consul/vendor/github.com/miekg/dns/msg.go

1946 lines
52 KiB
Go

// DNS packet assembly, see RFC 1035. Converting from - Unpack() -
// and to - Pack() - wire format.
// All the packers and unpackers take a (msg []byte, off int)
// and return (off1 int, ok bool). If they return ok==false, they
// also return off1==len(msg), so that the next unpacker will
// also fail. This lets us avoid checks of ok until the end of a
// packing sequence.
package dns
import (
"encoding/base32"
"encoding/base64"
"encoding/hex"
"math/big"
"math/rand"
"net"
"reflect"
"strconv"
"time"
)
const maxCompressionOffset = 2 << 13 // We have 14 bits for the compression pointer
var (
// ErrAlg indicates an error with the (DNSSEC) algorithm.
ErrAlg error = &Error{err: "bad algorithm"}
// ErrAuth indicates an error in the TSIG authentication.
ErrAuth error = &Error{err: "bad authentication"}
// ErrBuf indicates that the buffer used it too small for the message.
ErrBuf error = &Error{err: "buffer size too small"}
// ErrConnEmpty indicates a connection is being uses before it is initialized.
ErrConnEmpty error = &Error{err: "conn has no connection"}
// ErrExtendedRcode ...
ErrExtendedRcode error = &Error{err: "bad extended rcode"}
// ErrFqdn indicates that a domain name does not have a closing dot.
ErrFqdn error = &Error{err: "domain must be fully qualified"}
// ErrId indicates there is a mismatch with the message's ID.
ErrId error = &Error{err: "id mismatch"}
// ErrKeyAlg indicates that the algorithm in the key is not valid.
ErrKeyAlg error = &Error{err: "bad key algorithm"}
ErrKey error = &Error{err: "bad key"}
ErrKeySize error = &Error{err: "bad key size"}
ErrNoSig error = &Error{err: "no signature found"}
ErrPrivKey error = &Error{err: "bad private key"}
ErrRcode error = &Error{err: "bad rcode"}
ErrRdata error = &Error{err: "bad rdata"}
ErrRRset error = &Error{err: "bad rrset"}
ErrSecret error = &Error{err: "no secrets defined"}
ErrShortRead error = &Error{err: "short read"}
// ErrSig indicates that a signature can not be cryptographically validated.
ErrSig error = &Error{err: "bad signature"}
// ErrSOA indicates that no SOA RR was seen when doing zone transfers.
ErrSoa error = &Error{err: "no SOA"}
// ErrTime indicates a timing error in TSIG authentication.
ErrTime error = &Error{err: "bad time"}
// ErrTruncated indicates that we failed to unpack a truncated message.
// We unpacked as much as we had so Msg can still be used, if desired.
ErrTruncated error = &Error{err: "failed to unpack truncated message"}
)
// Id, by default, returns a 16 bits random number to be used as a
// message id. The random provided should be good enough. This being a
// variable the function can be reassigned to a custom function.
// For instance, to make it return a static value:
//
// dns.Id = func() uint16 { return 3 }
var Id func() uint16 = id
// MsgHdr is a a manually-unpacked version of (id, bits).
type MsgHdr struct {
Id uint16
Response bool
Opcode int
Authoritative bool
Truncated bool
RecursionDesired bool
RecursionAvailable bool
Zero bool
AuthenticatedData bool
CheckingDisabled bool
Rcode int
}
// Msg contains the layout of a DNS message.
type Msg struct {
MsgHdr
Compress bool `json:"-"` // If true, the message will be compressed when converted to wire format. This not part of the official DNS packet format.
Question []Question // Holds the RR(s) of the question section.
Answer []RR // Holds the RR(s) of the answer section.
Ns []RR // Holds the RR(s) of the authority section.
Extra []RR // Holds the RR(s) of the additional section.
}
// StringToType is the reverse of TypeToString, needed for string parsing.
var StringToType = reverseInt16(TypeToString)
// StringToClass is the reverse of ClassToString, needed for string parsing.
var StringToClass = reverseInt16(ClassToString)
// Map of opcodes strings.
var StringToOpcode = reverseInt(OpcodeToString)
// Map of rcodes strings.
var StringToRcode = reverseInt(RcodeToString)
// ClassToString is a maps Classes to strings for each CLASS wire type.
var ClassToString = map[uint16]string{
ClassINET: "IN",
ClassCSNET: "CS",
ClassCHAOS: "CH",
ClassHESIOD: "HS",
ClassNONE: "NONE",
ClassANY: "ANY",
}
// OpcodeToString maps Opcodes to strings.
var OpcodeToString = map[int]string{
OpcodeQuery: "QUERY",
OpcodeIQuery: "IQUERY",
OpcodeStatus: "STATUS",
OpcodeNotify: "NOTIFY",
OpcodeUpdate: "UPDATE",
}
// RcodeToString maps Rcodes to strings.
var RcodeToString = map[int]string{
RcodeSuccess: "NOERROR",
RcodeFormatError: "FORMERR",
RcodeServerFailure: "SERVFAIL",
RcodeNameError: "NXDOMAIN",
RcodeNotImplemented: "NOTIMPL",
RcodeRefused: "REFUSED",
RcodeYXDomain: "YXDOMAIN", // From RFC 2136
RcodeYXRrset: "YXRRSET",
RcodeNXRrset: "NXRRSET",
RcodeNotAuth: "NOTAUTH",
RcodeNotZone: "NOTZONE",
RcodeBadSig: "BADSIG", // Also known as RcodeBadVers, see RFC 6891
// RcodeBadVers: "BADVERS",
RcodeBadKey: "BADKEY",
RcodeBadTime: "BADTIME",
RcodeBadMode: "BADMODE",
RcodeBadName: "BADNAME",
RcodeBadAlg: "BADALG",
RcodeBadTrunc: "BADTRUNC",
}
// Rather than write the usual handful of routines to pack and
// unpack every message that can appear on the wire, we use
// reflection to write a generic pack/unpack for structs and then
// use it. Thus, if in the future we need to define new message
// structs, no new pack/unpack/printing code needs to be written.
// Domain names are a sequence of counted strings
// split at the dots. They end with a zero-length string.
// PackDomainName packs a domain name s into msg[off:].
// If compression is wanted compress must be true and the compression
// map needs to hold a mapping between domain names and offsets
// pointing into msg.
func PackDomainName(s string, msg []byte, off int, compression map[string]int, compress bool) (off1 int, err error) {
off1, _, err = packDomainName(s, msg, off, compression, compress)
return
}
func packDomainName(s string, msg []byte, off int, compression map[string]int, compress bool) (off1 int, labels int, err error) {
// special case if msg == nil
lenmsg := 256
if msg != nil {
lenmsg = len(msg)
}
ls := len(s)
if ls == 0 { // Ok, for instance when dealing with update RR without any rdata.
return off, 0, nil
}
// If not fully qualified, error out, but only if msg == nil #ugly
switch {
case msg == nil:
if s[ls-1] != '.' {
s += "."
ls++
}
case msg != nil:
if s[ls-1] != '.' {
return lenmsg, 0, ErrFqdn
}
}
// Each dot ends a segment of the name.
// We trade each dot byte for a length byte.
// Except for escaped dots (\.), which are normal dots.
