when tdb_chainunlock() fails, print the tdb error that occured
[metze/ctdb/wip.git] / server / eventscript.c
index 54d914b35cf3d3c8f4e8316df0f5b9bb3c1b2b84..8011269e26f625595841a5eae5302fa91e70bf57 100644 (file)
@@ -18,6 +18,7 @@
 */
 
 #include "includes.h"
+#include <time.h>
 #include "system/filesys.h"
 #include "system/wait.h"
 #include "system/dir.h"
 #include "lib/events/events.h"
 #include "../common/rb_tree.h"
 
-static struct {
-       struct timeval start;
-       const char *script_running;
-} child_state;
+static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te, struct timeval t, void *p);
 
 /*
-  ctdbd sends us a SIGTERM when we should time out the current script
+  ctdbd sends us a SIGTERM when we should die.
  */
 static void sigterm(int sig)
 {
-       DEBUG(DEBUG_ERR,("Timed out running script '%s' after %.1f seconds\n", 
-                child_state.script_running, timeval_elapsed(&child_state.start)));
        /* all the child processes will be running in the same process group */
        kill(-getpgrp(), SIGKILL);
-       exit(1);
+       _exit(1);
 }
 
 struct ctdb_event_script_state {
        struct ctdb_context *ctdb;
        pid_t child;
+       /* Warning: this can free us! */
        void (*callback)(struct ctdb_context *, int, void *);
        int fd[2];
        void *private_data;
+       bool from_user;
+       enum ctdb_eventscript_call call;
        const char *options;
+       struct timeval timeout;
+
+       unsigned int current;
+       struct ctdb_scripts_wire *scripts;
 };
 
-/*
-  run the event script - varargs version
-  this function is called and run in the context of a forked child
-  which allows it to do blocking calls such as system()
+static struct ctdb_script_wire *get_current_script(struct ctdb_event_script_state *state)
+{
+       return &state->scripts->scripts[state->current];
+}
+
+/* called from ctdb_logging when we have received output on STDERR from
+ * one of the eventscripts
  */
-static int ctdb_event_script_v(struct ctdb_context *ctdb, const char *options)
+static void log_event_script_output(const char *str, uint16_t len, void *p)
 {
-       char *cmdstr;
-       int ret;
+       struct ctdb_event_script_state *state
+               = talloc_get_type(p, struct ctdb_event_script_state);
+       struct ctdb_script_wire *current;
+       unsigned int slen, min;
+
+       /* We may have been aborted to run something else.  Discard */
+       if (state->scripts == NULL) {
+               return;
+       }
+
+       current = get_current_script(state);
+
+       /* Append, but don't overfill buffer.  It starts zero-filled. */
+       slen = strlen(current->output);
+       min = MIN(len, sizeof(current->output) - slen - 1);
+
+       memcpy(current->output + slen, str, min);
+}
+
+int32_t ctdb_control_get_event_script_status(struct ctdb_context *ctdb,
+                                            uint32_t call_type,
+                                            TDB_DATA *outdata)
+{
+       if (call_type >= CTDB_EVENT_MAX) {
+               return -1;
+       }
+
+       if (ctdb->last_status[call_type] == NULL) {
+               /* If it's never been run, return nothing so they can tell. */
+               outdata->dsize = 0;
+       } else {
+               outdata->dsize = talloc_get_size(ctdb->last_status[call_type]);
+               outdata->dptr  = (uint8_t *)ctdb->last_status[call_type];
+       }
+       return 0;
+}
+
+struct ctdb_script_tree_item {
+       const char *name;
+       int error;
+};
+
+/* Return true if OK, otherwise set errno. */
+static bool check_executable(const char *dir, const char *name)
+{
+       char *full;
        struct stat st;
-       TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
-       trbt_tree_t *tree;
-       DIR *dir;
-       struct dirent *de;
-       char *script;
 
-       if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
-               /* we guarantee that only some specifically allowed event scripts are run
-                  while in recovery */
-               const char *allowed_scripts[] = {"startrecovery", "shutdown" };
-               int i;
-               for (i=0;i<ARRAY_SIZE(allowed_scripts);i++) {
-                       if (strcmp(options, allowed_scripts[i]) == 0) break;
-               }
-               if (i == ARRAY_SIZE(allowed_scripts)) {
-                       DEBUG(0,("Refusing to run event scripts with option '%s' while in recovery\n",
-                                options));
-                       return -1;
-               }
+       full = talloc_asprintf(NULL, "%s/%s", dir, name);
+       if (!full)
+               return false;
+
+       if (stat(full, &st) != 0) {
+               DEBUG(DEBUG_ERR,("Could not stat event script %s: %s\n",
+                                full, strerror(errno)));
+               talloc_free(full);
+               return false;
        }
 
-       if (setpgid(0,0) != 0) {
-               DEBUG(DEBUG_ERR,("Failed to create process group for event scripts - %s\n",
-                        strerror(errno)));
-               talloc_free(tmp_ctx);
-               return -1;              
+       if (!(st.st_mode & S_IXUSR)) {
+               DEBUG(DEBUG_DEBUG,("Event script %s is not executable. Ignoring this event script\n", full));
+               errno = ENOEXEC;
+               talloc_free(full);
+               return false;
        }
 
