RunEvery() and better Error-handling added

1. symbolyze.Scanner now implements the RunEvery(time.Duration) method
   that will call scanner.Run() periodically.  It should be called to
   run in its own goroutine.  The loop can be stopped by calling
   scanner.Stop().

2. The scanner now collects all errors from all observers in a private
   error type `errors []error`. It's Error() returns a cumulated list of
   errors, seperated by a newline.
This commit is contained in:
Özgür Kesim 2020-01-18 20:27:23 +01:00
parent baf998232a
commit d800683dce
2 changed files with 125 additions and 39 deletions

View File

@ -4,6 +4,7 @@ import (
"flag"
"fmt"
"os"
"time"
"github.com/optimyze-interviews/OezguerKesim/GetRuntimeAddresses/ebpf"
"github.com/optimyze-interviews/OezguerKesim/GetRuntimeAddresses/symbolyze"
@ -28,9 +29,12 @@ func main() {
scanner := symbolyze.NewScanner(*symbol, *glob)
scanner.OnFound(mapFD.Set)
if *debug {
scanner.DebugOn()
}
scanner.Debug(*debug)
go scanner.RunEvery(time.Second)
time.Sleep(10 * time.Second)
scanner.Stop()
err = scanner.Run()
if err != nil {

View File

@ -10,6 +10,7 @@ import (
"strconv"
"strings"
"sync"
"time"
)
// Scanner represents an engine for scanning for a specific symbol in all
@ -24,25 +25,51 @@ import (
// /proc. Whenever a match is found, all observers will be called with the
// (pid, offset), concurrently.
type Scanner struct {
rwmutex
symbol string
pathglob string
cache map[string]uint64 // Contains (pathname, offset)
observers []Observer // Callbacks
logger // Embedded logger
// Instead of using a boolean to indicate debugging, we use function
// members. This way we can populate them with noop-functions in the
// non-debug case and not polute the code with if-statements.
debugf func(format string, v ...interface{})
debugln func(v ...interface{})
logger // Embedded logger
err error // error state of the scanner.
errors errors
ticker *time.Ticker // Used to run the scanner repeatedly
}
type errors []error
func (e errors) Error() string {
switch len(e) {
case 0:
return "nil"
case 1:
return e[0].Error()
default:
list := make([]string, len(e))
for i := range e {
list[i] = e[i].Error()
}
return fmt.Sprintf("multiple errors:\n%s", strings.Join(list, "\n"))
}
}
// We use a lowercase type alias for *log.Logger so that we can embedd it in
// Scanner without exporting it.
type logger = *log.Logger
type rwmutex = sync.RWMutex
var (
nodebugf = func(format string, v ...interface{}) {}
nodebugln = func(v ...interface{}) {}
)
// An Observer is a callback that can be registered with Scanner.OnFound. It
// will be called with a pid and an offset. Observers are called concurrently.
@ -62,41 +89,76 @@ func NewScanner(symbol, pathglob string) *Scanner {
logger: log.New(os.Stderr, "[symbolyze] ", log.Ltime|log.Lmicroseconds),
// debugging is off per default.
debugf: func(string, ...interface{}) {},
debugln: func(...interface{}) {},
debugf: nodebugf,
debugln: nodebugln,
}
}
// Debug sets the scanner into debugging mode. It must be called only once
// before a call to Scanner.Run().
func (S *Scanner) DebugOn() {
func (S *Scanner) Debug(on bool) {
S.Lock()
defer S.Unlock()
if on {
// Use the embedded *log.Logger for debugging.
S.debugf = S.Printf
S.debugln = S.Println
S.debugln("starting in debug-mode")
} else {
S.debugf = nodebugf
S.debugln = nodebugln
}
}
// setErrorf puts the Scanner into an error state with the given error
// statement. It also logs the error. setErrorf is not thread-safe.
func (S *Scanner) setErrorf(format string, a ...interface{}) {
S.err = fmt.Errorf(format, a...)
S.Lock()
S.errors = append(S.errors, fmt.Errorf(format, a...))
S.Unlock()
S.Printf(format, a...)
}
// OnFound puts an Observer function into the internal queue. The functions
// are called in sequence in their own goroutine whenever the scanner finds the
// symbol in a running program. That implies that an Observer has to be
// thread-safe. Errors from the observers will be logged.
//
// Calling OnFound is not thread-safe.
func (S *Scanner) OnFound(fun Observer) {
S.observers = append(S.observers, fun)
func (S *Scanner) HasErrors() bool {
S.RLock()
defer S.RUnlock()
return len(S.errors) > 0
}
func (S *Scanner) Errors() error {
S.RLock()
defer S.RUnlock()
if len(S.errors) == 0 {
return nil
} else {
e2 := make(errors, len(S.errors))
copy(e2, S.errors)
return e2
}
}
func (S *Scanner) addError(err error) {
S.Lock()
defer S.Unlock()
S.errors = append(S.errors, err)
}
// OnFound puts Observer functions at the end of the internal queue. All
