ctdb/server/eventscript.c

   1 /*
   2    event script handling
   3
   4    Copyright (C) Andrew Tridgell  2007
   5
   6    This program is free software; you can redistribute it and/or modify
   7    it under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3 of the License, or
   9    (at your option) any later version.
  10
  11    This program is distributed in the hope that it will be useful,
  12    but WITHOUT ANY WARRANTY; without even the implied warranty of
  13    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14    GNU General Public License for more details.
  15
  16    You should have received a copy of the GNU General Public License
  17    along with this program; if not, see <http://www.gnu.org/licenses/>.
  18 */
  19
  20 #include "includes.h"
  21 #include <time.h>
  22 #include "system/filesys.h"
  23 #include "system/wait.h"
  24 #include "system/dir.h"
  25 #include "system/locale.h"
  26 #include "../include/ctdb_private.h"
  27 #include "../common/rb_tree.h"
  28 #include "lib/util/dlinklist.h"
  29
  30 static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te, struct timeval t, void *p);
  31
  32 /* This is attached to the event script state. */
  33 struct event_script_callback {
  34         struct event_script_callback *next, *prev;
  35         struct ctdb_context *ctdb;
  36
  37         /* Warning: this can free us! */
  38         void (*fn)(struct ctdb_context *, int, void *);
  39         void *private_data;
  40 };
  41
  42
  43 struct ctdb_event_script_state {
  44         struct ctdb_context *ctdb;
  45         struct event_script_callback *callback;
  46         pid_t child;
  47         int fd[2];
  48         enum ctdb_eventscript_call call;
  49         const char *options;
  50         struct timeval timeout;
  51
  52         unsigned int current;
  53         struct ctdb_scripts_wire *scripts;
  54 };
  55
  56 static struct ctdb_script_wire *get_current_script(struct ctdb_event_script_state *state)
  57 {
  58         return &state->scripts->scripts[state->current];
  59 }
  60
  61 /* called from ctdb_logging when we have received output on STDERR from
  62  * one of the eventscripts
  63  */
  64 static void log_event_script_output(const char *str, uint16_t len, void *p)
  65 {
  66         struct ctdb_event_script_state *state
  67                 = talloc_get_type(p, struct ctdb_event_script_state);
  68         struct ctdb_script_wire *current;
  69         unsigned int slen, min;
  70
  71         /* We may have been aborted to run something else.  Discard */
  72         if (state->scripts == NULL) {
  73                 return;
  74         }
  75
  76         current = get_current_script(state);
  77
  78         /* Append, but don't overfill buffer.  It starts zero-filled. */
  79         slen = strlen(current->output);
  80         min = MIN(len, sizeof(current->output) - slen - 1);
  81
  82         memcpy(current->output + slen, str, min);
  83 }
  84
  85 int32_t ctdb_control_get_event_script_status(struct ctdb_context *ctdb,
  86                                              uint32_t call_type,
  87                                              TDB_DATA *outdata)
  88 {
  89         if (call_type >= CTDB_EVENT_MAX) {
  90                 return -1;
  91         }
  92
  93         if (ctdb->last_status[call_type] == NULL) {
  94                 /* If it's never been run, return nothing so they can tell. */
  95                 outdata->dsize = 0;
  96         } else {
  97                 outdata->dsize = talloc_get_size(ctdb->last_status[call_type]);
  98                 outdata->dptr  = (uint8_t *)ctdb->last_status[call_type];
  99         }
 100         return 0;
 101 }
 102
 103 /* To ignore directory entry return 0, else return non-zero */
 104 static int script_filter(const struct dirent *de)
 105 {
 106         int namelen = strlen(de->d_name);
 107
 108         /* Ignore . and .. */
 109         if (namelen < 3) {
 110                 return 0;
 111         }
 112
 113         /* Skip temporary files left behind by emacs */
 114         if (de->d_name[namelen-1] == '~') {
 115                 return 0;
 116         }
 117
 118         /* Filename should start with [0-9][0-9]. */
 119         if (!isdigit(de->d_name[0]) || !isdigit(de->d_name[1]) ||
 120             de->d_name[2] != '.') {
 121                 return 0;
 122         }
 123
 124         if (namelen > MAX_SCRIPT_NAME) {
 125                 return 0;
 126         }
 127
 128         return 1;
 129 }
 130
 131 /* Return true if OK, otherwise set errno. */
 132 static bool check_executable(const char *dir, const char *name)
 133 {
 134         char *full;
 135         struct stat st;
 136
 137         full = talloc_asprintf(NULL, "%s/%s", dir, name);
 138         if (!full)
 139                 return false;
 140
 141         if (stat(full, &st) != 0) {
 142                 DEBUG(DEBUG_ERR,("Could not stat event script %s: %s\n",
 143                                  full, strerror(errno)));
 144                 talloc_free(full);
 145                 return false;
 146         }
 147
 148         if (!(st.st_mode & S_IXUSR)) {
 149                 DEBUG(DEBUG_DEBUG,("Event script %s is not executable. Ignoring this event script\n", full));
 150                 errno = ENOEXEC;
 151                 talloc_free(full);
 152                 return false;
 153         }
 154
 155         talloc_free(full);
 156         return true;
 157 }
 158
 159 static struct ctdb_scripts_wire *ctdb_get_script_list(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx)
 160 {
 161         struct dirent **namelist;
 162         struct ctdb_scripts_wire *scripts;
 163         int count;
 164
 165         /* scan all directory entries and insert all valid scripts into the
 166            tree
 167         */
 168         count = scandir(ctdb->event_script_dir, &namelist, script_filter, alphasort);