-       signal(SIGTERM, sigterm);
+       talloc_free(full);
+       return true;
+}
 
-       child_state.start = timeval_current();
-       child_state.script_running = "startup";
+static struct ctdb_scripts_wire *ctdb_get_script_list(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx)
+{
+       DIR *dir;
+       struct dirent *de;
+       struct stat st;
+       trbt_tree_t *tree;
+       struct ctdb_scripts_wire *scripts;
+       TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
+       struct ctdb_script_tree_item *tree_item;
+       int count;
 
        /*
          the service specific event scripts 
@@ -102,7 +153,7 @@ static int ctdb_event_script_v(struct ctdb_context *ctdb, const char *options)
            errno == ENOENT) {
                DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
                talloc_free(tmp_ctx);
-               return -1;
+               return NULL;
        }
 
        /* create a tree to store all the script names in */
@@ -115,13 +166,13 @@ static int ctdb_event_script_v(struct ctdb_context *ctdb, const char *options)
        if (dir == NULL) {
                DEBUG(DEBUG_CRIT,("Failed to open event script directory '%s'\n", ctdb->event_script_dir));
                talloc_free(tmp_ctx);
-               return -1;
+               return NULL;
        }
 
+       count = 0;
        while ((de=readdir(dir)) != NULL) {
                int namlen;
                unsigned num;
-               char *str;
 
                namlen = strlen(de->d_name);
 
@@ -142,65 +193,242 @@ static int ctdb_event_script_v(struct ctdb_context *ctdb, const char *options)
                        continue;
                }
 
-               /* Make sure the event script is executable */
-               str = talloc_asprintf(tree, "%s/%s", ctdb->event_script_dir, de->d_name);
-               if (stat(str, &st) != 0) {
-                       DEBUG(DEBUG_ERR,("Could not stat event script %s. Ignoring this event script\n", str));
+               if (strlen(de->d_name) > MAX_SCRIPT_NAME) {
+                       DEBUG(DEBUG_ERR,("Script name %s too long! %u chars max",
+                                        de->d_name, MAX_SCRIPT_NAME));
                        continue;
                }
-               if (!(st.st_mode & S_IXUSR)) {
-                       DEBUG(DEBUG_ERR,("Event script %s is not executable. Ignoring this event script\n", str));
-                       continue;
+
+               tree_item = talloc(tree, struct ctdb_script_tree_item);
+               if (tree_item == NULL) {
+                       DEBUG(DEBUG_ERR, (__location__ " Failed to allocate new tree item\n"));
+                       talloc_free(tmp_ctx);
+                       return NULL;
+               }
+       
+               tree_item->error = 0;
+               if (!check_executable(ctdb->event_script_dir, de->d_name)) {
+                       tree_item->error = errno;
                }
-               
-               
-               /* store the event script in the tree */                
-               script = trbt_insert32(tree, num, talloc_strdup(tree, de->d_name));
-               if (script != NULL) {
-                       DEBUG(DEBUG_CRIT,("CONFIG ERROR: Multiple event scripts with the same prefix : '%s' and '%s'. Each event script MUST have a unique prefix\n", script, de->d_name));
+
+               tree_item->name = talloc_strdup(tree_item, de->d_name);
+               if (tree_item->name == NULL) {
+                       DEBUG(DEBUG_ERR,(__location__ " Failed to allocate script name.\n"));
                        talloc_free(tmp_ctx);
-                       closedir(dir);
-                       return -1;
+                       return NULL;
                }
+
+               /* store the event script in the tree */
+               trbt_insert32(tree, (num<<16)|count++, tree_item);
        }
        closedir(dir);
 
-       /* fetch the scripts from the tree one by one and execute
-          them
-        */
-       while ((script=trbt_findfirstarray32(tree, 1)) != NULL) {
-               cmdstr = talloc_asprintf(tmp_ctx, "%s/%s %s", 
-                               ctdb->event_script_dir,
-                               script, options);
-               CTDB_NO_MEMORY(ctdb, cmdstr);
-
-               DEBUG(DEBUG_INFO,("Executing event script %s\n",cmdstr));
-
-               child_state.start = timeval_current();
-               child_state.script_running = cmdstr;
-
-               ret = system(cmdstr);
-               /* if the system() call was successful, translate ret into the
-                  return code from the command
-               */
-               if (ret != -1) {
-                       ret = WEXITSTATUS(ret);
-               }
-               /* return an error if the script failed */
-               if (ret != 0) {
-                       DEBUG(DEBUG_ERR,("Event script %s failed with error %d\n", cmdstr, ret));
-                       talloc_free(tmp_ctx);
-                       return ret;
-               }
+       /* Overallocates by one, but that's OK */
+       scripts = talloc_zero_size(tmp_ctx,
+                                  sizeof(*scripts)
+                                  + sizeof(scripts->scripts[0]) * count);
+       if (scripts == NULL) {
+               DEBUG(DEBUG_ERR, (__location__ " Failed to allocate scripts\n"));
+               talloc_free(tmp_ctx);
+               return NULL;
+       }
+       scripts->num_scripts = count;
+
+       for (count = 0; count < scripts->num_scripts; count++) {
+               tree_item = trbt_findfirstarray32(tree, 1);
+
+               strcpy(scripts->scripts[count].name, tree_item->name);
+               scripts->scripts[count].status = -tree_item->error;
 
