syz-ci: switch to using syz-manager for smoke testing

Add smoke testing mode to manager and use it in syz-ci
instead of pkg/instance which uses syz-fuzzer binary.

3 files changed, 128 insertions, 51 deletions

diff --git a/pkg/report/report.go b/pkg/report/report.go
index 514a47734..2003b8b83 100644
--- a/pkg/report/report.go
+++ b/pkg/report/report.go
@@ -65,6 +65,8 @@ type Report struct {
 	Recipients vcs.Recipients
 	// GuiltyFile is the source file that we think is to blame for the crash  (filled in by Symbolize).
 	GuiltyFile string
+	// Arbitrary information about the test VM, may be attached to the report by users of the package.
+	MachineInfo []byte
 	// reportPrefixLen is length of additional prefix lines that we added before actual crash report.
 	reportPrefixLen int
 	// symbolized is set if the report is symbolized.
diff --git a/syz-ci/manager.go b/syz-ci/manager.go
index 38fcc4a77..65b5385ef 100644
--- a/syz-ci/manager.go
+++ b/syz-ci/manager.go
@@ -445,60 +445,63 @@ func (mgr *Manager) testImage(imageDir string, info *BuildInfo) error {
 	if err != nil {
 		return fmt.Errorf("failed to create manager config: %w", err)
 	}
-	defer os.RemoveAll(mgrcfg.Workdir)
 	if !vm.AllowsOvercommit(mgrcfg.Type) {
 		return nil // No support for creating machines out of thin air.
 	}
-	env, err := instance.NewEnv(mgrcfg, buildSem, testSem)
-	if err != nil {
+	osutil.MkdirAll(mgrcfg.Workdir)
+	configFile := filepath.Join(mgrcfg.Workdir, "manager.cfg")
+	if err := config.SaveFile(configFile, mgrcfg); err != nil {
 		return err
 	}
-	const (
-		testVMs     = 3
-		maxFailures = 1
-	)
-	results, err := env.Test(testVMs, nil, nil, nil)
-	if err != nil {
-		return err
+
+	testSem.Wait()
+	defer testSem.Signal()
+
+	timeout := 30 * time.Minute * mgrcfg.Timeouts.Scale
+	bin := filepath.Join(mgrcfg.Syzkaller, "bin", "syz-manager")
+	output, retErr := osutil.RunCmd(timeout, "", bin, "-config", configFile, "-mode=smoke-test")
+	if retErr == nil {
+		return nil
 	}
-	failures := 0
-	var failureErr error
-	for _, res := range results {
-		if res.Error == nil {
-			continue
-		}
-		failures++
-		var err *instance.TestError
-		switch {
-		case errors.As(res.Error, &err):
-			if rep := err.Report; rep != nil {
-				what := "test"
-				if err.Boot {
-					what = "boot"
-				}
-				rep.Title = fmt.Sprintf("%v %v error: %v",
-					mgr.mgrcfg.RepoAlias, what, rep.Title)
-				// There are usually no duplicates for boot errors, so we reset AltTitles.
-				// But if we pass them, we would need to add the same prefix as for Title
-				// in order to avoid duping boot bugs with non-boot bugs.
-				rep.AltTitles = nil
-				if err := mgr.reportBuildError(rep, info, imageDir); err != nil {
-					mgr.Errorf("failed to report image error: %v", err)
-				}
-			}
-			if err.Boot {
-				failureErr = fmt.Errorf("VM boot failed with: %w", err)
+
+	var verboseErr *osutil.VerboseError
+	if errors.As(retErr, &verboseErr) {
+		// Caller will log the error, so don't include full output.
+		retErr = errors.New(verboseErr.Title)
+	}
+	// If there was a kernel bug, report it to dashboard.
+	// Otherwise just save the output in a temp file and log an error, unclear what else we can do.
+	reportData, err := os.ReadFile(filepath.Join(mgrcfg.Workdir, "report.json"))
+	if err != nil {
+		if os.IsNotExist(err) {
+			mgr.Errorf("image testing failed w/o kernel bug")
+			tmp, err := os.CreateTemp(mgr.workDir, "smoke-test-error")
+			if err != nil {
+				mgr.Errorf("failed to create smoke test error file: %v", err)
 			} else {
-				failureErr = fmt.Errorf("VM testing failed with: %w", err)
+				tmp.Write(output)
+				tmp.Close()
+			}
+		} else {
+			mgr.Errorf("failed to read smoke test report: %v", err)
+		}
+	} else {
+		rep := new(report.Report)
+		if err := json.Unmarshal(reportData, rep); err != nil {
+			mgr.Errorf("failed to unmarshal smoke test report: %v", err)
+		} else {
+			rep.Title = fmt.Sprintf("%v test error: %v", mgr.mgrcfg.RepoAlias, rep.Title)
+			retErr = errors.New(rep.Title)
+			// There are usually no duplicates for boot errors, so we reset AltTitles.
+			// But if we pass them, we would need to add the same prefix as for Title
+			// in order to avoid duping boot bugs with non-boot bugs.
+			rep.AltTitles = nil
+			if err := mgr.reportBuildError(rep, info, imageDir); err != nil {
+				mgr.Errorf("failed to report image error: %v", err)
 			}
-		default:
-			failureErr = res.Error
 		}
 	}
-	if failures > maxFailures {
-		return failureErr
-	}
-	return nil
+	return retErr
 }
 
