open.source/magic-wormhole-transit-relay
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Compare commits

base: open.source:master
Branches Tags
open.source:master
open.source:no-pypiwin32
open.source:docs
open.source:rewrite
open.source:tests
open.source:0.2.1
open.source:0.2.0
open.source:0.1.2
open.source:0.1.1
open.source:0.1.0
..
compare: open.source:rewrite
Branches Tags
open.source:master
open.source:no-pypiwin32
open.source:docs
open.source:rewrite
open.source:tests
open.source:0.2.1
open.source:0.2.0
open.source:0.1.2
open.source:0.1.1
open.source:0.1.0

2 Commits
master ... rewrite

Author SHA1 Message Date
Brian Warner
790f29d4ba WIP db 2017-11-04 12:40:54 -07:00
Brian Warner
d7800f6337 rewrite connection handling, not sure it's a good idea 2017-11-04 12:23:32 -07:00
38 changed files with 984 additions and 3308 deletions
Show Stats Expand all files Collapse all files

55
.appveyor.yml
View File

@ -1,55 +0,0 @@
# adapted from https://packaging.python.org/en/latest/appveyor/
environment:
# we tell Tox to use "twisted[windows]", to get pypiwin32 installed
#TWISTED_EXTRAS: "[windows]"
# that didn't work (it seems to work when I run it locally, but on appveyor
# it fails to install the pypiwin32 package). So don't bother telling
# Twisted to support windows: just install it ourselves.
# EXTRA_DEPENDENCY: "pypiwin32"
matrix:
# For Python versions available on Appveyor, see
# http://www.appveyor.com/docs/installed-software#python
- PYTHON: "C:\\Python27"
- PYTHON: "C:\\Python27-x64"
DISTUTILS_USE_SDK: "1"
- PYTHON: "C:\\Python35"
- PYTHON: "C:\\Python36"
- PYTHON: "C:\\Python36-x64"
install:
- |
%PYTHON%\python.exe -m pip install wheel tox
# note:
# %PYTHON% has: python.exe
# %PYTHON%\Scripts has: pip.exe, tox.exe (and others installed by bare pip)
build: off
test_script:
# Put your test command here.
# Note that you must use the environment variable %PYTHON% to refer to
# the interpreter you're using - Appveyor does not do anything special
# to put the Python evrsion you want to use on PATH.
- |
misc\windows-build.cmd %PYTHON%\Scripts\tox.exe -e py
after_test:
# This step builds your wheels.
# Again, you only need build.cmd if you're building C extensions for
# 64-bit Python 3.3/3.4. And you need to use %PYTHON% to get the correct
# interpreter
- |
misc\windows-build.cmd %PYTHON%\python.exe setup.py bdist_wheel
artifacts:
# bdist_wheel puts your built wheel in the dist directory
- path: dist\*
#on_success:
# You can use this step to upload your artifacts to a public website.
# See Appveyor's documentation for more details. Or you can simply
# access your wheels from the Appveyor "artifacts" tab for your build.

35
.github/workflows/test.yml vendored
View File

@ -1,35 +0,0 @@
name: Tests
on:
push:
branches: [ master ]
pull_request:
branches: [ master ]
jobs:
testing:
runs-on: ubuntu-latest
strategy:
matrix:
python-version: [3.7, 3.8, 3.9]
steps:
- uses: actions/checkout@v2
- name: Set up Python
uses: actions/setup-python@v2
with:
python-version: ${{ matrix.python-version }}
- name: Install dependencies
run: |
python -m pip install --upgrade pip tox codecov
tox --notest -e coverage
- name: Test
run: |
python --version
tox -e coverage
- name: Upload Coverage
run: codecov

28
.travis.yml
View File

@ -1,17 +1,18 @@
arch:
- amd64
- ppc64le
language: python
# defaults: the py3.7 environment overrides these
dist: trusty
sudo: false
language: python
cache: pip
before_cache:
- rm -f $HOME/.cache/pip/log/debug.log
branches:
except:
- /^WIP-.*$/
python:
- "2.7"
- "3.3"
- "3.4"
- "3.5"
- "3.6"
- "nightly"
install:
- pip install -U pip tox virtualenv codecov
before_script:
@ -24,15 +25,6 @@ script:
after_success:
- codecov
matrix:
include:
- python: 2.7
- python: 3.5
- python: 3.6
- python: 3.7
# we don't actually need sudo, but that kicks us onto GCE, which lets
# us get xenial
sudo: true
dist: xenial
- python: nightly
allow_failures:
- python: nightly
- python: "3.3"
- python: "nightly"

5
MANIFEST.in
View File

@ -3,6 +3,5 @@ include src/wormhole_transit_relay/_version.py
include LICENSE README.md NEWS.md
recursive-include docs *.md *.rst *.dot
include .coveragerc tox.ini
include misc/*.py
include misc/munin/wormhole_transit*
include misc/munin/wormhole_transit
include misc/munin/wormhole_transit_alltime

34
NEWS.md
View File

@ -1,37 +1,5 @@
User-visible changes in "magic-wormhole-transit-relay":
## unreleased
* drop Python 2, Python 3.5 and 3.6 support
## Release 0.2.1 (11-Sep-2019)
* listen on IPv4+IPv6 properly (#12)
## Release 0.2.0 (10-Sep-2019)
* listen on IPv4+IPv6 socket by default (#12)
* enable SO_KEEPALIVE on all connections (#9)
* drop support for py3.3 and py3.4
* improve munin plugins
## Release 0.1.2 (19-Mar-2018)
* Allow more simultaneous connections, by increasing the rlimits() ceiling at
startup
* Improve munin plugins
* Get tests working on Windows
## Release 0.1.1 (14-Feb-2018)
Improve logging and munin graphing tools: previous version would count bad
handshakes twice (once as "errory", and again as "lonely"). The munin plugins
have been renamed.
## Release 0.1.0 (12-Nov-2017)
Initial release. Forked from magic-wormhole-0.10.3 (12-Sep-2017).
## forked from magic-wormhole-0.10.3 (12-Sep-2017)

20
README.md
View File

@ -1,19 +1,13 @@
# magic-wormhole-transit-relay
[![PyPI](http://img.shields.io/pypi/v/magic-wormhole-transit-relay.svg)](https://pypi.python.org/pypi/magic-wormhole-transit-relay)
![Tests](https://github.com/magic-wormhole/magic-wormhole-transit-relay/workflows/Tests/badge.svg)
[![codecov.io](https://codecov.io/github/magic-wormhole/magic-wormhole-transit-relay/coverage.svg?branch=master)](https://codecov.io/github/magic-wormhole/magic-wormhole-transit-relay?branch=master)
[![Build Status](https://travis-ci.org/warner/magic-wormhole-transit-relay.svg?branch=master)](https://travis-ci.org/warner/magic-wormhole-transit-relay)
[![codecov.io](https://codecov.io/github/warner/magic-wormhole-transit-relay/coverage.svg?branch=master)](https://codecov.io/github/warner/magic-wormhole-transit-relay?branch=master)
Transit Relay server for Magic-Wormhole
This repository implements the Magic-Wormhole "Transit Relay", a server that
helps clients establish bulk-data transit connections even when both are
behind NAT boxes. Each side makes a TCP connection to this server and
presents a handshake. Two connections with identical handshakes are glued
together, allowing them to pretend they have a direct connection.
This server used to be included in the magic-wormhole repository, but was
split out into a separate repo to aid deployment and development.
See docs/running.md for instructions to launch the server.
This repo implements the Magic-Wormhole "Transit Relay", a server that helps
clients establish bulk-data transit connections even when both are behind NAT
boxes. Each side makes a TCP connection to this server and presents a
handshake. Two connections with identical handshakes are glued together,
allowing them to pretend they have a direct connection.

54
client.py
View File

@ -1,54 +0,0 @@
"""
This is a test-client for the transit-relay that uses TCP. It
doesn't send any data, only prints out data that is received. Uses a
fixed token of 64 'a' characters. Always connects on localhost:4001
"""
from twisted.internet import endpoints
from twisted.internet.defer import (
Deferred,
)
from twisted.internet.task import react
from twisted.internet.error import (
ConnectionDone,
)
from twisted.internet.protocol import (
Protocol,
Factory,
)
class RelayEchoClient(Protocol):
"""
Speaks the version1 magic wormhole transit relay protocol (as a client)
"""
def connectionMade(self):
print(">CONNECT")
self.data = b""
self.transport.write(u"please relay {}\n".format(self.factory.token).encode("ascii"))
def dataReceived(self, data):
print(">RECV {} bytes".format(len(data)))
print(data.decode("ascii"))
self.data += data
if data == "ok\n":
self.transport.write("ding\n")
def connectionLost(self, reason):
if isinstance(reason.value, ConnectionDone):
self.factory.done.callback(None)
else:
print(">DISCONNCT: {}".format(reason))
self.factory.done.callback(reason)
@react
def main(reactor):
ep = endpoints.clientFromString(reactor, "tcp:localhost:4001")
f = Factory.forProtocol(RelayEchoClient)
f.token = "a" * 64
f.done = Deferred()
ep.connect(f)
return f.done

91
docs/logging.md
View File

@ -1,91 +0,0 @@
# Usage Logs
The transit relay does not emit or record any logging by default. By adding
option flags to the twist/twistd command line, you can enable one of two
different kinds of logs.
To avoid collecting information which could later be used to correlate
clients with external network traces, logged information can be "blurred".
This reduces the resolution of the data, retaining enough to answer questions
about how much the server is being used, but discarding fine-grained
timestamps or exact transfer sizes. The ``--blur-usage=`` option enables
this, and it takes an integer value (in seconds) to specify the desired time
window.
## Logging JSON Upon Each Connection
If --log-fd is provided, a line will be written to the given (numeric) file
descriptor after each connection is done. These events could be delivered to
a comprehensive logging system like XXX for offline analysis.
Each line will be a complete JSON object (starting with ``{``, ending with
``}\n``, and containing no internal newlines). The keys will be:
* ``started``: number, seconds since epoch
* ``total_time``: number, seconds from open to last close
* ``waiting_time``: number, seconds from start to 2nd side appearing, or null
* ``total_bytes``: number, total bytes relayed (sum of both directions)
* ``mood``: string, one of: happy, lonely, errory
A mood of ``happy`` means both sides gave a correct handshake. ``lonely``
means a second matching side never appeared (and thus ``waiting_time`` will
be null). ``errory`` means the first side gave an invalid handshake.
If --blur-usage= is provided, then ``started`` will be rounded to the given
time interval, and ``total_bytes`` will be rounded to a fixed set of buckets:
* file sizes less than 1MB: rounded to the next largest multiple of 10kB
* less than 1GB: multiple of 1MB
* 1GB or larger: multiple of 100MB
## Usage Database
If --usage-db= is provided, the server will maintain a SQLite database in the
given file. Current, recent, and historical usage data will be written to the
database, and external tools can query the DB for metrics: the munin plugins
in misc/ may be useful. Timestamps and sizes in this file will respect
--blur-usage. The four tables are:
``current`` contains a single row, with these columns:
* connected: number of paired connections
* waiting: number of not-yet-paired connections
* partal_bytes: bytes transmitted over not-yet-complete connections
``since_reboot`` contains a single row, with these columns:
* bytes: sum of ``total_bytes``
* connections: number of completed connections
* mood_happy: count of connections that finished "happy": both sides gave correct handshake
* mood_lonely: one side gave good handshake, other side never showed up
* mood_errory: one side gave a bad handshake
``all_time`` contains a single row, with these columns:
* bytes:
* connections:
* mood_happy:
* mood_lonely:
* mood_errory:
``usage`` contains one row per closed connection, with these columns:
* started: seconds since epoch, rounded to "blur time"
* total_time: seconds from first open to last close
* waiting_time: seconds from first open to second open, or None
* bytes: total bytes relayed (in both directions)
* result: (string) the mood: happy, lonely, errory
All tables will be updated after each connection is finished. In addition,
the ``current`` table will be updated at least once every 5 minutes.
## Logfiles for twistd
If daemonized by twistd, the server will write ``twistd.pid`` and
``twistd.log`` files as usual. By default ``twistd.log`` will only contain
startup, shutdown, and exception messages.
Setting ``--log-fd=1`` (file descriptor 1 is always stdout) will cause the
per-connection JSON lines to be interleaved with any messages sent to
Twisted's logging system. It may be better to use a different file
descriptor.

159
docs/running.md
View File

@ -1,159 +1,8 @@
# Running the Transit Relay
First off, you probably don't need to run a relay. The ``wormhole`` command,
as shipped from magic-wormhole.io, is configured to use a default Transit
Relay operated by the author of Magic-Wormhole. This can be changed with the
``--transit-helper=`` argument, and other applications that import the
Wormhole library might point elsewhere.
The only reasons to run a separate relay are:
* You are a kind-hearted server admin who wishes to support the project by
paying the bandwidth costs incurred by your friends, who you instruct in
the use of ``--transit-helper=``.
* You publish a different application, and want to provide your users with a
relay that fails at different times than the official one
## Installation
To run a transit relay, first you need an environment to install it.
* create a virtualenv
* ``pip install magic-wormhole-transit-relay`` into this virtualenv
# How to Run the Transit Relay
```
% virtualenv tr-venv
...
% tr-venv/bin/pip install magic-wormhole-transit-relay
...
pip install magic-wormhole-transit-relay
twist wormhole-transit-relay --port tcp:4001
```
## Running
The transit relay is not a standalone program: rather it is a plugin for the
Twisted application-running tools named ``twist`` (which only runs in the
foreground) and ``twistd`` (which daemonizes). To run the relay for testing,
use something like this:
```
% tr-venv/bin/twist transitrelay [ARGS]
2017-11-09T17:07:28-0800 [-] not blurring access times
2017-11-09T17:07:28-0800 [-] Transit starting on 4001
2017-11-09T17:07:28-0800 [wormhole_transit_relay.transit_server.Transit#info] Starting factory <wormhole_transit_relay.transit_server.Transit object at 0x7f01164b4550>
...
```
The relevant arguments are:
* ``--port=``: the endpoint to listen on, like ``tcp:4001``
* ``--log-fd=``: writes JSON lines to the given file descriptor for each connection
* ``--usage-db=``: maintains a SQLite database with current and historical usage data
* ``--blur-usage=``: round logged timestamps and data sizes
For WebSockets support, two additional arguments:
* ``--websocket``: the endpoint to listen for websocket connections
on, like ``tcp:4002``
* ``--websocket-url``: the URL of the WebSocket connection. This may
be different from the listening endpoint because of port-forwarding
and so forth. By default it will be ``ws://localhost:<port>`` if not
provided
When you use ``twist``, the relay runs in the foreground, so it will
generally exit as soon as the controlling terminal exits. For persistent
environments, you should daemonize the server.
## Minimizing Log Data
The server code attempts to strike a balance between minimizing data
collected about users, and recording enough information to manage the server
and monitor its operation. The standard `twistd.log` file does not record IP
addresses unless an error occurs. The optional `--log-fd=` file (and the
SQLite database generated if `--usage-db=` is enabled) record the time at
which the first side connected, the time until the second side connected, the
total transfer time, the total number of bytes transferred, and the
success/failure status (the "mood").
If `--blur-usage=` is provided, these recorded file sizes are rounded down:
sizes less than 1kB are recorded as 0, sizes up to 1MB are rounded to the
nearest kB, sizes up to 1GB are rounded to the nearest MB, and sizes above
1GB are rounded to the nearest 100MB.
The argument to `--blur-usage=` is treated as a number of seconds, and the
"first side connects" timestamp is rounded to a multiple of this. For
example, `--blur-usage=3600` means all timestamps are rounded down to the
nearest hour. The waiting time and total time deltas are recorded without
rounding.
## Daemonization
A production installation will want to daemonize the server somehow. One
option is to use ``twistd`` (the daemonizing version of ``twist``). This
takes the same plugin name and arguments as ``twist``, but forks into the
background, detaches from the controlling terminal, and writes all output
into a logfile:
```
% tr-venv/bin/twistd transitrelay [ARGS]
% cat twistd.log
2017-11-09T17:07:28-0800 [-] not blurring access times
2017-11-09T17:07:28-0800 [-] Transit starting on 4001
2017-11-09T17:07:28-0800 [wormhole_transit_relay.transit_server.Transit#info] Starting factory <wormhole_transit_relay.transit_server.Transit object at 0x7f01164b4550>
...
% cat twistd.pid; echo
18985
```
To shut down a ``twistd``-based server, you'll need to look in the
``twistd.pid`` file for the process id, and kill it:
```
% kill `cat twistd.pid`
```
To start the server each time the host reboots, you might use a crontab
"@reboot" job, or a systemd unit.
Another option is to run ``twist`` underneath a daemonization tool like
``daemontools`` or ``start-stop-daemon``. Since ``twist`` is just a regular
program, this leaves the daemonization tool in charge of issues like
restarting a process that exits unexpectedly, limiting the rate of
respawning, and switching to the correct user-id and base directory.
Packagers who create an installable transit-relay server package should
choose a suitable daemonization tool that matches the practices of the target
operating system. For example, Debian/Ubuntu packages should probably include
a systemd unit that runs ``twist transitrelay`` in some
``/var/run/magic-wormhole-transit-relay/`` directory.
Production environments that want to monitor the server for capacity
management can use the ``--log-fd=`` option to emit logs, then route those
logs into a suitable analysis tool. Other environments might be content to
use ``--usage-db=`` and run the included Munin plugins to monitor usage.
There is also a
[Dockerfile](https://github.com/ggeorgovassilis/magic-wormhole-transit-relay-docker),
written by George Georgovassilis, which you might find useful.
## Configuring Clients
The transit relay will listen on an "endpoint" (usually a TCP port, but it
could be a unix-domain socket or any other Endpoint that Twisted knows how to
listen on). By default this is ``tcp:4001``. The relay does not know what
hostname or IP address might point at it.
Clients are configured with a "Transit Helper" setting that includes both the
hostname and the port number, like the default
``tcp:transit.magic-wormhole.io:4001``. The standard ``wormhole`` tool takes
a ``--transit-helper=`` argument to override this. Other applications that
use ``wormhole`` as a library will have internal means to configure which
transit relay they use.
If you run your own transit relay, you will need to provide the new settings
to your clients for it to be used.
The standard ``wormhole`` tool is used by two sides: the sender and the
receiver. Both sides exchange their configured transit relay with their
partner. So if the sender overrides ``--transit-helper=`` but the receiver
does not, they might wind up using either relay server, depending upon which
one gets an established connection first.
The relay runs as a twist/twistd plugin. To