                /* remove this script from the tree */
-               talloc_free(script);
+               talloc_free(tree_item);
        }
 
-       child_state.start = timeval_current();
-       child_state.script_running = "finished";
-       
+       talloc_steal(mem_ctx, scripts);
+       talloc_free(tmp_ctx);
+       return scripts;
+}
+
+static int child_setup(struct ctdb_context *ctdb)
+{
+       if (setpgid(0,0) != 0) {
+               int ret = -errno;
+               DEBUG(DEBUG_ERR,("Failed to create process group for event scripts - %s\n",
+                        strerror(errno)));
+               return ret;
+       }
+
+       signal(SIGTERM, sigterm);
+       return 0;
+}
+
+static char *child_command_string(struct ctdb_context *ctdb,
+                                      TALLOC_CTX *ctx,
+                                      bool from_user,
+                                      const char *scriptname,
+                                      enum ctdb_eventscript_call call,
+                                      const char *options)
+{
+       const char *str = from_user ? "CTDB_CALLED_BY_USER=1 " : "";
+
+       /* Allow a setting where we run the actual monitor event
+          from an external source and replace it with
+          a "status" event that just picks up the actual
+          status of the event asynchronously.
+       */
+       if ((ctdb->tunable.use_status_events_for_monitoring != 0)
+           &&  (call == CTDB_EVENT_MONITOR)
+           &&  !from_user) {
+               return talloc_asprintf(ctx, "%s%s/%s %s",
+                                      str,
+                                      ctdb->event_script_dir,
+                                      scriptname, "status");
+       } else {
+               return talloc_asprintf(ctx, "%s%s/%s %s %s",
+                                      str,
+                                      ctdb->event_script_dir,
+                                      scriptname,
+                                      ctdb_eventscript_call_names[call],
+                                      options);
+       }
+}
+
+static int child_run_one(struct ctdb_context *ctdb,
+                        const char *scriptname, const char *cmdstr)
+{
+       int ret;
+
+       ret = system(cmdstr);
+       /* if the system() call was successful, translate ret into the
+          return code from the command
+       */
+       if (ret != -1) {
+               ret = WEXITSTATUS(ret);
+       } else {
+               ret = -errno;
+       }
+
+       /* 127 could mean it does not exist, 126 non-executable. */
+       if (ret == 127 || ret == 126) {
+               /* Re-check it... */
+               if (!check_executable(ctdb->event_script_dir, scriptname)) {
+                       DEBUG(DEBUG_ERR,("Script %s returned status %u. Someone just deleted it?\n",
+                                        cmdstr, ret));
+                       ret = -errno;
+               }
+       }
+       return ret;
+}
+
+/*
+  Actually run one event script
+  this function is called and run in the context of a forked child
+  which allows it to do blocking calls such as system()
+ */
+static int child_run_script(struct ctdb_context *ctdb,
+                           bool from_user,
+                           enum ctdb_eventscript_call call,
+                           const char *options,
+                           struct ctdb_script_wire *current)
+{
+       char *cmdstr;
+       int ret;
+       TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
+
+       ret = child_setup(ctdb);
+       if (ret != 0)
+               goto out;
+
+       cmdstr = child_command_string(ctdb, tmp_ctx, from_user,
+                                     current->name, call, options);
+       CTDB_NO_MEMORY(ctdb, cmdstr);
+
+       DEBUG(DEBUG_DEBUG,("Executing event script %s\n",cmdstr));
+
+       if (current->status) {
+               ret = current->status;
+               goto out;
+       }
+
+       ret = child_run_one(ctdb, current->name, cmdstr);
+out:
        talloc_free(tmp_ctx);
+       return ret;
+}
+
+static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
+                                     uint16_t flags, void *p);
+
+static int fork_child_for_script(struct ctdb_context *ctdb,
+                                struct ctdb_event_script_state *state)
+{
+       int r;
+       struct ctdb_script_wire *current = get_current_script(state);
+
+       current->start = timeval_current();
+
+       r = pipe(state->fd);
+       if (r != 0) {
+               DEBUG(DEBUG_ERR, (__location__ " pipe failed for child eventscript process\n"));
+               return -errno;
+       }
+
+       if (!ctdb_fork_with_logging(state, ctdb, log_event_script_output,
+                                   state, &state->child)) {
+               r = -errno;
+               close(state->fd[0]);
+               close(state->fd[1]);
+               return r;
+       }
+
+       /* If we are the child, do the work. */
+       if (state->child == 0) {
+               int rt;
+
+               close(state->fd[0]);
+               set_close_on_exec(state->fd[1]);
+
+               rt = child_run_script(ctdb, state->from_user, state->call, state->options, current);
+               /* We must be able to write PIPEBUF bytes at least; if this
+                  somehow fails, the read above will be short. */
+               write(state->fd[1], &rt, sizeof(rt));
+               close(state->fd[1]);
+               _exit(rt);
+       }
+
+       close(state->fd[1]);
+       set_close_on_exec(state->fd[0]);
+
+       DEBUG(DEBUG_DEBUG, (__location__ " Created PIPE FD:%d to child eventscript process\n", state->fd[0]));
+
+       /* Set ourselves up to be called when that's done. */
+       event_add_fd(ctdb->ev, state, state->fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
+                    ctdb_event_script_handler, state);
+       return 0;
+}
+
+/*
+ Summarize status of this run of scripts.
+ */
+static int script_status(struct ctdb_scripts_wire *scripts)
+{
+       unsigned int i;
+
+       for (i = 0; i < scripts->num_scripts; i++) {
+               switch (scripts->scripts[i].status) {
+               case -ENOENT:
+               case -ENOEXEC:
+                       /* Disabled or missing; that's OK. */
+                       break;
+               case 0:
+                       /* No problem. */
+                       break;
+               default:
+                       return scripts->scripts[i].status;
+               }
+       }
+
+       /* All OK! */
        return 0;
 }
 
