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 struct ctdb_script_tree_item {
 104         const char *name;
 105         int error;
 106 };
 107
 108 /* Return true if OK, otherwise set errno. */
 109 static bool check_executable(const char *dir, const char *name)
 110 {
 111         char *full;
 112         struct stat st;
 113
 114         full = talloc_asprintf(NULL, "%s/%s", dir, name);
 115         if (!full)
 116                 return false;
 117
 118         if (stat(full, &st) != 0) {
 119                 DEBUG(DEBUG_ERR,("Could not stat event script %s: %s\n",
 120                                  full, strerror(errno)));
 121                 talloc_free(full);
 122                 return false;
 123         }
 124
 125         if (!(st.st_mode & S_IXUSR)) {
 126                 DEBUG(DEBUG_DEBUG,("Event script %s is not executable. Ignoring this event script\n", full));
 127                 errno = ENOEXEC;
 128                 talloc_free(full);
 129                 return false;
 130         }
 131
 132         talloc_free(full);
 133         return true;
 134 }
 135
 136 static struct ctdb_scripts_wire *ctdb_get_script_list(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx)
 137 {
 138         DIR *dir;
 139         struct dirent *de;
 140         struct stat st;
 141         trbt_tree_t *tree;
 142         struct ctdb_scripts_wire *scripts;
 143         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
 144         struct ctdb_script_tree_item *tree_item;
 145         int count;
 146
 147         /*
 148           the service specific event scripts
 149         */
 150         if (stat(ctdb->event_script_dir, &st) != 0 &&
 151             errno == ENOENT) {
 152                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
 153                 talloc_free(tmp_ctx);
 154                 return NULL;
 155         }
 156
 157         /* create a tree to store all the script names in */
 158         tree = trbt_create(tmp_ctx, 0);
 159
 160         /* scan all directory entries and insert all valid scripts into the
 161            tree
 162         */
 163         dir = opendir(ctdb->event_script_dir);
 164         if (dir == NULL) {
 165                 DEBUG(DEBUG_CRIT,("Failed to open event script directory '%s'\n", ctdb->event_script_dir));
 166                 talloc_free(tmp_ctx);
 167                 return NULL;
 168         }
 169
 170         count = 0;
 171         while ((de=readdir(dir)) != NULL) {
 172                 int namlen;
 173                 unsigned num;
 174
 175                 namlen = strlen(de->d_name);
 176
 177                 if (namlen < 3) {
 178                         continue;
 179                 }
 180
 181                 if (de->d_name[namlen-1] == '~') {
 182                         /* skip files emacs left behind */
 183                         continue;
 184                 }
 185
 186                 if (de->d_name[2] != '.') {
 187                         continue;
 188                 }
 189
 190                 if (sscanf(de->d_name, "%02u.", &num) != 1) {
 191                         continue;
 192                 }
 193
 194                 if (strlen(de->d_name) > MAX_SCRIPT_NAME) {
 195                         DEBUG(DEBUG_ERR,("Script name %s too long! %u chars max",
 196                                          de->d_name, MAX_SCRIPT_NAME));
 197                         continue;
 198                 }
 199
 200                 tree_item = talloc(tree, struct ctdb_script_tree_item);
 201                 if (tree_item == NULL) {
 202                         DEBUG(DEBUG_ERR, (__location__ " Failed to allocate new tree item\n"));
 203                         closedir(dir);
 204                         talloc_free(tmp_ctx);
 205                         return NULL;
 206                 }
 207
 208                 tree_item->error = 0;
 209                 if (!check_executable(ctdb->event_script_dir, de->d_name)) {
 210                         tree_item->error = errno;
 211                 }
 212
 213                 tree_item->name = talloc_strdup(tree_item, de->d_name);
 214                 if (tree_item->name == NULL) {
 215                         DEBUG(DEBUG_ERR,(__location__ " Failed to allocate script name.\n"));
 216                         closedir(dir);
 217                         talloc_free(tmp_ctx);
 218                         return NULL;
 219                 }
 220
 221                 /* store the event script in the tree */
 222                 trbt_insert32(tree, (num<<16)|count++, tree_item);
 223         }
 224         closedir(dir);
 225
 226         /* Overallocates by one, but that's OK */
 227         scripts = talloc_zero_size(tmp_ctx,
 228                                    sizeof(*scripts)
 229                                    + sizeof(scripts->scripts[0]) * count);
 230         if (scripts == NULL) {
 231                 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate scripts\n"));
 232                 talloc_free(tmp_ctx);
 233                 return NULL;
 234         }
 235         scripts->num_scripts = count;
 236
 237         for (count = 0; count < scripts->num_scripts; count++) {
