Commit Graph

138 Commits (3b70968c256979a1f08ff02f716c058729ad69c2)

Author SHA1 Message Date
Brad Fitzpatrick fc160f80ee metrics: move currentFDs code to the metrics package
Updates #2784

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Brad Fitzpatrick 73280595a8 derp: accept dup clients without closing prior's connection
A public key should only have max one connection to a given
DERP node (or really: one connection to a node in a region).

But if people clone their machine keys (e.g. clone their VM, Raspbery
Pi SD card, etc), then we can get into a situation where a public key
is connected multiple times.

Originally, the DERP server handled this by just kicking out a prior
connections whenever a new one came. But this led to reconnect fights
where 2+ nodes were in hard loops trying to reconnect and kicking out
their peer.

Then a909d37a59 tried to add rate
limiting to how often that dup-kicking can happen, but empirically it
just doesn't work and ~leaks a bunch of goroutines and TCP
connections, tying them up for hour+ while more and more accumulate
and waste memory. Mostly because we were doing a time.Sleep forever
while not reading from their TCP connections.

Instead, just accept multiple connections per public key but track
which is the most recent. And if two both are writing back & forth,
then optionally disable them both. That last part is only enabled in
tests for now. The current default policy is just last-sender-wins
while we gather the next round of stats.

Updates #2751

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Brad Fitzpatrick ffd22050c0 derp: export current file descriptor metric 3 years ago
slowy07 ac0353e982 fix: typo spelling grammar
Signed-off-by: slowy07 <slowy.arfy@gmail.com>
3 years ago
Brad Fitzpatrick f35b8c3ead derp: fix meshing accounting edge case bug
If a peer is connected to multiple nodes in a region (so
multiForwarder is in use) and then a node restarts and re-sends all
its additions, this bug about whether an element is in the
multiForwarder could cause a one-time flip in the which peer node we
forward to.  Note a huge deal, but not written as intended.

Thanks to @lewgun for the bug report in #2141.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Brad Fitzpatrick fd7b738e5b derp: use pad32 package for padding, reduce duplication
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Brad Fitzpatrick 7298e777d4 derp: reduce server memory by 30% by removing persistent bufio.Writer
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Brad Fitzpatrick b622c60ed0 derp,wgengine/magicsock: don't assume stringer is in $PATH for go:generate
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Brad Fitzpatrick a909d37a59 derp: rate limit how often same-key clients can kick each other off server
Updates #392
Updates #506

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
Josh Bleecher Snyder 4dbbd0aa4a cmd/addlicense: add command to add licenseheaders to generated code
And use it to make our stringer invocations match the existing code.

Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
3 years ago
maddie d976a84d7e derp: allow self node when verifying clients
Fixes #2408

Signed-off-by: Maddie Zhan <maddie.zhan@cynovan.com>
3 years ago
Brad Fitzpatrick 4c0494185b derp: remove "fine for now" intentional slow memory leak from derp server
It was once believed that it might be useful. It wasn't. We never used it.

Remove it so we don't slowly leak memory.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
3 years ago
David Anderson d98829583a derp: use a dedicated queue for disco traffic.
Signed-off-by: David Anderson <danderson@tailscale.com>
3 years ago
David Anderson 67158549ab derp: actually export the new drop counter.
Signed-off-by: David Anderson <danderson@tailscale.com>
3 years ago
David Anderson 36492ace9d derp: add counters to track the type of dropped packets.
Signed-off-by: David Anderson <danderson@tailscale.com>
3 years ago
julianknodt 148602a89a derp,cmd/derper: allow server to verify clients
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>
3 years ago
julianknodt 3687e5352b derp: fix traffic handler peer addresses
Before it was using the local address and port, so fix that.
The fields in the response from `ss` are:

State, Recv-Q, Send-Q, Local Address:Port, Peer Address:Port, Process

Signed-off-by: julianknodt <julianknodt@gmail.com>
3 years ago
julianknodt 3728634af9 derp: add debug traffic handler
This adds a handler on the DERP server for logging bytes send and received by clients of the
server, by holding open a connection and recording if there is a difference between the number
of bytes sent and received. It sends a JSON marshalled object if there is an increase in the
number of bytes.

