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 /*
  33   ctdbd sends us a SIGTERM when we should die.
  34  */
  35 static void sigterm(int sig)
  36 {
  37         pid_t pid;
  38
  39         /* all the child processes will be running in the same process group */
  40         pid = getpgrp();
  41         if (pid == -1) {
  42                 kill(-getpid(), SIGKILL);
  43         } else {
  44                 kill(-pid, SIGKILL);
  45         }
  46         _exit(1);
  47 }
  48
  49 /* This is attached to the event script state. */
  50 struct event_script_callback {
  51         struct event_script_callback *next, *prev;
  52         struct ctdb_context *ctdb;
  53
  54         /* Warning: this can free us! */
  55         void (*fn)(struct ctdb_context *, int, void *);
  56         void *private_data;
  57 };
  58
  59
  60 struct ctdb_event_script_state {
  61         struct ctdb_context *ctdb;
  62         struct event_script_callback *callback;
  63         pid_t child;
  64         int fd[2];
  65         enum ctdb_eventscript_call call;
  66         const char *options;
  67         struct timeval timeout;
  68
  69         unsigned int current;
  70         struct ctdb_scripts_wire *scripts;
  71 };
  72
  73 static struct ctdb_script_wire *get_current_script(struct ctdb_event_script_state *state)
  74 {
  75         return &state->scripts->scripts[state->current];
  76 }
  77
  78 /* called from ctdb_logging when we have received output on STDERR from
  79  * one of the eventscripts
  80  */
  81 static void log_event_script_output(const char *str, uint16_t len, void *p)
  82 {
  83         struct ctdb_event_script_state *state
  84                 = talloc_get_type(p, struct ctdb_event_script_state);
  85         struct ctdb_script_wire *current;
  86         unsigned int slen, min;
  87
  88         /* We may have been aborted to run something else.  Discard */
  89         if (state->scripts == NULL) {
  90                 return;
  91         }
  92
  93         current = get_current_script(state);
  94
  95         /* Append, but don't overfill buffer.  It starts zero-filled. */
  96         slen = strlen(current->output);
  97         min = MIN(len, sizeof(current->output) - slen - 1);
  98
  99         memcpy(current->output + slen, str, min);
 100 }
 101
 102 int32_t ctdb_control_get_event_script_status(struct ctdb_context *ctdb,
 103                                              uint32_t call_type,
 104                                              TDB_DATA *outdata)
 105 {
 106         if (call_type >= CTDB_EVENT_MAX) {
 107                 return -1;
 108         }
 109
 110         if (ctdb->last_status[call_type] == NULL) {
 111                 /* If it's never been run, return nothing so they can tell. */
 112                 outdata->dsize = 0;
 113         } else {
 114                 outdata->dsize = talloc_get_size(ctdb->last_status[call_type]);
 115                 outdata->dptr  = (uint8_t *)ctdb->last_status[call_type];
 116         }
 117         return 0;
 118 }
 119
 120 struct ctdb_script_tree_item {
 121         const char *name;
 122         int error;
 123 };
 124
 125 /* Return true if OK, otherwise set errno. */
 126 static bool check_executable(const char *dir, const char *name)
 127 {
 128         char *full;
 129         struct stat st;
 130
 131         full = talloc_asprintf(NULL, "%s/%s", dir, name);
 132         if (!full)
 133                 return false;
 134
 135         if (stat(full, &st) != 0) {
 136                 DEBUG(DEBUG_ERR,("Could not stat event script %s: %s\n",
 137                                  full, strerror(errno)));
 138                 talloc_free(full);
 139                 return false;
 140         }
 141
 142         if (!(st.st_mode & S_IXUSR)) {
 143                 DEBUG(DEBUG_DEBUG,("Event script %s is not executable. Ignoring this event script\n", full));
 144                 errno = ENOEXEC;
 145                 talloc_free(full);
 146                 return false;
 147         }
 148
 149         talloc_free(full);
 150         return true;
 151 }
 152
 153 static struct ctdb_scripts_wire *ctdb_get_script_list(struct ctdb_context *ctdb, TALLOC_CTX *mem_ctx)
 154 {
 155         DIR *dir;
 156         struct dirent *de;
 157         struct stat st;
 158         trbt_tree_t *tree;
 159         struct ctdb_scripts_wire *scripts;
 160         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
 161         struct ctdb_script_tree_item *tree_item;
 162         int count;
 163
 164         /*
 165           the service specific event scripts
 166         */
 167         if (stat(ctdb->event_script_dir, &st) != 0 &&
 168             errno == ENOENT) {
 169                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
 170                 talloc_free(tmp_ctx);
 171                 return NULL;
 172         }
 173
 174         /* create a tree to store all the script names in */
 175         tree = trbt_create(tmp_ctx, 0);
 176
 177         /* scan all directory entries and insert all valid scripts into the
 178            tree
 179         */
 180         dir = opendir(ctdb->event_script_dir);
 181         if (dir == NULL) {
 182                 DEBUG(DEBUG_CRIT,("Failed to open event script directory '%s'\n", ctdb->event_script_dir));
 183                 talloc_free(tmp_ctx);
 184                 return NULL;
 185         }
 186
 187         count = 0;
 188         while ((de=readdir(dir)) != NULL) {
 189                 int namlen;
