is
configured via environment variables from a k8s Secret. Currently
only supports a single tsrecorder replica, but I've tried to take early
steps towards supporting multiple replicas by e.g. having a separate
secret for auth and state storage.
Example CR:
```yaml
apiVersion: tailscale.com/v1alpha1
kind: Recorder
metadata:
name: rec
spec:
enableUI: true
```
Updates #13298
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
Rename kube/{types,client,api} -> kube/{kubetypes,kubeclient,kubeapi}
so that we don't need to rename the package on each import to
convey that it's kubernetes specific.
Updates#cleanup
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Further split kube package into kube/{client,api,types}. This is so that
consumers who only need constants/static types don't have to import
the client and api bits.
Updates#cleanup
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
* cmd/k8s-operator,k8s-operator/sessonrecording: ensure CastHeader contains terminal size
For tsrecorder to be able to play session recordings, the recording's
CastHeader must have '.Width' and '.Height' fields set to non-zero.
Kubectl (or whoever is the client that initiates the 'kubectl exec'
session recording) sends the terminal dimensions in a resize message that
the API server proxy can intercept, however that races with the first server
message that we need to record.
This PR ensures we wait for the terminal dimensions to be processed from
the first resize message before any other data is sent, so that for all
sessions with terminal attached, the header of the session recording
contains the terminal dimensions and the recording can be played by tsrecorder.
Updates tailscale/tailscale#19821
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Previously, despite what the commit said, we were using a raw IP socket
that was *not* an AF_PACKET socket, and thus was subject to the host
firewall rules. Switch to using a real AF_PACKET socket to actually get
the functionality we want.
Updates #13140
Signed-off-by: Andrew Dunham <andrew@du.nham.ca>
Change-Id: If657daeeda9ab8d967e75a4f049c66e2bca54b78
Currently, we use PermitRead/PermitWrite/PermitCert permission flags to determine which operations are allowed for a LocalAPI client.
These checks are performed when localapi.Handler handles a request. Additionally, certain operations (e.g., changing the serve config)
requires the connected user to be a local admin. This approach is inherently racey and is subject to TOCTOU issues.
We consider it to be more critical on Windows environments, which are inherently multi-user, and therefore we prevent more than one
OS user from connecting and utilizing the LocalBackend at the same time. However, the same type of issues is also applicable to other
platforms when switching between profiles that have different OperatorUser values in ipn.Prefs.
We'd like to allow more than one Windows user to connect, but limit what they can see and do based on their access rights on the device
(e.g., an local admin or not) and to the currently active LoginProfile (e.g., owner/operator or not), while preventing TOCTOU issues on Windows
and other platforms. Therefore, we'd like to pass an actor from the LocalAPI to the LocalBackend to represent the user performing the operation.
The LocalBackend, or the profileManager down the line, will then check the actor's access rights to perform a given operation on the device
and against the current (and/or the target) profile.
This PR does not change the current permission model in any way, but it introduces the concept of an actor and includes some preparatory
work to pass it around. Temporarily, the ipnauth.Actor interface has methods like IsLocalSystem and IsLocalAdmin, which are only relevant
to the current permission model. It also lacks methods that will actually be used in the new model. We'll be adding these gradually in the next
PRs and removing the deprecated methods and the Permit* flags at the end of the transition.
Updates tailscale/corp#18342
Signed-off-by: Nick Khyl <nickk@tailscale.com>
This commit adds a new usermetric package and wires
up metrics across the tailscale client.
Updates tailscale/corp#22075
Co-authored-by: Anton Tolchanov <anton@tailscale.com>
Signed-off-by: Kristoffer Dalby <kristoffer@tailscale.com>
After the upstream PR is merged, we can point directly at github.com/vishvananda/netlink
and retire github.com/tailscale/netlink.
See https://github.com/vishvananda/netlink/pull/1006
Updates #12298
Signed-off-by: Percy Wegmann <percy@tailscale.com>
In prep for updating to new staticcheck required for Go 1.23.
Updates #12912
Change-Id: If77892a023b79c6fa798f936fc80428fd4ce0673
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
In 2f27319baf we disabled GRO due to a
data race around concurrent calls to tstun.Wrapper.Write(). This commit
refactors GRO to be thread-safe, and re-enables it on Linux.
