763 lines
17 KiB
JavaScript
763 lines
17 KiB
JavaScript
var types = require('./types')
|
|
var rcodes = require('./rcodes')
|
|
var opcodes = require('./opcodes')
|
|
var ip = require('ip')
|
|
var Buffer = require('safe-buffer').Buffer
|
|
|
|
var QUERY_FLAG = 0
|
|
var RESPONSE_FLAG = 1 << 15
|
|
var FLUSH_MASK = 1 << 15
|
|
var NOT_FLUSH_MASK = ~FLUSH_MASK
|
|
var QU_MASK = 1 << 15
|
|
var NOT_QU_MASK = ~QU_MASK
|
|
|
|
var name = exports.txt = exports.name = {}
|
|
|
|
name.encode = function (str, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(name.encodingLength(str))
|
|
if (!offset) offset = 0
|
|
var oldOffset = offset
|
|
|
|
// strip leading and trailing .
|
|
var n = str.replace(/^\.|\.$/gm, '')
|
|
if (n.length) {
|
|
var list = n.split('.')
|
|
|
|
for (var i = 0; i < list.length; i++) {
|
|
var len = buf.write(list[i], offset + 1)
|
|
buf[offset] = len
|
|
offset += len + 1
|
|
}
|
|
}
|
|
|
|
buf[offset++] = 0
|
|
|
|
name.encode.bytes = offset - oldOffset
|
|
return buf
|
|
}
|
|
|
|
name.encode.bytes = 0
|
|
|
|
name.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var list = []
|
|
var oldOffset = offset
|
|
var len = buf[offset++]
|
|
|
|
if (len === 0) {
|
|
name.decode.bytes = 1
|
|
return '.'
|
|
}
|
|
if (len >= 0xc0) {
|
|
var res = name.decode(buf, buf.readUInt16BE(offset - 1) - 0xc000)
|
|
name.decode.bytes = 2
|
|
return res
|
|
}
|
|
|
|
while (len) {
|
|
if (len >= 0xc0) {
|
|
list.push(name.decode(buf, buf.readUInt16BE(offset - 1) - 0xc000))
|
|
offset++
|
|
break
|
|
}
|
|
|
|
list.push(buf.toString('utf-8', offset, offset + len))
|
|
offset += len
|
|
len = buf[offset++]
|
|
}
|
|
|
|
name.decode.bytes = offset - oldOffset
|
|
return list.join('.')
|
|
}
|
|
|
|
name.decode.bytes = 0
|
|
|
|
name.encodingLength = function (n) {
|
|
return Buffer.byteLength(n) + 2
|
|
}
|
|
|
|
var string = {}
|
|
|
|
string.encode = function (s, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(string.encodingLength(s))
|
|
if (!offset) offset = 0
|
|
|
|
var len = buf.write(s, offset + 1)
|
|
buf[offset] = len
|
|
string.encode.bytes = len + 1
|
|
return buf
|
|
}
|
|
|
|
string.encode.bytes = 0
|
|
|
|
string.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var len = buf[offset]
|
|
var s = buf.toString('utf-8', offset + 1, offset + 1 + len)
|
|
string.decode.bytes = len + 1
|
|
return s
|
|
}
|
|
|
|
string.decode.bytes = 0
|
|
|
|
string.encodingLength = function (s) {
|
|
return Buffer.byteLength(s) + 1
|
|
}
|
|
|
|
var header = {}
|
|
|
|
header.encode = function (h, buf, offset) {
|
|
if (!buf) buf = header.encodingLength(h)
|
|
if (!offset) offset = 0
|
|
|
|
var flags = (h.flags || 0) & 32767
|
|
var type = h.type === 'response' ? RESPONSE_FLAG : QUERY_FLAG
|
|
|
|
buf.writeUInt16BE(h.id || 0, offset)
|
|
buf.writeUInt16BE(flags | type, offset + 2)
|
|
buf.writeUInt16BE(h.questions.length, offset + 4)
|
|
buf.writeUInt16BE(h.answers.length, offset + 6)
|
|
buf.writeUInt16BE(h.authorities.length, offset + 8)
|
|
buf.writeUInt16BE(h.additionals.length, offset + 10)
|
|
|
|
return buf
|
|
}
|
|
|
|
header.encode.bytes = 12
|
|
|
|
header.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
