3
0
mirror of https://github.com/ergochat/ergo.git synced 2024-12-23 19:22:41 +01:00
ergo/vendor/github.com/go-ldap/ldap/v3/dn.go

208 lines
6.0 KiB
Go
Raw Normal View History

package ldap
import (
"bytes"
enchex "encoding/hex"
"errors"
"fmt"
"strings"
ber "github.com/go-asn1-ber/asn1-ber"
)
// AttributeTypeAndValue represents an attributeTypeAndValue from https://tools.ietf.org/html/rfc4514
type AttributeTypeAndValue struct {
// Type is the attribute type
Type string
// Value is the attribute value
Value string
}
// RelativeDN represents a relativeDistinguishedName from https://tools.ietf.org/html/rfc4514
type RelativeDN struct {
Attributes []*AttributeTypeAndValue
}
// DN represents a distinguishedName from https://tools.ietf.org/html/rfc4514
type DN struct {
RDNs []*RelativeDN
}
// ParseDN returns a distinguishedName or an error.
// The function respects https://tools.ietf.org/html/rfc4514
func ParseDN(str string) (*DN, error) {
dn := new(DN)
dn.RDNs = make([]*RelativeDN, 0)
rdn := new(RelativeDN)
rdn.Attributes = make([]*AttributeTypeAndValue, 0)
buffer := bytes.Buffer{}
attribute := new(AttributeTypeAndValue)
escaping := false
unescapedTrailingSpaces := 0
stringFromBuffer := func() string {
s := buffer.String()
s = s[0 : len(s)-unescapedTrailingSpaces]
buffer.Reset()
unescapedTrailingSpaces = 0
return s
}
for i := 0; i < len(str); i++ {
char := str[i]
switch {
case escaping:
unescapedTrailingSpaces = 0
escaping = false
switch char {
case ' ', '"', '#', '+', ',', ';', '<', '=', '>', '\\':
buffer.WriteByte(char)
continue
}
// Not a special character, assume hex encoded octet
if len(str) == i+1 {
return nil, errors.New("got corrupted escaped character")
}
dst := []byte{0}
n, err := enchex.Decode([]byte(dst), []byte(str[i:i+2]))
if err != nil {
return nil, fmt.Errorf("failed to decode escaped character: %s", err)
} else if n != 1 {
return nil, fmt.Errorf("expected 1 byte when un-escaping, got %d", n)
}
buffer.WriteByte(dst[0])
i++
case char == '\\':
unescapedTrailingSpaces = 0
escaping = true
case char == '=':
attribute.Type = stringFromBuffer()
// Special case: If the first character in the value is # the
// following data is BER encoded so we can just fast forward
// and decode.
if len(str) > i+1 && str[i+1] == '#' {
i += 2
index := strings.IndexAny(str[i:], ",+")
data := str
if index > 0 {
data = str[i : i+index]
} else {
data = str[i:]
}
rawBER, err := enchex.DecodeString(data)
if err != nil {
return nil, fmt.Errorf("failed to decode BER encoding: %s", err)
}
packet, err := ber.DecodePacketErr(rawBER)
if err != nil {
return nil, fmt.Errorf("failed to decode BER packet: %s", err)
}
buffer.WriteString(packet.Data.String())
i += len(data) - 1
}
case char == ',' || char == '+':
// We're done with this RDN or value, push it
if len(attribute.Type) == 0 {
return nil, errors.New("incomplete type, value pair")
}
attribute.Value = stringFromBuffer()
rdn.Attributes = append(rdn.Attributes, attribute)
attribute = new(AttributeTypeAndValue)
if char == ',' {
dn.RDNs = append(dn.RDNs, rdn)
rdn = new(RelativeDN)
rdn.Attributes = make([]*AttributeTypeAndValue, 0)
}
case char == ' ' && buffer.Len() == 0:
// ignore unescaped leading spaces
continue
default:
if char == ' ' {
// Track unescaped spaces in case they are trailing and we need to remove them
unescapedTrailingSpaces++
} else {
// Reset if we see a non-space char
unescapedTrailingSpaces = 0
}
buffer.WriteByte(char)
}
}
if buffer.Len() > 0 {
if len(attribute.Type) == 0 {
return nil, errors.New("DN ended with incomplete type, value pair")
}
attribute.Value = stringFromBuffer()
rdn.Attributes = append(rdn.Attributes, attribute)
dn.RDNs = append(dn.RDNs, rdn)
}
return dn, nil
}
// Equal returns true if the DNs are equal as defined by rfc4517 4.2.15 (distinguishedNameMatch).
// Returns true if they have the same number of relative distinguished names
// and corresponding relative distinguished names (by position) are the same.
func (d *DN) Equal(other *DN) bool {
if len(d.RDNs) != len(other.RDNs) {
return false
}
for i := range d.RDNs {
if !d.RDNs[i].Equal(other.RDNs[i]) {
return false
}
}
return true
}
// AncestorOf returns true if the other DN consists of at least one RDN followed by all the RDNs of the current DN.
// "ou=widgets,o=acme.com" is an ancestor of "ou=sprockets,ou=widgets,o=acme.com"
// "ou=widgets,o=acme.com" is not an ancestor of "ou=sprockets,ou=widgets,o=foo.com"
// "ou=widgets,o=acme.com" is not an ancestor of "ou=widgets,o=acme.com"
func (d *DN) AncestorOf(other *DN) bool {
if len(d.RDNs) >= len(other.RDNs) {
return false
}
// Take the last `len(d.RDNs)` RDNs from the other DN to compare against
otherRDNs := other.RDNs[len(other.RDNs)-len(d.RDNs):]
for i := range d.RDNs {
if !d.RDNs[i].Equal(otherRDNs[i]) {
return false
}
}
return true
}
// Equal returns true if the RelativeDNs are equal as defined by rfc4517 4.2.15 (distinguishedNameMatch).
// Relative distinguished names are the same if and only if they have the same number of AttributeTypeAndValues
// and each attribute of the first RDN is the same as the attribute of the second RDN with the same attribute type.
// The order of attributes is not significant.
// Case of attribute types is not significant.
func (r *RelativeDN) Equal(other *RelativeDN) bool {
if len(r.Attributes) != len(other.Attributes) {
return false
}
return r.hasAllAttributes(other.Attributes) && other.hasAllAttributes(r.Attributes)
}
func (r *RelativeDN) hasAllAttributes(attrs []*AttributeTypeAndValue) bool {
for _, attr := range attrs {
found := false
for _, myattr := range r.Attributes {
if myattr.Equal(attr) {
found = true
break
}
}
if !found {
return false
}
}
return true
}
// Equal returns true if the AttributeTypeAndValue is equivalent to the specified AttributeTypeAndValue
// Case of the attribute type is not significant
func (a *AttributeTypeAndValue) Equal(other *AttributeTypeAndValue) bool {
return strings.EqualFold(a.Type, other.Type) && a.Value == other.Value
}