 190                 unsigned num;
 191
 192                 namlen = strlen(de->d_name);
 193
 194                 if (namlen < 3) {
 195                         continue;
 196                 }
 197
 198                 if (de->d_name[namlen-1] == '~') {
 199                         /* skip files emacs left behind */
 200                         continue;
 201                 }
 202
 203                 if (de->d_name[2] != '.') {
 204                         continue;
 205                 }
 206
 207                 if (sscanf(de->d_name, "%02u.", &num) != 1) {
 208                         continue;
 209                 }
 210
 211                 if (strlen(de->d_name) > MAX_SCRIPT_NAME) {
 212                         DEBUG(DEBUG_ERR,("Script name %s too long! %u chars max",
 213                                          de->d_name, MAX_SCRIPT_NAME));
 214                         continue;
 215                 }
 216
 217                 tree_item = talloc(tree, struct ctdb_script_tree_item);
 218                 if (tree_item == NULL) {
 219                         DEBUG(DEBUG_ERR, (__location__ " Failed to allocate new tree item\n"));
 220                         closedir(dir);
 221                         talloc_free(tmp_ctx);
 222                         return NULL;
 223                 }
 224
 225                 tree_item->error = 0;
 226                 if (!check_executable(ctdb->event_script_dir, de->d_name)) {
 227                         tree_item->error = errno;
 228                 }
 229
 230                 tree_item->name = talloc_strdup(tree_item, de->d_name);
 231                 if (tree_item->name == NULL) {
 232                         DEBUG(DEBUG_ERR,(__location__ " Failed to allocate script name.\n"));
 233                         closedir(dir);
 234                         talloc_free(tmp_ctx);
 235                         return NULL;
 236                 }
 237
 238                 /* store the event script in the tree */
 239                 trbt_insert32(tree, (num<<16)|count++, tree_item);
 240         }
 241         closedir(dir);
 242
 243         /* Overallocates by one, but that's OK */
 244         scripts = talloc_zero_size(tmp_ctx,
 245                                    sizeof(*scripts)
 246                                    + sizeof(scripts->scripts[0]) * count);
 247         if (scripts == NULL) {
 248                 DEBUG(DEBUG_ERR, (__location__ " Failed to allocate scripts\n"));
 249                 talloc_free(tmp_ctx);
 250                 return NULL;
 251         }
 252         scripts->num_scripts = count;
 253
 254         for (count = 0; count < scripts->num_scripts; count++) {
 255                 tree_item = trbt_findfirstarray32(tree, 1);
 256
 257                 strcpy(scripts->scripts[count].name, tree_item->name);
 258                 scripts->scripts[count].status = -tree_item->error;
 259
 260                 /* remove this script from the tree */
 261                 talloc_free(tree_item);
 262         }
 263
 264         talloc_steal(mem_ctx, scripts);
 265         talloc_free(tmp_ctx);
 266         return scripts;
 267 }
 268
 269 static int child_setup(struct ctdb_context *ctdb)
 270 {
 271         if (setpgid(0,0) != 0) {
 272                 int ret = -errno;
 273                 DEBUG(DEBUG_ERR,("Failed to create process group for event scripts - %s\n",
 274                          strerror(errno)));
 275                 return ret;
 276         }
 277
 278         signal(SIGTERM, sigterm);
 279         return 0;
 280 }
 281
 282 static char *child_command_string(struct ctdb_context *ctdb,
 283                                        TALLOC_CTX *ctx,
 284                                        const char *scriptname,
 285                                        enum ctdb_eventscript_call call,
 286                                        const char *options)
 287 {
 288         return talloc_asprintf(ctx, "%s/%s %s %s",
 289                                ctdb->event_script_dir,
 290                                scriptname,
 291                                ctdb_eventscript_call_names[call],
 292                                options);
 293 }
 294
 295 static int child_run_one(struct ctdb_context *ctdb,
 296                          const char *scriptname, const char *cmdstr)
 297 {
 298         int ret;
 299
 300         ret = system(cmdstr);
 301         /* if the system() call was successful, translate ret into the
 302            return code from the command
 303         */
 304         if (ret != -1) {
 305                 ret = WEXITSTATUS(ret);
 306         } else {
 307                 ret = -errno;
 308         }
 309
 310         /* 127 could mean it does not exist, 126 non-executable. */
 311         if (ret == 127 || ret == 126) {
 312                 /* Re-check it... */
 313                 if (!check_executable(ctdb->event_script_dir, scriptname)) {
 314                         DEBUG(DEBUG_ERR,("Script %s returned status %u. Someone just deleted it?\n",
 315                                          cmdstr, ret));
 316                         ret = -errno;
 317                 }
 318         }
 319         return ret;
 320 }
 321
 322 /*
 323   Actually run one event script
 324   this function is called and run in the context of a forked child
 325   which allows it to do blocking calls such as system()
 326  */
 327 static int child_run_script(struct ctdb_context *ctdb,
 328                             enum ctdb_eventscript_call call,