if (buf.length < 12) throw new Error('Header must be 12 bytes')
|
|
var flags = buf.readUInt16BE(offset + 2)
|
|
|
|
return {
|
|
id: buf.readUInt16BE(offset),
|
|
type: flags & RESPONSE_FLAG ? 'response' : 'query',
|
|
flags: flags & 32767,
|
|
flag_qr: ((flags >> 15) & 0x1) === 1,
|
|
opcode: opcodes.toString((flags >> 11) & 0xf),
|
|
flag_auth: ((flags >> 10) & 0x1) === 1,
|
|
flag_trunc: ((flags >> 9) & 0x1) === 1,
|
|
flag_rd: ((flags >> 8) & 0x1) === 1,
|
|
flag_ra: ((flags >> 7) & 0x1) === 1,
|
|
flag_z: ((flags >> 6) & 0x1) === 1,
|
|
flag_ad: ((flags >> 5) & 0x1) === 1,
|
|
flag_cd: ((flags >> 4) & 0x1) === 1,
|
|
rcode: rcodes.toString(flags & 0xf),
|
|
questions: new Array(buf.readUInt16BE(offset + 4)),
|
|
answers: new Array(buf.readUInt16BE(offset + 6)),
|
|
authorities: new Array(buf.readUInt16BE(offset + 8)),
|
|
additionals: new Array(buf.readUInt16BE(offset + 10))
|
|
}
|
|
}
|
|
|
|
header.decode.bytes = 12
|
|
|
|
header.encodingLength = function () {
|
|
return 12
|
|
}
|
|
|
|
var runknown = exports.unknown = {}
|
|
|
|
runknown.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(runknown.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
buf.writeUInt16BE(data.length, offset)
|
|
data.copy(buf, offset + 2)
|
|
|
|
runknown.encode.bytes = data.length + 2
|
|
return buf
|
|
}
|
|
|
|
runknown.encode.bytes = 0
|
|
|
|
runknown.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var len = buf.readUInt16BE(offset)
|
|
var data = buf.slice(offset + 2, offset + 2 + len)
|
|
runknown.decode.bytes = len + 2
|
|
return data
|
|
}
|
|
|
|
runknown.decode.bytes = 0
|
|
|
|
runknown.encodingLength = function (data) {
|
|
return data.length + 2
|
|
}
|
|
|
|
var rns = exports.ns = {}
|
|
|
|
rns.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(rns.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
name.encode(data, buf, offset + 2)
|
|
buf.writeUInt16BE(name.encode.bytes, offset)
|
|
rns.encode.bytes = name.encode.bytes + 2
|
|
return buf
|
|
}
|
|
|
|
rns.encode.bytes = 0
|
|
|
|
rns.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var len = buf.readUInt16BE(offset)
|
|
var dd = name.decode(buf, offset + 2)
|
|
|
|
rns.decode.bytes = len + 2
|
|
return dd
|
|
}
|
|
|
|
rns.decode.bytes = 0
|
|
|
|
rns.encodingLength = function (data) {
|
|
return name.encodingLength(data) + 2
|
|
}
|
|
|
|
var rsoa = exports.soa = {}
|
|
|
|
rsoa.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(rsoa.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
offset += 2
|
|
name.encode(data.mname, buf, offset)
|
|
offset += name.encode.bytes
|
|
name.encode(data.rname, buf, offset)
|
|
offset += name.encode.bytes
|
|
buf.writeUInt32BE(data.serial || 0, offset)
|
|
offset += 4
|
|
buf.writeUInt32BE(data.refresh || 0, offset)
|
|
offset += 4
|
|
buf.writeUInt32BE(data.retry || 0, offset)
|
|
offset += 4
|
|
buf.writeUInt32BE(data.expire || 0, offset)
|
|
offset += 4
|
|
buf.writeUInt32BE(data.minimum || 0, offset)
|
|
offset += 4
|
|
|
|
buf.writeUInt16BE(offset - oldOffset - 2, oldOffset)
|
|
rsoa.encode.bytes = offset - oldOffset
|
|
return buf
|
|
}
|
|
|
|
rsoa.encode.bytes = 0
|
|
|
|
