utils.go

// Copyright 2019 Keybase, Inc. All rights reserved. Use of
// this source code is governed by the included BSD license.

package status

import (
	"archive/tar"
	"bufio"
	"bytes"
	"compress/gzip"
	"encoding/json"
	"fmt"
	"io"

	"mime/multipart"
	"os"
	"os/exec"
	"path/filepath"
	"runtime"
	"sort"
	"strconv"
	"strings"
	"time"

	humanize "github.com/dustin/go-humanize"
	"github.com/keybase/client/go/install"
	"github.com/keybase/client/go/libkb"
	"github.com/keybase/client/go/logger"
	"github.com/keybase/client/go/protocol/keybase1"
	jsonw "github.com/keybase/go-jsonw"
)

// MergeStatusJSON merges the given `obj` into the given `status` JSON blob.
// If any errors occur the original `status` is returned. Otherwise a new JSON
// blob is created of the form {"status": status, key: obj}
func MergeStatusJSON(obj interface{}, key, status string) string {
	if err := jsonw.EnsureMaxDepthBytesDefault([]byte(status)); err != nil {
		return status
	}

	var statusObj map[string]interface{}
	if err := json.Unmarshal([]byte(status), &statusObj); err != nil {
		return status
	}

	statusMap := make(map[string]interface{})
	statusMap["status"] = statusObj
	statusMap[key] = obj

	fullStatus, err := json.Marshal(statusMap)
	if err != nil {
		return status
	}
	return string(fullStatus)
}

// getServiceLog uses the EffectiveLogPath if available and falls back to the
// ServiceLogFileName otherwise.
func getServiceLog(mctx libkb.MetaContext, logDir string) string {
	serviceLogPath, ok := mctx.G().Env.GetEffectiveLogFile()
	if ok && serviceLogPath != "" {
		var err error
		serviceLogPath, err = filepath.Abs(serviceLogPath)
		if err != nil {
			mctx.Debug("Unable to get abspath for effective log file %v", err)
			serviceLogPath = ""
		}
	}
	if serviceLogPath == "" {
		return filepath.Join(logDir, libkb.ServiceLogFileName)
	}
	return serviceLogPath
}

func logFilesFromStatus(g *libkb.GlobalContext, fstatus *keybase1.FullStatus) Logs {
	logDir := g.Env.GetLogDir()
	installLogPath, err := install.InstallLogPath()
	if err != nil {
		g.Log.Errorf("Error (InstallLogPath): %s", err)
	}

	watchdogLogPath, err := install.WatchdogLogPath(filepath.Join(logDir, "watchdog*.log"))
	if err != nil {
		g.Log.Errorf("Error (WatchdogLogPath): %s", err)
	}

	traceDir := logDir
	cpuProfileDir := logDir
	if fstatus != nil {
		return Logs{
			GUI:        fstatus.Desktop.Log,
			Kbfs:       fstatus.Kbfs.Log,
			KbfsPerf:   fstatus.Kbfs.PerfLog,
			Service:    fstatus.Service.Log,
			EK:         fstatus.Service.EkLog,
			Perf:       fstatus.Service.PerfLog,
			Updater:    fstatus.Updater.Log,
			Start:      fstatus.Start.Log,
			System:     install.SystemLogPath(),
			Git:        fstatus.Git.Log,
			GitPerf:    fstatus.Git.PerfLog,
			Install:    installLogPath,
			Trace:      traceDir,
			CPUProfile: cpuProfileDir,
			Watchdog:   watchdogLogPath,
		}
	}

	return Logs{
		GUI:      filepath.Join(logDir, libkb.GUILogFileName),
		Kbfs:     filepath.Join(logDir, libkb.KBFSLogFileName),
		Service:  getServiceLog(libkb.NewMetaContextTODO(g), logDir),
		EK:       filepath.Join(logDir, libkb.EKLogFileName),
		Perf:     filepath.Join(logDir, libkb.PerfLogFileName),
		KbfsPerf: filepath.Join(logDir, libkb.KBFSPerfLogFileName),
		Updater:  filepath.Join(logDir, libkb.UpdaterLogFileName),
		Start:    filepath.Join(logDir, libkb.StartLogFileName),
		Git:      filepath.Join(logDir, libkb.GitLogFileName),
		GitPerf:  filepath.Join(logDir, libkb.GitPerfLogFileName),
		Install:  installLogPath,
		Trace:    traceDir,
		Watchdog: watchdogLogPath,
	}
}

func addFile(mpart *multipart.Writer, param, filename string, data []byte) error {
	if len(data) == 0 {
		return nil
	}

	part, err := mpart.CreateFormFile(param, filename)
	if err != nil {
		return err
	}
	_, err = io.Copy(part, bytes.NewBuffer(data))
	return err
}

func addGzippedFile(mpart *multipart.Writer, param, filename, data string) error {
	if len(data) == 0 {
		return nil
	}

	part, err := mpart.CreateFormFile(param, filename)
	if err != nil {
		return err
	}
	gz := gzip.NewWriter(part)
	if _, err := gz.Write([]byte(data)); err != nil {
		return err
	}
	return gz.Close()
}

