Switch deephash to use sha256x.Hash.
We add sha256x.HashString to efficiently hash a string.
It uses unsafe under the hood to convert a string to a []byte.
We also modify sha256x.Hash to export the underlying hash.Hash
for testing purposes so that we can intercept all hash.Hash calls.
Performance:
name old time/op new time/op delta
Hash-24 19.8µs ± 1% 19.2µs ± 1% -3.01% (p=0.000 n=10+10)
HashPacketFilter-24 2.61µs ± 0% 2.53µs ± 1% -3.01% (p=0.000 n=8+10)
HashMapAcyclic-24 31.3µs ± 1% 29.8µs ± 0% -4.80% (p=0.000 n=10+9)
TailcfgNode-24 1.83µs ± 1% 1.82µs ± 2% ~ (p=0.305 n=10+10)
HashArray-24 344ns ± 2% 323ns ± 1% -6.02% (p=0.000 n=9+10)
The performance gains is not as dramatic as sha256x over sha256 due to:
1. most of the hashing already occurring through the direct memory hashing logic, and
2. what does not go through direct memory hashing is slowed down by reflect.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Following the pattern elsewhere, we create a new tka-specific types package for the types
that need to couple between the serialized structure types, and tka.
Signed-off-by: Tom DNetto <tom@tailscale.com>
- A network-lock key is generated if it doesn't already exist, and stored in the StateStore. The public component is communicated to control during registration.
- If TKA state exists on the filesystem, a tailnet key authority is initialized (but nothing is done with it for now).
Signed-off-by: Tom DNetto <tom@tailscale.com>
This adds the inverse to CapabilityFileSharingSend so that senders can
identify who they can Taildrop to.
Updates #2101
Signed-off-by: Maisem Ali <maisem@tailscale.com>
3f686688a6 regressed the Windows beFirewallKillswitch code,
preventing the /firewall subprocess from running.
Fixestailscale/corp#6063
Change-Id: Ibd105759e5fecfeffc54f587f8ddcd0f1cbc4dca
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
And rewrite cloud detection to try to do only zero or one metadata
discovery request for all clouds, only doing a first (or second) as
confidence increases. Work remains for Windows, but a start.
And add Cloud to tailcfg.Hostinfo, which helped with testing using
"tailcfg debug hostinfo".
Updates #4983 (Linux only)
Updates #4984
Change-Id: Ib03337089122ce0cb38c34f724ba4b4812bc614e
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
And remove the GCP special-casing from ipn/ipnlocal; do it only in the
forwarder for *.internal.
Fixes#4980Fixes#4981
Change-Id: I5c481e96d91f3d51d274a80fbd37c38f16dfa5cb
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This does three things:
* If you're on GCP, it adds a *.internal DNS split route to the
metadata server, so we never break GCP DNS names. This lets people
have some Tailscale nodes on GCP and some not (e.g. laptops at home)
without having to add a Tailnet-wide *.internal DNS route.
If you already have such a route, though, it won't overwrite it.
* If the 100.100.100.100 DNS forwarder has nowhere to forward to,
it forwards it to the GCP metadata IP, which forwards to 8.8.8.8.
This means there are never errNoUpstreams ("upstream nameservers not set")
errors on GCP due to e.g. mangled /etc/resolv.conf (GCP default VMs
don't have systemd-resolved, so it's likely a DNS supremacy fight)
* makes the DNS fallback mechanism use the GCP metadata IP as a
fallback before our hosted HTTP-based fallbacks
I created a default GCP VM from their web wizard. It has no
systemd-resolved.
I then made its /etc/resolv.conf be empty and deleted its GCP
hostnames in /etc/hosts.
I then logged in to a tailnet with no global DNS settings.
With this, tailscaled writes /etc/resolv.conf (direct mode, as no
systemd-resolved) and sets it to 100.100.100.100, which then has
regular DNS via the metadata IP and *.internal DNS via the metadata IP
as well. If the tailnet configures explicit DNS servers, those are used
instead, except for *.internal.
This also adds a new util/cloudenv package based on version/distro
where the cloud type is only detected once. We'll likely expand it in
the future for other clouds, doing variants of this change for other
popular cloud environments.
Fixes#4911
RELNOTES=Google Cloud DNS improvements
Change-Id: I19f3c2075983669b2b2c0f29a548da8de373c7cf
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
We weren't wiring up netstack.Impl to the LocalBackend in some cases
on Windows. This fixes Windows 7 when run as a service.
Updates #4750 (fixes after pull in to corp repo)
Change-Id: I9ce51b797710f2bedfa90545776b7628c7528e99
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
We can't do Noise-over-HTTP in Wasm/JS (because we don't have bidirectional
communication), but we should be able to do it over WebSockets. Reuses
derp WebSocket support that allows us to turn a WebSocket connection
into a net.Conn.
Updates #3157
Signed-off-by: Mihai Parparita <mihai@tailscale.com>
AFAICT this isn't documented on MSDN, but based on the issue referenced below,
NRPT rules are not working when a rule specifies > 50 domains.
This patch modifies our NRPT rule generator to split the list of domains
into chunks as necessary, and write a separate rule for each chunk.
For compatibility reasons, we continue to use the hard-coded rule ID, but
as additional rules are required, we generate new GUIDs. Those GUIDs are
stored under the Tailscale registry path so that we know which rules are ours.
I made some changes to winutils to add additional helper functions in support
of both the code and its test: I added additional registry accessors, and also
moved some token accessors from paths to util/winutil.
