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 = ETCDIR "/ctdb/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         argv = talloc_array(state, const char *, 5);
 460
 461         argv[0] = talloc_asprintf(argv, "%d", fd[1]);
 462         argv[1] = talloc_strdup(argv, debug_hung_script);
 463         argv[2] = talloc_asprintf(argv, "%d", state->child);
 464         argv[3] = talloc_strdup(argv, ctdb_eventscript_call_names[state->call]);
 465         argv[4] = NULL;
 466
 467         for (i=0; i<4; i++) {
 468                 if (argv[i] == NULL) {
 469                         close(fd[0]);
 470                         close(fd[1]);
 471                         talloc_free(argv);
 472                         return;
 473                 }
 474         }
 475
 476
 477         if (!ctdb_vfork_with_logging(state, ctdb, "Hung-script",
 478                                      helper_prog, 5, argv, NULL, NULL, &pid)) {
 479                 DEBUG(DEBUG_ERR,("Failed to fork a child to track hung event script\n"));
 480                 talloc_free(argv);
 481                 close(fd[0]);
 482                 close(fd[1]);
 483                 return;
 484         }
 485
 486         talloc_free(argv);
 487         close(fd[1]);
 488
 489         ttimer = tevent_add_timer(ctdb->ev, state,
 490                                   timeval_current_ofs(ctdb->tunable.script_timeout, 0),
 491                                   debug_hung_script_timeout, state);
 492         if (ttimer == NULL) {
 493                 close(fd[0]);
 494                 return;
 495         }
 496
 497         tfd = tevent_add_fd(ctdb->ev, state, fd[0], EVENT_FD_READ,
 498                             debug_hung_script_done, state);
 499         if (tfd == NULL) {
 500                 talloc_free(ttimer);
 501                 close(fd[0]);
 502                 return;
 503         }
 504         tevent_fd_set_auto_close(tfd);
 505 }
 506
 507 /* called when child times out */
 508 static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te,
 509                                       struct timeval t, void *p)
 510 {
 511         struct ctdb_event_script_state *state = talloc_get_type(p, struct ctdb_event_script_state);
 512         struct ctdb_context *ctdb = state->ctdb;
 513         struct ctdb_script_wire *current = get_current_script(state);
 514         struct debug_hung_script_state *debug_state;
 515
 516         DEBUG(DEBUG_ERR,("Event script '%s %s %s' timed out after %.1fs, count: %u, pid: %d\n",
 517                          current->name, ctdb_eventscript_call_names[state->call], state->options,
 518                          timeval_elapsed(&current->start),
 519                          ctdb->event_script_timeouts, state->child));
 520
 521         /* ignore timeouts for these events */
 522         switch (state->call) {
 523         case CTDB_EVENT_START_RECOVERY:
 524         case CTDB_EVENT_RECOVERED:
 525         case CTDB_EVENT_TAKE_IP:
 526         case CTDB_EVENT_RELEASE_IP:
 527                 state->scripts->scripts[state->current].status = 0;
 528                 DEBUG(DEBUG_ERR,("Ignoring hung script for %s call %d\n", state->options, state->call));
 529                 break;
 530         default:
 531                 state->scripts->scripts[state->current].status = -ETIME;
 532         }
 533
 534         debug_state = talloc_zero(ctdb, struct debug_hung_script_state);
 535         if (debug_state == NULL) {
 536                 talloc_free(state);
 537                 return;
 538         }
 539
 540         /* Save information useful for running debug hung script, so
 541          * eventscript state can be freed.
 542          */
 543         debug_state->ctdb = ctdb;
 544         debug_state->child = state->child;
 545         debug_state->call = state->call;
 546
 547         /* This destructor will actually kill the hung event script */
 548         talloc_set_destructor(debug_state, debug_hung_script_state_destructor);
 549
 550         state->child = 0;
 551         talloc_free(state);
 552
 553         ctdb_run_debug_hung_script(ctdb, debug_state);
 554 }
 555
 556 /*
 557   destroy an event script: kill it if ->child != 0.