// There is also a trailing zero.
// Compression
nameoffset := -1
pointer := -1
// Emit sequence of counted strings, chopping at dots.
begin := 0
bs := []byte(s)
roBs, bsFresh, escapedDot := s, true, false
for i := 0; i < ls; i++ {
if bs[i] == '\\' {
for j := i; j < ls-1; j++ {
bs[j] = bs[j+1]
}
ls--
if off+1 > lenmsg {
return lenmsg, labels, ErrBuf
}
// check for \DDD
if i+2 < ls && isDigit(bs[i]) && isDigit(bs[i+1]) && isDigit(bs[i+2]) {
bs[i] = dddToByte(bs[i:])
for j := i + 1; j < ls-2; j++ {
bs[j] = bs[j+2]
}
ls -= 2
} else if bs[i] == 't' {
bs[i] = '\t'
} else if bs[i] == 'r' {
bs[i] = '\r'
} else if bs[i] == 'n' {
bs[i] = '\n'
}
escapedDot = bs[i] == '.'
bsFresh = false
continue
}
if bs[i] == '.' {
if i > 0 && bs[i-1] == '.' && !escapedDot {
// two dots back to back is not legal
return lenmsg, labels, ErrRdata
}
if i-begin >= 1<<6 { // top two bits of length must be clear
return lenmsg, labels, ErrRdata
}
// off can already (we're in a loop) be bigger than len(msg)
// this happens when a name isn't fully qualified
if off+1 > lenmsg {
return lenmsg, labels, ErrBuf
}
if msg != nil {
msg[off] = byte(i - begin)
}
offset := off
off++
for j := begin; j < i; j++ {
if off+1 > lenmsg {
return lenmsg, labels, ErrBuf
}
if msg != nil {
msg[off] = bs[j]
}
off++
}
if compress && !bsFresh {
roBs = string(bs)
bsFresh = true
}
// Dont try to compress '.'
if compress && roBs[begin:] != "." {
if p, ok := compression[roBs[begin:]]; !ok {
// Only offsets smaller than this can be used.
if offset < maxCompressionOffset {
compression[roBs[begin:]] = offset
}
} else {
// The first hit is the longest matching dname
// keep the pointer offset we get back and store
// the offset of the current name, because that's
// where we need to insert the pointer later
// If compress is true, we're allowed to compress this dname
if pointer == -1 && compress {
pointer = p // Where to point to
nameoffset = offset // Where to point from
break
}
}
}
labels++
begin = i + 1
}
escapedDot = false
}
// Root label is special
if len(bs) == 1 && bs[0] == '.' {
return off, labels, nil
}
// If we did compression and we find something add the pointer here
if pointer != -1 {
// We have two bytes (14 bits) to put the pointer in
// if msg == nil, we will never do compression
msg[nameoffset], msg[nameoffset+1] = packUint16(uint16(pointer ^ 0xC000))
off = nameoffset + 1
goto End
}
if msg != nil {
msg[off] = 0
}
End:
off++
return off, labels, nil
}
// Unpack a domain name.
// In addition to the simple sequences of counted strings above,
// domain names are allowed to refer to strings elsewhere in the
// packet, to avoid repeating common suffixes when returning
// many entries in a single domain. The pointers are marked
// by a length byte with the top two bits set. Ignoring those
// two bits, that byte and the next give a 14 bit offset from msg[0]
// where we should pick up the trail.
// Note that if we jump elsewhere in the packet,
// we return off1 == the offset after the first pointer we found,
// which is where the next record will start.
// In theory, the pointers are only allowed to jump backward.
// We let them jump anywhere and stop jumping after a while.
// UnpackDomainName unpacks a domain name into a string.
func UnpackDomainName(msg []byte, off int) (string, int, error) {
s := make([]byte, 0, 64)
off1 := 0
lenmsg := len(msg)
ptr := 0 // number of pointers followed
Loop:
for {
if off >= lenmsg {
return "", lenmsg, ErrBuf
}
c := int(msg[off])
off++
switch c & 0xC0 {
case 0x00:
if c == 0x00 {
// end of name
break Loop
}
// literal string
if off+c > lenmsg {
return "", lenmsg, ErrBuf
}
for j := off; j < off+c; j++ {
switch b := msg[j]; b {
case '.', '(', ')', ';', ' ', '@':
fallthrough
case '"', '\\':
s = append(s, '\\', b)
case '\t':
s = append(s, '\\', 't')
case '\r':
s = append(s, '\\', 'r')
default:
if b < 32 || b >= 127 { // unprintable use \DDD
var buf [3]byte
bufs := strconv.AppendInt(buf[:0], int64(b), 10)
s = append(s, '\\')
for i := 0; i < 3-len(bufs); i++ {
s = append(s, '0')
}
for _, r := range bufs {
s = append(s, r)
}
} else {
s = append(s, b)
}
}
}
s = append(s, '.')
off += c
case 0xC0:
// pointer to somewhere else in msg.
// remember location after first ptr,
// since that's how many bytes we consumed.
// also, don't follow too many pointers --
// maybe there's a loop.
if off >= lenmsg {
return "", lenmsg, ErrBuf
}
c1 := msg[off]
off++
if ptr == 0 {
off1 = off
}
if ptr++; ptr > 10 {
return "", lenmsg, &Error{err: "too many compression pointers"}
}
off = (c^0xC0)<<8 | int(c1)
default:
// 0x80 and 0x40 are reserved
return "", lenmsg, ErrRdata
}
}
if ptr == 0 {
off1 = off
}
if len(s) == 0 {
s = []byte(".")
}
return string(s), off1, nil
}
func packTxt(txt []string, msg []byte, offset int, tmp []byte) (int, error) {
var err error
if len(txt) == 0 {
if offset >= len(msg) {
return offset, ErrBuf
}
msg[offset] = 0
return offset, nil
}
for i := range txt {
if len(txt[i]) > len(tmp) {
return offset, ErrBuf
}
offset, err = packTxtString(txt[i], msg, offset, tmp)
if err != nil {
return offset, err
}
}
return offset, err
}
func packTxtString(s string, msg []byte, offset int, tmp []byte) (int, error) {
lenByteOffset := offset
if offset >= len(msg) {
return offset, ErrBuf
}
offset++
bs := tmp[:len(s)]
copy(bs, s)
for i := 0; i < len(bs); i++ {
if len(msg) <= offset {
return offset, ErrBuf
}
if bs[i] == '\\' {
i++
if i == len(bs) {
break
}
// check for \DDD
if i+2 < len(bs) && isDigit(bs[i]) && isDigit(bs[i+1]) && isDigit(bs[i+2]) {
msg[offset] = dddToByte(bs[i:])
i += 2
} else if bs[i] == 't' {
msg[offset] = '\t'
} else if bs[i] == 'r' {
msg[offset] = '\r'
} else if bs[i] == 'n' {
msg[offset] = '\n'
} else {
msg[offset] = bs[i]
}
} else {
msg[offset] = bs[i]
}
offset++
}
l := offset - lenByteOffset - 1
if l > 255 {
return offset, &Error{err: "string exceeded 255 bytes in txt"}
}
msg[lenByteOffset] = byte(l)
return offset, nil
}
func packOctetString(s string, msg []byte, offset int, tmp []byte) (int, error) {
if offset >= len(msg) {
return offset, ErrBuf
}
bs := tmp[:len(s)]
copy(bs, s)
for i := 0; i < len(bs); i++ {
if len(msg) <= offset {
return offset, ErrBuf
}
if bs[i] == '\\' {
i++
if i == len(bs) {
break
}
// check for \DDD
if i+2 < len(bs) && isDigit(bs[i]) && isDigit(bs[i+1]) && isDigit(bs[i+2]) {
msg[offset] = dddToByte(bs[i:])
i += 2
} else {
msg[offset] = bs[i]
}
} else {
msg[offset] = bs[i]
}
offset++
}
return offset, nil
}
func unpackTxt(msg []byte, off0 int) (ss []string, off int, err error) {
off = off0
var s string
for off < len(msg) && err == nil {
s, off, err = unpackTxtString(msg, off)
if err == nil {
ss = append(ss, s)
}
}
return
}
func unpackTxtString(msg []byte, offset int) (string, int, error) {
if offset+1 > len(msg) {
return "", offset, &Error{err: "overflow unpacking txt"}
}
l := int(msg[offset])
if offset+l+1 > len(msg) {
return "", offset, &Error{err: "overflow unpacking txt"}
}
s := make([]byte, 0, l)
for _, b := range msg[offset+1 : offset+1+l] {
switch b {
case '"', '\\':
s = append(s, '\\', b)
case '\t':
s = append(s, `\t`...)
