Care about children of supervised processes

rts.tests/supervise-orphanage.exp

@@ -0,0 +1,7 @@
+--- supervise properly runs an orphanage
+the first run
+test.sv: up (pid x) x seconds, normally down, running
+Caught TERM
+test.sv: up (pid x) x seconds, normally down, orphanage
+the second run
+Caught HUP

rts.tests/supervise-orphanage.sh

@@ -0,0 +1,43 @@
+echo '--- supervise properly runs an orphanage'
+catexe test.sv/run <<EOF
+#!/bin/sh
+nohup ../../sleeper >output &
+mv run2 run
+echo the first run
+exec ../../sleeper
+EOF
+catexe test.sv/run2 <<EOF
+#!/bin/sh
+echo the second run
+svc -x .
+EOF
+touch test.sv/orphanage
+supervise test.sv &
+until svok test.sv || ! jobs %% >/dev/null 2>&1
+do
+  sleep 1
+done
+if svok test.sv
+then
+    svc -u test.sv
+    while [ -e test.sv/run2 ]
+    do
+      sleep 1
+    done
+    svstat test.sv | filter_svstat
+    svc -t test.sv
+    until svstat test.sv | grep -q orphanage
+    do
+      sleep 1
+    done
+    svstat test.sv | filter_svstat
+    cat test.sv/output
+    svc -+h test.sv
+    wait
+    cat test.sv/output
+else
+    echo "This test fails on older Unix systems"
+    echo "(everything which is not Linux or FreeBSD)"
+    echo "as they have no subprocess reapers"
+fi
+rm test.sv/orphanage

subreaper.c

@@ -0,0 +1,34 @@
+#include "strerr.h"
+
+#ifdef __linux__
+#include <sys/prctl.h>
+
+#ifdef PR_SET_CHILD_SUBREAPER
+#define HAS_SUBREAPER 1
+
+extern int set_subreaper(void)
+{
+    return prctl(PR_SET_CHILD_SUBREAPER,1,0,0,0);
+}
+
+#endif
+#endif
+
+#ifdef __FreeBSD__
+#include <sys/procctl.h>
+#ifdef PROC_REAP_ACQUIRE
+#define HAS_SUBREAPER 1
+
+extern int set_subreaper(void)
+{
+    return procctl(P_PID,0,PROC_REAP_ACQUIRE,NULL);
+}
+#endif
+#endif
+
+#ifndef HAS_SUBREAPER
+extern int set_subreaper(void)
+{
+    return -1;
+}
+#endif

subreaper.h

@@ -0,0 +1,8 @@
+/* Public domain. */
+
+#ifndef SUBREAPER_H
+#define SUBREAPER_H
+
+extern int set_subreaper(void);
+
+#endif

supervise.8.in

@@ -88,6 +88,14 @@ The fourth is the exit code or kill signal number, or
 .B 0
 on start.
 
+If the file
+.IB s /orphanage
+exists,
+.B supervise
+will wait until all ancestors of the supervised process have also exited,
+before it will consider the process stopped. In the meantime, the process
+is considered to be in an "orphanage" state.
+
 .B supervise
 maintains status information in a binary format inside the directory
 .IR s\fB/supervise ,

supervise.c

@@ -17,6 +17,7 @@
 #include "iopause.h"
 #include "taia.h"
 #include "deepsleep.h"
+#include "subreaper.h"
 #include "stralloc.h"
 #include "svpath.h"
 #include "svstatus.h"
@@ -44,6 +45,7 @@ int fdok;
 int fdstatus;
 
 int flagexit = 0;
+int flagorphanage = 0;
 int firstrun = 1;
 const char *runscript = 0;
 
@@ -295,11 +297,18 @@ void doit(void)
     for (;;) {
       r = wait_nohang(&wstat);
       if (!r) break;
-      if ((r == -1) && (errno != error_intr)) break;
-      if (r == svcmain.pid)
+      if (flagorphanage && r == svcmain.pid) {
+        svcmain.flagstatus = svstatus_orphanage;
+        announce();
+        continue;
+      }
+      if ((r == svcmain.pid) || ((svcmain.flagstatus == svstatus_orphanage)
+                                 && (r == -1) && (errno == ECHILD)))
 	svc = &svcmain;
       else if (r == svclog.pid)
 	svc = &svclog;
+      else if ((r == -1) && (errno != error_intr))
+        break;
       else
 	continue;
       killpid = svc->pid;
@@ -444,9 +453,16 @@ int main(int argc,char **argv)
   if (!svpath_init())
     strerr_die3sys(111,FATAL,"unable to setup control path for ",dir);
 
+  if (stat_exists("orphanage") != 0) {
+    flagorphanage = 1;
+    if (set_subreaper())
+      strerr_die2sys(111,FATAL,"could not set subreaper attribute");
+  }
   if (stat_isexec("log") > 0) {
     if (pipe(logpipe) != 0)
       strerr_die3sys(111,FATAL,"unable to create pipe for ",dir);
+    else if (flagorphanage)
+      strerr_die2sys(111,FATAL,"orphanage and log are mutually exclusive");
     svclog.flagwantup = 1;
   }
   if (stat("down",&st) != -1) {

supervise=x

@@ -1,5 +1,6 @@
 deepsleep.o
 svpath.o
+subreaper.o
 time.a
 unix.a
 byte.a

svstat.c

@@ -84,6 +84,7 @@ static void showstatus(const char status[19], int r, int normallyup)
     case svstatus_running: x = ", running"; break;
     case svstatus_stopping: x = ", stopping"; break;
     case svstatus_failed: x=", failed"; break;
+    case svstatus_orphanage: x=", orphanage"; break;
     default: x = ", status unknown";
     }
     if (pid && (want == '\000'))

svstatus.h

@@ -8,6 +8,7 @@ enum svstatus {
   svstatus_running,
   svstatus_stopping,
   svstatus_failed,
+  svstatus_orphanage,
 };
 
 #endif