 558  */
 559 static int event_script_destructor(struct ctdb_event_script_state *state)
 560 {
 561         int status;
 562         struct event_script_callback *callback;
 563
 564         if (state->child) {
 565                 DEBUG(DEBUG_ERR,(__location__ " Sending SIGTERM to child pid:%d\n", state->child));
 566
 567                 if (ctdb_kill(state->ctdb, state->child, SIGTERM) != 0) {
 568                         DEBUG(DEBUG_ERR,("Failed to kill child process for eventscript, errno %s(%d)\n", strerror(errno), errno));
 569                 }
 570         }
 571
 572         /* If we were the current monitor, we no longer are. */
 573         if (state->ctdb->current_monitor == state) {
 574                 state->ctdb->current_monitor = NULL;
 575         }
 576
 577         /* Save our scripts as the last executed status, if we have them.
 578          * See ctdb_event_script_callback_v where we abort monitor event. */
 579         if (state->scripts) {
 580                 talloc_free(state->ctdb->last_status[state->call]);
 581                 state->ctdb->last_status[state->call] = state->scripts;
 582                 if (state->current < state->ctdb->last_status[state->call]->num_scripts) {
 583                         state->ctdb->last_status[state->call]->num_scripts = state->current+1;
 584                 }
 585         }
 586
 587         /* Use last status as result, or "OK" if none. */
 588         if (state->ctdb->last_status[state->call]) {
 589                 status = script_status(state->ctdb->last_status[state->call]);
 590         } else {
 591                 status = 0;
 592         }
 593
 594         state->ctdb->active_events--;
 595         if (state->ctdb->active_events < 0) {
 596                 ctdb_fatal(state->ctdb, "Active events < 0");
 597         }
 598
 599         /* This is allowed to free us; talloc will prevent double free anyway,
 600          * but beware if you call this outside the destructor!
 601          * the callback hangs off a different context so we walk the list
 602          * of "active" callbacks until we find the one state points to.
 603          * if we cant find it it means the callback has been removed.
 604          */
 605         for (callback = state->ctdb->script_callbacks; callback != NULL; callback = callback->next) {
 606                 if (callback == state->callback) {
 607                         break;
 608                 }
 609         }
 610
 611         state->callback = NULL;
 612
 613         if (callback) {
 614                 /* Make sure destructor doesn't free itself! */
 615                 talloc_steal(NULL, callback);
 616                 callback->fn(state->ctdb, status, callback->private_data);
 617                 talloc_free(callback);
 618         }
 619
 620         return 0;
 621 }
 622
 623 static unsigned int count_words(const char *options)
 624 {
 625         unsigned int words = 0;
 626
 627         options += strspn(options, " \t");
 628         while (*options) {
 629                 words++;
 630                 options += strcspn(options, " \t");
 631                 options += strspn(options, " \t");
 632         }
 633         return words;
 634 }
 635
 636 static bool check_options(enum ctdb_eventscript_call call, const char *options)
 637 {
 638         switch (call) {
 639         /* These all take no arguments. */
 640         case CTDB_EVENT_INIT:
 641         case CTDB_EVENT_SETUP:
 642         case CTDB_EVENT_STARTUP:
 643         case CTDB_EVENT_START_RECOVERY:
 644         case CTDB_EVENT_RECOVERED:
 645         case CTDB_EVENT_MONITOR:
 646         case CTDB_EVENT_SHUTDOWN:
 647         case CTDB_EVENT_IPREALLOCATED:
 648                 return count_words(options) == 0;
 649
 650         case CTDB_EVENT_TAKE_IP: /* interface, IP address, netmask bits. */
 651         case CTDB_EVENT_RELEASE_IP:
 652                 return count_words(options) == 3;
 653
 654         case CTDB_EVENT_UPDATE_IP: /* old interface, new interface, IP address, netmask bits. */
 655                 return count_words(options) == 4;
 656
