Implement configurable actions depending on the exit code of

a component.

* pies/pies.c: New configuration file section "return-code".
(capa): Add mailer and debug.
(main): Register "return-code" section and all mailer formats.
* pies/progman.c (enum prog_status): New type.
(struct prog): Remove `disabled'. New fields `status' and `act'.
(register_prog): Initialize act.
(prog_start): Exit with EX_SOFTWARE code if the component cannot
be started.
(progman_start): Do not start disabled components.
(progman_wake_disabled): Wake only components marked as
status_stopped.
(notify): New function.
(progman_cleanup): Act in accordance with the `act' field if the
component terminates.
(progman_stop_component,progman_dump_stats): Take into account
prog->v.p.status.
* pies/pies.h (MAX_RETURN_CODE): New define.
(enum return_action): New type.
(struct component.act): New member.
(default_component): New global.


git-svn-id: file:///svnroot/mailfromd/branches/gmach@1600 7a8a7f39-df28-0410-adc6-e0d955640f24

4 files changed, 356 insertions, 12 deletions

diff --git a/ChangeLog b/ChangeLog
index a93ef6bb..b1eb2200 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,29 @@
+2008-01-31  Sergey Poznyakoff  <gray@gnu.org.ua>
+
+	Implement configurable actions depending on the exit code of
+	a component.
+	
+	* pies/pies.c: New configuration file section "return-code".
+	(capa): Add mailer and debug.
+	(main): Register "return-code" section and all mailer formats.
+	* pies/progman.c (enum prog_status): New type.
+	(struct prog): Remove `disabled'. New fields `status' and `act'.
+	(register_prog): Initialize act.
+	(prog_start): Exit with EX_SOFTWARE code if the component cannot
+	be started.
+	(progman_start): Do not start disabled components.
+	(progman_wake_disabled): Wake only components marked as
+	status_stopped.
+	(notify): New function.
+	(progman_cleanup): Act in accordance with the `act' field if the
+	component terminates.
+	(progman_stop_component,progman_dump_stats): Take into account
+	prog->v.p.status.
+	* pies/pies.h (MAX_RETURN_CODE): New define.
+	(enum return_action): New type.
+	(struct component.act): New member.
+	(default_component): New global.
+
 2008-01-28  Sergey Poznyakoff  <gray@gnu.org.ua>
 
 	* mfd/bi_db.m4 (dbdel): Fix format arguments.
diff --git a/pies/pies.c b/pies/pies.c
index f922fc18..4b5fa762 100644
--- a/pies/pies.c
+++ b/pies/pies.c
@@ -63,6 +63,154 @@ stderr_closed_p()
 