 329                             const char *options,
 330                             struct ctdb_script_wire *current)
 331 {
 332         char *cmdstr;
 333         int ret;
 334         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
 335
 336         ret = child_setup(ctdb);
 337         if (ret != 0)
 338                 goto out;
 339
 340         cmdstr = child_command_string(ctdb, tmp_ctx,
 341                                       current->name, call, options);
 342         CTDB_NO_MEMORY(ctdb, cmdstr);
 343
 344         DEBUG(DEBUG_DEBUG,("Executing event script %s\n",cmdstr));
 345
 346         if (current->status) {
 347                 ret = current->status;
 348                 goto out;
 349         }
 350
 351         ret = child_run_one(ctdb, current->name, cmdstr);
 352 out:
 353         talloc_free(tmp_ctx);
 354         return ret;
 355 }
 356
 357 /* There cannot be more than 10 arguments to command helper. */
 358 #define MAX_HELPER_ARGS         (10)
 359
 360 static bool child_helper_args(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
 361                               enum ctdb_eventscript_call call,
 362                               const char *options,
 363                               struct ctdb_script_wire *current, int fd,
 364                               int *argc, const char ***argv)
 365 {
 366         const char **tmp;
 367         int n, i;
 368         char *t, *saveptr, *opt;
 369
 370         tmp = talloc_array(mem_ctx, const char *, 10+1);
 371         if (tmp == NULL)  goto failed;
 372
 373         tmp[0] = talloc_asprintf(tmp, "%d", fd);
 374         tmp[1] = talloc_asprintf(tmp, "%s/%s", ctdb->event_script_dir, current->name);
 375         tmp[2] = talloc_asprintf(tmp, "%s", ctdb_eventscript_call_names[call]);
 376         n = 3;
 377
 378         /* Split options into individual arguments */
 379         opt = talloc_strdup(mem_ctx, options);
 380         if (opt == NULL) {
 381                 goto failed;
 382         }
 383
 384         t = strtok_r(opt, " ", &saveptr);
 385         while (t != NULL) {
 386                 tmp[n++] = talloc_strdup(tmp, t);
 387                 if (n > MAX_HELPER_ARGS) {
 388                         goto args_failed;
 389                 }
 390                 t = strtok_r(NULL, " ", &saveptr);
 391         }
 392
 393         for (i=0; i<n; i++) {
 394                 if (tmp[i] == NULL) {
 395                         goto failed;
 396                 }
 397         }
 398
 399         /* Last argument should be NULL */
 400         tmp[n++] = NULL;
 401
 402         *argc = n;
 403         *argv = tmp;
 404         return true;
 405
 406
 407 args_failed:
 408         DEBUG(DEBUG_ERR, (__location__ " too many arguments '%s' to eventscript '%s'\n",
 409                           options, ctdb_eventscript_call_names[call]));
 410
 411 failed:
 412         if (tmp) {
 413                 talloc_free(tmp);
 414         }
 415         return false;
 416
 417 }
 418
 419 static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
 420                                       uint16_t flags, void *p);
 421
 422 static const char *helper_prog = NULL;
 423
 424 static int fork_child_for_script(struct ctdb_context *ctdb,
 425                                  struct ctdb_event_script_state *state)
 426 {
 427         int r;
 428         struct tevent_fd *fde;
 429         struct ctdb_script_wire *current = get_current_script(state);
 430         int argc;
 431         const char **argv;
 432         static const char *helper = BINDIR "/ctdb_event_helper";
 433
 434         if (helper_prog == NULL) {
 435                 const char *t = getenv("CTDB_EVENT_HELPER");
 436                 if (t != NULL) {
 437                         helper_prog = t;
 438                 } else {
 439                         helper_prog = helper;
 440                 }
 441         }
 442
 443         current->start = timeval_current();
 444
 445         r = pipe(state->fd);
 446         if (r != 0) {
 447                 DEBUG(DEBUG_ERR, (__location__ " pipe failed for child eventscript process\n"));
 448                 return -errno;
 449         }
 450
 451         /* Arguments for helper */
 452         if (!child_helper_args(state, ctdb, state->call, state->options, current,
 453                                state->fd[1], &argc, &argv)) {
 454                 DEBUG(DEBUG_ERR, (__location__ " failed to create arguments for eventscript helper\n"));
 455                 r = -ENOMEM;
 456                 close(state->fd[0]);
 457                 close(state->fd[1]);
 458                 return r;
 459         }
 460
 461         if (!ctdb_vfork_with_logging(state, ctdb, current->name,
 462                                      helper_prog, argc, argv,
 463                                      log_event_script_output,
 464                                      state, &state->child)) {
 465                 talloc_free(argv);
 466                 r = -errno;
 467                 close(state->fd[0]);
 468                 close(state->fd[1]);
 469                 return r;
 470         }
 471
 472         talloc_free(argv);
 473
 474         close(state->fd[1]);
 475         set_close_on_exec(state->fd[0]);
 476
 477         /* Set ourselves up to be called when that's done. */
 478         fde = event_add_fd(ctdb->ev, state, state->fd[0], EVENT_FD_READ,
 479                            ctdb_event_script_handler, state);
 480         tevent_fd_set_auto_close(fde);
 481
 482         return 0;