 657         default:
 658                 DEBUG(DEBUG_ERR,(__location__ "Unknown ctdb_eventscript_call %u\n", call));
 659                 return false;
 660         }
 661 }
 662
 663 static int remove_callback(struct event_script_callback *callback)
 664 {
 665         DLIST_REMOVE(callback->ctdb->script_callbacks, callback);
 666         return 0;
 667 }
 668
 669 /*
 670   run the event script in the background, calling the callback when
 671   finished
 672  */
 673 static int ctdb_event_script_callback_v(struct ctdb_context *ctdb,
 674                                         const void *mem_ctx,
 675                                         void (*callback)(struct ctdb_context *, int, void *),
 676                                         void *private_data,
 677                                         enum ctdb_eventscript_call call,
 678                                         const char *fmt, va_list ap)
 679 {
 680         struct ctdb_event_script_state *state;
 681
 682         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 683                 /* we guarantee that only some specifically allowed event scripts are run
 684                    while in recovery */
 685                 const enum ctdb_eventscript_call allowed_calls[] = {
 686                         CTDB_EVENT_INIT,
 687                         CTDB_EVENT_SETUP,
 688                         CTDB_EVENT_START_RECOVERY,
 689                         CTDB_EVENT_SHUTDOWN,
 690                         CTDB_EVENT_RELEASE_IP,
 691                         CTDB_EVENT_IPREALLOCATED,
 692                 };
 693                 int i;
 694                 for (i=0;i<ARRAY_SIZE(allowed_calls);i++) {
 695                         if (call == allowed_calls[i]) break;
 696                 }
 697                 if (i == ARRAY_SIZE(allowed_calls)) {
 698                         DEBUG(DEBUG_ERR,("Refusing to run event scripts call '%s' while in recovery\n",
 699                                  ctdb_eventscript_call_names[call]));
 700                         return -1;
 701                 }
 702         }
 703
 704         /* Do not run new monitor events if some event is already running */
 705         if (call == CTDB_EVENT_MONITOR && ctdb->active_events > 0) {
 706                 if (callback != NULL) {
 707                         callback(ctdb, -ECANCELED, private_data);
 708                 }
 709                 return 0;
 710         }
 711
 712         /* Kill off any running monitor events to run this event. */
 713         if (ctdb->current_monitor) {
 714                 struct ctdb_event_script_state *ms = talloc_get_type(ctdb->current_monitor, struct ctdb_event_script_state);
 715
 716                 /* Cancel current monitor callback state only if monitoring
 717                  * context ctdb->monitor->monitor_context has not been freed */
 718                 if (ms->callback != NULL && !ctdb_stopped_monitoring(ctdb)) {
 719                         ms->callback->fn(ctdb, -ECANCELED, ms->callback->private_data);
 720                         talloc_free(ms->callback);
 721                 }
 722
 723                 /* Discard script status so we don't save to last_status */
 724                 talloc_free(ctdb->current_monitor->scripts);
 725                 ctdb->current_monitor->scripts = NULL;
 726                 talloc_free(ctdb->current_monitor);
 727                 ctdb->current_monitor = NULL;
 728         }
 729
 730         state = talloc(ctdb->event_script_ctx, struct ctdb_event_script_state);
 731         CTDB_NO_MEMORY(ctdb, state);
 732
 733         /* The callback isn't done if the context is freed. */
 734         state->callback = talloc(mem_ctx, struct event_script_callback);
 735         CTDB_NO_MEMORY(ctdb, state->callback);
 736         DLIST_ADD(ctdb->script_callbacks, state->callback);
 737         talloc_set_destructor(state->callback, remove_callback);
 738         state->callback->ctdb         = ctdb;
 739         state->callback->fn           = callback;
 740         state->callback->private_data = private_data;
 741
 742         state->ctdb = ctdb;
