lib/tdb2/tools/tdb2torture.c

   1 /* this tests tdb by doing lots of ops from several simultaneous
   2    writers - that stresses the locking code.
   3 */
   4
   5 #include "config.h"
   6 #include "tdb2.h"
   7 #include <ccan/err/err.h>
   8 #ifdef HAVE_LIBREPLACE
   9 #include <replace.h>
  10 #else
  11 #include <stdlib.h>
  12 #include <getopt.h>
  13 #include <stdarg.h>
  14 #include <stdio.h>
  15 #include <string.h>
  16 #include <errno.h>
  17 #include <unistd.h>
  18 #include <sys/types.h>
  19 #include <fcntl.h>
  20 #include <time.h>
  21 #include <sys/wait.h>
  22 #endif
  23
  24 /* Currently we default to creating a tdb1.  This will change! */
  25 #define TDB2_IS_DEFAULT false
  26
  27 //#define REOPEN_PROB 30
  28 #define DELETE_PROB 8
  29 #define STORE_PROB 4
  30 #define APPEND_PROB 6
  31 #define TRANSACTION_PROB 10
  32 #define TRANSACTION_PREPARE_PROB 2
  33 #define LOCKSTORE_PROB 5
  34 #define TRAVERSE_PROB 20
  35 #define TRAVERSE_MOD_PROB 100
  36 #define TRAVERSE_ABORT_PROB 500
  37 #define CULL_PROB 100
  38 #define KEYLEN 3
  39 #define DATALEN 100
  40
  41 static struct tdb_context *db;
  42 static int in_transaction;
  43 static int in_traverse;
  44 static int error_count;
  45 #if TRANSACTION_PROB
  46 static int always_transaction = 0;
  47 #endif
  48 static int loopnum;
  49 static int count_pipe;
  50 static union tdb_attribute log_attr;
  51 static union tdb_attribute seed_attr;
  52
  53 static void tdb_log(struct tdb_context *tdb,
  54                     enum tdb_log_level level,
  55                     enum TDB_ERROR ecode,
  56                     const char *message,
  57                     void *data)
  58 {
  59         printf("tdb:%s:%s:%s\n",
  60                tdb_name(tdb), tdb_errorstr(ecode), message);
  61         fflush(stdout);
  62 #if 0
  63         {
  64                 char str[200];
  65                 signal(SIGUSR1, SIG_IGN);
  66                 sprintf(str,"xterm -e gdb /proc/%d/exe %d", getpid(), getpid());
  67                 system(str);
  68         }
  69 #endif
  70 }
  71
  72 #include "../private.h"
  73
  74 static void segv_handler(int sig, siginfo_t *info, void *p)
  75 {
  76         char string[100];
  77
  78         sprintf(string, "%u: death at %p (map_ptr %p, map_size %zu)\n",
  79                 getpid(), info->si_addr, db->file->map_ptr,
  80                 (size_t)db->file->map_size);
  81         if (write(2, string, strlen(string)) > 0)
  82                 sleep(60);
  83         _exit(11);
  84 }
  85
  86 static void fatal(struct tdb_context *tdb, const char *why)
  87 {
  88         fprintf(stderr, "%u:%s:%s\n", getpid(), why,
  89                 tdb ? tdb_errorstr(tdb_error(tdb)) : "(no tdb)");
  90         error_count++;
  91 }
  92
  93 static char *randbuf(int len)
  94 {
  95         char *buf;
  96         int i;
  97         buf = (char *)malloc(len+1);
  98
  99         for (i=0;i<len;i++) {
 100                 buf[i] = 'a' + (rand() % 26);
 101         }
 102         buf[i] = 0;
 103         return buf;
 104 }
 105
 106 static void addrec_db(void);
 107 static int modify_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 108                            void *state)
 109 {
 110 #if CULL_PROB
 111         if (random() % CULL_PROB == 0) {
 112                 tdb_delete(tdb, key);
 113         }
 114 #endif
 115
 116 #if TRAVERSE_MOD_PROB
 117         if (random() % TRAVERSE_MOD_PROB == 0) {
 118                 addrec_db();
 119         }
 120 #endif
 121
 122 #if TRAVERSE_ABORT_PROB
 123         if (random() % TRAVERSE_ABORT_PROB == 0)
 124                 return 1;
 125 #endif
 126
 127         return 0;
 128 }
 129
 130 static void addrec_db(void)
 131 {
 132         int klen, dlen;
 133         char *k, *d;
 134         TDB_DATA key, data;
 135
 136         klen = 1 + (rand() % KEYLEN);
 137         dlen = 1 + (rand() % DATALEN);
 138
 139         k = randbuf(klen);
 140         d = randbuf(dlen);
 141
 142         key.dptr = (unsigned char *)k;
 143         key.dsize = klen+1;
 144
 145         data.dptr = (unsigned char *)d;
 146         data.dsize = dlen+1;
 147
 148 #if REOPEN_PROB
 149         if (in_traverse == 0 && in_transaction == 0 && random() % REOPEN_PROB == 0) {
 150                 tdb_reopen_all(0);
 151                 goto next;
 152         }
 153 #endif
 154
 155 #if TRANSACTION_PROB
 156         if (in_traverse == 0 && in_transaction == 0 && (always_transaction || random() % TRANSACTION_PROB == 0)) {
 157                 if (tdb_transaction_start(db) != 0) {
 158                         fatal(db, "tdb_transaction_start failed");
 159                 }
 160                 in_transaction++;
 161                 goto next;