Fixes https://github.com/tailscale/coral/issues/63
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
I wrote this code way back at the beginning of my tenure at Tailscale when we
had concerns about needing to restore deleted machine keys from backups.
We never ended up using this functionality, and the code is now getting in the
way, so we might as well remove it.
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
Enables the behavior described in the statepath flag, where if only
statedir is passed, then state is statedir/tailscaled.state.
Signed-off-by: James Tucker <james@tailscale.com>
Once a stop request is received and the service updates its status to `svc.StopPending`,
it should continue running *until the shutdown sequence is complete*, and then
return out of `(*ipnService).Execute`, which automatically sends a `svc.Stopped`
notification to Windows.
To make this happen, I changed the loop so that it runs until `doneCh` is
closed, and then returns. I also removed a spurious `svc.StopPending` notification
that the Windows Service Control Manager might be interpreting as a request for
more time to shut down.
Finally, I added some optional logging that sends a record of service notifications
to the Windows event log, allowing us to more easily correlate with any Service
Control Manager errors that are sent to the same log.
Change-Id: I5b596122e5e89c4c655fe747a612a52cb4e8f1e0
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
Well, goimports actually (which adds the normal import grouping order we do)
Change-Id: I0ce1b1c03185f3741aad67c14a7ec91a838de389
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The connections returned from SystemDial are automatically closed when
there is a major link change.
Also plumb through the dialer to the noise client so that connections
are auto-reset when moving from cellular to WiFi etc.
Updates #3363
Signed-off-by: Maisem Ali <maisem@tailscale.com>
No CLI support yet. Just the curl'able version if you know the peerapi
port. (like via a TSMP ping)
Updates #306
Change-Id: I0662ba6530f7ab58d0ddb24e3664167fcd1c4bcf
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
I've done this a handful of times in the past and again today.
Time to make it a supported thing for the future.
Used while debugging tailscale/corp#4559 (macsys CLI issues)
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This reverts commit 8d6793fd70.
Reason: breaks Android build (cgo/pthreads addition)
We can try again next cycle.
Change-Id: I5e7e1730a8bf399a8acfce546a6d22e11fb835d5
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Attempt to load the xt_mark kernel module when it is not present. If the
load fails, log error information.
It may be tempting to promote this failure to an error once it has been
in use for some time, so as to avoid reaching an error with the iptables
invocation, however, there are conditions under which the two stages may
disagree - this change adds more useful breadcrumbs.
Example new output from tailscaled running under my WSL2:
```
router: ensure module xt_mark: "/usr/sbin/modprobe xt_mark" failed: exit status 1; modprobe: FATAL: Module xt_mark not found in directory /lib/modules/5.10.43.3-microsoft-standard-WSL2
```
Background:
There are two places to lookup modules, one is `/proc/modules` "old",
the other is `/sys/module/` "new".
There was query_modules(2) in linux <2.6, alas, it is gone.
In a docker container in the default configuration, you would get
/proc/modules and /sys/module/ both populated. lsmod may work file,
modprobe will fail with EPERM at `finit_module()` for an unpriviliged
container.
In a priviliged container the load may *succeed*, if some conditions are
met. This condition should be avoided, but the code landing in this
change does not attempt to avoid this scenario as it is both difficult
to detect, and has a very uncertain impact.
In an nspawn container `/proc/modules` is populated, but `/sys/module`
does not exist. Modern `lsmod` versions will fail to gather most module
information, without sysfs being populated with module information.
In WSL2 modules are likely missing, as the in-use kernel typically is
not provided by the distribution filesystem, and WSL does not mount in a
module filesystem of its own. Notably the WSL2 kernel supports iptables
marks without listing the xt_mark module in /sys/module, and
/proc/modules is empty.
On a recent kernel, we can ask the capabilities system about SYS_MODULE,
that will help to disambiguate between the non-privileged container case
and just being root. On older kernels these calls may fail.
Update #4329
Signed-off-by: James Tucker <james@tailscale.com>
Remove the weird netstack -> tailssh dependency and instead have tailssh
register itself with ipnlocal when linked.
This makes tailssh.server a singleton, so we can have a global map of
all sessions.
Updates #3802
Change-Id: Iad5caec3a26a33011796878ab66b8e7b49339f29
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This extracts DOH mapping of known public DNS providers in
forwarder.go into its own package, to be consumed by other repos
Signed-off-by: Jenny Zhang <jz@tailscale.com>
Usage of userspace-networking is increasing, and the aggressive GC
tuning causes a significant reduction in performance in that mode.
Signed-off-by: James Tucker <james@tailscale.com>
In tracking down issue #4144 and reading through the netstack code in
detail, I discovered that the packet buf Clone path did not reset the
packetbuf it was getting from the sync.Pool. The fix was sent upstream
https://github.com/google/gvisor/pull/7385, and this bump pulls that in.
At this time there is no known path that this fixes, however at the time
of upstream submission this reset at least one field that could lead to
incorrect packet routing if exercised, a situation that could therefore
lead to an information leak.
Signed-off-by: James Tucker <james@tailscale.com>
In addition an envknob (TS_DEBUG_NETSTACK_LEAK_MODE) now provides access
to set leak tracking to more useful values.
Fixes#4309
Signed-off-by: James Tucker <james@tailscale.com>
While we rearrange/upstream things.
gliderlabs/ssh is forked into tempfork from our prior fork
at be8b7add40
x/crypto/ssh OTOH is forked at
https://github.com/tailscale/golang-x-crypto because it was gnarlier
to vendor with various internal packages, etc.
