x-pack/auditbeat/module/system/socket: get full length path and arg from /proc when not available from kprobe (#29410) (#29958)

Also use first arg from sysinfo.Processes in place of Name to avoid process name
truncation.

(cherry picked from commit d46bb5f)

Co-authored-by: Dan Kortschak <[email protected]>

Author: mergify[bot]

CHANGELOG.next.asciidoc

@@ -44,6 +44,11 @@ https://github.com/elastic/beats/compare/v7.0.0-alpha2...master[Check the HEAD d
 *Auditbeat*
 
 - system/socket: Fix startup errors on newer 5.x kernels due to missing _do_fork function. {issue}29607[29607] {pull}29744[29744]
+- system/package: Fix parsing of Installed-Size field of DEB packages. {issue}16661[16661] {pull}17188[17188]
+- system module: Fix panic during initialisation when /proc/stat can't be read. {pull}17569[17569]
+- system/package: Fix an error that can occur while trying to persist package metadata. {issue}18536[18536] {pull}18887[18887]
+- system/socket: Fix bugs leading to wrong process being attributed to flows. {pull}29166[29166] {issue}17165[17165]
+- system/socket: Fix process name and arg truncation for long names, paths and args lists. {issue}24667[24667] {pull}29410[29410]
 
 *Filebeat*
 

x-pack/auditbeat/module/system/socket/events.go

@@ -859,8 +859,10 @@ func (e *inetReleaseCall) Update(s *state) error {
 // kernels it needs to dump fixed-size arrays in 8-byte chunks. As the total
 // number of fetchargs available is limited, we have to dump only the first
 // 128 bytes of every argument.
-const maxProgArgLen = 128
-const maxProgArgs = 5
+const (
+	maxProgArgLen = 128
+	maxProgArgs   = 5
+)
 
 type execveCall struct {
 	Meta tracing.Metadata    `kprobe:"metadata"`
@@ -880,38 +882,78 @@ type execveCall struct {
 func (e *execveCall) getProcess() *process {
 	p := &process{
 		pid:     e.Meta.PID,
-		path:    readCString(e.Path[:]),
 		created: kernelTime(e.Meta.Timestamp),
 	}
-	p.name = filepath.Base(p.path)
-	var argc int
-	for argc = 0; argc <= maxProgArgs; argc++ {
-		if e.Ptrs[argc] == 0 {
-			break
+
+	if idx := bytes.IndexByte(e.Path[:], 0); idx >= 0 {
+		// Fast path if we already have the path.
+		p.path = string(e.Path[:idx])
+	} else {
+		// Attempt to get the path from the /prox/<pid>/exe symlink.
+		var err error
+		p.path, err = filepath.EvalSymlinks(fmt.Sprintf("/proc/%d/exe", e.Meta.PID))
+		if err != nil {
+			if pe, ok := err.(*os.PathError); ok && strings.Contains(pe.Path, "(deleted)") {
+				// Keep the deleted path from the PathError.
+				p.path = pe.Path
+			} else {
+				// Fallback to the truncated path.
+				p.path = string(e.Path[:]) + " ..."
+			}
 		}
 	}
-	p.args = make([]string, argc)
-	params := [maxProgArgs][]byte{
+
+	// Check for truncation of arg list or arguments.
+	params := [...][]byte{
 		e.Param0[:],
 		e.Param1[:],
 		e.Param2[:],
 		e.Param3[:],
 		e.Param4[:],
 	}
-	limit := argc
-	if limit > maxProgArgs {
-		limit = maxProgArgs
-		p.args[limit] = "..."
+	var (
+		argc         int
+		truncatedArg bool
+	)
+	for argc = 0; argc < len(e.Ptrs); argc++ {
+		if e.Ptrs[argc] == 0 {
+			break
+		}
+		if argc < len(params) && bytes.IndexByte(params[argc], 0) < 0 {
+			truncatedArg = true
+		}
+	}
+	if argc > maxProgArgs || truncatedArg {
+		// Attempt to get complete args list from /proc/<pid>/cmdline.
+		cmdline, err := os.ReadFile(fmt.Sprintf("/proc/%d/cmdline", e.Meta.PID))
+		if err == nil {
+			p.args = strings.Split(strings.TrimRight(string(cmdline), "\x00"), "\x00")
+		}
 	}
-	for i := 0; i < limit; i++ {
-		p.args[i] = readCString(params[i])
+
+	if p.args == nil {
+		// Fallback to arg list if unsuccessful or no truncation.
+		p.args = make([]string, argc)
+		if argc > maxProgArgs {
+			argc = maxProgArgs
+			p.args[argc] = "..."
+		}
+		for i, par := range params[:argc] {
+			p.args[i] = readCString(par)
+		}
 	}
-	if p.hasCreds = e.creds != nil; p.hasCreds {
+
+	// Get name from first argument.
+	p.name = filepath.Base(p.args[0])
+
+	if e.creds != nil {
+		p.hasCreds = true
 		p.uid = e.creds.UID
 		p.gid = e.creds.GID
 		p.euid = e.creds.EUID
 		p.egid = e.creds.EGID
 	}
+
 	return p
 }
 

x-pack/auditbeat/module/system/socket/socket_linux.go

@@ -187,6 +187,15 @@ func (m *MetricSet) Run(r mb.PushReporterV2) {
 	} else {
 		for _, p := range procs {
 			if i, err := p.Info(); err == nil {
+				if len(i.Name) == 16 && len(i.Args) != 0 {
+					// github.com/prometheus/procfs uses /proc/<pid>/stat for
+					// the process name which is truncated to 16 bytes, so get
+					// the name from the cmdline data if it might be truncated.
+					// The guard for length of i.Args is for cases where there
+					// is no command line reported by proc fs; this should never
+					// happen, but does.
+					i.Name = filepath.Base(i.Args[0])
+				}
 				process := &process{
 					name:        i.Name,
 					pid:         uint32(i.PID),