git.ozlabs.org Git - ccan/blob - ccan/tdb2/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 <ccan/tdb2/tdb2.h>
   6 #include <stdlib.h>
   7 #include <err.h>
   8 #include <getopt.h>
   9 #include <stdarg.h>
  10 #include <stdio.h>
  11 #include <string.h>
  12 #include <errno.h>
  13 #include <unistd.h>
  14 #include <sys/types.h>
  15 #include <fcntl.h>
  16 #include <time.h>
  17 #include <sys/wait.h>
  18
  19 //#define REOPEN_PROB 30
  20 #define DELETE_PROB 8
  21 #define STORE_PROB 4
  22 #define APPEND_PROB 6
  23 //#define TRANSACTION_PROB 10
  24 //#define TRANSACTION_PREPARE_PROB 2
  25 #define LOCKSTORE_PROB 5
  26 #define TRAVERSE_PROB 20
  27 #define TRAVERSE_READ_PROB 20
  28 #define TRAVERSE_MOD_PROB 100
  29 #define TRAVERSE_ABORT_PROB 500
  30 #define CULL_PROB 100
  31 #define KEYLEN 3
  32 #define DATALEN 100
  33
  34 static struct tdb_context *db;
  35 static int in_transaction;
  36 static int in_traverse;
  37 static int error_count;
  38 #if TRANSACTION_PROB
  39 static int always_transaction = 0;
  40 #endif
  41 static int loopnum;
  42 static int count_pipe;
  43 static union tdb_attribute log_attr;
  44 static union tdb_attribute seed_attr;
  45
  46 #ifdef PRINTF_FMT
  47 static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, void *private, const char *format, ...) PRINTF_FMT(4,5);
  48 #endif
  49 static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, void *private, const char *format, ...)
  50 {
  51         va_list ap;
  52
  53         if (level != TDB_DEBUG_TRACE)
  54                 error_count++;
  55
  56         va_start(ap, format);
  57         vfprintf(stdout, format, ap);
  58         va_end(ap);
  59         fflush(stdout);
  60 #if 0
  61         if (level != TDB_DEBUG_TRACE) {
  62                 char *ptr;
  63                 signal(SIGUSR1, SIG_IGN);
  64                 asprintf(&ptr,"xterm -e gdb /proc/%d/exe %d", getpid(), getpid());
  65                 system(ptr);
  66                 free(ptr);
  67         }
  68 #endif
  69 }
  70
  71 #include "../private.h"
  72
  73 static void segv_handler(int signal, siginfo_t *info, void *p)
  74 {
  75         char string[100];
  76
  77         sprintf(string, "%u: death at %p (map_ptr %p, map_size %llu)\n",
  78                 getpid(), info->si_addr, db->map_ptr, db->map_size);
  79         write(2, string, strlen(string));
  80         sleep(60);
  81         _exit(11);
  82 }
  83
  84 static void fatal(const char *why)
  85 {
  86         perror(why);
  87         error_count++;
  88 }
  89
  90 static char *randbuf(int len)
  91 {
  92         char *buf;
  93         int i;
  94         buf = (char *)malloc(len+1);
  95
  96         for (i=0;i<len;i++) {
  97                 buf[i] = 'a' + (rand() % 26);
  98         }
  99         buf[i] = 0;
 100         return buf;
 101 }
 102
 103 static void addrec_db(void);
 104 static int modify_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 105                            void *state)
 106 {
 107 #if CULL_PROB
 108         if (random() % CULL_PROB == 0) {
 109                 tdb_delete(tdb, key);
 110         }
 111 #endif
 112
 113 #if TRAVERSE_MOD_PROB
 114         if (random() % TRAVERSE_MOD_PROB == 0) {
 115                 addrec_db();
 116         }
 117 #endif
 118
 119 #if TRAVERSE_ABORT_PROB
 120         if (random() % TRAVERSE_ABORT_PROB == 0)
 121                 return 1;
 122 #endif
 123
 124         return 0;
 125 }
 126
 127 static void addrec_db(void)
 128 {
 129         int klen, dlen;
 130         char *k, *d;
 131         TDB_DATA key, data;
 132
 133         klen = 1 + (rand() % KEYLEN);
 134         dlen = 1 + (rand() % DATALEN);
 135
 136         k = randbuf(klen);
 137         d = randbuf(dlen);
 138
 139         key.dptr = (unsigned char *)k;
 140         key.dsize = klen+1;
 141
 142         data.dptr = (unsigned char *)d;
 143         data.dsize = dlen+1;
 144
 145 #if REOPEN_PROB
 146         if (in_traverse == 0 && in_transaction == 0 && random() % REOPEN_PROB == 0) {
 147                 tdb_reopen_all(0);
 148                 goto next;
 149         }
 150 #endif
 151
 152 #if TRANSACTION_PROB
 153         if (in_traverse == 0 && in_transaction == 0 && (always_transaction || random() % TRANSACTION_PROB == 0)) {