Its git history shows where it starts (2c7772ba30643b7a2026cbea938420dce7c6384d).
Updates #3802
Change-Id: I546e5cdf831cfc030a6c42557c0ad2c58766c65f
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The best flag to use on Win7 and Win8.0 is deprecated in Win8.1, so we resolve
the flag depending on OS version info.
Fixes https://github.com/tailscale/tailscale/issues/4201
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
Due to a bug in Go (golang/go#51778), cmd/go doesn't warn about your
Go version being older than the go.mod's declared Go version in that
case that package loading fails before the build starts, such as when
you use packages that are only in the current version of Go, like our
use of net/netip.
This change works around that Go bug by adding build tags and a
pre-Go1.18-only file that will cause Go 1.17 and earlier to fail like:
$ ~/sdk/go1.17/bin/go install ./cmd/tailscaled
# tailscale.com/cmd/tailscaled
./required_version.go:11:2: undefined: you_need_Go_1_18_to_compile_Tailscale
note: module requires Go 1.18
Change-Id: I39f5820de646703e19dde448dd86a7022252f75c
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The docs say:
Note that while correct uses of TryLock do exist, they are rare,
and use of TryLock is often a sign of a deeper problem in a particular use of mutexes.
Rare code! Or bad code! Who can tell!
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
It includes a fix to allow us to use Go 1.18.
We can now remove our Tailscale-only build tags.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
The certstore code is impacted by golang/go#51726.
The Tailscale Go toolchain fork contains a temporary workaround,
so it can compile it. Once the upstream toolchain can compile certstore,
presumably in Go 1.18.1, we can revert this change.
Note that depaware runs with the upstream toolchain.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
Still not sure the exact rules of how/when/who's supposed to set
these, but this works for now on making them match. Baby steps.
Will research more and adjust later.
Updates #4146 (but not enough to fix it, something's still wrong)
Updates #3802
Change-Id: I496d8cd7e31d45fe9ede88fc8894f35dc096de67
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
One of the current few steps to run Tailscale on gokrazy is to
specify the --tun=userspace-networking flag:
https://gokrazy.org/userguide/install/tailscale/
Instead, make it the default for now. Later we can change the
default to kernel mode if available and fall back to userspace
mode like Synology, once #391 is done.
Likewise, set default paths for Gokrazy, as its filesystem hierarchy
is not the Linux standard one. Instead, use the conventional paths as
documented at https://gokrazy.org/userguide/install/tailscale/.
Updates #1866
RELNOTE=default to userspace-networking mode on gokrazy
Change-Id: I3766159a294738597b4b30629d2860312dbb7609
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
If it's in a non-standard table, as it is on Unifi UDM Pro, apparently.
Updates #4038 (probably fixes, but don't have hardware to verify)
Change-Id: I2cb9a098d8bb07d1a97a6045b686aca31763a937
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Also move KubeStore and MemStore into their own package.
RELNOTE: tsnet now supports providing a custom ipn.StateStore.
Signed-off-by: Maisem Ali <maisem@tailscale.com>
Turns out we're pretty good already at init-time work in tailscaled.
The regexp/syntax shows up but it's hard to get rid of that; zstd even
uses regexp. *shrug*
Change-Id: I856aca056dcb7489f5fc22ef07f55f34ddf19bd6
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For ssh and maybe windows service babysitter later.
Updates #3802
Change-Id: I7492b98df98971b3fb72d148ba92c2276cca491f
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For local dev testing initially. Product-wise, it'll probably only be
workable on the two unsandboxed builds.
Updates #3802
Change-Id: Ic352f966e7fb29aff897217d79b383131bf3f92b
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Still largely incomplete, but in a better home now.
Updates #3802
Change-Id: I46c5ffdeb12e306879af801b06266839157bc624
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
I was about to add a third copy, so unify them now instead.
Change-Id: I3b93896aa1249b1250a6b1df4829d57717f2311a
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This patch adds new functions to be used when accessing system policies,
and revises callers to use the new functions. They first attempt the new
registry path for policies, and if that fails, attempt to fall back to the
legacy path.
We keep non-policy variants of these functions because we should be able to
retain the ability to read settings from locations that are not exposed to
sysadmins for group policy edits.
The remaining changes will be done in corp.
Updates https://github.com/tailscale/tailscale/issues/3584
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
Now that Go 1.17 has module graph pruning
(https://go.dev/doc/go1.17#go-command), we should be able to use
upstream netstack without breaking our private repo's build
that then depends on the tailscale.com Go module.
This is that experiment.
Updates #1518 (the original bug to break out netstack to own module)
Updates #2642 (this updates netstack, but doesn't remove workaround)
Change-Id: I27a252c74a517053462e5250db09f379de8ac8ff
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Disabled by default.
To use, run tailscaled with:
TS_SSH_ALLOW_LOGIN=you@bar.com
And enable with:
$ TAILSCALE_USE_WIP_CODE=true tailscale up --ssh=true
Then ssh [any-user]@[your-tailscale-ip] for a root bash shell.
(both the "root" and "bash" part are temporary)
Updates #3802
Change-Id: I268f8c3c95c8eed5f3231d712a5dc89615a406f0
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
A new package can also later record/report which knobs are checked and
set. It also makes the code cleaner & easier to grep for env knobs.
Change-Id: Id8a123ab7539f1fadbd27e0cbeac79c2e4f09751
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Treat UDP send EPERM errors as a lost UDP packet, not something super
fatal. That's just the Linux firewall preventing it from going out.
And add a leaf package net/neterror for that (and future) policy that
all three packages can share, with tests.