@@ -210,152 +438,316 @@ static void ctdb_event_script_handler(struct event_context *ev, struct fd_event
 {
        struct ctdb_event_script_state *state = 
                talloc_get_type(p, struct ctdb_event_script_state);
-       int status = -1;
-       void (*callback)(struct ctdb_context *, int, void *) = state->callback;
-       void *private_data = state->private_data;
+       struct ctdb_script_wire *current = get_current_script(state);
        struct ctdb_context *ctdb = state->ctdb;
+       int r, status;
 
-       waitpid(state->child, &status, 0);
-       if (status != -1) {
-               status = WEXITSTATUS(status);
+       r = read(state->fd[0], &current->status, sizeof(current->status));
+       if (r < 0) {
+               current->status = -errno;
+       } else if (r != sizeof(current->status)) {
+               current->status = -EIO;
        }
-       talloc_set_destructor(state, NULL);
-       talloc_free(state);
-       callback(ctdb, status, private_data);
 
-       ctdb->event_script_timeouts = 0;
-}
+       current->finished = timeval_current();
+       /* valgrind gets overloaded if we run next script as it's still doing
+        * post-execution analysis, so kill finished child here. */
+       if (ctdb->valgrinding) {
+               kill(state->child, SIGKILL);
+       }
 
-static void ctdb_ban_self(struct ctdb_context *ctdb, uint32_t ban_period)
-{
-       int ret;
-       struct ctdb_ban_info b;
-       TDB_DATA data;
+       state->child = 0;
 
-       b.pnn      = ctdb->pnn;
-       b.ban_time = ban_period;
+       status = script_status(state->scripts);
 
-       data.dptr = (uint8_t *)&b;
-       data.dsize = sizeof(b);
+       /* Aborted or finished all scripts?  We're done. */
+       if (status != 0 || state->current+1 == state->scripts->num_scripts) {
+               DEBUG(DEBUG_INFO,(__location__ " Eventscript %s %s finished with state %d\n",
+                                 ctdb_eventscript_call_names[state->call], state->options, status));
 
-       ret = ctdb_daemon_send_message(ctdb, CTDB_BROADCAST_CONNECTED,
-               CTDB_SRVID_BAN_NODE, data);
-       if (ret != 0) {
-               DEBUG(DEBUG_ERR,(__location__ " Failed to send ban message\n"));
+               ctdb->event_script_timeouts = 0;
+               talloc_free(state);
+               return;
+       }
+
+       /* Forget about that old fd. */
+       talloc_free(fde);
+
+       /* Next script! */
+       state->current++;
+       current++;
+       current->status = fork_child_for_script(ctdb, state);
+       if (current->status != 0) {
+               /* This calls the callback. */
+               talloc_free(state);
        }
 }
 
+static void debug_timeout(struct ctdb_event_script_state *state)
+{
+       struct ctdb_script_wire *current = get_current_script(state);
+       char *cmd;
+       pid_t pid;
+       time_t t;
+       char tbuf[100], buf[200];
+
+       cmd = child_command_string(state->ctdb, state,
+                                  state->from_user, current->name,
+                                  state->call, state->options);
+       CTDB_NO_MEMORY_VOID(state->ctdb, cmd);
+
+       DEBUG(DEBUG_ERR,("Timed out running script '%s' after %.1f seconds pid :%d\n",
+                        cmd, timeval_elapsed(&current->start), state->child));
+       talloc_free(cmd);
+
+       t = time(NULL);
+       strftime(tbuf, sizeof(tbuf)-1, "%Y%m%d%H%M%S",  localtime(&t));
+       sprintf(buf, "{ pstree -p; cat /proc/locks; ls -li /var/ctdb/ /var/ctdb/persistent; }"
+                       " >/tmp/ctdb.event.%s.%d", tbuf, getpid());
+
+       pid = fork();
+       if (pid == 0) {
+               ctdb_reduce_priority(state->ctdb);
+               system(buf);
+               /* Now we can kill the child */
+               kill(state->child, SIGTERM);
+               exit(0);
+       }
+       if (pid == -1) {
+               DEBUG(DEBUG_ERR,("Fork for debug script failed : %s\n",
+                                strerror(errno)));
+       } else {
+               DEBUG(DEBUG_ERR,("Logged timedout eventscript : %s\n", buf));
+               /* Don't kill child until timeout done. */
+               state->child = 0;
+       }
+}
 