 154                 if (tdb_transaction_start(db) != 0) {
 155                         fatal("tdb_transaction_start failed");
 156                 }
 157                 in_transaction++;
 158                 goto next;
 159         }
 160         if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 161                 if (random() % TRANSACTION_PREPARE_PROB == 0) {
 162                         if (tdb_transaction_prepare_commit(db) != 0) {
 163                                 fatal("tdb_transaction_prepare_commit failed");
 164                         }
 165                 }
 166                 if (tdb_transaction_commit(db) != 0) {
 167                         fatal("tdb_transaction_commit failed");
 168                 }
 169                 in_transaction--;
 170                 goto next;
 171         }
 172
 173         if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 174                 if (tdb_transaction_cancel(db) != 0) {
 175                         fatal("tdb_transaction_cancel failed");
 176                 }
 177                 in_transaction--;
 178                 goto next;
 179         }
 180 #endif
 181
 182 #if DELETE_PROB
 183         if (random() % DELETE_PROB == 0) {
 184                 tdb_delete(db, key);
 185                 goto next;
 186         }
 187 #endif
 188
 189 #if STORE_PROB
 190         if (random() % STORE_PROB == 0) {
 191                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 192                         fatal("tdb_store failed");
 193                 }
 194                 goto next;
 195         }
 196 #endif
 197
 198 #if APPEND_PROB
 199         if (random() % APPEND_PROB == 0) {
 200                 if (tdb_append(db, key, data) != 0) {
 201                         fatal("tdb_append failed");
 202                 }
 203                 goto next;
 204         }
 205 #endif
 206
 207 #if LOCKSTORE_PROB
 208         if (random() % LOCKSTORE_PROB == 0) {
 209                 tdb_chainlock(db, key);
 210                 data = tdb_fetch(db, key);
 211                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 212                         fatal("tdb_store failed");
 213                 }
 214                 if (data.dptr) free(data.dptr);
 215                 tdb_chainunlock(db, key);
 216                 goto next;
 217         }
 218 #endif
 219
 220 #if TRAVERSE_PROB
 221         /* FIXME: recursive traverses break transactions? */
 222         if (in_traverse == 0 && random() % TRAVERSE_PROB == 0) {
 223                 in_traverse++;
 224                 tdb_traverse(db, modify_traverse, NULL);
 225                 in_traverse--;
 226                 goto next;
 227         }
 228 #endif
 229
 230 #if TRAVERSE_READ_PROB
 231         if (in_traverse == 0 && random() % TRAVERSE_READ_PROB == 0) {
 232                 in_traverse++;
 233                 tdb_traverse_read(db, NULL, NULL);
 234                 in_traverse--;
 235                 goto next;
 236         }
 237 #endif
 238
 239         data = tdb_fetch(db, key);
 240         if (data.dptr) free(data.dptr);
 241
 242 next:
 243         free(k);
 244         free(d);
 245 }
 246
 247 static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 248                        void *state)
 249 {
 250         tdb_delete(tdb, key);
 251         return 0;
 252 }
 253
 254 static void usage(void)
 255 {
 256         printf("Usage: tdbtorture"
 257 #if TRANSACTION_PROB
 258                " [-t]"
 259 #endif
 260                " [-k] [-n NUM_PROCS] [-l NUM_LOOPS] [-s SEED]\n");
 261         exit(0);
 262 }
 263
 264 static void send_count_and_suicide(int sig)
 265 {
 266         /* This ensures our successor can continue where we left off. */
 267         write(count_pipe, &loopnum, sizeof(loopnum));
 268         /* This gives a unique signature. */
 269         kill(getpid(), SIGUSR2);
 270 }
 271
 272 static int run_child(int i, int seed, unsigned num_loops, unsigned start)
 273 {
 274         struct sigaction act = { .sa_sigaction = segv_handler,
 275                                  .sa_flags = SA_SIGINFO };
 276         sigaction(11, &act, NULL);
 277
 278         db = tdb_open("torture.tdb", TDB_DEFAULT, O_RDWR | O_CREAT, 0600,
 279                       &log_attr);