 169         if (count == -1) {
 170                 DEBUG(DEBUG_CRIT, ("Failed to read event script directory '%s' - %s\n",
 171                                    ctdb->event_script_dir, strerror(errno)));
 172                 return NULL;
 173         }
 174
 175         /* Overallocates by one, but that's OK */
 176         scripts = talloc_zero_size(mem_ctx,
 177                                    sizeof(*scripts)
 178                                    + sizeof(scripts->scripts[0]) * count);
 179         if (scripts == NULL) {
 180                 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate scripts\n"));
 181                 free(namelist);
 182                 return NULL;
 183         }
 184         scripts->num_scripts = count;
 185
 186         for (count = 0; count < scripts->num_scripts; count++) {
 187                 strcpy(scripts->scripts[count].name, namelist[count]->d_name);
 188                 scripts->scripts[count].status = 0;
 189                 if (!check_executable(ctdb->event_script_dir, namelist[count]->d_name)) {
 190                         scripts->scripts[count].status = -errno;
 191                 }
 192         }
 193
 194         free(namelist);
 195         return scripts;
 196 }
 197
 198
 199 /* There cannot be more than 10 arguments to command helper. */
 200 #define MAX_HELPER_ARGS         (10)
 201
 202 static bool child_helper_args(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
 203                               enum ctdb_eventscript_call call,
 204                               const char *options,
 205                               struct ctdb_script_wire *current, int fd,
 206                               int *argc, const char ***argv)
 207 {
 208         const char **tmp;
 209         int n, i;
 210         char *t, *saveptr, *opt;
 211
 212         tmp = talloc_array(mem_ctx, const char *, 10+1);
 213         if (tmp == NULL)  goto failed;
 214
 215         tmp[0] = talloc_asprintf(tmp, "%d", fd);
 216         tmp[1] = talloc_asprintf(tmp, "%s/%s", ctdb->event_script_dir, current->name);
 217         tmp[2] = talloc_asprintf(tmp, "%s", ctdb_eventscript_call_names[call]);
 218         n = 3;
 219
 220         /* Split options into individual arguments */
 221         opt = talloc_strdup(mem_ctx, options);
 222         if (opt == NULL) {
 223                 goto failed;
 224         }
 225
 226         t = strtok_r(opt, " ", &saveptr);
 227         while (t != NULL) {
 228                 tmp[n++] = talloc_strdup(tmp, t);
 229                 if (n > MAX_HELPER_ARGS) {
 230                         goto args_failed;
 231                 }
 232                 t = strtok_r(NULL, " ", &saveptr);
 233         }
 234
 235         for (i=0; i<n; i++) {
 236                 if (tmp[i] == NULL) {
 237                         goto failed;
 238                 }
 239         }
 240
 241         /* Last argument should be NULL */
 242         tmp[n++] = NULL;
 243
 244         *argc = n;
 245         *argv = tmp;
 246         return true;
 247
 248
 249 args_failed:
 250         DEBUG(DEBUG_ERR, (__location__ " too many arguments '%s' to eventscript '%s'\n",
 251                           options, ctdb_eventscript_call_names[call]));
 252
 253 failed:
 254         if (tmp) {
 255                 talloc_free(tmp);
 256         }
 257         return false;
 258
 259 }
 260
 261 static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
 262                                       uint16_t flags, void *p);
 263
 264 static const char *helper_prog = NULL;
 265
 266 static int fork_child_for_script(struct ctdb_context *ctdb,
 267                                  struct ctdb_event_script_state *state)
 268 {
 269         int r;
 270         struct tevent_fd *fde;
 271         struct ctdb_script_wire *current = get_current_script(state);
 272         int argc;
 273         const char **argv;
 274         static const char *helper = BINDIR "/ctdb_event_helper";
 275
 276         if (helper_prog == NULL) {
 277                 const char *t = getenv("CTDB_EVENT_HELPER");
 278                 if (t != NULL) {
 279                         helper_prog = t;
 280                 } else {
 281                         helper_prog = helper;
 282                 }
 283         }
 284
 285         current->start = timeval_current();
 286
 287         r = pipe(state->fd);
 288         if (r != 0) {
 289                 DEBUG(DEBUG_ERR, (__location__ " pipe failed for child eventscript process\n"));
 290                 return -errno;
 291         }
 292
 293         /* Arguments for helper */
 294         if (!child_helper_args(state, ctdb, state->call, state->options, current,
 295                                state->fd[1], &argc, &argv)) {
 296                 DEBUG(DEBUG_ERR, (__location__ " failed to create arguments for eventscript helper\n"));
 297                 r = -ENOMEM;
 298                 close(state->fd[0]);
 299                 close(state->fd[1]);
 300                 return r;
 301         }
 302
 303         if (!ctdb_vfork_with_logging(state, ctdb, current->name,
 304                                      helper_prog, argc, argv,
 305                                      log_event_script_output,
 306                                      state, &state->child)) {
 307                 talloc_free(argv);
 308                 r = -errno;
 309                 close(state->fd[0]);
 310                 close(state->fd[1]);
 311                 return r;
 312         }
 313
 314         talloc_free(argv);
 315
 316         close(state->fd[1]);
 317         set_close_on_exec(state->fd[0]);
 318
 319         /* Set ourselves up to be called when that's done. */
 320         fde = event_add_fd(ctdb->ev, state, state->fd[0], EVENT_FD_READ,
 321                            ctdb_event_script_handler, state);
 322         tevent_fd_set_auto_close(fde);
 323
 324         return 0;