 743         state->call = call;
 744         state->options = talloc_vasprintf(state, fmt, ap);
 745         state->timeout = timeval_set(ctdb->tunable.script_timeout, 0);
 746         state->scripts = NULL;
 747         if (state->options == NULL) {
 748                 DEBUG(DEBUG_ERR, (__location__ " could not allocate state->options\n"));
 749                 talloc_free(state);
 750                 return -1;
 751         }
 752         if (!check_options(state->call, state->options)) {
 753                 DEBUG(DEBUG_ERR, ("Bad eventscript options '%s' for %s\n",
 754                                   ctdb_eventscript_call_names[state->call], state->options));
 755                 talloc_free(state);
 756                 return -1;
 757         }
 758
 759         DEBUG(DEBUG_INFO,(__location__ " Starting eventscript %s %s\n",
 760                           ctdb_eventscript_call_names[state->call],
 761                           state->options));
 762
 763         /* This is not a child of state, since we save it in destructor. */
 764         state->scripts = ctdb_get_script_list(ctdb, ctdb);
 765         if (state->scripts == NULL) {
 766                 talloc_free(state);
 767                 return -1;
 768         }
 769         state->current = 0;
 770         state->child = 0;
 771
 772         if (call == CTDB_EVENT_MONITOR) {
 773                 ctdb->current_monitor = state;
 774         }
 775
 776         talloc_set_destructor(state, event_script_destructor);
 777
 778         ctdb->active_events++;
 779
 780         /* Nothing to do? */
 781         if (state->scripts->num_scripts == 0) {
 782                 talloc_free(state);
 783                 return 0;
 784         }
 785
 786         state->scripts->scripts[0].status = fork_child_for_script(ctdb, state);
 787         if (state->scripts->scripts[0].status != 0) {
 788                 /* Callback is called from destructor, with fail result. */
 789                 talloc_free(state);
 790                 return 0;
 791         }
 792
 793         if (!timeval_is_zero(&state->timeout)) {
 794                 event_add_timed(ctdb->ev, state, timeval_current_ofs(state->timeout.tv_sec, state->timeout.tv_usec), ctdb_event_script_timeout, state);
 795         } else {
 796                 DEBUG(DEBUG_ERR, (__location__ " eventscript %s %s called with no timeout\n",
 797                                   ctdb_eventscript_call_names[state->call],
 798                                   state->options));
 799         }
 800
 801         return 0;
 802 }
 803
 804
 805 /*
 806   run the event script in the background, calling the callback when
 807   finished.  If mem_ctx is freed, callback will never be called.
 808  */
 809 int ctdb_event_script_callback(struct ctdb_context *ctdb,
 810                                TALLOC_CTX *mem_ctx,
 811                                void (*callback)(struct ctdb_context *, int, void *),
 812                                void *private_data,
 813                                enum ctdb_eventscript_call call,
 814                                const char *fmt, ...)
 815 {
 816         va_list ap;
 817         int ret;
 818
 819         va_start(ap, fmt);
 820         ret = ctdb_event_script_callback_v(ctdb, mem_ctx, callback, private_data, call, fmt, ap);
 821         va_end(ap);
 822
 823         return ret;
 824 }
 825
 826
 827 struct callback_status {
 828         bool done;
 829         int status;
 830 };
 831
 832 /*
 833   called when ctdb_event_script() finishes
 834  */
 835 static void event_script_callback(struct ctdb_context *ctdb, int status, void *private_data)
 836 {
 837         struct callback_status *s = (struct callback_status *)private_data;
 838         s->done = true;
 839         s->status = status;
 840 }
 841
 842 /*
 843   run the event script, waiting for it to complete. Used when the caller
 844   doesn't want to continue till the event script has finished.
 845  */
 846 int ctdb_event_script_args(struct ctdb_context *ctdb, enum ctdb_eventscript_call call,
 847                            const char *fmt, ...)