 162         }
 163         if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 164                 if (random() % TRANSACTION_PREPARE_PROB == 0) {
 165                         if (tdb_transaction_prepare_commit(db) != 0) {
 166                                 fatal(db, "tdb_transaction_prepare_commit failed");
 167                         }
 168                 }
 169                 if (tdb_transaction_commit(db) != 0) {
 170                         fatal(db, "tdb_transaction_commit failed");
 171                 }
 172                 in_transaction--;
 173                 goto next;
 174         }
 175
 176         if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 177                 tdb_transaction_cancel(db);
 178                 in_transaction--;
 179                 goto next;
 180         }
 181 #endif
 182
 183 #if DELETE_PROB
 184         if (random() % DELETE_PROB == 0) {
 185                 tdb_delete(db, key);
 186                 goto next;
 187         }
 188 #endif
 189
 190 #if STORE_PROB
 191         if (random() % STORE_PROB == 0) {
 192                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 193                         fatal(db, "tdb_store failed");
 194                 }
 195                 goto next;
 196         }
 197 #endif
 198
 199 #if APPEND_PROB
 200         if (random() % APPEND_PROB == 0) {
 201                 if (tdb_append(db, key, data) != 0) {
 202                         fatal(db, "tdb_append failed");
 203                 }
 204                 goto next;
 205         }
 206 #endif
 207
 208 #if LOCKSTORE_PROB
 209         if (random() % LOCKSTORE_PROB == 0) {
 210                 tdb_chainlock(db, key);
 211                 if (tdb_fetch(db, key, &data) != TDB_SUCCESS) {
 212                         data.dsize = 0;
 213                         data.dptr = NULL;
 214                 }
 215                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 216                         fatal(db, "tdb_store failed");
 217                 }
 218                 if (data.dptr) free(data.dptr);
 219                 tdb_chainunlock(db, key);
 220                 goto next;
 221         }
 222 #endif
 223
 224 #if TRAVERSE_PROB
 225         /* FIXME: recursive traverses break transactions? */
 226         if (in_traverse == 0 && random() % TRAVERSE_PROB == 0) {
 227                 in_traverse++;
 228                 tdb_traverse(db, modify_traverse, NULL);
 229                 in_traverse--;
 230                 goto next;
 231         }
 232 #endif
 233
 234         if (tdb_fetch(db, key, &data) == TDB_SUCCESS)
 235                 free(data.dptr);
 236
 237 next:
 238         free(k);
 239         free(d);
 240 }
 241
 242 static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 243                        void *state)
 244 {
 245         tdb_delete(tdb, key);
 246         return 0;
 247 }
 248
 249 static void usage(void)
 250 {
 251         printf("Usage: tdbtorture"
 252 #if TRANSACTION_PROB
 253                " [-t]"
 254 #endif
 255                " [-k] [-n NUM_PROCS] [-l NUM_LOOPS] [-s SEED] [-S]\n");
 256         exit(0);
 257 }
 258
 259 static void send_count_and_suicide(int sig)
 260 {
 261         /* This ensures our successor can continue where we left off. */
 262         if (write(count_pipe, &loopnum, sizeof(loopnum)) != sizeof(loopnum))
 263                 exit(2);
 264         /* This gives a unique signature. */
 265         kill(getpid(), SIGUSR2);
 266 }
 267
 268 static int run_child(const char *filename, int i, int seed, unsigned num_loops,
 269                      unsigned start, int tdb_flags)
 270 {
 271         struct sigaction act = { .sa_sigaction = segv_handler,
 272                                  .sa_flags = SA_SIGINFO };
 273         sigaction(11, &act, NULL);
 274
 275         db = tdb_open(filename, tdb_flags, O_RDWR | O_CREAT, 0600,
 276                       &log_attr);
 277         if (!db) {
 278                 fatal(NULL, "db open failed");
 279         }
 280
 281 #if 0
 282         if (i == 0) {
 283                 printf("pid %i\n", getpid());
 284                 sleep(9);
 285         } else
 286                 sleep(10);
 287 #endif
 288
 289         srand(seed + i);
 290         srandom(seed + i);
 291
 292         /* Set global, then we're ready to handle being killed. */
 293         loopnum = start;
 294         signal(SIGUSR1, send_count_and_suicide);
 295
 296         for (;loopnum<num_loops && error_count == 0;loopnum++) {
 297                 addrec_db();
 298         }
 299
 300         if (error_count == 0) {
 301                 tdb_traverse(db, NULL, NULL);
 302 #if TRANSACTION_PROB
 303                 if (always_transaction) {