 238                 tree_item = trbt_findfirstarray32(tree, 1);
 239
 240                 strcpy(scripts->scripts[count].name, tree_item->name);
 241                 scripts->scripts[count].status = -tree_item->error;
 242
 243                 /* remove this script from the tree */
 244                 talloc_free(tree_item);
 245         }
 246
 247         talloc_steal(mem_ctx, scripts);
 248         talloc_free(tmp_ctx);
 249         return scripts;
 250 }
 251
 252
 253 /* There cannot be more than 10 arguments to command helper. */
 254 #define MAX_HELPER_ARGS         (10)
 255
 256 static bool child_helper_args(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
 257                               enum ctdb_eventscript_call call,
 258                               const char *options,
 259                               struct ctdb_script_wire *current, int fd,
 260                               int *argc, const char ***argv)
 261 {
 262         const char **tmp;
 263         int n, i;
 264         char *t, *saveptr, *opt;
 265
 266         tmp = talloc_array(mem_ctx, const char *, 10+1);
 267         if (tmp == NULL)  goto failed;
 268
 269         tmp[0] = talloc_asprintf(tmp, "%d", fd);
 270         tmp[1] = talloc_asprintf(tmp, "%s/%s", ctdb->event_script_dir, current->name);
 271         tmp[2] = talloc_asprintf(tmp, "%s", ctdb_eventscript_call_names[call]);
 272         n = 3;
 273
 274         /* Split options into individual arguments */
 275         opt = talloc_strdup(mem_ctx, options);
 276         if (opt == NULL) {
 277                 goto failed;
 278         }
 279
 280         t = strtok_r(opt, " ", &saveptr);
 281         while (t != NULL) {
 282                 tmp[n++] = talloc_strdup(tmp, t);
 283                 if (n > MAX_HELPER_ARGS) {
 284                         goto args_failed;
 285                 }
 286                 t = strtok_r(NULL, " ", &saveptr);
 287         }
 288
 289         for (i=0; i<n; i++) {
 290                 if (tmp[i] == NULL) {
 291                         goto failed;
 292                 }
 293         }
 294
 295         /* Last argument should be NULL */
 296         tmp[n++] = NULL;
 297
 298         *argc = n;
 299         *argv = tmp;
 300         return true;
 301
 302
 303 args_failed:
 304         DEBUG(DEBUG_ERR, (__location__ " too many arguments '%s' to eventscript '%s'\n",
 305                           options, ctdb_eventscript_call_names[call]));
 306
 307 failed:
 308         if (tmp) {
 309                 talloc_free(tmp);
 310         }
 311         return false;
 312
 313 }
 314
 315 static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
 316                                       uint16_t flags, void *p);
 317
 318 static const char *helper_prog = NULL;
 319
 320 static int fork_child_for_script(struct ctdb_context *ctdb,
 321                                  struct ctdb_event_script_state *state)
 322 {
 323         int r;
 324         struct tevent_fd *fde;
 325         struct ctdb_script_wire *current = get_current_script(state);
 326         int argc;
 327         const char **argv;
 328         static const char *helper = BINDIR "/ctdb_event_helper";
 329
 330         if (helper_prog == NULL) {
 331                 const char *t = getenv("CTDB_EVENT_HELPER");
 332                 if (t != NULL) {
 333                         helper_prog = t;
 334                 } else {
 335                         helper_prog = helper;
 336                 }
 337         }
 338
 339         current->start = timeval_current();
 340
 341         r = pipe(state->fd);
 342         if (r != 0) {
 343                 DEBUG(DEBUG_ERR, (__location__ " pipe failed for child eventscript process\n"));
 344                 return -errno;
 345         }
 346
 347         /* Arguments for helper */
 348         if (!child_helper_args(state, ctdb, state->call, state->options, current,
 349                                state->fd[1], &argc, &argv)) {
 350                 DEBUG(DEBUG_ERR, (__location__ " failed to create arguments for eventscript helper\n"));
 351                 r = -ENOMEM;
 352                 close(state->fd[0]);
 353                 close(state->fd[1]);
 354                 return r;
 355         }
 356
 357         if (!ctdb_vfork_with_logging(state, ctdb, current->name,
 358                                      helper_prog, argc, argv,
 359                                      log_event_script_output,
 360                                      state, &state->child)) {
 361                 talloc_free(argv);
 362                 r = -errno;
 363                 close(state->fd[0]);
 364                 close(state->fd[1]);
 365                 return r;
 366         }
 367
 368         talloc_free(argv);
 369
 370         close(state->fd[1]);
 371         set_close_on_exec(state->fd[0]);
 372
 373         /* Set ourselves up to be called when that's done. */
 374         fde = event_add_fd(ctdb->ev, state, state->fd[0], EVENT_FD_READ,
 375                            ctdb_event_script_handler, state);
 376         tevent_fd_set_auto_close(fde);
 377
 378         return 0;