 325 }
 326
 327 /*
 328  Summarize status of this run of scripts.
 329  */
 330 static int script_status(struct ctdb_scripts_wire *scripts)
 331 {
 332         unsigned int i;
 333
 334         for (i = 0; i < scripts->num_scripts; i++) {
 335                 switch (scripts->scripts[i].status) {
 336                 case -ENOENT:
 337                 case -ENOEXEC:
 338                         /* Disabled or missing; that's OK. */
 339                         break;
 340                 case 0:
 341                         /* No problem. */
 342                         break;
 343                 default:
 344                         return scripts->scripts[i].status;
 345                 }
 346         }
 347
 348         /* All OK! */
 349         return 0;
 350 }
 351
 352 /* called when child is finished */
 353 static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
 354                                       uint16_t flags, void *p)
 355 {
 356         struct ctdb_event_script_state *state =
 357                 talloc_get_type(p, struct ctdb_event_script_state);
 358         struct ctdb_script_wire *current = get_current_script(state);
 359         struct ctdb_context *ctdb = state->ctdb;
 360         int r, status;
 361
 362         if (ctdb == NULL) {
 363                 DEBUG(DEBUG_ERR,("Eventscript finished but ctdb is NULL\n"));
 364                 return;
 365         }
 366
 367         r = read(state->fd[0], &current->status, sizeof(current->status));
 368         if (r < 0) {
 369                 current->status = -errno;
 370         } else if (r != sizeof(current->status)) {
 371                 current->status = -EIO;
 372         }
 373
 374         current->finished = timeval_current();
 375         /* valgrind gets overloaded if we run next script as it's still doing
 376          * post-execution analysis, so kill finished child here. */
 377         if (ctdb->valgrinding) {
 378                 ctdb_kill(ctdb, state->child, SIGKILL);
 379         }
 380
 381         state->child = 0;
 382
 383         status = script_status(state->scripts);
 384
 385         /* Aborted or finished all scripts?  We're done. */
 386         if (status != 0 || state->current+1 == state->scripts->num_scripts) {
 387                 DEBUG(DEBUG_INFO,(__location__ " Eventscript %s %s finished with state %d\n",
 388                                   ctdb_eventscript_call_names[state->call], state->options, status));
 389
 390                 ctdb->event_script_timeouts = 0;
 391                 talloc_free(state);
 392                 return;
 393         }
 394
 395         /* Forget about that old fd. */
 396         talloc_free(fde);
 397
 398         /* Next script! */
 399         state->current++;
 400         current++;
 401         current->status = fork_child_for_script(ctdb, state);
 402         if (current->status != 0) {
 403                 /* This calls the callback. */
 404                 talloc_free(state);
 405         }
 406 }
 407
 408 struct debug_hung_script_state {
 409         struct ctdb_context *ctdb;
 410         pid_t child;
 411         enum ctdb_eventscript_call call;
 412 };
 413
 414 static int debug_hung_script_state_destructor(struct debug_hung_script_state *state)
 415 {
 416         if (state->child) {
 417                 ctdb_kill(state->ctdb, state->child, SIGKILL);
 418         }
 419         return 0;
 420 }
 421
 422 static void debug_hung_script_timeout(struct tevent_context *ev, struct tevent_timer *te,
 423                                       struct timeval t, void *p)
 424 {
 425         struct debug_hung_script_state *state =
 426                 talloc_get_type(p, struct debug_hung_script_state);
 427
 428         talloc_free(state);
 429 }
 430
 431 static void debug_hung_script_done(struct tevent_context *ev, struct tevent_fd *fde,
 432                                    uint16_t flags, void *p)
 433 {
 434         struct debug_hung_script_state *state =
 435                 talloc_get_type(p, struct debug_hung_script_state);
 436
 437         talloc_free(state);
 438 }
 439
 440 static void ctdb_run_debug_hung_script(struct ctdb_context *ctdb, struct debug_hung_script_state *state)
 441 {
 442         pid_t pid;
 443         const char * debug_hung_script = CTDB_ETCDIR "/debug-hung-script.sh";
 444         int fd[2];
 445         struct tevent_timer *ttimer;
 446         struct tevent_fd *tfd;
 447         const char **argv;
 448         int i;
 449