200
docs/transit.md
View File

@ -1,4 +1,4 @@
# Transit Protocol
= Transit Protocol =
The Transit protocol is responsible for establishing an encrypted
bidirectional record stream between two programs. It must be given a "transit
@ -9,11 +9,109 @@ The protocol tries hard to create a **direct** connection between the two
ends, but if that fails, it uses a centralized relay server to ferry data
between two separate TCP streams (one to each client).
This repository provides that centralized relay server. For details of the
protocol spoken by the clients, and the client-side API, please see
``transit.md`` in the magic-wormhole repository.
The current implementation starts with the following:
## Relay
* detect all of the host's IP addresses
* listen on a random TCP port
* offers the (address,port) pairs as hints
The other side will attempt to connect to each of those ports, as well as
listening on its own socket. After a few seconds without success, they will
both connect to a relay server.
== Roles ==
The Transit protocol has pre-defined "Sender" and "Receiver" roles (unlike
Wormhole, which is symmetric/nobody-goes-first). Each connection must have
exactly one Sender and exactly one Receiver.
The connection itself is bidirectional: either side can send or receive
records. However the connection establishment mechanism needs to know who is
in charge, and the encryption layer needs a way to produce separate keys for
each side..
This may be relaxed in the future, much as Wormhole was.
== Records ==
Transit establishes a **record-pipe**, so the two sides can send and receive
whole records, rather than unframed bytes. This is a side-effect of the
encryption (which uses the NaCl "secretbox" function). The encryption adds 44
bytes of overhead to each record (4-byte length, 24-byte nonce, 32-byte MAC),
so you might want to use slightly larger records for efficiency. The maximum
record size is 2^32 bytes (4GiB). The whole record must be held in memory at
the same time, plus its ciphertext, so very large ciphertexts are not
recommended.
Transit provides **confidentiality**, **integrity**, and **ordering** of
records. Passive attackers can only do the following:
* learn the size and transmission time of each record
* learn the sending and destination IP addresses
In addition, an active attacker is able to:
* delay delivery of individual records, while maintaining ordering (if they
delay record #4, they must delay #5 and later as well)
* terminate the connection at any time
If either side receives a corrupted or out-of-order record, they drop the
connection. Attackers cannot modify the contents of a record, or change the
order of the records, without being detected and the connection being
dropped. If a record is lost (e.g. the receiver observers records #1,#2,#4,
but not #3), the connection is dropped when the unexpected sequence number is
received.
== Handshake ==
The transit key is used to derive several secondary keys. Two of them are
used as a "handshake", to distinguish correct Transit connections from other
programs that happen to connect to the Transit sockets by mistake or malice.
The handshake is also responsible for choosing exactly one TCP connection to
use, even though multiple outbound and inbound connections are being
attempted.
The SENDER-HANDSHAKE is the string `transit sender %s ready\n\n`, with the
`%s` replaced by a hex-encoded 32-byte HKDF derivative of the transit key,
using a "context string" of `transit_sender`. The RECEIVER-HANDSHAKE is the
same but with `receiver` instead of `sender` (both for the string and the
HKDF context).
The handshake protocol is like this:
* immediately upon socket connection being made, the Sender writes
SENDER-HANDSHAKE to the socket (regardless of whether the Sender initiated
the TCP connection, or was listening on a socket and just accepted the
connection)
* likewise the Receiver immediately writes RECEIVER-HANDSHAKE to either kind
of socket
* if the Sender sees anything other than RECEIVER-HANDSHAKE as the first
bytes on the wire, it hangs up
* likewise with the Receiver and SENDER-HANDSHAKE
* if the Sender sees that this is the first connection to get
RECEIVER-HANDSHAKE, it sends `go\n`. If some other connection got there
first, it hangs up (or sends `nevermind\n` and then hangs up, but this is
mostly for debugging, and implementations should not depend upon it). After
sending `go`, it switches to encrypted-record mode.
* if the Receiver sees `go\n`, it switches to encrypted-record mode. If the
receiver sees anything else, or a disconnected socket, it disconnects.
To tolerate the inevitable race conditions created by multiple contending
sockets, only the Sender gets to decide which one wins: the first one to make
it past negotiation. Hopefully this is correlated with the fastest connection
pathway. The protocol ignores any socket that is not somewhat affiliated with
the matching Transit instance.
Hints will frequently point to local IP addresses (local to the other end)
which might be in use by unrelated nearby computers. The handshake helps to
ignore these spurious connections. It is still possible for an attacker to
cause the connection to fail, by intercepting both connections (to learn the
two handshakes), then making new connections to play back the recorded
handshakes, but this level of attacker could simply drop the user's packets
directly.
== Relay ==
The **Transit Relay** is a host which offers TURN-like services for
magic-wormhole instances. It uses a TCP-based protocol with a handshake to
@ -40,3 +138,95 @@ hints available, the Transit instance will wait a few seconds before
attempting to use the relay. If it has no viable direct hints, it will start
using the relay right away. This prefers direct connections, but doesn't
introduce completely unnecessary stalls.
== API ==
First, create a Transit instance, giving it the connection information of the
transit relay. The application must know whether it should use a Sender or a
Receiver:
```python
from wormhole.blocking.transit import TransitSender
s = TransitSender("tcp:relayhost.example.org:12345")
```
Next, ask the Transit for its direct and relay hints. This should be
delivered to the other side via a Wormhole message (i.e. add them to a dict,
serialize it with JSON, send the result as a message with `wormhole.send()`).
```python
direct_hints = s.get_direct_hints()
relay_hints = s.get_relay_hints()
```
Then, perform the Wormhole exchange, which ought to give you the direct and
relay hints of the other side. Tell your Transit instance about their hints.
```python
s.add_their_direct_hints(their_direct_hints)
s.add_their_relay_hints(their_relay_hints)
```
Then use `wormhole.derive_key()` to obtain a shared key for Transit purposes,
and tell your Transit about it. Both sides must use the same derivation
string, and this string must not be used for any other purpose, but beyond
that it doesn't much matter what the exact string is.
```python
key = w.derive_key(application_id + "/transit-key")
s.set_transit_key(key)
```
Finally, tell the Transit instance to connect. This will yield a "record
pipe" object, on which records can be sent and received. If no connection can
be established within a timeout (defaults to 30 seconds), `connect()` will
throw an exception instead. The pipe can be closed with `close()`.
```python
rp = s.connect()
rp.send_record(b"my first record")
their_record = rp.receive_record()
rp.send_record(b"Greatest Hits)
other = rp.receive_record()
rp.close()
```
Records can be sent and received arbitrarily (you are not limited to taking
turns). However the blocking API does not provide a way to send records while
waiting for an inbound record. This *might* work with threads, but it has not
been tested.
== Twisted API ==
The same facilities are available in the asynchronous Twisted environment.
The difference is that some functions return Deferreds instead of immediate
values. The final record-pipe object is a Protocol (TBD: maybe this is a job
for Tubes?), which exposes `receive_record()` as a Deferred-returning
function that internally holds a queue of inbound records.
```python
from twisted.internet.defer import inlineCallbacks
from wormhole.twisted.transit import TransitSender
@inlineCallbacks
def do_transit():
s = TransitSender(relay)
my_relay_hints = s.get_relay_hints()
my_direct_hints = yield s.get_direct_hints()
# (send hints via wormhole)
s.add_their_relay_hints(their_relay_hints)
s.add_their_direct_hints(their_direct_hints)
s.set_transit_key(key)
rp = yield s.connect()
rp.send_record(b"eponymous")
them = yield rp.receive_record()
yield rp.close()
```
This object also implements the `IConsumer`/`IProducer` protocols for
**bytes**, which means you can transfer a file by wiring up a file reader as
a Producer. Each chunk of bytes that the Producer generates will be put into
a single record. The Consumer interface works the same way. This enables
backpressure and flow-control: if the far end (or the network) cannot keep up
with the stream of data, the sender will wait for them to catch up before
filling buffers without bound.

46
misc/migrate_usage_db.py
View File

@ -1,46 +0,0 @@
"""Migrate the usage data from the old bundled Transit Relay database.
The magic-wormhole package used to include both servers (Rendezvous and
Transit). "wormhole server" started both of these, and used the
"relay.sqlite" database to store both immediate server state and long-term
usage data.
These were split out to their own packages: version 0.11 omitted the Transit
Relay in favor of the new "magic-wormhole-transit-relay" distribution.
This script reads the long-term Transit usage data from the pre-0.11
wormhole-server relay.sqlite, and copies it into a new "usage.sqlite"
database in the current directory.
It will refuse to touch an existing "usage.sqlite" file.
The resuting "usage.sqlite" should be passed into --usage-db=, e.g. "twist
transitrelay --usage=.../PATH/TO/usage.sqlite".
"""
import sys
from wormhole_transit_relay.database import open_existing_db, create_db
source_fn = sys.argv[1]
source_db = open_existing_db(source_fn)
target_db = create_db("usage.sqlite")
num_rows = 0
for row in source_db.execute("SELECT * FROM `transit_usage`"
" ORDER BY `started`").fetchall():
target_db.execute("INSERT INTO `usage`"
" (`started`, `total_time`, `waiting_time`,"
" `total_bytes`, `result`)"
" VALUES(?,?,?,?,?)",
(row["started"], row["total_time"], row["waiting_time"],
row["total_bytes"], row["result"]))
num_rows += 1
target_db.execute("INSERT INTO `current`"
" (`rebooted`, `updated`, `connected`, `waiting`,"
" `incomplete_bytes`)"
" VALUES(?,?,?,?,?)",
(0, 0, 0, 0, 0))
target_db.commit()
print("usage database migrated (%d rows) into 'usage.sqlite'" % num_rows)
sys.exit(0)

33
misc/munin/wormhole_transit Executable file
View File

@ -0,0 +1,33 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.serverdir /path/to/your/wormhole/server
"""
import os, sys, time, json
CONFIG = """\
graph_title Magic-Wormhole Transit Usage (since reboot)
graph_vlabel Bytes Since Reboot
graph_category network
bytes.label Transit Bytes
bytes.draw LINE1
bytes.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print CONFIG.rstrip()
sys.exit(0)
serverdir = os.environ["serverdir"]
fn = os.path.join(serverdir, "stats.json")
with open(fn) as f:
data = json.load(f)
if time.time() > data["valid_until"]:
sys.exit(1) # expired
t = data["transit"]["since_reboot"]
print "bytes.value", t["bytes"]

39
misc/munin/wormhole_transit_active
View File

@ -1,39 +0,0 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.usagedb /path/to/your/wormhole/server/usage.sqlite
"""
import os, sys, time, sqlite3
CONFIG = """\
graph_title Magic-Wormhole Transit Active Channels
graph_vlabel Channels
graph_category wormhole
waiting.label Transit Waiting
waiting.draw LINE1
waiting.type GAUGE
connected.label Transit Connected
connected.draw LINE1
connected.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print(CONFIG.rstrip())
sys.exit(0)
dbfile = os.environ["usagedb"]
assert os.path.exists(dbfile)
db = sqlite3.connect(dbfile)
MINUTE = 60.0
updated,waiting,connected = db.execute("SELECT `updated`,`waiting`,`connected`"
" FROM `current`").fetchone()
if time.time() > updated + 5*MINUTE:
sys.exit(1) # expired
print("waiting.value", waiting)
print("connected.value", connected)

33
misc/munin/wormhole_transit_alltime Normal file
View File

@ -0,0 +1,33 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.serverdir /path/to/your/wormhole/server
"""
import os, sys, time, json
CONFIG = """\
graph_title Magic-Wormhole Transit Usage (all time)
graph_vlabel Bytes Since DB Creation
graph_category network
bytes.label Transit Bytes
bytes.draw LINE1
bytes.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print CONFIG.rstrip()
sys.exit(0)
serverdir = os.environ["serverdir"]
fn = os.path.join(serverdir, "stats.json")
with open(fn) as f:
data = json.load(f)
if time.time() > data["valid_until"]:
sys.exit(1) # expired
t = data["transit"]["all_time"]
print "bytes.value", t["bytes"]

41
misc/munin/wormhole_transit_bytes
View File

@ -1,41 +0,0 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.usagedb /path/to/your/wormhole/server/usage.sqlite
"""
import os, sys, time, sqlite3
CONFIG = """\
graph_title Magic-Wormhole Transit Usage (since reboot)
graph_vlabel Bytes Since Reboot
graph_category wormhole
bytes.label Transit Bytes (complete)
bytes.draw LINE1
bytes.type GAUGE
incomplete.label Transit Bytes (incomplete)
incomplete.draw LINE1
incomplete.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print(CONFIG.rstrip())
sys.exit(0)
dbfile = os.environ["usagedb"]
assert os.path.exists(dbfile)
db = sqlite3.connect(dbfile)
MINUTE = 60.0
updated,rebooted,incomplete = db.execute("SELECT `updated`,`rebooted`,`incomplete_bytes` FROM `current`").fetchone()
if time.time() > updated + 5*MINUTE:
sys.exit(1) # expired
complete = db.execute("SELECT SUM(`total_bytes`) FROM `usage`"
" WHERE `started` > ?",
(rebooted,)).fetchone()[0] or 0
print("bytes.value", complete)
print("incomplete.value", complete+incomplete)

41
misc/munin/wormhole_transit_bytes_alltime
View File

@ -1,41 +0,0 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.usagedb /path/to/your/wormhole/server/usage.sqlite
"""
import os, sys, time, sqlite3
CONFIG = """\
graph_title Magic-Wormhole Transit Usage (all time)
graph_vlabel Bytes Since DB Creation
graph_category wormhole
bytes.label Transit Bytes (complete)
bytes.draw LINE1
bytes.type GAUGE
incomplete.label Transit Bytes (incomplete)
incomplete.draw LINE1
incomplete.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print(CONFIG.rstrip())
sys.exit(0)
dbfile = os.environ["usagedb"]
assert os.path.exists(dbfile)
db = sqlite3.connect(dbfile)
MINUTE = 60.0
updated,incomplete = db.execute("SELECT `updated`,`incomplete_bytes`"
" FROM `current`").fetchone()
if time.time() > updated + 5*MINUTE:
sys.exit(1) # expired
complete = db.execute("SELECT SUM(`total_bytes`)"
" FROM `usage`").fetchone()[0] or 0
print("bytes.value", complete)
print("incomplete.value", complete+incomplete)

69
misc/munin/wormhole_transit_events
View File

@ -1,69 +0,0 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.usagedb /path/to/your/wormhole/server/usage.sqlite
"""
import os, sys, time, sqlite3
CONFIG = """\
graph_title Magic-Wormhole Transit Server Events (since reboot)
graph_vlabel Events Since Reboot
graph_category wormhole
happy.label Happy
happy.draw LINE1
happy.type GAUGE
errory.label Errory
errory.draw LINE1
errory.type GAUGE
lonely.label Lonely
lonely.draw LINE1
lonely.type GAUGE
redundant.label Redundant
redundant.draw LINE1
redundant.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print(CONFIG.rstrip())
sys.exit(0)
dbfile = os.environ["usagedb"]
assert os.path.exists(dbfile)
db = sqlite3.connect(dbfile)
MINUTE = 60.0
rebooted,updated = db.execute("SELECT `rebooted`, `updated` FROM `current`").fetchone()
if time.time() > updated + 5*MINUTE:
sys.exit(1) # expired
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE"
" `started` > ? AND"
" `result` = 'happy'",
(rebooted,)).fetchone()[0]
print("happy.value", count)
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE"
" `started` > ? AND"
" `result` = 'errory'",
(rebooted,)).fetchone()[0]
print("errory.value", count)
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE"
" `started` > ? AND"
" `result` = 'lonely'",
(rebooted,)).fetchone()[0]
print("lonely.value", count)
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE"
" `started` > ? AND"
" `result` = 'redundant'",
(rebooted,)).fetchone()[0]
print("redundant.value", count)

