lib/tdb/tools/tdbtorture.c

   1 /* this tests tdb by doing lots of ops from several simultaneous
   2    writers - that stresses the locking code.
   3 */
   4
   5 #include "replace.h"
   6 #include "system/time.h"
   7 #include "system/wait.h"
   8 #include "system/filesys.h"
   9 #include "tdb.h"
  10
  11 #ifdef HAVE_GETOPT_H
  12 #include <getopt.h>
  13 #endif
  14
  15
  16 #define REOPEN_PROB 30
  17 #define DELETE_PROB 8
  18 #define STORE_PROB 4
  19 #define APPEND_PROB 6
  20 #define TRANSACTION_PROB 10
  21 #define TRANSACTION_PREPARE_PROB 2
  22 #define LOCKSTORE_PROB 5
  23 #define TRAVERSE_PROB 20
  24 #define TRAVERSE_READ_PROB 20
  25 #define CULL_PROB 100
  26 #define KEYLEN 3
  27 #define DATALEN 100
  28
  29 static struct tdb_context *db;
  30 static int in_transaction;
  31 static int error_count;
  32 static int always_transaction = 0;
  33 static int hash_size = 2;
  34 static int loopnum;
  35 static int count_pipe;
  36 static bool mutex = false;
  37 static struct tdb_logging_context log_ctx;
  38
  39 #ifdef PRINTF_ATTRIBUTE
  40 static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const char *format, ...) PRINTF_ATTRIBUTE(3,4);
  41 #endif
  42 static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const char *format, ...)
  43 {
  44         va_list ap;
  45
  46         /* trace level messages do not indicate an error */
  47         if (level != TDB_DEBUG_TRACE) {
  48                 error_count++;
  49         }
  50
  51         va_start(ap, format);
  52         vfprintf(stdout, format, ap);
  53         va_end(ap);
  54         fflush(stdout);
  55 #if 0
  56         if (level != TDB_DEBUG_TRACE) {
  57                 char *ptr;
  58                 signal(SIGUSR1, SIG_IGN);
  59                 asprintf(&ptr,"xterm -e gdb /proc/%d/exe %d", getpid(), getpid());
  60                 system(ptr);
  61                 free(ptr);
  62         }
  63 #endif
  64 }
  65
  66 static void fatal(const char *why)
  67 {
  68         perror(why);
  69         error_count++;
  70 }
  71
  72 static char *randbuf(int len)
  73 {
  74         char *buf;
  75         int i;
  76         buf = (char *)malloc(len+1);
  77
  78         for (i=0;i<len;i++) {
  79                 buf[i] = 'a' + (rand() % 26);
  80         }
  81         buf[i] = 0;
  82         return buf;
  83 }
  84
  85 static int cull_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
  86                          void *state)
  87 {
  88 #if CULL_PROB
  89         if (random() % CULL_PROB == 0) {
  90                 tdb_delete(tdb, key);
  91         }
  92 #endif
  93         return 0;
  94 }
  95
  96 static void addrec_db(void)
  97 {
  98         int klen, dlen;
  99         char *k, *d;
 100         TDB_DATA key, data;
 101
 102         klen = 1 + (rand() % KEYLEN);
 103         dlen = 1 + (rand() % DATALEN);
 104
 105         k = randbuf(klen);
 106         d = randbuf(dlen);
 107
 108         key.dptr = (unsigned char *)k;
 109         key.dsize = klen+1;
 110
 111         data.dptr = (unsigned char *)d;
 112         data.dsize = dlen+1;
 113
 114 #if REOPEN_PROB
 115         if (in_transaction == 0 && random() % REOPEN_PROB == 0) {
 116                 tdb_reopen_all(0);
 117                 goto next;
 118         }
 119 #endif
 120
 121 #if TRANSACTION_PROB
 122         if (in_transaction == 0 &&
 123             ((tdb_get_flags(db) & TDB_MUTEX_LOCKING) == 0) &&
 124             (always_transaction || random() % TRANSACTION_PROB == 0)) {
 125                 if (tdb_transaction_start(db) != 0) {
 126                         fatal("tdb_transaction_start failed");
 127                 }
 128                 in_transaction++;
 129                 goto next;
 130         }
 131         if (in_transaction && random() % TRANSACTION_PROB == 0) {
 132                 if (random() % TRANSACTION_PREPARE_PROB == 0) {
 133                         if (tdb_transaction_prepare_commit(db) != 0) {
 134                                 fatal("tdb_transaction_prepare_commit failed");
 135                         }
 136                 }
 137                 if (tdb_transaction_commit(db) != 0) {
 138                         fatal("tdb_transaction_commit failed");