 483 }
 484
 485 /*
 486  Summarize status of this run of scripts.
 487  */
 488 static int script_status(struct ctdb_scripts_wire *scripts)
 489 {
 490         unsigned int i;
 491
 492         for (i = 0; i < scripts->num_scripts; i++) {
 493                 switch (scripts->scripts[i].status) {
 494                 case -ENOENT:
 495                 case -ENOEXEC:
 496                         /* Disabled or missing; that's OK. */
 497                         break;
 498                 case 0:
 499                         /* No problem. */
 500                         break;
 501                 default:
 502                         return scripts->scripts[i].status;
 503                 }
 504         }
 505
 506         /* All OK! */
 507         return 0;
 508 }
 509
 510 /* called when child is finished */
 511 static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde,
 512                                       uint16_t flags, void *p)
 513 {
 514         struct ctdb_event_script_state *state =
 515                 talloc_get_type(p, struct ctdb_event_script_state);
 516         struct ctdb_script_wire *current = get_current_script(state);
 517         struct ctdb_context *ctdb = state->ctdb;
 518         int r, status;
 519
 520         if (ctdb == NULL) {
 521                 DEBUG(DEBUG_ERR,("Eventscript finished but ctdb is NULL\n"));
 522                 return;
 523         }
 524
 525         r = read(state->fd[0], &current->status, sizeof(current->status));
 526         if (r < 0) {
 527                 current->status = -errno;
 528         } else if (r != sizeof(current->status)) {
 529                 current->status = -EIO;
 530         }
 531
 532         current->finished = timeval_current();
 533         /* valgrind gets overloaded if we run next script as it's still doing
 534          * post-execution analysis, so kill finished child here. */
 535         if (ctdb->valgrinding) {
 536                 ctdb_kill(ctdb, state->child, SIGKILL);
 537         }
 538
 539         state->child = 0;
 540
 541         status = script_status(state->scripts);
 542
 543         /* Aborted or finished all scripts?  We're done. */
 544         if (status != 0 || state->current+1 == state->scripts->num_scripts) {
 545                 DEBUG(DEBUG_INFO,(__location__ " Eventscript %s %s finished with state %d\n",
 546                                   ctdb_eventscript_call_names[state->call], state->options, status));
 547
 548                 ctdb->event_script_timeouts = 0;
 549                 talloc_free(state);
 550                 return;
 551         }
 552
 553         /* Forget about that old fd. */
 554         talloc_free(fde);
 555
 556         /* Next script! */
 557         state->current++;
 558         current++;
 559         current->status = fork_child_for_script(ctdb, state);
 560         if (current->status != 0) {
 561                 /* This calls the callback. */
 562                 talloc_free(state);
 563         }
 564 }
 565
 566 struct debug_hung_script_state {
 567         struct ctdb_context *ctdb;
 568         pid_t child;
 569         enum ctdb_eventscript_call call;
 570 };
 571
 572 static int debug_hung_script_state_destructor(struct debug_hung_script_state *state)
 573 {
 574         if (state->child) {
 575                 ctdb_kill(state->ctdb, state->child, SIGKILL);
 576         }
 577         return 0;
 578 }
 579
 580 static void debug_hung_script_timeout(struct tevent_context *ev, struct tevent_timer *te,
 581                                       struct timeval t, void *p)
 582 {
 583         struct debug_hung_script_state *state =
 584                 talloc_get_type(p, struct debug_hung_script_state);
 585
 586         talloc_free(state);
 587 }
 588
 589 static void debug_hung_script_done(struct tevent_context *ev, struct tevent_fd *fde,
 590                                    uint16_t flags, void *p)
 591 {
 592         struct debug_hung_script_state *state =
 593                 talloc_get_type(p, struct debug_hung_script_state);
 594
 595         talloc_free(state);
 596 }
 597
 598 static void ctdb_run_debug_hung_script(struct ctdb_context *ctdb, struct debug_hung_script_state *state)
 599 {
 600         pid_t pid;
 601         const char * debug_hung_script = ETCDIR "/ctdb/debug-hung-script.sh";
 602         int fd[2];
 603         struct tevent_timer *ttimer;
 604         struct tevent_fd *tfd;
 605
 606         if (pipe(fd) < 0) {
 607                 DEBUG(DEBUG_ERR,("Failed to create pipe fd for debug hung script\n"));
 608                 return;
 609         }
 610
 611         if (!ctdb_fork_with_logging(ctdb, ctdb, "Hung script", NULL, NULL, &pid)) {
 612                 DEBUG(DEBUG_ERR,("Failed to fork a child process with logging to track hung event script\n"));
 613                 close(fd[0]);
 614                 close(fd[1]);
 615                 return;
 616         }
 617         if (pid == -1) {
 618                 DEBUG(DEBUG_ERR,("Fork for debug script failed : %s\n",
 619                                  strerror(errno)));
 620                 close(fd[0]);
 621                 close(fd[1]);
 622                 return;
 623         }
 624         if (pid == 0) {
 625                 char *buf;
 626
 627                 ctdb_set_process_name("ctdb_debug_hung_script");
 628                 if (getenv("CTDB_DEBUG_HUNG_SCRIPT") != NULL) {
 629                         debug_hung_script = getenv("CTDB_DEBUG_HUNG_SCRIPT");
 630                 }
 631
 632                 close(fd[0]);
 633
 634                 buf = talloc_asprintf(NULL, "%s %d %s",
 635                                       debug_hung_script, state->child,
 636                                       ctdb_eventscript_call_names[state->call]);
 637                 system(buf);
 638                 talloc_free(buf);
 639
 640                 _exit(0);
 641         }
 642
 643         close(fd[1]);
 644
 645         ttimer = tevent_add_timer(ctdb->ev, state,
 646                                   timeval_current_ofs(ctdb->tunable.script_timeout, 0),
 647                                   debug_hung_script_timeout, state);
 648         if (ttimer == NULL) {
 649                 close(fd[0]);
 650                 return;
 651         }
 652
 653         tfd = tevent_add_fd(ctdb->ev, state, fd[0], EVENT_FD_READ,
 654                             debug_hung_script_done, state);
 655         if (tfd == NULL) {
 656                 talloc_free(ttimer);
 657                 close(fd[0]);
 658                 return;
 659         }
 660         tevent_fd_set_auto_close(tfd);
 661 }
 662
 663 /* called when child times out */
 664 static void ctdb_event_script_timeout(struct event_context *ev, struct timed_event *te,
 665                                       struct timeval t, void *p)
 666 {
 667         struct ctdb_event_script_state *state = talloc_get_type(p, struct ctdb_event_script_state);
 668         struct ctdb_context *ctdb = state->ctdb;
 669         struct ctdb_script_wire *current = get_current_script(state);
 670         struct debug_hung_script_state *debug_state;
 671
 672         DEBUG(DEBUG_ERR,("Event script '%s %s %s' timed out after %.1fs, count: %u, pid: %d\n",
 673                          current->name, ctdb_eventscript_call_names[state->call], state->options,
 674                          timeval_elapsed(&current->start),
 675                          ctdb->event_script_timeouts, state->child));
 676
 677         /* ignore timeouts for these events */
 678         switch (state->call) {
 679         case CTDB_EVENT_START_RECOVERY:
 680         case CTDB_EVENT_RECOVERED:
 681         case CTDB_EVENT_TAKE_IP:
 682         case CTDB_EVENT_RELEASE_IP:
 683                 state->scripts->scripts[state->current].status = 0;
 684                 DEBUG(DEBUG_ERR,("Ignoring hung script for %s call %d\n", state->options, state->call));
 685                 break;
 686         default:
 687                 state->scripts->scripts[state->current].status = -ETIME;
 688         }
 689
 690         debug_state = talloc_zero(ctdb, struct debug_hung_script_state);
 691         if (debug_state == NULL) {
 692                 talloc_free(state);
 693                 return;
 694         }
 695
 696         /* Save information useful for running debug hung script, so
 697          * eventscript state can be freed.