rsoa.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
|
|
var data = {}
|
|
offset += 2
|
|
data.mname = name.decode(buf, offset)
|
|
offset += name.decode.bytes
|
|
data.rname = name.decode(buf, offset)
|
|
offset += name.decode.bytes
|
|
data.serial = buf.readUInt32BE(offset)
|
|
offset += 4
|
|
data.refresh = buf.readUInt32BE(offset)
|
|
offset += 4
|
|
data.retry = buf.readUInt32BE(offset)
|
|
offset += 4
|
|
data.expire = buf.readUInt32BE(offset)
|
|
offset += 4
|
|
data.minimum = buf.readUInt32BE(offset)
|
|
offset += 4
|
|
|
|
rsoa.decode.bytes = offset - oldOffset
|
|
return data
|
|
}
|
|
|
|
rsoa.decode.bytes = 0
|
|
|
|
rsoa.encodingLength = function (data) {
|
|
return 22 + name.encodingLength(data.mname) + name.encodingLength(data.rname)
|
|
}
|
|
|
|
var rtxt = exports.txt = exports.null = {}
|
|
var rnull = rtxt
|
|
|
|
rtxt.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(rtxt.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
if (typeof data === 'string') data = Buffer.from(data)
|
|
if (!data) data = Buffer.allocUnsafe(0)
|
|
|
|
var oldOffset = offset
|
|
offset += 2
|
|
|
|
var len = data.length
|
|
data.copy(buf, offset, 0, len)
|
|
offset += len
|
|
|
|
buf.writeUInt16BE(offset - oldOffset - 2, oldOffset)
|
|
rtxt.encode.bytes = offset - oldOffset
|
|
return buf
|
|
}
|
|
|
|
rtxt.encode.bytes = 0
|
|
|
|
rtxt.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
var oldOffset = offset
|
|
var len = buf.readUInt16BE(offset)
|
|
|
|
offset += 2
|
|
|
|
var data = buf.slice(offset, offset + len)
|
|
offset += len
|
|
|
|
rtxt.decode.bytes = offset - oldOffset
|
|
return data
|
|
}
|
|
|
|
rtxt.decode.bytes = 0
|
|
|
|
rtxt.encodingLength = function (data) {
|
|
if (!data) return 2
|
|
return (Buffer.isBuffer(data) ? data.length : Buffer.byteLength(data)) + 2
|
|
}
|
|
|
|
var rhinfo = exports.hinfo = {}
|
|
|
|
rhinfo.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(rhinfo.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
offset += 2
|
|
string.encode(data.cpu, buf, offset)
|
|
offset += string.encode.bytes
|
|
string.encode(data.os, buf, offset)
|
|
offset += string.encode.bytes
|
|
buf.writeUInt16BE(offset - oldOffset - 2, oldOffset)
|
|
rhinfo.encode.bytes = offset - oldOffset
|
|
return buf
|
|
}
|
|
|
|
rhinfo.encode.bytes = 0
|
|
|
|
rhinfo.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
|
|
var data = {}
|
|
offset += 2
|
|
data.cpu = string.decode(buf, offset)
|
|
offset += string.decode.bytes
|
|
data.os = string.decode(buf, offset)
|
|
offset += string.decode.bytes
|
|
rhinfo.decode.bytes = offset - oldOffset
|
|
return data
|
|
}
|
|
|
|
rhinfo.decode.bytes = 0
|
|
|
|
rhinfo.encodingLength = function (data) {
|
|
return string.encodingLength(data.cpu) + string.encodingLength(data.os) + 2
|
|
}
|
|
|
|
var rptr = exports.ptr = {}
|
|
var rcname = exports.cname = rptr
|
|
var rdname = exports.dname = rptr
|
|
|
|
rptr.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(rptr.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
name.encode(data, buf, offset + 2)
|
|
buf.writeUInt16BE(name.encode.bytes, offset)
|
|
rptr.encode.bytes = name.encode.bytes + 2
|
|
return buf
|
|
}
|
|
|
|