 379 }
 380
 381 /*
 382  Summarize status of this run of scripts.
 383  */
 384 static int script_status(struct ctdb_scripts_wire *scripts)
 385 {
 386         unsigned int i;
 387
 388         for (i = 0; i < scripts->num_scripts; i++) {
 389                 switch (scripts->scripts[i].status) {
 390                 case -ENOENT:
 391                 case -ENOEXEC:
 392                         /* Disabled or missing; that's OK. */
 393                         break;
 394                 case 0:
 395                         /* No problem. */
 396                         break;
 397                 default:
 398                         return scripts->scripts[i].status;
 399                 }
 400         }
 401
 402         /* All OK! */
 403         return 0;
 404 }
 405
 406 /* called when child is finished */
 407 static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
 408                                       uint16_t flags, void *p)
 409 {
 410         struct ctdb_event_script_state *state =
 411                 talloc_get_type(p, struct ctdb_event_script_state);
 412         struct ctdb_script_wire *current = get_current_script(state);
 413         struct ctdb_context *ctdb = state->ctdb;
 414         int r, status;
 415
 416         if (ctdb == NULL) {
 417                 DEBUG(DEBUG_ERR,("Eventscript finished but ctdb is NULL\n"));
 418                 return;
 419         }
 420
 421         r = read(state->fd[0], &current->status, sizeof(current->status));
 422         if (r < 0) {
 423                 current->status = -errno;
 424         } else if (r != sizeof(current->status)) {
 425                 current->status = -EIO;
 426         }
 427
 428         current->finished = timeval_current();
 429         /* valgrind gets overloaded if we run next script as it's still doing
 430          * post-execution analysis, so kill finished child here. */
 431         if (ctdb->valgrinding) {
 432                 ctdb_kill(ctdb, state->child, SIGKILL);
 433         }
 434
 435         state->child = 0;
 436
 437         status = script_status(state->scripts);
 438
 439         /* Aborted or finished all scripts?  We're done. */
 440         if (status != 0 || state->current+1 == state->scripts->num_scripts) {
 441                 DEBUG(DEBUG_INFO,(__location__ " Eventscript %s %s finished with state %d\n",
 442                                   ctdb_eventscript_call_names[state->call], state->options, status));
 443
 444                 ctdb->event_script_timeouts = 0;
 445                 talloc_free(state);
 446                 return;
 447         }
 448
 449         /* Forget about that old fd. */
 450         talloc_free(fde);
 451
 452         /* Next script! */
 453         state->current++;
 454         current++;
 455         current->status = fork_child_for_script(ctdb, state);
 456         if (current->status != 0) {
 457                 /* This calls the callback. */
 458                 talloc_free(state);
 459         }
 460 }
 461
 462 struct debug_hung_script_state {
 463         struct ctdb_context *ctdb;
 464         pid_t child;
 465         enum ctdb_eventscript_call call;
 466 };
 467
 468 static int debug_hung_script_state_destructor(struct debug_hung_script_state *state)
 469 {
 470         if (state->child) {
 471                 ctdb_kill(state->ctdb, state->child, SIGKILL);
 472         }
 473         return 0;
 474 }
 475
 476 static void debug_hung_script_timeout(struct tevent_context *ev, struct tevent_timer *te,
 477                                       struct timeval t, void *p)
 478 {
 479         struct debug_hung_script_state *state =
 480                 talloc_get_type(p, struct debug_hung_script_state);
 481
 482         talloc_free(state);
 483 }
 484
 485 static void debug_hung_script_done(struct tevent_context *ev, struct tevent_fd *fde,
 486                                    uint16_t flags, void *p)
 487 {
 488         struct debug_hung_script_state *state =
 489                 talloc_get_type(p, struct debug_hung_script_state);
 490
 491         talloc_free(state);
 492 }
 493
 494 static void ctdb_run_debug_hung_script(struct ctdb_context *ctdb, struct debug_hung_script_state *state)
 495 {
 496         pid_t pid;
 497         const char * debug_hung_script = ETCDIR "/ctdb/debug-hung-script.sh";
 498         int fd[2];
 499         struct tevent_timer *ttimer;
 500         struct tevent_fd *tfd;
 501         const char **argv;
 502         int i;
 503
 504         if (helper_prog == NULL) {
 505                 return;
 506         }
 507
 508         if (pipe(fd) < 0) {
 509                 DEBUG(DEBUG_ERR,("Failed to create pipe fd for debug hung script\n"));
 510                 return;
 511         }
 512
 513         argv = talloc_array(state, const char *, 5);
 514
 515         argv[0] = talloc_asprintf(argv, "%d", fd[1]);