 139                 }
 140                 in_transaction--;
 141                 goto next;
 142         }
 143         if (in_transaction && random() % TRANSACTION_PROB == 0) {
 144                 if (tdb_transaction_cancel(db) != 0) {
 145                         fatal("tdb_transaction_cancel failed");
 146                 }
 147                 in_transaction--;
 148                 goto next;
 149         }
 150 #endif
 151
 152 #if DELETE_PROB
 153         if (random() % DELETE_PROB == 0) {
 154                 tdb_delete(db, key);
 155                 goto next;
 156         }
 157 #endif
 158
 159 #if STORE_PROB
 160         if (random() % STORE_PROB == 0) {
 161                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 162                         fatal("tdb_store failed");
 163                 }
 164                 goto next;
 165         }
 166 #endif
 167
 168 #if APPEND_PROB
 169         if (random() % APPEND_PROB == 0) {
 170                 if (tdb_append(db, key, data) != 0) {
 171                         fatal("tdb_append failed");
 172                 }
 173                 goto next;
 174         }
 175 #endif
 176
 177 #if LOCKSTORE_PROB
 178         if (random() % LOCKSTORE_PROB == 0) {
 179                 tdb_chainlock(db, key);
 180                 data = tdb_fetch(db, key);
 181                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 182                         fatal("tdb_store failed");
 183                 }
 184                 if (data.dptr) free(data.dptr);
 185                 tdb_chainunlock(db, key);
 186                 goto next;
 187         }
 188 #endif
 189
 190 #if TRAVERSE_PROB
 191         if (random() % TRAVERSE_PROB == 0) {
 192                 tdb_traverse(db, cull_traverse, NULL);
 193                 goto next;
 194         }
 195 #endif
 196
 197 #if TRAVERSE_READ_PROB
 198         if (random() % TRAVERSE_READ_PROB == 0) {
 199                 tdb_traverse_read(db, NULL, NULL);
 200                 goto next;
 201         }
 202 #endif
 203
 204         data = tdb_fetch(db, key);
 205         if (data.dptr) free(data.dptr);
 206
 207 next:
 208         free(k);
 209         free(d);
 210 }
 211
 212 static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 213                        void *state)
 214 {
 215         tdb_delete(tdb, key);
 216         return 0;
 217 }
 218
 219 static void usage(void)
 220 {
 221         printf("Usage: tdbtorture [-t] [-k] [-m] [-n NUM_PROCS] [-l NUM_LOOPS] [-s SEED] [-H HASH_SIZE]\n");
 222         exit(0);
 223 }
 224
 225 static void send_count_and_suicide(int sig)
 226 {
 227         /* This ensures our successor can continue where we left off. */
 228         write(count_pipe, &loopnum, sizeof(loopnum));
 229         /* This gives a unique signature. */
 230         kill(getpid(), SIGUSR2);
 231 }
 232
 233 static int run_child(const char *filename, int i, int seed, unsigned num_loops, unsigned start)
 234 {
 235         int tdb_flags = TDB_DEFAULT|TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH;
 236
 237         if (mutex) {
 238                 tdb_flags |= TDB_MUTEX_LOCKING;
 239         }
 240
 241         db = tdb_open_ex(filename, hash_size, tdb_flags,
 242                          O_RDWR | O_CREAT, 0600, &log_ctx, NULL);
 243         if (!db) {
 244                 fatal("db open failed");
 245         }
 246
 247         srand(seed + i);
 248         srandom(seed + i);
 249
 250         /* Set global, then we're ready to handle being killed. */
 251         loopnum = start;
 252         signal(SIGUSR1, send_count_and_suicide);
 253
 254         for (;loopnum<num_loops && error_count == 0;loopnum++) {
 255                 addrec_db();
 256         }
 257
 258         if (error_count == 0) {
 259                 tdb_traverse_read(db, NULL, NULL);
 260                 if (always_transaction) {
 261                         while (in_transaction) {
 262                                 tdb_transaction_cancel(db);
 263                                 in_transaction--;
 264                         }
 265                         if (tdb_transaction_start(db) != 0)