 
 static int
+_cb_action (mu_debug_t debug, void *data, char *arg)
+{
+  enum return_action *pact = data;
+  static struct mu_kwd actab[] = {
+    { "disable", action_disable },
+    { "restart", action_restart },
+    { NULL }
+  };
+  int res;
+  
+  if (mu_kwd_xlat_name (actab, arg, &res))
+    {
+      mu_cfg_format_error (debug, MU_DEBUG_ERROR,
+			   _("unknown action code: %s"), arg);
+      return 0;
+    }
+  *pact = res;
+  return 0;
+}
+
+static int
+_cb_notify_addr (mu_debug_t debug, void *data, char *arg)
+{
+  mu_address_t *paddr = data;
+  mu_address_t addr;
+  int rc;
+
+  rc = mu_address_create (&addr, arg);
+  if (rc)
+    mu_cfg_format_error (debug, MU_DEBUG_ERROR,
+			 _("%s: invalid e-mail address: %s"),
+			 arg, mu_strerror (rc));
+  if (*paddr)
+    {
+      mu_address_union (paddr, addr);
+      mu_address_destroy (&addr);
+    }
+  else
+    *paddr = addr;
+  return 0;
+}
+
+struct mu_cfg_param return_code_cfg_param[] = {
+  { "action", mu_cfg_callback, NULL,
+    mu_offsetof (struct action, act), _cb_action,
+    N_("Specifies action to take when a component finishes with this "
+       "return code."),
+    N_("arg: {disable | restart}") },
+  { "notify", mu_cfg_callback, NULL,
+    mu_offsetof (struct action, addr), _cb_notify_addr,
+    N_("Notify this address when then component terminates."),
+    N_("arg: email-list") },
+  { "message", mu_cfg_string, NULL,
+    mu_offsetof (struct action, message), NULL,
+    N_("Notification message text (with headers).") },
+  { NULL }
+};
+  
+static int
+return_code_section_parser (enum mu_cfg_section_stage stage,
+			    const mu_cfg_node_t *node,
+			    const char *section_label, void **section_data,
+			    void *call_data,
+			    mu_cfg_tree_t *tree)
+{
+  struct component *comp = *section_data;
+  struct action *act;
+  int argc, i;
+  char **argv;
+  int *retv;
+  int retc = 0;
+  int allflag = 0;
+  int rc;
+  
+  switch (stage)
+    {
+    case mu_cfg_section_start:
+      rc = mu_argcv_get (node->tag_label, NULL, NULL, &argc, &argv);
+      if (rc)
+	{
+	  mu_cfg_format_error (tree->debug, MU_DEBUG_ERROR,
+			       _("cannot parse return codes: %s"),
+			       mu_strerror (rc));
+	  return 1;
+	}
+      retv = xcalloc (argc, sizeof *retv);
+      if (argc == 0 || (argc == 1 && strcmp (argv[0], "*") == 0))
+	allflag = 1;
+      else
+	{
+	  for (i = 0; i < argc; i++)
+	    {
+	      char *p;
+	      int n = strtoul (argv[i], &p, 0);
+	      if (*p)
+		mu_cfg_format_error (tree->debug, MU_DEBUG_ERROR,
+				     _("%s: not a number"), p);
+	      else if (n > MAX_RETURN_CODE)
+		mu_cfg_format_error (tree->debug, MU_DEBUG_ERROR,
+				     _("%s: invalid return code"), p);
+	      else
+		{
+		  /* Alles in ordnung */
+		  retv[retc++] = n;
+	      }
+	    }
+	}
+      
+      mu_argcv_free (argc, argv);
+      if (retc == 0 && !allflag)
+	{
+	  free (retv);
+	  return 1;
+	}
+
+      act = xzalloc (sizeof *act);
+      if (allflag)
+	{
+	  for (i = 0; i <= MAX_RETURN_CODE; i++)
+	    if (comp->act[i] == NULL)
+	      comp->act[i] = act;
+	}
+      else
+	for (i = 0; i < retc; i++)
+	  comp->act[retv[i]] = act;
+      *section_data = act;
+      break;
+
+    case mu_cfg_section_end:
+      break;
+    }
+  return 0;
+}
+
+void
+return_code_cfg_init ()
+{
+  struct mu_cfg_section *section;
+
+  if (mu_create_canned_section ("return-code", &section))
+    exit (EX_SOFTWARE);
+  section->parser = return_code_section_parser;
+  section->label = N_("<tag: exit-code-list>");
+  mu_cfg_section_add_params (section, return_code_cfg_param);
+}
+
+
+static int
 _cb_group (mu_debug_t debug, void *data, char *arg)
 {
   int argc, i;
@@ -224,6 +372,7 @@ struct mu_cfg_param component_cfg_param[] = {
     N_("Set program environment.  Argument is a list of assignments "
        "separated by white space."),
     N_("arg: list") },
+  { "return-code", mu_cfg_section },
   { NULL }
 };
 
@@ -264,6 +413,7 @@ component_cfg_init ()
 
     
 
+struct component default_component;
 
 static int
 _cb_debug (mu_debug_t debug, void *data, char *arg)
@@ -299,6 +449,7 @@ struct mu_cfg_param pies_cfg_param[] = {
   { "shutdown-timeout", mu_cfg_uint, &shutdown_timeout, 0, NULL,
     N_("Wait <n> seconds for all children to shut down."),
     "n" },
+  { "return-code", mu_cfg_section, &default_component },
   { NULL }
 };
 
@@ -311,6 +462,8 @@ static char args_doc[] = "";
 static const char *capa[] = {
   "common",
   "logging",
+  "mailer",
+  "debug",
   NULL
 };
 
@@ -694,8 +847,10 @@ main (int argc, char **argv)
   argp_program_version_hook = version;
   /* Set default logging */
   log_setup (!stderr_closed_p ());
+  return_code_cfg_init ();
   component_cfg_init ();
   mu_argp_init (program_version, package_bugreport);
+  mu_register_all_mailer_formats ();
   rc = mu_app_init (&argp, capa, pies_cfg_param, argc, argv, 0, &index, NULL);
   if (rc)
     exit (EX_CONFIG);
diff --git a/pies/pies.h b/pies/pies.h
index 66badb6c..2f27b802 100644
--- a/pies/pies.h
+++ b/pies/pies.h
@@ -54,6 +54,21 @@ struct pies_privs_data
   mu_list_t groups;
 };
 