Updates #3619
Change-Id: Ibdb838c43ee9efe70f4f25f7fc7fdf4607ba9c1d
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This is for use by the Windows GUI client to log via when an
exit node is in use, so the logs don't go out via the exit node and
instead go directly, like tailscaled's. The dialer tried to do that
in the unprivileged GUI by binding to a specific interface, but the
"Internet Kill Switch" installed by tailscaled for exit nodes
precludes that from working and instead the GUI fails to dial out.
So, go through tailscaled (with a CONNECT request) instead.
Fixestailscale/corp#3169
Change-Id: I17a8efdc1d4b8fed53a29d1c19995592b651b215
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This moves the Windows-only initialization of the filelogger into
logpolicy. Previously we only did it when babysitting the tailscaled
subprocess, but this meant that log messages from the service itself
never made it to disk. Examples that weren't logged to disk:
* logtail unable to dial out,
* DNS flush messages from the service
* svc.ChangeRequest messages (#3581)
This is basically the same fix as #3571 but staying in the Logf type,
and avoiding build-tagged file (which wasn't quite a goal, but
happened and seemed nice)
Fixes#3570
Co-authored-by: Aaron Klotz <aaron@tailscale.com>
Change-Id: Iacd80c4720b7218365ec80ae143339d030842702
I broke it in 1.17.x sometime while rewiring some logs stuff,
mostly in 0653efb092 (but with a handful
of logs-related changes around that time)
Fixestailscale/corp#3265
Change-Id: Icb5c07412dc6d55f1d9244c5d0b51dceca6a7e34
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This is enough to handle the DNS queries as generated by Go's
net package (which our HTTP/SOCKS client uses), and the responses
generated by the ExitDNS DoH server.
This isn't yet suitable for putting on 100.100.100.100 where a number
of different DNS clients would hit it, as this doesn't yet do
EDNS0. It might work, but it's untested and likely incomplete.
Likewise, this doesn't handle anything about truncation, as the
exchanges are entirely in memory between Go or DoH. That would also
need to be handled later, if/when it's hooked up to 100.100.100.100.
Updates #3507
Change-Id: I1736b0ad31eea85ea853b310c52c5e6bf65c6e2a
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The caller of func run said:
// No need to log; the func already did
But that wasn't true. Some return paths didn't log.
So instead, return rich errors and have func main do the logging,
so we can't miss anything in the future.
Prior to this, safesocket.Listen for instance was causing tailscaled
to os.Exit(1) on failure without any clue as to why.
Change-Id: I9d71cc4d73d0fed4aa1b1902cae199f584f25793
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
To make ExitDNS cheaper.
Might not finish client-side support in December before 1.20, but at
least server support can start rolling out ahead of clients being
ready for it.
Tested with curl against peerapi.
Updates #1713
Change-Id: I676fed5fb1aef67e78c542a3bc93bddd04dd11fe
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
And simplify, unexport some tsdial/netstack stuff in the the process.
Fixes#3475
Change-Id: I186a5a5cbd8958e25c075b4676f7f6e70f3ff76e
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This starts to refactor tsdial.Dialer's name resolution to have
different stages: in-memory MagicDNS vs system resolution. A future
change will plug in ExitDNS resolution.
This also plumbs a Dialer into netstack and unexports the dnsMap
internals.
And it removes some of the async AddNetworkMapCallback usage and
replaces it with synchronous updates of the Dialer's netmap
from LocalBackend, since the LocalBackend has the Dialer too.
Updates #3475
Change-Id: Idcb7b1169878c74f0522f5151031ccbc49fe4cb4
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
In prep for moving stuff out of LocalBackend.
Change-Id: I9725aa9c3ebc7275f8c40e040b326483c0340127
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Not done yet, but this move more of the outbound dial special casing
from random packages into tsdial, which aspires to be the one unified
place for all outbound dialing shenanigans.
Then this plumbs it all around, so everybody is ultimately
holding on to the same dialer.
As of this commit, macOS/iOS using an exit node should be able to
reach to the exit node's DoH DNS proxy over peerapi, doing the sockopt
to stay within the Network Extension.
A number of steps remain, including but limited to:
* move a bunch more random dialing stuff
* make netstack-mode tailscaled be able to use exit node's DNS proxy,
teaching tsdial's resolver to use it when an exit node is in use.
Updates #1713
Change-Id: I1e8ee378f125421c2b816f47bc2c6d913ddcd2f5
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For now this just deletes the net/socks5/tssocks implementation (and
the DNSMap stuff from wgengine/netstack) and moves it into net/tsdial.
Then initialize a Dialer early in tailscaled, currently only use for the
outbound and SOCKS5 proxies. It will be plumbed more later. Notably, it
needs to get down into the DNS forwarder for exit node DNS forwading
in netstack mode. But it will also absorb all the peerapi setsockopt
and netns Dial and tlsdial complexity too.
Updates #1713
Change-Id: Ibc6d56ae21a22655b2fa1002d8fc3f2b2ae8b6df
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
When this happens, it is incredibly noisy in the logs.
It accounts for about a third of all remaining
"unexpected" log lines from a recent investigation.
It's not clear that we know how to fix this,
we have a functioning workaround,
and we now have a (cheap and efficient) metric for this
that we can use for measurements.
So reduce the logging to approximately once per minute.
Signed-off-by: Josh Bleecher Snyder <josh@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>
And annotate magicsock as a start.
And add localapi and debug handlers with the Prometheus-format
exporter.