 450         if (helper_prog == NULL) {
 451                 return;
 452         }
 453
 454         if (pipe(fd) < 0) {
 455                 DEBUG(DEBUG_ERR,("Failed to create pipe fd for debug hung script\n"));
 456                 return;
 457         }
 458
 459         if (getenv("CTDB_DEBUG_HUNG_SCRIPT") != NULL) {
 460                 debug_hung_script = getenv("CTDB_DEBUG_HUNG_SCRIPT");
 461         }
 462
 463         argv = talloc_array(state, const char *, 5);
 464
 465         argv[0] = talloc_asprintf(argv, "%d", fd[1]);
 466         argv[1] = talloc_strdup(argv, debug_hung_script);
 467         argv[2] = talloc_asprintf(argv, "%d", state->child);
 468         argv[3] = talloc_strdup(argv, ctdb_eventscript_call_names[state->call]);
 469         argv[4] = NULL;
 470
 471         for (i=0; i<4; i++) {
 472                 if (argv[i] == NULL) {
 473                         close(fd[0]);
 474                         close(fd[1]);
 475                         talloc_free(argv);
 476                         return;
 477                 }
 478         }
 479
 480
 481         if (!ctdb_vfork_with_logging(state, ctdb, "Hung-script",
 482                                      helper_prog, 5, argv, NULL, NULL, &pid)) {
 483                 DEBUG(DEBUG_ERR,("Failed to fork a child to track hung event script\n"));
 484                 talloc_free(argv);
 485                 close(fd[0]);
 486                 close(fd[1]);
 487                 return;
 488         }
 489
 490         talloc_free(argv);
 491         close(fd[1]);
 492
 493         ttimer = tevent_add_timer(ctdb->ev, state,
 494                                   timeval_current_ofs(ctdb->tunable.script_timeout, 0),
 495                                   debug_hung_script_timeout, state);
 496         if (ttimer == NULL) {
 497                 close(fd[0]);
 498                 return;
 499         }
 500
 501         tfd = tevent_add_fd(ctdb->ev, state, fd[0], EVENT_FD_READ,
 502                             debug_hung_script_done, state);
 503         if (tfd == NULL) {
 504                 talloc_free(ttimer);
 505                 close(fd[0]);
 506                 return;
 507         }
 508         tevent_fd_set_auto_close(tfd);
 509 }
 510
 511 /* called when child times out */
 512 static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te,
 513                                       struct timeval t, void *p)
 514 {
 515         struct ctdb_event_script_state *state = talloc_get_type(p, struct ctdb_event_script_state);
 516         struct ctdb_context *ctdb = state->ctdb;
 517         struct ctdb_script_wire *current = get_current_script(state);
 518         struct debug_hung_script_state *debug_state;
 519
 520         DEBUG(DEBUG_ERR,("Event script '%s %s %s' timed out after %.1fs, count: %u, pid: %d\n",
 521                          current->name, ctdb_eventscript_call_names[state->call], state->options,
 522                          timeval_elapsed(&current->start),
 523                          ctdb->event_script_timeouts, state->child));
 524
 525         /* ignore timeouts for these events */
 526         switch (state->call) {
 527         case CTDB_EVENT_START_RECOVERY:
 528         case CTDB_EVENT_RECOVERED:
 529         case CTDB_EVENT_TAKE_IP:
 530         case CTDB_EVENT_RELEASE_IP:
 531                 state->scripts->scripts[state->current].status = 0;
 532                 DEBUG(DEBUG_ERR,("Ignoring hung script for %s call %d\n", state->options, state->call));
 533                 break;
 534         default:
 535                 state->scripts->scripts[state->current].status = -ETIME;
 536         }
 537
 538         debug_state = talloc_zero(ctdb, struct debug_hung_script_state);
 539         if (debug_state == NULL) {
 540                 talloc_free(state);
 541                 return;
 542         }
 543
 544         /* Save information useful for running debug hung script, so
 545          * eventscript state can be freed.
 546          */
 547         debug_state->ctdb = ctdb;
 548         debug_state->child = state->child;
 549         debug_state->call = state->call;
 550
 551         /* This destructor will actually kill the hung event script */
 552         talloc_set_destructor(debug_state, debug_hung_script_state_destructor);
 553
 554         state->child = 0;
 555         talloc_free(state);
 556
 557         ctdb_run_debug_hung_script(ctdb, debug_state);
 558 }
 559
 560 /*
 561   destroy an event script: kill it if ->child != 0.
