mirror of
https://github.com/StackExchange/dnscontrol.git
synced 2024-11-15 04:36:19 +08:00
b7c251190f
* govet and golint corrections
104 lines
2.4 KiB
Go
104 lines
2.4 KiB
Go
package spflib
|
|
|
|
import (
|
|
"fmt"
|
|
"strings"
|
|
)
|
|
|
|
// TXT outputs s as a TXT record.
|
|
func (s *SPFRecord) TXT() string {
|
|
text := "v=spf1"
|
|
for _, p := range s.Parts {
|
|
text += " " + p.Text
|
|
}
|
|
return text
|
|
}
|
|
|
|
const maxLen = 255
|
|
|
|
// TXTSplit returns a set of txt records to use for SPF.
|
|
// pattern given is used to name all chained spf records.
|
|
// patern should include %d, which will be replaced by a counter.
|
|
// should result in fqdn after replacement
|
|
// returned map will have keys with fqdn of resulting records.
|
|
// root record will be under key "@"
|
|
func (s *SPFRecord) TXTSplit(pattern string) map[string]string {
|
|
m := map[string]string{}
|
|
s.split("@", pattern, 1, m)
|
|
return m
|
|
|
|
}
|
|
|
|
func (s *SPFRecord) split(thisfqdn string, pattern string, nextIdx int, m map[string]string) {
|
|
base := s.TXT()
|
|
// simple case. it fits
|
|
if len(base) <= maxLen {
|
|
m[thisfqdn] = base
|
|
return
|
|
}
|
|
|
|
// we need to trim.
|
|
// take parts while we fit
|
|
nextFQDN := fmt.Sprintf(pattern, nextIdx)
|
|
lastPart := s.Parts[len(s.Parts)-1]
|
|
tail := " include:" + nextFQDN + " " + lastPart.Text
|
|
thisText := "v=spf1"
|
|
|
|
newRec := &SPFRecord{}
|
|
over := false
|
|
addedCount := 0
|
|
for _, part := range s.Parts {
|
|
if !over {
|
|
if len(thisText)+1+len(part.Text)+len(tail) <= maxLen {
|
|
thisText += " " + part.Text
|
|
addedCount++
|
|
} else {
|
|
over = true
|
|
if addedCount == 0 {
|
|
// the first part is too big to include. We kinda have to give up here.
|
|
m[thisfqdn] = base
|
|
return
|
|
}
|
|
}
|
|
}
|
|
if over {
|
|
newRec.Parts = append(newRec.Parts, part)
|
|
}
|
|
}
|
|
m[thisfqdn] = thisText + tail
|
|
newRec.split(nextFQDN, pattern, nextIdx+1, m)
|
|
}
|
|
|
|
// Flatten optimizes s.
|
|
func (s *SPFRecord) Flatten(spec string) *SPFRecord {
|
|
newRec := &SPFRecord{}
|
|
for _, p := range s.Parts {
|
|
if p.IncludeRecord == nil {
|
|
// non-includes copy straight over
|
|
newRec.Parts = append(newRec.Parts, p)
|
|
} else if !matchesFlatSpec(spec, p.IncludeDomain) {
|
|
// includes that don't match get copied straight across
|
|
newRec.Parts = append(newRec.Parts, p)
|
|
} else {
|
|
// flatten child recursively
|
|
flattenedChild := p.IncludeRecord.Flatten(spec)
|
|
// include their parts (skipping final all term)
|
|
for _, childPart := range flattenedChild.Parts[:len(flattenedChild.Parts)-1] {
|
|
newRec.Parts = append(newRec.Parts, childPart)
|
|
}
|
|
}
|
|
}
|
|
return newRec
|
|
}
|
|
|
|
func matchesFlatSpec(spec, fqdn string) bool {
|
|
if spec == "*" {
|
|
return true
|
|
}
|
|
for _, p := range strings.Split(spec, ",") {
|
|
if p == fqdn {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|