This refactor now carries a GRO type across tstun and netstack APIs
with a lifetime that is scoped to a single tstun.Wrapper.Write() call.
In 25f0a3fc8f we used build tags to
prevent importation of gVisor's GRO package on iOS as at the time we
believed it was contributing to additional memory usage on that
platform. It wasn't, so this commit simplifies and removes those
build tags.
Updates tailscale/corp#22353
Updates tailscale/corp#22125
Updates #6816
Signed-off-by: Jordan Whited <jordan@tailscale.com>
cmd/k8s-operator/deploy: replace wildcards in Kubernetes Operator RBAC role definitions with verbs
fixes: #13168
Signed-off-by: Pierig Le Saux <pierig@n3xt.io>
Coder has just adopted nhooyr/websocket which unfortunately changes the import path.
`github.com/coder/coder` imports `tailscale.com/net/wsconn` which was still pointing
to `nhooyr.io/websocket`, but this change updates it.
See https://coder.com/blog/websocket
Updates #13154
Change-Id: I3dec6512472b14eae337ae22c5bcc1e3758888d5
Signed-off-by: Kyle Carberry <kyle@carberry.com>
cmd/k8s-operator,k8s-operator/sessionrecording: support recording WebSocket sessions
Kubernetes currently supports two streaming protocols, SPDY and WebSockets.
WebSockets are replacing SPDY, see
https://github.com/kubernetes/enhancements/issues/4006.
We were currently only supporting SPDY, erroring out if session
was not SPDY and relying on the kube's built-in SPDY fallback.
This PR:
- adds support for parsing contents of 'kubectl exec' sessions streamed
over WebSockets
- adds logic to distinguish 'kubectl exec' requests for a SPDY/WebSockets
sessions and call the relevant handler
Updates tailscale/corp#19821
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Co-authored-by: Tom Proctor <tomhjp@users.noreply.github.com>
* cmd/k8s-operator: fix DNS reconciler for dual-stack clusters
This fixes a bug where DNS reconciler logic was always assuming
that no more than one EndpointSlice exists for a Service.
In fact, there can be multiple, for example, in dual-stack
clusters, but also in other cases this is valid (as per kube docs).
This PR:
- allows for multiple EndpointSlices
- picks out the ones for IPv4 family
- deduplicates addresses
Updates tailscale/tailscale#13056
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Co-authored-by: Tom Proctor <tomhjp@users.noreply.github.com>
Package setting contains types for defining and representing policy settings.
It facilitates the registration of setting definitions using Register and RegisterDefinition,
and the retrieval of registered setting definitions via Definitions and DefinitionOf.
This package is intended for use primarily within the syspolicy package hierarchy,
and added in a preparation for the next PRs.
Updates #12687
Signed-off-by: Nick Khyl <nickk@tailscale.com>
This commit implements TCP GRO for packets being written to gVisor on
Linux. Windows support will follow later. The wireguard-go dependency is
updated in order to make use of newly exported IP checksum functions.
gVisor is updated in order to make use of newly exported
stack.PacketBuffer GRO logic.
TCP throughput towards gVisor, i.e. TUN write direction, is dramatically
improved as a result of this commit. Benchmarks show substantial
improvement, sometimes as high as 2x. High bandwidth-delay product
paths remain receive window limited, bottlenecked by gVisor's default
TCP receive socket buffer size. This will be addressed in a follow-on
commit.
The iperf3 results below demonstrate the effect of this commit between
two Linux computers with i5-12400 CPUs. There is roughly ~13us of round
trip latency between them.
The first result is from commit 57856fc without TCP GRO.
Starting Test: protocol: TCP, 1 streams, 131072 byte blocks
- - - - - - - - - - - - - - - - - - - - - - - - -
Test Complete. Summary Results:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 4.77 GBytes 4.10 Gbits/sec 20 sender
[ 5] 0.00-10.00 sec 4.77 GBytes 4.10 Gbits/sec receiver
The second result is from this commit with TCP GRO.
