In some configurations, user explicitly do not want to store
tailscale state in k8s secrets, because doing that leads to
some annoying permission issues with sidecar containers.
With this change, TS_KUBE_SECRET="" and TS_STATE_DIR=/foo
will force storage to file when running in kubernetes.
Fixes#6704.
Signed-off-by: David Anderson <danderson@tailscale.com>
The operator creates a fair bit of internal cluster state to manage proxying,
dumping it all in the default namespace is handy for development but rude
for production.
Updates #502
Signed-off-by: David Anderson <danderson@tailscale.com>
We used to need to do timed requeues in a few places in the reconcile logic,
and the easiest way to do that was to plumb reconcile.Result return values
around. But now we're purely event-driven, so the only thing we care about
is whether or not an error occurred.
Incidentally also fix a very minor bug where headless services would get
completely ignored, rather than reconciled into the correct state. This
shouldn't matter in practice because you can't transition from a headful
to a headless service without a deletion, but for consistency let's avoid
having a path that takes no definite action if a service of interest does
exist.
Updates #502.
Signed-off-by: David Anderson <danderson@tailscale.com>
Previously, we had to do blind timed requeues while waiting for
the tailscale hostname, because we looked up the hostname through
the API. But now the proxy container image writes back its hostname
to the k8s secret, so we get an event-triggered reconcile automatically
when the time is right.
Updates #502
Signed-off-by: David Anderson <danderson@tailscale.com>
As is convention in the k8s world, use zap for structured logging. For
development, OPERATOR_LOGGING=dev switches to a more human-readable output
than JSON.
Updates #502
Signed-off-by: David Anderson <danderson@tailscale.com>
Our reconcile loop gets triggered again when the StatefulSet object
finally disappears (in addition to when its deletion starts, as indicated
by DeletionTimestamp != 0). So, we don't need to queue additional
reconciliations to proceed with the remainder of the cleanup, that
happens organically.
Signed-off-by: David Anderson <danderson@tailscale.com>
Tests cover configuring a proxy through an annotation rather than a
LoadBalancerClass, and converting between those two modes at runtime.
Updates #502.
Signed-off-by: David Anderson <danderson@tailscale.com>
For other test cases, the operator is going to produce similar generated
objects in several codepaths, and those objects are large. Move them out
to helpers so that the main test code stays a bit more intelligible.
The top-level Service that we start and end with remains in the main test
body, because its shape at the start and end is one of the main things that
varies a lot between test cases.
Updates #502.
Signed-off-by: David Anderson <danderson@tailscale.com>
The test verifies one of the successful reconcile paths, where
a client requests an exposed service via a LoadBalancer class.
Updates #502.
Signed-off-by: David Anderson <danderson@tailscale.com>
Also introduces an intermediary interface for the tailscale client, in
preparation for operator tests that fake out the Tailscale API interaction.
Updates #502.
Signed-off-by: David Anderson <danderson@tailscale.com>
This was initially developed in a separate repo, but for build/release
reasons and because go module management limits the damage of importing
k8s things now, moving it into this repo.
At time of commit, the operator enables exposing services over tailscale,
with the 'tailscale' loadBalancerClass. It also currently requires an
unreleased feature to access the Tailscale API, so is not usable yet.
Updates #502.
Signed-off-by: David Anderson <danderson@tailscale.com>
We still have to shell out to `tailscale up` because the container image's
API includes "arbitrary flags to tailscale up", unfortunately. But this
should still speed up startup a little, and also enables k8s-bound containers
to update their device information as new netmap updates come in.
Fixes#6657
Signed-off-by: David Anderson <danderson@tailscale.com>
Only enable forwarding for an IP family if any forwarding is required
for that family.
Fixes#6221.
Signed-off-by: David Anderson <danderson@tailscale.com>
This implements the same functionality as the former run.sh, but in Go
and with a little better awareness of tailscaled's lifecycle.
Also adds TS_AUTH_ONCE, which fixes the unfortunate behavior run.sh had
where it would unconditionally try to reauth every time if you gave it
an authkey, rather than try to use it only if auth is actually needed.
This makes it a bit nicer to deploy these containers in automation, since
you don't have to run the container once, then go and edit its definition
to remove authkeys.
Signed-off-by: David Anderson <danderson@tailscale.com>
From the original commit that implemented it:
It accepts Postgres connections over Tailscale only, dials
out to the configured upstream database with TLS (using
strong settings, not the swiss cheese that postgres defaults to),
and proxies the client through.
It also keeps an audit log of the sessions it passed through,
along with the Tailscale-provided machine and user identity
of the connecting client.
In our other repo, this was:
commit 92e5edf98e8c2be362f564a408939a5fc3f8c539,
Change-Id I742959faaa9c7c302bc312c7dc0d3327e677dc28.
Co-authored-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Signed-off-by: David Anderson <danderson@tailscale.com>
This lets us distinguish "no IPv6 because the device's ISP doesn't
offer IPv6" from "IPv6 is unavailable/disabled in the OS".
Signed-off-by: David Anderson <danderson@tailscale.com>
Doing so makes development unpleasant, because we have to first break the
client by bumping to a version the control server rejects, then upgrade
the control server to make it accept the new version.