 266                                 fatal("tdb_transaction_start failed");
 267                 }
 268                 tdb_traverse(db, traverse_fn, NULL);
 269                 tdb_traverse(db, traverse_fn, NULL);
 270                 if (always_transaction) {
 271                         if (tdb_transaction_commit(db) != 0)
 272                                 fatal("tdb_transaction_commit failed");
 273                 }
 274         }
 275
 276         tdb_close(db);
 277
 278         return (error_count < 100 ? error_count : 100);
 279 }
 280
 281 static char *test_path(const char *filename)
 282 {
 283         const char *prefix = getenv("TEST_DATA_PREFIX");
 284
 285         if (prefix) {
 286                 char *path = NULL;
 287                 int ret;
 288
 289                 ret = asprintf(&path, "%s/%s", prefix, filename);
 290                 if (ret == -1) {
 291                         return NULL;
 292                 }
 293                 return path;
 294         }
 295
 296         return strdup(filename);
 297 }
 298
 299 int main(int argc, char * const *argv)
 300 {
 301         int i, seed = -1;
 302         int num_loops = 5000;
 303         int num_procs = 3;
 304         int c, pfds[2];
 305         extern char *optarg;
 306         pid_t *pids;
 307         int kill_random = 0;
 308         int *done;
 309         char *test_tdb;
 310
 311         log_ctx.log_fn = tdb_log;
 312
 313         while ((c = getopt(argc, argv, "n:l:s:H:thkm")) != -1) {
 314                 switch (c) {
 315                 case 'n':
 316                         num_procs = strtol(optarg, NULL, 0);
 317                         break;
 318                 case 'l':
 319                         num_loops = strtol(optarg, NULL, 0);
 320                         break;
 321                 case 'H':
 322                         hash_size = strtol(optarg, NULL, 0);
 323                         break;
 324                 case 's':
 325                         seed = strtol(optarg, NULL, 0);
 326                         break;
 327                 case 't':
 328                         always_transaction = 1;
 329                         break;
 330                 case 'k':
 331                         kill_random = 1;
 332                         break;
 333                 case 'm':
 334                         mutex = tdb_runtime_check_for_robust_mutexes();
 335                         if (!mutex) {
 336                                 printf("tdb_runtime_check_for_robust_mutexes() returned false\n");
 337                                 exit(1);
 338                         }
 339                         break;
 340                 default:
 341                         usage();
 342                 }
 343         }
 344
 345         test_tdb = test_path("torture.tdb");
 346
 347         unlink(test_tdb);
 348
 349         if (seed == -1) {
 350                 seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
 351         }
 352
 353         printf("Testing with %d processes, %d loops, %d hash_size, seed=%d%s\n",
 354                num_procs, num_loops, hash_size, seed,
 355                (always_transaction ? " (all within transactions)" : ""));
 356
 357         if (num_procs == 1 && !kill_random) {
 358                 /* Don't fork for this case, makes debugging easier. */
 359                 error_count = run_child(test_tdb, 0, seed, num_loops, 0);
 360                 goto done;
 361         }
 362
 363         pids = (pid_t *)calloc(sizeof(pid_t), num_procs);
 364         if (pids == NULL) {
 365                 perror("Unable to allocate memory for pids");
 366                 exit(1);
 367         }
 368         done = (int *)calloc(sizeof(int), num_procs);
 369         if (done == NULL) {
 370                 perror("Unable to allocate memory for done");
 371                 exit(1);
 372         }
 373
 374         if (pipe(pfds) != 0) {
 375                 perror("Creating pipe");
 376                 exit(1);
 377         }
 378         count_pipe = pfds[1];
 379
 380         for (i=0;i<num_procs;i++) {
 381                 if ((pids[i]=fork()) == 0) {
 382                         close(pfds[0]);
 383                         exit(run_child(test_tdb, i, seed, num_loops, 0));