case '\r':
s = append(s, `\r`...)
case '\n':
s = append(s, `\n`...)
default:
if b < 32 || b > 127 { // unprintable
var buf [3]byte
bufs := strconv.AppendInt(buf[:0], int64(b), 10)
s = append(s, '\\')
for i := 0; i < 3-len(bufs); i++ {
s = append(s, '0')
}
for _, r := range bufs {
s = append(s, r)
}
} else {
s = append(s, b)
}
}
}
offset += 1 + l
return string(s), offset, nil
}
// Pack a reflect.StructValue into msg. Struct members can only be uint8, uint16, uint32, string,
// slices and other (often anonymous) structs.
func packStructValue(val reflect.Value, msg []byte, off int, compression map[string]int, compress bool) (off1 int, err error) {
var txtTmp []byte
lenmsg := len(msg)
numfield := val.NumField()
for i := 0; i < numfield; i++ {
typefield := val.Type().Field(i)
if typefield.Tag == `dns:"-"` {
continue
}
switch fv := val.Field(i); fv.Kind() {
default:
return lenmsg, &Error{err: "bad kind packing"}
case reflect.Interface:
// PrivateRR is the only RR implementation that has interface field.
// therefore it's expected that this interface would be PrivateRdata
switch data := fv.Interface().(type) {
case PrivateRdata:
n, err := data.Pack(msg[off:])
if err != nil {
return lenmsg, err
}
off += n
default:
return lenmsg, &Error{err: "bad kind interface packing"}
}
case reflect.Slice:
switch typefield.Tag {
default:
return lenmsg, &Error{"bad tag packing slice: " + typefield.Tag.Get("dns")}
case `dns:"domain-name"`:
for j := 0; j < val.Field(i).Len(); j++ {
element := val.Field(i).Index(j).String()
off, err = PackDomainName(element, msg, off, compression, false && compress)
if err != nil {
return lenmsg, err
}
}
case `dns:"txt"`:
if txtTmp == nil {
txtTmp = make([]byte, 256*4+1)
}
off, err = packTxt(fv.Interface().([]string), msg, off, txtTmp)
if err != nil {
return lenmsg, err
}
case `dns:"opt"`: // edns
for j := 0; j < val.Field(i).Len(); j++ {
element := val.Field(i).Index(j).Interface()
b, e := element.(EDNS0).pack()
if e != nil {
return lenmsg, &Error{err: "overflow packing opt"}
}
// Option code
msg[off], msg[off+1] = packUint16(element.(EDNS0).Option())
// Length
msg[off+2], msg[off+3] = packUint16(uint16(len(b)))
off += 4
if off+len(b) > lenmsg {
copy(msg[off:], b)
off = lenmsg
continue
}
// Actual data
copy(msg[off:off+len(b)], b)
off += len(b)
}
case `dns:"a"`:
if val.Type().String() == "dns.IPSECKEY" {
// Field(2) is GatewayType, must be 1
if val.Field(2).Uint() != 1 {
continue
}
}
// It must be a slice of 4, even if it is 16, we encode
// only the first 4
if off+net.IPv4len > lenmsg {
return lenmsg, &Error{err: "overflow packing a"}
}
switch fv.Len() {
case net.IPv6len:
msg[off] = byte(fv.Index(12).Uint())
msg[off+1] = byte(fv.Index(13).Uint())
msg[off+2] = byte(fv.Index(14).Uint())
msg[off+3] = byte(fv.Index(15).Uint())
off += net.IPv4len
case net.IPv4len:
msg[off] = byte(fv.Index(0).Uint())
msg[off+1] = byte(fv.Index(1).Uint())
msg[off+2] = byte(fv.Index(2).Uint())
msg[off+3] = byte(fv.Index(3).Uint())
off += net.IPv4len
case 0:
// Allowed, for dynamic updates
default:
return lenmsg, &Error{err: "overflow packing a"}
}
case `dns:"aaaa"`:
if val.Type().String() == "dns.IPSECKEY" {
// Field(2) is GatewayType, must be 2
if val.Field(2).Uint() != 2 {
continue
}
}
if fv.Len() == 0 {
break
}
if fv.Len() > net.IPv6len || off+fv.Len() > lenmsg {
return lenmsg, &Error{err: "overflow packing aaaa"}
}
for j := 0; j < net.IPv6len; j++ {
msg[off] = byte(fv.Index(j).Uint())
off++
}
case `dns:"wks"`:
// TODO(miek): this is wrong should be lenrd
if off == lenmsg {
break // dyn. updates
}
if val.Field(i).Len() == 0 {
break
}
off1 := off
for j := 0; j < val.Field(i).Len(); j++ {
serv := int(fv.Index(j).Uint())
if off+serv/8+1 > len(msg) {
return len(msg), &Error{err: "overflow packing wks"}
}
msg[off+serv/8] |= byte(1 << (7 - uint(serv%8)))
if off+serv/8+1 > off1 {
off1 = off + serv/8 + 1
}
}
off = off1
case `dns:"nsec"`: // NSEC/NSEC3
// This is the uint16 type bitmap
if val.Field(i).Len() == 0 {
// Do absolutely nothing
break
}
var lastwindow, lastlength uint16
for j := 0; j < val.Field(i).Len(); j++ {
t := uint16(fv.Index(j).Uint())
window := t / 256
length := (t-window*256)/8 + 1
if window > lastwindow && lastlength != 0 {
// New window, jump to the new offset
off += int(lastlength) + 2
lastlength = 0
}
if window < lastwindow || length < lastlength {
return len(msg), &Error{err: "nsec bits out of order"}
}
if off+2+int(length) > len(msg) {
return len(msg), &Error{err: "overflow packing nsec"}
}
// Setting the window #
msg[off] = byte(window)
// Setting the octets length
msg[off+1] = byte(length)
// Setting the bit value for the type in the right octet
msg[off+1+int(length)] |= byte(1 << (7 - (t % 8)))
lastwindow, lastlength = window, length
}
off += int(lastlength) + 2
}
case reflect.Struct:
off, err = packStructValue(fv, msg, off, compression, compress)
if err != nil {
return lenmsg, err
}
case reflect.Uint8:
if off+1 > lenmsg {
return lenmsg, &Error{err: "overflow packing uint8"}
}
msg[off] = byte(fv.Uint())
off++
case reflect.Uint16:
if off+2 > lenmsg {