 516         argv[1] = talloc_strdup(argv, debug_hung_script);
 517         argv[2] = talloc_asprintf(argv, "%d", state->child);
 518         argv[3] = talloc_strdup(argv, ctdb_eventscript_call_names[state->call]);
 519         argv[4] = NULL;
 520
 521         for (i=0; i<4; i++) {
 522                 if (argv[i] == NULL) {
 523                         close(fd[0]);
 524                         close(fd[1]);
 525                         talloc_free(argv);
 526                         return;
 527                 }
 528         }
 529
 530
 531         if (!ctdb_vfork_with_logging(state, ctdb, "Hung-script",
 532                                      helper_prog, 5, argv, NULL, NULL, &pid)) {
 533                 DEBUG(DEBUG_ERR,("Failed to fork a child to track hung event script\n"));
 534                 talloc_free(argv);
 535                 close(fd[0]);
 536                 close(fd[1]);
 537                 return;
 538         }
 539
 540         talloc_free(argv);
 541         close(fd[1]);
 542
 543         ttimer = tevent_add_timer(ctdb->ev, state,
 544                                   timeval_current_ofs(ctdb->tunable.script_timeout, 0),
 545                                   debug_hung_script_timeout, state);
 546         if (ttimer == NULL) {
 547                 close(fd[0]);
 548                 return;
 549         }
 550
 551         tfd = tevent_add_fd(ctdb->ev, state, fd[0], EVENT_FD_READ,
 552                             debug_hung_script_done, state);
 553         if (tfd == NULL) {
 554                 talloc_free(ttimer);
 555                 close(fd[0]);
 556                 return;
 557         }
 558         tevent_fd_set_auto_close(tfd);
 559 }
 560
 561 /* called when child times out */
 562 static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te,
 563                                       struct timeval t, void *p)
 564 {
 565         struct ctdb_event_script_state *state = talloc_get_type(p, struct ctdb_event_script_state);
 566         struct ctdb_context *ctdb = state->ctdb;
 567         struct ctdb_script_wire *current = get_current_script(state);
 568         struct debug_hung_script_state *debug_state;
 569
 570         DEBUG(DEBUG_ERR,("Event script '%s %s %s' timed out after %.1fs, count: %u, pid: %d\n",
 571                          current->name, ctdb_eventscript_call_names[state->call], state->options,
 572                          timeval_elapsed(&current->start),
 573                          ctdb->event_script_timeouts, state->child));
 574
 575         /* ignore timeouts for these events */
 576         switch (state->call) {
 577         case CTDB_EVENT_START_RECOVERY:
 578         case CTDB_EVENT_RECOVERED:
 579         case CTDB_EVENT_TAKE_IP:
 580         case CTDB_EVENT_RELEASE_IP:
 581                 state->scripts->scripts[state->current].status = 0;
 582                 DEBUG(DEBUG_ERR,("Ignoring hung script for %s call %d\n", state->options, state->call));
 583                 break;
 584         default:
 585                 state->scripts->scripts[state->current].status = -ETIME;
 586         }
 587
 588         debug_state = talloc_zero(ctdb, struct debug_hung_script_state);
 589         if (debug_state == NULL) {
 590                 talloc_free(state);
 591                 return;
 592         }
 593
 594         /* Save information useful for running debug hung script, so
 595          * eventscript state can be freed.
 596          */
 597         debug_state->ctdb = ctdb;
 598         debug_state->child = state->child;
 599         debug_state->call = state->call;
 600
 601         /* This destructor will actually kill the hung event script */
 602         talloc_set_destructor(debug_state, debug_hung_script_state_destructor);
 603
 604         state->child = 0;
 605         talloc_free(state);
 606
 607         ctdb_run_debug_hung_script(ctdb, debug_state);
 608 }
 609
 610 /*
 611   destroy an event script: kill it if ->child != 0.
 612  */
 613 static int event_script_destructor(struct ctdb_event_script_state *state)
 614 {
 615         int status;
 616         struct event_script_callback *callback;
 617
 618         if (state->child) {
 619                 DEBUG(DEBUG_ERR,(__location__ " Sending SIGTERM to child pid:%d\n", state->child));
 620
 621                 if (ctdb_kill(state->ctdb, state->child, SIGTERM) != 0) {
 622                         DEBUG(DEBUG_ERR,("Failed to kill child process for eventscript, errno %s(%d)\n", strerror(errno), errno));
 623                 }
 624         }
 625
 626         /* If we were the current monitor, we no longer are. */
 627         if (state->ctdb->current_monitor == state) {
 628                 state->ctdb->current_monitor = NULL;
 629         }
 630
 631         /* Save our scripts as the last executed status, if we have them.