 func (mgr *Manager) reportBuildError(rep *report.Report, info *BuildInfo, imageDir string) error {
@@ -550,6 +553,9 @@ func (mgr *Manager) createTestConfig(imageDir string, info *BuildInfo) (*mgrconf
 	if err := instance.SetConfigImage(mgrcfg, imageDir, true); err != nil {
 		return nil, err
 	}
+	if err := instance.OverrideVMCount(mgrcfg, 3); err != nil {
+		return nil, err
+	}
 	mgrcfg.KernelSrc = mgr.kernelSrcDir
 	if err := mgrconfig.Complete(mgrcfg); err != nil {
 		return nil, fmt.Errorf("bad manager config: %w", err)
diff --git a/syz-manager/manager.go b/syz-manager/manager.go
index 0758930c8..ff29ed73b 100644
--- a/syz-manager/manager.go
+++ b/syz-manager/manager.go
@@ -7,6 +7,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"flag"
 	"fmt"
 	"io"
@@ -47,10 +48,19 @@ var (
 	flagConfig = flag.String("config", "", "configuration file")
 	flagDebug  = flag.Bool("debug", false, "dump all VM output to console")
 	flagBench  = flag.String("bench", "", "write execution statistics into this file periodically")
+
+	flagMode = flag.String("mode", "fuzzing", "mode of operation, one of:\n"+
+		" - fuzzing: the default continuous fuzzing mode\n"+
+		" - smoke-test: run smoke test for syzkaller+kernel\n"+
+		"	The test consists of booting VMs and running some simple test programs\n"+
+		"	to ensure that fuzzing can proceed in general. After completing the test\n"+
+		"	the process exits and the exit status indicates success/failure.\n"+
+		"	If the kernel oopses during testing, the report is saved to workdir/report.json.\n")
 )
 
 type Manager struct {
 	cfg             *mgrconfig.Config
+	mode            Mode
 	vmPool          *vm.Pool
 	target          *prog.Target
 	sysTarget       *targets.Target
@@ -100,6 +110,14 @@ type Manager struct {
 	Stats
 }
 
+type Mode int
+
+// For description of modes see flagMode help.
+const (
+	ModeFuzzing Mode = iota
+	ModeSmokeTest
+)
+
 const (
 	// Just started, nothing done yet.
 	phaseInit = iota
@@ -122,7 +140,6 @@ type Crash struct {
 	fromHub       bool // this crash was created based on a repro from syz-hub
 	fromDashboard bool // .. or from dashboard
 	*report.Report
-	machineInfo []byte
 }
 
 func main() {
@@ -143,6 +160,19 @@ func main() {
 }
 
 func RunManager(cfg *mgrconfig.Config) {
+	var mode Mode
+	switch *flagMode {
+	case "fuzzing":
+		mode = ModeFuzzing
+	case "smoke-test":
+		mode = ModeSmokeTest
+		cfg.DashboardClient = ""
+		cfg.HubClient = ""
+	default:
+		flag.PrintDefaults()
+		log.Fatalf("unknown mode: %v", *flagMode)
+	}
+
 	var vmPool *vm.Pool
 	// Type "none" is a special case for debugging/development when manager
 	// does not start any VMs, but instead you start them manually
@@ -166,6 +196,7 @@ func RunManager(cfg *mgrconfig.Config) {
 	corpusUpdates := make(chan corpus.NewItemEvent, 32)
 	mgr := &Manager{
 		cfg:                cfg,
+		mode:               mode,
 		vmPool:             vmPool,
 		corpus:             corpus.NewMonitoredCorpus(context.Background(), corpusUpdates),
 		corpusPreloaded:    make(chan bool),
@@ -217,7 +248,9 @@ func RunManager(cfg *mgrconfig.Config) {
 	}
 