 562  */
 563 static int event_script_destructor(struct ctdb_event_script_state *state)
 564 {
 565         int status;
 566         struct event_script_callback *callback;
 567
 568         if (state->child) {
 569                 DEBUG(DEBUG_ERR,(__location__ " Sending SIGTERM to child pid:%d\n", state->child));
 570
 571                 if (ctdb_kill(state->ctdb, state->child, SIGTERM) != 0) {
 572                         DEBUG(DEBUG_ERR,("Failed to kill child process for eventscript, errno %s(%d)\n", strerror(errno), errno));
 573                 }
 574         }
 575
 576         /* If we were the current monitor, we no longer are. */
 577         if (state->ctdb->current_monitor == state) {
 578                 state->ctdb->current_monitor = NULL;
 579         }
 580
 581         /* Save our scripts as the last executed status, if we have them.
 582          * See ctdb_event_script_callback_v where we abort monitor event. */
 583         if (state->scripts) {
 584                 talloc_free(state->ctdb->last_status[state->call]);
 585                 state->ctdb->last_status[state->call] = state->scripts;
 586                 if (state->current < state->ctdb->last_status[state->call]->num_scripts) {
 587                         state->ctdb->last_status[state->call]->num_scripts = state->current+1;
 588                 }
 589         }
 590
 591         /* Use last status as result, or "OK" if none. */
 592         if (state->ctdb->last_status[state->call]) {
 593                 status = script_status(state->ctdb->last_status[state->call]);
 594         } else {
 595                 status = 0;
 596         }
 597
 598         state->ctdb->active_events--;
 599         if (state->ctdb->active_events < 0) {
 600                 ctdb_fatal(state->ctdb, "Active events < 0");
 601         }
 602
 603         /* This is allowed to free us; talloc will prevent double free anyway,
 604          * but beware if you call this outside the destructor!
 605          * the callback hangs off a different context so we walk the list
 606          * of "active" callbacks until we find the one state points to.
 607          * if we cant find it it means the callback has been removed.
 608          */
 609         for (callback = state->ctdb->script_callbacks; callback != NULL; callback = callback->next) {
 610                 if (callback == state->callback) {
 611                         break;
 612                 }
 613         }
 614
 615         state->callback = NULL;
 616
 617         if (callback) {
 618                 /* Make sure destructor doesn't free itself! */
 619                 talloc_steal(NULL, callback);
 620                 callback->fn(state->ctdb, status, callback->private_data);
 621                 talloc_free(callback);
 622         }
 623
 624         return 0;
 625 }
 626
 627 static unsigned int count_words(const char *options)
 628 {
 629         unsigned int words = 0;
 630
 631         options += strspn(options, " \t");
 632         while (*options) {
 633                 words++;
 634                 options += strcspn(options, " \t");
 635                 options += strspn(options, " \t");
 636         }
 637         return words;
 638 }
 639
 640 static bool check_options(enum ctdb_eventscript_call call, const char *options)
 641 {
 642         switch (call) {
 643         /* These all take no arguments. */
 644         case CTDB_EVENT_INIT:
 645         case CTDB_EVENT_SETUP:
 646         case CTDB_EVENT_STARTUP:
 647         case CTDB_EVENT_START_RECOVERY:
 648         case CTDB_EVENT_RECOVERED:
 649         case CTDB_EVENT_MONITOR:
 650         case CTDB_EVENT_SHUTDOWN:
 651         case CTDB_EVENT_IPREALLOCATED:
 652                 return count_words(options) == 0;
 653
 654         case CTDB_EVENT_TAKE_IP: /* interface, IP address, netmask bits. */
 655         case CTDB_EVENT_RELEASE_IP:
 656                 return count_words(options) == 3;
 657
 658         case CTDB_EVENT_UPDATE_IP: /* old interface, new interface, IP address, netmask bits. */
 659                 return count_words(options) == 4;
 660
 661         default:
 662                 DEBUG(DEBUG_ERR,(__location__ "Unknown ctdb_eventscript_call %u\n", call));
 663                 return false;
 664         }
 665 }
 666
 667 static int remove_callback(struct event_script_callback *callback)
 668 {
 669         DLIST_REMOVE(callback->ctdb->script_callbacks, callback);
 670         return 0;
 671 }
 672
 673 /*
 674   run the event script in the background, calling the callback when
 675   finished
 676  */
 677 static int ctdb_event_script_callback_v(struct ctdb_context *ctdb,
 678                                         const void *mem_ctx,
 679                                         void (*callback)(struct ctdb_context *, int, void *),
 680                                         void *private_data,
 681                                         enum ctdb_eventscript_call call,
 682                                         const char *fmt, va_list ap)
 683 {
 684         struct ctdb_event_script_state *state;
 685
 686         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 687                 /* we guarantee that only some specifically allowed event scripts are run
 688                    while in recovery */
 689                 const enum ctdb_eventscript_call allowed_calls[] = {
 690                         CTDB_EVENT_INIT,
 691                         CTDB_EVENT_SETUP,
 692                         CTDB_EVENT_START_RECOVERY,