 632          * See ctdb_event_script_callback_v where we abort monitor event. */
 633         if (state->scripts) {
 634                 talloc_free(state->ctdb->last_status[state->call]);
 635                 state->ctdb->last_status[state->call] = state->scripts;
 636                 if (state->current < state->ctdb->last_status[state->call]->num_scripts) {
 637                         state->ctdb->last_status[state->call]->num_scripts = state->current+1;
 638                 }
 639         }
 640
 641         /* Use last status as result, or "OK" if none. */
 642         if (state->ctdb->last_status[state->call]) {
 643                 status = script_status(state->ctdb->last_status[state->call]);
 644         } else {
 645                 status = 0;
 646         }
 647
 648         /* This is allowed to free us; talloc will prevent double free anyway,
 649          * but beware if you call this outside the destructor!
 650          * the callback hangs off a different context so we walk the list
 651          * of "active" callbacks until we find the one state points to.
 652          * if we cant find it it means the callback has been removed.
 653          */
 654         for (callback = state->ctdb->script_callbacks; callback != NULL; callback = callback->next) {
 655                 if (callback == state->callback) {
 656                         break;
 657                 }
 658         }
 659
 660         state->callback = NULL;
 661
 662         if (callback) {
 663                 /* Make sure destructor doesn't free itself! */
 664                 talloc_steal(NULL, callback);
 665                 callback->fn(state->ctdb, status, callback->private_data);
 666                 talloc_free(callback);
 667         }
 668
 669         return 0;
 670 }
 671
 672 static unsigned int count_words(const char *options)
 673 {
 674         unsigned int words = 0;
 675
 676         options += strspn(options, " \t");
 677         while (*options) {
 678                 words++;
 679                 options += strcspn(options, " \t");
 680                 options += strspn(options, " \t");
 681         }
 682         return words;
 683 }
 684
 685 static bool check_options(enum ctdb_eventscript_call call, const char *options)
 686 {
 687         switch (call) {
 688         /* These all take no arguments. */
 689         case CTDB_EVENT_INIT:
 690         case CTDB_EVENT_SETUP:
 691         case CTDB_EVENT_STARTUP:
 692         case CTDB_EVENT_START_RECOVERY:
 693         case CTDB_EVENT_RECOVERED:
 694         case CTDB_EVENT_MONITOR:
 695         case CTDB_EVENT_SHUTDOWN:
 696         case CTDB_EVENT_IPREALLOCATED:
 697                 return count_words(options) == 0;
 698
 699         case CTDB_EVENT_TAKE_IP: /* interface, IP address, netmask bits. */
 700         case CTDB_EVENT_RELEASE_IP:
 701                 return count_words(options) == 3;
 702
 703         case CTDB_EVENT_UPDATE_IP: /* old interface, new interface, IP address, netmask bits. */
 704                 return count_words(options) == 4;
 705
 706         default:
 707                 DEBUG(DEBUG_ERR,(__location__ "Unknown ctdb_eventscript_call %u\n", call));
 708                 return false;
 709         }
 710 }
 711
 712 static int remove_callback(struct event_script_callback *callback)
 713 {
 714         DLIST_REMOVE(callback->ctdb->script_callbacks, callback);
 715         return 0;
 716 }
 717
 718 /*
 719   run the event script in the background, calling the callback when
 720   finished
 721  */
 722 static int ctdb_event_script_callback_v(struct ctdb_context *ctdb,
 723                                         const void *mem_ctx,
 724                                         void (*callback)(struct ctdb_context *, int, void *),
 725                                         void *private_data,
 726                                         enum ctdb_eventscript_call call,
 727                                         const char *fmt, va_list ap)
 728 {
 729         struct ctdb_event_script_state *state;
 730
 731         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 732                 /* we guarantee that only some specifically allowed event scripts are run
 733                    while in recovery */
 734                 const enum ctdb_eventscript_call allowed_calls[] = {
 735                         CTDB_EVENT_INIT,
 736                         CTDB_EVENT_SETUP,
 737                         CTDB_EVENT_START_RECOVERY,
 738                         CTDB_EVENT_SHUTDOWN,
 739                         CTDB_EVENT_RELEASE_IP,
 740                         CTDB_EVENT_IPREALLOCATED,
 741                 };
 742                 int i;
 743                 for (i=0;i<ARRAY_SIZE(allowed_calls);i++) {
 744                         if (call == allowed_calls[i]) break;
 745                 }
 746                 if (i == ARRAY_SIZE(allowed_calls)) {
 747                         DEBUG(DEBUG_ERR,("Refusing to run event scripts call '%s' while in recovery\n",