// tail the logs that start with the stem `stem`, which are of type `which`.
// Get the most recent `numBytes` from the concatenation of the files.
func tail(log logger.Logger, which string, stem string, numBytes int) (ret string) {

	numFiles := 0

	log.Debug("+ tailing %s file with stem %q", which, stem)
	defer func() {
		log.Debug("- collected %d bytes from %d files", len(ret), numFiles)
	}()

	if len(stem) == 0 {
		log.Debug("| skipping %s logs (since no stem given)", which)
		return
	}

	lognames := listLogFiles(log, stem)
	var parts []string
	remaining := numBytes

	// Keep reading logs in reverse chronological order until we've read nothing
	// more, or we've filled up numBytes worth of buffer, or we didn't have to read
	// the whole file.
	for _, logname := range lognames {
		data, seeked := tailFile(log, which, logname, remaining)
		if len(data) == 0 {
			break
		}
		parts = append(parts, data)
		numFiles++
		remaining -= len(data)
		if remaining <= 0 || seeked {
			break
		}
	}

	// Reverse the array; took this one-line from StackOverflow answer
	for i, j := 0, len(parts)-1; i < j; i, j = i+1, j-1 {
		parts[i], parts[j] = parts[j], parts[i]
	}

	return strings.Join(parts, "")
}

type nameAndMTime struct {
	name  string
	mtime time.Time
}

type nameAndMTimes []nameAndMTime

func (a nameAndMTimes) Len() int           { return len(a) }
func (a nameAndMTimes) Swap(i, j int)      { a[i], a[j] = a[j], a[i] }
func (a nameAndMTimes) Less(i, j int) bool { return a[i].mtime.After(a[j].mtime) }

// List logfiles that match the glob filename*, and return then in reverse chronological order.
// We'll need to read the first, and maybe the second
func listLogFiles(log logger.Logger, stem string) (ret []string) {
	stem = filepath.Clean(stem)
	dir := filepath.Dir(stem)
	base := filepath.Base(stem)
	files, err := os.ReadDir(dir)

	defer func() {
		log.Debug("listLogFiles(%q) -> %v", stem, ret)
	}()

	// In the worst case, just return the stem in an array
	defret := []string{stem}

	if err != nil {
		log.Debug("failed to list directory %q: %s", dir, err)
		return defret
	}

	var tmp []nameAndMTime
	for _, d := range files {
		fullname := filepath.Clean(filepath.Join(dir, d.Name()))
		// Use the stat on the file (and not from the directory read)
		// since the latter doesn't work reliably on Windows
		finfo, err := os.Stat(fullname)
		if err != nil {
			log.Debug("Cannot stat %q: %s", fullname, err)
			continue
		}
		if strings.HasPrefix(d.Name(), base) {
			tmp = append(tmp, nameAndMTime{fullname, finfo.ModTime()})
		}
	}
	if len(tmp) == 0 {
		log.Debug("no files found matching \"%s*\"; falling back to default glob", stem)
		return defret
	}

	// Sort the files in reverse chronological mtime order. We should get the raw stem first.
	sort.Sort(nameAndMTimes(tmp))
	log.Debug("Sorted file list: %+v", tmp)

	for _, f := range tmp {
		ret = append(ret, f.name)
	}

	// If we didn't get the raw stem first, then we have a problem, so just use only the
	// raw stem and ignore the rest of our work.
	if stem != ret[0] {
		log.Debug("Expected to get %q first, but got %q instead! Falling back to one log only", stem, ret[0])
		return defret
	}
	return ret
}

// findFirstNewline first the first newline in the given byte array, and then returns the
// rest of the byte array. Should be safe to use on utf-8 strings.
func findFirstNewline(b []byte) []byte {
	index := bytes.IndexByte(b, '\n')
	if index < 0 || index == len(b)-1 {
		return nil
	}
	return b[(index + 1):]
}

func appendError(log logger.Logger, collected []byte, format string, args ...interface{}) []byte {
	msg := "Error reading logs: " + fmt.Sprintf(format, args...)
	log.Errorf(msg)
	return append(collected, []byte("\n"+msg+"\n")...)
}