 698          */
 699         debug_state->ctdb = ctdb;
 700         debug_state->child = state->child;
 701         debug_state->call = state->call;
 702
 703         /* This destructor will actually kill the hung event script */
 704         talloc_set_destructor(debug_state, debug_hung_script_state_destructor);
 705
 706         state->child = 0;
 707         talloc_free(state);
 708
 709         ctdb_run_debug_hung_script(ctdb, debug_state);
 710 }
 711
 712 /*
 713   destroy an event script: kill it if ->child != 0.
 714  */
 715 static int event_script_destructor(struct ctdb_event_script_state *state)
 716 {
 717         int status;
 718         struct event_script_callback *callback;
 719
 720         if (state->child) {
 721                 DEBUG(DEBUG_ERR,(__location__ " Sending SIGTERM to child pid:%d\n", state->child));
 722
 723                 if (ctdb_kill(state->ctdb, state->child, SIGTERM) != 0) {
 724                         DEBUG(DEBUG_ERR,("Failed to kill child process for eventscript, errno %s(%d)\n", strerror(errno), errno));
 725                 }
 726         }
 727
 728         /* If we were the current monitor, we no longer are. */
 729         if (state->ctdb->current_monitor == state) {
 730                 state->ctdb->current_monitor = NULL;
 731         }
 732
 733         /* Save our scripts as the last executed status, if we have them.
 734          * See ctdb_event_script_callback_v where we abort monitor event. */
 735         if (state->scripts) {
 736                 talloc_free(state->ctdb->last_status[state->call]);
 737                 state->ctdb->last_status[state->call] = state->scripts;
 738                 if (state->current < state->ctdb->last_status[state->call]->num_scripts) {
 739                         state->ctdb->last_status[state->call]->num_scripts = state->current+1;
 740                 }
 741         }
 742
 743         /* Use last status as result, or "OK" if none. */
 744         if (state->ctdb->last_status[state->call]) {
 745                 status = script_status(state->ctdb->last_status[state->call]);
 746         } else {
 747                 status = 0;
 748         }
 749
 750         /* This is allowed to free us; talloc will prevent double free anyway,
 751          * but beware if you call this outside the destructor!
 752          * the callback hangs off a different context so we walk the list
 753          * of "active" callbacks until we find the one state points to.
 754          * if we cant find it it means the callback has been removed.