 748                                  ctdb_eventscript_call_names[call]));
 749                         return -1;
 750                 }
 751         }
 752
 753         /* Kill off any running monitor events to run this event. */
 754         if (ctdb->current_monitor) {
 755                 struct ctdb_event_script_state *ms = talloc_get_type(ctdb->current_monitor, struct ctdb_event_script_state);
 756
 757                 /* Cancel current monitor callback state only if monitoring
 758                  * context ctdb->monitor->monitor_context has not been freed */
 759                 if (ms->callback != NULL && !ctdb_stopped_monitoring(ctdb)) {
 760                         ms->callback->fn(ctdb, -ECANCELED, ms->callback->private_data);
 761                         talloc_free(ms->callback);
 762                 }
 763
 764                 /* Discard script status so we don't save to last_status */
 765                 talloc_free(ctdb->current_monitor->scripts);
 766                 ctdb->current_monitor->scripts = NULL;
 767                 talloc_free(ctdb->current_monitor);
 768                 ctdb->current_monitor = NULL;
 769         }
 770
 771         state = talloc(ctdb->event_script_ctx, struct ctdb_event_script_state);
 772         CTDB_NO_MEMORY(ctdb, state);
 773
 774         /* The callback isn't done if the context is freed. */
 775         state->callback = talloc(mem_ctx, struct event_script_callback);
 776         CTDB_NO_MEMORY(ctdb, state->callback);
 777         DLIST_ADD(ctdb->script_callbacks, state->callback);
 778         talloc_set_destructor(state->callback, remove_callback);
 779         state->callback->ctdb         = ctdb;
 780         state->callback->fn           = callback;
 781         state->callback->private_data = private_data;
 782
 783         state->ctdb = ctdb;
 784         state->call = call;
 785         state->options = talloc_vasprintf(state, fmt, ap);
 786         state->timeout = timeval_set(ctdb->tunable.script_timeout, 0);
 787         state->scripts = NULL;
 788         if (state->options == NULL) {
 789                 DEBUG(DEBUG_ERR, (__location__ " could not allocate state->options\n"));
 790                 talloc_free(state);
 791                 return -1;
 792         }
 793         if (!check_options(state->call, state->options)) {
 794                 DEBUG(DEBUG_ERR, ("Bad eventscript options '%s' for %s\n",
 795                                   ctdb_eventscript_call_names[state->call], state->options));
 796                 talloc_free(state);
 797                 return -1;
 798         }
 799
 800         DEBUG(DEBUG_INFO,(__location__ " Starting eventscript %s %s\n",
 801                           ctdb_eventscript_call_names[state->call],
 802                           state->options));
 803
 804         /* This is not a child of state, since we save it in destructor. */
 805         state->scripts = ctdb_get_script_list(ctdb, ctdb);
 806         if (state->scripts == NULL) {
 807                 talloc_free(state);
 808                 return -1;
 809         }
 810         state->current = 0;
 811         state->child = 0;
 812
 813         if (call == CTDB_EVENT_MONITOR) {
 814                 ctdb->current_monitor = state;
 815         }
 816
 817         talloc_set_destructor(state, event_script_destructor);
 818
 819         /* Nothing to do? */
 820         if (state->scripts->num_scripts == 0) {
 821                 talloc_free(state);
 822                 return 0;
 823         }
 824
 825         state->scripts->scripts[0].status = fork_child_for_script(ctdb, state);
 826         if (state->scripts->scripts[0].status != 0) {
 827                 /* Callback is called from destructor, with fail result. */
 828                 talloc_free(state);
 829                 return 0;
 830         }
 831
 832         if (!timeval_is_zero(&state->timeout)) {
 833                 event_add_timed(ctdb->ev, state, timeval_current_ofs(state->timeout.tv_sec, state->timeout.tv_usec), ctdb_event_script_timeout, state);
 834         } else {
 835                 DEBUG(DEBUG_ERR, (__location__ " eventscript %s %s called with no timeout\n",
 836                                   ctdb_eventscript_call_names[state->call],
 837                                   state->options));
 838         }
 839
 840         return 0;
 841 }
 842
 843
 844 /*
 845   run the event script in the background, calling the callback when
 846   finished.  If mem_ctx is freed, callback will never be called.
 847  */
 848 int ctdb_event_script_callback(struct ctdb_context *ctdb,
 849                                TALLOC_CTX *mem_ctx,
 850                                void (*callback)(struct ctdb_context *, int, void *),
 851                                void *private_data,
 852                                enum ctdb_eventscript_call call,
 853                                const char *fmt, ...)