// Observer functions are called in sequence in their own goroutine whenever
// the scanner finds the symbol in a running program. That implies that an
// Observer has to be thread-safe. Errors from the observers will be logged.
func (S *Scanner) OnFound(fun ...Observer) {
S.Lock()
defer S.Unlock()
S.observers = append(S.observers, fun...)
return
}
// Run starts the scanning process. It scans the entries of all /proc/NNN/maps
// files for pathnames that match the provided path-glob and are executables or
// shared libraries in ELF formmat. It searches for the provided symbol in
// shared libraries in ELF format. It searches for the provided symbol in
// those files and calls the registered Observer functions, concurrently, with
// the pid and offset of the symbol.
//
@ -104,20 +166,20 @@ func (S *Scanner) OnFound(fun Observer) {
// it will try to continue to loop over all pids, writing potential errors to
// the console. Errors from the observer functions are logged.
func (S *Scanner) Run() error {
if S.err != nil {
return S.err
if S.HasErrors() {
return S.Errors()
}
proc, err := os.Open("/proc")
if err != nil {
S.setErrorf("Failed to open /proc: %v\n", err)
return S.err
return S.Errors()
}
infos, err := proc.Readdir(-1)
if err != nil {
S.setErrorf("Failed to read /proc: %v\n", err)
return S.err
return S.Errors()
}
proc.Close()
@ -132,7 +194,7 @@ func (S *Scanner) Run() error {
continue
} else if pid, err := strconv.Atoi(pid_s); err != nil {
continue
} else if offset, found := S.searchSymbolIn(pid); !found {
} else if offset, found := S.searchSymbolInPid(pid); !found {
continue
} else {
// Call the observers with (pid, offset), in the
@ -143,8 +205,7 @@ func (S *Scanner) Run() error {
go func() {
err = observer(pid, offset)
if err != nil {
S.Printf("S.observer[%d](%d, %d) error: %v", n, pid, offset, err)
// TODO: accumulate errors from all Observers.
S.addError(fmt.Errorf("S.observer[%d](%d, %d) error: %v", n, pid, offset, err))
}
wg.Done()
}()
@ -154,10 +215,31 @@ func (S *Scanner) Run() error {
}
wg.Wait() // Wait for all observers to finish
return S.err
return S.Errors()
}
// searchSymbolIn loops over the entries in /proc/<pid>/maps and searches for
// RunEvery() starts a scanning process and repeats at the given time step.
func (S *Scanner) RunEvery(step time.Duration) {
S.ticker = time.NewTicker(step)
for {
select {
case <-S.ticker.C:
err := S.Run()
if err != nil {
S.Println(err)
}
}
}
}
func (S *Scanner) Stop() {
if S.ticker != nil {
S.debugln("Stopping ticker")
S.ticker.Stop()
}
}
// searchSymbolInPid loops over the entries in /proc/<pid>/maps and searches for
// the symbol in the mapped files.
//
// The current implementation makes the following assumptions:
@ -167,12 +249,12 @@ func (S *Scanner) Run() error {
// 4. The symbol is present at most in one mapped file at the same time.
//
// It returns the offsets in memory of the running program, if found.
func (S *Scanner) searchSymbolIn(pid int) (offset uint64, found bool) {
func (S *Scanner) searchSymbolInPid(pid int) (offset uint64, found bool) {
path := filepath.Join("/proc", strconv.Itoa(pid), "maps")
maps, err := os.Open(path)
if err != nil {
S.Printf("%v\n", err)
S.debugf("%v\n", err)
return 0, false
}
@ -230,7 +312,7 @@ func (S *Scanner) searchSymbolIn(pid int) (offset uint64, found bool) {
// Finally, find the symbol in the binary. If found,
// findSymbol returns the offset of the symbol in memory,
// taking alignment into account.
memOffset, found := S.findSymbol(pathname)
memOffset, found := S.findSymbolInELF(pathname)
if !found {
continue
}
@ -242,15 +324,15 @@ func (S *Scanner) searchSymbolIn(pid int) (offset uint64, found bool) {
return 0, false
}
// findSymbol searches for the provided symbol in the given pathname to an
// findSymbolInELF searches for the provided symbol in the given pathname to an
// ELF-file. If found, it returns the offset of the symbol in the virtual
// memory according to the fomula:
//
// vmOffset = alignedOffset(section) + offsetInSection(symbol)
//
// The result will be cached so that subsequent calls to findSymbol with the
// same pathname can quickly return.
func (S *Scanner) findSymbol(pathname string) (offset uint64, found bool) {
// The result will be cached so that subsequent calls to findSymbolInELF with
// the same pathname can quickly return.
func (S *Scanner) findSymbolInELF(pathname string) (offset uint64, found bool) {
// 0. Return the value from the cache, if found.
if offset, found = S.cache[pathname]; found {