 693                         CTDB_EVENT_SHUTDOWN,
 694                         CTDB_EVENT_RELEASE_IP,
 695                         CTDB_EVENT_IPREALLOCATED,
 696                 };
 697                 int i;
 698                 for (i=0;i<ARRAY_SIZE(allowed_calls);i++) {
 699                         if (call == allowed_calls[i]) break;
 700                 }
 701                 if (i == ARRAY_SIZE(allowed_calls)) {
 702                         DEBUG(DEBUG_ERR,("Refusing to run event scripts call '%s' while in recovery\n",
 703                                  ctdb_eventscript_call_names[call]));
 704                         return -1;
 705                 }
 706         }
 707
 708         /* Do not run new monitor events if some event is already running */
 709         if (call == CTDB_EVENT_MONITOR && ctdb->active_events > 0) {
 710                 if (callback != NULL) {
 711                         callback(ctdb, -ECANCELED, private_data);
 712                 }
 713                 return 0;
 714         }
 715
 716         /* Kill off any running monitor events to run this event. */
 717         if (ctdb->current_monitor) {
 718                 struct ctdb_event_script_state *ms = talloc_get_type(ctdb->current_monitor, struct ctdb_event_script_state);
 719
 720                 /* Cancel current monitor callback state only if monitoring
 721                  * context ctdb->monitor->monitor_context has not been freed */
 722                 if (ms->callback != NULL && !ctdb_stopped_monitoring(ctdb)) {
 723                         ms->callback->fn(ctdb, -ECANCELED, ms->callback->private_data);
 724                         talloc_free(ms->callback);
 725                 }
 726
 727                 /* Discard script status so we don't save to last_status */
 728                 talloc_free(ctdb->current_monitor->scripts);
 729                 ctdb->current_monitor->scripts = NULL;
 730                 talloc_free(ctdb->current_monitor);
 731                 ctdb->current_monitor = NULL;
 732         }
 733
 734         state = talloc(ctdb->event_script_ctx, struct ctdb_event_script_state);
 735         CTDB_NO_MEMORY(ctdb, state);
 736
 737         /* The callback isn't done if the context is freed. */
 738         state->callback = talloc(mem_ctx, struct event_script_callback);
 739         CTDB_NO_MEMORY(ctdb, state->callback);
 740         DLIST_ADD(ctdb->script_callbacks, state->callback);
 741         talloc_set_destructor(state->callback, remove_callback);
 742         state->callback->ctdb         = ctdb;
 743         state->callback->fn           = callback;
 744         state->callback->private_data = private_data;
 745
 746         state->ctdb = ctdb;
 747         state->call = call;
 748         state->options = talloc_vasprintf(state, fmt, ap);
 749         state->timeout = timeval_set(ctdb->tunable.script_timeout, 0);
 750         state->scripts = NULL;
 751         if (state->options == NULL) {
 752                 DEBUG(DEBUG_ERR, (__location__ " could not allocate state->options\n"));
 753                 talloc_free(state);
 754                 return -1;
 755         }
 756         if (!check_options(state->call, state->options)) {
 757                 DEBUG(DEBUG_ERR, ("Bad eventscript options '%s' for '%s'\n",
 758                                   state->options,
 759                                   ctdb_eventscript_call_names[state->call]));
 760                 talloc_free(state);
 761                 return -1;
 762         }
 763
 764         DEBUG(DEBUG_INFO,(__location__ " Starting eventscript %s %s\n",
 765                           ctdb_eventscript_call_names[state->call],
 766                           state->options));
 767
 768         /* This is not a child of state, since we save it in destructor. */
 769         state->scripts = ctdb_get_script_list(ctdb, ctdb);
 770         if (state->scripts == NULL) {
 771                 talloc_free(state);
 772                 return -1;
 773         }
 774         state->current = 0;
 775         state->child = 0;
 776
 777         if (call == CTDB_EVENT_MONITOR) {
 778                 ctdb->current_monitor = state;
 779         }
 780
 781         talloc_set_destructor(state, event_script_destructor);
 782
 783         ctdb->active_events++;
 784
 785         /* Nothing to do? */
 786         if (state->scripts->num_scripts == 0) {
 787                 talloc_free(state);
 788                 return 0;
 789         }
 790
 791         state->scripts->scripts[0].status = fork_child_for_script(ctdb, state);
 792         if (state->scripts->scripts[0].status != 0) {
 793                 /* Callback is called from destructor, with fail result. */
 794                 talloc_free(state);
 795                 return 0;
 796         }
 797
 798         if (!timeval_is_zero(&state->timeout)) {
 799                 event_add_timed(ctdb->ev, state, timeval_current_ofs(state->timeout.tv_sec, state->timeout.tv_usec), ctdb_event_script_timeout, state);
 800         } else {
 801                 DEBUG(DEBUG_ERR, (__location__ " eventscript %s %s called with no timeout\n",
 802                                   ctdb_eventscript_call_names[state->call],
 803                                   state->options));
 804         }
 805
 806         return 0;
 807 }
 808
 809
 810 /*
 811   run the event script in the background, calling the callback when
 812   finished.  If mem_ctx is freed, callback will never be called.
