The resulting empty Prefs had AllowSingleHosts=false and
Routeall=false, so that on iOS if you did these steps:
- Login and leave running
- Terminate the frontend
- Restart the frontend (fast path restart, missing prefs)
- Set WantRunning=false
- Set WantRunning=true
...then you would have Tailscale running, but with no routes. You would
also accidentally disable the ExitNodeID/IP prefs (symptom: the current
exit node setting didn't appear in the UI), but since nothing
else worked either, you probably didn't notice.
The fix was easy enough. It turns out we already knew about the
problem, so this also fixes one of the BUG entries in state_test.
Fixes: #1918 (BUG-1) and some as-yet-unreported bugs with exit nodes.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Previously, there was no server round trip required to log out, so when
you asked ipnlocal to Logout(), it could clear the netmap immediately
and switch to NeedsLogin state.
In v1.8, we added a true Logout operation. ipn.Logout() would trigger
an async cc.StartLogout() and *also* immediately switch to NeedsLogin.
Unfortunately, some frontends would see NeedsLogin and immediately
trigger a new StartInteractiveLogin() operation, before the
controlclient auth state machine actually acted on the Logout command,
thus accidentally invalidating the entire logout operation, retaining
the netmap, and violating the user's expectations.
Instead, add a new LogoutFinished signal from controlclient
(paralleling LoginFinished) and, upon starting a logout, don't update
the ipn state machine until it's received.
Updates: #1918 (BUG-2)
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This code path is very tricky since it was originally designed for the
"re-authenticate to refresh my keys" use case, which didn't want to
lose the original session even if the refresh cycle failed. This is why
it acts differently from the Logout(); Login(); case.
Maybe that's too fancy, considering that it probably never quite worked
at all, for switching between users without logging out first. But it
works now.
This was more invasive than I hoped, but the necessary fixes actually
removed several other suspicious BUG: lines from state_test.go, so I'm
pretty confident this is a significant net improvement.
Fixestailscale/corp#1756.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
If the engine was shutting down from a previous session
(e.closing=true), it would return an error code when trying to get
status. In that case, ipnlocal would never unblock any callers that
were waiting on the status.
Not sure if this ever happened in real life, but I accidentally
triggered it while writing a test.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This is needed because the original opts.Prefs field was at some point
subverted for use in frontend->backend state migration for backward
compatibility on some platforms. We still need that feature, but we
also need the feature of providing the full set of prefs from
`tailscale up`, *not* including overwriting the prefs.Persist keys, so
we can't use the original field from `tailscale up`.
`tailscale up` had attempted to compensate for that by doing SetPrefs()
before Start(), but that violates the ipn.Backend contract, which says
you should call Start() before anything else (that's why it's called
Start()). As a result, doing SetPrefs({ControlURL=...,
WantRunning=true}) would cause a connection to the *previous* control
server (because WantRunning=true), and then connect to the *new*
control server only after running Start().
This problem may have been avoided before, but only by pure luck.
It turned out to be relatively harmless since the connection to the old
control server was immediately closed and replaced anyway, but it
created a race condition that could have caused spurious notifications
or rejected keys if the server responded quickly.
As already covered by existing TODOs, a better fix would be to have
Start() get out of the business of state migration altogether. But
we're approaching a release so I want to make the minimum possible fix.
Fixes#1840.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
There's no need to warn that it was not provided on the command line
after doing a sequence of up; logout; up --args. If you're asking for
tailscale to be up, you always mean that you prefer LoggedOut to become
false.
Fixes#1828
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Without this, macOS would fail to display its menu state correctly if you
started it while !WantRunning. It relies on the netmap in order to show
the logged-in username.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
There was logic that would make a "down" tailscale backend (ie.
!WantRunning) refuse to do any network activity. Unfortunately, this
makes the macOS and iOS UI unable to render correctly if they start
while !WantRunning.
Now that we have Prefs.LoggedOut, use that instead. So `tailscale down`
will still allow the controlclient to connect its authroutine, but
pause the maproutine. `tailscale logout` will entirely stop all
activity.
This new behaviour is not obviously correct; it's a bit annoying that
`tailsale down` doesn't terminate all activity like you might expect.