 755          */
 756         for (callback = state->ctdb->script_callbacks; callback != NULL; callback = callback->next) {
 757                 if (callback == state->callback) {
 758                         break;
 759                 }
 760         }
 761
 762         state->callback = NULL;
 763
 764         if (callback) {
 765                 /* Make sure destructor doesn't free itself! */
 766                 talloc_steal(NULL, callback);
 767                 callback->fn(state->ctdb, status, callback->private_data);
 768                 talloc_free(callback);
 769         }
 770
 771         return 0;
 772 }
 773
 774 static unsigned int count_words(const char *options)
 775 {
 776         unsigned int words = 0;
 777
 778         options += strspn(options, " \t");
 779         while (*options) {
 780                 words++;
 781                 options += strcspn(options, " \t");
 782                 options += strspn(options, " \t");
 783         }
 784         return words;
 785 }
 786
 787 static bool check_options(enum ctdb_eventscript_call call, const char *options)
 788 {
 789         switch (call) {
 790         /* These all take no arguments. */
 791         case CTDB_EVENT_INIT:
 792         case CTDB_EVENT_SETUP:
 793         case CTDB_EVENT_STARTUP:
 794         case CTDB_EVENT_START_RECOVERY:
 795         case CTDB_EVENT_RECOVERED:
 796         case CTDB_EVENT_MONITOR:
 797         case CTDB_EVENT_SHUTDOWN:
 798         case CTDB_EVENT_IPREALLOCATED:
 799                 return count_words(options) == 0;
 800
 801         case CTDB_EVENT_TAKE_IP: /* interface, IP address, netmask bits. */
 802         case CTDB_EVENT_RELEASE_IP:
 803                 return count_words(options) == 3;
 804
 805         case CTDB_EVENT_UPDATE_IP: /* old interface, new interface, IP address, netmask bits. */
 806                 return count_words(options) == 4;
 807
 808         default:
 809                 DEBUG(DEBUG_ERR,(__location__ "Unknown ctdb_eventscript_call %u\n", call));
 810                 return false;
 811         }
 812 }
 813
 814 static int remove_callback(struct event_script_callback *callback)
 815 {
 816         DLIST_REMOVE(callback->ctdb->script_callbacks, callback);
 817         return 0;
 818 }
 819
 820 /*
 821   run the event script in the background, calling the callback when
 822   finished
 823  */
 824 static int ctdb_event_script_callback_v(struct ctdb_context *ctdb,
 825                                         const void *mem_ctx,
 826                                         void (*callback)(struct ctdb_context *, int, void *),
 827                                         void *private_data,
 828                                         enum ctdb_eventscript_call call,
 829                                         const char *fmt, va_list ap)
 830 {
 831         struct ctdb_event_script_state *state;
 832
 833         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
 834                 /* we guarantee that only some specifically allowed event scripts are run
 835                    while in recovery */
 836                 const enum ctdb_eventscript_call allowed_calls[] = {
 837                         CTDB_EVENT_INIT,
 838                         CTDB_EVENT_SETUP,
 839                         CTDB_EVENT_START_RECOVERY,
 840                         CTDB_EVENT_SHUTDOWN,
 841                         CTDB_EVENT_RELEASE_IP,
 842                         CTDB_EVENT_IPREALLOCATED,
 843                 };
 844                 int i;
 845                 for (i=0;i<ARRAY_SIZE(allowed_calls);i++) {
 846                         if (call == allowed_calls[i]) break;
 847                 }
 848                 if (i == ARRAY_SIZE(allowed_calls)) {
 849                         DEBUG(DEBUG_ERR,("Refusing to run event scripts call '%s' while in recovery\n",
 850                                  ctdb_eventscript_call_names[call]));
 851                         return -1;
 852                 }
 853         }
 854
 855         /* Kill off any running monitor events to run this event. */
 856         if (ctdb->current_monitor) {
 857                 struct ctdb_event_script_state *ms = talloc_get_type(ctdb->current_monitor, struct ctdb_event_script_state);
 858
 859                 /* Cancel current monitor callback state only if monitoring
 860                  * context ctdb->monitor->monitor_context has not been freed */
 861                 if (ms->callback != NULL && !ctdb_stopped_monitoring(ctdb)) {
 862                         ms->callback->fn(ctdb, -ECANCELED, ms->callback->private_data);
 863                         talloc_free(ms->callback);
 864                 }
 865
 866                 /* Discard script status so we don't save to last_status */
 867                 talloc_free(ctdb->current_monitor->scripts);
 868                 ctdb->current_monitor->scripts = NULL;
 869                 talloc_free(ctdb->current_monitor);
 870                 ctdb->current_monitor = NULL;
 871         }
 872
 873         state = talloc(ctdb->event_script_ctx, struct ctdb_event_script_state);
 874         CTDB_NO_MEMORY(ctdb, state);
 875
 876         /* The callback isn't done if the context is freed. */
 877         state->callback = talloc(mem_ctx, struct event_script_callback);
 878         CTDB_NO_MEMORY(ctdb, state->callback);
 879         DLIST_ADD(ctdb->script_callbacks, state->callback);
 880         talloc_set_destructor(state->callback, remove_callback);
 881         state->callback->ctdb         = ctdb;
 882         state->callback->fn           = callback;
 883         state->callback->private_data = private_data;
 884
 885         state->ctdb = ctdb;
 886         state->call = call;
 887         state->options = talloc_vasprintf(state, fmt, ap);
 888         state->timeout = timeval_set(ctdb->tunable.script_timeout, 0);
 889         state->scripts = NULL;
 890         if (state->options == NULL) {
 891                 DEBUG(DEBUG_ERR, (__location__ " could not allocate state->options\n"));
 892                 talloc_free(state);
 893                 return -1;
 894         }
 895         if (!check_options(state->call, state->options)) {
 896                 DEBUG(DEBUG_ERR, ("Bad eventscript options '%s' for %s\n",
 897                                   ctdb_eventscript_call_names[state->call], state->options));
 898                 talloc_free(state);
 899                 return -1;
 900         }
 901
 902         DEBUG(DEBUG_INFO,(__location__ " Starting eventscript %s %s\n",
 903                           ctdb_eventscript_call_names[state->call],
 904                           state->options));
 905
 906         /* This is not a child of state, since we save it in destructor. */
 907         state->scripts = ctdb_get_script_list(ctdb, ctdb);
 908         if (state->scripts == NULL) {
 909                 talloc_free(state);
 910                 return -1;
 911         }
 912         state->current = 0;
 913         state->child = 0;
 914
 915         if (call == CTDB_EVENT_MONITOR) {
 916                 ctdb->current_monitor = state;
 917         }
 918
 919         talloc_set_destructor(state, event_script_destructor);
 920
 921         /* Nothing to do? */
 922         if (state->scripts->num_scripts == 0) {
 923                 talloc_free(state);
 924                 return 0;
 925         }
 926
 927         state->scripts->scripts[0].status = fork_child_for_script(ctdb, state);
 928         if (state->scripts->scripts[0].status != 0) {
 929                 /* Callback is called from destructor, with fail result. */
 930                 talloc_free(state);
 931                 return 0;
 932         }
 933
 934         if (!timeval_is_zero(&state->timeout)) {
 935                 event_add_timed(ctdb->ev, state, timeval_current_ofs(state->timeout.tv_sec, state->timeout.tv_usec), ctdb_event_script_timeout, state);
 936         } else {
 937                 DEBUG(DEBUG_ERR, (__location__ " eventscript %s %s called with no timeout\n",
 938                                   ctdb_eventscript_call_names[state->call],
 939                                   state->options));
 940         }
 941
 942         return 0;
 943 }
 944
 945
 946 /*
 947   run the event script in the background, calling the callback when
 948   finished.  If mem_ctx is freed, callback will never be called.