 304                         while (in_transaction) {
 305                                 tdb_transaction_cancel(db);
 306                                 in_transaction--;
 307                         }
 308                         if (tdb_transaction_start(db) != 0)
 309                                 fatal(db, "tdb_transaction_start failed");
 310                 }
 311 #endif
 312                 tdb_traverse(db, traverse_fn, NULL);
 313                 tdb_traverse(db, traverse_fn, NULL);
 314
 315 #if TRANSACTION_PROB
 316                 if (always_transaction) {
 317                         if (tdb_transaction_commit(db) != 0)
 318                                 fatal(db, "tdb_transaction_commit failed");
 319                 }
 320 #endif
 321         }
 322
 323         tdb_close(db);
 324
 325         return (error_count < 100 ? error_count : 100);
 326 }
 327
 328 static char *test_path(const char *filename)
 329 {
 330         const char *prefix = getenv("TEST_DATA_PREFIX");
 331
 332         if (prefix) {
 333                 char *path = NULL;
 334                 int ret;
 335
 336                 ret = asprintf(&path, "%s/%s", prefix, filename);
 337                 if (ret == -1) {
 338                         return NULL;
 339                 }
 340                 return path;
 341         }
 342
 343         return strdup(filename);
 344 }
 345
 346 int main(int argc, char * const *argv)
 347 {
 348         int i, seed = -1;
 349         int num_loops = 5000;
 350         int num_procs = 3;
 351         int c, pfds[2];
 352         extern char *optarg;
 353         pid_t *pids;
 354         int kill_random = 0;
 355         int *done;
 356         int tdb_flags = TDB_DEFAULT;
 357         bool tdb2 = TDB2_IS_DEFAULT;
 358         char *test_tdb;
 359
 360         log_attr.base.attr = TDB_ATTRIBUTE_LOG;
 361         log_attr.base.next = &seed_attr;
 362         log_attr.log.fn = tdb_log;
 363         seed_attr.base.attr = TDB_ATTRIBUTE_SEED;
 364         seed_attr.base.next = NULL;
 365
 366         while ((c = getopt(argc, argv, "n:l:s:thkS12")) != -1) {
 367                 switch (c) {
 368                 case 'n':
 369                         num_procs = strtol(optarg, NULL, 0);
 370                         break;
 371                 case 'l':
 372                         num_loops = strtol(optarg, NULL, 0);
 373                         break;
 374                 case 's':
 375                         seed = strtol(optarg, NULL, 0);
 376                         break;
 377                 case 'S':
 378                         tdb_flags = TDB_NOSYNC;
 379                         break;
 380                 case 't':
 381 #if TRANSACTION_PROB
 382                         always_transaction = 1;
 383 #else
 384                         fprintf(stderr, "Transactions not supported\n");
 385                         usage();
 386 #endif
 387                         break;
 388                 case 'k':
 389                         kill_random = 1;
 390                         break;
 391                 case '1':
 392                         tdb2 = false;
 393                         break;
 394                 case '2':
 395                         tdb2 = true;
 396                         break;
 397                 default:
 398                         usage();
 399                 }
 400         }
 401
 402         if (!tdb2) {
 403                 tdb_flags |= TDB_VERSION1;
 404                 /* TDB1 tdbs don't use seed. */
 405                 log_attr.base.next = NULL;
 406         }
 407
 408         test_tdb = test_path("torture.tdb");
 409
 410         unlink(test_tdb);
 411
 412         if (seed == -1) {
 413                 seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
 414         }
 415         seed_attr.seed.seed = (((uint64_t)seed) << 32) | seed;
 416
 417         if (num_procs == 1 && !kill_random) {
 418                 /* Don't fork for this case, makes debugging easier. */
 419                 error_count = run_child(test_tdb, 0, seed, num_loops, 0,
 420                                         tdb_flags);
 421                 goto done;
 422         }
 423
 424         pids = (pid_t *)calloc(sizeof(pid_t), num_procs);
 425         done = (int *)calloc(sizeof(int), num_procs);
 426
 427         if (pipe(pfds) != 0) {
 428                 perror("Creating pipe");
 429                 exit(1);
 430         }
 431         count_pipe = pfds[1];
 432
 433         for (i=0;i<num_procs;i++) {
 434                 if ((pids[i]=fork()) == 0) {
 435                         close(pfds[0]);
 436                         if (i == 0) {
 437                                 printf("testing with %d processes, %d loops, seed=%d%s\n",
