Compare commits

..

No commits in common. "727d06c5d81d8c2d178a97caef37d3f5e8391475" and "9f18e08bda973941b62e096cb5c2e61ac39c6746" have entirely different histories.

2 changed files with 146 additions and 112 deletions

256
index.js
View File

@ -11,55 +11,19 @@ Packer.create = function (opts) {
machine = {}; machine = {};
} }
machine.onMessage = opts.onmessage || opts.onMessage;
machine.onmessage = opts.onmessage || opts.onMessage; machine.onmessage = opts.onmessage || opts.onMessage;
machine.oncontrol = opts.oncontrol || opts.onControl; machine.onError = opts.onerror || opts.onError;
machine.onerror = opts.onerror || opts.onError; machine.onerror = opts.onerror || opts.onError;
machine.onEnd = opts.onend || opts.onEnd;
machine.onend = opts.onend || opts.onEnd; machine.onend = opts.onend || opts.onEnd;
machine._version = 1; machine._version = 1;
machine.fns = {}; machine.state = 0;
machine.states = { 0: 'version', 1: 'headerLength', 2: 'header', 3: 'data'/*, 4: 'error'*/ };
machine.states_length = Object.keys(machine.states).length;
machine.chunkIndex = 0; machine.chunkIndex = 0;
machine.buf = null; machine.fns = {};
machine.bufIndex = 0;
machine.fns.collectData = function (chunk, size) {
var chunkLeft = chunk.length - machine.chunkIndex;
if (size <= 0) {
return Buffer.alloc(0);
}
// First handle case where we don't have all the data we need yet. We need to save
// what we have in a buffer, and increment the index for both the buffer and the chunk.
if (machine.bufIndex + chunkLeft < size) {
if (!machine.buf) {
machine.buf = Buffer.alloc(size);
}
chunk.copy(machine.buf, machine.bufIndex, machine.chunkIndex);
machine.bufIndex += chunkLeft;
machine.chunkIndex += chunkLeft;
return null;
}
// Read and mark as read however much data we need from the chunk to complete our buffer.
var partLen = size - machine.bufIndex;
var part = chunk.slice(machine.chunkIndex, machine.chunkIndex+partLen);
machine.chunkIndex += partLen;
// If we had nothing buffered than the part of the chunk we just read is all we need.
if (!machine.buf) {
return part;
}
// Otherwise we need to copy the new data into the buffer.
part.copy(machine.buf, machine.bufIndex);
// Before returning the buffer we need to clear our reference to it.
var buf = machine.buf;
machine.buf = null;
machine.bufIndex = 0;
return buf;
};
machine.fns.version = function (chunk) { machine.fns.version = function (chunk) {
//console.log(''); //console.log('');
@ -84,58 +48,118 @@ Packer.create = function (opts) {
return true; return true;
}; };
machine.buf = null;
machine.bufIndex = 0;
//var buf = Buffer.alloc(4096);
machine.fns.header = function (chunk) { machine.fns.header = function (chunk) {
//console.log(''); //console.log('');
//console.log('[header]'); //console.log('[header]');
var header = machine.fns.collectData(chunk, machine.headerLen); var curSize = machine.bufIndex + (chunk.length - machine.chunkIndex);
var partLen = 0;
var str = '';
var part;
if (curSize < machine.headerLen) {
// I still don't have the whole header,
// so just create a large enough buffer,
// write these bits, and wait for the next chunk.
if (!machine.buf) {
machine.buf = Buffer.alloc(machine.headerLen);
}
// partLen should be no more than the available size
partLen = Math.min(machine.headerLen - machine.bufIndex, chunk.length - machine.chunkIndex);
part = chunk.slice(machine.chunkIndex, machine.chunkIndex + partLen);
chunk.copy(machine.buf, machine.bufIndex, machine.chunkIndex, machine.chunkIndex + partLen);
machine.chunkIndex += partLen; // this MUST be chunk.length
machine.bufIndex += partLen;
// We don't have the entire header yet so return false.
if (!header) {
return false; return false;
} }
else {
// it's now ready to discover the whole header
if (machine.buf) {
str += machine.buf.slice(0, machine.bufIndex).toString();
}
machine._headers = header.toString().split(/,/g); partLen = machine.headerLen - str.length;