Signed-off-by: julianknodt <julianknodt@gmail.com>
3 years ago
julianknodt fe54721e31 derp: add pkt queue latency timer
It would be useful to know the time that packets spend inside of a queue before they are sent
off, as that can be indicative of the load the server is handling (and there was also an
existing TODO). This adds a simple exponential moving average metric to track the average packet
queue duration.
Changes during review:
Add CAS loop for recording queue timing w/ expvar.Func, rm snake_case, annotate in milliseconds,
convert

Signed-off-by: julianknodt <julianknodt@gmail.com>
3 years ago
David Anderson 54e6c3a290 version: use OSS repo's version when building.
When building with redo, also include the git commit hash
from the proprietary repo, so that we have a precise commit
that identifies all build info (including Go toolchain version).

Add a top-level build script demonstrating to downstream distros
how to burn the right information into builds.

Adjust `tailscale version` to print commit hashes when available.

Fixes #841.

Signed-off-by: David Anderson <danderson@tailscale.com>
4 years ago
Brad Fitzpatrick 169ff22a84 derp: set NotBefore and NotAfter in DERP server's metacert
Fixes regression from e415991256 that
only affected Windows users because Go only on Windows delegates x509
cert validation to the OS and Windows as unhappy with our "metacert"
lacking NotBefore and NotAfter.

Fixes #705
4 years ago
Brad Fitzpatrick e415991256 derp, derp/derphttp: remove one RTT from DERP setup
* advertise server's DERP public key following its ServerHello
* have client look for that DEPR public key in the response
  PeerCertificates
* let client advertise it's going into a "fast start" mode
  if it finds it
* modify server to support that fast start mode, just not
  sending the HTTP response header

Cuts down another round trip, bringing the latency of being able to
write our first DERP frame from SF to Bangalore from ~725ms
(3 RTT) to ~481ms (2 RTT: TCP and TLS).

Fixes #693

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
4 years ago
Brad Fitzpatrick 805850add9 derp: remove JSON struct tags in comments
They don't work in comments.

Added a test too to show that there's no change in behavior.
(It does case insensitive matching on parse anyway)
4 years ago
Brad Fitzpatrick 93ffc565e5 derp: remove protocol version 1 support
It hasn't existed for a long time and there are no current users.

Fixes #199
4 years ago
Josh Bleecher Snyder 062bd67d3b derp: use rand instead of crypto/rand to generate jitter
We don't need crypto/rand. Let the OS keep its entropy bits.

Signed-off-by: Josh Bleecher Snyder <josharian@gmail.com>
4 years ago
David Anderson 37c19970b3 derp: add a debug option to verbosely log drops to a destination.
Signed-off-by: David Anderson <danderson@tailscale.com>
4 years ago
Josh Bleecher Snyder 909c165382 derp: remove two key.Public allocations
Reading and writing a [32]byte key to a bufio.Reader/bufio.Writer
can easily by done without allocating. Do so.

It is slower; on my machine, it adds about 100ns per read/write.
However, the overall request takes a minimum of several µs,
and it cuts allocations meaningfully, so it is probably worth it.

name                      old time/op    new time/op    delta
SendRecv/msgsize=10-8       9.21µs ± 9%    9.08µs ± 8%     ~     (p=0.250 n=15+15)
SendRecv/msgsize=100-8      6.51µs ± 9%    6.60µs ± 7%     ~     (p=0.259 n=15+13)
SendRecv/msgsize=1000-8     7.24µs ±13%    7.61µs ±36%     ~     (p=1.000 n=11+15)
SendRecv/msgsize=10000-8    19.5µs ±15%    19.9µs ±25%     ~     (p=0.890 n=14+15)