 384                 }
 385         }
 386
 387         while (num_procs) {
 388                 int status, j;
 389                 pid_t pid;
 390
 391                 if (error_count != 0) {
 392                         /* try and stop the test on any failure */
 393                         for (j=0;j<num_procs;j++) {
 394                                 if (pids[j] != 0) {
 395                                         kill(pids[j], SIGTERM);
 396                                 }
 397                         }
 398                 }
 399
 400                 pid = waitpid(-1, &status, kill_random ? WNOHANG : 0);
 401                 if (pid == 0) {
 402                         struct timeval tv;
 403
 404                         /* Sleep for 1/10 second. */
 405                         tv.tv_sec = 0;
 406                         tv.tv_usec = 100000;
 407                         select(0, NULL, NULL, NULL, &tv);
 408
 409                         /* Kill someone. */
 410                         kill(pids[random() % num_procs], SIGUSR1);
 411                         continue;
 412                 }
 413
 414                 if (pid == -1) {
 415                         perror("failed to wait for child\n");
 416                         exit(1);
 417                 }
 418
 419                 for (j=0;j<num_procs;j++) {
 420                         if (pids[j] == pid) break;
 421                 }
 422                 if (j == num_procs) {
 423                         printf("unknown child %d exited!?\n", (int)pid);
 424                         exit(1);
 425                 }
 426                 if (WIFSIGNALED(status)) {
 427                         if (WTERMSIG(status) == SIGUSR2
 428                             || WTERMSIG(status) == SIGUSR1) {
 429                                 /* SIGUSR2 means they wrote to pipe. */
 430                                 if (WTERMSIG(status) == SIGUSR2) {
 431                                         read(pfds[0], &done[j],
 432                                              sizeof(done[j]));
 433                                 }
 434                                 pids[j] = fork();
 435                                 if (pids[j] == 0)
 436                                         exit(run_child(test_tdb, j, seed,
 437                                                        num_loops, done[j]));
 438                                 printf("Restarting child %i for %u-%u\n",
 439                                        j, done[j], num_loops);
 440                                 continue;
 441                         }
 442                         printf("child %d exited with signal %d\n",
 443                                (int)pid, WTERMSIG(status));
 444                         error_count++;
 445                 } else {
 446                         if (WEXITSTATUS(status) != 0) {
 447                                 printf("child %d exited with status %d\n",
 448                                        (int)pid, WEXITSTATUS(status));
 449                                 error_count++;
 450                         }
 451                 }
 452                 memmove(&pids[j], &pids[j+1],
 453                         (num_procs - j - 1)*sizeof(pids[0]));
 454                 num_procs--;
 455         }
 456
 457         free(pids);
 458
 459 done:
 460         if (error_count == 0) {
 461                 int tdb_flags = TDB_DEFAULT;
 462
 463                 if (mutex) {
 464                         tdb_flags |= TDB_NOLOCK;
 465                 }
 466
 467                 db = tdb_open_ex(test_tdb, hash_size, tdb_flags,
 468                                  O_RDWR, 0, &log_ctx, NULL);
 469                 if (!db) {
 470                         fatal("db open failed\n");
 471                         exit(1);
 472                 }
 473                 if (tdb_check(db, NULL, NULL) == -1) {
 474                         printf("db check failed\n");
 475                         exit(1);
 476                 }
 477                 tdb_close(db);
 478                 printf("OK\n");
 479         }
 480
 481         free(test_tdb);
 482         return error_count;
 483 }