61
misc/munin/wormhole_transit_events_alltime
View File

@ -1,61 +0,0 @@
#! /usr/bin/env python
"""
Use the following in /etc/munin/plugin-conf.d/wormhole :
[wormhole_*]
env.usagedb /path/to/your/wormhole/server/usage.sqlite
"""
import os, sys, time, sqlite3
CONFIG = """\
graph_title Magic-Wormhole Transit Server Events (all time)
graph_vlabel Events
graph_category wormhole
happy.label Happy
happy.draw LINE1
happy.type GAUGE
errory.label Errory
errory.draw LINE1
errory.type GAUGE
lonely.label Lonely
lonely.draw LINE1
lonely.type GAUGE
redundant.label Redundant
redundant.draw LINE1
redundant.type GAUGE
"""
if len(sys.argv) > 1 and sys.argv[1] == "config":
print(CONFIG.rstrip())
sys.exit(0)
dbfile = os.environ["usagedb"]
assert os.path.exists(dbfile)
db = sqlite3.connect(dbfile)
MINUTE = 60.0
rebooted,updated = db.execute("SELECT `rebooted`, `updated` FROM `current`").fetchone()
if time.time() > updated + 5*MINUTE:
sys.exit(1) # expired
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE `result` = 'happy'",
).fetchone()[0]
print("happy.value", count)
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE `result` = 'errory'",
).fetchone()[0]
print("errory.value", count)
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE `result` = 'lonely'",
).fetchone()[0]
print("lonely.value", count)
count = db.execute("SELECT COUNT() FROM `usage`"
" WHERE `result` = 'redundant'",
).fetchone()[0]
print("redundant.value", count)

21
misc/windows-build.cmd
View File

@ -1,21 +0,0 @@
@echo off
:: To build extensions for 64 bit Python 3, we need to configure environment
:: variables to use the MSVC 2010 C++ compilers from GRMSDKX_EN_DVD.iso of:
:: MS Windows SDK for Windows 7 and .NET Framework 4
::
:: More details at:
:: https://github.com/cython/cython/wiki/64BitCythonExtensionsOnWindows
IF "%DISTUTILS_USE_SDK%"=="1" (
ECHO Configuring environment to build with MSVC on a 64bit architecture
ECHO Using Windows SDK 7.1
"C:\Program Files\Microsoft SDKs\Windows\v7.1\Setup\WindowsSdkVer.exe" -q -version:v7.1
CALL "C:\Program Files\Microsoft SDKs\Windows\v7.1\Bin\SetEnv.cmd" /x64 /release
SET MSSdk=1
REM Need the following to allow tox to see the SDK compiler
SET TOX_TESTENV_PASSENV=DISTUTILS_USE_SDK MSSdk INCLUDE LIB
) ELSE (
ECHO Using default MSVC build environment
)
CALL %*

4
setup.py
View File

@ -14,12 +14,10 @@ setup(name="magic-wormhole-transit-relay",
package_dir={"": "src"},
packages=["wormhole_transit_relay",
"wormhole_transit_relay.test",
"twisted.plugins",
],
package_data={"wormhole_transit_relay": ["db-schemas/*.sql"]},
install_requires=[
"twisted >= 21.2.0",
"autobahn >= 21.3.1",
"twisted >= 17.5.0",
],
extras_require={
':sys_platform=="win32"': ["pypiwin32"],

2
src/wormhole_transit_relay/_version.py
View File

@ -495,7 +495,7 @@ def get_versions():
# versionfile_source is the relative path from the top of the source
# tree (where the .git directory might live) to this file. Invert
# this to find the root from __file__.
for _ in cfg.versionfile_source.split('/'):
for i in cfg.versionfile_source.split('/'):
root = os.path.dirname(root)
except NameError:
return {"version": "0+unknown", "full-revisionid": None,

76
src/wormhole_transit_relay/database.py
View File

@ -1,3 +1,4 @@
from __future__ import unicode_literals
import os
import sqlite3
import tempfile
@ -8,16 +9,16 @@ class DBError(Exception):
pass
def get_schema(version):
schema_bytes = resource_string("wormhole_transit_relay",
schema_bytes = resource_string("wormhole.server",
"db-schemas/v%d.sql" % version)
return schema_bytes.decode("utf-8")
## def get_upgrader(new_version):
## schema_bytes = resource_string("wormhole_transit_relay",
## "db-schemas/upgrade-to-v%d.sql" % new_version)
## return schema_bytes.decode("utf-8")
def get_upgrader(new_version):
schema_bytes = resource_string("wormhole.server",
"db-schemas/upgrade-to-v%d.sql" % new_version)
return schema_bytes.decode("utf-8")
TARGET_VERSION = 1
TARGET_VERSION = 3
def dict_factory(cursor, row):
d = {}
@ -50,11 +51,9 @@ def _open_db_connection(dbfile):
"""
try:
db = sqlite3.connect(dbfile)
_initialize_db_connection(db)
except (EnvironmentError, sqlite3.OperationalError, sqlite3.DatabaseError) as e:
# this indicates that the file is not a compatible database format.
# Perhaps it was created with an old version, or it might be junk.
except (EnvironmentError, sqlite3.OperationalError) as e:
raise DBError("Unable to create/open db file %s: %s" % (dbfile, e))
_initialize_db_connection(db)
return db
def _get_temporary_dbfile(dbfile):
@ -92,52 +91,31 @@ def get_db(dbfile, target_version=TARGET_VERSION):
else:
db = _atomic_create_and_initialize_db(dbfile, target_version)
version = db.execute("SELECT version FROM version").fetchone()["version"]
try:
version = db.execute("SELECT version FROM version").fetchone()["version"]
except sqlite3.DatabaseError as e:
# this indicates that the file is not a compatible database format.
# Perhaps it was created with an old version, or it might be junk.
raise DBError("db file is unusable: %s" % e)
## while version < target_version:
## log.msg(" need to upgrade from %s to %s" % (version, target_version))
## try:
## upgrader = get_upgrader(version+1)
## except ValueError: # ResourceError??
## log.msg(" unable to upgrade %s to %s" % (version, version+1))
## raise DBError("Unable to upgrade %s to version %s, left at %s"
## % (dbfile, version+1, version))
## log.msg(" executing upgrader v%s->v%s" % (version, version+1))
## db.executescript(upgrader)
## db.commit()
## version = version+1
while version < target_version:
log.msg(" need to upgrade from %s to %s" % (version, target_version))
try:
upgrader = get_upgrader(version+1)
except ValueError: # ResourceError??
log.msg(" unable to upgrade %s to %s" % (version, version+1))
raise DBError("Unable to upgrade %s to version %s, left at %s"
% (dbfile, version+1, version))
log.msg(" executing upgrader v%s->v%s" % (version, version+1))
db.executescript(upgrader)
db.commit()
version = version+1
if version != target_version:
raise DBError("Unable to handle db version %s" % version)
return db
class DBDoesntExist(Exception):
pass
def open_existing_db(dbfile):
assert dbfile != ":memory:"
if not os.path.exists(dbfile):
raise DBDoesntExist()
return _open_db_connection(dbfile)
class DBAlreadyExists(Exception):
pass
def create_db(dbfile):
"""Create the given db file. Refuse to touch a pre-existing file.
This is meant for use by migration tools, to create the output target"""
if dbfile == ":memory:":
db = _open_db_connection(dbfile)
_initialize_db_schema(db, TARGET_VERSION)
elif os.path.exists(dbfile):
raise DBAlreadyExists()
else:
db = _atomic_create_and_initialize_db(dbfile, TARGET_VERSION)
return db
def dump_db(db):
# to let _iterdump work, we need to restore the original row factory
orig = db.row_factory

9
src/wormhole_transit_relay/db-schemas/v1.sql
View File

@ -7,8 +7,8 @@ CREATE TABLE `version` -- contains one row
CREATE TABLE `current` -- contains one row
(
`rebooted` INTEGER, -- seconds since epoch of most recent reboot
`updated` INTEGER, -- when `current` was last updated
`reboot` INTEGER, -- seconds since epoch of most recent reboot
`last_update` INTEGER, -- when `current` was last updated
`connected` INTEGER, -- number of current paired connections
`waiting` INTEGER, -- number of not-yet-paired connections
`incomplete_bytes` INTEGER -- bytes sent through not-yet-complete connections
@ -24,8 +24,7 @@ CREATE TABLE `usage`
-- transit moods:
-- "errory": one side gave the wrong handshake
-- "lonely": good handshake, but the other side never showed up
-- "redundant": good handshake, abandoned in favor of different connection
-- "happy": both sides gave correct handshake
);
CREATE INDEX `usage_started_index` ON `usage` (`started`);
CREATE INDEX `usage_result_index` ON `usage` (`result`);
CREATE INDEX `transit_usage_idx` ON `transit_usage` (`started`);
CREATE INDEX `transit_usage_result_idx` ON `transit_usage` (`result`);

35
src/wormhole_transit_relay/increase_rlimits.py
View File

@ -1,35 +0,0 @@
try:
# 'resource' is unix-only
from resource import getrlimit, setrlimit, RLIMIT_NOFILE
except ImportError: # pragma: nocover
getrlimit, setrlimit, RLIMIT_NOFILE = None, None, None # pragma: nocover
from twisted.python import log
def increase_rlimits():
if getrlimit is None:
log.msg("unable to import 'resource', leaving rlimit alone")
return
soft, hard = getrlimit(RLIMIT_NOFILE)
if soft >= 10000:
log.msg("RLIMIT_NOFILE.soft was %d, leaving it alone" % soft)
return
# OS-X defaults to soft=7168, and reports a huge number for 'hard',
# but won't accept anything more than soft=10240, so we can't just
# set soft=hard. Linux returns (1024, 1048576) and is fine with
# soft=hard. Cygwin is reported to return (256,-1) and accepts up to
# soft=3200. So we try multiple values until something works.
for newlimit in [hard, 10000, 3200, 1024]:
log.msg("changing RLIMIT_NOFILE from (%s,%s) to (%s,%s)" %
(soft, hard, newlimit, hard))
try:
setrlimit(RLIMIT_NOFILE, (newlimit, hard))
log.msg("setrlimit successful")
return
except ValueError as e:
log.msg("error during setrlimit: %s" % e)
continue
except:
log.msg("other error during setrlimit, leaving it alone")
log.err()
return
log.msg("unable to change rlimit, leaving it alone")

477
src/wormhole_transit_relay/server_state.py
View File

@ -1,477 +0,0 @@
from collections import defaultdict
import automat
from twisted.python import log
from zope.interface import (
Interface,
Attribute,
)
class ITransitClient(Interface):
"""
Represents the client side of a connection to this transit
relay. This is used by TransitServerState instances.
"""
started_time = Attribute("timestamp when the connection was established")
def send(data):
"""
Send some byets to the client
"""
def disconnect():
"""
Disconnect the client transport
"""
def connect_partner(other):
"""
Hook up to our partner.
:param ITransitClient other: our partner
"""
def disconnect_partner():
"""
Disconnect our partner's transport
"""
class ActiveConnections(object):
"""
Tracks active connections.
A connection is 'active' when both sides have shown up and they
are glued together (and thus could be passing data back and forth
if any is flowing).
"""
def __init__(self):
self._connections = set()
def register(self, side0, side1):
"""
A connection has become active so register both its sides
:param TransitConnection side0: one side of the connection
:param TransitConnection side1: one side of the connection
"""
self._connections.add(side0)
self._connections.add(side1)
def unregister(self, side):
"""
One side of a connection has become inactive.
:param TransitConnection side: an inactive side of a connection
"""
self._connections.discard(side)
class PendingRequests(object):
"""
Tracks outstanding (non-"active") requests.
We register client connections against the tokens we have
received. When the other side shows up we can thus match it to the
correct partner connection. At this point, the connection becomes
"active" is and is thus no longer "pending" and so will no longer
be in this collection.
"""
def __init__(self, active_connections):
"""
:param active_connections: an instance of ActiveConnections where
connections are put when both sides arrive.
"""
self._requests = defaultdict(set) # token -> set((side, TransitConnection))
self._active = active_connections
def unregister(self, token, side, tc):
"""
We no longer care about a particular client (e.g. it has
disconnected).
"""
if token in self._requests:
self._requests[token].discard((side, tc))
if not self._requests[token]:
# no more sides; token is dead
del self._requests[token]
self._active.unregister(tc)
def register(self, token, new_side, new_tc):
"""
A client has connected and successfully offered a token (and
optional 'side' token). If this is the first one for this
token, we merely remember it. If it is the second side for
this token we connect them together.
:param bytes token: the token for this connection.
:param bytes new_side: None or the side token for this connection
:param TransitServerState new_tc: the state-machine of the connection
:returns bool: True if we are the first side to register this
token
"""
potentials = self._requests[token]
for old in potentials:
(old_side, old_tc) = old
if ((old_side is None)
or (new_side is None)
or (old_side != new_side)):
# we found a match
# drop and stop tracking the rest
potentials.remove(old)
for (_, leftover_tc) in potentials.copy():
# Don't record this as errory. It's just a spare connection
# from the same side as a connection that got used. This
# can happen if the connection hint contains multiple
# addresses (we don't currently support those, but it'd
# probably be useful in the future).
leftover_tc.partner_connection_lost()
self._requests.pop(token, None)
# glue the two ends together
self._active.register(new_tc, old_tc)
new_tc.got_partner(old_tc)
old_tc.got_partner(new_tc)
return False
potentials.add((new_side, new_tc))
return True
# TODO: timer
class TransitServerState(object):
"""
Encapsulates the state-machine of the server side of a transit
relay connection.
Once the protocol has been told to relay (or to relay for a side)
it starts passing all received bytes to the other side until it
closes.
"""
_machine = automat.MethodicalMachine()
_client = None
_buddy = None
_token = None
_side = None
_first = None
_mood = "empty"
_total_sent = 0
def __init__(self, pending_requests, usage_recorder):
self._pending_requests = pending_requests
self._usage = usage_recorder
def get_token(self):
"""
:returns str: a string describing our token. This will be "-" if
we have no token yet, or "{16 chars}-<unsided>" if we have
just a token or "{16 chars}-{16 chars}" if we have a token and
a side.
"""
d = "-"
if self._token is not None:
d = self._token[:16].decode("ascii")
if self._side is not None:
d += "-" + self._side.decode("ascii")
else:
d += "-<unsided>"
return d
@_machine.input()
def connection_made(self, client):
"""
A client has connected. May only be called once.
:param ITransitClient client: our client.
"""
# NB: the "only called once" is enforced by the state-machine;
# this input is only valid for the "listening" state, to which
# we never return.
@_machine.input()
def please_relay(self, token):
"""
A 'please relay X' message has been received (the original version
of the protocol).
"""
@_machine.input()
def please_relay_for_side(self, token, side):
"""
A 'please relay X for side Y' message has been received (the
second version of the protocol).
"""
@_machine.input()
def bad_token(self):
"""
A bad token / relay line was received (e.g. couldn't be parsed)
"""
@_machine.input()
def got_partner(self, client):
"""
The partner for this relay session has been found
"""
@_machine.input()
def connection_lost(self):
"""
Our transport has failed.
"""
@_machine.input()
def partner_connection_lost(self):
"""
Our partner's transport has failed.
"""
@_machine.input()
def got_bytes(self, data):
"""
Some bytes have arrived (that aren't part of the handshake)
"""
@_machine.output()
def _remember_client(self, client):
self._client = client
# note that there is no corresponding "_forget_client" because we
# may still want to access it after it is gone .. for example, to
# get the .started_time for logging purposes
@_machine.output()
def _register_token(self, token):
return self._real_register_token_for_side(token, None)
@_machine.output()
def _register_token_for_side(self, token, side):
return self._real_register_token_for_side(token, side)
@_machine.output()
def _unregister(self):
"""
remove us from the thing that remembers tokens and sides
"""
return self._pending_requests.unregister(self._token, self._side, self)
@_machine.output()
def _send_bad(self):
self._mood = "errory"
self._client.send(b"bad handshake\n")
if self._client.factory.log_requests:
log.msg("transit handshake failure")
@_machine.output()
def _send_ok(self):
self._client.send(b"ok\n")
@_machine.output()
def _send_impatient(self):
self._client.send(b"impatient\n")
if self._client.factory.log_requests:
log.msg("transit impatience failure")
@_machine.output()
def _count_bytes(self, data):
self._total_sent += len(data)
@_machine.output()
def _send_to_partner(self, data):
self._buddy._client.send(data)
@_machine.output()
def _connect_partner(self, client):
self._buddy = client
self._client.connect_partner(client)
@_machine.output()
def _disconnect(self):
self._client.disconnect()
@_machine.output()
def _disconnect_partner(self):
self._client.disconnect_partner()
# some outputs to record "usage" information ..
@_machine.output()
def _record_usage(self):
if self._mood == "jilted":
if self._buddy and self._buddy._mood == "happy":
return
self._usage.record(
started=self._client.started_time,
buddy_started=self._buddy._client.started_time if self._buddy is not None else None,
result=self._mood,
bytes_sent=self._total_sent,
buddy_bytes=self._buddy._total_sent if self._buddy is not None else None
)
# some outputs to record the "mood" ..
@_machine.output()
def _mood_happy(self):
self._mood = "happy"
@_machine.output()
def _mood_lonely(self):
self._mood = "lonely"
@_machine.output()
def _mood_redundant(self):
self._mood = "redundant"
@_machine.output()
def _mood_impatient(self):
self._mood = "impatient"
@_machine.output()
def _mood_errory(self):
self._mood = "errory"
@_machine.output()
def _mood_happy_if_first(self):
"""
We disconnected first so we're only happy if we also connected
first.
"""
if self._first:
self._mood = "happy"
else:
self._mood = "jilted"
def _real_register_token_for_side(self, token, side):
"""
A client has connected and sent a valid version 1 or version 2
handshake. If the former, `side` will be None.
In either case, we remember the tokens and register
ourselves. This might result in 'got_partner' notifications to
two state-machines if this is the second side for a given token.
:param bytes token: the token
:param bytes side: The side token (or None)
"""
self._token = token
self._side = side
self._first = self._pending_requests.register(token, side, self)
@_machine.state(initial=True)
def listening(self):
"""
Initial state, awaiting connection.
"""
@_machine.state()
def wait_relay(self):
"""
Waiting for a 'relay' message
"""
@_machine.state()
def wait_partner(self):
"""
Waiting for our partner to connect
"""
@_machine.state()
def relaying(self):
"""
Relaying bytes to our partner
"""
@_machine.state()
def done(self):
"""
Terminal state
"""
listening.upon(
connection_made,
enter=wait_relay,
outputs=[_remember_client],
)
listening.upon(
connection_lost,
enter=done,
outputs=[_mood_errory],
)
wait_relay.upon(
please_relay,
enter=wait_partner,
outputs=[_mood_lonely, _register_token],
)
wait_relay.upon(
please_relay_for_side,
enter=wait_partner,
outputs=[_mood_lonely, _register_token_for_side],
)
wait_relay.upon(
bad_token,
enter=done,
outputs=[_mood_errory, _send_bad, _disconnect, _record_usage],
)
wait_relay.upon(
got_bytes,
enter=done,
outputs=[_count_bytes, _mood_errory, _disconnect, _record_usage],
)
wait_relay.upon(
connection_lost,
enter=done,
outputs=[_disconnect, _record_usage],
)
wait_partner.upon(
got_partner,
enter=relaying,
outputs=[_mood_happy, _send_ok, _connect_partner],
)
wait_partner.upon(
connection_lost,
enter=done,
outputs=[_mood_lonely, _unregister, _record_usage],
)
wait_partner.upon(
got_bytes,
enter=done,
outputs=[_mood_impatient, _send_impatient, _disconnect, _unregister, _record_usage],
)
wait_partner.upon(
partner_connection_lost,
enter=done,
outputs=[_mood_redundant, _disconnect, _record_usage],
)
relaying.upon(
got_bytes,
enter=relaying,
outputs=[_count_bytes, _send_to_partner],
)
relaying.upon(
connection_lost,
enter=done,
outputs=[_mood_happy_if_first, _disconnect_partner, _unregister, _record_usage],
)
done.upon(
connection_lost,
enter=done,
outputs=[],
)
done.upon(
partner_connection_lost,
enter=done,
outputs=[],
)
# uncomment to turn on state-machine tracing
# set_trace_function = _machine._setTrace