 438                                        num_procs, num_loops, seed,
 439 #if TRANSACTION_PROB
 440                                        always_transaction ? " (all within transactions)" : ""
 441 #else
 442                                        ""
 443 #endif
 444                                         );
 445                         }
 446                         exit(run_child(test_tdb, i, seed, num_loops, 0,
 447                                        tdb_flags));
 448                 }
 449         }
 450
 451         while (num_procs) {
 452                 int status, j;
 453                 pid_t pid;
 454
 455                 if (error_count != 0) {
 456                         /* try and stop the test on any failure */
 457                         for (j=0;j<num_procs;j++) {
 458                                 if (pids[j] != 0) {
 459                                         kill(pids[j], SIGTERM);
 460                                 }
 461                         }
 462                 }
 463
 464                 pid = waitpid(-1, &status, kill_random ? WNOHANG : 0);
 465                 if (pid == 0) {
 466                         struct timespec ts;
 467
 468                         /* Sleep for 1/10 second. */
 469                         ts.tv_sec = 0;
 470                         ts.tv_nsec = 100000000;
 471                         nanosleep(&ts, NULL);
 472
 473                         /* Kill someone. */
 474                         kill(pids[random() % num_procs], SIGUSR1);
 475                         continue;
 476                 }
 477
 478                 if (pid == -1) {
 479                         perror("failed to wait for child\n");
 480                         exit(1);
 481                 }
 482
 483                 for (j=0;j<num_procs;j++) {
 484                         if (pids[j] == pid) break;
 485                 }
 486                 if (j == num_procs) {
 487                         printf("unknown child %d exited!?\n", (int)pid);
 488                         exit(1);
 489                 }
 490                 if (WIFSIGNALED(status)) {
 491                         if (WTERMSIG(status) == SIGUSR2
 492                             || WTERMSIG(status) == SIGUSR1) {
 493                                 /* SIGUSR2 means they wrote to pipe. */
 494                                 if (WTERMSIG(status) == SIGUSR2) {
 495                                         if (read(pfds[0], &done[j],
 496                                                  sizeof(done[j]))
 497                                             != sizeof(done[j]))
 498                                                 err(1,
 499                                                     "Short read from child?");
 500                                 }
 501                                 pids[j] = fork();
 502                                 if (pids[j] == 0)
 503                                         exit(run_child(test_tdb, j, seed,
 504                                                        num_loops, done[j],
 505                                                        tdb_flags));
 506                                 printf("Restarting child %i for %u-%u\n",
 507                                        j, done[j], num_loops);
 508                                 continue;
 509                         }
 510                         printf("child %d exited with signal %d\n",
 511                                (int)pid, WTERMSIG(status));
 512                         error_count++;
 513                 } else {
 514                         if (WEXITSTATUS(status) != 0) {
 515                                 printf("child %d exited with status %d\n",
 516                                        (int)pid, WEXITSTATUS(status));
 517                                 error_count++;
 518                         }
 519                 }
 520                 memmove(&pids[j], &pids[j+1],
 521                         (num_procs - j - 1)*sizeof(pids[0]));
 522                 num_procs--;
 523         }
 524
 525         free(pids);
 526
 527 done:
 528         if (error_count == 0) {
 529                 db = tdb_open(test_tdb, TDB_DEFAULT, O_RDWR | O_CREAT,
 530                               0600, &log_attr);
 531                 if (!db) {
 532                         fatal(db, "db open failed");
 533                         exit(1);
 534                 }
 535                 if (tdb_check(db, NULL, NULL) != 0) {
 536                         fatal(db, "db check failed");
 537                         exit(1);
 538                 }
 539                 tdb_close(db);
 540                 printf("OK\n");
 541         }
 542
 543         free(test_tdb);
 544         return error_count;
 545 }