Updates #3307
Change-Id: I47c5d535fe54424741df143d052760387248f8d3
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
More work towards removing the massive ipnserver.Run and ipnserver.Options
and making composable pieces.
Work remains. (The getEngine retry loop on Windows complicates things.)
For now some duplicate code exists. Once the Windows side is fixed
to either not need the retry loop or to move the retry loop into a
custom wgengine.Engine wrapper, then we can unify tailscaled_windows.go
too.
Change-Id: If84d16e3cd15b54ead3c3bb301f27ae78d055f80
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
github.com/go-multierror/multierror served us well.
But we need a few feature from it (implement Is),
and it's not worth maintaining a fork of such a small module.
Instead, I did a clean room implementation inspired by its API.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
Using temporary netlink fork in github.com/tailscale/netlink until we
get the necessary changes upstream in either vishvananda/netlink
or jsimonetti/rtnetlink.
Updates #391
Change-Id: I6e1de96cf0750ccba53dabff670aca0c56dffb7c
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Even if not in use. We plan to use it for more stuff later.
(not for iOS or macOS-GUIs yet; only tailscaled)
Change-Id: Idaef719d2a009be6a39f158fd8f57f8cca68e0ee
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Temporary measure until we switch to Go 1.18.
$ go run ./cmd/tailscale version
1.17.0-date.20211022
go version: go1.17
Updates #81
Change-Id: Ic82ebffa5f46789089e5fb9810b3f29e36a47f1a
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
utils/winutil/vss contains just enough COM wrapping to query the Volume Shadow Copy service for snapshots.
WalkSnapshotsForLegacyStateDir is the friendlier interface that adds awareness of our actual use case,
mapping the snapshots and locating our legacy state directory.
Updates #3011
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
From https://github.com/tailscale/tailscale/pull/1919 with
edits by bradfitz@.
This change introduces a new storage provider for the state file. It
allows users to leverage AWS SSM parameter store natively within
tailscaled, like:
$ tailscaled --state=arn:aws:ssm:eu-west-1:123456789:parameter/foo
Known limitations:
- it is not currently possible to specific a custom KMS key ID
RELNOTE=tailscaled on Linux supports using AWS SSM for state
Edits-By: Brad Fitzpatrick <bradfitz@tailscale.com>
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Signed-off-by: Maxime VISONNEAU <maxime.visonneau@gmail.com>
This feature wasn't working until I realized that we also need to opt into
the events. MSDN wasn't so generous as to make this easy to deduce.
Updates #2956
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
The old name invited confusion:
* is this the HTTP proxy to use ourselves? (no, that's
via an environment variable, per proxy conventions)
* is this for LetsEncrypt https-to-localhost-http
proxying? (no, that'll come later)
So rename to super verbose --outbound-http-proxy-listen
before the 1.16.0 release to make it clear what it is.
It listens (serves) and it's for outbound, not inbound.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For the service, all we need to do is handle the `svc.SessionChange` command.
Upon receipt of a `windows.WTS_SESSION_UNLOCK` event, we fire off a goroutine to flush the DNS cache.
(Windows expects responses to service requests to be quick, so we don't want to do that synchronously.)
This is gated on an integral registry value named `FlushDNSOnSessionUnlock`,
whose value we obtain during service initialization.
(See [this link](https://docs.microsoft.com/en-us/windows/win32/api/winsvc/nc-winsvc-lphandler_function_ex) for information re: handling `SERVICE_CONTROL_SESSIONCHANGE`.)
Fixes#2956
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
This adds support for tailscaled to be an HTTP proxy server.
It shares the same backend dialing code as the SOCK5 server, but the
client protocol is HTTP (including CONNECT), rather than SOCKS.
Fixes#2289
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This was already possible on Linux if you ran tailscaled with --debug
(which runs net/http/pprof), but it requires the user have the Go
toolchain around.
Also, it wasn't possible on macOS, as there's no way to run the IPNExtension
with a debug server (it doesn't run tailscaled).
And on Windows it's super tedious: beyond what users want to do or
what we want to explain.
Instead, put it in "tailscale debug" so it works and works the same on
all platforms. Then we can ask users to run it when we're debugging something
and they can email us the output files.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
There are two reasons this can't ever go to actual logs,
but rewrite it to make it happy.
Fixestailscale/corp#2695
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
ProgramData has a permissive ACL. For us to safely store machine-wide
state information, we must set a more restrictive ACL on our state directory.
We set the ACL so that only talescaled's user (ie, LocalSystem) and the
Administrators group may access our directory.
We must include Administrators to ensure that logs continue to be easily
accessible; omitting that group would force users to use special tools to
log in interactively as LocalSystem, which is not ideal.
(Note that the ACL we apply matches the ACL that was used for LocalSystem's
AppData\Local).
There are two cases where we need to reset perms: One is during migration
from the old location to the new. The second case is for clean installations
where we are creating the file store for the first time.
Updates #2856
Signed-off-by: Aaron Klotz <aaron@tailscale.com>
The earlier 382b349c54 was too late,
as engine creation itself needed to listen on things.
Fixes#2827
Updates #2822
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
* Revert "Revert "types/key: add MachinePrivate and MachinePublic.""
This reverts commit 61c3b98a24.
Signed-off-by: David Anderson <danderson@tailscale.com>
* types/key: add ControlPrivate, with custom serialization.
ControlPrivate is just a MachinePrivate that serializes differently
in JSON, to be compatible with how the Tailscale control plane
historically serialized its private key.
Signed-off-by: David Anderson <danderson@tailscale.com>
Plumb throughout the codebase as a replacement for the mixed use of
tailcfg.MachineKey and wgkey.Private/Public.