136
src/wormhole_transit_relay/server_tap.py
View File

@ -1,83 +1,97 @@
import os
from . import transit_server
from twisted.internet import reactor
from twisted.python import usage
from twisted.application.service import MultiService
from twisted.application.internet import (TimerService,
StreamServerEndpointService)
from twisted.application.internet import StreamServerEndpointService
from twisted.internet import endpoints
from twisted.internet import protocol
from autobahn.twisted.websocket import WebSocketServerFactory
from . import transit_server
from .usage import create_usage_tracker
from .increase_rlimits import increase_rlimits
from .database import get_db
LONGDESC = """\
This plugin sets up a 'Transit Relay' server for magic-wormhole. This service
listens for TCP connections, finds pairs which present the same handshake, and
glues the two TCP sockets together.
If --log-stdout is provided, a line will be written to stdout after each
connection is done. This line will be a complete JSON object (starting with
"{", ending with "}\n", and containing no internal newlines). The keys will
be:
* 'started': number, seconds since epoch
* 'total_time': number, seconds from open to last close
* 'waiting_time': number, seconds from start to 2nd side appearing, or null
* 'total_bytes': number, total bytes relayed (sum of both directions)
* 'mood': string, one of: happy, lonely, errory
A mood of "happy" means both sides gave a correct handshake. "lonely" means a
second matching side never appeared (and thus 'waiting_time' will be null).
"errory" means the first side gave an invalid handshake.
If --blur-usage= is provided, then 'started' will be rounded to the given time
interval, and 'total_bytes' will be rounded as well.
If --usage-db= is provided, the server will maintain a SQLite database in the
given file. Current, recent, and historical usage data will be written to the
database, and external tools can query the DB for metrics: the munin plugins
in misc/ may be useful. Timestamps and sizes in this file will respect
--blur-usage. The four tables are:
"current" contains a single row, with these columns:
* connected: number of paired connections
* waiting: number of not-yet-paired connections
* partal_bytes: bytes transmitted over not-yet-complete connections
"since_reboot" contains a single row, with these columns:
* bytes: sum of 'total_bytes'
* connections: number of completed connections
* mood_happy: count of connections that finished "happy": both sides gave correct handshake
* mood_lonely: one side gave good handshake, other side never showed up
* mood_errory: one side gave a bad handshake
"all_time" contains a single row, with these columns:
* bytes:
* connections:
* mood_happy:
* mood_lonely:
* mood_errory:
"usage" contains one row per closed connection, with these columns:
* started: seconds since epoch, rounded to "blur time"
* total_time: seconds from first open to last close
* waiting_time: seconds from first open to second open, or None
* bytes: total bytes relayed (in both directions)
* result: (string) the mood: happy, lonely, errory
All tables will be updated after each connection is finished. In addition,
the "current" table will be updated at least once every 5 minutes.
If daemonized by twistd, the server will write twistd.pid and twistd.log
files as usual. By default twistd.log will only contain startup, shutdown,
and exception messages. Adding --log-stdout will add per-connection JSON
lines to twistd.log.
"""
class Options(usage.Options):
synopsis = "[--port=] [--log-fd] [--blur-usage=] [--usage-db=]"
#synopsis = "[--port=] [--log-stdout] [--blur-usage=] [--usage-db=]"
longdesc = LONGDESC
optFlags = {
("log-stdout", None, "write JSON usage logs to stdout"),
}
optParameters = [
("port", "p", "tcp:4001:interface=\:\:", "endpoint to listen on"),
("websocket", "w", None, "endpoint to listen for WebSocket connections"),
("websocket-url", "u", None, "WebSocket URL (derived from endpoint if not provided)"),
("port", "p", "tcp:4001", "endpoint to listen on"),
("blur-usage", None, None, "blur timestamps and data sizes in logs"),
("log-fd", None, None, "write JSON usage logs to this file descriptor"),
("usage-db", None, None, "record usage data (SQLite)"),
]
def opt_blur_usage(self, arg):
self["blur-usage"] = int(arg)
self["blur_usage"] = int(arg)
def makeService(config, reactor=reactor):
increase_rlimits()
tcp_ep = endpoints.serverFromString(reactor, config["port"]) # to listen
ws_ep = (
endpoints.serverFromString(reactor, config["websocket"])
if config["websocket"] is not None
else None
)
log_file = (
os.fdopen(int(config["log-fd"]), "w")
if config["log-fd"] is not None
else None
)
db = None if config["usage-db"] is None else get_db(config["usage-db"])
usage = create_usage_tracker(
blur_usage=config["blur-usage"],
log_file=log_file,
usage_db=db,
)
transit = transit_server.Transit(usage, reactor.seconds)
tcp_factory = protocol.ServerFactory()
tcp_factory.protocol = transit_server.TransitConnection
tcp_factory.log_requests = False
if ws_ep is not None:
ws_url = config["websocket-url"]
if ws_url is None:
# we're using a "private" attribute here but I don't see
# any useful alternative unless we also want to parse
# Twisted endpoint-strings.
ws_url = "ws://localhost:{}/".format(ws_ep._port)
print("Using WebSocket URL '{}'".format(ws_url))
ws_factory = WebSocketServerFactory(ws_url)
ws_factory.protocol = transit_server.WebSocketTransitConnection
ws_factory.transit = transit
ws_factory.log_requests = False
tcp_factory.transit = transit
parent = MultiService()
StreamServerEndpointService(tcp_ep, tcp_factory).setServiceParent(parent)
if ws_ep is not None:
StreamServerEndpointService(ws_ep, ws_factory).setServiceParent(parent)
TimerService(5*60.0, transit.update_stats).setServiceParent(parent)
return parent
ep = endpoints.serverFromString(reactor, config["port"]) # to listen
f = transit_server.Transit(blur_usage=config["blur-usage"],
log_stdout=config["log-stdout"],
usage_db=config["usage-db"])
return StreamServerEndpointService(ep, f)

151
src/wormhole_transit_relay/test/common.py
View File

@ -1,144 +1,23 @@
from twisted.internet.protocol import (
ClientFactory,
Protocol,
)
from twisted.test import iosim
from zope.interface import (
Interface,
Attribute,
implementer,
)
from ..transit_server import (
Transit,
TransitConnection,
)
from twisted.internet.protocol import ServerFactory
from ..usage import create_usage_tracker
class IRelayTestClient(Interface):
"""
The client interface used by tests.
"""
connected = Attribute("True if we are currently connected else False")
def send(data):
"""
Send some bytes.
:param bytes data: the data to send
"""
def disconnect():
"""
Terminate the connection.
"""
def get_received_data():
"""
:returns: all the bytes received from the server on this
connection.
"""
def reset_data():
"""
Erase any received data to this point.
"""
#from __future__ import unicode_literals
from twisted.internet import reactor, endpoints
from twisted.internet.defer import inlineCallbacks
from ..transit_server import Transit
class ServerBase:
log_requests = False
def setUp(self):
self._pumps = []
self._lp = None
if self.log_requests:
blur_usage = None
else:
blur_usage = 60.0
self._setup_relay(blur_usage=blur_usage)
self._setup_relay()
def flush(self):
did_work = False
for pump in self._pumps:
did_work = pump.flush() or did_work
if did_work:
self.flush()
def _setup_relay(self, blur_usage=None, log_file=None, usage_db=None):
usage = create_usage_tracker(
blur_usage=blur_usage,
log_file=log_file,
usage_db=usage_db,
)
self._transit_server = Transit(usage, lambda: 123456789.0)
def new_protocol(self):
"""
This should be overridden by derived test-case classes to decide
if they want a TCP or WebSockets protocol.
"""
raise NotImplementedError()
def new_protocol_tcp(self):
"""
Create a new client protocol connected to the server.
:returns: a IRelayTestClient implementation
"""
server_factory = ServerFactory()
server_factory.protocol = TransitConnection
server_factory.transit = self._transit_server
server_factory.log_requests = self.log_requests
server_protocol = server_factory.buildProtocol(('127.0.0.1', 0))
@implementer(IRelayTestClient)
class TransitClientProtocolTcp(Protocol):
"""
Speak the transit client protocol used by the tests over TCP
"""
_received = b""
connected = False
# override Protocol callbacks
def connectionMade(self):
self.connected = True
return Protocol.connectionMade(self)
def connectionLost(self, reason):
self.connected = False
return Protocol.connectionLost(self, reason)
def dataReceived(self, data):
self._received = self._received + data
# IRelayTestClient
def send(self, data):
self.transport.write(data)
def disconnect(self):
self.transport.loseConnection()
def reset_received_data(self):
self._received = b""
def get_received_data(self):
return self._received
client_factory = ClientFactory()
client_factory.protocol = TransitClientProtocolTcp
client_protocol = client_factory.buildProtocol(('127.0.0.1', 31337))
pump = iosim.connect(
server_protocol,
iosim.makeFakeServer(server_protocol),
client_protocol,
iosim.makeFakeClient(client_protocol),
)
pump.flush()
self._pumps.append(pump)
return client_protocol
@inlineCallbacks
def _setup_relay(self, blur_usage=None, usage_logfile=None, stats_file=None):
ep = endpoints.TCP4ServerEndpoint(reactor, 0, interface="127.0.0.1")
self._transit_server = Transit(blur_usage=blur_usage,
usage_logfile=usage_logfile,
stats_file=stats_file)
self._lp = yield ep.listen(self._transit_server)
addr = self._lp.getHost()
# ws://127.0.0.1:%d/wormhole-relay/ws
self.transit = u"tcp:127.0.0.1:%d" % addr.port
def tearDown(self):
if self._lp:

208
src/wormhole_transit_relay/test/test_backpressure.py
View File

@ -1,208 +0,0 @@
from io import (
StringIO,
)
import sys
import shutil
from twisted.trial import unittest
from twisted.internet.interfaces import (
IPullProducer,
)
from twisted.internet.protocol import (
ProcessProtocol,
)
from twisted.internet.defer import (
inlineCallbacks,
Deferred,
)
from autobahn.twisted.websocket import (
WebSocketClientProtocol,
create_client_agent,
)
from zope.interface import implementer
class _CollectOutputProtocol(ProcessProtocol):
"""
Internal helper. Collects all output (stdout + stderr) into
self.output, and callback's on done with all of it after the
process exits (for any reason).
"""
def __init__(self):
self.done = Deferred()
self.running = Deferred()
self.output = StringIO()
def processEnded(self, reason):
if not self.done.called:
self.done.callback(self.output.getvalue())
def outReceived(self, data):
print(data.decode(), end="", flush=True)
self.output.write(data.decode(sys.getfilesystemencoding()))
if not self.running.called:
if "on 8088" in self.output.getvalue():
self.running.callback(None)
def errReceived(self, data):
print("ERR: {}".format(data.decode(sys.getfilesystemencoding())))
self.output.write(data.decode(sys.getfilesystemencoding()))
def run_transit(reactor, proto, tcp_port=None, websocket_port=None):
exe = shutil.which("twistd")
args = [
exe, "-n", "transitrelay",
]
if tcp_port is not None:
args.append("--port")
args.append(tcp_port)
if websocket_port is not None:
args.append("--websocket")
args.append(websocket_port)
proc = reactor.spawnProcess(proto, exe, args)
return proc
class Sender(WebSocketClientProtocol):
"""
"""
def __init__(self, *args, **kw):
WebSocketClientProtocol.__init__(self, *args, **kw)
self.done = Deferred()
self.got_ok = Deferred()
def onMessage(self, payload, is_binary):
print("onMessage")
if not self.got_ok.called:
if payload == b"ok\n":
self.got_ok.callback(None)
print("send: {}".format(payload.decode("utf8")))
def onClose(self, clean, code, reason):
print(f"close: {clean} {code} {reason}")
self.done.callback(None)
class Receiver(WebSocketClientProtocol):
"""
"""
def __init__(self, *args, **kw):
WebSocketClientProtocol.__init__(self, *args, **kw)
self.done = Deferred()
self.first_message = Deferred()
self.received = 0
def onMessage(self, payload, is_binary):
print("recv: {}".format(len(payload)))
self.received += len(payload)
if not self.first_message.called:
self.first_message.callback(None)
def onClose(self, clean, code, reason):
print(f"close: {clean} {code} {reason}")
self.done.callback(None)
class TransitWebSockets(unittest.TestCase):
"""
Integration-style tests of the transit WebSocket relay, using the
real reactor (and running transit as a subprocess).
"""
@inlineCallbacks
def test_buffer_fills(self):
"""
A running transit relay stops accepting incoming data at a
reasonable amount if the peer isn't reading. This test defines
that as 'less than 100MiB' although in practice Twisted seems
to stop before 10MiB.
"""
from twisted.internet import reactor
transit_proto = _CollectOutputProtocol()
transit_proc = run_transit(reactor, transit_proto, websocket_port="tcp:8088")
def cleanup_process():
transit_proc.signalProcess("HUP")
return transit_proto.done
self.addCleanup(cleanup_process)
yield transit_proto.running
print("Transit running")
agent = create_client_agent(reactor)
side_a = yield agent.open("ws://localhost:8088", {}, lambda: Sender())
side_b = yield agent.open("ws://localhost:8088", {}, lambda: Receiver())
side_a.sendMessage(b"please relay aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa for side aaaaaaaaaaaaaaaa", True)
side_b.sendMessage(b"please relay aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa for side bbbbbbbbbbbbbbbb", True)
yield side_a.got_ok
yield side_b.first_message
# remove side_b's filedescriptor from the reactor .. this
# means it will not read any more data
reactor.removeReader(side_b.transport)
# attempt to send up to 100MiB through side_a .. we should get
# backpressure before that works which only manifests itself
# as this producer not being asked to produce more
max_data = 1024*1024*100 # 100MiB
@implementer(IPullProducer)
class ProduceMessages:
def __init__(self, ws, on_produce):
self._ws = ws
self._sent = 0
self._max = max_data
self._on_produce = on_produce
def resumeProducing(self):
self._on_produce()
if self._sent >= self._max:
self._ws.sendClose()
return
data = b"a" * 1024*1024
self._ws.sendMessage(data, True)
self._sent += len(data)
print("sent {}, total {}".format(len(data), self._sent))
# our only signal is, "did our producer get asked to produce
# more data" which it should do periodically. We want to stop
# if we haven't seen a new data request for a while -- defined
# as "more than 5 seconds".
done = Deferred()
last_produce = None
timeout = 2 # seconds
def asked_for_data():
nonlocal last_produce
last_produce = reactor.seconds()
data = ProduceMessages(side_a, asked_for_data)
side_a.transport.registerProducer(data, False)
data.resumeProducing()
def check_if_done():
if last_produce is not None:
if reactor.seconds() - last_produce > timeout:
done.callback(None)
return
# recursive call to ourselves to check again soon
reactor.callLater(.1, check_if_done)
check_if_done()
yield done
mib = 1024*1024.0
print("Sent {}MiB of {}MiB before backpressure".format(data._sent / mib, max_data / mib))
self.assertTrue(data._sent < max_data, "Too much data sent")
side_a.sendClose()
side_b.sendClose()
yield side_a.done
yield side_b.done

