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