part = chunk.slice(machine.chunkIndex, machine.chunkIndex + partLen);
str += part.toString();
machine.family = machine._headers[0]; machine.chunkIndex += partLen;
machine.address = machine._headers[1]; machine.buf = null; // back to null
machine.port = machine._headers[2]; machine.bufIndex = 0; // back to 0
machine.bodyLen = parseInt(machine._headers[3], 10) || 0;
machine.service = machine._headers[4];
//console.log('machine.service', machine.service);
return true; machine._headers = str.split(/,/g);
machine.family = machine._headers[0];
machine.address = machine._headers[1];
machine.port = machine._headers[2];
machine.bodyLen = parseInt(machine._headers[3], 10) || -1;
machine.service = machine._headers[4];
//console.log('machine.service', machine.service);
return true;
}
}; };
machine.fns.data = function (chunk) { machine.fns.data = function (chunk) {
//console.log(''); //console.log('');
//console.log('[data]'); //console.log('[data]');
var data = machine.fns.collectData(chunk, machine.bodyLen); var curSize = machine.bufIndex + (chunk.length - machine.chunkIndex);
//console.log('curSize:', curSize);
//console.log('bodyLen:', machine.bodyLen, typeof machine.bodyLen);
var partLen = 0;
var msg;
var data;
partLen = Math.min(machine.bodyLen - machine.bufIndex, chunk.length - machine.chunkIndex);
if (curSize < machine.bodyLen) {
//console.log('curSize < bodyLen');
// I still don't have the whole header,
// so just create a large enough buffer,
// write these bits, and wait for the next chunk.
if (!machine.buf) {
machine.buf = Buffer.alloc(machine.bodyLen);
}
chunk.copy(machine.buf, machine.bufIndex, machine.chunkIndex, machine.chunkIndex + partLen);
machine.chunkIndex += partLen; // this MUST be chunk.length
machine.bufIndex += partLen;
// We don't have the entire body yet so return false.
if (!data) {
return false; return false;
} }
if (machine.bufIndex > 0) {
// the completing remainder of the body is in the current slice
chunk.copy(machine.buf, machine.bufIndex, machine.chunkIndex, machine.chunkIndex + partLen);
}
else {
// the whole body is in the current slice
machine.buf = chunk.slice(machine.chunkIndex, machine.chunkIndex + partLen);
}
machine.bufIndex += partLen;
machine.service = machine.service;
data = machine.buf.slice(0, machine.bufIndex);
//console.log('machine.service', machine.service);
// //
// data, end, error // data, end, error
// //
var msg = {};
if ('error' === machine.service) {
try {
msg = JSON.parse(data.toString());
} catch(e) {
msg.message = data.toString();
msg.code = 'E_UNKNOWN_ERR';
}
}
msg.family = machine.family;
msg.address = machine.address;
msg.port = machine.port;
msg.service = machine.service;
msg.data = data;
if ('end' === machine.service) { if ('end' === machine.service) {
msg = {};
msg.family = machine.family;
msg.address = machine.address;
msg.port = machine.port;
msg.service = 'end';
msg.data = data;
if (machine.emit) { if (machine.emit) {
machine.emit('tunnelEnd', msg); machine.emit('tunnelEnd', msg);
} }
@ -144,6 +168,18 @@ Packer.create = function (opts) {
} }
} }
else if ('error' === machine.service) { else if ('error' === machine.service) {
try {
msg = JSON.parse(machine.data.toString());
} catch(e) {
msg = new Error('unknown error');
}
msg.family = machine.family;
msg.address = machine.address;
msg.port = machine.port;
msg.service = 'error';
msg.data = data;
if (machine.emit) { if (machine.emit) {
machine.emit('tunnelError', msg); machine.emit('tunnelError', msg);
} }
@ -151,15 +187,15 @@ Packer.create = function (opts) {
(machine.onerror||machine.onmessage)(msg); (machine.onerror||machine.onmessage)(msg);
} }
} }
else if ('control' === machine.service) {
if (machine.emit) {
machine.emit('tunnelControl', msg);
}
else {
(machine.oncontrol||machine.onmessage)(msg);
}
}
else { else {
msg = {};
msg.family = machine.family;
msg.address = machine.address;
msg.port = machine.port;
msg.service = machine.service;