+#define MAX_RETURN_CODE 127
+
+enum return_action
+{
+  action_restart,
+  action_disable,
+};
+
+struct action
+{
+  enum return_action act;  /* Action to take when the component terminates */
+  mu_address_t addr;       /* Addresses to notify about it. */
+  char *message;           /* Notification mail. */
+};
+
 struct component
 {
   char *tag;             /* Entry tag (for diagnostics purposes) */
@@ -64,7 +79,8 @@ struct component
   char *rmfile;
   int retr[2];
   struct pies_privs_data privs;
-  mode_t umask;	
+  mode_t umask;
+  struct action *act[MAX_RETURN_CODE+1];
 };
 
 void register_prog (struct component *comp);
@@ -82,4 +98,4 @@ void priv_setup (struct pies_privs_data *pr);
 
 extern char *syslog_tag;
 extern unsigned long shutdown_timeout;
-
+extern struct component default_component;
diff --git a/pies/progman.c b/pies/progman.c
index bfc9512e..7d029392 100644
--- a/pies/progman.c
+++ b/pies/progman.c
@@ -15,10 +15,19 @@
    with this program. If not, see <http://www.gnu.org/licenses/>. */
 
 #include "pies.h"
+/* FIXME */
+const char *mu_umaxtostr (unsigned slot, uintmax_t val);
 
 #define TYPE_COMPONENT 0
 #define TYPE_RETR      1
 
+enum prog_status
+  {
+    status_enabled,
+    status_stopped,
+    status_disabled
+  };
+
 struct prog
 {
   struct prog *next;
@@ -39,7 +48,8 @@ struct prog
       mode_t umask;
       time_t timestamp;      /* Time of last startup */
       size_t count;          /* Number of failed starts since timestamp */
-      int disabled;          /* 1 if this entry is disabled */
+      enum prog_status status; /* Current component status */
+      struct action *act[MAX_RETURN_CODE+1];
     } p;
     
     struct
@@ -159,6 +169,10 @@ register_prog (struct component *comp)
   newp->v.p.argc = i;
   newp->v.p.argv = comp->argv;
   newp->v.p.env = comp->env;
+  if (comp->disabled)
+    newp->v.p.status = status_disabled;
+  for (i = 0; i <= MAX_RETURN_CODE; i++)
+    newp->v.p.act[i] = comp->act[i] ? comp->act[i] : default_component.act[i];
   
   pstr = (char*) (newp + 1);
   
@@ -384,7 +398,7 @@ prog_start (struct prog *prog)
       mu_error (_("%s is respawning too fast, disabled for %d minutes"),
 		prog->tag, SLEEPTIME / 60);
       prog->v.p.timestamp = now;
-      prog->v.p.disabled = 1;
+      prog->v.p.status = status_disabled;
       
       old_alarm = alarm (0);
       if (old_alarm > SLEEPTIME || old_alarm <= 0)
@@ -445,7 +459,7 @@ prog_start (struct prog *prog)
        openlog (MU_LOG_TAG(), LOG_PID, mu_log_facility);
        syslog (LOG_CRIT, _("cannot start `%s': %s"), prog->tag,
 	       mu_strerror (errno));
-       exit (1);
+       exit (EX_SOFTWARE);
 
     case -1:
       mu_diag_output (MU_DIAG_CRIT,
@@ -464,8 +478,8 @@ progman_start ()
   struct prog *prog;
 
   for (prog = proghead; prog; prog = prog->next)
-    if (IS_PROG (prog))
-	prog_start (prog);
+    if (IS_PROG (prog) && prog->v.p.status != status_disabled)
+      prog_start (prog);
 }
 
 void
@@ -474,9 +488,9 @@ progman_wake_disabled ()
   struct prog *prog;
 
   for (prog = proghead; prog; prog = prog->next)
-    if (IS_PROG (prog) && prog->v.p.disabled)
+    if (IS_PROG (prog) && prog->v.p.status == status_stopped)
       {
-	prog->v.p.disabled = 0;
+	prog->v.p.status = status_enabled;
 	prog->v.p.count = 0;
 	prog->v.p.timestamp = 0;
 	prog_start (prog);
@@ -580,6 +594,109 @@ prog_stop_dependent (struct prog *prog)
       }
 }
 