41
src/wormhole_transit_relay/test/test_config.py
View File

@ -1,41 +0,0 @@
from twisted.trial import unittest
from .. import server_tap
PORT = "tcp:4001:interface=\:\:"
class Config(unittest.TestCase):
def test_defaults(self):
o = server_tap.Options()
o.parseOptions([])
self.assertEqual(o, {"blur-usage": None, "log-fd": None,
"usage-db": None, "port": PORT,
"websocket": None, "websocket-url": None})
def test_blur(self):
o = server_tap.Options()
o.parseOptions(["--blur-usage=60"])
self.assertEqual(o, {"blur-usage": 60, "log-fd": None,
"usage-db": None, "port": PORT,
"websocket": None, "websocket-url": None})
def test_websocket(self):
o = server_tap.Options()
o.parseOptions(["--websocket=tcp:4004"])
self.assertEqual(o, {"blur-usage": None, "log-fd": None,
"usage-db": None, "port": PORT,
"websocket": "tcp:4004", "websocket-url": None})
def test_websocket_url(self):
o = server_tap.Options()
o.parseOptions(["--websocket=tcp:4004", "--websocket-url=ws://example.com/"])
self.assertEqual(o, {"blur-usage": None, "log-fd": None,
"usage-db": None, "port": PORT,
"websocket": "tcp:4004",
"websocket-url": "ws://example.com/"})
def test_string(self):
o = server_tap.Options()
s = str(o)
self.assertIn("This plugin sets up a 'Transit Relay'", s)
self.assertIn("--blur-usage=", s)
self.assertIn("blur timestamps and data sizes in logs", s)

138
src/wormhole_transit_relay/test/test_database.py
View File

@ -1,138 +0,0 @@
import os
from twisted.python import filepath
from twisted.trial import unittest
from .. import database
from ..database import get_db, TARGET_VERSION, dump_db, DBError
class Get(unittest.TestCase):
def test_create_default(self):
db_url = ":memory:"
db = get_db(db_url)
rows = db.execute("SELECT * FROM version").fetchall()
self.assertEqual(len(rows), 1)
self.assertEqual(rows[0]["version"], TARGET_VERSION)
def test_open_existing_file(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "normal.db")
db = get_db(fn)
rows = db.execute("SELECT * FROM version").fetchall()
self.assertEqual(len(rows), 1)
self.assertEqual(rows[0]["version"], TARGET_VERSION)
db2 = get_db(fn)
rows = db2.execute("SELECT * FROM version").fetchall()
self.assertEqual(len(rows), 1)
self.assertEqual(rows[0]["version"], TARGET_VERSION)
def test_open_bad_version(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "old.db")
db = get_db(fn)
db.execute("UPDATE version SET version=999")
db.commit()
with self.assertRaises(DBError) as e:
get_db(fn)
self.assertIn("Unable to handle db version 999", str(e.exception))
def test_open_corrupt(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "corrupt.db")
with open(fn, "wb") as f:
f.write(b"I am not a database")
with self.assertRaises(DBError) as e:
get_db(fn)
self.assertIn("not a database", str(e.exception))
def test_failed_create_allows_subsequent_create(self):
patch = self.patch(database, "get_schema", lambda version: b"this is a broken schema")
dbfile = filepath.FilePath(self.mktemp())
self.assertRaises(Exception, lambda: get_db(dbfile.path))
patch.restore()
get_db(dbfile.path)
def OFF_test_upgrade(self): # disabled until we add a v2 schema
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "upgrade.db")
self.assertNotEqual(TARGET_VERSION, 2)
# create an old-version DB in a file
db = get_db(fn, 2)
rows = db.execute("SELECT * FROM version").fetchall()
self.assertEqual(len(rows), 1)
self.assertEqual(rows[0]["version"], 2)
del db
# then upgrade the file to the latest version
dbA = get_db(fn, TARGET_VERSION)
rows = dbA.execute("SELECT * FROM version").fetchall()
self.assertEqual(len(rows), 1)
self.assertEqual(rows[0]["version"], TARGET_VERSION)
dbA_text = dump_db(dbA)
del dbA
# make sure the upgrades got committed to disk
dbB = get_db(fn, TARGET_VERSION)
dbB_text = dump_db(dbB)
del dbB
self.assertEqual(dbA_text, dbB_text)
# The upgraded schema should be equivalent to that of a new DB.
# However a text dump will differ because ALTER TABLE always appends
# the new column to the end of a table, whereas our schema puts it
# somewhere in the middle (wherever it fits naturally). Also ALTER
# TABLE doesn't include comments.
if False:
latest_db = get_db(":memory:", TARGET_VERSION)
latest_text = dump_db(latest_db)
with open("up.sql","w") as f: f.write(dbA_text)
with open("new.sql","w") as f: f.write(latest_text)
# check with "diff -u _trial_temp/up.sql _trial_temp/new.sql"
self.assertEqual(dbA_text, latest_text)
class Create(unittest.TestCase):
def test_memory(self):
db = database.create_db(":memory:")
latest_text = dump_db(db)
self.assertIn("CREATE TABLE", latest_text)
def test_preexisting(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "preexisting.db")
with open(fn, "w"):
pass
with self.assertRaises(database.DBAlreadyExists):
database.create_db(fn)
def test_create(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "created.db")
db = database.create_db(fn)
latest_text = dump_db(db)
self.assertIn("CREATE TABLE", latest_text)
class Open(unittest.TestCase):
def test_open(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "created.db")
db1 = database.create_db(fn)
latest_text = dump_db(db1)
self.assertIn("CREATE TABLE", latest_text)
db2 = database.open_existing_db(fn)
self.assertIn("CREATE TABLE", dump_db(db2))
def test_doesnt_exist(self):
basedir = self.mktemp()
os.mkdir(basedir)
fn = os.path.join(basedir, "created.db")
with self.assertRaises(database.DBDoesntExist):
database.open_existing_db(fn)

56
src/wormhole_transit_relay/test/test_rlimits.py
View File

@ -1,56 +0,0 @@
from unittest import mock
from twisted.trial import unittest
from ..increase_rlimits import increase_rlimits
class RLimits(unittest.TestCase):
def test_rlimit(self):
def patch_r(name, *args, **kwargs):
return mock.patch("wormhole_transit_relay.increase_rlimits." + name, *args, **kwargs)
fakelog = []
def checklog(*expected):
self.assertEqual(fakelog, list(expected))
fakelog[:] = []
NF = "NOFILE"
mock_NF = patch_r("RLIMIT_NOFILE", NF)
with patch_r("log.msg", fakelog.append):
with patch_r("getrlimit", None):
increase_rlimits()
checklog("unable to import 'resource', leaving rlimit alone")
with mock_NF:
with patch_r("getrlimit", return_value=(20000, 30000)) as gr:
increase_rlimits()
self.assertEqual(gr.mock_calls, [mock.call(NF)])
checklog("RLIMIT_NOFILE.soft was 20000, leaving it alone")
with patch_r("getrlimit", return_value=(10, 30000)) as gr:
with patch_r("setrlimit", side_effect=TypeError("other")):
with patch_r("log.err") as err:
increase_rlimits()
self.assertEqual(err.mock_calls, [mock.call()])
checklog("changing RLIMIT_NOFILE from (10,30000) to (30000,30000)",
"other error during setrlimit, leaving it alone")
for maxlimit in [40000, 20000, 9000, 2000, 1000]:
def setrlimit(which, newlimit):
if newlimit[0] > maxlimit:
raise ValueError("nope")
return None
calls = []
expected = []
for tries in [30000, 10000, 3200, 1024]:
calls.append(mock.call(NF, (tries, 30000)))
expected.append("changing RLIMIT_NOFILE from (10,30000) to (%d,30000)" % tries)
if tries > maxlimit:
expected.append("error during setrlimit: nope")
else:
expected.append("setrlimit successful")
break
else:
expected.append("unable to change rlimit, leaving it alone")
with patch_r("setrlimit", side_effect=setrlimit) as sr:
increase_rlimits()
self.assertEqual(sr.mock_calls, calls)
checklog(*expected)

70
src/wormhole_transit_relay/test/test_service.py
View File

@ -1,70 +0,0 @@
from twisted.trial import unittest
from unittest import mock
from twisted.application.service import MultiService
from autobahn.twisted.websocket import WebSocketServerFactory
from .. import server_tap
class Service(unittest.TestCase):
def test_defaults(self):
o = server_tap.Options()
o.parseOptions([])
with mock.patch("wormhole_transit_relay.server_tap.create_usage_tracker") as t:
s = server_tap.makeService(o)
self.assertEqual(t.mock_calls,
[mock.call(blur_usage=None,
log_file=None, usage_db=None)])
self.assertIsInstance(s, MultiService)
def test_blur(self):
o = server_tap.Options()
o.parseOptions(["--blur-usage=60"])
with mock.patch("wormhole_transit_relay.server_tap.create_usage_tracker") as t:
server_tap.makeService(o)
self.assertEqual(t.mock_calls,
[mock.call(blur_usage=60,
log_file=None, usage_db=None)])
def test_log_fd(self):
o = server_tap.Options()
o.parseOptions(["--log-fd=99"])
fd = object()
with mock.patch("wormhole_transit_relay.server_tap.create_usage_tracker") as t:
with mock.patch("wormhole_transit_relay.server_tap.os.fdopen",
return_value=fd) as f:
server_tap.makeService(o)
self.assertEqual(f.mock_calls, [mock.call(99, "w")])
self.assertEqual(t.mock_calls,
[mock.call(blur_usage=None,
log_file=fd, usage_db=None)])
def test_websocket(self):
"""
A websocket factory is created when passing --websocket
"""
o = server_tap.Options()
o.parseOptions(["--websocket=tcp:4004"])
services = server_tap.makeService(o)
self.assertTrue(
any(
isinstance(s.factory, WebSocketServerFactory)
for s in services.services
)
)
def test_websocket_explicit_url(self):
"""
A websocket factory is created with --websocket and
--websocket-url
"""
o = server_tap.Options()
o.parseOptions([
"--websocket=tcp:4004",
"--websocket-url=ws://example.com:4004",
])
services = server_tap.makeService(o)
self.assertTrue(
any(
isinstance(s.factory, WebSocketServerFactory)
for s in services.services
)
)

165
src/wormhole_transit_relay/test/test_stats.py
View File

@ -1,130 +1,59 @@
import os, io, json
from unittest import mock
from __future__ import print_function, unicode_literals
import os, json
import mock
from twisted.trial import unittest
from ..transit_server import Transit
from ..usage import create_usage_tracker
from .. import database
class DB(unittest.TestCase):
def test_db(self):
T = 1519075308.0
class Timer:
t = T
def __call__(self):
return self.t
get_time = Timer()
class UsageLog(unittest.TestCase):
def test_log(self):
d = self.mktemp()
os.mkdir(d)
usage_db = os.path.join(d, "usage.sqlite")
db = database.get_db(usage_db)
t = Transit(
create_usage_tracker(blur_usage=None, log_file=None, usage_db=db),
get_time,
)
self.assertEqual(len(t.usage._backends), 1)
usage = list(t.usage._backends)[0]
usage_logfile = os.path.join(d, "usage.log")
def read():
with open(usage_logfile, "r") as f:
return [json.loads(line) for line in f.readlines()]
t = Transit(None, usage_logfile, None)
t.recordUsage(started=123, result="happy", total_bytes=100,
total_time=10, waiting_time=2)
self.assertEqual(read(), [dict(started=123, mood="happy",
total_time=10, waiting_time=2,
total_bytes=100)])
get_time.t = T + 1
usage.record_usage(started=123, mood="happy", total_bytes=100,
total_time=10, waiting_time=2)
t.update_stats()
t.recordUsage(started=150, result="errory", total_bytes=200,
total_time=11, waiting_time=3)
self.assertEqual(read(), [dict(started=123, mood="happy",
total_time=10, waiting_time=2,
total_bytes=100),
dict(started=150, mood="errory",
total_time=11, waiting_time=3,
total_bytes=200),
])
self.assertEqual(db.execute("SELECT * FROM `usage`").fetchall(),
[dict(result="happy", started=123,
total_bytes=100, total_time=10, waiting_time=2),
])
self.assertEqual(db.execute("SELECT * FROM `current`").fetchone(),
dict(rebooted=T+0, updated=T+1,
incomplete_bytes=0,
waiting=0, connected=0))
if False:
# the current design opens the logfile exactly once, at process
# start, in the faint hopes of surviving an exhaustion of available
# file descriptors. This should be rethought.
os.unlink(usage_logfile)
get_time.t = T + 2
usage.record_usage(started=150, mood="errory", total_bytes=200,
total_time=11, waiting_time=3)
t.update_stats()
self.assertEqual(db.execute("SELECT * FROM `usage`").fetchall(),
[dict(result="happy", started=123,
total_bytes=100, total_time=10, waiting_time=2),
dict(result="errory", started=150,
total_bytes=200, total_time=11, waiting_time=3),
])
self.assertEqual(db.execute("SELECT * FROM `current`").fetchone(),
dict(rebooted=T+0, updated=T+2,
incomplete_bytes=0,
waiting=0, connected=0))
t.recordUsage(started=200, result="lonely", total_bytes=300,
total_time=12, waiting_time=4)
self.assertEqual(read(), [dict(started=200, mood="lonely",
total_time=12, waiting_time=4,
total_bytes=300)])
get_time.t = T + 3
t.update_stats()
self.assertEqual(db.execute("SELECT * FROM `current`").fetchone(),
dict(rebooted=T+0, updated=T+3,
incomplete_bytes=0,
waiting=0, connected=0))
def test_no_db(self):
t = Transit(
create_usage_tracker(blur_usage=None, log_file=None, usage_db=None),
lambda: 0,
)
self.assertEqual(0, len(t.usage._backends))
class LogToStdout(unittest.TestCase):
class StandardLogfile(unittest.TestCase):
def test_log(self):
# emit lines of JSON to log_file, if set
log_file = io.StringIO()
t = Transit(
create_usage_tracker(blur_usage=None, log_file=log_file, usage_db=None),
lambda: 0,
)
with mock.patch("time.time", return_value=133):
t.usage.record(
started=123,
buddy_started=125,
result="happy",
bytes_sent=100,
buddy_bytes=0,
)
self.assertEqual(json.loads(log_file.getvalue()),
{"started": 123, "total_time": 10,
"waiting_time": 2, "total_bytes": 100,
"mood": "happy"})
def test_log_blurred(self):
# if blurring is enabled, timestamps should be rounded to the
# requested amount, and sizes should be rounded up too
log_file = io.StringIO()
t = Transit(
create_usage_tracker(blur_usage=60, log_file=log_file, usage_db=None),
lambda: 0,
)
with mock.patch("time.time", return_value=123 + 10):
t.usage.record(
started=123,
buddy_started=125,
result="happy",
bytes_sent=11999,
buddy_bytes=0,
)
print(log_file.getvalue())
self.assertEqual(json.loads(log_file.getvalue()),
{"started": 120, "total_time": 10,
"waiting_time": 2, "total_bytes": 20000,
"mood": "happy"})
# the default, when _blur_usage is None, will log to twistd.log
t = Transit(blur_usage=None, usage_logfile=None, stats_file=None)
with mock.patch("twisted.python.log.msg") as m:
t.recordUsage(started=123, result="happy", total_bytes=100,
total_time=10, waiting_time=2)
self.assertEqual(m.mock_calls, [mock.call(format="Transit.recordUsage {bytes}B", bytes=100)])
def test_do_not_log(self):
t = Transit(
create_usage_tracker(blur_usage=60, log_file=None, usage_db=None),
lambda: 0,
)
t.usage.record(
started=123,
buddy_started=124,
result="happy",
bytes_sent=11999,
buddy_bytes=12,
)
# the default, when _blur_usage is None, will log to twistd.log
t = Transit(blur_usage=60, usage_logfile=None, stats_file=None)
with mock.patch("twisted.python.log.msg") as m:
t.recordUsage(started=123, result="happy", total_bytes=100,
total_time=10, waiting_time=2)
self.assertEqual(m.mock_calls, [])