// Get logs from the systemd journal. Currently we don't use this for most of
// our logging, since it's not persisted across boot on some systems. But we do
// use it for startup logs.
func tailSystemdJournal(log logger.Logger, userUnits []string, numBytes int) (ret string) {
	log.Debug("+ tailing journalctl for %#v (%d bytes)", userUnits, numBytes)
	defer func() {
		log.Debug("- scanned %d bytes", len(ret))
	}()

	// Journalctl doesn't provide a "last N bytes" flag directly, so instead we
	// use "last N lines". Large log files in practice seem to be about 150
	// bits per line. We'll request lines on that assumption, but if we get
	// more than 2x as many bytes as we expected, we'll stop reading and
	// include a big error.
	guessedLines := numBytes / 150
	maxBytes := numBytes * 2

	// We intentionally avoid the --user flag to journalctl. That would make us
	// skip over the system journal, but in e.g. Ubuntu 16.04, that's where
	// user units write their logs.
	// Unfortunately, this causes permission errors in some operating systems
	// like Debian Stretch, but it is not fatal. as we ignore errors in this
	// function.
	args := []string{
		"--lines=" + strconv.Itoa(guessedLines),
	}
	if len(userUnits) == 0 {
		panic("without --user-unit we would scrape all system logs!!!")
	}
	for _, userUnit := range userUnits {
		args = append(args, "--user-unit="+userUnit)
	}
	journalCmd := exec.Command("journalctl", args...)
	journalCmd.Stderr = os.Stderr
	stdout, err := journalCmd.StdoutPipe()
	if err != nil {
		msg := fmt.Sprintf("Failed to open a pipe for journalctl: %s", err)
		log.Errorf(msg)
		return msg
	}
	err = journalCmd.Start()
	if err != nil {
		msg := fmt.Sprintf("Failed to run journalctl: %s", err)
		log.Errorf(msg)
		return msg
	}

	// Once we start reading output, don't short-circuit on errors. Just log
	// them, and return whatever we got.
	stdoutLimited := io.LimitReader(stdout, int64(maxBytes))
	output, err := io.ReadAll(stdoutLimited)
	if err != nil {
		output = appendError(log, output, "Error reading from journalctl pipe: %s", err)
	}
	// We must close stdout before Wait, or else Wait might deadlock.
	stdout.Close()
	err = journalCmd.Wait()
	if err != nil {
		output = appendError(log, output, "Journalctl exited with an error: %s", err)
	}
	if len(output) >= maxBytes {
		output = appendError(log, output, "Journal lines longer than expected. Logs truncated.")
	}
	return string(output)
}

// tailFile takes the last n bytes, but advances to the first newline. Returns the log (as a string)
// and a bool, indicating if we read the full log, or we had to advance into the log to find the newline.
func tailFile(log logger.Logger, which string, filename string, numBytes int) (ret string, seeked bool) {

	log.Debug("+ tailing %s log %q (%d bytes)", which, filename, numBytes)
	defer func() {
		log.Debug("- scanned %d bytes", len(ret))
	}()

	seeked = false
	if filename == "" {
		return ret, seeked
	}
	finfo, err := os.Stat(filename)
	if os.IsNotExist(err) {
		log.Debug("log %q doesn't exist", filename)
		return ret, seeked
	}
	f, err := os.Open(filename)
	if err != nil {
		log.Errorf("error opening log %q: %s", filename, err)
		return ret, seeked
	}
	defer f.Close()
	if finfo.Size() > int64(numBytes) {
		seeked = true
		_, err = f.Seek(int64(-numBytes), io.SeekEnd)
		if err != nil {
			log.Errorf("Can't seek log %q: %s", filename, err)
			return ret, seeked
		}
	}
	buf, err := io.ReadAll(f)
	if err != nil {
		log.Errorf("Failure in reading file %q: %s", filename, err)
		return ret, seeked
	}
	if seeked {
		buf = findFirstNewline(buf)
	}
	return string(buf), seeked
}

func addFileToTar(tw *tar.Writer, path string) error {
	file, err := os.Open(path)
	if err != nil {
		return err
	}
	defer file.Close()

	if stat, err := file.Stat(); err == nil {
		header := tar.Header{
			Typeflag: tar.TypeReg,
			Name:     filepath.Base(path),
			Size:     stat.Size(),
			Mode:     int64(0600),
			ModTime:  stat.ModTime(),
		}
		if err := tw.WriteHeader(&header); err != nil {
			return err
		}
		if _, err := io.Copy(tw, file); err != nil {
			return err
		}
	}
	return nil
}

func addFilesToTarGz(log logger.Logger, w io.Writer, paths []string) bool {
	gw := gzip.NewWriter(w)
	defer gw.Close()
	tw := tar.NewWriter(gw)
	defer tw.Close()

	added := false
	for _, path := range paths {
		err := addFileToTar(tw, path)
		if err != nil {
			log.Warning("Error adding %q to tar file: %s", path, err)
			continue
		}
		log.Debug("Added trace file %q", path)
		added = true
	}
	return added
}