+static void
+send_msg (mu_address_t rcpt, mu_message_t msg)
+{
+  mu_mailer_t mailer;
+  int rc;
+
+  rc = mu_mailer_create (&mailer, NULL);
+  if (rc)
+    {
+      const char *mailer_url;
+      mu_mailer_get_url_default (&mailer_url);
+
+      mu_error (_("Cannot create mailer `%s': %s"),
+		mailer_url, mu_strerror (rc));
+      return;
+    }
+  
+  /* FIXME: mailer flags? */
+  rc = mu_mailer_open (mailer, 0);
+  if (rc)
+    {
+      const char *mailer_url;
+      mu_mailer_get_url_default (&mailer_url);
+
+      mu_mailer_destroy(&mailer);
+      mu_error (_("Opening mailer `%s' failed: %s"),
+		mailer_url, mu_strerror (rc));
+      return;
+    }
+	
+  rc = mu_mailer_send_message (mailer, msg, NULL, rcpt);
+  mu_mailer_destroy (&mailer);
+  if (rc)
+    mu_error (_("Cannot send message: %s"), mu_strerror (rc));
+}
+
+static const char default_termination_message[] =
+"From: <>\n"
+"X-Agent: ${canonical-program-name} (${package} ${version})\n"
+"Subject: Component ${component} terminated with code ${retcode}.\n"
+"\n";
+
+static void
+vartab_define_project (mu_vartab_t vtab)
+{
+  static struct {
+    char *name;
+    char *value;
+  } ptab[] = {
+    { "canonical-program-name", "pies" },
+    { "package", PACKAGE },
+    { "version", PACKAGE_VERSION },
+    { "mu-version", MAILUTILS_VERSION },
+    { NULL }
+  };
+  int i;
+
+  for (i = 0; ptab[i].name; i++)
+    mu_vartab_define (vtab, ptab[i].name, ptab[i].value, 1);
+}
+    
+
+static void
+notify (const char *tag, int status, struct action *act)
+{
+  mu_vartab_t vtab;
+  char *msg_text = NULL;
+  mu_message_t msg = NULL;
+  mu_stream_t stream = NULL;
+  int rc;
+  
+  mu_vartab_create (&vtab);
+  mu_vartab_define (vtab, "component", tag, 1);
+  mu_vartab_define (vtab, "retcode", mu_umaxtostr (0, status), 1);
+  mu_vartab_define (vtab, "program-name", mu_program_name, 1);
+  vartab_define_project (vtab);
+  rc = mu_vartab_expand (vtab,
+			 act->message ?
+			   act->message : default_termination_message,
+			 &msg_text);
+  mu_vartab_destroy (&vtab);
+  if (rc)
+    {
+      mu_error (_("cannot expand message body: %s"), mu_strerror (rc));
+      /* FIXME: Notify anyway? */
+      return;
+    }
+
+  rc = mu_message_create (&msg, NULL);
+  if (rc)
+    {
+      mu_error (_("cannot create message: %s"), mu_strerror (rc));
+      free (msg_text);
+    }
+  
+  mu_message_get_stream (msg, &stream);
+  mu_stream_write (stream, msg_text, strlen (msg_text), 0, NULL);
+  free (msg_text);
+  
+  send_msg (act->addr, msg);
+  mu_message_destroy (&msg, mu_message_get_owner (msg));
+}
+
 void
 progman_cleanup (int expect_term)
 {
@@ -601,7 +718,35 @@ progman_cleanup (int expect_term)
 	{
 	  prog_stop_dependent (prog);
 	  if (!expect_term)
-	    prog_start (prog);
+	    {
+	      if (WIFEXITED (status))
+		{
+		  struct action *act;
+		  status = WEXITSTATUS (status);
+		  if (status <= MAX_RETURN_CODE
+		      && prog->v.p.act && (act = prog->v.p.act[status]))
+		    {
+		      switch (act->act)
+			{
+			case action_restart:
+			  prog_start (prog);
+			  break;
+			  
+			case action_disable:
+			  mu_diag_output (MU_DIAG_NOTICE,
+					  _("Disabling component %s: exited "
+					    "with code %d"),
+					  prog->tag, status);
+			  prog->v.p.status = status_disabled;
+			}
+		      if (act->addr)
+			notify (prog->tag, status, act);
+		      continue;
+		    }
+		}
+	      /* Default action: */
+	      prog_start (prog);
+	    }
 	}
     }
 }
@@ -615,7 +760,7 @@ progman_stop_component (const char *name)
   for (prog = proghead; prog; prog = prog->next)
     if (IS_PROG (prog) && strcmp (prog->tag, name) == 0)
       {
-	if (prog->v.p.disabled)
+	if (prog->v.p.status != status_enabled)
 	  mu_diag_output (MU_DIAG_INFO,
 			  _("stopping component `%s': component not started"),
 		  name);
@@ -650,7 +795,7 @@ progman_dump_stats (const char *filename)
 		   prog->tag);
 	  if (prog->pid)
 	    fprintf (fp, "%lu", (unsigned long) prog->pid);
-	  else if (prog->v.p.disabled)
+	  else if (prog->v.p.status == status_stopped)
 	    {
 	      char buf[48];
 	      time_t t = prog->v.p.timestamp + SLEEPTIME;
@@ -658,6 +803,8 @@ progman_dump_stats (const char *filename)
 			localtime (&t));
 	      fprintf (fp, _("[disabled; scheduled for %s]"), buf);
 	    }
+	  else if (prog->v.p.status == status_disabled)
+	    fprintf (fp, _("[disabled]"));
 	  else
 	    fprintf (fp, "[not running]");
 	  if (rc = mu_argcv_string (prog->v.p.argc, prog->v.p.argv, &s))