Maybe we should redesign the UI code to render differently when
disconnected, and then revert this change.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
EditPrefs should be just a wrapper around the action of changing prefs,
but someone had added a side effect of calling Login() sometimes. The
side effect happened *after* running the state machine, which would
sometimes result in us going into NeedsLogin immediately before calling
cc.Login().
This manifested as the macOS app not being able to Connect if you
launched it with LoggedOut=false and WantRunning=false. Trying to
Connect() would sent us to the NeedsLogin state instead.
Signed-off-by: Avery Pennarun <apenwarr@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>
A very long unit test that verifies the way the controlclient and
ipn.Backend interact.
This is a giant sequential test of the state machine. The test passes,
but only because it's asserting all the wrong behaviour. I marked all
the behaviour I think is wrong with BUG comments, and several
additional test opportunities with TODO.
Note: the new test supercedes TestStartsInNeedsLoginState, which was
checking for incorrect behaviour (although the new test still checks
for the same incorrect behaviour) and assumed .Start() would converge
before returning, which it happens to do, but only for this very
specific case, for the current implementation. You're supposed to wait
for the notifications.
Updates: tailscale/corp#1660
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
The old decay-based one took a while to converge. This new one (based
very loosely on TCP BBR) seems to converge quickly on what seems to be
the best speed.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This tries to generate traffic at a rate that will saturate the
receiver, without overdoing it, even in the event of packet loss. It's
unrealistically more aggressive than TCP (which will back off quickly
in case of packet loss) but less silly than a blind test that just
generates packets as fast as it can (which can cause all the CPU to be
absorbed by the transmitter, giving an incorrect impression of how much
capacity the total system has).
Initial indications are that a syscall about every 10 packets (TCP bulk
delivery) is roughly the same speed as sending every packet through a
channel. A syscall per packet is about 5x-10x slower than that.
The whole tailscale wireguard-go + magicsock + packet filter
combination is about 4x slower again, which is better than I thought
we'd do, but probably has room for improvement.
Note that in "full" tailscale, there is also a tundev read/write for
every packet, effectively doubling the syscall overhead per packet.
Given these numbers, it seems like read/write syscalls are only 25-40%
of the total CPU time used in tailscale proper, so we do have
significant non-syscall optimization work to do too.
Sample output:
$ GOMAXPROCS=2 go test -bench . -benchtime 5s ./cmd/tailbench
goos: linux
goarch: amd64
pkg: tailscale.com/cmd/tailbench
cpu: Intel(R) Core(TM) i7-4785T CPU @ 2.20GHz
BenchmarkTrivialNoAlloc/32-2 56340248 93.85 ns/op 340.98 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTrivialNoAlloc/124-2 57527490 99.27 ns/op 1249.10 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTrivialNoAlloc/1024-2 52537773 111.3 ns/op 9200.39 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTrivial/32-2 41878063 135.6 ns/op 236.04 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTrivial/124-2 41270439 138.4 ns/op 896.02 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTrivial/1024-2 36337252 154.3 ns/op 6635.30 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkBlockingChannel/32-2 12171654 494.3 ns/op 64.74 MB/s 0 %lost 1791 B/op 0 allocs/op
BenchmarkBlockingChannel/124-2 12149956 507.8 ns/op 244.17 MB/s 0 %lost 1792 B/op 1 allocs/op
BenchmarkBlockingChannel/1024-2 11034754 528.8 ns/op 1936.42 MB/s 0 %lost 1792 B/op 1 allocs/op
BenchmarkNonlockingChannel/32-2 8960622 2195 ns/op 14.58 MB/s 8.825 %lost 1792 B/op 1 allocs/op
BenchmarkNonlockingChannel/124-2 3014614 2224 ns/op 55.75 MB/s 11.18 %lost 1792 B/op 1 allocs/op
BenchmarkNonlockingChannel/1024-2 3234915 1688 ns/op 606.53 MB/s 3.765 %lost 1792 B/op 1 allocs/op
BenchmarkDoubleChannel/32-2 8457559 764.1 ns/op 41.88 MB/s 5.945 %lost 1792 B/op 1 allocs/op
BenchmarkDoubleChannel/124-2 5497726 1030 ns/op 120.38 MB/s 12.14 %lost 1792 B/op 1 allocs/op
BenchmarkDoubleChannel/1024-2 7985656 1360 ns/op 752.86 MB/s 13.57 %lost 1792 B/op 1 allocs/op