 949  */
 950 int ctdb_event_script_callback(struct ctdb_context *ctdb,
 951                                TALLOC_CTX *mem_ctx,
 952                                void (*callback)(struct ctdb_context *, int, void *),
 953                                void *private_data,
 954                                enum ctdb_eventscript_call call,
 955                                const char *fmt, ...)
 956 {
 957         va_list ap;
 958         int ret;
 959
 960         va_start(ap, fmt);
 961         ret = ctdb_event_script_callback_v(ctdb, mem_ctx, callback, private_data, call, fmt, ap);
 962         va_end(ap);
 963
 964         return ret;
 965 }
 966
 967
 968 struct callback_status {
 969         bool done;
 970         int status;
 971 };
 972
 973 /*
 974   called when ctdb_event_script() finishes
 975  */
 976 static void event_script_callback(struct ctdb_context *ctdb, int status, void *private_data)
 977 {
 978         struct callback_status *s = (struct callback_status *)private_data;
 979         s->done = true;
 980         s->status = status;
 981 }
 982
 983 /*
 984   run the event script, waiting for it to complete. Used when the caller
 985   doesn't want to continue till the event script has finished.
 986  */
 987 int ctdb_event_script_args(struct ctdb_context *ctdb, enum ctdb_eventscript_call call,
 988                            const char *fmt, ...)
 989 {
 990         va_list ap;
 991         int ret;
 992         struct callback_status status;
 993
 994         va_start(ap, fmt);
 995         ret = ctdb_event_script_callback_v(ctdb, ctdb,
 996                         event_script_callback, &status, call, fmt, ap);
 997         va_end(ap);
 998         if (ret != 0) {
 999                 return ret;
1000         }
1001
1002         status.status = -1;
1003         status.done = false;
1004
1005         while (status.done == false && event_loop_once(ctdb->ev) == 0) /* noop */;
1006
1007         if (status.status == -ETIME) {
1008                 DEBUG(DEBUG_ERR, (__location__ " eventscript for '%s' timedout."
1009                                   " Immediately banning ourself for %d seconds\n",
1010                                   ctdb_eventscript_call_names[call],
1011                                   ctdb->tunable.recovery_ban_period));
1012
1013                 /* Don't ban self if CTDB is starting up or shutting down */
1014                 if (call != CTDB_EVENT_INIT && call != CTDB_EVENT_SHUTDOWN) {
1015                         ctdb_ban_self(ctdb);
1016                 }
1017         }
1018
1019         return status.status;
1020 }
1021
1022 int ctdb_event_script(struct ctdb_context *ctdb, enum ctdb_eventscript_call call)
1023 {
1024         /* GCC complains about empty format string, so use %s and "". */
1025         return ctdb_event_script_args(ctdb, call, "%s", "");
1026 }
1027
1028 struct eventscript_callback_state {
1029         struct ctdb_req_control *c;
1030 };
1031
1032 /*
1033   called when a forced eventscript run has finished
1034  */
1035 static void run_eventscripts_callback(struct ctdb_context *ctdb, int status,
1036                                  void *private_data)
1037 {
1038         struct eventscript_callback_state *state =
1039                 talloc_get_type(private_data, struct eventscript_callback_state);
1040
1041         ctdb_enable_monitoring(ctdb);
1042
1043         if (status != 0) {
1044                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts\n"));
1045         }
1046
1047         ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
1048         /* This will free the struct ctdb_event_script_state we are in! */
1049         talloc_free(state);
1050         return;
1051 }
1052
1053
1054 /* Returns rest of string, or NULL if no match. */
1055 static const char *get_call(const char *p, enum ctdb_eventscript_call *call)
1056 {
1057         unsigned int len;
1058
1059         /* Skip any initial whitespace. */
1060         p += strspn(p, " \t");
1061
1062         /* See if we match any. */
1063         for (*call = 0; *call < CTDB_EVENT_MAX; (*call)++) {
1064                 len = strlen(ctdb_eventscript_call_names[*call]);
1065                 if (strncmp(p, ctdb_eventscript_call_names[*call], len) == 0) {
1066                         /* If end of string or whitespace, we're done. */
1067                         if (strcspn(p + len, " \t") == 0) {
1068                                 return p + len;
1069                         }
1070                 }
1071         }
1072         return NULL;
1073 }
1074
1075 /*
1076   A control to force running of the eventscripts from the ctdb client tool
1077 */
1078 int32_t ctdb_run_eventscripts(struct ctdb_context *ctdb,
1079                 struct ctdb_req_control *c,
1080                 TDB_DATA indata, bool *async_reply)
1081 {
1082         int ret;
1083         struct eventscript_callback_state *state;
1084         const char *options;
1085         enum ctdb_eventscript_call call;
1086
1087         /* Figure out what call they want. */
1088         options = get_call((const char *)indata.dptr, &call);
1089         if (!options) {
1090                 DEBUG(DEBUG_ERR, (__location__ " Invalid event name \"%s\"\n", (const char *)indata.dptr));
1091                 return -1;
1092         }
1093
1094         if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