773
src/wormhole_transit_relay/test/test_transit_server.py
View File

@ -1,273 +1,256 @@
from __future__ import print_function, unicode_literals
from binascii import hexlify
from twisted.trial import unittest
from twisted.test import iosim
from autobahn.twisted.websocket import (
WebSocketServerFactory,
WebSocketClientFactory,
WebSocketClientProtocol,
)
from autobahn.twisted.testing import (
create_pumper,
MemoryReactorClockResolver,
)
from autobahn.exception import Disconnected
from zope.interface import implementer
from .common import (
ServerBase,
IRelayTestClient,
)
from ..usage import (
MemoryUsageRecorder,
blur_size,
)
from ..transit_server import (
WebSocketTransitConnection,
TransitServerState,
)
from twisted.internet import protocol, reactor, defer
from twisted.internet.endpoints import clientFromString, connectProtocol
from .common import ServerBase
from .. import transit_server
class Accumulator(protocol.Protocol):
def __init__(self):
self.data = b""
self.count = 0
self._wait = None
self._disconnect = defer.Deferred()
def waitForBytes(self, more):
assert self._wait is None
self.count = more
self._wait = defer.Deferred()
self._check_done()
return self._wait
def dataReceived(self, data):
self.data = self.data + data
self._check_done()
def _check_done(self):
if self._wait and len(self.data) >= self.count:
d = self._wait
self._wait = None
d.callback(self)
def connectionLost(self, why):
if self._wait:
self._wait.errback(RuntimeError("closed"))
self._disconnect.callback(None)
def handshake(token, side=None):
hs = b"please relay " + hexlify(token)
if side is not None:
hs += b" for side " + hexlify(side)
hs += b"\n"
return hs
class _Transit:
def count(self):
return sum([
len(potentials)
for potentials
in self._transit_server.pending_requests._requests.values()
])
class Transit(ServerBase, unittest.TestCase):
def test_blur_size(self):
self.failUnlessEqual(blur_size(0), 0)
self.failUnlessEqual(blur_size(1), 10e3)
self.failUnlessEqual(blur_size(10e3), 10e3)
self.failUnlessEqual(blur_size(10e3+1), 20e3)
self.failUnlessEqual(blur_size(15e3), 20e3)
self.failUnlessEqual(blur_size(20e3), 20e3)
self.failUnlessEqual(blur_size(1e6), 1e6)
self.failUnlessEqual(blur_size(1e6+1), 2e6)
self.failUnlessEqual(blur_size(1.5e6), 2e6)
self.failUnlessEqual(blur_size(2e6), 2e6)
self.failUnlessEqual(blur_size(900e6), 900e6)
self.failUnlessEqual(blur_size(1000e6), 1000e6)
self.failUnlessEqual(blur_size(1050e6), 1100e6)
self.failUnlessEqual(blur_size(1100e6), 1100e6)
self.failUnlessEqual(blur_size(1150e6), 1200e6)
blur = transit_server.blur_size
self.failUnlessEqual(blur(0), 0)
self.failUnlessEqual(blur(1), 10e3)
self.failUnlessEqual(blur(10e3), 10e3)
self.failUnlessEqual(blur(10e3+1), 20e3)
self.failUnlessEqual(blur(15e3), 20e3)
self.failUnlessEqual(blur(20e3), 20e3)
self.failUnlessEqual(blur(1e6), 1e6)
self.failUnlessEqual(blur(1e6+1), 2e6)
self.failUnlessEqual(blur(1.5e6), 2e6)
self.failUnlessEqual(blur(2e6), 2e6)
self.failUnlessEqual(blur(900e6), 900e6)
self.failUnlessEqual(blur(1000e6), 1000e6)
self.failUnlessEqual(blur(1050e6), 1100e6)
self.failUnlessEqual(blur(1100e6), 1100e6)
self.failUnlessEqual(blur(1150e6), 1200e6)
@defer.inlineCallbacks
def test_register(self):
p1 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
side1 = b"\x01"*8
a1.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
p1.send(handshake(token1, side1))
self.flush()
# let that arrive
while self.count() == 0:
yield self.wait()
self.assertEqual(self.count(), 1)
p1.disconnect()
self.flush()
a1.transport.loseConnection()
# let that get removed
while self.count() > 0:
yield self.wait()
self.assertEqual(self.count(), 0)
# the token should be removed too
self.assertEqual(len(self._transit_server.pending_requests._requests), 0)
self.assertEqual(len(self._transit_server._pending_requests), 0)
@defer.inlineCallbacks
def test_both_unsided(self):
p1 = self.new_protocol()
p2 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
a2 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
p1.send(handshake(token1, side=None))
self.flush()
p2.send(handshake(token1, side=None))
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) + b"\n")
a2.transport.write(b"please relay " + hexlify(token1) + b"\n")
# a correct handshake yields an ack, after which we can send
exp = b"ok\n"
self.assertEqual(p1.get_received_data(), exp)
self.assertEqual(p2.get_received_data(), exp)
p1.reset_received_data()
p2.reset_received_data()
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
s1 = b"data1"
p1.send(s1)
self.flush()
self.assertEqual(p2.get_received_data(), s1)
a1.transport.write(s1)
p1.disconnect()
self.flush()
exp = b"ok\n"
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
# all data they sent after the handshake should be given to us
exp = b"ok\n"+s1
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
a1.transport.loseConnection()
a2.transport.loseConnection()
@defer.inlineCallbacks
def test_sided_unsided(self):
p1 = self.new_protocol()
p2 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
a2 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
side1 = b"\x01"*8
p1.send(handshake(token1, side=side1))
self.flush()
p2.send(handshake(token1, side=None))
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
a2.transport.write(b"please relay " + hexlify(token1) + b"\n")
# a correct handshake yields an ack, after which we can send
exp = b"ok\n"
self.assertEqual(p1.get_received_data(), exp)
self.assertEqual(p2.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
s1 = b"data1"
a1.transport.write(s1)
p1.reset_received_data()
p2.reset_received_data()
exp = b"ok\n"
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
# all data they sent after the handshake should be given to us
s1 = b"data1"
p1.send(s1)
self.flush()
self.assertEqual(p2.get_received_data(), s1)
exp = b"ok\n"+s1
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
p1.disconnect()
self.flush()
a1.transport.loseConnection()
a2.transport.loseConnection()
@defer.inlineCallbacks
def test_unsided_sided(self):
p1 = self.new_protocol()
p2 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
a2 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
side1 = b"\x01"*8
p1.send(handshake(token1, side=None))
p2.send(handshake(token1, side=side1))
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) + b"\n")
a2.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
# a correct handshake yields an ack, after which we can send
exp = b"ok\n"
self.assertEqual(p1.get_received_data(), exp)
self.assertEqual(p2.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
s1 = b"data1"
a1.transport.write(s1)
p1.reset_received_data()
p2.reset_received_data()
exp = b"ok\n"
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
# all data they sent after the handshake should be given to us
s1 = b"data1"
p1.send(s1)
self.flush()
self.assertEqual(p2.get_received_data(), s1)
exp = b"ok\n"+s1
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
p1.disconnect()
p2.disconnect()
a1.transport.loseConnection()
a2.transport.loseConnection()
@defer.inlineCallbacks
def test_both_sided(self):
p1 = self.new_protocol()
p2 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
a2 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
side1 = b"\x01"*8
side2 = b"\x02"*8
p1.send(handshake(token1, side=side1))
self.flush()
p2.send(handshake(token1, side=side2))
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
a2.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side2) + b"\n")
# a correct handshake yields an ack, after which we can send
exp = b"ok\n"
self.assertEqual(p1.get_received_data(), exp)
self.assertEqual(p2.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
s1 = b"data1"
a1.transport.write(s1)
p1.reset_received_data()
p2.reset_received_data()
exp = b"ok\n"
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
# all data they sent after the handshake should be given to us
s1 = b"data1"
p1.send(s1)
self.flush()
self.assertEqual(p2.get_received_data(), s1)
exp = b"ok\n"+s1
yield a2.waitForBytes(len(exp))
self.assertEqual(a2.data, exp)
p1.disconnect()
p2.disconnect()
a1.transport.loseConnection()
a2.transport.loseConnection()
def count(self):
return sum([len(potentials)
for potentials
in self._transit_server._pending_requests.values()])
def wait(self):
d = defer.Deferred()
reactor.callLater(0.001, d.callback, None)
return d
@defer.inlineCallbacks
def test_ignore_same_side(self):
p1 = self.new_protocol()
p2 = self.new_protocol()
p3 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
a2 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
side1 = b"\x01"*8
p1.send(handshake(token1, side=side1))
self.flush()
self.assertEqual(self.count(), 1)
p2.send(handshake(token1, side=side1))
self.flush()
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
# let that arrive
while self.count() == 0:
yield self.wait()
a2.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
# let that arrive
while self.count() == 1:
yield self.wait()
self.assertEqual(self.count(), 2) # same-side connections don't match
# when the second side arrives, the spare first connection should be
# closed
side2 = b"\x02"*8
p3.send(handshake(token1, side=side2))
self.flush()
self.assertEqual(self.count(), 0)
self.assertEqual(len(self._transit_server.pending_requests._requests), 0)
self.assertEqual(len(self._transit_server.active_connections._connections), 2)
# That will trigger a disconnect on exactly one of (p1 or p2).
# The other connection should still be connected
self.assertEqual(sum([int(t.connected) for t in [p1, p2]]), 1)
a1.transport.loseConnection()
a2.transport.loseConnection()
p1.disconnect()
p2.disconnect()
p3.disconnect()
def test_bad_handshake_old(self):
p1 = self.new_protocol()
token1 = b"\x00"*32
p1.send(b"please DELAY " + hexlify(token1) + b"\n")
self.flush()
exp = b"bad handshake\n"
self.assertEqual(p1.get_received_data(), exp)
p1.disconnect()
def test_bad_handshake_old_slow(self):
p1 = self.new_protocol()
p1.send(b"please DELAY ")
self.flush()
# As in test_impatience_new_slow, the current state machine has code
# that can only be reached if we insert a stall here, so dataReceived
# gets called twice. Hopefully we can delete this test once
# dataReceived is refactored to remove that state.
@defer.inlineCallbacks
def test_bad_handshake(self):
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
# the server waits for the exact number of bytes in the expected
# handshake message. to trigger "bad handshake", we must match.
p1.send(hexlify(token1) + b"\n")
self.flush()
a1.transport.write(b"please DELAY " + hexlify(token1) + b"\n")
exp = b"bad handshake\n"
self.assertEqual(p1.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
p1.disconnect()
def test_bad_handshake_new(self):
p1 = self.new_protocol()
token1 = b"\x00"*32
side1 = b"\x01"*8
# the server waits for the exact number of bytes in the expected
# handshake message. to trigger "bad handshake", we must match.
p1.send(b"please DELAY " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
self.flush()
exp = b"bad handshake\n"
self.assertEqual(p1.get_received_data(), exp)
p1.disconnect()
a1.transport.loseConnection()
@defer.inlineCallbacks
def test_binary_handshake(self):
p1 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
binary_bad_handshake = b"\x00\x01\xe0\x0f\n\xff"
# the embedded \n makes the server trigger early, before the full
@ -276,412 +259,42 @@ class _Transit:
# UnicodeDecodeError when it tried to coerce the incoming handshake
# to unicode, due to the ("\n" in buf) check. This was fixed to use
# (b"\n" in buf). This exercises the old failure.
p1.send(binary_bad_handshake)
self.flush()
a1.transport.write(binary_bad_handshake)
exp = b"bad handshake\n"
self.assertEqual(p1.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
p1.disconnect()
a1.transport.loseConnection()
@defer.inlineCallbacks
def test_impatience_old(self):
p1 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
# sending too many bytes is impatience.
p1.send(b"please relay " + hexlify(token1))
p1.send(b"\nNOWNOWNOW")
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) + b"\nNOWNOWNOW")
exp = b"impatient\n"
self.assertEqual(p1.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
p1.disconnect()
a1.transport.loseConnection()
@defer.inlineCallbacks
def test_impatience_new(self):
p1 = self.new_protocol()
ep = clientFromString(reactor, self.transit)
a1 = yield connectProtocol(ep, Accumulator())
token1 = b"\x00"*32
side1 = b"\x01"*8
# sending too many bytes is impatience.
p1.send(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1))
p1.send(b"\nNOWNOWNOW")
self.flush()
a1.transport.write(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\nNOWNOWNOW")
exp = b"impatient\n"
self.assertEqual(p1.get_received_data(), exp)
yield a1.waitForBytes(len(exp))
self.assertEqual(a1.data, exp)
p1.disconnect()
def test_impatience_new_slow(self):
p1 = self.new_protocol()
# For full coverage, we need dataReceived to see a particular framing
# of these two pieces of data, and ITCPTransport doesn't have flush()
# (which probably wouldn't work anyways). For now, force a 100ms
# stall between the two writes. I tried setTcpNoDelay(True) but it
# didn't seem to help without the stall. The long-term fix is to
# rewrite dataReceived() to remove the multiple "impatient"
# codepaths, deleting the particular clause that this test exercises,
# then remove this test.
token1 = b"\x00"*32
side1 = b"\x01"*8
# sending too many bytes is impatience.
p1.send(b"please relay " + hexlify(token1) +
b" for side " + hexlify(side1) + b"\n")
self.flush()
p1.send(b"NOWNOWNOW")
self.flush()
exp = b"impatient\n"
self.assertEqual(p1.get_received_data(), exp)
p1.disconnect()
def test_short_handshake(self):
p1 = self.new_protocol()
# hang up before sending a complete handshake
p1.send(b"short")
self.flush()
p1.disconnect()
def test_empty_handshake(self):
p1 = self.new_protocol()
# hang up before sending anything
p1.disconnect()
class TransitWithLogs(_Transit, ServerBase, unittest.TestCase):
log_requests = True
def new_protocol(self):
return self.new_protocol_tcp()
class TransitWithoutLogs(_Transit, ServerBase, unittest.TestCase):
log_requests = False
def new_protocol(self):
return self.new_protocol_tcp()
def _new_protocol_ws(transit_server, log_requests):
"""
Internal helper for test-suites that need to provide WebSocket
client/server pairs.
:returns: a 2-tuple: (iosim.IOPump, protocol)
"""
ws_factory = WebSocketServerFactory("ws://localhost:4002")
ws_factory.protocol = WebSocketTransitConnection
ws_factory.transit = transit_server
ws_factory.log_requests = log_requests
ws_protocol = ws_factory.buildProtocol(('127.0.0.1', 0))
@implementer(IRelayTestClient)
class TransitWebSocketClientProtocol(WebSocketClientProtocol):
_received = b""
connected = False
def connectionMade(self):
self.connected = True
return super(TransitWebSocketClientProtocol, self).connectionMade()
def connectionLost(self, reason):
self.connected = False
return super(TransitWebSocketClientProtocol, self).connectionLost(reason)
def onMessage(self, data, isBinary):
self._received = self._received + data
def send(self, data):
self.sendMessage(data, True)
def get_received_data(self):
return self._received
def reset_received_data(self):
self._received = b""
def disconnect(self):
self.sendClose(1000, True)
client_factory = WebSocketClientFactory()
client_factory.protocol = TransitWebSocketClientProtocol
client_protocol = client_factory.buildProtocol(('127.0.0.1', 31337))
client_protocol.disconnect = client_protocol.dropConnection
pump = iosim.connect(
ws_protocol,
iosim.makeFakeServer(ws_protocol),
client_protocol,
iosim.makeFakeClient(client_protocol),
)
return pump, client_protocol
class TransitWebSockets(_Transit, ServerBase, unittest.TestCase):
def new_protocol(self):
return self.new_protocol_ws()
def new_protocol_ws(self):
pump, proto = _new_protocol_ws(self._transit_server, self.log_requests)
self._pumps.append(pump)
return proto
def test_websocket_to_tcp(self):
"""
One client is WebSocket and one is TCP
"""
p1 = self.new_protocol_ws()
p2 = self.new_protocol_tcp()
token1 = b"\x00"*32
side1 = b"\x01"*8
side2 = b"\x02"*8
p1.send(handshake(token1, side=side1))
self.flush()
p2.send(handshake(token1, side=side2))
self.flush()
# a correct handshake yields an ack, after which we can send
exp = b"ok\n"
self.assertEqual(p1.get_received_data(), exp)
self.assertEqual(p2.get_received_data(), exp)
p1.reset_received_data()
p2.reset_received_data()
# all data they sent after the handshake should be given to us
s1 = b"data1"
p1.send(s1)
self.flush()
self.assertEqual(p2.get_received_data(), s1)
p1.disconnect()
p2.disconnect()
self.flush()
def test_bad_handshake_old_slow(self):
"""
This test only makes sense for TCP
"""
def test_send_closed_partner(self):
"""
Sending data to a closed partner causes an error that propogates
to the sender.
"""
p1 = self.new_protocol()
p2 = self.new_protocol()
# set up a successful connection
token = b"a" * 32
p1.send(handshake(token))
p2.send(handshake(token))
self.flush()
# p2 loses connection, then p1 sends a message
p2.transport.loseConnection()
self.flush()
# at this point, p1 learns that p2 is disconnected (because it
# tried to relay "a message" but failed)
# try to send more (our partner p2 is gone now though so it
# should be an immediate error)
with self.assertRaises(Disconnected):
p1.send(b"more message")
self.flush()
class Usage(ServerBase, unittest.TestCase):
log_requests = True
def setUp(self):
super(Usage, self).setUp()
self._usage = MemoryUsageRecorder()
self._transit_server.usage.add_backend(self._usage)
def new_protocol(self):
return self.new_protocol_tcp()
def test_empty(self):
p1 = self.new_protocol()
# hang up before sending anything
p1.disconnect()
self.flush()
# that will log the "empty" usage event
self.assertEqual(len(self._usage.events), 1, self._usage)
self.assertEqual(self._usage.events[0]["mood"], "empty", self._usage)
def test_short(self):
# Note: this test only runs on TCP clients because WebSockets
# already does framing (so it's either "a bad handshake" or
# there's no handshake at all yet .. you can't have a "short"
# one).
p1 = self.new_protocol()
# hang up before sending a complete handshake
p1.send(b"short")
p1.disconnect()
self.flush()
# that will log the "empty" usage event
self.assertEqual(len(self._usage.events), 1, self._usage)
self.assertEqual("empty", self._usage.events[0]["mood"])
def test_errory(self):
p1 = self.new_protocol()
p1.send(b"this is a very bad handshake\n")
self.flush()
# that will log the "errory" usage event, then drop the connection
p1.disconnect()
self.assertEqual(len(self._usage.events), 1, self._usage)
self.assertEqual(self._usage.events[0]["mood"], "errory", self._usage)
def test_lonely(self):
p1 = self.new_protocol()
token1 = b"\x00"*32
side1 = b"\x01"*8
p1.send(handshake(token1, side=side1))
self.flush()
# now we disconnect before the peer connects
p1.disconnect()
self.flush()
self.assertEqual(len(self._usage.events), 1, self._usage)
self.assertEqual(self._usage.events[0]["mood"], "lonely", self._usage)
self.assertIdentical(self._usage.events[0]["waiting_time"], None)
def test_one_happy_one_jilted(self):
p1 = self.new_protocol()
p2 = self.new_protocol()
token1 = b"\x00"*32
side1 = b"\x01"*8
side2 = b"\x02"*8
p1.send(handshake(token1, side=side1))
self.flush()
p2.send(handshake(token1, side=side2))
self.flush()
self.assertEqual(self._usage.events, []) # no events yet
p1.send(b"\x00" * 13)
self.flush()
p2.send(b"\xff" * 7)
self.flush()
p1.disconnect()
self.flush()
self.assertEqual(len(self._usage.events), 1, self._usage)
self.assertEqual(self._usage.events[0]["mood"], "happy", self._usage)
self.assertEqual(self._usage.events[0]["total_bytes"], 20)
self.assertNotIdentical(self._usage.events[0]["waiting_time"], None)
def test_redundant(self):
p1a = self.new_protocol()
p1b = self.new_protocol()
p1c = self.new_protocol()
p2 = self.new_protocol()
token1 = b"\x00"*32
side1 = b"\x01"*8
side2 = b"\x02"*8
p1a.send(handshake(token1, side=side1))
self.flush()
p1b.send(handshake(token1, side=side1))
self.flush()
# connect and disconnect a third client (for side1) to exercise the
# code that removes a pending connection without removing the entire
# token
p1c.send(handshake(token1, side=side1))
p1c.disconnect()
self.flush()
self.assertEqual(len(self._usage.events), 1, self._usage)
self.assertEqual(self._usage.events[0]["mood"], "lonely")
p2.send(handshake(token1, side=side2))
self.flush()
self.assertEqual(len(self._transit_server.pending_requests._requests), 0)
self.assertEqual(len(self._usage.events), 2, self._usage)
self.assertEqual(self._usage.events[1]["mood"], "redundant")
# one of the these is unecessary, but probably harmless
p1a.disconnect()
p1b.disconnect()
self.flush()
self.assertEqual(len(self._usage.events), 3, self._usage)
self.assertEqual(self._usage.events[2]["mood"], "happy")
class UsageWebSockets(Usage):
"""
All the tests of 'Usage' except with a WebSocket (instead of TCP)
transport.
This overrides ServerBase.new_protocol to achieve this. It might
be nicer to parametrize these tests in a way that doesn't use
inheritance .. but all the support etc classes are set up that way
already.
"""
def setUp(self):
super(UsageWebSockets, self).setUp()
self._pump = create_pumper()
self._reactor = MemoryReactorClockResolver()
return self._pump.start()
def tearDown(self):
return self._pump.stop()
def new_protocol(self):
return self.new_protocol_ws()
def new_protocol_ws(self):
pump, proto = _new_protocol_ws(self._transit_server, self.log_requests)
self._pumps.append(pump)
return proto
def test_short(self):
"""
This test essentially just tests the framing of the line-oriented
TCP protocol; it doesnt' make sense for the WebSockets case
because WS handles frameing: you either sent a 'bad handshake'
because it is semantically invalid or no handshake (yet).
"""
def test_send_non_binary_message(self):
"""
A non-binary WebSocket message is an error
"""
ws_factory = WebSocketServerFactory("ws://localhost:4002")
ws_factory.protocol = WebSocketTransitConnection
ws_protocol = ws_factory.buildProtocol(('127.0.0.1', 0))
with self.assertRaises(ValueError):
ws_protocol.onMessage(u"foo", isBinary=False)
class State(unittest.TestCase):
"""
Tests related to server_state.TransitServerState
"""
def setUp(self):
self.state = TransitServerState(None, None)
def test_empty_token(self):
self.assertEqual(
"-",
self.state.get_token(),
)
a1.transport.loseConnection()