name                      old speed      new speed      delta
SendRecv/msgsize=10-8     1.09MB/s ± 8%  1.10MB/s ± 8%     ~     (p=0.286 n=15+15)
SendRecv/msgsize=100-8    15.4MB/s ± 8%  15.1MB/s ± 6%     ~     (p=0.129 n=15+12)
SendRecv/msgsize=1000-8    139MB/s ±15%   135MB/s ±28%     ~     (p=1.000 n=11+15)
SendRecv/msgsize=10000-8   516MB/s ±17%   506MB/s ±21%     ~     (p=0.880 n=14+15)

name                      old alloc/op   new alloc/op   delta
SendRecv/msgsize=10-8         170B ± 1%      108B ± 1%  -36.63%  (p=0.000 n=15+15)
SendRecv/msgsize=100-8        265B ± 1%      203B ± 1%  -23.34%  (p=0.000 n=15+15)
SendRecv/msgsize=1000-8     1.18kB ± 1%    1.12kB ± 0%   -5.31%  (p=0.000 n=14+14)
SendRecv/msgsize=10000-8    18.8kB ± 2%    18.8kB ± 2%     ~     (p=0.443 n=12+12)

name                      old allocs/op  new allocs/op  delta
SendRecv/msgsize=10-8         4.00 ± 0%      2.00 ± 0%  -50.00%  (p=0.000 n=15+15)
SendRecv/msgsize=100-8        4.00 ± 0%      2.00 ± 0%  -50.00%  (p=0.000 n=15+15)
SendRecv/msgsize=1000-8       4.00 ± 0%      2.00 ± 0%  -50.00%  (p=0.000 n=15+15)
SendRecv/msgsize=10000-8      5.00 ± 0%      3.00 ± 0%  -40.00%  (p=0.000 n=13+14)

Signed-off-by: Josh Bleecher Snyder <josharian@gmail.com>
4 years ago
Brad Fitzpatrick 75225368a4 derp: fix 32-bit struct field alignment 4 years ago
David Anderson 15949ad77d derp: export the new expvar. 4 years ago
David Anderson 13661e195a derp: rename "wireguard" packet type to "other".
Strictly speaking, we don't know that it's a wireguard packet, just that
it doesn't look like a disco packet.

Signed-off-by: David Anderson <danderson@tailscale.com>
4 years ago
David Anderson 1b5b59231b derp: break down received packets by kind (disco vs. wireguard).
Signed-off-by: David Anderson <danderson@tailscale.com>
4 years ago
Josh Bleecher Snyder b23f2263c1 derp: add server version to /debug, expvars
This will make it easier for a human to tell what
version is deployed, for (say) correlating line numbers
in profiles or panics to corresponding source code.

It'll also let us observe version changes in prometheus.

Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
4 years ago
David Anderson c3994fd77c derp: remove OnlyDisco option.
Active discovery lets us introspect the state of the network stack precisely
enough that it's unnecessary, and dropping the initial DERP packets greatly
slows down tests. Additionally, it's unrealistic since our production network
will never deliver _only_ discovery packets, it'll be all or nothing.

Signed-off-by: David Anderson <danderson@tailscale.com>
4 years ago
Brad Fitzpatrick 52969bdfb0 derp: fix atomic padding on 32-bit again
Broken by earlier OnlyDisco addition.
4 years ago
Brad Fitzpatrick a6559a8924 wgengine/magicsock: run test DERP in mode where only disco packets allowed
So we don't accidentally pass a NAT traversal test by having DERP pick up our slack
when we really just wanted DERP as an OOB messaging channel.
4 years ago
Brad Fitzpatrick c1cabe75dc derp: fix server struct fielfd alignment on 32-bit
Mostly so the GitHub CI will pass on 32-bit.
4 years ago
Brad Fitzpatrick 4732722b87 derp: add frameClosePeer to move around clients within a region
For various reasons (mostly during rollouts or config changes on our
side), nodes may end up connecting to a fallback DERP node in a
region, rather than the primary one we tell them about in the DERP
map.

Connecting to the "wrong" node is fine, but it's in our best interest
for all nodes in a domain to connect to the same node, to reduce
intra-region packet forwarding.

