lxd/instance/drivers/qemu: Lock the vsock ID selection

If multiple VMs are started (e.g. parent and imported clone) using the same vsock ID
a race condition might occur in which the already cleared vsock ID isn't bound yet to the socket.
This leads to the next VM picking the exact same one since it doesn't seem to be acquired.

Signed-off-by: Julian Pelizäus <[email protected]>

lxd/instance/drivers/driver_qemu.go

@@ -24,6 +24,7 @@ import (
 	"sort"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/flosch/pongo2"
@@ -134,6 +135,8 @@ const qemuSparseUSBPorts = 8
 
 var errQemuAgentOffline = fmt.Errorf("LXD VM agent isn't currently running")
 
+var nextVsockIDMutex = sync.Mutex{}
+
 type monitorHook func(m *qmp.Monitor) error
 
 // qemuLoad creates a Qemu instance from the supplied InstanceArgs.
@@ -7477,6 +7480,8 @@ func (d *qemu) freeVsockID(vsockID uint32) bool {
 
 // nextVsockID returns the next free vsock Context ID for the VM.
 // It tries to acquire one randomly until the timeout exceeds.
+// Concurrent requests from different routines will wait on each other
+// if a new vsock ID needs to be selected.
 func (d *qemu) nextVsockID() (uint32, error) {
 	// Check if vsock ID from last VM start is present in volatile, then use that.
 	// This allows a running VM to be recovered after DB record deletion and that an agent connection still works
@@ -7502,6 +7507,23 @@ func (d *qemu) nextVsockID() (uint32, error) {
 
 	timeout := 5 * time.Second
 
+	waitForVsockID := func(ctx context.Context, vsockID uint32) {
+		defer nextVsockIDMutex.Unlock()
+		ticker := time.NewTicker(50 * time.Millisecond)
+		defer ticker.Stop()
+
+		for {
+			select {
+			case <-ctx.Done():
+				return
+			case <-ticker.C:
+				if !d.freeVsockID(vsockID) {
+					return
+				}
+			}
+		}
+	}
+
 	// Try to find a new Context ID.
 	for start := time.Now(); time.Since(start) <= timeout; {
 		candidateVsockID := r.Uint32()
@@ -7511,6 +7533,8 @@ func (d *qemu) nextVsockID() (uint32, error) {
 		}
 
 		if d.freeVsockID(candidateVsockID) {
+			nextVsockIDMutex.Lock()
+			go waitForVsockID(d.state.ShutdownCtx, candidateVsockID)
 			return candidateVsockID, nil
 		}