Merge pull request #12405 from tomponline/tp-vm-vsockid

lxd-agent: Fixes intermittent exec EOF closure when vsock listener is restarted just after boot

lxd-agent/api_1.0.go

@@ -1,14 +1,12 @@
 package main
 
 import (
-	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
 	"io"
 	"net/http"
 	"os"
-	"time"
 
 	"github.com/mdlayher/vsock"
 
@@ -193,41 +191,19 @@ func getClient(CID uint32, port int, serverCertificate string) (*http.Client, er
 	return client, nil
 }
 
-// waitVsockContextID checks for valid local context ID and returns it.
-// If no valid context ID has been ascertained when the context is cancelled, the last error is returned.
-func waitVsockContextID(ctx context.Context) (uint32, error) {
+func startHTTPServer(d *Daemon, debug bool) error {
 	const CIDAny uint32 = 4294967295 // Equivalent to VMADDR_CID_ANY.
 
-	for {
-		cid, err := vsock.ContextID()
-		if cid == CIDAny {
-			// Ignore VMADDR_CID_ANY as this seems to indicate the vsock module is still initialising.
-			err = fmt.Errorf("Invalid context ID %d", cid)
-		} else if err == nil {
-			return cid, nil
-		}
-
-		ctxErr := ctx.Err()
-		if ctxErr != nil {
-			if err != nil {
-				return 0, err
-			}
-
-			return 0, ctxErr
-		}
-
-		time.Sleep(time.Second)
-	}
-}
-
-func startHTTPServer(d *Daemon, debug bool) error {
-	// Setup the listener on VM's context ID for inbound connections from LXD.
-	l, err := vsock.ListenContextID(d.localCID, shared.HTTPSDefaultPort, nil)
+	// Setup the listener on wildcard CID for inbound connections from LXD.
+	// We use the VMADDR_CID_ANY CID so that if the VM's CID changes in the future the listener still works.
+	// A CID change can occur when restoring a stateful VM that was previously using one CID but is
+	// subsequently restored using a different one.
+	l, err := vsock.ListenContextID(CIDAny, shared.HTTPSDefaultPort, nil)
 	if err != nil {
 		return fmt.Errorf("Failed to listen on vsock: %w", err)
 	}
 
-	logger.Info("Started vsock listener", logger.Ctx{"contextID": d.localCID})
+	logger.Info("Started vsock listener")
 
 	// Load the expected server certificate.
 	cert, err := shared.ReadCert("server.crt")

lxd-agent/daemon.go

@@ -16,8 +16,6 @@ type Daemon struct {
 	serverPort        uint32
 	serverCertificate string
 
-	localCID uint32
-
 	// The channel which is used to indicate that the lxd-agent was able to connect to LXD.
 	chConnected chan struct{}
 

lxd-agent/main_agent.go

@@ -129,45 +129,12 @@ func (c *cmdAgent) Run(cmd *cobra.Command, args []string) error {
 
 	d := newDaemon(c.global.flagLogDebug, c.global.flagLogVerbose)
 
-	// Wait up to 30s to get a valid local vsock context ID.
-	ctx, cancel := context.WithTimeout(context.Background(), time.Second*30)
-	d.localCID, err = waitVsockContextID(ctx)
-	if err != nil {
-		cancel()
-		return fmt.Errorf("Failed getting vsock context ID: %w", err)
-	}
-
-	cancel()
-
 	// Start the server.
 	err = startHTTPServer(d, c.global.flagLogDebug)
 	if err != nil {
 		return fmt.Errorf("Failed to start HTTP server: %w", err)
 	}
 
-	// Check context ID periodically, and restart the HTTP server if needed.
-	go func() {
-		for {
-			time.Sleep(30 * time.Second)
-			cid, err := waitVsockContextID(context.Background())
-			if err != nil || d.localCID == cid {
-				continue
-			}
-
-			// Restart server
-			logger.Warn("Restarting the vsock server due to context ID change", logger.Ctx{"oldID": d.localCID, "newID": cid})
-			servers["http"].Close()
-
-			// Update context ID.
-			d.localCID = cid
-
-			err = startHTTPServer(d, c.global.flagLogDebug)
-			if err != nil {
-				errChan <- err
-			}
-		}
-	}()
-
 	// Check whether we should start the devlxd server in the early setup. This way, /dev/lxd/sock
 	// will be available for any systemd services starting after the lxd-agent.
 	if shared.PathExists("agent.conf") {

lxd/vsock/vsock.go

@@ -12,11 +12,6 @@ import (
 	"github.com/canonical/lxd/shared"
 )
 
-// ContextID returns the local VM sockets context ID.
-func ContextID() (uint32, error) {
-	return vsock.ContextID()
-}
-
 // Dial connects to a remote vsock.
 func Dial(cid, port uint32) (net.Conn, error) {
 	return vsock.Dial(cid, port, nil)