This strict rejection at handshake time is only necessary if we want to
blocklist some vulnerable protocol versions in the future. So, switch
to a default-permissive stance: until we have such a version that we
have to eagerly block early, we'll accept whatever version the client
presents, and leave it to the user of controlbase.Conn to make decisions
based on that version.
Noise still enforces that the client and server *agree* on what protocol
version is being used, and the control server still has the option to
finish the handshake and then hang up with an in-noise error, rather
than abort at the handshake level.
Updates #3488
Signed-off-by: David Anderson <danderson@tailscale.com>
This is so that we can plumb our client capability version through
the protocol as the Noise version. The capability version increments
more frequently than strictly required (the Noise version only needs
to change when cryptographically-significant changes are made to
the protocol, whereas the capability version also indicates changes
in non-cryptographically-significant parts of the protocol), but this
gives us a safe pre-auth way to determine if the client supports
future protocol features, while still relying on Noise's strong
assurance that the client and server have agreed on the same version.
Currently, the server executes the same protocol regardless of the
version number, and just presents the version to the caller so they
can do capability-based things in the upper RPC protocol. In future,
we may add a ratchet to disallow obsolete protocols, or vary the
Noise handshake behavior based on requested version.
Updates #3488
Signed-off-by: David Anderson <danderson@tailscale.com>
prober: add labels to Probe instances.
This allows especially dynamically-registered probes to have a bunch
more dimensions along which they can be sliced in Prometheus.
Signed-off-by: David Anderson <danderson@tailscale.com>
This enables the infrequent use of more complex Prometheus types, such as
timeseries with high/irregular label cardinality, without needing to
discover and implement generic abstracted type like LabelMap for each one.
Signed-off-by: David Anderson <danderson@tailscale.com>
Turns out, it's annoying to have to wait the entire interval
before getting any monitorable data, especially for very long
interval probes like hourly/daily checks.
Signed-off-by: David Anderson <danderson@tailscale.com>
Incidentally, simplify the go generate CI workflow, by
marking the dnsfallback update non-hermetic (so CI will
skip it) rather than manually filter it out of `go list`.
Updates #4194
Signed-off-by: David Anderson <danderson@tailscale.com>
This unbreaks some downstream users of tailscale who end up
with build errors from importing a v0 indirect dependency.
Signed-off-by: David Anderson <danderson@tailscale.com>
Salamanders also have no scales. I checked the interweb, and there
doesn't seem to be any subspecies that would let us claim that
*some* salamanders are scaley.
But they are tailey, for sure.
Signed-off-by: David Anderson <danderson@tailscale.com>
Mudpuppies are salamanders, and as such have tails but no scales.
The management apologizes for the error.
Signed-off-by: David Anderson <danderson@tailscale.com>
With this change, the client can obtain the initial handshake message
separately from the rest of the handshake, for embedding into another
protocol. This enables things like RTT reduction by stuffing the
handshake initiation message into an HTTP header.
Similarly, the server API optionally accepts a pre-read Noise initiation
message, in addition to reading the message directly off a net.Conn.
Updates #3488
Signed-off-by: David Anderson <danderson@tailscale.com>
Go 1.17 added a HandshakeContext func to take care of timeouts during
TLS handshaking, so switch from our homegrown goroutine implementation
to the standard way.
Signed-off-by: David Anderson <danderson@tailscale.com>
Cancelling the context makes the timeout goroutine race with the write that
reports a successful TLS handshake, so you can end up with a successful TLS
handshake that mysteriously reports that it timed out after ~0s in flight.
The context is always canceled and cleaned up as the function exits, which
happens mere microseconds later, so just let function exit clean up and
thereby avoid races.
Signed-off-by: David Anderson <danderson@tailscale.com>
Given our development cycle, we'll instead do big-bang updates
after every release, to give time for all the updates to soak in
unstable.
This does _not_ disable dependabot security-critical PRs.
Signed-off-by: David Anderson <danderson@tailscale.com>
Without this, enabling an exit node immediately blackholes all traffic,
but doesn't correctly let it flow to the exit node until the next netmap
update.
Fixes#3447
Signed-off-by: David Anderson <danderson@tailscale.com>
If IP forwarding is disabled globally, but enabled per-interface on all interfaces,
don't complain. If only some interfaces have forwarding enabled, warn that some
subnet routing/exit node traffic may not work.
Fixes#1586
Signed-off-by: David Anderson <danderson@tailscale.com>
It's a basic "deny everything" policy, since DERP's HTTP
server is very uninteresting from a browser POV. But it
stops every security scanner under the sun from reporting
"dangerously configured" HTTP servers.
Updates tailscale/corp#3119
Signed-off-by: David Anderson <danderson@tailscale.com>
Lets the systemd-resolved OSConfigurator report health changes
for out of band config resyncs.
Updates #3327
Signed-off-by: David Anderson <danderson@tailscale.com>
wgengine/wgcfg: introduce wgcfg.NewDevice helper to disable roaming
at all call sites (one real plus several tests).
Fixestailscale/corp#3016.
Signed-off-by: David Anderson <danderson@tailscale.com>
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