This adds a privileged frame type used by the control system that can
kick off a client connection when they're connected to the wrong node
in a region. Then they hopefully reconnect immediately to the correct
location. (If not, we can leave them alone and stop closing them.)

Updates tailscale/corp#372
4 years ago
Brad Fitzpatrick dd43d9bc5f derp: fix varz typo
Updates tailscale/corp#391
4 years ago
Brad Fitzpatrick b87396b5d9 cmd/derper, derp: add some more varz and consistency check handler
I'm trying to hunt down a slow drift in numbers not agreeing.
4 years ago
Brad Fitzpatrick 6fbd1abcd3 derp: update peerGone code to work with regional DERP mesh clusters too
Updates #150
Updates #388
4 years ago
Brad Fitzpatrick 07ca0c1c29 derp: fix tracking problem if conn starts local, then also joins mesh peer 4 years ago
Brad Fitzpatrick 1cb7dab881 cmd/derper: support forwarding packets amongst set of peer DERP servers
Updates #388

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
4 years ago
Brad Fitzpatrick 4d599d194f derp, derp/derphttp: add key accessors, add Client.RecvDetail
Client.RecvDetail returns a connection generation so interested clients
can detect when a reconnect happened. (Will be needed for #388)
5 years ago
Brad Fitzpatrick b33c86b542 derp: add an unexported key.Public zero value variable to be less verbose 5 years ago
Brad Fitzpatrick b663ab4685 cmd/derper: treat self-connection connection watch as no-op
Updates #388
5 years ago
Brad Fitzpatrick 484b7fc9a3 derp, cmd/derper: add frameWatchConns, framePeerPresent for inter-DERP routing
This lets a trusted DERP client that knows a pre-shared key subscribe
to the connection list. Upon subscribing, they get the current set
of connected public keys, and then all changes over time.

This lets a set of DERP server peers within a region all stay connected to
each other and know which clients are connected to which nodes.

Updates #388

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Dmitry Adamushko f2c2d0de68 derp/derp_server: unregisterClient() for replaced client connections.
When unregistering a replaced client connection, move the
still-connected peers to the current client connecition. Inform
the peers that we've gone only when unregistering the active
client connection.

Signed-off-by: Dmitry Adamushko <da@stablebits.net>
5 years ago
Dmitry Adamushko 44434fdc82 derp/derp_server: fixed unbalanced {register,unregister}Client() calls.
Signed-off-by: Dmitry Adamushko <da@stablebits.net>
5 years ago
Brad Fitzpatrick 1ab5b31c4b derp, magicsock: send new "peer gone" frames when previous sender disconnects
Updates #150 (not yet enabled by default in magicsock)

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 8c4cef60f8 derp: wait for send goroutine to finish before returning from run
I saw a test flake due to the sender goroutine logging (ultimately to
t.Logf) after the server was closed.

This makes sure the all goroutines are cleaned up before Server.Close
returns.
5 years ago
Brad Fitzpatrick 521ad7b0fc derp: only flush writes to clients when we're out of things to write
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 0a25abcdaa derp: remove TODO comment
It was done in bf704a5218
5 years ago
Brad Fitzpatrick 8519a1b29f derp: revert rename of pkt to sendMsg
New plan, sorry.

Most reverts 77921a31b1, except some doc changes.
5 years ago
Brad Fitzpatrick c34b350efa derp: remove the mutex around and closing of send channel
Makes it less complicated.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 77921a31b1 derp: rename pkt to sendMsg, in prep for other types of writes
Updates #150
5 years ago
Brad Fitzpatrick 1453aecb44 derp: add sclient.done channel, simplify some context passing
This is mostly prep for a few future CLs, making sure we always have a
close-on-dead done channel available to select on when doing other
channel operations.
5 years ago
David Anderson ea90780066 derp: specify type of the by-reason drop varz.
Signed-off-by: David Anderson <dave@natulte.net>
5 years ago
Brad Fitzpatrick fd824df1fa derp: export metric for server's initial MemStats.Sys reading
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
David Anderson 64b5248929 derp: sprinkle comments and docstrings.
Signed-off-by: David Anderson <dave@natulte.net>
5 years ago
David Anderson 58b8ebd290 derp: add per-reason packet drop counter.
In parallel with the summed counter for now, for dashboard migration.