 	go mgr.heartbeatLoop()
-	osutil.HandleInterrupts(vm.Shutdown)
+	if mgr.mode != ModeSmokeTest {
+		osutil.HandleInterrupts(vm.Shutdown)
+	}
 	if mgr.vmPool == nil {
 		log.Logf(0, "no VMs started (type=none)")
 		log.Logf(0, "you are supposed to start syz-fuzzer manually as:")
@@ -730,8 +763,11 @@ func (mgr *Manager) runInstance(index int) (*Crash, error) {
 
 	rep, vmInfo, err := mgr.runInstanceInner(index, instanceName, injectLog)
 	machineInfo := mgr.serv.shutdownInstance(instanceName, rep != nil)
-	if len(vmInfo) != 0 {
-		machineInfo = append(append(vmInfo, '\n'), machineInfo...)
+	if rep != nil {
+		if len(vmInfo) != 0 {
+			machineInfo = append(append(vmInfo, '\n'), machineInfo...)
+		}
+		rep.MachineInfo = machineInfo
 	}
 
 	// Error that is not a VM crash.
@@ -745,7 +781,6 @@ func (mgr *Manager) runInstance(index int) (*Crash, error) {
 	crash := &Crash{
 		instanceName: instanceName,
 		Report:       rep,
-		machineInfo:  machineInfo,
 	}
 	return crash, nil
 }
@@ -756,6 +791,22 @@ func (mgr *Manager) runInstanceInner(index int, instanceName string, injectLog <
 
 	inst, err := mgr.vmPool.Create(index)
 	if err != nil {
+		var bootErr vm.BootErrorer
+		if errors.As(err, &bootErr) {
+			title, output := bootErr.BootError()
+			rep := mgr.reporter.Parse(output)
+			if rep != nil && rep.Type == crash_pkg.UnexpectedReboot {
+				// Avoid detecting any boot crash as "unexpected kernel reboot".
+				rep = mgr.reporter.ParseFrom(output, rep.SkipPos)
+			}
+			if rep == nil {
+				rep = &report.Report{
+					Title:  title,
+					Output: output,
+				}
+			}
+			return rep, nil, nil
+		}
 		return nil, nil, fmt.Errorf("failed to create instance: %w", err)
 	}
 	defer inst.Close()
@@ -873,6 +924,17 @@ func (mgr *Manager) saveCrash(crash *Crash) bool {
 	}
 	log.Logf(0, "%s: crash: %v%v", crash.instanceName, crash.Title, flags)
 
+	if mgr.mode == ModeSmokeTest {
+		data, err := json.Marshal(crash.Report)
+		if err != nil {
+			log.Fatalf("failed to serialize crash report: %v", err)
+		}
+		if err := osutil.WriteFile(filepath.Join(mgr.cfg.Workdir, "report.json"), data); err != nil {
+			log.Fatal(err)
+		}
+		log.Fatalf("kernel crashed in smoke testing mode, exiting")
+	}
+
 	if crash.Suppressed {
 		// Collect all of them into a single bucket so that it's possible to control and assess them,
 		// e.g. if there are some spikes in suppressed reports.
@@ -901,7 +963,7 @@ func (mgr *Manager) saveCrash(crash *Crash) bool {
 			Recipients:  crash.Recipients.ToDash(),
 			Log:         crash.Output,
 			Report:      crash.Report.Report,
-			MachineInfo: crash.machineInfo,
+			MachineInfo: crash.MachineInfo,
 		}
 		setGuiltyFiles(dc, crash.Report)
 		resp, err := mgr.dash.ReportCrash(dc)
@@ -952,7 +1014,7 @@ func (mgr *Manager) saveCrash(crash *Crash) bool {
 	writeOrRemove("log", crash.Output)
 	writeOrRemove("tag", []byte(mgr.cfg.Tag))
 	writeOrRemove("report", crash.Report.Report)
-	writeOrRemove("machineInfo", crash.machineInfo)
+	writeOrRemove("machineInfo", crash.MachineInfo)
 	return mgr.needLocalRepro(crash)
 }
 
@@ -1335,6 +1397,13 @@ func (mgr *Manager) currentBugFrames() BugFrames {
 
 func (mgr *Manager) machineChecked(features flatrpc.Feature, enabledSyscalls map[*prog.Syscall]bool,
 	opts flatrpc.ExecOpts) queue.Source {
+	if mgr.mode == ModeSmokeTest {
+		log.Logf(0, "smoke test succeeded, shutting down...")
+		close(vm.Shutdown)
+		time.Sleep(10 * time.Second)
+		os.Exit(0)
+	}
+
 	mgr.mu.Lock()
 	defer mgr.mu.Unlock()
 	if mgr.checkDone {