rptr.encode.bytes = 0
|
|
|
|
rptr.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var data = name.decode(buf, offset + 2)
|
|
rptr.decode.bytes = name.decode.bytes + 2
|
|
return data
|
|
}
|
|
|
|
rptr.decode.bytes = 0
|
|
|
|
rptr.encodingLength = function (data) {
|
|
return name.encodingLength(data) + 2
|
|
}
|
|
|
|
var rsrv = exports.srv = {}
|
|
|
|
rsrv.encode = function (data, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(rsrv.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
buf.writeUInt16BE(data.priority || 0, offset + 2)
|
|
buf.writeUInt16BE(data.weight || 0, offset + 4)
|
|
buf.writeUInt16BE(data.port || 0, offset + 6)
|
|
name.encode(data.target, buf, offset + 8)
|
|
|
|
var len = name.encode.bytes + 6
|
|
buf.writeUInt16BE(len, offset)
|
|
|
|
rsrv.encode.bytes = len + 2
|
|
return buf
|
|
}
|
|
|
|
rsrv.encode.bytes = 0
|
|
|
|
rsrv.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var len = buf.readUInt16BE(offset)
|
|
|
|
var data = {}
|
|
data.priority = buf.readUInt16BE(offset + 2)
|
|
data.weight = buf.readUInt16BE(offset + 4)
|
|
data.port = buf.readUInt16BE(offset + 6)
|
|
data.target = name.decode(buf, offset + 8)
|
|
|
|
rsrv.decode.bytes = len + 2
|
|
return data
|
|
}
|
|
|
|
rsrv.decode.bytes = 0
|
|
|
|
rsrv.encodingLength = function (data) {
|
|
return 8 + name.encodingLength(data.target)
|
|
}
|
|
|
|
var rcaa = exports.caa = {}
|
|
|
|
rcaa.ISSUER_CRITICAL = 1 << 7
|
|
|
|
rcaa.encode = function (data, buf, offset) {
|
|
var len = rcaa.encodingLength(data)
|
|
|
|
if (!buf) buf = Buffer.allocUnsafe(rcaa.encodingLength(data))
|
|
if (!offset) offset = 0
|
|
|
|
if (data.issuerCritical) {
|
|
data.flags = rcaa.ISSUER_CRITICAL
|
|
}
|
|
|
|
buf.writeUInt16BE(len - 2, offset)
|
|
offset += 2
|
|
buf.writeUInt8(data.flags || 0, offset)
|
|
offset += 1
|
|
string.encode(data.tag, buf, offset)
|
|
offset += string.encode.bytes
|
|
buf.write(data.value, offset)
|
|
offset += Buffer.byteLength(data.value)
|
|
|
|
rcaa.encode.bytes = len
|
|
return buf
|
|
}
|
|
|
|
rcaa.encode.bytes = 0
|
|
|
|
rcaa.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var len = buf.readUInt16BE(offset)
|
|
offset += 2
|
|
|
|
var oldOffset = offset
|
|
var data = {}
|
|
data.flags = buf.readUInt8(offset)
|
|
offset += 1
|
|
data.tag = string.decode(buf, offset)
|
|
offset += string.decode.bytes
|
|
data.value = buf.toString('utf-8', offset, oldOffset + len)
|
|
|
|
data.issuerCritical = !!(data.flags & rcaa.ISSUER_CRITICAL)
|
|
|
|
rcaa.decode.bytes = len + 2
|
|
|
|
return data
|
|
}
|
|
|
|
rcaa.decode.bytes = 0
|
|
|
|
rcaa.encodingLength = function (data) {
|
|
return string.encodingLength(data.tag) + string.encodingLength(data.value) + 2
|
|
}
|
|
|
|
var ra = exports.a = {}
|
|
|
|
ra.encode = function (host, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(ra.encodingLength(host))
|
|
if (!offset) offset = 0
|
|
|
|
buf.writeUInt16BE(4, offset)
|
|
offset += 2
|
|
ip.toBuffer(host, buf, offset)
|
|
ra.encode.bytes = 6
|
|
return buf
|
|
}
|
|
|
|
ra.encode.bytes = 0
|
|
|
|
ra.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
offset += 2
|
|
var host = ip.toString(buf, offset, 4)
|
|