 813  */
 814 int ctdb_event_script_callback(struct ctdb_context *ctdb,
 815                                TALLOC_CTX *mem_ctx,
 816                                void (*callback)(struct ctdb_context *, int, void *),
 817                                void *private_data,
 818                                enum ctdb_eventscript_call call,
 819                                const char *fmt, ...)
 820 {
 821         va_list ap;
 822         int ret;
 823
 824         va_start(ap, fmt);
 825         ret = ctdb_event_script_callback_v(ctdb, mem_ctx, callback, private_data, call, fmt, ap);
 826         va_end(ap);
 827
 828         return ret;
 829 }
 830
 831
 832 struct callback_status {
 833         bool done;
 834         int status;
 835 };
 836
 837 /*
 838   called when ctdb_event_script() finishes
 839  */
 840 static void event_script_callback(struct ctdb_context *ctdb, int status, void *private_data)
 841 {
 842         struct callback_status *s = (struct callback_status *)private_data;
 843         s->done = true;
 844         s->status = status;
 845 }
 846
 847 /*
 848   run the event script, waiting for it to complete. Used when the caller
 849   doesn't want to continue till the event script has finished.
 850  */
 851 int ctdb_event_script_args(struct ctdb_context *ctdb, enum ctdb_eventscript_call call,
 852                            const char *fmt, ...)
 853 {
 854         va_list ap;
 855         int ret;
 856         struct callback_status status;
 857
 858         va_start(ap, fmt);
 859         ret = ctdb_event_script_callback_v(ctdb, ctdb,
 860                         event_script_callback, &status, call, fmt, ap);
 861         va_end(ap);
 862         if (ret != 0) {
 863                 return ret;
 864         }
 865
 866         status.status = -1;
 867         status.done = false;
 868
 869         while (status.done == false && event_loop_once(ctdb->ev) == 0) /* noop */;
 870
 871         if (status.status == -ETIME) {
 872                 DEBUG(DEBUG_ERR, (__location__ " eventscript for '%s' timedout."
 873                                   " Immediately banning ourself for %d seconds\n",
 874                                   ctdb_eventscript_call_names[call],
 875                                   ctdb->tunable.recovery_ban_period));
 876
 877                 /* Don't ban self if CTDB is starting up or shutting down */
 878                 if (call != CTDB_EVENT_INIT && call != CTDB_EVENT_SHUTDOWN) {
 879                         ctdb_ban_self(ctdb);
 880                 }
 881         }
 882
 883         return status.status;
 884 }
 885
 886 int ctdb_event_script(struct ctdb_context *ctdb, enum ctdb_eventscript_call call)
 887 {
 888         /* GCC complains about empty format string, so use %s and "". */
 889         return ctdb_event_script_args(ctdb, call, "%s", "");
 890 }
 891
 892 struct eventscript_callback_state {
 893         struct ctdb_req_control *c;
 894 };
 895
 896 /*
 897   called when a forced eventscript run has finished
 898  */
 899 static void run_eventscripts_callback(struct ctdb_context *ctdb, int status,
 900                                  void *private_data)
 901 {
 902         struct eventscript_callback_state *state =
 903                 talloc_get_type(private_data, struct eventscript_callback_state);
 904
 905         if (status != 0) {
 906                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts\n"));
 907         }
 908
 909         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
 910         /* This will free the struct ctdb_event_script_state we are in! */
 911         talloc_free(state);
 912         return;
 913 }
 914
 915
 916 /* Returns rest of string, or NULL if no match. */
 917 static const char *get_call(const char *p, enum ctdb_eventscript_call *call)
 918 {
 919         unsigned int len;
 920
 921         /* Skip any initial whitespace. */
 922         p += strspn(p, " \t");
 923
 924         /* See if we match any. */
 925         for (*call = 0; *call < CTDB_EVENT_MAX; (*call)++) {
 926                 len = strlen(ctdb_eventscript_call_names[*call]);
 927                 if (strncmp(p, ctdb_eventscript_call_names[*call], len) == 0) {
 928                         /* If end of string or whitespace, we're done. */
 929                         if (strcspn(p + len, " \t") == 0) {
 930                                 return p + len;
 931                         }
 932                 }
 933         }
 934         return NULL;
 935 }
 936
 937 /*
 938   A control to force running of the eventscripts from the ctdb client tool
 939 */
 940 int32_t ctdb_run_eventscripts(struct ctdb_context *ctdb,
 941                 struct ctdb_req_control *c,
 942                 TDB_DATA indata, bool *async_reply)
 943 {
 944         int ret;
 945         struct eventscript_callback_state *state;
 946         const char *options;
 947         enum ctdb_eventscript_call call;
 948
 949         /* Figure out what call they want. */
 950         options = get_call((const char *)indata.dptr, &call);
 951         if (!options) {
 952                 DEBUG(DEBUG_ERR, (__location__ " Invalid event name \"%s\"\n", (const char *)indata.dptr));
 953                 return -1;
 954         }
 955