Starting Test: protocol: TCP, 1 streams, 131072 byte blocks
- - - - - - - - - - - - - - - - - - - - - - - - -
Test Complete. Summary Results:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 10.6 GBytes 9.14 Gbits/sec 20 sender
[ 5] 0.00-10.00 sec 10.6 GBytes 9.14 Gbits/sec receiver
Updates #6816
Signed-off-by: Jordan Whited <jordan@tailscale.com>
This commit implements TCP GSO for packets being read from gVisor on
Linux. Windows support will follow later. The wireguard-go dependency is
updated in order to make use of newly exported GSO logic from its tun
package.
A new gVisor stack.LinkEndpoint implementation has been established
(linkEndpoint) that is loosely modeled after its predecessor
(channel.Endpoint). This new implementation supports GSO of monster TCP
segments up to 64K in size, whereas channel.Endpoint only supports up to
32K. linkEndpoint will also be required for GRO, which will be
implemented in a follow-on commit.
TCP throughput from gVisor, i.e. TUN read direction, is dramatically
improved as a result of this commit. Benchmarks show substantial
improvement through a wide range of RTT and loss conditions, sometimes
as high as 5x.
The iperf3 results below demonstrate the effect of this commit between
two Linux computers with i5-12400 CPUs. There is roughly ~13us of round
trip latency between them.
The first result is from commit 57856fc without TCP GSO.
Starting Test: protocol: TCP, 1 streams, 131072 byte blocks
- - - - - - - - - - - - - - - - - - - - - - - - -
Test Complete. Summary Results:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 2.51 GBytes 2.15 Gbits/sec 154 sender
[ 5] 0.00-10.00 sec 2.49 GBytes 2.14 Gbits/sec receiver
The second result is from this commit with TCP GSO.
Starting Test: protocol: TCP, 1 streams, 131072 byte blocks
- - - - - - - - - - - - - - - - - - - - - - - - -
Test Complete. Summary Results:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 12.6 GBytes 10.8 Gbits/sec 6 sender
[ 5] 0.00-10.00 sec 12.6 GBytes 10.8 Gbits/sec receiver
Updates #6816
Signed-off-by: Jordan Whited <jordan@tailscale.com>
cmd/k8s-operator,k8s-operator/sessionrecording,sessionrecording,ssh/tailssh: refactor session recording functionality
Refactor SSH session recording functionality (mostly the bits related to
Kubernetes API server proxy 'kubectl exec' session recording):
- move the session recording bits used by both Tailscale SSH
and the Kubernetes API server proxy into a shared sessionrecording package,
to avoid having the operator to import ssh/tailssh
- move the Kubernetes API server proxy session recording functionality
into a k8s-operator/sessionrecording package, add some abstractions
in preparation for adding support for a second streaming protocol (WebSockets)
Updates tailscale/corp#19821
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Re-instates the functionality that generates CRD API docs, but using
a different library as the one we were using earlier seemed to have
some issues with its Git history.
Also regenerates the docs (make kube-generate-all).
Updates tailscale/tailscale#12859
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Updates tailscale/tailscale#1634
This PR introduces a new `captive-portal-detected` Warnable which is set to an unhealthy state whenever a captive portal is detected on the local network, preventing Tailscale from connecting.
ipn/ipnlocal: fix captive portal loop shutdown
Change-Id: I7cafdbce68463a16260091bcec1741501a070c95
net/captivedetection: fix mutex misuse
ipn/ipnlocal: ensure that we don't fail to start the timer
Change-Id: I3e43fb19264d793e8707c5031c0898e48e3e7465
Signed-off-by: Andrew Dunham <andrew@du.nham.ca>
Signed-off-by: Andrea Gottardo <andrea@gottardo.me>
Remove fybrik.io/crdoc dependency as it is causing issues for folks attempting
to vendor tailscale using GOPROXY=direct.
This means that the CRD API docs in ./k8s-operator/api.md will no longer
be generated- I am going to look at replacing it with another tool
in a follow-up.
Updates tailscale/tailscale#12859
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
cmd/k8s-operator,ssh/tailssh,tsnet: optionally record kubectl exec sessions
The Kubernetes operator's API server proxy, when it receives a request
for 'kubectl exec' session now reads 'RecorderAddrs', 'EnforceRecorder'
fields from tailcfg.KubernetesCapRule.
