syzkaller/syz-ci/jobs.go
Dmitry Vyukov b48312c59d syz-ci: retry test jobs up to 3 times
GCE can be buggy, kernel can be buggy.
2018-01-16 13:37:14 +01:00

399 lines
12 KiB
Go

// Copyright 2017 syzkaller project authors. All rights reserved.
// Use of this source code is governed by Apache 2 LICENSE that can be found in the LICENSE file.
package main
import (
"fmt"
"io/ioutil"
"os"
"path/filepath"
"time"
"github.com/google/syzkaller/dashboard/dashapi"
"github.com/google/syzkaller/pkg/config"
"github.com/google/syzkaller/pkg/csource"
"github.com/google/syzkaller/pkg/git"
"github.com/google/syzkaller/pkg/kernel"
. "github.com/google/syzkaller/pkg/log"
"github.com/google/syzkaller/pkg/osutil"
"github.com/google/syzkaller/pkg/report"
"github.com/google/syzkaller/prog"
"github.com/google/syzkaller/syz-manager/mgrconfig"
"github.com/google/syzkaller/vm"
)
type JobProcessor struct {
name string
managers []*Manager
dash *dashapi.Dashboard
syzkallerRepo string
syzkallerBranch string
}
func newJobProcessor(cfg *Config, managers []*Manager) *JobProcessor {
jp := &JobProcessor{
name: fmt.Sprintf("%v-job", cfg.Name),
managers: managers,
syzkallerRepo: cfg.Syzkaller_Repo,
syzkallerBranch: cfg.Syzkaller_Branch,
}
if cfg.Dashboard_Addr != "" && cfg.Dashboard_Client != "" {
jp.dash = dashapi.New(cfg.Dashboard_Client, cfg.Dashboard_Addr, cfg.Dashboard_Key)
}
return jp
}
func (jp *JobProcessor) loop(stop chan struct{}) {
if jp.dash == nil {
return
}
ticker := time.NewTicker(time.Minute)
defer ticker.Stop()
for {
select {
case <-ticker.C:
jp.poll()
case <-stop:
Logf(0, "job loop stopped")
return
}
}
}
func (jp *JobProcessor) poll() {
var names []string
for _, mgr := range jp.managers {
names = append(names, mgr.name)
}
req, err := jp.dash.JobPoll(names)
if err != nil {
jp.Errorf("failed to poll jobs: %v", err)
return
}
if req.ID == "" {
return
}
var mgr *Manager
for _, m := range jp.managers {
if m.name == req.Manager {
mgr = m
break
}
}
if mgr == nil {
jp.Errorf("got job for unknown manager: %v", req.Manager)
return
}
job := &Job{
req: req,
mgr: mgr,
}
Logf(0, "starting job %v for manager %v on %v/%v",
req.ID, req.Manager, req.KernelRepo, req.KernelBranch)
resp := jp.process(job)
Logf(0, "done job %v: commit %v, crash %q, error: %s",
resp.ID, resp.Build.KernelCommit, resp.CrashTitle, resp.Error)
if err := jp.dash.JobDone(resp); err != nil {
jp.Errorf("failed to mark job as done: %v", err)
return
}
}
type Job struct {
req *dashapi.JobPollResp
resp *dashapi.JobDoneReq
mgr *Manager
mgrcfg *mgrconfig.Config
}
func (jp *JobProcessor) process(job *Job) *dashapi.JobDoneReq {
req, mgr := job.req, job.mgr
build := dashapi.Build{
Manager: mgr.name,
ID: req.ID,
OS: mgr.managercfg.TargetOS,
Arch: mgr.managercfg.TargetArch,
VMArch: mgr.managercfg.TargetVMArch,
CompilerID: mgr.compilerID,
KernelRepo: req.KernelRepo,
KernelBranch: req.KernelBranch,
KernelCommit: "[unknown]",
SyzkallerCommit: "[unknown]",
}
job.resp = &dashapi.JobDoneReq{
ID: req.ID,
Build: build,
}
required := []struct {
name string
ok bool
}{
{"kernel repository", req.KernelRepo != ""},
{"kernel branch", req.KernelBranch != ""},
{"kernel config", len(req.KernelConfig) != 0},
{"syzkaller commit", req.SyzkallerCommit != ""},
{"test patch", len(req.Patch) != 0},
{"reproducer options", len(req.ReproOpts) != 0},
{"reproducer program", len(req.ReproSyz) != 0},
}
for _, req := range required {
if !req.ok {
job.resp.Error = []byte(req.name + " is empty")
jp.Errorf("%s", job.resp.Error)
return job.resp
}
}
// TODO(dvyukov): this will only work for qemu/gce,
// because e.g. adb requires unique device IDs and we can't use what
// manager already uses. For qemu/gce this is also bad, because we
// override resource limits specified in config (e.g. can OOM), but works.