 /* called when child times out */
 static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te, 
                                      struct timeval t, void *p)
 {
        struct ctdb_event_script_state *state = talloc_get_type(p, struct ctdb_event_script_state);
-       void (*callback)(struct ctdb_context *, int, void *) = state->callback;
-       void *private_data = state->private_data;
        struct ctdb_context *ctdb = state->ctdb;
-
-       DEBUG(DEBUG_ERR,("Event script timed out : %s count : %u\n", state->options, ctdb->event_script_timeouts));
-
-       if (!strcmp(state->options, "monitor")) {
-               /* if it is a monitor event, we allow it to "hang" a few times
-                  before we declare it a failure and ban ourself (and make
-                  ourself unhealthy)
-               */
-               DEBUG(DEBUG_ERR, (__location__ " eventscript for monitor event timedout.\n"));
-
-               ctdb->event_script_timeouts++;
-               if (ctdb->event_script_timeouts > ctdb->tunable.script_ban_count) {
-                       ctdb->event_script_timeouts = 0;
-                       DEBUG(DEBUG_ERR, ("Maximum timeout count %u reached for eventscript. Banning self for %d seconds\n", ctdb->tunable.script_ban_count, ctdb->tunable.recovery_ban_period));
-                       ctdb_ban_self(ctdb, ctdb->tunable.recovery_ban_period);
-                       callback(ctdb, -1, private_data);
-               } else {
-                       callback(ctdb, 0, private_data);
-               }
-       } else if (!strcmp(state->options, "startup")) {
-               DEBUG(DEBUG_ERR, (__location__ " eventscript for startup event timedout.\n"));
-               callback(ctdb, -1, private_data);
-       } else {
-               /* if it is not a monitor event we ban ourself immediately */
-               DEBUG(DEBUG_ERR, (__location__ " eventscript for NON-monitor/NON-startup event timedout. Immediately banning ourself for %d seconds\n", ctdb->tunable.recovery_ban_period));
-               ctdb_ban_self(ctdb, ctdb->tunable.recovery_ban_period);
-               callback(ctdb, -1, private_data);
+       struct ctdb_script_wire *current = get_current_script(state);
+
+       DEBUG(DEBUG_ERR,("Event script timed out : %s %s %s count : %u  pid : %d\n",
+                        current->name, ctdb_eventscript_call_names[state->call], state->options, ctdb->event_script_timeouts, state->child));
+
+       /* ignore timeouts for these events */
+       switch (state->call) {
+       case CTDB_EVENT_START_RECOVERY:
+       case CTDB_EVENT_RECOVERED:
+       case CTDB_EVENT_TAKE_IP:
+       case CTDB_EVENT_RELEASE_IP:
+       case CTDB_EVENT_STOPPED:
+       case CTDB_EVENT_MONITOR:
+       case CTDB_EVENT_STATUS:
+               state->scripts->scripts[state->current].status = 0;
+               DEBUG(DEBUG_ERR,("Ignoring hung script for %s call %d\n", state->options, state->call));
+               break;
+        default:
+               state->scripts->scripts[state->current].status = -ETIME;
+               debug_timeout(state);
        }
 
        talloc_free(state);
 }
 
 /*
-  destroy a running event script
+  destroy an event script: kill it if ->child != 0.
  */
 static int event_script_destructor(struct ctdb_event_script_state *state)
 {
-       DEBUG(DEBUG_ERR,(__location__ " Sending SIGTERM to child pid:%d\n", state->child));
-       kill(state->child, SIGTERM);
-       waitpid(state->child, NULL, 0);
+       int status;
+
+       if (state->child) {
+               DEBUG(DEBUG_ERR,(__location__ " Sending SIGTERM to child pid:%d\n", state->child));
+
+               if (kill(state->child, SIGTERM) != 0) {
+                       DEBUG(DEBUG_ERR,("Failed to kill child process for eventscript, errno %s(%d)\n", strerror(errno), errno));
+               }
+       }
+
+       /* If we were the current monitor, we no longer are. */
+       if (state->ctdb->current_monitor == state) {
+               state->ctdb->current_monitor = NULL;
+       }
+
+       /* Save our scripts as the last executed status, if we have them.
+        * See ctdb_event_script_callback_v where we abort monitor event. */
+       if (state->scripts) {
+               talloc_free(state->ctdb->last_status[state->call]);
+               state->ctdb->last_status[state->call] = state->scripts;
+               if (state->current < state->ctdb->last_status[state->call]->num_scripts) {
+                       state->ctdb->last_status[state->call]->num_scripts = state->current+1;
+               }
+       }
+
+       /* Use last status as result, or "OK" if none. */
+       if (state->ctdb->last_status[state->call]) {
+               status = script_status(state->ctdb->last_status[state->call]);
+       } else {
+               status = 0;
+       }
+
+       /* This is allowed to free us; talloc will prevent double free anyway,
+        * but beware if you call this outside the destructor! */
+       if (state->callback) {
+               state->callback(state->ctdb, status, state->private_data);
+       }
+
        return 0;
 }
 