BenchmarkUDP/32-2 1652134 3695 ns/op 8.66 MB/s 0 %lost 176 B/op 3 allocs/op
BenchmarkUDP/124-2 1621024 3765 ns/op 32.94 MB/s 0 %lost 176 B/op 3 allocs/op
BenchmarkUDP/1024-2 1553750 3825 ns/op 267.72 MB/s 0 %lost 176 B/op 3 allocs/op
BenchmarkTCP/32-2 11056336 503.2 ns/op 63.60 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTCP/124-2 11074869 533.7 ns/op 232.32 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkTCP/1024-2 8934968 671.4 ns/op 1525.20 MB/s 0 %lost 0 B/op 0 allocs/op
BenchmarkWireGuardTest/32-2 1403702 4547 ns/op 7.04 MB/s 14.37 %lost 467 B/op 3 allocs/op
BenchmarkWireGuardTest/124-2 780645 7927 ns/op 15.64 MB/s 1.537 %lost 420 B/op 3 allocs/op
BenchmarkWireGuardTest/1024-2 512671 11791 ns/op 86.85 MB/s 0.5206 %lost 411 B/op 3 allocs/op
PASS
ok tailscale.com/wgengine/bench 195.724s
Updates #414.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This fixes the problem where, while running `redo version-info.sh`, the
repo would always show up as dirty, because redo creates a temp file
named *.tmp. This caused the version code to always have a -dirty tag,
but not when you run version.sh by hand.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Instead of reverting to 0.0.0, keep the same version number (eg. 1.2.4)
but add an extra suffix with the change count,
eg. 1.2.4-6-tb35d95ad7-gcb8be72e6. This avoids the problem where a
small patch causes the code to report a totally different version to
the server, which might change its behaviour based on version code.
(The server might enable various bug workarounds since it thinks
0.0.0 is very old.)
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This option isn't available on slightly older versions of git. We were
no longer using the real describe functionality anyway, so let's just do
something simpler to detect a dirty worktree.
While we're here, fix up a little bit of sh style.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
These accidentally make the tag syntax more flexible than was intended,
which will create forward compatibility problems later. Let's go back
to the old stricter parser.
Revert "cmd/tailscale/cli: fix double tag: prefix in tailscale up"
Revert "cmd/tailscale/cli, tailcfg: allow tag without "tag:" prefix in 'tailscale up'"
This reverts commit a702921620f7b6e386f393a9a1340d4218597469.
This reverts commit cd07437ade.
Affects #861.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Turns out for the particular error I was chasing, it actually returns
200 and zero data. But this code mirrors the same check in the map
poll, and is the right thing to do in the name of future debugging.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This function is only called in fake mode, which won't do anything more
with the packet after we respond to it anyway, so dropping it in the
prefilter is not necessary. And it's kinda semantically wrong: we did
not reject it, so telling the upper layer that it was rejected produces
an ugly error message.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
At some point faketun got implemented as a loopback (put a packet in
from wireguard, the same packet goes back to wireguard) which is not
useful. It's supposed to be an interface that just sinks all packets,
and then wgengine adds *only* and ICMP Echo responder as a layer on
top.
This caused extremely odd bugs on darwin, where the special case that
reinjects packets from local->local was filling the loopback channel
and creating an infinite loop (which became jammed since the reader and
writer were in the same goroutine).
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We need to emit Prefs when it *has* changed, not when it hasn't.
Test is added in our e2e test, separately.
Fixes: #620
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We were using the Go 'path' module, which apparently doesn't handle
backslashes correctly. path/filepath does.
However, the main bug turned out to be that we were not calling .Base()
on the path if version.ReadExe() fails, which it seems to do at least
on Windows 7. As a result, our logfile persistence was not working on
Windows, and logids would be regenerated on every restart.
Affects: #620
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We want to run bo.Backoff() after every upload, regardless. If
upload==true but err!=nil, we weren't backing off, which caused some
very-high-throughput log upload retries in bad network conditions.