If 'RecorderAddrs' is set to one or more addresses (of a tsrecorder instance(s)),
it attempts to connect to those and sends the session contents
to the recorder before forwarding the request to the kube API
server. If connection cannot be established or fails midway,
it is only allowed if 'EnforceRecorder' is not true (fail open).
Updates tailscale/corp#19821
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Co-authored-by: Maisem Ali <maisem@tailscale.com>
cmd/containerboot,cmd/k8s-operator: enable IPv6 for fqdn egress proxies
Don't skip installing egress forwarding rules for IPv6 (as long as the host
supports IPv6), and set headless services `ipFamilyPolicy` to
`PreferDualStack` to optionally enable both IP families when possible. Note
that even with `PreferDualStack` set, testing a dual-stack GKE cluster with
the default DNS setup of kube-dns did not correctly set both A and
AAAA records for the headless service, and instead only did so when
switching the cluster DNS to Cloud DNS. For both IPv4 and IPv6 to work
simultaneously in a dual-stack cluster, we require headless services to
return both A and AAAA records.
If the host doesn't support IPv6 but the FQDN specified only has IPv6
addresses available, containerboot will exit with error code 1 and an
error message because there is no viable egress route.
Fixes#12215
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
Adds a new TailscaleProxyReady condition type for use in corev1.Service
conditions.
Also switch our CRDs to use metav1.Condition instead of
ConnectorCondition. The Go structs are seralized identically, but it
updates some descriptions and validation rules. Update k8s
controller-tools and controller-runtime deps to fix the documentation
generation for metav1.Condition so that it excludes comments and
TODOs.
Stop expecting the fake client to populate TypeMeta in tests. See
kubernetes-sigs/controller-runtime#2633 for details of the change.
Finally, make some minor improvements to validation for service hostnames.
Fixes#12216
Co-authored-by: Irbe Krumina <irbe@tailscale.com>
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
* cmd/containerboot: store device ID before setting up proxy routes.
For containerboot instances whose state needs to be stored
in a Kubernetes Secret, we additonally store the device's
ID, FQDN and IPs.
This is used, between other, by the Kubernetes operator,
who uses the ID to delete the device when resources need
cleaning up and writes the FQDN and IPs on various kube
resource statuses for visibility.
This change shifts storing device ID earlier in the proxy setup flow,
to ensure that if proxy routing setup fails,
the device can still be deleted.
Updates tailscale/tailscale#12146
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
* code review feedback
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
---------
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
This PR is in prep of adding logic to control to be able to parse
tailscale.com/cap/kubernetes grants in control:
- moves the type definition of PeerCapabilityKubernetes cap to a location
shared with control.
- update the Kubernetes cap rule definition with fields for granting
kubectl exec session recording capabilities.
- adds a convenience function to produce tailcfg.RawMessage from an
arbitrary cap rule and a test for it.
An example grant defined via ACLs:
"grants": [{
"src": ["tag:eng"],
"dst": ["tag:k8s-operator"],
"app": {
"tailscale.com/cap/kubernetes": [{
"recorder": ["tag:my-recorder"]
“enforceRecorder”: true
}],
},
}
]
This grant enforces `kubectl exec` sessions from tailnet clients,
matching `tag:eng` via API server proxy matching `tag:k8s-operator`
to be recorded and recording to be sent to a tsrecorder instance,
matching `tag:my-recorder`.
The type needs to be shared with control because we want
control to parse this cap and resolve tags to peer IPs.
Updates tailscale/corp#19821
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Add a new .spec.tailscale.acceptRoutes field to ProxyClass,
that can be optionally set to true for the proxies to
accept routes advertized by other nodes on tailnet (equivalent of
setting --accept-routes to true).
Updates tailscale/tailscale#12322,tailscale/tailscale#10684
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Also removes hardcoded image repo/tag from example DNSConfig resource
as the operator now knows how to default those.
Updates tailscale/tailscale#11019
Signed-off-by: Irbe Krumina <irbe@tailscale.com>
Add new fields TailnetIPs and Hostname to Connector Status. These
contain the addresses of the Tailscale node that the operator created
for the Connector to aid debugging.
Fixes#12214
Signed-off-by: Tom Proctor <tomhjp@users.noreply.github.com>