Signed-off-by: David Anderson <danderson@tailscale.com>
Instead of using the legacy codepath, teach discoEndpoint to handle
peers that have a home DERP, but no disco key. We can still communicate
with them, but only over DERP.
Signed-off-by: David Anderson <danderson@tailscale.com>
Allows FreeBSD to function as an exit node in the same way
that Windows and Tailscaled-on-MacOS do.
RELNOTE=FreeBSD can now function as an exit node.
Fixes https://github.com/tailscale/tailscale/issues/2498
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
rsc.io/goversion is really expensive.
Running version.ReadExe on tailscaled on darwin
allocates 47k objects, almost 11mb.
All we want is the module info. For that, all we need to do
is scan through the binary looking for the magic start/end strings
and then grab the bytes in between them.
We can do that easily and quickly with nothing but a 64k buffer.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
Still very much a prototype (hard-coded IPs, etc) but should be
non-invasive enough to submit at this point and iterate from here.
Updates #2589
Co-Author: David Crawshaw <crawshaw@tailscale.com>
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Prior to Tailscale 1.12 it detected UPnP on any port.
Starting with Tailscale 1.11.x, it stopped detecting UPnP on all ports.
Then start plumbing its discovered Location header port number to the
code that was assuming port 5000.
Fixes#2109
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
For testing pfSense clients "behind" pfSense on Digital Ocean where
the main interface still exists. This is easier for debugging.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This is a simplified rate limiter geared for exactly our needs:
A fast, mono.Time-based rate limiter for use in tstun.
It was generated by stripping down the x/time/rate rate limiter
to just our needs and switching it to use mono.Time.
It removes one time.Now call per packet.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
magicsock makes multiple calls to Now per packet.
Move to mono.Now. Changing some of the calls to
use package mono has a cascading effect,
causing non-per-packet call sites to also switch.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
There's a call to Now once per packet.
Move to mono.Now.
Though the current implementation provides high precision,
we document it to be coarse, to preserve the ability
to switch to a coarse monotonic time later.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
The previous algorithm used a map of all visited pointers.
The strength of this approach is that it quickly prunes any nodes
that we have ever visited before. The detriment of the approach
is that pruning is heavily dependent on the order that pointers
were visited. This is especially relevant for hashing a map
where map entries are visited in a non-deterministic manner,
which would cause the map hash to be non-deterministic
(which defeats the point of a hash).
This new algorithm uses a stack of all visited pointers,
similar to how github.com/google/go-cmp performs cycle detection.
When we visit a pointer, we push it onto the stack, and when
we leave a pointer, we pop it from the stack.
Before visiting a pointer, we first check whether the pointer exists
anywhere in the stack. If yes, then we prune the node.
The detriment of this approach is that we may hash a node more often
than before since we do not prune as aggressively.
The set of visited pointers up until any node is only the
path of nodes up to that node and not any other pointers
that may have been visited elsewhere. This provides us
deterministic hashing regardless of visit order.
We can now delete hashMapFallback and associated complexity,
which only exists because the previous approach was non-deterministic
in the presence of cycles.
This fixes a failure of the old algorithm where obviously different
values are treated as equal because the pruning was too aggresive.
See https://github.com/tailscale/tailscale/issues/2443#issuecomment-883653534
The new algorithm is slightly slower since it prunes less aggresively:
name old time/op new time/op delta
Hash-8 66.1µs ± 1% 68.8µs ± 1% +4.09% (p=0.000 n=19+19)
HashMapAcyclic-8 63.0µs ± 1% 62.5µs ± 1% -0.76% (p=0.000 n=18+19)
TailcfgNode-8 9.79µs ± 2% 9.88µs ± 1% +0.95% (p=0.000 n=19+17)
HashArray-8 643ns ± 1% 653ns ± 1% +1.64% (p=0.000 n=19+19)
However, a slower but more correct algorithm seems
more favorable than a faster but incorrect algorithm.
Signed-off-by: Joe Tsai <joetsai@digital-static.net>
Prep for #1591 which will need to make Linux's router react to changes
that the link monitor observes.
The router package already depended on the monitor package
transitively. Now it's explicit.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
To unify the Windows service and non-service/non-Windows paths a bit.
And provides a way to make Linux act like Windows for testing.
(notably, for testing the fix to #2137)
One perhaps visible change of this is that tailscaled.exe when run in
cmd.exe/powershell (not as a Windows Service) no longer uses the
"_daemon" autostart key. But in addition to being naturally what falls
out of this change, that's also what Windows users would likely want,
as otherwise the unattended mode user is ignored when the "_daemon"
autostart key is specified. Notably, this would let people debug what
their normally-run-as-a-service tailscaled is doing, even when they're
running in Unattended Mode.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add in UPnP portmapping, using goupnp library in order to get the UPnP client and run the
portmapping functions. This rips out anywhere where UPnP used to be in portmapping, and has a
flow separate from PMP and PCP.
RELNOTE=portmapper now supports UPnP mappings
Fixes#682
Updates #2109
Signed-off-by: julianknodt <julianknodt@gmail.com>
This adapts the existing in-process logcatcher from tstest/integration
into a public type and uses it on the side of testcontrol. This also
fixes a bug in the Alpine Linux OpenRC unit that makes every value in
`/etc/default/tailscaled` exported into tailscaled's environment, a-la
systemd [Service].EnviromentFile.