Updates #282.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We'll use SO_BINDTODEVICE instead of fancy policy routing. This has
some limitations: for example, we will route all traffic through the
interface that has the main "default" (0.0.0.0/0) route, so machines
that have multiple physical interfaces might have to go through DERP to
get to some peers. But machines with multiple physical interfaces are
very likely to have policy routing (ip rule) support anyway.
So far, the only OS I know of that needs this feature is ChromeOS
(crostini). Fixes#245.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
* 'master' of github.com:tailscale/tailscale:
tailcfg: remove unused, unimplemented DERPNode.CertFingerprint for now
net/netns: also don't err on tailscaled -fake as a regular user
net/netcheck: fix HTTPS fallback bug from earlier today
net/netns: don't return an error if we're not root and running the tailscale binary
Otherwise iOS/macOS will reconfigure their routing every time anything
minor changes in the netmap (in particular, endpoints and DERP homes),
which is way too often.
Some users reported "network reconfigured" errors from Chrome when this
happens.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Specifically, this sequence:
iptables -N ts-forward
iptables -A ts-forward -m mark --mark 0x10000 -j ACCEPT
iptables -A FORWARD -j ts-forward
doesn't work on Debian-9-using-nftables, but this sequence:
iptables -N ts-forward
iptables -A FORWARD -j ts-forward
iptables -A ts-forward -m mark --mark 0x10000 -j ACCEPT
does work.
I'm sure the reason why is totally fascinating, but it's an old version
of iptables and the bug doesn't seem to exist on modern nftables, so
let's refactor our code to add rules in the always-safe order and
pretend this never happened.
Fixes#401.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
On startup, and when switching into =off and =nodivert, we were
deleting netfilter rules even if we weren't the ones that added them.
In order to avoid interfering with rules added by the sysadmin, we have
to be sure to delete rules only in the case that we added them in the
first place.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Instead of retrieving the list of chains, or the list of rules in a
chain, just try deleting the ones we don't want and then adding the
ones we do want. An error in flushing/deleting still means the rule
doesn't exist anymore, so there was no need to check for it first.
This avoids the need to parse iptables output, which avoids the need to
ever call iptables -S, which fixes#403, among other things. It's also
much more future proof in case the iptables command line changes.
Unfortunately the iptables go module doesn't properly pass the iptables
command exit code back up when doing .Delete(), so we can't correctly
check the exit code there. (exit code 1 really means the rule didn't
exist, rather than some other weird problem).
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This removes the use of suppress_ifgroup and fwmark "x/y" notation,
which are, among other things, not available in busybox and centos6.
We also use the return codes from the 'ip' program instead of trying to
parse its output.
I also had to remove the previous hack that routed all of 100.64.0.0/10
by default, because that would add the /10 route into the 'main' route
table instead of the new table 88, which is no good. It was a terrible
hack anyway; if we wanted to capture that route, we should have
captured it explicitly as a subnet route, not as part of the addr. Note
however that this change affects all platforms, so hopefully there
won't be any surprises elsewhere.
Fixes#405
Updates #320, #144
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Let's actually list the file we checked
(/proc/sys/net/ipv4/ip_forward). That gives the admin something
specific to look for when they get this message.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We would print a message about "nothing more to do", which some people
thought was an error or warning. Let's only print a message after
authenticating if we previously asked for interaction, and let's
shorten that message to just "Success," which is what it means.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Under some conditions, code would try to look things up in the maps
before the first call to updateLatency. I don't see any reason to delay
initialization of the maps, so let's just init them right away when
creating the Report instance.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
- Reformat the warning about a message being rate limited to print the
format string, rather than the formatted message. This helps give a
clue what "type" of message is being limited.
- Change the rate limit warning to be [RATE LIMITED] in all caps. This
uses less space on each line, plus is more noticeable.
- In tailscaled, change the frequency to be less often (once every 5
seconds per format string) but to allow bursts of up to 5 messages.