+static unsigned int count_words(const char *options)
+{
+       unsigned int words = 0;
+
+       options += strspn(options, " \t");
+       while (*options) {
+               words++;
+               options += strcspn(options, " \t");
+               options += strspn(options, " \t");
+       }
+       return words;
+}
+
+static bool check_options(enum ctdb_eventscript_call call, const char *options)
+{
+       switch (call) {
+       /* These all take no arguments. */
+       case CTDB_EVENT_INIT:
+       case CTDB_EVENT_SETUP:
+       case CTDB_EVENT_STARTUP:
+       case CTDB_EVENT_START_RECOVERY:
+       case CTDB_EVENT_RECOVERED:
+       case CTDB_EVENT_STOPPED:
+       case CTDB_EVENT_MONITOR:
+       case CTDB_EVENT_STATUS:
+       case CTDB_EVENT_SHUTDOWN:
+       case CTDB_EVENT_RELOAD:
+               return count_words(options) == 0;
+
+       case CTDB_EVENT_TAKE_IP: /* interface, IP address, netmask bits. */
+       case CTDB_EVENT_RELEASE_IP:
+               return count_words(options) == 3;
+
+       case CTDB_EVENT_UPDATE_IP: /* old interface, new interface, IP address, netmask bits. */
+               return count_words(options) == 4;
+
+       default:
+               DEBUG(DEBUG_ERR,(__location__ "Unknown ctdb_eventscript_call %u\n", call));
+               return false;
+       }
+}
+
 /*
   run the event script in the background, calling the callback when 
   finished
  */
 static int ctdb_event_script_callback_v(struct ctdb_context *ctdb, 
-                                       struct timeval timeout,
-                                       TALLOC_CTX *mem_ctx,
                                        void (*callback)(struct ctdb_context *, int, void *),
                                        void *private_data,
+                                       bool from_user,
+                                       enum ctdb_eventscript_call call,
                                        const char *fmt, va_list ap)
 {
        struct ctdb_event_script_state *state;
-       int ret;
 
-       state = talloc(mem_ctx, struct ctdb_event_script_state);
+       state = talloc(ctdb->event_script_ctx, struct ctdb_event_script_state);
        CTDB_NO_MEMORY(ctdb, state);
 
        state->ctdb = ctdb;
        state->callback = callback;
        state->private_data = private_data;
+       state->from_user = from_user;
+       state->call = call;
        state->options = talloc_vasprintf(state, fmt, ap);
-       CTDB_NO_MEMORY(ctdb, state->options);
-       
-       ret = pipe(state->fd);
-       if (ret != 0) {
+       state->timeout = timeval_set(ctdb->tunable.script_timeout, 0);
+       state->scripts = NULL;
+       if (state->options == NULL) {
+               DEBUG(DEBUG_ERR, (__location__ " could not allocate state->options\n"));
                talloc_free(state);
                return -1;
        }
+       if (!check_options(state->call, state->options)) {
+               DEBUG(DEBUG_ERR, ("Bad eventscript options '%s' for %s\n",
+                                 ctdb_eventscript_call_names[state->call], state->options));
+               talloc_free(state);
+               return -1;
+       }
+
+       if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
+               /* we guarantee that only some specifically allowed event scripts are run
+                  while in recovery */
+               const enum ctdb_eventscript_call allowed_calls[] = {
+                       CTDB_EVENT_INIT,
+                       CTDB_EVENT_SETUP,
+                       CTDB_EVENT_START_RECOVERY,
+                       CTDB_EVENT_SHUTDOWN,
+                       CTDB_EVENT_RELEASE_IP,
+                       CTDB_EVENT_STOPPED
+               };
+               int i;
+               for (i=0;i<ARRAY_SIZE(allowed_calls);i++) {
+                       if (call == allowed_calls[i]) break;
+               }
+               if (i == ARRAY_SIZE(allowed_calls)) {
+                       DEBUG(DEBUG_ERR,("Refusing to run event scripts call '%s' while in recovery\n",
+                                ctdb_eventscript_call_names[call]));
+                       talloc_free(state);
+                       return -1;
+               }
+       }
 
-       state->child = fork();
+       /* Kill off any running monitor events to run this event. */
+       if (ctdb->current_monitor) {
+               /* Discard script status so we don't save to last_status */
+               talloc_free(ctdb->current_monitor->scripts);
+               ctdb->current_monitor->scripts = NULL;
+               talloc_free(ctdb->current_monitor);
+               ctdb->current_monitor = NULL;
+       }
 
-       if (state->child == (pid_t)-1) {
-               close(state->fd[0]);
-               close(state->fd[1]);
+       DEBUG(DEBUG_INFO,(__location__ " Starting eventscript %s %s\n",
+                         ctdb_eventscript_call_names[state->call],
+                         state->options));
+
+       /* This is not a child of state, since we save it in destructor. */
+       state->scripts = ctdb_get_script_list(ctdb, ctdb);
+       if (state->scripts == NULL) {
                talloc_free(state);
                return -1;
        }
+       state->current = 0;
+       state->child = 0;
 