 280         if (!db) {
 281                 fatal("db open failed");
 282         }
 283
 284 #if 0
 285         if (i == 0) {
 286                 printf("pid %i\n", getpid());
 287                 sleep(9);
 288         } else
 289                 sleep(10);
 290 #endif
 291
 292         srand(seed + i);
 293         srandom(seed + i);
 294
 295         /* Set global, then we're ready to handle being killed. */
 296         loopnum = start;
 297         signal(SIGUSR1, send_count_and_suicide);
 298
 299         for (;loopnum<num_loops && error_count == 0;loopnum++) {
 300                 addrec_db();
 301         }
 302
 303         if (error_count == 0) {
 304                 tdb_traverse_read(db, NULL, NULL);
 305 #if TRANSACTION_PROB
 306                 if (always_transaction) {
 307                         while (in_transaction) {
 308                                 tdb_transaction_cancel(db);
 309                                 in_transaction--;
 310                         }
 311                         if (tdb_transaction_start(db) != 0)
 312                                 fatal("tdb_transaction_start failed");
 313                 }
 314 #endif
 315                 tdb_traverse(db, traverse_fn, NULL);
 316                 tdb_traverse(db, traverse_fn, NULL);
 317
 318 #if TRANSACTION_PROB
 319                 if (always_transaction) {
 320                         if (tdb_transaction_commit(db) != 0)
 321                                 fatal("tdb_transaction_commit failed");
 322                 }
 323 #endif
 324         }
 325
 326         tdb_close(db);
 327
 328         return (error_count < 100 ? error_count : 100);
 329 }
 330
 331 int main(int argc, char * const *argv)
 332 {
 333         int i, seed = -1;
 334         int num_loops = 5000;
 335         int num_procs = 3;
 336         int c, pfds[2];
 337         extern char *optarg;
 338         pid_t *pids;
 339         int kill_random = 0;
 340         int *done;
 341
 342         log_attr.base.attr = TDB_ATTRIBUTE_LOG;
 343         log_attr.base.next = &seed_attr;
 344         log_attr.log.log_fn = tdb_log;
 345         seed_attr.base.attr = TDB_ATTRIBUTE_SEED;
 346
 347         while ((c = getopt(argc, argv, "n:l:s:thk")) != -1) {
 348                 switch (c) {
 349                 case 'n':
 350                         num_procs = strtol(optarg, NULL, 0);
 351                         break;
 352                 case 'l':
 353                         num_loops = strtol(optarg, NULL, 0);
 354                         break;
 355                 case 's':
 356                         seed = strtol(optarg, NULL, 0);
 357                         break;
 358                 case 't':
 359 #if TRANSACTION_PROB
 360                         always_transaction = 1;
 361 #else
 362                         fprintf(stderr, "Transactions not supported\n");
 363                         usage();
 364 #endif
 365                         break;
 366                 case 'k':
 367                         kill_random = 1;
 368                         break;
 369                 default:
 370                         usage();
 371                 }
 372         }
 373
 374         unlink("torture.tdb");
 375
 376         if (seed == -1) {
 377                 seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
 378         }
 379         seed_attr.seed.seed = (((uint64_t)seed) << 32) | seed;
 380
 381         if (num_procs == 1 && !kill_random) {
 382                 /* Don't fork for this case, makes debugging easier. */
 383                 error_count = run_child(0, seed, num_loops, 0);
 384                 goto done;
 385         }
 386
 387         pids = (pid_t *)calloc(sizeof(pid_t), num_procs);
 388         done = (int *)calloc(sizeof(int), num_procs);
 389
 390         if (pipe(pfds) != 0) {
 391                 perror("Creating pipe");
 392                 exit(1);
 393         }
 394         count_pipe = pfds[1];
 395
 396         for (i=0;i<num_procs;i++) {
 397                 if ((pids[i]=fork()) == 0) {
 398                         close(pfds[0]);
 399                         if (i == 0) {
 400                                 printf("testing with %d processes, %d loops, seed=%d%s\n",
 401                                        num_procs, num_loops, seed,