1095                 DEBUG(DEBUG_ERR, (__location__ " Aborted running eventscript \"%s\" while in RECOVERY mode\n", indata.dptr));
1096                 return -1;
1097         }
1098
1099         state = talloc(ctdb->event_script_ctx, struct eventscript_callback_state);
1100         CTDB_NO_MEMORY(ctdb, state);
1101
1102         state->c = talloc_steal(state, c);
1103
1104         DEBUG(DEBUG_NOTICE,("Running eventscripts with arguments %s\n", indata.dptr));
1105
1106         ctdb_disable_monitoring(ctdb);
1107
1108         ret = ctdb_event_script_callback(ctdb,
1109                          state, run_eventscripts_callback, state,
1110                          call, "%s", options);
1111
1112         if (ret != 0) {
1113                 ctdb_enable_monitoring(ctdb);
1114                 DEBUG(DEBUG_ERR,(__location__ " Failed to run eventscripts with arguments %s\n", indata.dptr));
1115                 talloc_free(state);
1116                 return -1;
1117         }
1118
1119         /* tell ctdb_control.c that we will be replying asynchronously */
1120         *async_reply = true;
1121
1122         return 0;
1123 }
1124
1125
1126
1127 int32_t ctdb_control_enable_script(struct ctdb_context *ctdb, TDB_DATA indata)
1128 {
1129         const char *script;
1130         struct stat st;
1131         char *filename;
1132         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1133
1134         script = (char *)indata.dptr;
1135         if (indata.dsize == 0) {
1136                 DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
1137                 talloc_free(tmp_ctx);
1138                 return -1;
1139         }
1140         if (indata.dptr[indata.dsize - 1] != '\0') {
1141                 DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
1142                 talloc_free(tmp_ctx);
1143                 return -1;
1144         }
1145         if (index(script,'/') != NULL) {
1146                 DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to enable script %s\n", script));
1147                 talloc_free(tmp_ctx);
1148                 return -1;
1149         }
1150
1151
1152         if (stat(ctdb->event_script_dir, &st) != 0 &&
1153             errno == ENOENT) {
1154                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
1155                 talloc_free(tmp_ctx);
1156                 return -1;
1157         }
1158
1159
1160         filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
1161         if (filename == NULL) {
1162                 DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
1163                 talloc_free(tmp_ctx);
1164                 return -1;
1165         }
1166
1167         if (stat(filename, &st) != 0) {
1168                 DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to enable script.\n", filename));
1169                 talloc_free(tmp_ctx);
1170                 return -1;
1171         }
1172
1173         if (chmod(filename, st.st_mode | S_IXUSR) == -1) {
1174                 DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to enable script.\n", filename));
1175                 talloc_free(tmp_ctx);
1176                 return -1;
1177         }
1178
1179         talloc_free(tmp_ctx);
1180         return 0;
1181 }
1182
1183 int32_t ctdb_control_disable_script(struct ctdb_context *ctdb, TDB_DATA indata)
1184 {
1185         const char *script;
1186         struct stat st;
1187         char *filename;
1188         TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1189
1190         script = (char *)indata.dptr;
1191         if (indata.dsize == 0) {
1192                 DEBUG(DEBUG_ERR,(__location__ " No script specified.\n"));
1193                 talloc_free(tmp_ctx);
1194                 return -1;
1195         }
1196         if (indata.dptr[indata.dsize - 1] != '\0') {
1197                 DEBUG(DEBUG_ERR,(__location__ " String is not null terminated.\n"));
1198                 talloc_free(tmp_ctx);
1199                 return -1;
1200         }
1201         if (index(script,'/') != NULL) {
1202                 DEBUG(DEBUG_ERR,(__location__ " Script name contains '/'. Failed to disable script %s\n", script));
1203                 talloc_free(tmp_ctx);
1204                 return -1;
1205         }
1206
1207
1208         if (stat(ctdb->event_script_dir, &st) != 0 &&
1209             errno == ENOENT) {
1210                 DEBUG(DEBUG_CRIT,("No event script directory found at '%s'\n", ctdb->event_script_dir));
1211                 talloc_free(tmp_ctx);
1212                 return -1;
1213         }
1214
1215
1216         filename = talloc_asprintf(tmp_ctx, "%s/%s", ctdb->event_script_dir, script);
1217         if (filename == NULL) {
1218                 DEBUG(DEBUG_ERR,(__location__ " Failed to create script path\n"));
1219                 talloc_free(tmp_ctx);
1220                 return -1;
1221         }
1222
1223         if (stat(filename, &st) != 0) {
1224                 DEBUG(DEBUG_ERR,("Could not stat event script %s. Failed to disable script.\n", filename));
1225                 talloc_free(tmp_ctx);
1226                 return -1;
1227         }
1228
1229         if (chmod(filename, st.st_mode & ~(S_IXUSR|S_IXGRP|S_IXOTH)) == -1) {
1230                 DEBUG(DEBUG_ERR,("Could not chmod %s. Failed to disable script.\n", filename));
1231                 talloc_free(tmp_ctx);
1232                 return -1;
1233         }
1234
1235         talloc_free(tmp_ctx);
1236         return 0;
1237 }