-       if (state->child == 0) {
-               close(state->fd[0]);
-               if (ctdb->do_setsched) {
-                       ctdb_restore_scheduler(ctdb);
-               }
-               set_close_on_exec(state->fd[1]);
-               ret = ctdb_event_script_v(ctdb, state->options);
-               _exit(ret);
+       if (!from_user && (call == CTDB_EVENT_MONITOR || call == CTDB_EVENT_STATUS)) {
+               ctdb->current_monitor = state;
        }
 
        talloc_set_destructor(state, event_script_destructor);
 
-       close(state->fd[1]);
+       /* Nothing to do? */
+       if (state->scripts->num_scripts == 0) {
+               talloc_free(state);
+               return 0;
+       }
 
-       event_add_fd(ctdb->ev, state, state->fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
-                    ctdb_event_script_handler, state);
+       state->scripts->scripts[0].status = fork_child_for_script(ctdb, state);
+       if (state->scripts->scripts[0].status != 0) {
+               /* Callback is called from destructor, with fail result. */
+               talloc_free(state);
+               return 0;
+       }
 
-       if (!timeval_is_zero(&timeout)) {
-               event_add_timed(ctdb->ev, state, timeout, ctdb_event_script_timeout, state);
+       if (!timeval_is_zero(&state->timeout)) {
+               event_add_timed(ctdb->ev, state, timeval_current_ofs(state->timeout.tv_sec, state->timeout.tv_usec), ctdb_event_script_timeout, state);
        } else {
-               DEBUG(DEBUG_ERR, (__location__ " eventscript %s called with no timeout\n", state->options));
+               DEBUG(DEBUG_ERR, (__location__ " eventscript %s %s called with no timeout\n",
+                                 ctdb_eventscript_call_names[state->call],
+                                 state->options));
        }
 
        return 0;
@@ -367,17 +759,18 @@ static int ctdb_event_script_callback_v(struct ctdb_context *ctdb,
   finished
  */
 int ctdb_event_script_callback(struct ctdb_context *ctdb, 
-                              struct timeval timeout,
                               TALLOC_CTX *mem_ctx,
                               void (*callback)(struct ctdb_context *, int, void *),
                               void *private_data,
+                              bool from_user,
+                              enum ctdb_eventscript_call call,
                               const char *fmt, ...)
 {
        va_list ap;
        int ret;
 
        va_start(ap, fmt);
-       ret = ctdb_event_script_callback_v(ctdb, timeout, mem_ctx, callback, private_data, fmt, ap);
+       ret = ctdb_event_script_callback_v(ctdb, callback, private_data, from_user, call, fmt, ap);
        va_end(ap);
 
        return ret;
@@ -400,42 +793,52 @@ static void event_script_callback(struct ctdb_context *ctdb, int status, void *p
 }
 
 /*
-  run the event script, waiting for it to complete. Used when the caller doesn't want to 
-  continue till the event script has finished.
+  run the event script, waiting for it to complete. Used when the caller
+  doesn't want to continue till the event script has finished.
  */
-int ctdb_event_script(struct ctdb_context *ctdb, const char *fmt, ...)
+int ctdb_event_script_args(struct ctdb_context *ctdb, enum ctdb_eventscript_call call,
+                          const char *fmt, ...)
 {
        va_list ap;
        int ret;
-       TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
        struct callback_status status;
 
        va_start(ap, fmt);
-       ret = ctdb_event_script_callback_v(ctdb, timeval_zero(), tmp_ctx, event_script_callback, &status, fmt, ap);
-       va_end(ap);
-
+       ret = ctdb_event_script_callback_v(ctdb,
+                       event_script_callback, &status, false, call, fmt, ap);
        if (ret != 0) {
-               talloc_free(tmp_ctx);
                return ret;
        }
+       va_end(ap);
 
        status.status = -1;
        status.done = false;
 
        while (status.done == false && event_loop_once(ctdb->ev) == 0) /* noop */;
 
-       talloc_free(tmp_ctx);
+       if (status.status == -ETIME) {
+               DEBUG(DEBUG_ERR, (__location__ " eventscript for '%s' timedout."
+                                 " Immediately banning ourself for %d seconds\n",
+                                 ctdb_eventscript_call_names[call],
+                                 ctdb->tunable.recovery_ban_period));
+               ctdb_ban_self(ctdb);
+       }
 
        return status.status;
 }
 
+int ctdb_event_script(struct ctdb_context *ctdb, enum ctdb_eventscript_call call)
+{
+       /* GCC complains about empty format string, so use %s and "". */
+       return ctdb_event_script_args(ctdb, call, "%s", "");
+}
 
 struct eventscript_callback_state {
        struct ctdb_req_control *c;
 };
 
 /*
-  called when takeip event finishes
+  called when a forced eventscript run has finished
  */
 static void run_eventscripts_callback(struct ctdb_context *ctdb, int status, 
                                 void *private_data)
@@ -447,17 +850,36 @@ static void run_eventscripts_callback(struct ctdb_context *ctdb, int status,
 
        if (status != 0) {
                DEBUG(DEBUG_ERR,(__location__ " Failed to forcibly run eventscripts\n"));
-               ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
-               talloc_free(state);
-               return;
        }
 