 402 #if TRANSACTION_PROB
 403                                        always_transaction ? " (all within transactions)" : ""
 404 #else
 405                                        ""
 406 #endif
 407                                         );
 408                         }
 409                         exit(run_child(i, seed, num_loops, 0));
 410                 }
 411         }
 412
 413         while (num_procs) {
 414                 int status, j;
 415                 pid_t pid;
 416
 417                 if (error_count != 0) {
 418                         /* try and stop the test on any failure */
 419                         for (j=0;j<num_procs;j++) {
 420                                 if (pids[j] != 0) {
 421                                         kill(pids[j], SIGTERM);
 422                                 }
 423                         }
 424                 }
 425
 426                 pid = waitpid(-1, &status, kill_random ? WNOHANG : 0);
 427                 if (pid == 0) {
 428                         struct timespec ts;
 429
 430                         /* Sleep for 1/10 second. */
 431                         ts.tv_sec = 0;
 432                         ts.tv_nsec = 100000000;
 433                         nanosleep(&ts, NULL);
 434
 435                         /* Kill someone. */
 436                         kill(pids[random() % num_procs], SIGUSR1);
 437                         continue;
 438                 }
 439
 440                 if (pid == -1) {
 441                         perror("failed to wait for child\n");
 442                         exit(1);
 443                 }
 444
 445                 for (j=0;j<num_procs;j++) {
 446                         if (pids[j] == pid) break;
 447                 }
 448                 if (j == num_procs) {
 449                         printf("unknown child %d exited!?\n", (int)pid);
 450                         exit(1);
 451                 }
 452                 if (WIFSIGNALED(status)) {
 453                         if (WTERMSIG(status) == SIGUSR2
 454                             || WTERMSIG(status) == SIGUSR1) {
 455                                 /* SIGUSR2 means they wrote to pipe. */
 456                                 if (WTERMSIG(status) == SIGUSR2) {
 457                                         read(pfds[0], &done[j],
 458                                              sizeof(done[j]));
 459                                 }
 460                                 pids[j] = fork();
 461                                 if (pids[j] == 0)
 462                                         exit(run_child(j, seed, num_loops,
 463                                                        done[j]));
 464                                 printf("Restarting child %i for %u-%u\n",
 465                                        j, done[j], num_loops);
 466                                 continue;
 467                         }
 468                         printf("child %d exited with signal %d\n",
 469                                (int)pid, WTERMSIG(status));
 470                         error_count++;
 471                 } else {
 472                         if (WEXITSTATUS(status) != 0) {
 473                                 printf("child %d exited with status %d\n",
 474                                        (int)pid, WEXITSTATUS(status));
 475                                 error_count++;
 476                         }
 477                 }
 478                 memmove(&pids[j], &pids[j+1],
 479                         (num_procs - j - 1)*sizeof(pids[0]));
 480                 num_procs--;
 481         }
 482
 483         free(pids);
 484
 485 done:
 486         if (error_count == 0) {
 487                 db = tdb_open("torture.tdb", TDB_DEFAULT, O_RDWR | O_CREAT,
 488                               0600, &log_attr);
 489                 if (!db) {
 490                         fatal("db open failed");
 491                 }
 492                 if (tdb_check(db, NULL, NULL) == -1) {
 493                         printf("db check failed");
 494                         exit(1);
 495                 }
 496                 tdb_close(db);
 497                 printf("OK\n");
 498         }
 499
 500         return error_count;
 501 }