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