return lenmsg, &Error{err: "overflow packing uint16"}
}
i := fv.Uint()
msg[off] = byte(i >> 8)
msg[off+1] = byte(i)
off += 2
case reflect.Uint32:
if off+4 > lenmsg {
return lenmsg, &Error{err: "overflow packing uint32"}
}
i := fv.Uint()
msg[off] = byte(i >> 24)
msg[off+1] = byte(i >> 16)
msg[off+2] = byte(i >> 8)
msg[off+3] = byte(i)
off += 4
case reflect.Uint64:
switch typefield.Tag {
default:
if off+8 > lenmsg {
return lenmsg, &Error{err: "overflow packing uint64"}
}
i := fv.Uint()
msg[off] = byte(i >> 56)
msg[off+1] = byte(i >> 48)
msg[off+2] = byte(i >> 40)
msg[off+3] = byte(i >> 32)
msg[off+4] = byte(i >> 24)
msg[off+5] = byte(i >> 16)
msg[off+6] = byte(i >> 8)
msg[off+7] = byte(i)
off += 8
case `dns:"uint48"`:
// Used in TSIG, where it stops at 48 bits, so we discard the upper 16
if off+6 > lenmsg {
return lenmsg, &Error{err: "overflow packing uint64 as uint48"}
}
i := fv.Uint()
msg[off] = byte(i >> 40)
msg[off+1] = byte(i >> 32)
msg[off+2] = byte(i >> 24)
msg[off+3] = byte(i >> 16)
msg[off+4] = byte(i >> 8)
msg[off+5] = byte(i)
off += 6
}
case reflect.String:
// There are multiple string encodings.
// The tag distinguishes ordinary strings from domain names.
s := fv.String()
switch typefield.Tag {
default:
return lenmsg, &Error{"bad tag packing string: " + typefield.Tag.Get("dns")}
case `dns:"base64"`:
b64, e := fromBase64([]byte(s))
if e != nil {
return lenmsg, e
}
copy(msg[off:off+len(b64)], b64)
off += len(b64)
case `dns:"domain-name"`:
if val.Type().String() == "dns.IPSECKEY" {
// Field(2) is GatewayType, 1 and 2 or used for addresses
x := val.Field(2).Uint()
if x == 1 || x == 2 {
continue
}
}
if off, err = PackDomainName(s, msg, off, compression, false && compress); err != nil {
return lenmsg, err
}
case `dns:"cdomain-name"`:
if off, err = PackDomainName(s, msg, off, compression, true && compress); err != nil {
return lenmsg, err
}
case `dns:"size-base32"`:
// This is purely for NSEC3 atm, the previous byte must
// holds the length of the encoded string. As NSEC3
// is only defined to SHA1, the hashlength is 20 (160 bits)
msg[off-1] = 20
fallthrough
case `dns:"base32"`:
b32, e := fromBase32([]byte(s))
if e != nil {
return lenmsg, e
}
copy(msg[off:off+len(b32)], b32)
off += len(b32)
case `dns:"size-hex"`:
fallthrough
case `dns:"hex"`:
// There is no length encoded here
h, e := hex.DecodeString(s)
if e != nil {
return lenmsg, e
}
if off+hex.DecodedLen(len(s)) > lenmsg {
return lenmsg, &Error{err: "overflow packing hex"}
}
copy(msg[off:off+hex.DecodedLen(len(s))], h)
off += hex.DecodedLen(len(s))
case `dns:"size"`:
// the size is already encoded in the RR, we can safely use the
// length of string. String is RAW (not encoded in hex, nor base64)
copy(msg[off:off+len(s)], s)
off += len(s)
case `dns:"octet"`:
bytesTmp := make([]byte, 256)
off, err = packOctetString(fv.String(), msg, off, bytesTmp)
if err != nil {
return lenmsg, err
}
case `dns:"txt"`:
fallthrough
case "":
if txtTmp == nil {
txtTmp = make([]byte, 256*4+1)
}
off, err = packTxtString(fv.String(), msg, off, txtTmp)
if err != nil {
return lenmsg, err
}
}
}
}
return off, nil
}
func structValue(any interface{}) reflect.Value {
return reflect.ValueOf(any).Elem()
}
// PackStruct packs any structure to wire format.
func PackStruct(any interface{}, msg []byte, off int) (off1 int, err error) {
off, err = packStructValue(structValue(any), msg, off, nil, false)
return off, err
}
func packStructCompress(any interface{}, msg []byte, off int, compression map[string]int, compress bool) (off1 int, err error) {
off, err = packStructValue(structValue(any), msg, off, compression, compress)
return off, err
}
// Unpack a reflect.StructValue from msg.
// Same restrictions as packStructValue.
func unpackStructValue(val reflect.Value, msg []byte, off int) (off1 int, err error) {
lenmsg := len(msg)
for i := 0; i < val.NumField(); i++ {
if off > lenmsg {
return lenmsg, &Error{"bad offset unpacking"}
}
switch fv := val.Field(i); fv.Kind() {
default:
return lenmsg, &Error{err: "bad kind unpacking"}
case reflect.Interface:
// PrivateRR is the only RR implementation that has interface field.
// therefore it's expected that this interface would be PrivateRdata
switch data := fv.Interface().(type) {
case PrivateRdata:
n, err := data.Unpack(msg[off:])
if err != nil {
return lenmsg, err
}
off += n
default:
return lenmsg, &Error{err: "bad kind interface unpacking"}
}
case reflect.Slice:
switch val.Type().Field(i).Tag {
default:
return lenmsg, &Error{"bad tag unpacking slice: " + val.Type().Field(i).Tag.Get("dns")}
case `dns:"domain-name"`:
// HIP record slice of name (or none)
var servers []string
var s string
for off < lenmsg {
s, off, err = UnpackDomainName(msg, off)
if err != nil {
return lenmsg, err
}
servers = append(servers, s)
}
fv.Set(reflect.ValueOf(servers))
case `dns:"txt"`:
if off == lenmsg {
break
}
var txt []string
txt, off, err = unpackTxt(msg, off)
if err != nil {
return lenmsg, err
}
fv.Set(reflect.ValueOf(txt))
case `dns:"opt"`: // edns0
if off == lenmsg {
// This is an EDNS0 (OPT Record) with no rdata
// We can safely return here.