Signed-off-by: Christine Dodrill <xe@tailscale.com>
After allowing for custom DERP maps, it's convenient to be able to see their latency in
netcheck. This adds a query to the local tailscaled for the current DERPMap.
Updates #1264
Signed-off-by: julianknodt <julianknodt@gmail.com>
Move derpmap.Prod to a static JSON file (go:generate'd) instead,
to make its role explicit. And add a TODO about making dnsfallback
use an update-over-time DERP map file instead of a baked-in one.
Updates #1264
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This adds a flag to the DERP server which specifies to verify clients through a local
tailscaled. It is opt-in, so should not affect existing clients, and is mainly intended for
users who want to run their own DERP servers. It assumes there is a local tailscaled running and
will attempt to hit it for peer status information.
Updates #1264
Signed-off-by: julianknodt <julianknodt@gmail.com>
The only connectivity an AWS Lambda container has is an IPv4 link-local
169.254.x.x address using NAT:
12: vtarget_1@if11: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500
qdisc noqueue state UP group default qlen 1000
link/ether 7e:1c:3f:00:00:00 brd ff:ff:ff:ff:ff:ff link-netnsid 1
inet 169.254.79.1/32 scope global vtarget_1
valid_lft forever preferred_lft forever
If there are no other IPv4/v6 addresses available, and we are running
in AWS Lambda, allow IPv4 169.254.x.x addresses to be used.
----
Similarly, a Google Cloud Run container's only connectivity is
a Unique Local Address fddf:3978:feb1:d745::c001/128.
If there are no other addresses available then allow IPv6
Unique Local Addresses to be used.
We actually did this in an earlier release, but now refactor it to
work the same way as the IPv4 link-local support is being done.
Signed-off-by: Denton Gentry <dgentry@tailscale.com>
The dependency is a "soft" ordering dependency only, meaning that
tailscaled will start after those services if those services were
going to be run anyway, but doesn't force either of them to run.
That's why it's safe to specify this dependency unconditionally,
even for systems that don't run those services.
Updates #2127.
Signed-off-by: David Anderson <danderson@tailscale.com>
Alpine Linux[1] is a minimal Linux distribution built around musl libc.
It boots very quickly, requires very little ram and is as close as you
can get to an ideal citizen for testing Tailscale on musl. Alpine has a
Tailscale package already[2], but this patch also makes it easier for us
to provide an Alpine Linux package off of pkgs in the future.
Alpine only offers Tailscale on the rolling-release edge branch.
[1]: https://alpinelinux.org/
[2]: https://pkgs.alpinelinux.org/packages?name=tailscale&branch=edge
Updates #1988
Signed-off-by: Christine Dodrill <xe@tailscale.com>
The cyolosecurity fork of certstore did not update its module name and
thus can only be used with a replace directive. This interferes with
installing using `go install` so I created a tailscale fork with an
updated module name.
Signed-off-by: Adrian Dewhurst <adrian@tailscale.com>
Pull in the latest version of wireguard-windows.
Switch to upstream wireguard-go.
This requires reverting all of our import paths.
Unfortunately, this has to happen at the same time.
The wireguard-go change is very low risk,
as that commit matches our fork almost exactly.
(The only changes are import paths, CI files, and a go.mod entry.)
So if there are issues as a result of this commit,
the first place to look is wireguard-windows changes.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
At the start of a dev cycle we'll upgrade all dependencies.
Done with:
$ for Dep in $(cat go.mod | perl -ne '/(\S+) v/ and print "$1\n"'); do go get $Dep@upgrade; done
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Yes, it printed, but that was an implementation detail for hashing.
And coming optimization will make it print even less.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
- Switch to our own simpler token bucket, since x/time/rate is missing
necessary stuff (can't provide your own time func; can't check the
current bucket contents) and it's overkill anyway.
- Add tests that actually include advancing time.
- Don't remove the rate limit on a message until there's enough room to
print at least two more of them. When we do, we'll also print how
many we dropped, as a contextual reminder that some were previously
lost. (This is more like how the Linux kernel does it.)
- Reformat the [RATE LIMITED] messages to be shorter, and to not
corrupt original message. Instead, we print the message, then print
its format string.
- Use %q instead of \"%s\", for more accurate parsing later, if the
format string contained quotes.
Fixes#1772
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We had two separate code paths for the initial UDP listener bind
and any subsequent rebinds.
IPv6 got left out of the rebind code.
Rather than duplicate it there, unify the two code paths.
Then improve the resulting code:
* Rebind had nested listen attempts to try the user-specified port first,
and then fall back to :0 if that failed. Convert that into a loop.
* Initial bind tried only the user-specified port.
Rebind tried the user-specified port and 0.
But there are actually three ports of interest:
The one the user specified, the most recent port in use, and 0.
We now try all three in order, as appropriate.
* In the extremely rare case in which binding to port 0 fails,
use a dummy net.PacketConn whose reads block until close.
This will keep the wireguard-go receive func goroutine alive.
As a pleasant side-effect of this, if we decide that
we need to resuscitate #1796, it will now be much easier.
Fixes#1799
Co-authored-by: David Anderson <danderson@tailscale.com>
Signed-off-by: Josh Bleecher Snyder <josharian@gmail.com>
Assume it'll stay at 0 forever, so hard-code it
and delete code conditional on it being non-0.
Signed-off-by: Josh Bleecher Snyder <josharian@gmail.com>
NetworkManager fixed the bug that forced us to use NetworkManager
if it's programming systemd-resolved, and in the same release also
made NetworkManager ignore DNS settings provided for unmanaged
interfaces... Which breaks what we used to do. So, with versions
1.26.6 and above, we MUST NOT use NetworkManager to indirectly
program systemd-resolved, but thankfully we can talk to resolved
directly and get the right outcome.
