[client] Use native windows sock opts to avoid routing loops (#4314)

- Move `util/grpc` and `util/net` to `client` so `internal` packages can be accessed
 - Add methods to return the next best interface after the NetBird interface.
- Use `IP_UNICAST_IF` sock opt to force the outgoing interface for the NetBird `net.Dialer` and `net.ListenerConfig` to avoid routing loops. The interface is picked by the new route lookup method.
- Some refactoring to avoid import cycles
- Old behavior is available through `NB_USE_LEGACY_ROUTING=true` env var
This commit is contained in:
Viktor Liu
2025-09-20 09:31:04 +02:00
committed by GitHub
parent 90577682e4
commit 55126f990c
77 changed files with 1180 additions and 606 deletions

View File

@@ -3,7 +3,6 @@
package systemops
import (
"context"
"errors"
"fmt"
"net"
@@ -22,7 +21,7 @@ import (
"github.com/netbirdio/netbird/client/internal/routemanager/util"
"github.com/netbirdio/netbird/client/internal/routemanager/vars"
"github.com/netbirdio/netbird/client/internal/statemanager"
nbnet "github.com/netbirdio/netbird/util/net"
"github.com/netbirdio/netbird/client/net/hooks"
)
const localSubnetsCacheTTL = 15 * time.Minute
@@ -96,9 +95,9 @@ func (r *SysOps) cleanupRefCounter(stateManager *statemanager.Manager) error {
return nil
}
// TODO: Remove hooks selectively
nbnet.RemoveDialerHooks()
nbnet.RemoveListenerHooks()
hooks.RemoveWriteHooks()
hooks.RemoveCloseHooks()
hooks.RemoveAddressRemoveHooks()
if err := r.refCounter.Flush(); err != nil {
return fmt.Errorf("flush route manager: %w", err)
@@ -290,12 +289,7 @@ func (r *SysOps) genericRemoveVPNRoute(prefix netip.Prefix, intf *net.Interface)
}
func (r *SysOps) setupHooks(initAddresses []net.IP, stateManager *statemanager.Manager) error {
beforeHook := func(connID nbnet.ConnectionID, ip net.IP) error {
prefix, err := util.GetPrefixFromIP(ip)
if err != nil {
return fmt.Errorf("convert ip to prefix: %w", err)
}
beforeHook := func(connID hooks.ConnectionID, prefix netip.Prefix) error {
if _, err := r.refCounter.IncrementWithID(string(connID), prefix, struct{}{}); err != nil {
return fmt.Errorf("adding route reference: %v", err)
}
@@ -304,7 +298,7 @@ func (r *SysOps) setupHooks(initAddresses []net.IP, stateManager *statemanager.M
return nil
}
afterHook := func(connID nbnet.ConnectionID) error {
afterHook := func(connID hooks.ConnectionID) error {
if err := r.refCounter.DecrementWithID(string(connID)); err != nil {
return fmt.Errorf("remove route reference: %w", err)
}
@@ -317,36 +311,20 @@ func (r *SysOps) setupHooks(initAddresses []net.IP, stateManager *statemanager.M
var merr *multierror.Error
for _, ip := range initAddresses {
if err := beforeHook("init", ip); err != nil {
merr = multierror.Append(merr, fmt.Errorf("add initial route for %s: %w", ip, err))
prefix, err := util.GetPrefixFromIP(ip)
if err != nil {
merr = multierror.Append(merr, fmt.Errorf("invalid IP address %s: %w", ip, err))
continue
}
if err := beforeHook("init", prefix); err != nil {
merr = multierror.Append(merr, fmt.Errorf("add initial route for %s: %w", prefix, err))
}
}
nbnet.AddDialerHook(func(ctx context.Context, connID nbnet.ConnectionID, resolvedIPs []net.IPAddr) error {
if ctx.Err() != nil {
return ctx.Err()
}
hooks.AddWriteHook(beforeHook)
hooks.AddCloseHook(afterHook)
var merr *multierror.Error
for _, ip := range resolvedIPs {
merr = multierror.Append(merr, beforeHook(connID, ip.IP))
}
return nberrors.FormatErrorOrNil(merr)
})
nbnet.AddDialerCloseHook(func(connID nbnet.ConnectionID, conn *net.Conn) error {
return afterHook(connID)
})
nbnet.AddListenerWriteHook(func(connID nbnet.ConnectionID, ip *net.IPAddr, data []byte) error {
return beforeHook(connID, ip.IP)
})
nbnet.AddListenerCloseHook(func(connID nbnet.ConnectionID, conn net.PacketConn) error {
return afterHook(connID)
})
nbnet.AddListenerAddressRemoveHook(func(connID nbnet.ConnectionID, prefix netip.Prefix) error {
hooks.AddAddressRemoveHook(func(connID hooks.ConnectionID, prefix netip.Prefix) error {
if _, err := r.refCounter.Decrement(prefix); err != nil {
return fmt.Errorf("remove route reference: %w", err)
}