ra.decode.bytes = 6
|
|
return host
|
|
}
|
|
|
|
ra.decode.bytes = 0
|
|
|
|
ra.encodingLength = function () {
|
|
return 6
|
|
}
|
|
|
|
var raaaa = exports.aaaa = {}
|
|
|
|
raaaa.encode = function (host, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(raaaa.encodingLength(host))
|
|
if (!offset) offset = 0
|
|
|
|
buf.writeUInt16BE(16, offset)
|
|
offset += 2
|
|
ip.toBuffer(host, buf, offset)
|
|
raaaa.encode.bytes = 18
|
|
return buf
|
|
}
|
|
|
|
raaaa.encode.bytes = 0
|
|
|
|
raaaa.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
offset += 2
|
|
var host = ip.toString(buf, offset, 16)
|
|
raaaa.decode.bytes = 18
|
|
return host
|
|
}
|
|
|
|
raaaa.decode.bytes = 0
|
|
|
|
raaaa.encodingLength = function () {
|
|
return 18
|
|
}
|
|
|
|
var renc = exports.record = function (type) {
|
|
switch (type.toUpperCase()) {
|
|
case 'A': return ra
|
|
case 'PTR': return rptr
|
|
case 'CNAME': return rcname
|
|
case 'DNAME': return rdname
|
|
case 'TXT': return rtxt
|
|
case 'NULL': return rnull
|
|
case 'AAAA': return raaaa
|
|
case 'SRV': return rsrv
|
|
case 'HINFO': return rhinfo
|
|
case 'CAA': return rcaa
|
|
case 'NS': return rns
|
|
case 'SOA': return rsoa
|
|
}
|
|
return runknown
|
|
}
|
|
|
|
var answer = exports.answer = {}
|
|
|
|
answer.encode = function (a, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(answer.encodingLength(a))
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
|
|
name.encode(a.name, buf, offset)
|
|
offset += name.encode.bytes
|
|
|
|
buf.writeUInt16BE(types.toType(a.type), offset)
|
|
|
|
var klass = a.class === undefined ? 1 : a.class
|
|
if (a.flush) klass |= FLUSH_MASK // the 1st bit of the class is the flush bit
|
|
buf.writeUInt16BE(klass, offset + 2)
|
|
|
|
buf.writeUInt32BE(a.ttl || 0, offset + 4)
|
|
|
|
var enc = renc(a.type)
|
|
enc.encode(a.data, buf, offset + 8)
|
|
offset += 8 + enc.encode.bytes
|
|
|
|
answer.encode.bytes = offset - oldOffset
|
|
return buf
|
|
}
|
|
|
|
answer.encode.bytes = 0
|
|
|
|
answer.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var a = {}
|
|
var oldOffset = offset
|
|
|
|
a.name = name.decode(buf, offset)
|
|
offset += name.decode.bytes
|
|
a.type = types.toString(buf.readUInt16BE(offset))
|
|
a.class = buf.readUInt16BE(offset + 2)
|
|
a.ttl = buf.readUInt32BE(offset + 4)
|
|
|
|
a.flush = !!(a.class & FLUSH_MASK)
|
|
if (a.flush) a.class &= NOT_FLUSH_MASK
|
|
|
|
var enc = renc(a.type)
|
|
a.data = enc.decode(buf, offset + 8)
|
|
offset += 8 + enc.decode.bytes
|
|
|
|
answer.decode.bytes = offset - oldOffset
|
|
return a
|
|
}
|
|
|
|
answer.decode.bytes = 0
|
|
|
|
answer.encodingLength = function (a) {
|
|
return name.encodingLength(a.name) + 8 + renc(a.type).encodingLength(a.data)
|
|
}
|
|
|
|
var question = exports.question = {}
|
|
|
|
question.encode = function (q, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(question.encodingLength(q))
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
|
|
name.encode(q.name, buf, offset)
|
|
offset += name.encode.bytes
|
|
|
|
buf.writeUInt16BE(types.toType(q.type), offset)
|
|
offset += 2
|
|
|
|
buf.writeUInt16BE(q.class === undefined ? 1 : q.class, offset)
|
|
offset += 2
|
|
|