break
}
var edns []EDNS0
Option:
code := uint16(0)
if off+4 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking opt"}
}
code, off = unpackUint16(msg, off)
optlen, off1 := unpackUint16(msg, off)
if off1+int(optlen) > lenmsg {
return lenmsg, &Error{err: "overflow unpacking opt"}
}
switch code {
case EDNS0NSID:
e := new(EDNS0_NSID)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
case EDNS0SUBNET, EDNS0SUBNETDRAFT:
e := new(EDNS0_SUBNET)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
if code == EDNS0SUBNETDRAFT {
e.DraftOption = true
}
case EDNS0UL:
e := new(EDNS0_UL)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
case EDNS0LLQ:
e := new(EDNS0_LLQ)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
case EDNS0DAU:
e := new(EDNS0_DAU)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
case EDNS0DHU:
e := new(EDNS0_DHU)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
case EDNS0N3U:
e := new(EDNS0_N3U)
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
default:
e := new(EDNS0_LOCAL)
e.Code = code
if err := e.unpack(msg[off1 : off1+int(optlen)]); err != nil {
return lenmsg, err
}
edns = append(edns, e)
off = off1 + int(optlen)
}
if off < lenmsg {
goto Option
}
fv.Set(reflect.ValueOf(edns))
case `dns:"a"`:
if val.Type().String() == "dns.IPSECKEY" {
// Field(2) is GatewayType, must be 1
if val.Field(2).Uint() != 1 {
continue
}
}
if off == lenmsg {
break // dyn. update
}
if off+net.IPv4len > lenmsg {
return lenmsg, &Error{err: "overflow unpacking a"}
}
fv.Set(reflect.ValueOf(net.IPv4(msg[off], msg[off+1], msg[off+2], msg[off+3])))
off += net.IPv4len
case `dns:"aaaa"`:
if val.Type().String() == "dns.IPSECKEY" {
// Field(2) is GatewayType, must be 2
if val.Field(2).Uint() != 2 {
continue
}
}
if off == lenmsg {
break
}
if off+net.IPv6len > lenmsg {
return lenmsg, &Error{err: "overflow unpacking aaaa"}
}
fv.Set(reflect.ValueOf(net.IP{msg[off], msg[off+1], msg[off+2], msg[off+3], msg[off+4],
msg[off+5], msg[off+6], msg[off+7], msg[off+8], msg[off+9], msg[off+10],
msg[off+11], msg[off+12], msg[off+13], msg[off+14], msg[off+15]}))
off += net.IPv6len
case `dns:"wks"`:
// Rest of the record is the bitmap
var serv []uint16
j := 0
for off < lenmsg {
if off+1 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking wks"}
}
b := msg[off]
// Check the bits one by one, and set the type
if b&0x80 == 0x80 {
serv = append(serv, uint16(j*8+0))
}
if b&0x40 == 0x40 {
serv = append(serv, uint16(j*8+1))
}
if b&0x20 == 0x20 {
serv = append(serv, uint16(j*8+2))
}
if b&0x10 == 0x10 {
serv = append(serv, uint16(j*8+3))
}
if b&0x8 == 0x8 {
serv = append(serv, uint16(j*8+4))
}
if b&0x4 == 0x4 {
serv = append(serv, uint16(j*8+5))
}
if b&0x2 == 0x2 {
serv = append(serv, uint16(j*8+6))
}
if b&0x1 == 0x1 {
serv = append(serv, uint16(j*8+7))
}
j++
off++
}
fv.Set(reflect.ValueOf(serv))
case `dns:"nsec"`: // NSEC/NSEC3
if off == len(msg) {
break
}
// Rest of the record is the type bitmap
var nsec []uint16
length := 0
window := 0
lastwindow := -1
for off < len(msg) {
if off+2 > len(msg) {
return len(msg), &Error{err: "overflow unpacking nsecx"}
}
window = int(msg[off])
length = int(msg[off+1])
off += 2
if window <= lastwindow {
// RFC 4034: Blocks are present in the NSEC RR RDATA in
// increasing numerical order.
return len(msg), &Error{err: "out of order NSEC block"}
}
if length == 0 {
// RFC 4034: Blocks with no types present MUST NOT be included.
return len(msg), &Error{err: "empty NSEC block"}
}
if length > 32 {
return len(msg), &Error{err: "NSEC block too long"}
}
if off+length > len(msg) {
return len(msg), &Error{err: "overflowing NSEC block"}
}
// Walk the bytes in the window and extract the type bits
for j := 0; j < length; j++ {
b := msg[off+j]
// Check the bits one by one, and set the type
if b&0x80 == 0x80 {
nsec = append(nsec, uint16(window*256+j*8+0))
}
if b&0x40 == 0x40 {
nsec = append(nsec, uint16(window*256+j*8+1))
}
if b&0x20 == 0x20 {
nsec = append(nsec, uint16(window*256+j*8+2))
}
if b&0x10 == 0x10 {
nsec = append(nsec, uint16(window*256+j*8+3))
}
if b&0x8 == 0x8 {
nsec = append(nsec, uint16(window*256+j*8+4))
}
if b&0x4 == 0x4 {
nsec = append(nsec, uint16(window*256+j*8+5))
}
if b&0x2 == 0x2 {
nsec = append(nsec, uint16(window*256+j*8+6))
}
if b&0x1 == 0x1 {
nsec = append(nsec, uint16(window*256+j*8+7))
}
}
off += length
lastwindow = window
}
fv.Set(reflect.ValueOf(nsec))
}
case reflect.Struct:
off, err = unpackStructValue(fv, msg, off)
if err != nil {
return lenmsg, err
}
if val.Type().Field(i).Name == "Hdr" {
lenrd := off + int(val.FieldByName("Hdr").FieldByName("Rdlength").Uint())
if lenrd > lenmsg {
return lenmsg, &Error{err: "overflowing header size"}
}
msg = msg[:lenrd]
lenmsg = len(msg)
}
case reflect.Uint8:
if off == lenmsg {
break
}
if off+1 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking uint8"}
}
fv.SetUint(uint64(uint8(msg[off])))
off++
case reflect.Uint16:
if off == lenmsg {
break
}
var i uint16
if off+2 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking uint16"}
}
i, off = unpackUint16(msg, off)
fv.SetUint(uint64(i))
case reflect.Uint32:
if off == lenmsg {
break
}
if off+4 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking uint32"}
}
fv.SetUint(uint64(uint32(msg[off])<<24 | uint32(msg[off+1])<<16 | uint32(msg[off+2])<<8 | uint32(msg[off+3])))
off += 4
case reflect.Uint64:
if off == lenmsg {
break
}
switch val.Type().Field(i).Tag {
default:
if off+8 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking uint64"}
}
fv.SetUint(uint64(uint64(msg[off])<<56 | uint64(msg[off+1])<<48 | uint64(msg[off+2])<<40 |
uint64(msg[off+3])<<32 | uint64(msg[off+4])<<24 | uint64(msg[off+5])<<16 | uint64(msg[off+6])<<8 | uint64(msg[off+7])))
off += 8
case `dns:"uint48"`:
// Used in TSIG where the last 48 bits are occupied, so for now, assume a uint48 (6 bytes)
if off+6 > lenmsg {
return lenmsg, &Error{err: "overflow unpacking uint64 as uint48"}
}
fv.SetUint(uint64(uint64(msg[off])<<40 | uint64(msg[off+1])<<32 | uint64(msg[off+2])<<24 | uint64(msg[off+3])<<16 |
uint64(msg[off+4])<<8 | uint64(msg[off+5])))
off += 6
}
case reflect.String:
var s string
if off == lenmsg {
break
}
switch val.Type().Field(i).Tag {
default:
return lenmsg, &Error{"bad tag unpacking string: " + val.Type().Field(i).Tag.Get("dns")}
case `dns:"octet"`:
s = string(msg[off:])