Signed-off-by: David Anderson <dave@natulte.net>
5 years ago
David Anderson 1e031001db derp: move write timeout to package constant.
Signed-off-by: David Anderson <dave@natulte.net>
5 years ago
David Anderson dbfc916273 derp: add short queues between reader and writer, drop on overload.
This avoids the server blocking on misbehaving or heavily contended
clients. We attempt to drop from the head of the queue to keep
overall queueing time lower.

Also:
 - fixes server->client keepalives, which weren't happening.
 - removes read rate-limiter, deferring instead to kernel-level
   global limiter/fair queuer.

Signed-off-by: David Anderson <dave@natulte.net>
5 years ago
David Crawshaw e838b3fb59 derp: use a write timeout when sending to clients
Signed-off-by: David Crawshaw <crawshaw@tailscale.com>
5 years ago
David Crawshaw 3df1b97ea8 derp: do not treat failure to relay as the fault of the sender
If Alice attempts to send a packet to Bob and the DERP server
encounters an error on the socket to Bob, we should not disconnect
Alice for that.

Signed-off-by: David Crawshaw <crawshaw@tailscale.com>
5 years ago
David Crawshaw 43aa8595dd derp: introduce Conn interface
This lets us test with something other than a net.Conn.

Signed-off-by: David Crawshaw <crawshaw@tailscale.com>
5 years ago
Brad Fitzpatrick 4fd29349b9 derp: add clients_replaced counter
Updates #109
5 years ago
Brad Fitzpatrick 5d8001d0ad derp: add varz for home moves
Updates #162
Updates #163 (maybe)

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick dd456f04c5 derp: actually record client's preferred bit 5 years ago
Brad Fitzpatrick 6978b93bdd derp, magicsock: track home (preferred) vs visiting connections for stats 5 years ago
Brad Fitzpatrick bf704a5218 derp: protocol negotiation, add v2: send src pub keys to clients in packets
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick e371520cc5 tsweb, derp: add expvar http.Handler for Prometheus's format
And add some opinions.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick d59d0576af derp: don't log on all dropped packets; super spammy
Signed-off-by: Brad Fitzpatrick <brad@danga.com>
5 years ago
Brad Fitzpatrick 752146a70f derp: clean up derphttp client code, use contexts
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 3317531021 derp: fix JSON field typo, sort fields to make it easier to see inconsistencies 5 years ago
Brad Fitzpatrick 6cd81d5d1f derp: add more derp stats for dropped and received packets
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 2612e54ad1 derp, cmd/derper: add debug handlers, stats
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 379a3125fd derp, wgengine/magicsock: support more than just packets from Client.Recv
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 88f1cc0c98 derp, cmd/derper: add rate limiting support, add default 5Mbps limit
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 322cb58b14 derp: deflake test I flaked up in earlier change
I broke an invariant in 11048b8932 (it was even nicely
documented then).

Also clean up the test a bit from while I was debugging it.

Fixes #84

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick f029c4c82d derp: change the protocol framing to always include a length
Addresses one of crawshaw's TODOs.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 11048b8932 derp: add docs on current protocol overview, break accept apart a bit
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 79af6054bf derp: don't start sending keepalives to client until after serverInfo sent
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 4b461c2e77 derp: rename some things in the server, add some docs 5 years ago
Brad Fitzpatrick 619697063e derp: more misc cleanups
And add an explicit derphttp.Client.Connect in the test now that it's
done lazily.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 259406e797 derp: move away from [32]byte key types
And some minor cleanup in the process.

Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick 2896be60db Move "logger" package to under types, now that we have it.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Brad Fitzpatrick a59dc5f155 Use logger.Logf consistently. It was used in most places.
Signed-off-by: Brad Fitzpatrick <bradfitz@tailscale.com>
5 years ago
Earl Lee a8d8b8719a Move Linux client & common packages into a public repo. 5 years ago