 854 {
 855         va_list ap;
 856         int ret;
 857
 858         va_start(ap, fmt);
 859         ret = ctdb_event_script_callback_v(ctdb, mem_ctx, callback, private_data, call, fmt, ap);
 860         va_end(ap);
 861
 862         return ret;
 863 }
 864
 865
 866 struct callback_status {
 867         bool done;
 868         int status;
 869 };
 870
 871 /*
 872   called when ctdb_event_script() finishes
 873  */
 874 static void event_script_callback(struct ctdb_context *ctdb, int status, void *private_data)
 875 {
 876         struct callback_status *s = (struct callback_status *)private_data;
 877         s->done = true;
 878         s->status = status;
 879 }
 880
 881 /*
 882   run the event script, waiting for it to complete. Used when the caller
 883   doesn't want to continue till the event script has finished.
 884  */
 885 int ctdb_event_script_args(struct ctdb_context *ctdb, enum ctdb_eventscript_call call,
 886                            const char *fmt, ...)
 887 {
 888         va_list ap;
 889         int ret;
 890         struct callback_status status;
 891
 892         va_start(ap, fmt);
 893         ret = ctdb_event_script_callback_v(ctdb, ctdb,
 894                         event_script_callback, &status, call, fmt, ap);
 895         va_end(ap);
 896         if (ret != 0) {
 897                 return ret;
 898         }
 899
 900         status.status = -1;
 901         status.done = false;
 902
 903         while (status.done == false && event_loop_once(ctdb->ev) == 0) /* noop */;
 904
 905         if (status.status == -ETIME) {
 906                 DEBUG(DEBUG_ERR, (__location__ " eventscript for '%s' timedout."
 907                                   " Immediately banning ourself for %d seconds\n",
 908                                   ctdb_eventscript_call_names[call],
 909                                   ctdb->tunable.recovery_ban_period));
 910
 911                 /* Don't ban self if CTDB is starting up or shutting down */
 912                 if (call != CTDB_EVENT_INIT && call != CTDB_EVENT_SHUTDOWN) {
 913                         ctdb_ban_self(ctdb);
 914                 }
 915         }
 916
 917         return status.status;
 918 }
 919
 920 int ctdb_event_script(struct ctdb_context *ctdb, enum ctdb_eventscript_call call)
 921 {
 922         /* GCC complains about empty format string, so use %s and "". */
 923         return ctdb_event_script_args(ctdb, call, "%s", "");
 924 }
 925
 926 struct eventscript_callback_state {
 927         struct ctdb_req_control *c;
 928 };
 929
 930 /*
 931   called when a forced eventscript run has finished
 932  */
 933 static void run_eventscripts_callback(struct ctdb_context *ctdb, int status,
 934                                  void *private_data)
 935 {
 936         struct eventscript_callback_state *state =
 937                 talloc_get_type(private_data, struct eventscript_callback_state);
 938
 939         ctdb_enable_monitoring(ctdb);
 940
 941         if (status != 0) {
 942                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts\n"));
 943         }
 944
 945         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
 946         /* This will free the struct ctdb_event_script_state we are in! */
 947         talloc_free(state);
 948         return;
 949 }
 950
 951
 952 /* Returns rest of string, or NULL if no match. */
 953 static const char *get_call(const char *p, enum ctdb_eventscript_call *call)
 954 {
 955         unsigned int len;
 956
 957         /* Skip any initial whitespace. */
 958         p += strspn(p, " \t");
 959
 960         /* See if we match any. */
 961         for (*call = 0; *call < CTDB_EVENT_MAX; (*call)++) {
 962                 len = strlen(ctdb_eventscript_call_names[*call]);
 963                 if (strncmp(p, ctdb_eventscript_call_names[*call], len) == 0) {
 964                         /* If end of string or whitespace, we're done. */
 965                         if (strcspn(p + len, " \t") == 0) {
 966                                 return p + len;
 967                         }
 968                 }
 969         }
 970         return NULL;
 971 }
 972
 973 /*
 974   A control to force running of the eventscripts from the ctdb client tool
 975 */
 976 int32_t ctdb_run_eventscripts(struct ctdb_context *ctdb,
 977                 struct ctdb_req_control *c,
 978                 TDB_DATA indata, bool *async_reply)
 979 {
 980         int ret;
 981         struct eventscript_callback_state *state;
 982         const char *options;
 983         enum ctdb_eventscript_call call;
 984
 985         /* Figure out what call they want. */
 986         options = get_call((const char *)indata.dptr, &call);
 987         if (!options) {
 988                 DEBUG(DEBUG_ERR, (__location__ " Invalid event name \"%s\"\n", (const char *)indata.dptr));
 989                 return -1;
 990         }
 991
 992         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 993                 DEBUG(DEBUG_ERR, (__location__ " Aborted running eventscript \"%s\" while in RECOVERY mode\n", indata.dptr));
 994                 return -1;
 995         }
 996