msg.data = data;
if (machine.emit) { if (machine.emit) {
machine.emit('tunnelData', msg); machine.emit('tunnelData', msg);
} }
@ -168,11 +204,12 @@ Packer.create = function (opts) {
} }
} }
machine.chunkIndex += partLen; // === chunk.length
machine.buf = null; // reset to null
machine.bufIndex = 0; // reset to 0
return true; return true;
}; };
machine.state = 0;
machine.states = ['version', 'headerLength', 'header', 'data'];
machine.fns.addChunk = function (chunk) { machine.fns.addChunk = function (chunk) {
//console.log(''); //console.log('');
//console.log('[addChunk]'); //console.log('[addChunk]');
@ -182,19 +219,17 @@ Packer.create = function (opts) {
if (true === machine.fns[machine.states[machine.state]](chunk)) { if (true === machine.fns[machine.states[machine.state]](chunk)) {
machine.state += 1; machine.state += 1;
machine.state %= machine.states.length; machine.state %= machine.states_length;
} }
} }
}; };
return machine; return machine;
}; };
Packer.pack = function (address, data, service) { Packer.pack = function (address, data, service) {
data = data || Buffer.alloc(1); data = data || Buffer.alloc(1);
if (!Buffer.isBuffer(data)) {
data = new Buffer(JSON.stringify(data));
}
if (!data.byteLength) { if (!data.byteLength) {
data = Buffer.alloc(1); data = Buffer.alloc(1);
} }
@ -207,21 +242,16 @@ Packer.pack = function (address, data, service) {
} }
var version = 1; var version = 1;
var header; var header = Buffer.from([
if (service === 'control') { /*servername,*/ address.family, address.address, address.port, data.byteLength
header = Buffer.from(['', '', '', data.byteLength, service].join(',')); , (address.service || '')
} ].join(','));
else {
header = Buffer.from([
address.family, address.address, address.port, data.byteLength, (address.service || '')
].join(','));
}
var meta = Buffer.from([ 255 - version, header.length ]); var meta = Buffer.from([ 255 - version, header.length ]);
var buf = Buffer.alloc(meta.byteLength + header.byteLength + data.byteLength); var buf = Buffer.alloc(meta.byteLength + header.byteLength + data.byteLength);
meta.copy(buf, 0); meta.copy(buf, 0, 0, meta.byteLength);
header.copy(buf, 2); header.copy(buf, 2, 0, header.byteLength);
data.copy(buf, 2 + header.byteLength); data.copy(buf, 2 + header.byteLength, 0, data.byteLength);
return buf; return buf;
}; };
@ -232,18 +262,22 @@ Packer.socketToAddr = function (socket) {
// tlsSocket.remoteAddress = remoteAddress; // causes core dump // tlsSocket.remoteAddress = remoteAddress; // causes core dump
// console.log(tlsSocket.remoteAddress); // console.log(tlsSocket.remoteAddress);
function extractValue(name) {
return socket[name]
|| socket['_'+name]
|| socket._handle._parentWrap[name]
|| socket._handle._parentWrap._handle.owner.stream[name]
;
}
return { return {
family: extractValue('remoteFamily') family:
, address: extractValue('remoteAddress') socket.remoteFamily
, port: extractValue('remotePort') || socket._remoteFamily
|| socket._handle._parentWrap.remoteFamily
|| socket._handle._parentWrap._handle.owner.stream.remoteFamily
, address:
socket.remoteAddress
|| socket._remoteAddress
|| socket._handle._parentWrap.remoteAddress
|| socket._handle._parentWrap._handle.owner.stream.remoteAddress
, port:
socket.remotePort
|| socket._remotePort
|| socket._handle._parentWrap.remotePort
|| socket._handle._parentWrap._handle.owner.stream.remotePort
}; };
}; };

View File

@ -1,6 +1,6 @@
{ {
"name": "tunnel-packer", "name": "tunnel-packer",
"version": "1.2.1", "version": "1.1.0",
"description": "A strategy for packing and unpacking a proxy stream (i.e. packets through a tunnel). Handles multiplexed and tls connections. Used by telebit and telebitd.", "description": "A strategy for packing and unpacking a proxy stream (i.e. packets through a tunnel). Handles multiplexed and tls connections. Used by telebit and telebitd.",
"main": "index.js", "main": "index.js",
"scripts": { "scripts": {