After reducing UDP stack traversal overhead via GSO and GRO,
runtime.chanrecv() began to account for a high percentage (20% in one
environment) of perf samples during a throughput benchmark. The
individual packet channel ops with the crypto goroutines was the primary
contributor to this overhead.
Updating these channels to pass vectors, which the device package
already handles at its ends, reduced this overhead substantially, and
improved throughput.
The iperf3 results below demonstrate the effect of this commit between
two Linux computers with i5-12400 CPUs. There is roughly ~13us of round
trip latency between them.
The first result is with UDP GSO and GRO, and with single element
channels.
Starting Test: protocol: TCP, 1 streams, 131072 byte blocks
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 12.3 GBytes 10.6 Gbits/sec 232 3.15 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
Test Complete. Summary Results:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 12.3 GBytes 10.6 Gbits/sec 232 sender
[ 5] 0.00-10.04 sec 12.3 GBytes 10.6 Gbits/sec receiver
The second result is with channels updated to pass a slice of
elements.
Starting Test: protocol: TCP, 1 streams, 131072 byte blocks
[ ID] Interval Transfer Bitrate Retr Cwnd
[ 5] 0.00-10.00 sec 13.2 GBytes 11.3 Gbits/sec 182 3.15 MBytes
- - - - - - - - - - - - - - - - - - - - - - - - -
Test Complete. Summary Results:
[ ID] Interval Transfer Bitrate Retr
[ 5] 0.00-10.00 sec 13.2 GBytes 11.3 Gbits/sec 182 sender
[ 5] 0.00-10.04 sec 13.2 GBytes 11.3 Gbits/sec receiver
Reviewed-by: Adrian Dewhurst <adrian@tailscale.com>
Signed-off-by: Jordan Whited <jordan@tailscale.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Accept packet vectors for reading and writing in the tun.Device and
conn.Bind interfaces, so that the internal plumbing between these
interfaces now passes a vector of packets. Vectors move untouched
between these interfaces, i.e. if 128 packets are received from
conn.Bind.Read(), 128 packets are passed to tun.Device.Write(). There is
no internal buffering.
Currently, existing implementations are only adjusted to have vectors
of length one. Subsequent patches will improve that.
Also, as a related fixup, use the unix and windows packages rather than
the syscall package when possible.
Co-authored-by: James Tucker <james@tailscale.com>
Signed-off-by: James Tucker <james@tailscale.com>
Signed-off-by: Jordan Whited <jordan@tailscale.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Before, the code attached a finalizer to an object that wasn't returned,
resulting in immediate garbage collection. Instead return the actual
pointer.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Without this, we wind up freeing packets that the encryption/decryption
queues still have, resulting in a UaF.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
The immediate motivation for this change is an observed deadlock.
1. A goroutine calls peer.Stop. That calls peer.queue.Lock().
2. Another goroutine is in RoutineSequentialReceiver.
It receives an elem from peer.queue.inbound.
3. The peer.Stop goroutine calls close(peer.queue.inbound),
close(peer.queue.outbound), and peer.stopping.Wait().
It blocks waiting for RoutineSequentialReceiver
and RoutineSequentialSender to exit.
4. The RoutineSequentialReceiver goroutine calls peer.SendStagedPackets().
SendStagedPackets attempts peer.queue.RLock().
That blocks forever because the peer.Stop
goroutine holds a write lock on that mutex.
A background motivation for this change is that it can be expensive
to have a mutex in the hot code path of RoutineSequential*.
The mutex was necessary to avoid attempting to send elems on a closed channel.
This commit removes that danger by never closing the channel.
Instead, we send a sentinel nil value on the channel to indicate
to the receiver that it should exit.
The only problem with this is that if the receiver exits,
we could write an elem into the channel which would never get received.
If it never gets received, it cannot get returned to the device pools.
To work around this, we use a finalizer. When the channel can be GC'd,
the finalizer drains any remaining elements from the channel and
restores them to the device pool.
After that change, peer.queue.RWMutex no longer makes sense where it is.
It is only used to prevent concurrent calls to Start and Stop.
Move it to a more sensible location and make it a plain sync.Mutex.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
We have a bunch of stupid channel tricks, and I'm about to add more.
Give them their own file. This commit is 100% code movement.
Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>