 997         state = talloc(ctdb->event_script_ctx, struct eventscript_callback_state);
 998         CTDB_NO_MEMORY(ctdb, state);
 999
1000         state->c = talloc_steal(state, c);
1001
1002         DEBUG(DEBUG_NOTICE,("Running eventscripts with arguments %s\n", indata.dptr));
1003
1004         ctdb_disable_monitoring(ctdb);
1005
1006         ret = ctdb_event_script_callback(ctdb,
1007                          state, run_eventscripts_callback, state,
1008                          call, "%s", options);
1009
1010         if (ret != 0) {
1011                 ctdb_enable_monitoring(ctdb);
1012                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts with arguments %s\n", indata.dptr));
1013                 talloc_free(state);
1014                 return -1;
1015         }
1016
1017         /* tell ctdb_control.c that we will be replying asynchronously */
1018         *async_reply = true;
1019
1020         return 0;
1021 }
1022
1023
1024
1025 int32_t ctdb_control_enable_script(struct ctdb_context *ctdb, TDB_DATA indata)
1026 {
1027         const char *script;
1028         struct stat st;
1029         char *filename;
1030         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1031
1032         script = (char *)indata.dptr;
1033         if (indata.dsize == 0) {
1034                 DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
1035                 talloc_free(tmp_ctx);
1036                 return -1;
1037         }
1038         if (indata.dptr[indata.dsize - 1] != '\0') {
1039                 DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
1040                 talloc_free(tmp_ctx);
1041                 return -1;
1042         }
1043         if (index(script,'/') != NULL) {
1044                 DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to enable script %s\n", script));
1045                 talloc_free(tmp_ctx);
1046                 return -1;
1047         }
1048
1049
1050         if (stat(ctdb->event_script_dir, &st) != 0 &&
1051             errno == ENOENT) {
1052                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
1053                 talloc_free(tmp_ctx);
1054                 return -1;
1055         }
1056
1057
1058         filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
1059         if (filename == NULL) {
1060                 DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
1061                 talloc_free(tmp_ctx);
1062                 return -1;
1063         }
1064
1065         if (stat(filename, &st) != 0) {
1066                 DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to enable script.\n", filename));
1067                 talloc_free(tmp_ctx);
1068                 return -1;
1069         }
1070
1071         if (chmod(filename, st.st_mode | S_IXUSR) == -1) {
1072                 DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to enable script.\n", filename));
1073                 talloc_free(tmp_ctx);
1074                 return -1;
1075         }
1076
1077         talloc_free(tmp_ctx);
1078         return 0;
1079 }
1080
1081 int32_t ctdb_control_disable_script(struct ctdb_context *ctdb, TDB_DATA indata)
1082 {
1083         const char *script;
1084         struct stat st;
1085         char *filename;
1086         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1087
1088         script = (char *)indata.dptr;
1089         if (indata.dsize == 0) {
1090                 DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
1091                 talloc_free(tmp_ctx);
1092                 return -1;
1093         }
1094         if (indata.dptr[indata.dsize - 1] != '\0') {
1095                 DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
1096                 talloc_free(tmp_ctx);
1097                 return -1;
1098         }
1099         if (index(script,'/') != NULL) {
1100                 DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to disable script %s\n", script));
1101                 talloc_free(tmp_ctx);
1102                 return -1;
1103         }
1104
1105
1106         if (stat(ctdb->event_script_dir, &st) != 0 &&
1107             errno == ENOENT) {
1108                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
1109                 talloc_free(tmp_ctx);
1110                 return -1;
1111         }
1112
1113
1114         filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
1115         if (filename == NULL) {
1116                 DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
1117                 talloc_free(tmp_ctx);
1118                 return -1;
1119         }
1120
1121         if (stat(filename, &st) != 0) {
1122                 DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to disable script.\n", filename));
1123                 talloc_free(tmp_ctx);
1124                 return -1;
1125         }
1126
1127         if (chmod(filename, st.st_mode & ~(S_IXUSR|S_IXGRP|S_IXOTH)) == -1) {
1128                 DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to disable script.\n", filename));
1129                 talloc_free(tmp_ctx);
1130                 return -1;
1131         }
1132
1133         talloc_free(tmp_ctx);
1134         return 0;
1135 }