 848 {
 849         va_list ap;
 850         int ret;
 851         struct callback_status status;
 852
 853         va_start(ap, fmt);
 854         ret = ctdb_event_script_callback_v(ctdb, ctdb,
 855                         event_script_callback, &status, call, fmt, ap);
 856         va_end(ap);
 857         if (ret != 0) {
 858                 return ret;
 859         }
 860
 861         status.status = -1;
 862         status.done = false;
 863
 864         while (status.done == false && event_loop_once(ctdb->ev) == 0) /* noop */;
 865
 866         if (status.status == -ETIME) {
 867                 DEBUG(DEBUG_ERR, (__location__ " eventscript for '%s' timedout."
 868                                   " Immediately banning ourself for %d seconds\n",
 869                                   ctdb_eventscript_call_names[call],
 870                                   ctdb->tunable.recovery_ban_period));
 871
 872                 /* Don't ban self if CTDB is starting up or shutting down */
 873                 if (call != CTDB_EVENT_INIT && call != CTDB_EVENT_SHUTDOWN) {
 874                         ctdb_ban_self(ctdb);
 875                 }
 876         }
 877
 878         return status.status;
 879 }
 880
 881 int ctdb_event_script(struct ctdb_context *ctdb, enum ctdb_eventscript_call call)
 882 {
 883         /* GCC complains about empty format string, so use %s and "". */
 884         return ctdb_event_script_args(ctdb, call, "%s", "");
 885 }
 886
 887 struct eventscript_callback_state {
 888         struct ctdb_req_control *c;
 889 };
 890
 891 /*
 892   called when a forced eventscript run has finished
 893  */
 894 static void run_eventscripts_callback(struct ctdb_context *ctdb, int status,
 895                                  void *private_data)
 896 {
 897         struct eventscript_callback_state *state =
 898                 talloc_get_type(private_data, struct eventscript_callback_state);
 899
 900         ctdb_enable_monitoring(ctdb);
 901
 902         if (status != 0) {
 903                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts\n"));
 904         }
 905
 906         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
 907         /* This will free the struct ctdb_event_script_state we are in! */
 908         talloc_free(state);
 909         return;
 910 }
 911
 912
 913 /* Returns rest of string, or NULL if no match. */
 914 static const char *get_call(const char *p, enum ctdb_eventscript_call *call)
 915 {
 916         unsigned int len;
 917
 918         /* Skip any initial whitespace. */
 919         p += strspn(p, " \t");
 920
 921         /* See if we match any. */
 922         for (*call = 0; *call < CTDB_EVENT_MAX; (*call)++) {
 923                 len = strlen(ctdb_eventscript_call_names[*call]);
 924                 if (strncmp(p, ctdb_eventscript_call_names[*call], len) == 0) {
 925                         /* If end of string or whitespace, we're done. */
 926                         if (strcspn(p + len, " \t") == 0) {
 927                                 return p + len;
 928                         }
 929                 }
 930         }
 931         return NULL;
 932 }
 933
 934 /*
 935   A control to force running of the eventscripts from the ctdb client tool
 936 */
 937 int32_t ctdb_run_eventscripts(struct ctdb_context *ctdb,
 938                 struct ctdb_req_control *c,
 939                 TDB_DATA indata, bool *async_reply)
 940 {
 941         int ret;
 942         struct eventscript_callback_state *state;
 943         const char *options;
 944         enum ctdb_eventscript_call call;
 945
 946         /* Figure out what call they want. */
 947         options = get_call((const char *)indata.dptr, &call);
 948         if (!options) {
 949                 DEBUG(DEBUG_ERR, (__location__ " Invalid event name \"%s\"\n", (const char *)indata.dptr));
 950                 return -1;
 951         }
 952
 953         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 954                 DEBUG(DEBUG_ERR, (__location__ " Aborted running eventscript \"%s\" while in RECOVERY mode\n", indata.dptr));
 955                 return -1;
 956         }
 957
 958         state = talloc(ctdb->event_script_ctx, struct eventscript_callback_state);
 959         CTDB_NO_MEMORY(ctdb, state);
 960
 961         state->c = talloc_steal(state, c);
 962
 963         DEBUG(DEBUG_NOTICE,("Running eventscripts with arguments %s\n", indata.dptr));
 964
 965         ctdb_disable_monitoring(ctdb);
 966
 967         ret = ctdb_event_script_callback(ctdb,
 968                          state, run_eventscripts_callback, state,
 969                          call, "%s", options);
 970
 971         if (ret != 0) {
 972                 ctdb_enable_monitoring(ctdb);
 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 }