off = lenmsg
case `dns:"hex"`:
hexend := lenmsg
if val.FieldByName("Hdr").FieldByName("Rrtype").Uint() == uint64(TypeHIP) {
hexend = off + int(val.FieldByName("HitLength").Uint())
}
if hexend > lenmsg {
return lenmsg, &Error{err: "overflow unpacking HIP hex"}
}
s = hex.EncodeToString(msg[off:hexend])
off = hexend
case `dns:"base64"`:
// Rest of the RR is base64 encoded value
b64end := lenmsg
if val.FieldByName("Hdr").FieldByName("Rrtype").Uint() == uint64(TypeHIP) {
b64end = off + int(val.FieldByName("PublicKeyLength").Uint())
}
if b64end > lenmsg {
return lenmsg, &Error{err: "overflow unpacking HIP base64"}
}
s = toBase64(msg[off:b64end])
off = b64end
case `dns:"cdomain-name"`:
fallthrough
case `dns:"domain-name"`:
if val.Type().String() == "dns.IPSECKEY" {
// Field(2) is GatewayType, 1 and 2 or used for addresses
x := val.Field(2).Uint()
if x == 1 || x == 2 {
continue
}
}
if off == lenmsg && int(val.FieldByName("Hdr").FieldByName("Rdlength").Uint()) == 0 {
// zero rdata is ok for dyn updates, but only if rdlength is 0
break
}
s, off, err = UnpackDomainName(msg, off)
if err != nil {
return lenmsg, err
}
case `dns:"size-base32"`:
var size int
switch val.Type().Name() {
case "NSEC3":
switch val.Type().Field(i).Name {
case "NextDomain":
name := val.FieldByName("HashLength")
size = int(name.Uint())
}
}
if off+size > lenmsg {
return lenmsg, &Error{err: "overflow unpacking base32"}
}
s = toBase32(msg[off : off+size])
off += size
case `dns:"size-hex"`:
// a "size" string, but it must be encoded in hex in the string
var size int
switch val.Type().Name() {
case "NSEC3":
switch val.Type().Field(i).Name {
case "Salt":
name := val.FieldByName("SaltLength")
size = int(name.Uint())
case "NextDomain":
name := val.FieldByName("HashLength")
size = int(name.Uint())
}
case "TSIG":
switch val.Type().Field(i).Name {
case "MAC":
name := val.FieldByName("MACSize")
size = int(name.Uint())
case "OtherData":
name := val.FieldByName("OtherLen")
size = int(name.Uint())
}
}
if off+size > lenmsg {
return lenmsg, &Error{err: "overflow unpacking hex"}
}
s = hex.EncodeToString(msg[off : off+size])
off += size
case `dns:"txt"`:
fallthrough
case "":
s, off, err = unpackTxtString(msg, off)
}
fv.SetString(s)
}
}
return off, nil
}
// Helpers for dealing with escaped bytes
func isDigit(b byte) bool { return b >= '0' && b <= '9' }
func dddToByte(s []byte) byte {
return byte((s[0]-'0')*100 + (s[1]-'0')*10 + (s[2] - '0'))
}
// UnpackStruct unpacks a binary message from offset off to the interface
// value given.
func UnpackStruct(any interface{}, msg []byte, off int) (int, error) {
return unpackStructValue(structValue(any), msg, off)
}
// Helper function for packing and unpacking
func intToBytes(i *big.Int, length int) []byte {
buf := i.Bytes()
if len(buf) < length {
b := make([]byte, length)
copy(b[length-len(buf):], buf)
return b
}
return buf
}
func unpackUint16(msg []byte, off int) (uint16, int) {
return uint16(msg[off])<<8 | uint16(msg[off+1]), off + 2
}
func packUint16(i uint16) (byte, byte) {
return byte(i >> 8), byte(i)
}
func toBase32(b []byte) string {
return base32.HexEncoding.EncodeToString(b)
}
func fromBase32(s []byte) (buf []byte, err error) {
buflen := base32.HexEncoding.DecodedLen(len(s))
buf = make([]byte, buflen)
n, err := base32.HexEncoding.Decode(buf, s)
buf = buf[:n]
return
}
func toBase64(b []byte) string {
return base64.StdEncoding.EncodeToString(b)
}
func fromBase64(s []byte) (buf []byte, err error) {
buflen := base64.StdEncoding.DecodedLen(len(s))
buf = make([]byte, buflen)
n, err := base64.StdEncoding.Decode(buf, s)
buf = buf[:n]
return
}
// PackRR packs a resource record rr into msg[off:].
// See PackDomainName for documentation about the compression.
func PackRR(rr RR, msg []byte, off int, compression map[string]int, compress bool) (off1 int, err error) {
if rr == nil {
return len(msg), &Error{err: "nil rr"}
}
off1, err = packStructCompress(rr, msg, off, compression, compress)
if err != nil {
return len(msg), err
}
if rawSetRdlength(msg, off, off1) {
return off1, nil
}
return off, ErrRdata
}
// UnpackRR unpacks msg[off:] into an RR.
func UnpackRR(msg []byte, off int) (rr RR, off1 int, err error) {
// unpack just the header, to find the rr type and length
var h RR_Header
off0 := off
if off, err = UnpackStruct(&h, msg, off); err != nil {
return nil, len(msg), err
}
end := off + int(h.Rdlength)
// make an rr of that type and re-unpack.
mk, known := TypeToRR[h.Rrtype]
if !known {
rr = new(RFC3597)
} else {
rr = mk()
}
off, err = UnpackStruct(rr, msg, off0)
if off != end {
return &h, end, &Error{err: "bad rdlength"}
}
return rr, off, err
}
// unpackRRslice unpacks msg[off:] into an []RR.
// If we cannot unpack the whole array, then it will return nil
func unpackRRslice(l int, msg []byte, off int) (dst1 []RR, off1 int, err error) {
var r RR
// Optimistically make dst be the length that was sent
dst := make([]RR, 0, l)
for i := 0; i < l; i++ {
off1 := off
r, off, err = UnpackRR(msg, off)
if err != nil {
off = len(msg)
break
}
// If offset does not increase anymore, l is a lie
if off1 == off {
l = i
break
}
dst = append(dst, r)
}
if err != nil && off == len(msg) {
dst = nil
}
return dst, off, err
}
// Reverse a map
func reverseInt8(m map[uint8]string) map[string]uint8 {
n := make(map[string]uint8)
for u, s := range m {
n[s] = u
}
return n
}
func reverseInt16(m map[uint16]string) map[string]uint16 {
n := make(map[string]uint16)
for u, s := range m {
n[s] = u
}
return n
}
func reverseInt(m map[int]string) map[string]int {
n := make(map[string]int)
for u, s := range m {
n[s] = u
}
return n
}
// Convert a MsgHdr to a string, with dig-like headers:
//
//;; opcode: QUERY, status: NOERROR, id: 48404
//
//;; flags: qr aa rd ra;
func (h *MsgHdr) String() string {
if h == nil {
return "<nil> MsgHdr"
}
s := ";; opcode: " + OpcodeToString[h.Opcode]
s += ", status: " + RcodeToString[h.Rcode]
s += ", id: " + strconv.Itoa(int(h.Id)) + "\n"
s += ";; flags:"
if h.Response {
s += " qr"
}
if h.Authoritative {
s += " aa"
}
if h.Truncated {
s += " tc"
}
if h.RecursionDesired {
s += " rd"
}
if h.RecursionAvailable {
s += " ra"
}
if h.Zero { // Hmm
s += " z"
}
if h.AuthenticatedData {
s += " ad"
}
if h.CheckingDisabled {
s += " cd"
}
s += ";"
return s
}
// Pack packs a Msg: it is converted to to wire format.