564
src/wormhole_transit_relay/transit_server.py
View File

@ -1,9 +1,8 @@
import re
import time
from __future__ import print_function, unicode_literals
import os, re, time, json
from twisted.python import log
from twisted.protocols.basic import LineReceiver
from autobahn.twisted.websocket import WebSocketServerProtocol
from twisted.internet import protocol
from .database import get_db
SECONDS = 1.0
MINUTE = 60*SECONDS
@ -11,256 +10,363 @@ HOUR = 60*MINUTE
DAY = 24*HOUR
MB = 1000*1000
def round_to(size, coarseness):
return int(coarseness*(1+int((size-1)/coarseness)))
from wormhole_transit_relay.server_state import (
TransitServerState,
PendingRequests,
ActiveConnections,
ITransitClient,
)
from zope.interface import implementer
def blur_size(size):
if size == 0:
return 0
if size < 1e6:
return round_to(size, 10e3)
if size < 1e9:
return round_to(size, 1e6)
return round_to(size, 100e6)
@implementer(ITransitClient)
class TransitConnection(LineReceiver):
delimiter = b'\n'
# maximum length of a line we will accept before the handshake is complete.
# This must be >= to the longest possible handshake message.
MAX_LENGTH = 1024
started_time = None
def send(self, data):
"""
ITransitClient API
"""
self.transport.write(data)
def disconnect(self):
"""
ITransitClient API
"""
self.transport.loseConnection()
def connect_partner(self, other):
"""
ITransitClient API
"""
self._buddy = other
self._buddy._client.transport.registerProducer(self.transport, True)
def disconnect_partner(self):
"""
ITransitClient API
"""
assert self._buddy is not None, "internal error: no buddy"
if self.factory.log_requests:
log.msg("buddy_disconnected {}".format(self._buddy.get_token()))
self._buddy._client.disconnect()
class TransitConnection(protocol.Protocol):
def __init__(self):
self._got_token = False
self._got_side = False
self._token_buffer = b""
self._sent_ok = False
self._buddy = None
self._total_sent = 0
def describeToken(self):
d = "-"
if self._got_token:
d = self._got_token[:16].decode("ascii")
if self._got_side:
d += "-" + self._got_side.decode("ascii")
else:
d += "-<unsided>"
return d
def connectionMade(self):
# ideally more like self._reactor.seconds() ... but Twisted
# doesn't have a good way to get the reactor for a protocol
# (besides "use the global one")
self.started_time = time.time()
self._state = TransitServerState(
self.factory.transit.pending_requests,
self.factory.transit.usage,
)
self._state.connection_made(self)
self.transport.setTcpKeepAlive(True)
self._started = time.time()
self._log_requests = self.factory._log_requests
# uncomment to turn on state-machine tracing
# def tracer(oldstate, theinput, newstate):
# print("TRACE: {}: {} --{}--> {}".format(id(self), oldstate, theinput, newstate))
# self._state.set_trace_function(tracer)
def dataReceived(self, data):
if self._sent_ok:
# We are an IPushProducer to our buddy's IConsumer, so they'll
# throttle us (by calling pauseProducing()) when their outbound
# buffer is full (e.g. when their downstream pipe is full). In
# practice, this buffers about 10MB per connection, after which
# point the sender will only transmit data as fast as the
# receiver can handle it.
self._total_sent += len(data)
self._buddy.transport.write(data)
return
if self._got_token: # but not yet sent_ok
self.transport.write(b"impatient\n")
if self._log_requests:
log.msg("transit impatience failure")
return self.disconnect() # impatience yields failure
# else this should be (part of) the token
self._token_buffer += data
buf = self._token_buffer
def lineReceived(self, line):
"""
LineReceiver API
"""
# old: "please relay {64}\n"
token = None
old = re.search(br"^please relay (\w{64})$", line)
if old:
token = old.group(1)
self._state.please_relay(token)
# new: "please relay {64} for side {16}\n"
new = re.search(br"^please relay (\w{64}) for side (\w{16})$", line)
if new:
token = new.group(1)
side = new.group(2)
self._state.please_relay_for_side(token, side)
(old, handshake_len, token) = self._check_old_handshake(buf)
assert old in ("yes", "waiting", "no")
if old == "yes":
# remember they aren't supposed to send anything past their
# handshake until we've said go
if len(buf) > handshake_len:
self.transport.write(b"impatient\n")
if self._log_requests:
log.msg("transit impatience failure")
return self.disconnect() # impatience yields failure
return self._got_handshake(token, None)
(new, handshake_len, token, side) = self._check_new_handshake(buf)
assert new in ("yes", "waiting", "no")
if new == "yes":
if len(buf) > handshake_len:
self.transport.write(b"impatient\n")
if self._log_requests:
log.msg("transit impatience failure")
return self.disconnect() # impatience yields failure
return self._got_handshake(token, side)
if (old == "no" and new == "no"):
self.transport.write(b"bad handshake\n")
if self._log_requests:
log.msg("transit handshake failure")
return self.disconnect() # incorrectness yields failure
# else we'll keep waiting
if token is None:
self._state.bad_token()
else:
self.setRawMode()
def _check_old_handshake(self, buf):
# old: "please relay {64}\n"
# return ("yes", handshake, token) if buf contains an old-style handshake
# return ("waiting", None, None) if it might eventually contain one
# return ("no", None, None) if it could never contain one
wanted = len("please relay \n")+32*2
if len(buf) < wanted-1 and b"\n" in buf:
return ("no", None, None)
if len(buf) < wanted:
return ("waiting", None, None)
def rawDataReceived(self, data):
"""
LineReceiver API
"""
# We are an IPushProducer to our buddy's IConsumer, so they'll
# throttle us (by calling pauseProducing()) when their outbound
# buffer is full (e.g. when their downstream pipe is full). In
# practice, this buffers about 10MB per connection, after which
# point the sender will only transmit data as fast as the
# receiver can handle it.
self._state.got_bytes(data)
mo = re.search(br"^please relay (\w{64})\n", buf, re.M)
if mo:
token = mo.group(1)
return ("yes", wanted, token)
return ("no", None, None)
def _check_new_handshake(self, buf):
# new: "please relay {64} for side {16}\n"
wanted = len("please relay for side \n")+32*2+8*2
if len(buf) < wanted-1 and b"\n" in buf:
return ("no", None, None, None)
if len(buf) < wanted:
return ("waiting", None, None, None)
mo = re.search(br"^please relay (\w{64}) for side (\w{16})\n", buf, re.M)
if mo:
token = mo.group(1)
side = mo.group(2)
return ("yes", wanted, token, side)
return ("no", None, None, None)
def _got_handshake(self, token, side):
self._got_token = token
self._got_side = side
self.factory.transitGotToken(token, side, self)
def buddy_connected(self, them):
self._buddy = them
self._had_buddy = True
self.transport.write(b"ok\n")
self._sent_ok = True
# Connect the two as a producer/consumer pair. We use streaming=True,
# so this expects the IPushProducer interface, and uses
# pauseProducing() to throttle, and resumeProducing() to unthrottle.
self._buddy.transport.registerProducer(self.transport, True)
# The Transit object calls buddy_connected() on both protocols, so
# there will be two producer/consumer pairs.
def buddy_disconnected(self):
if self._log_requests:
log.msg("buddy_disconnected %s" % self.describeToken())
self._buddy = None
self.transport.loseConnection()
def connectionLost(self, reason):
self._state.connection_lost()
if self._buddy:
self._buddy.buddy_disconnected() # hang up on the buddy
self.factory.transitFinished(self, self._got_token, self._got_side,
self.describeToken())
def disconnect(self):
# called when we hang up on a connection because they violated the
# protocol, or we abandon a losing connection because a different one
# from that side won
self.transport.loseConnection()
self.factory.transitFailed(self)
class Transit(object):
"""
I manage pairs of simultaneous connections to a secondary TCP port,
both forwarded to the other. Clients must begin each connection with
"please relay TOKEN for SIDE\n" (or a legacy form without the "for
SIDE"). Two connections match if they use the same TOKEN and have
different SIDEs (the redundant connections are dropped when a match is
made). Legacy connections match any with the same TOKEN, ignoring SIDE
(so two legacy connections will match each other).
PENDING, OPEN, LINGERING, EMPTY = range(4)
I will send "ok\n" when the matching connection is established, or
disconnect if no matching connection is made within MAX_WAIT_TIME
seconds. I will disconnect if you send data before the "ok\n". All data
you get after the "ok\n" will be from the other side. You will not
receive "ok\n" until the other side has also connected and submitted a
matching token (and differing SIDE).
class Channel(object):
def __init__(self, factory):
self._factory = factory
self._state = PENDING
self._connections = set() # (side, tc)
In addition, the connections will be dropped after MAXLENGTH bytes have
been sent by either side, or MAXTIME seconds have elapsed after the
matching connections were established. A future API will reveal these
limits to clients instead of causing mysterious spontaneous failures.
def gotConnection(self, side, tc):
if self._state == PENDING:
for old in self._connections:
(old_side, old_tc) = old
if ((old_side is None)
or (side is None)
or (old_side != side)):
# we found a match
if self._debug_log:
log.msg("transit relay 2: %s" % new_tc.describeToken())
self._state = OPEN
self._factory.channelOpen(self)
# drop and stop tracking the rest
self._connections.remove(old)
for (_, leftover_tc) in self._connections:
# TODO: not "errory"? the ones we drop are the parallel
# connections from the first client ('side' was the
# same), so it's not really an error. More of a "you
# lost, one of your other connections won, sorry"
leftover_tc.disconnect()
self._pending_tokens.pop(token)
These relay connections are not half-closeable (unlike full TCP
connections, applications will not receive any data after half-closing
their outgoing side). Applications must negotiate shutdown with their
peer and not close the connection until all data has finished
transferring in both directions. Applications which only need to send
data in one direction can use close() as usual.
"""
# glue the two ends together
self._active_connections.add(new_tc)
self._active_connections.add(old_tc)
new_tc.buddy_connected(old_tc)
old_tc.buddy_connected(new_tc)
return
if self._debug_log:
log.msg("transit relay 1: %s" % new_tc.describeToken())
potentials.add((new_side, new_tc))
# TODO: timer
class Transit(protocol.ServerFactory):
# I manage pairs of simultaneous connections to a secondary TCP port,
# both forwarded to the other. Clients must begin each connection with
# "please relay TOKEN for SIDE\n" (or a legacy form without the "for
# SIDE"). Two connections match if they use the same TOKEN and have
# different SIDEs (the redundant connections are dropped when a match is
# made). Legacy connections match any with the same TOKEN, ignoring SIDE
# (so two legacy connections will match each other).
# I will send "ok\n" when the matching connection is established, or
# disconnect if no matching connection is made within MAX_WAIT_TIME
# seconds. I will disconnect if you send data before the "ok\n". All data
# you get after the "ok\n" will be from the other side. You will not
# receive "ok\n" until the other side has also connected and submitted a
# matching token (and differing SIDE).
# In addition, the connections will be dropped after MAXLENGTH bytes have
# been sent by either side, or MAXTIME seconds have elapsed after the
# matching connections were established. A future API will reveal these
# limits to clients instead of causing mysterious spontaneous failures.
# These relay connections are not half-closeable (unlike full TCP
# connections, applications will not receive any data after half-closing
# their outgoing side). Applications must negotiate shutdown with their
# peer and not close the connection until all data has finished
# transferring in both directions. Applications which only need to send
# data in one direction can use close() as usual.
# TODO: unused
MAX_WAIT_TIME = 30*SECONDS
# TODO: unused
MAXLENGTH = 10*MB
# TODO: unused
MAXTIME = 60*SECONDS
protocol = TransitConnection
def __init__(self, usage, get_timestamp):
self.active_connections = ActiveConnections()
self.pending_requests = PendingRequests(self.active_connections)
self.usage = usage
self._timestamp = get_timestamp
self._rebooted = self._timestamp()
def __init__(self, blur_usage, log_stdout, usage_db):
self._blur_usage = blur_usage
self._debug_log = False
self._log_stdout = log_stdout
self._db = None
if usage_db:
self._db = get_db(usage_db)
# we don't track TransitConnections until they submit a token
def update_stats(self):
# Channels are indexed by token, and are either pending, open, or
# lingering
self._channels = {} # token -> Channel
self._pending_channels = set()
self._open_channels = set()
self._lingering_channels = set()
def transitGotToken(self, token, new_side, new_tc):
if token not in self._channels:
self._channels[token] = Channel(self)
self._channels[token].gotConnection(new_side, new_tc)
def channelOpen(self, c):
self._pending_channels.remove(c)
self._open_channels.add(c)
def channelClosed(self, c):
self._open_channels.remove(c)
self._lingering_channels.add(c)
def channelEmpty(self, c):
self._lingering_channels.remove(c)
def transitFinished(self, tc, token, side, description):
# we're called each time a TransitConnection shuts down
if token in self._pending_tokens:
side_tc = (side, tc)
if side_tc in self._pending_tokens[token]:
self._pending_tokens[token].remove(side_tc)
if not self._pending_tokens[token]: # set is now empty
del self._pending_tokens[token]
if self._debug_log:
log.msg("transitFinished %s" % (description,))
self._active_connections.discard(tc)
# Record usage. There are five cases:
# * 1: we connected, never had a buddy
# * 2: we connected first, we disconnect before the buddy
# * 3: we connected first, buddy disconnects first
# * 4: buddy connected first, we disconnect before buddy
# * 5: buddy connected first, buddy disconnects first
# whoever disconnects first gets to write the usage record (1,2,4)
finished = time.time()
if self._had_buddy:
if self._buddy: # 2,4: we disconnected first
total_bytes = self._total_sent + self._buddy._total_sent
starts = [self._started, self._buddy._started]
total_time = finished - min(starts)
waiting_time = max(starts) - min(starts)
self.factory.buddyIsLingering(self._buddy)
self.factory.recordUsage(self._started, "happy", total_bytes,
total_time, waiting_time)
else: # 3, 5: we disconnected last
self.factory.doneLingering(self)
else: # 1: we were the only one
total_time = finished - self._started
self.factory.recordUsage(self._started, "lonely", 0,
total_time, None)
def transitFailed(self, p):
if self._debug_log:
log.msg("transitFailed %r" % p)
finished = time.time()
total_time = finished - self._started
self.factory.recordUsage(self._started, "errory", 0,
total_time, None)
pass
def buddyIsLingering(self, buddy_tc):
self._active_connections.remove(buddy_tc)
self._lingering_connections.add(buddy_tc)
def doneLingering(self, old_tc):
self._lingering_connections.remove(buddy_tc)
def recordUsage(self, started, result, total_bytes,
total_time, waiting_time):
if self._debug_log:
log.msg(format="Transit.recordUsage {bytes}B", bytes=total_bytes)
if self._blur_usage:
started = self._blur_usage * (started // self._blur_usage)
total_bytes = blur_size(total_bytes)
if self._log_stdout:
data = {"started": started,
"total_time": total_time,
"waiting_time": waiting_time,
"total_bytes": total_bytes,
"mood": result,
}
sys.stdout.write(json.dumps(data))
sys.stdout.write("\n")
sys.stdout.flush()
if self._db:
self._db.execute("INSERT INTO `usage`"
" (`started`, `total_time`, `waiting_time`,"
" `total_bytes`, `result`)"
" VALUES (?,?,?, ?,?)",
(started, total_time, waiting_time,
total_bytes, result))
self._update_stats()
self._db.commit()
def _update_stats(self):
# current status: should be zero when idle
reboot = self._reboot
last_update = time.time()
connected = len(self._active_connections) / 2
# TODO: when a connection is half-closed, len(active) will be odd. a
# moment later (hopefully) the other side will disconnect, but
# _update_stats isn't updated until later.
waiting = len(self._pending_tokens)
# "waiting" doesn't count multiple parallel connections from the same
# side
self.usage.update_stats(
rebooted=self._rebooted,
updated=self._timestamp(),
connected=len(self.active_connections._connections),
waiting=len(self.pending_requests._requests),
incomplete_bytes=sum(
tc._total_sent
for tc in self.active_connections._connections
),
)
@implementer(ITransitClient)
class WebSocketTransitConnection(WebSocketServerProtocol):
started_time = None
def send(self, data):
"""
ITransitClient API
"""
self.sendMessage(data, isBinary=True)
def disconnect(self):
"""
ITransitClient API
"""
self.sendClose(1000, None)
def connect_partner(self, other):
"""
ITransitClient API
"""
self._buddy = other
self._buddy._client.transport.registerProducer(self.transport, True)
def disconnect_partner(self):
"""
ITransitClient API
"""
assert self._buddy is not None, "internal error: no buddy"
if self.factory.log_requests:
log.msg("buddy_disconnected {}".format(self._buddy.get_token()))
self._buddy._client.disconnect()
self._buddy = None
def connectionMade(self):
"""
IProtocol API
"""
super(WebSocketTransitConnection, self).connectionMade()
self.started_time = time.time()
self._first_message = True
self._state = TransitServerState(
self.factory.transit.pending_requests,
self.factory.transit.usage,
)
# uncomment to turn on state-machine tracing
# def tracer(oldstate, theinput, newstate):
# print("WSTRACE: {}: {} --{}--> {}".format(id(self), oldstate, theinput, newstate))
# self._state.set_trace_function(tracer)
def onOpen(self):
self._state.connection_made(self)
def onMessage(self, payload, isBinary):
"""
We may have a 'handshake' on our hands or we may just have some bytes to relay
"""
if not isBinary:
raise ValueError(
"All messages must be binary"
)
if self._first_message:
self._first_message = False
token = None
old = re.search(br"^please relay (\w{64})$", payload)
if old:
token = old.group(1)
self._state.please_relay(token)
# new: "please relay {64} for side {16}\n"
new = re.search(br"^please relay (\w{64}) for side (\w{16})$", payload)
if new:
token = new.group(1)
side = new.group(2)
self._state.please_relay_for_side(token, side)
if token is None:
self._state.bad_token()
else:
self._state.got_bytes(payload)
def onClose(self, wasClean, code, reason):
"""
IWebSocketChannel API
"""
self._state.connection_lost()
incomplete_bytes = sum(tc._total_sent
for tc in self._active_connections)
self._db.execute("DELETE FROM `current`")
self._db.execute("INSERT INTO `current`"
" (`reboot`, `last_update`, `connected`, `waiting`,"
" `incomplete_bytes`)"
" VALUES (?, ?, ?, ?, ?)",
(reboot, last_update, connected, waiting,
incomplete_bytes))