-       /* the control succeeded */
-       ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
+       ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
+       /* This will free the struct ctdb_event_script_state we are in! */
        talloc_free(state);
        return;
 }
 
+
+/* Returns rest of string, or NULL if no match. */
+static const char *get_call(const char *p, enum ctdb_eventscript_call *call)
+{
+       unsigned int len;
+
+       /* Skip any initial whitespace. */
+       p += strspn(p, " \t");
+
+       /* See if we match any. */
+       for (*call = 0; *call < CTDB_EVENT_MAX; (*call)++) {
+               len = strlen(ctdb_eventscript_call_names[*call]);
+               if (strncmp(p, ctdb_eventscript_call_names[*call], len) == 0) {
+                       /* If end of string or whitespace, we're done. */
+                       if (strcspn(p + len, " \t") == 0) {
+                               return p + len;
+                       }
+               }
+       }
+       return NULL;
+}
+
 /*
   A control to force running of the eventscripts from the ctdb client tool
 */
@@ -467,27 +889,33 @@ int32_t ctdb_run_eventscripts(struct ctdb_context *ctdb,
 {
        int ret;
        struct eventscript_callback_state *state;
+       const char *options;
+       enum ctdb_eventscript_call call;
+
+       /* Figure out what call they want. */
+       options = get_call((const char *)indata.dptr, &call);
+       if (!options) {
+               DEBUG(DEBUG_ERR, (__location__ " Invalid forced \"%s\"\n", (const char *)indata.dptr));
+               return -1;
+       }
 
-       /* kill off any previous invokations of forced eventscripts */
-       if (ctdb->eventscripts_ctx) {
-               talloc_free(ctdb->eventscripts_ctx);
+       if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
+               DEBUG(DEBUG_ERR, (__location__ " Aborted running eventscript \"%s\" while in RECOVERY mode\n", indata.dptr));
+               return -1;
        }
-       ctdb->eventscripts_ctx = talloc_new(ctdb);
-       CTDB_NO_MEMORY(ctdb, ctdb->eventscripts_ctx);
 
-       state = talloc(ctdb->eventscripts_ctx, struct eventscript_callback_state);
+       state = talloc(ctdb->event_script_ctx, struct eventscript_callback_state);
        CTDB_NO_MEMORY(ctdb, state);
 
-       state->c = talloc_steal(ctdb, c);
+       state->c = talloc_steal(state, c);
 
        DEBUG(DEBUG_NOTICE,("Forced running of eventscripts with arguments %s\n", indata.dptr));
 
        ctdb_disable_monitoring(ctdb);
 
        ret = ctdb_event_script_callback(ctdb, 
-                        timeval_current_ofs(ctdb->tunable.script_timeout, 0),
                         state, run_eventscripts_callback, state,
-                        (const char *)indata.dptr);
+                        true, call, "%s", options);
 
        if (ret != 0) {
                ctdb_enable_monitoring(ctdb);
@@ -502,3 +930,116 @@ int32_t ctdb_run_eventscripts(struct ctdb_context *ctdb,
        return 0;
 }
 
+
+
+int32_t ctdb_control_enable_script(struct ctdb_context *ctdb, TDB_DATA indata)
+{
+       const char *script;
+       struct stat st;
+       char *filename;
+       TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
+
+       script = (char *)indata.dptr;
+       if (indata.dsize == 0) {
+               DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+       if (indata.dptr[indata.dsize - 1] != '\0') {
+               DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+       if (index(script,'/') != NULL) {
+               DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to enable script %s\n", script));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+
+       if (stat(ctdb->event_script_dir, &st) != 0 && 
+           errno == ENOENT) {
+               DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+
+       filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
+       if (filename == NULL) {
+               DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+       if (stat(filename, &st) != 0) {
+               DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to enable script.\n", filename));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+       if (chmod(filename, st.st_mode | S_IXUSR) == -1) {
+               DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to enable script.\n", filename));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+       talloc_free(tmp_ctx);
+       return 0;
+}
+
+int32_t ctdb_control_disable_script(struct ctdb_context *ctdb, TDB_DATA indata)
+{
+       const char *script;
+       struct stat st;
+       char *filename;
+       TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
+
+       script = (char *)indata.dptr;
+       if (indata.dsize == 0) {
+               DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+       if (indata.dptr[indata.dsize - 1] != '\0') {
+               DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+       if (index(script,'/') != NULL) {
+               DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to disable script %s\n", script));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+
+       if (stat(ctdb->event_script_dir, &st) != 0 && 
+           errno == ENOENT) {
+               DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+
+       filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
+       if (filename == NULL) {
+               DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+       if (stat(filename, &st) != 0) {
+               DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to disable script.\n", filename));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+       if (chmod(filename, st.st_mode & ~(S_IXUSR|S_IXGRP|S_IXOTH)) == -1) {
+               DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to disable script.\n", filename));
+               talloc_free(tmp_ctx);
+               return -1;
+       }
+
+       talloc_free(tmp_ctx);
+       return 0;
+}