// If the dns.Compress is true the message will be in compressed wire format.
func (dns *Msg) Pack() (msg []byte, err error) {
return dns.PackBuffer(nil)
}
// PackBuffer packs a Msg, using the given buffer buf. If buf is too small
// a new buffer is allocated.
func (dns *Msg) PackBuffer(buf []byte) (msg []byte, err error) {
var dh Header
var compression map[string]int
if dns.Compress {
compression = make(map[string]int) // Compression pointer mappings
}
if dns.Rcode < 0 || dns.Rcode > 0xFFF {
return nil, ErrRcode
}
if dns.Rcode > 0xF {
// Regular RCODE field is 4 bits
opt := dns.IsEdns0()
if opt == nil {
return nil, ErrExtendedRcode
}
opt.SetExtendedRcode(uint8(dns.Rcode >> 4))
dns.Rcode &= 0xF
}
// Convert convenient Msg into wire-like Header.
dh.Id = dns.Id
dh.Bits = uint16(dns.Opcode)<<11 | uint16(dns.Rcode)
if dns.Response {
dh.Bits |= _QR
}
if dns.Authoritative {
dh.Bits |= _AA
}
if dns.Truncated {
dh.Bits |= _TC
}
if dns.RecursionDesired {
dh.Bits |= _RD
}
if dns.RecursionAvailable {
dh.Bits |= _RA
}
if dns.Zero {
dh.Bits |= _Z
}
if dns.AuthenticatedData {
dh.Bits |= _AD
}
if dns.CheckingDisabled {
dh.Bits |= _CD
}
// Prepare variable sized arrays.
question := dns.Question
answer := dns.Answer
ns := dns.Ns
extra := dns.Extra
dh.Qdcount = uint16(len(question))
dh.Ancount = uint16(len(answer))
dh.Nscount = uint16(len(ns))
dh.Arcount = uint16(len(extra))
// We need the uncompressed length here, because we first pack it and then compress it.
msg = buf
compress := dns.Compress
dns.Compress = false
if packLen := dns.Len() + 1; len(msg) < packLen {
msg = make([]byte, packLen)
}
dns.Compress = compress
// Pack it in: header and then the pieces.
off := 0
off, err = packStructCompress(&dh, msg, off, compression, dns.Compress)
if err != nil {
return nil, err
}
for i := 0; i < len(question); i++ {
off, err = packStructCompress(&question[i], msg, off, compression, dns.Compress)
if err != nil {
return nil, err
}
}
for i := 0; i < len(answer); i++ {
off, err = PackRR(answer[i], msg, off, compression, dns.Compress)
if err != nil {
return nil, err
}
}
for i := 0; i < len(ns); i++ {
off, err = PackRR(ns[i], msg, off, compression, dns.Compress)
if err != nil {
return nil, err
}
}
for i := 0; i < len(extra); i++ {
off, err = PackRR(extra[i], msg, off, compression, dns.Compress)
if err != nil {
return nil, err
}
}
return msg[:off], nil
}
// Unpack unpacks a binary message to a Msg structure.
func (dns *Msg) Unpack(msg []byte) (err error) {
// Header.
var dh Header
off := 0
if off, err = UnpackStruct(&dh, msg, off); err != nil {
return err
}
dns.Id = dh.Id
dns.Response = (dh.Bits & _QR) != 0
dns.Opcode = int(dh.Bits>>11) & 0xF
dns.Authoritative = (dh.Bits & _AA) != 0
dns.Truncated = (dh.Bits & _TC) != 0
dns.RecursionDesired = (dh.Bits & _RD) != 0
dns.RecursionAvailable = (dh.Bits & _RA) != 0
dns.Zero = (dh.Bits & _Z) != 0
dns.AuthenticatedData = (dh.Bits & _AD) != 0
dns.CheckingDisabled = (dh.Bits & _CD) != 0
dns.Rcode = int(dh.Bits & 0xF)
// Optimistically use the count given to us in the header
dns.Question = make([]Question, 0, int(dh.Qdcount))
var q Question
for i := 0; i < int(dh.Qdcount); i++ {
off1 := off
off, err = UnpackStruct(&q, msg, off)
if err != nil {
// Even if Truncated is set, we only will set ErrTruncated if we
// actually got the questions
return err
}
if off1 == off { // Offset does not increase anymore, dh.Qdcount is a lie!
dh.Qdcount = uint16(i)
break
}
dns.Question = append(dns.Question, q)
}
dns.Answer, off, err = unpackRRslice(int(dh.Ancount), msg, off)
// The header counts might have been wrong so we need to update it
dh.Ancount = uint16(len(dns.Answer))
if err == nil {
dns.Ns, off, err = unpackRRslice(int(dh.Nscount), msg, off)
}
// The header counts might have been wrong so we need to update it
dh.Nscount = uint16(len(dns.Ns))
if err == nil {
dns.Extra, off, err = unpackRRslice(int(dh.Arcount), msg, off)
}
// The header counts might have been wrong so we need to update it
dh.Arcount = uint16(len(dns.Extra))
if off != len(msg) {
// TODO(miek) make this an error?
// use PackOpt to let people tell how detailed the error reporting should be?
// println("dns: extra bytes in dns packet", off, "<", len(msg))
} else if dns.Truncated {
// Whether we ran into a an error or not, we want to return that it
// was truncated
err = ErrTruncated
}
return err
}
// Convert a complete message to a string with dig-like output.
func (dns *Msg) String() string {
if dns == nil {
return "<nil> MsgHdr"
}
s := dns.MsgHdr.String() + " "
s += "QUERY: " + strconv.Itoa(len(dns.Question)) + ", "
s += "ANSWER: " + strconv.Itoa(len(dns.Answer)) + ", "
s += "AUTHORITY: " + strconv.Itoa(len(dns.Ns)) + ", "
s += "ADDITIONAL: " + strconv.Itoa(len(dns.Extra)) + "\n"
if len(dns.Question) > 0 {
s += "\n;; QUESTION SECTION:\n"
for i := 0; i < len(dns.Question); i++ {
s += dns.Question[i].String() + "\n"
}
}
if len(dns.Answer) > 0 {
s += "\n;; ANSWER SECTION:\n"
for i := 0; i < len(dns.Answer); i++ {
if dns.Answer[i] != nil {
s += dns.Answer[i].String() + "\n"
}
}
}
if len(dns.Ns) > 0 {
s += "\n;; AUTHORITY SECTION:\n"
for i := 0; i < len(dns.Ns); i++ {
if dns.Ns[i] != nil {
s += dns.Ns[i].String() + "\n"
}
}
}
if len(dns.Extra) > 0 {
s += "\n;; ADDITIONAL SECTION:\n"
for i := 0; i < len(dns.Extra); i++ {
if dns.Extra[i] != nil {
s += dns.Extra[i].String() + "\n"
}
}
}
return s
}
// Len returns the message length when in (un)compressed wire format.