switch typ := mgr.managercfg.Type; typ {
case "gce", "qemu":
default:
job.resp.Error = []byte(fmt.Sprintf("testing is not yet supported for %v machine type.", typ))
jp.Errorf("%s", job.resp.Error)
return job.resp
}
if err := jp.buildImage(job); err != nil {
job.resp.Error = []byte(err.Error())
return job.resp
}
var err error
for try := 0; try < 3; try++ {
if err = jp.test(job); err == nil {
break
}
Logf(0, "job: testing failed, trying once again\n%v", err)
}
if err != nil {
job.resp.Error = []byte(err.Error())
}
return job.resp
}
func (jp *JobProcessor) buildImage(job *Job) error {
kernelBuildSem <- struct{}{}
defer func() { <-kernelBuildSem }()
req, resp, mgr := job.req, job.resp, job.mgr
dir := osutil.Abs(filepath.Join("jobs", mgr.managercfg.TargetOS))
kernelDir := filepath.Join(dir, "kernel")
if err := osutil.MkdirAll(kernelDir); err != nil {
return fmt.Errorf("failed to create temp dir: %v", err)
}
imageDir := filepath.Join(dir, "image")
os.RemoveAll(imageDir)
if err := osutil.MkdirAll(imageDir); err != nil {
return fmt.Errorf("failed to create temp dir: %v", err)
}
workDir := filepath.Join(dir, "workdir")
os.RemoveAll(workDir)
if err := osutil.MkdirAll(workDir); err != nil {
return fmt.Errorf("failed to create temp dir: %v", err)
}
gopathDir := filepath.Join(dir, "gopath")
syzkallerDir := filepath.Join(gopathDir, "src", "github.com", "google", "syzkaller")
if err := osutil.MkdirAll(syzkallerDir); err != nil {
return fmt.Errorf("failed to create temp dir: %v", err)
}
Logf(0, "job: fetching syzkaller on %v...", req.SyzkallerCommit)
err := git.CheckoutCommit(syzkallerDir, jp.syzkallerRepo, jp.syzkallerBranch, req.SyzkallerCommit)
if err != nil {
return fmt.Errorf("failed to checkout syzkaller repo: %v", err)
}
Logf(0, "job: building syzkaller...")
cmd := osutil.Command("make", "target")
cmd.Dir = syzkallerDir
cmd.Env = append([]string{}, os.Environ()...)
cmd.Env = append(cmd.Env,
"GOPATH="+gopathDir,
"TARGETOS="+mgr.managercfg.TargetOS,
"TARGETVMARCH="+mgr.managercfg.TargetVMArch,
"TARGETARCH="+mgr.managercfg.TargetArch,
)
if _, err := osutil.Run(time.Hour, cmd); err != nil {
return fmt.Errorf("syzkaller build failed: %v", err)
}
resp.Build.SyzkallerCommit = req.SyzkallerCommit
Logf(0, "job: fetching kernel...")
kernelCommit, err := git.Checkout(kernelDir, req.KernelRepo, req.KernelBranch)
if err != nil {
return fmt.Errorf("failed to checkout kernel repo: %v", err)
}
resp.Build.KernelCommit = kernelCommit
if err := git.Patch(kernelDir, req.Patch); err != nil {
return err
}
Logf(0, "job: building kernel...")
configFile := filepath.Join(dir, "kernel.config")
if err := osutil.WriteFile(configFile, req.KernelConfig); err != nil {
return fmt.Errorf("failed to write temp file: %v", err)
}
if err := kernel.Build(kernelDir, mgr.mgrcfg.Compiler, configFile); err != nil {
return fmt.Errorf("kernel build failed: %v", err)
}
kernelConfig, err := ioutil.ReadFile(filepath.Join(kernelDir, ".config"))
if err != nil {
return fmt.Errorf("failed to read config file: %v", err)
}
resp.Build.KernelConfig = kernelConfig
Logf(0, "job: creating image...")
image := filepath.Join(imageDir, "image")
key := filepath.Join(imageDir, "key")
err = kernel.CreateImage(kernelDir, mgr.mgrcfg.Userspace,
mgr.mgrcfg.Kernel_Cmdline, mgr.mgrcfg.Kernel_Sysctl, image, key)
if err != nil {
return fmt.Errorf("image build failed: %v", err)
}
mgrcfg := new(mgrconfig.Config)
*mgrcfg = *mgr.managercfg
mgrcfg.Name += "-job"
mgrcfg.Workdir = workDir
mgrcfg.Vmlinux = filepath.Join(kernelDir, "vmlinux")
mgrcfg.Kernel_Src = kernelDir
mgrcfg.Syzkaller = syzkallerDir
mgrcfg.Image = image
mgrcfg.Sshkey = key
// Reload config to fill derived fields (ugly hack).