This greatly reduces the number of messages that are rate limited
during startup, but allows us to tighten the limit even further during
normal runtime.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This cuts RSS from ~30MB to ~20MB on my machine, after the previous fix
to get rid of unnecessary zstd buffers.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
The compressed blobs we send back and forth are small and infrequent,
which doesn't justify the 8MB * GOMAXPROCS memory that was being
allocated. This was the overwhelming majority of memory use in
tailscaled. On my system it goes from ~100M RSS to ~15M RSS (which is
still suspiciously high, but we can worry about that more later).
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This could happen when a process disappeared while we were reading its
file descriptor list.
I was able to replicate the problem by running this in another
terminal:
while :; do for i in $(seq 10); do
/bin/true & done >&/dev/null; wait >&/dev/null;
done
And then running the portlist tests thousands of times.
Fixes#339.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
By default, nothing differentiates errors or fatals from regular logs, so they just
blend into the rest of the logs.
As a bonus, if you run a test using t.Run(), the log messages printed
via the sub-t.Run() are printed at a different time from log messages
printed via the parent t.Run(), making debugging almost impossible.
This doesn't actually fix the test flake I'm looking for, but at least
I can find it in the logs now.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Our new build scripts try to build ipn-go-bridge on more than just
linux and darwin, so let's enable this file so it can be successful on
every platform.
This didn't catch anything yet, but it's good practice for detecting
goroutine leaks that we might not find otherwise.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
If a test calls log.Printf, 'go test' horrifyingly rearranges the
output to no longer be in chronological order, which makes debugging
virtually impossible. Let's stop that from happening by making
log.Printf panic if called from any module, no matter how deep, during
tests.
This required us to change the default error handler in at least one
http.Server, as well as plumbing a bunch of logf functions around,
especially in magicsock and wgengine, but also in logtail and backoff.
To add insult to injury, 'go test' also rearranges the output when a
parent test has multiple sub-tests (all the sub-test's t.Logf is always
printed after all the parent tests t.Logf), so we need to screw around
with a special Logf that can point at the "current" t (current_t.Logf)
in some places. Probably our entire way of using subtests is wrong,
since 'go test' would probably like to run them all in parallel if you
called t.Parallel(), but it definitely can't because the're all
manipulating the shared state created by the parent test. They should
probably all be separate toplevel tests instead, with common
setup/teardown logic. But that's a job for another time.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Inclusion of the word "assert" made it seem like a failure, even though
it was supposed to be identifying the name of the function (Assert()).
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
These will be used for dynamically changing the identity of a node, so
its ACL rights can be different from your own.
Note: Not all implemented yet on the server side, but we need this so
we can request the tagged rights in the first place.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This depends on improved support from the control server, to send the
new subnet width (Bits) fields. If these are missing, we fall back to
assuming their value is /32.
Conversely, if the server sends Bits fields to an older client, it will
interpret them as /32 addresses. Since the only rules we allow are
"accept" rules, this will be narrower or equal to the intended rule, so
older clients will simply reject hosts on the wider subnet (fail
closed).
With this change, the internal filter.Matches format has diverged
from the wire format used by controlclient, so move the wire format
into tailcfg and convert it to filter.Matches in controlclient.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
We were printing "Shields Up" when the netmap wasn't initialized yet,
which while technically effectively true, turned out to be confusing
when trying to debug things.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Some programs use frequent short-duration backoffs even under non-error
conditions. They can set this to avoid logging short backoffs when
things are operating normally, but still get messages when longer
backoffs kick in.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
When shields are up, no services are available to connect to, so hide
them all. This will also help them disappear from the UI menu on
other nodes.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
This sets a default packet filter that blocks all incoming requests,
giving end users more control over who can get into their machine, even
if the admin hasn't set any central ACLs.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Longer term, we should probably update the packet filter to be fully
stateful, for both TCP and ICMP. That is, only ICMP packets related to
a session *we* initiated should be allowed back in. But this is
reasonably secure for now, since wireguard is already trimming most
traffic. The current code would not protect against eg. Ping-of-Death style
attacks from VPN nodes.
Fixestailscale/tailscale#290.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
- Reset() was not including a Version field, so was getting rejected;
the Logout operation no longer happened when the client got disconnected.
- Don't crash if we can't decode 0-byte messages, which I suspect might
sometimes come through on EOF.
Signed-off-by: Avery Pennarun <apenwarr@tailscale.com>
Avery PennarunandBrad Fitzpatrick