Fixes#1788
Signed-off-by: David Anderson <danderson@tailscale.com>
With this change, the ipnserver's safesocket.Listen (the localhost
tcp.Listen) happens right away, before any synchronous
TUN/DNS/Engine/etc setup work, which might be slow, especially on
early boot on Windows.
Because the safesocket.Listen starts up early, that means localhost
TCP dials (the safesocket.Connect from the GUI) complete successfully
and thus the GUI avoids the MessageBox error. (I verified that
pacifies it, even without a Listener.Accept; I'd feared that Windows
localhost was maybe special and avoided the normal listener backlog).
Once the GUI can then connect immediately without errors, the various
timeouts then matter less, because the backend is no longer trying to
race against the GUI's timeout. So keep retrying on errors for a
minute, or 10 minutes if the system just booted in the past 10
minutes.
This should fix the problem with Windows 10 desktops auto-logging in
and starting the Tailscale frontend which was then showing a
MessageBox error about failing to connect to tailscaled, which was
slow coming up because the Windows networking stack wasn't up
yet. Fingers crossed.
Fixes#1313 (previously #1187, etc)
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This change implements Windows version of install-system-daemon and
uninstall-system-daemon subcommands. When running the commands the
user will install or remove Tailscale Windows service.
Updates #1232
Signed-off-by: Alex Brainman <alex.brainman@gmail.com>
Logout used to be a no-op, so the ipnserver previously synthensized a Logout
on disconnect. Now that Logout actually invalidates the node key that was
forcing all GUI closes to log people out.
Instead, add a method to LocalBackend to specifically mean "the
Windows GUI closed, please forget all the state".
Fixestailscale/corp#1591 (ignoring the notification issues, tracked elsewhere)
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
They were scattered/duplicated in misc places before.
It can't be in the client package itself for circular dep reasons.
This new package is basically tailcfg but for localhost
communications, instead of to control.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This is usually the same as the requested interface, but on some
unixes can vary based on device number allocation, and on Windows
it's the GUID instead of the pretty name, since everything relating
to configuration wants the GUID.
Signed-off-by: David Anderson <danderson@tailscale.com>
Upstream wireguard-go has changed its receive model.
NewDevice now accepts a conn.Bind interface.
The conn.Bind is stateless; magicsock.Conns are stateful.
To work around this, we add a connBind type that supports
cheap teardown and bring-up, backed by a Conn.
The new conn.Bind allows us to specify a set of receive functions,
rather than having to shoehorn everything into ReceiveIPv4 and ReceiveIPv6.
This lets us plumbing DERP messages directly into wireguard-go,
instead of having to mux them via ReceiveIPv4.
One consequence of the new conn.Bind layer is that
closing the wireguard-go device is now indistinguishable
from the routine bring-up and tear-down normally experienced
by a conn.Bind. We thus have to explicitly close the magicsock.Conn
when the close the wireguard-go device.
One downside of this change is that we are reliant on wireguard-go
to call receiveDERP to process DERP messages. This is fine for now,
but is perhaps something we should fix in the future.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
The bool was already called useNetstack at the caller.
isUserspace (to mean netstack) is confusing next to wgengine.NewUserspaceEngine, as that's
a different type of 'userspace'.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
The resolver still only supports a single upstream config, and
ipn/wgengine still have to split up the DNS config, but this moves
closer to unifying the DNS configs.
As a handy side-effect of the refactor, IPv6 MagicDNS records exist
now.
Signed-off-by: David Anderson <danderson@tailscale.com>
For discovery when an explicit hostname/IP is known. We'll still
also send it via control for finding peers by a list.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This makes setup more explicit in prod codepaths, without
requiring a bunch of arguments or helpers for tests and
userspace mode.
Signed-off-by: David Anderson <danderson@tailscale.com>
control/controlclient: sign RegisterRequest
Some customers wish to verify eligibility for devices to join their
tailnets using machine identity certificates. TLS client certs could
potentially fulfill this role but the initial customer for this feature
has technical requirements that prevent their use. Instead, the
certificate is loaded from the Windows local machine certificate store
and uses its RSA public key to sign the RegisterRequest message.
There is room to improve the flexibility of this feature in future and
it is currently only tested on Windows (although Darwin theoretically
works too), but this offers a reasonable starting place for now.
Updates tailscale/coral#6
Signed-off-by: Adrian Dewhurst <adrian@tailscale.com>
So we can empty import the guts of cmd/tailscaled from another
module for go mod tidy reasons.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Add proto to flowtrack.Tuple.
Add types/ipproto leaf package to break a cycle.
Server-side ACL work remains.
Updates #1516
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
This reverts the revert commit 84aba349d9.
And changes us to use inet.af/netstack.
Updates #1518
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
gVisor fixed their google/gvisor#1446 so we can include gVisor mode
on 32-bit machines.
A few minor upstream API changes, as normal.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Part of overall effort to clean up, unify, use link monitoring more,
and make Tailscale quieter when all networks are down. This is especially
bad on macOS where we can get killed for not being polite it seems.
(But we should be polite in any case)
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Gets it out of wgengine so the Engine isn't responsible for being a
callback registration hub for it.
This also removes the Engine.LinkChange method, as it's no longer
necessary. The monitor tells us about changes; it doesn't seem to
need any help. (Currently it was only used by Swift, but as of
14dc790137 we just do the same from Go)
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
Joe Tsai