 956         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 957                 DEBUG(DEBUG_ERR, (__location__ " Aborted running eventscript \"%s\" while in RECOVERY mode\n", indata.dptr));
 958                 return -1;
 959         }
 960
 961         state = talloc(ctdb->event_script_ctx, struct eventscript_callback_state);
 962         CTDB_NO_MEMORY(ctdb, state);
 963
 964         state->c = talloc_steal(state, c);
 965
 966         DEBUG(DEBUG_NOTICE,("Running eventscripts with arguments %s\n", indata.dptr));
 967
 968         ret = ctdb_event_script_callback(ctdb,
 969                          state, run_eventscripts_callback, state,
 970                          call, "%s", options);
 971
 972         if (ret != 0) {
 973                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts with arguments %s\n", indata.dptr));
 974                 talloc_free(state);
 975                 return -1;
 976         }
 977
 978         /* tell ctdb_control.c that we will be replying asynchronously */
 979         *async_reply = true;
 980
 981         return 0;
 982 }
 983
 984
 985
 986 int32_t ctdb_control_enable_script(struct ctdb_context *ctdb, TDB_DATA indata)
 987 {
 988         const char *script;
 989         struct stat st;
 990         char *filename;
 991         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
 992
 993         script = (char *)indata.dptr;
 994         if (indata.dsize == 0) {
 995                 DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
 996                 talloc_free(tmp_ctx);
 997                 return -1;
 998         }
 999         if (indata.dptr[indata.dsize - 1] != '\0') {
1000                 DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
1001                 talloc_free(tmp_ctx);
1002                 return -1;
1003         }
1004         if (index(script,'/') != NULL) {
1005                 DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to enable script %s\n", script));
1006                 talloc_free(tmp_ctx);
1007                 return -1;
1008         }
1009
1010
1011         if (stat(ctdb->event_script_dir, &st) != 0 &&
1012             errno == ENOENT) {
1013                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
1014                 talloc_free(tmp_ctx);
1015                 return -1;
1016         }
1017
1018
1019         filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
1020         if (filename == NULL) {
1021                 DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
1022                 talloc_free(tmp_ctx);
1023                 return -1;
1024         }
1025
1026         if (stat(filename, &st) != 0) {
1027                 DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to enable script.\n", filename));
1028                 talloc_free(tmp_ctx);
1029                 return -1;
1030         }
1031
1032         if (chmod(filename, st.st_mode | S_IXUSR) == -1) {
1033                 DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to enable script.\n", filename));
1034                 talloc_free(tmp_ctx);
1035                 return -1;
1036         }
1037
1038         talloc_free(tmp_ctx);
1039         return 0;
1040 }
1041
1042 int32_t ctdb_control_disable_script(struct ctdb_context *ctdb, TDB_DATA indata)
1043 {
1044         const char *script;
1045         struct stat st;
1046         char *filename;
1047         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1048
1049         script = (char *)indata.dptr;
1050         if (indata.dsize == 0) {
1051                 DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
1052                 talloc_free(tmp_ctx);
1053                 return -1;
1054         }
1055         if (indata.dptr[indata.dsize - 1] != '\0') {
1056                 DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
1057                 talloc_free(tmp_ctx);
1058                 return -1;
1059         }
1060         if (index(script,'/') != NULL) {
1061                 DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to disable script %s\n", script));
1062                 talloc_free(tmp_ctx);
1063                 return -1;
1064         }
1065
1066
1067         if (stat(ctdb->event_script_dir, &st) != 0 &&
1068             errno == ENOENT) {
1069                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
1070                 talloc_free(tmp_ctx);
1071                 return -1;
1072         }
1073
1074
1075         filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
1076         if (filename == NULL) {
1077                 DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
1078                 talloc_free(tmp_ctx);
1079                 return -1;
1080         }
1081
1082         if (stat(filename, &st) != 0) {
1083                 DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to disable script.\n", filename));
1084                 talloc_free(tmp_ctx);
1085                 return -1;
1086         }
1087
1088         if (chmod(filename, st.st_mode & ~(S_IXUSR|S_IXGRP|S_IXOTH)) == -1) {
1089                 DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to disable script.\n", filename));
1090                 talloc_free(tmp_ctx);
1091                 return -1;
1092         }
1093
1094         talloc_free(tmp_ctx);
1095         return 0;
1096 }