func getBundledFiles(log logger.Logger, files []string, maxFileCount int) []byte {
	// Send the newest files.
	if len(files) > maxFileCount {
		files = files[len(files)-maxFileCount:]
	}

	buf := bytes.NewBuffer(nil)
	added := addFilesToTarGz(log, buf, files)
	if !added {
		return nil
	}
	return buf.Bytes()
}

func getTraceBundle(log logger.Logger, traceDir string) []byte {
	traceFiles, err := libkb.GetSortedTraceFiles(traceDir)
	if err != nil {
		log.Warning("Error getting trace files in %q: %s", traceDir, err)
		return nil
	}

	return getBundledFiles(log, traceFiles, libkb.MaxTraceFileCount)
}

func getCPUProfileBundle(log logger.Logger, cpuProfileDir string) []byte {
	cpuProfileFiles, err := libkb.GetSortedCPUProfileFiles(cpuProfileDir)
	if err != nil {
		log.Warning("Error getting CPU profile files in %q: %s", cpuProfileDir, err)
		return nil
	}

	return getBundledFiles(log, cpuProfileFiles, libkb.MaxCPUProfileFileCount)
}

func getPlatformInfo() keybase1.PlatformInfo {
	return keybase1.PlatformInfo{
		Os:        runtime.GOOS,
		Arch:      runtime.GOARCH,
		GoVersion: runtime.Version(),
	}
}

// DirSize walks the file tree the size of the given directory
func DirSize(dirPath string) (size uint64, numFiles int, err error) {
	err = filepath.Walk(dirPath, func(_ string, info os.FileInfo, err error) error {
		if err != nil {
			return err
		}
		if !info.IsDir() {
			size += uint64(info.Size())
			numFiles++
		}
		return nil
	})
	return size, numFiles, err
}

func CacheSizeInfo(g *libkb.GlobalContext) (info []keybase1.DirSizeInfo, err error) {
	cacheDir := g.GetCacheDir()
	files, err := os.ReadDir(cacheDir)
	if err != nil {
		return nil, err
	}

	var totalSize uint64
	var totalFiles int
	for _, file := range files {
		if !file.IsDir() {
			info, err := file.Info()
			if err != nil {
				return nil, err
			}
			totalSize += uint64(info.Size())
			continue
		}
		dirPath := filepath.Join(cacheDir, file.Name())
		size, numFiles, err := DirSize(dirPath)
		if err != nil {
			return nil, err
		}
		totalSize += size
		totalFiles += numFiles
		info = append(info, keybase1.DirSizeInfo{
			Name:      dirPath,
			NumFiles:  numFiles,
			HumanSize: humanize.Bytes(size),
		})
	}
	info = append(info, keybase1.DirSizeInfo{
		Name:      cacheDir,
		NumFiles:  totalFiles,
		HumanSize: humanize.Bytes(totalSize),
	})
	return info, nil
}

func GetFirstClient(v []keybase1.ClientStatus, typ keybase1.ClientType) *keybase1.ClientDetails {
	for _, cli := range v {
		if cli.Details.ClientType == typ {
			return &cli.Details
		}
	}
	return nil
}

// zipLogs takes a slice of logs and returns `numBytes` worth of the
// most recent log lines from the sorted set of all lines across all
// logs.  The lines in each log should already be sorted, and the logs
// should be sortable alphabetically between each other (e.g., each
// line in every log should start with a timestamp in the same
// format).
func zipLogs(numBytes int, logs ...string) (res string) {
	scanners := make([]*bufio.Scanner, len(logs))
	for i, log := range logs {
		scanners[i] = bufio.NewScanner(strings.NewReader(log))
	}

	// nextLines is the next chronological line in each log.  It's
	// empty if we need to get another line from the log.
	nextLines := make([]string, len(logs))
	var buf bytes.Buffer
	for {
		// Fill in the next line.
		minLine := 0
		for i, currLine := range nextLines {
			if currLine == "" {
				for scanners[i].Scan() {
					nextLines[i] = scanners[i].Text()
					if nextLines[i] != "" {
						break
					}
				}
			}

			// If we still don't have a line, just skip this log.
			if nextLines[i] == "" {
				continue
			}

			if minLine != i && nextLines[i] < nextLines[minLine] {
				minLine = i
			}
		}

		line := nextLines[minLine]
		if line == "" {
			// We're done!
			break
		}

		buf.WriteString(line + "\n")
		nextLines[minLine] = ""
	}

	if buf.Len() > numBytes {
		b := buf.Bytes()[buf.Len()-numBytes:]
		return string(findFirstNewline(b))
	}

	return buf.String()
}