lordwelch/router7
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

dhcp4: drop expired lease on server error (faster time to recovery)

Browse Source 
netconfigd still keeps the address configured for as long as possible,
but dhcp4 now more quickly returns to a from-scratch DHCP exchange.
This commit is contained in:
Michael Stapelberg 2024-12-21 16:07:56 +01:00
parent ed7523c311
commit af27264a03
1 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
cmd/dhcp4/dhcp4.go
View File

@ -35,6 +35,7 @@ import (
"github.com/google/gopacket/layers"
"github.com/google/renameio"
"github.com/jpillora/backoff"
rtr7dhcp4 "github.com/rtr7/dhcp4"
"github.com/rtr7/router7/internal/dhcp4"
"github.com/rtr7/router7/internal/netconfig"
"github.com/rtr7/router7/internal/notify"
@ -133,15 +134,36 @@ func logic() error {
Min: 10 * time.Second,
Max: 1 * time.Minute,
}
var lastSuccess time.Time
if st, err := os.Stat(ackFn); err == nil {
lastSuccess = st.ModTime()
}
log.Printf("last success: %v", lastSuccess)
ObtainOrRenew:
for c.ObtainOrRenew() {
if err := c.Err(); err != nil {
dur := backoff.Duration()
// Drop the lease if we do not get a reply from the DHCP server.
// I observed this in practice where over a period of days,
// the dhcp4 client would hang like this:
//
// dhcp4.go:140: Temporary error: DHCP: read packet
// 42:66:f1:f1:bd:e7: i/o timeout (waiting 1m0s)
//
// For brief periods of time, we probably want to paper over such
// issues, but after the lease expired, we should start the DHCP
// exchange from scratch.
if c.Ack != nil && time.Since(lastSuccess) > rtr7dhcp4.LeaseFromACK(c.Ack).RenewalTime {
log.Printf("Temporary error: %v (dropping lease and retrying)", err)
c.Ack = nil
continue
}
log.Printf("Temporary error: %v (waiting %v)", err, dur)
time.Sleep(dur)
continue
}
backoff.Reset()
lastSuccess = time.Now()
log.Printf("lease: %+v", c.Config())
b, err := json.Marshal(c.Config())
if err != nil {