|
question.encode.bytes = offset - oldOffset
|
|
return q
|
|
}
|
|
|
|
question.encode.bytes = 0
|
|
|
|
question.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
var q = {}
|
|
|
|
q.name = name.decode(buf, offset)
|
|
offset += name.decode.bytes
|
|
|
|
q.type = types.toString(buf.readUInt16BE(offset))
|
|
offset += 2
|
|
|
|
q.class = buf.readUInt16BE(offset)
|
|
offset += 2
|
|
|
|
var qu = !!(q.class & QU_MASK)
|
|
if (qu) q.class &= NOT_QU_MASK
|
|
|
|
question.decode.bytes = offset - oldOffset
|
|
return q
|
|
}
|
|
|
|
question.decode.bytes = 0
|
|
|
|
question.encodingLength = function (q) {
|
|
return name.encodingLength(q.name) + 4
|
|
}
|
|
|
|
exports.AUTHORITATIVE_ANSWER = 1 << 10
|
|
exports.TRUNCATED_RESPONSE = 1 << 9
|
|
exports.RECURSION_DESIRED = 1 << 8
|
|
exports.RECURSION_AVAILABLE = 1 << 7
|
|
exports.AUTHENTIC_DATA = 1 << 5
|
|
exports.CHECKING_DISABLED = 1 << 4
|
|
|
|
exports.encode = function (result, buf, offset) {
|
|
if (!buf) buf = Buffer.allocUnsafe(exports.encodingLength(result))
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
|
|
if (!result.questions) result.questions = []
|
|
if (!result.answers) result.answers = []
|
|
if (!result.authorities) result.authorities = []
|
|
if (!result.additionals) result.additionals = []
|
|
|
|
header.encode(result, buf, offset)
|
|
offset += header.encode.bytes
|
|
|
|
offset = encodeList(result.questions, question, buf, offset)
|
|
offset = encodeList(result.answers, answer, buf, offset)
|
|
offset = encodeList(result.authorities, answer, buf, offset)
|
|
offset = encodeList(result.additionals, answer, buf, offset)
|
|
|
|
exports.encode.bytes = offset - oldOffset
|
|
|
|
return buf
|
|
}
|
|
|
|
exports.encode.bytes = 0
|
|
|
|
exports.decode = function (buf, offset) {
|
|
if (!offset) offset = 0
|
|
|
|
var oldOffset = offset
|
|
var result = header.decode(buf, offset)
|
|
offset += header.decode.bytes
|
|
|
|
offset = decodeList(result.questions, question, buf, offset)
|
|
offset = decodeList(result.answers, answer, buf, offset)
|
|
offset = decodeList(result.authorities, answer, buf, offset)
|
|
offset = decodeList(result.additionals, answer, buf, offset)
|
|
|
|
exports.decode.bytes = offset - oldOffset
|
|
|
|
return result
|
|
}
|
|
|
|
exports.decode.bytes = 0
|
|
|
|
exports.encodingLength = function (result) {
|
|
return header.encodingLength(result) +
|
|
encodingLengthList(result.questions || [], question) +
|
|
encodingLengthList(result.answers || [], answer) +
|
|
encodingLengthList(result.authorities || [], answer) +
|
|
encodingLengthList(result.additionals || [], answer)
|
|
}
|
|
|
|
function encodingLengthList (list, enc) {
|
|
var len = 0
|
|
for (var i = 0; i < list.length; i++) len += enc.encodingLength(list[i])
|
|
return len
|
|
}
|
|
|
|
function encodeList (list, enc, buf, offset) {
|
|
for (var i = 0; i < list.length; i++) {
|
|
enc.encode(list[i], buf, offset)
|
|
offset += enc.encode.bytes
|
|
}
|
|
return offset
|
|
}
|
|
|
|
function decodeList (list, enc, buf, offset) {
|
|
for (var i = 0; i < list.length; i++) {
|
|
list[i] = enc.decode(buf, offset)
|
|
offset += enc.decode.bytes
|
|
}
|
|
return offset
|
|
}
|