238
src/wormhole_transit_relay/usage.py
View File

@ -1,238 +0,0 @@
import time
import json
from twisted.python import log
from zope.interface import (
implementer,
Interface,
)
def create_usage_tracker(blur_usage, log_file, usage_db):
"""
:param int blur_usage: see UsageTracker
:param log_file: None or a file-like object to write JSON-encoded
lines of usage information to.
:param usage_db: None or an sqlite3 database connection
:returns: a new UsageTracker instance configured with backends.
"""
tracker = UsageTracker(blur_usage)
if usage_db:
tracker.add_backend(DatabaseUsageRecorder(usage_db))
if log_file:
tracker.add_backend(LogFileUsageRecorder(log_file))
return tracker
class IUsageWriter(Interface):
"""
Records actual usage statistics in some way
"""
def record_usage(started=None, total_time=None, waiting_time=None, total_bytes=None, mood=None):
"""
:param int started: timestemp when this connection began
:param float total_time: total seconds this connection lasted
:param float waiting_time: None or the total seconds one side
waited for the other
:param int total_bytes: the total bytes sent. In case the
connection was concluded successfully, only one side will
record the total bytes (but count both).
:param str mood: the 'mood' of the connection
"""
@implementer(IUsageWriter)
class MemoryUsageRecorder:
"""
Remebers usage records in memory.
"""
def __init__(self):
self.events = []
def record_usage(self, started=None, total_time=None, waiting_time=None, total_bytes=None, mood=None):
"""
IUsageWriter.
"""
data = {
"started": started,
"total_time": total_time,
"waiting_time": waiting_time,
"total_bytes": total_bytes,
"mood": mood,
}
self.events.append(data)
@implementer(IUsageWriter)
class LogFileUsageRecorder:
"""
Writes usage records to a file. The records are written in JSON,
one record per line.
"""
def __init__(self, writable_file):
self._file = writable_file
def record_usage(self, started=None, total_time=None, waiting_time=None, total_bytes=None, mood=None):
"""
IUsageWriter.
"""
data = {
"started": started,
"total_time": total_time,
"waiting_time": waiting_time,
"total_bytes": total_bytes,
"mood": mood,
}
self._file.write(json.dumps(data) + "\n")
self._file.flush()
@implementer(IUsageWriter)
class DatabaseUsageRecorder:
"""
Write usage records into a database
"""
def __init__(self, db):
self._db = db
def record_usage(self, started=None, total_time=None, waiting_time=None, total_bytes=None, mood=None):
"""
IUsageWriter.
"""
self._db.execute(
"INSERT INTO `usage`"
" (`started`, `total_time`, `waiting_time`,"
" `total_bytes`, `result`)"
" VALUES (?,?,?,?,?)",
(started, total_time, waiting_time, total_bytes, mood)
)
# original code did "self._update_stats()" here, thus causing
# "global" stats update on every connection update .. should
# we repeat this behavior, or really only record every
# 60-seconds with the timer?
self._db.commit()
class UsageTracker(object):
"""
Tracks usage statistics of connections
"""
def __init__(self, blur_usage):
"""
:param int blur_usage: None or the number of seconds to use as a
window around which to blur time statistics (e.g. "60" means times
will be rounded to 1 minute intervals). When blur_usage is
non-zero, sizes will also be rounded into buckets of "one
megabyte", "one gigabyte" or "lots"
"""
self._backends = set()
self._blur_usage = blur_usage
if blur_usage:
log.msg("blurring access times to %d seconds" % self._blur_usage)
else:
log.msg("not blurring access times")
def add_backend(self, backend):
"""
Add a new backend.
:param IUsageWriter backend: the backend to add
"""
self._backends.add(backend)
def record(self, started, buddy_started, result, bytes_sent, buddy_bytes):
"""
:param int started: timestamp when our connection started
:param int buddy_started: None, or the timestamp when our
partner's connection started (will be None if we don't yet
have a partner).
:param str result: a label for the result of the connection
(one of the "moods").
:param int bytes_sent: number of bytes we sent
:param int buddy_bytes: number of bytes our partner sent
"""
# ideally self._reactor.seconds() or similar, but ..
finished = time.time()
if buddy_started is not None:
starts = [started, buddy_started]
total_time = finished - min(starts)
waiting_time = max(starts) - min(starts)
total_bytes = bytes_sent + buddy_bytes
else:
total_time = finished - started
waiting_time = None
total_bytes = bytes_sent
# note that "bytes_sent" should always be 0 here, but
# we're recording what the state-machine remembered in any
# case
if self._blur_usage:
started = self._blur_usage * (started // self._blur_usage)
total_bytes = blur_size(total_bytes)
# This is "a dict" instead of "kwargs" because we have to make
# it into a dict for the log use-case and in-memory/testing
# use-case anyway so this is less repeats of the names.
self._notify_backends({
"started": started,
"total_time": total_time,
"waiting_time": waiting_time,
"total_bytes": total_bytes,
"mood": result,
})
def update_stats(self, rebooted, updated, connected, waiting,
incomplete_bytes):
"""
Update general statistics.
"""
# in original code, this is only recorded in the database
# .. perhaps a better way to do this, but ..
for backend in self._backends:
if isinstance(backend, DatabaseUsageRecorder):
backend._db.execute("DELETE FROM `current`")
backend._db.execute(
"INSERT INTO `current`"
" (`rebooted`, `updated`, `connected`, `waiting`,"
" `incomplete_bytes`)"
" VALUES (?, ?, ?, ?, ?)",
(int(rebooted), int(updated), connected, waiting,
incomplete_bytes)
)
def _notify_backends(self, data):
"""
Internal helper. Tell every backend we have about a new usage record.
"""
for backend in self._backends:
backend.record_usage(**data)
def round_to(size, coarseness):
return int(coarseness*(1+int((size-1)/coarseness)))
def blur_size(size):
if size == 0:
return 0
if size < 1e6:
return round_to(size, 10e3)
if size < 1e9:
return round_to(size, 1e6)
return round_to(size, 100e6)

2
tox.ini
View File

@ -4,7 +4,7 @@
# and then run "tox" from this directory.
[tox]
envlist = {py37,py38,py39,py310,pypy}
envlist = {py27,py34,py35,py36,pypy}
skip_missing_interpreters = True
minversion = 2.4.0

82
ws_client.py
View File

@ -1,82 +0,0 @@
"""
This is a test-client for the transit-relay that uses WebSockets.
If an additional command-line argument (anything) is added, it will
send 5 messages upon connection. Otherwise, it just prints out what is
received. Uses a fixed token of 64 'a' characters. Always connects on
localhost:4002
"""
import sys
from twisted.internet import endpoints
from twisted.internet.defer import (
Deferred,
inlineCallbacks,
)
from twisted.internet.task import react, deferLater
from autobahn.twisted.websocket import (
WebSocketClientProtocol,
WebSocketClientFactory,
)
class RelayEchoClient(WebSocketClientProtocol):
def onOpen(self):
self._received = b""
self.sendMessage(
u"please relay {} for side {}".format(
self.factory.token,
self.factory.side,
).encode("ascii"),
True,
)
def onMessage(self, data, isBinary):
print(">onMessage: {} bytes".format(len(data)))
print(data, isBinary)
if data == b"ok\n":
self.factory.ready.callback(None)
else:
self._received += data
if False:
# test abrupt hangup from receiving side
self.transport.loseConnection()
def onClose(self, wasClean, code, reason):
print(">onClose", wasClean, code, reason)
self.factory.done.callback(reason)
if not self.factory.ready.called:
self.factory.ready.errback(RuntimeError(reason))
@react
@inlineCallbacks
def main(reactor):
will_send_message = len(sys.argv) > 1
ep = endpoints.clientFromString(reactor, "tcp:localhost:4002")
f = WebSocketClientFactory("ws://127.0.0.1:4002/")
f.reactor = reactor
f.protocol = RelayEchoClient
f.token = "a" * 64
f.side = "0" * 16 if will_send_message else "1" * 16
f.done = Deferred()
f.ready = Deferred()
proto = yield ep.connect(f)
print("proto", proto)
yield f.ready
print("ready")
if will_send_message:
for _ in range(5):
print("sending message")
proto.sendMessage(b"it's a message", True)
yield deferLater(reactor, 0.2)
yield proto.sendClose()
print("closing")
yield f.done
print("relayed {} bytes:".format(len(proto._received)))
print(proto._received.decode("utf8"))