// If dns.Compress is true compression it is taken into account. Len()
// is provided to be a faster way to get the size of the resulting packet,
// than packing it, measuring the size and discarding the buffer.
func (dns *Msg) Len() int {
// We always return one more than needed.
l := 12 // Message header is always 12 bytes
var compression map[string]int
if dns.Compress {
compression = make(map[string]int)
}
for i := 0; i < len(dns.Question); i++ {
l += dns.Question[i].len()
if dns.Compress {
compressionLenHelper(compression, dns.Question[i].Name)
}
}
for i := 0; i < len(dns.Answer); i++ {
l += dns.Answer[i].len()
if dns.Compress {
k, ok := compressionLenSearch(compression, dns.Answer[i].Header().Name)
if ok {
l += 1 - k
}
compressionLenHelper(compression, dns.Answer[i].Header().Name)
k, ok = compressionLenSearchType(compression, dns.Answer[i])
if ok {
l += 1 - k
}
compressionLenHelperType(compression, dns.Answer[i])
}
}
for i := 0; i < len(dns.Ns); i++ {
l += dns.Ns[i].len()
if dns.Compress {
k, ok := compressionLenSearch(compression, dns.Ns[i].Header().Name)
if ok {
l += 1 - k
}
compressionLenHelper(compression, dns.Ns[i].Header().Name)
k, ok = compressionLenSearchType(compression, dns.Ns[i])
if ok {
l += 1 - k
}
compressionLenHelperType(compression, dns.Ns[i])
}
}
for i := 0; i < len(dns.Extra); i++ {
l += dns.Extra[i].len()
if dns.Compress {
k, ok := compressionLenSearch(compression, dns.Extra[i].Header().Name)
if ok {
l += 1 - k
}
compressionLenHelper(compression, dns.Extra[i].Header().Name)
k, ok = compressionLenSearchType(compression, dns.Extra[i])
if ok {
l += 1 - k
}
compressionLenHelperType(compression, dns.Extra[i])
}
}
return l
}
// Put the parts of the name in the compression map.
func compressionLenHelper(c map[string]int, s string) {
pref := ""
lbs := Split(s)
for j := len(lbs) - 1; j >= 0; j-- {
pref = s[lbs[j]:]
if _, ok := c[pref]; !ok {
c[pref] = len(pref)
}
}
}
// Look for each part in the compression map and returns its length,
// keep on searching so we get the longest match.
func compressionLenSearch(c map[string]int, s string) (int, bool) {
off := 0
end := false
if s == "" { // don't bork on bogus data
return 0, false
}
for {
if _, ok := c[s[off:]]; ok {
return len(s[off:]), true
}
if end {
break
}
off, end = NextLabel(s, off)
}
return 0, false
}
// TODO(miek): should add all types, because the all can be *used* for compression.
func compressionLenHelperType(c map[string]int, r RR) {
switch x := r.(type) {
case *NS:
compressionLenHelper(c, x.Ns)
case *MX:
compressionLenHelper(c, x.Mx)
case *CNAME:
compressionLenHelper(c, x.Target)
case *PTR:
compressionLenHelper(c, x.Ptr)
case *SOA:
compressionLenHelper(c, x.Ns)
compressionLenHelper(c, x.Mbox)
case *MB:
compressionLenHelper(c, x.Mb)
case *MG:
compressionLenHelper(c, x.Mg)
case *MR:
compressionLenHelper(c, x.Mr)
case *MF:
compressionLenHelper(c, x.Mf)
case *MD:
compressionLenHelper(c, x.Md)
case *RT:
compressionLenHelper(c, x.Host)
case *MINFO:
compressionLenHelper(c, x.Rmail)
compressionLenHelper(c, x.Email)
case *AFSDB:
compressionLenHelper(c, x.Hostname)
}
}
// Only search on compressing these types.
func compressionLenSearchType(c map[string]int, r RR) (int, bool) {
switch x := r.(type) {
case *NS:
return compressionLenSearch(c, x.Ns)
case *MX:
return compressionLenSearch(c, x.Mx)
case *CNAME:
return compressionLenSearch(c, x.Target)
case *PTR:
return compressionLenSearch(c, x.Ptr)
case *SOA:
k, ok := compressionLenSearch(c, x.Ns)
k1, ok1 := compressionLenSearch(c, x.Mbox)
if !ok && !ok1 {
return 0, false
}
return k + k1, true
case *MB:
return compressionLenSearch(c, x.Mb)
case *MG:
return compressionLenSearch(c, x.Mg)
case *MR:
return compressionLenSearch(c, x.Mr)
case *MF:
return compressionLenSearch(c, x.Mf)
case *MD:
return compressionLenSearch(c, x.Md)
case *RT:
return compressionLenSearch(c, x.Host)
case *MINFO:
k, ok := compressionLenSearch(c, x.Rmail)
k1, ok1 := compressionLenSearch(c, x.Email)
if !ok && !ok1 {
return 0, false
}
return k + k1, true
case *AFSDB:
return compressionLenSearch(c, x.Hostname)
}
return 0, false
}
// id returns a 16 bits random number to be used as a
// message id. The random provided should be good enough.
func id() uint16 {
return uint16(rand.Int()) ^ uint16(time.Now().Nanosecond())
}
// Copy returns a new RR which is a deep-copy of r.
func Copy(r RR) RR {
r1 := r.copy()
return r1
}
// Copy returns a new *Msg which is a deep-copy of dns.
func (dns *Msg) Copy() *Msg {
return dns.CopyTo(new(Msg))
}
// CopyTo copies the contents to the provided message using a deep-copy and returns the copy.
func (dns *Msg) CopyTo(r1 *Msg) *Msg {
r1.MsgHdr = dns.MsgHdr
r1.Compress = dns.Compress
if len(dns.Question) > 0 {
r1.Question = make([]Question, len(dns.Question))
copy(r1.Question, dns.Question) // TODO(miek): Question is an immutable value, ok to do a shallow-copy
}
rrArr := make([]RR, len(dns.Answer)+len(dns.Ns)+len(dns.Extra))
var rri int
if len(dns.Answer) > 0 {
rrbegin := rri
for i := 0; i < len(dns.Answer); i++ {
rrArr[rri] = dns.Answer[i].copy()
rri++
}
r1.Answer = rrArr[rrbegin:rri:rri]
}
if len(dns.Ns) > 0 {
rrbegin := rri
for i := 0; i < len(dns.Ns); i++ {
rrArr[rri] = dns.Ns[i].copy()
rri++
}
r1.Ns = rrArr[rrbegin:rri:rri]
}
if len(dns.Extra) > 0 {
rrbegin := rri
for i := 0; i < len(dns.Extra); i++ {
rrArr[rri] = dns.Extra[i].copy()
rri++
}
r1.Extra = rrArr[rrbegin:rri:rri]
}
return r1
}