cfgdata, err := config.SaveData(mgrcfg)
if err != nil {
return fmt.Errorf("failed to save manager config: %v", err)
}
if job.mgrcfg, err = mgrconfig.LoadData(cfgdata); err != nil {
return fmt.Errorf("failed to reload manager config: %v", err)
}
return nil
}
func (jp *JobProcessor) test(job *Job) error {
req, mgrcfg := job.req, job.mgrcfg
Logf(0, "job: booting VM...")
inst, reporter, rep, err := bootInstance(mgrcfg)
if err != nil {
return err
}
if rep != nil {
// We should not put rep into resp.CrashTitle/CrashReport,
// because that will be treated as patch not fixing the bug.
return fmt.Errorf("%v\n\n%s\n\n%s", rep.Title, rep.Report, rep.Output)
}
defer inst.Close()
Logf(0, "job: testing instance...")
rep, err = testInstance(inst, reporter, mgrcfg)
if err != nil {
return err
}
if rep != nil {
// We should not put rep into resp.CrashTitle/CrashReport,
// because that will be treated as patch not fixing the bug.
return fmt.Errorf("%v\n\n%s\n\n%s", rep.Title, rep.Report, rep.Output)
}
Logf(0, "job: copying binaries...")
execprogBin, err := inst.Copy(mgrcfg.SyzExecprogBin)
if err != nil {
return fmt.Errorf("failed to copy test binary to VM: %v", err)
}
executorBin, err := inst.Copy(mgrcfg.SyzExecutorBin)
if err != nil {
return fmt.Errorf("failed to copy test binary to VM: %v", err)
}
progFile := filepath.Join(mgrcfg.Workdir, "repro.prog")
if err := osutil.WriteFile(progFile, req.ReproSyz); err != nil {
return fmt.Errorf("failed to write temp file: %v", err)
}
vmProgFile, err := inst.Copy(progFile)
if err != nil {
return fmt.Errorf("failed to copy to VM: %v", err)
}
Logf(0, "job: testing syzkaller program...")
opts, err := csource.DeserializeOptions(req.ReproOpts)
if err != nil {
return err
}
// Combine repro options and default options in a way that increases chances to reproduce the crash.
// First, we always enable threaded/collide as it should be [almost] strictly better.
// Executor does not support empty sandbox, so we use none instead.
// Finally, always use repeat and multiple procs.
if opts.Sandbox == "" {
opts.Sandbox = "none"
}
if !opts.Fault {
opts.FaultCall = -1
}
cmdSyz := fmt.Sprintf("%v -executor %v -arch=%v -procs=%v -sandbox=%v"+
" -fault_call=%v -fault_nth=%v -repeat=0 -cover=0 %v",
execprogBin, executorBin, mgrcfg.TargetArch, mgrcfg.Procs, opts.Sandbox,
opts.FaultCall, opts.FaultNth, vmProgFile)
crashed, err := jp.testProgram(job, inst, cmdSyz, reporter, 7*time.Minute)
if crashed || err != nil {
return err
}
if len(req.ReproC) != 0 {
Logf(0, "job: testing C program...")
cFile := filepath.Join(mgrcfg.Workdir, "repro.c")
if err := osutil.WriteFile(cFile, req.ReproC); err != nil {
return fmt.Errorf("failed to write temp file: %v", err)
}
target, err := prog.GetTarget(mgrcfg.TargetOS, mgrcfg.TargetArch)
if err != nil {
return err
}
bin, err := csource.Build(target, "c", cFile)
if err != nil {
return err
}
vmBin, err := inst.Copy(bin)
if err != nil {
return fmt.Errorf("failed to copy test binary to VM: %v", err)
}
// We should test for longer (e.g. 5 mins), but the problem is that
// reproducer does not print anything, so after 3 mins we detect "no output".
crashed, err := jp.testProgram(job, inst, vmBin, reporter, time.Minute)
if crashed || err != nil {
return err
}
}
return nil
}
func (jp *JobProcessor) testProgram(job *Job, inst *vm.Instance, command string,
reporter report.Reporter, testTime time.Duration) (bool, error) {
outc, errc, err := inst.Run(testTime, nil, command)
if err != nil {
return false, fmt.Errorf("failed to run binary in VM: %v", err)
}
rep := vm.MonitorExecution(outc, errc, reporter, true)
if rep == nil {
return false, nil
}
if err := reporter.Symbolize(rep); err != nil {
jp.Errorf("failed to symbolize report: %v", err)
}
job.resp.CrashTitle = rep.Title
job.resp.CrashReport = rep.Report
job.resp.CrashLog = rep.Output
return true, nil
}
// Errorf logs non-fatal error and sends it to dashboard.
func (jp *JobProcessor) Errorf(msg string, args ...interface{}) {
Logf(0, "job: "+msg, args...)
if jp.dash != nil {
jp.dash.LogError(jp.name, msg, args...)
}
}