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         va_start(ap, format);
  54         vfprintf(stdout, format, ap);
  55         va_end(ap);
  56         fflush(stdout);
  57 #if 0
  58         {
  59                 char *ptr;
  60                 signal(SIGUSR1, SIG_IGN);
  61                 asprintf(&ptr,"xterm -e gdb /proc/%d/exe %d", getpid(), getpid());
  62                 system(ptr);
  63                 free(ptr);
  64         }
  65 #endif
  66 }
  67
  68 #include "../private.h"
  69
  70 static void segv_handler(int signal, siginfo_t *info, void *p)
  71 {
  72         char string[100];
  73
  74         sprintf(string, "%u: death at %p (map_ptr %p, map_size %llu)\n",
  75                 getpid(), info->si_addr, db->map_ptr, db->map_size);
  76         write(2, string, strlen(string));
  77         sleep(60);
  78         _exit(11);
  79 }
  80
  81 static void fatal(const char *why)
  82 {
  83         perror(why);
  84         error_count++;
  85 }
  86
  87 static char *randbuf(int len)
  88 {
  89         char *buf;
  90         int i;
  91         buf = (char *)malloc(len+1);
  92
  93         for (i=0;i<len;i++) {
  94                 buf[i] = 'a' + (rand() % 26);
  95         }
  96         buf[i] = 0;
  97         return buf;
  98 }
  99
 100 static void addrec_db(void);
 101 static int modify_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 102                            void *state)
 103 {
 104 #if CULL_PROB
 105         if (random() % CULL_PROB == 0) {
 106                 tdb_delete(tdb, key);
 107         }
 108 #endif
 109
 110 #if TRAVERSE_MOD_PROB
 111         if (random() % TRAVERSE_MOD_PROB == 0) {
 112                 addrec_db();
 113         }
 114 #endif
 115
 116 #if TRAVERSE_ABORT_PROB
 117         if (random() % TRAVERSE_ABORT_PROB == 0)
 118                 return 1;
 119 #endif
 120
 121         return 0;
 122 }
 123
 124 static void addrec_db(void)
 125 {
 126         int klen, dlen;
 127         char *k, *d;
 128         TDB_DATA key, data;
 129
 130         klen = 1 + (rand() % KEYLEN);
 131         dlen = 1 + (rand() % DATALEN);
 132
 133         k = randbuf(klen);
 134         d = randbuf(dlen);
 135
 136         key.dptr = (unsigned char *)k;
 137         key.dsize = klen+1;
 138
 139         data.dptr = (unsigned char *)d;
 140         data.dsize = dlen+1;
 141
 142 #if REOPEN_PROB
 143         if (in_traverse == 0 && in_transaction == 0 && random() % REOPEN_PROB == 0) {
 144                 tdb_reopen_all(0);
 145                 goto next;
 146         }
 147 #endif
 148
 149 #if TRANSACTION_PROB
 150         if (in_traverse == 0 && in_transaction == 0 && (always_transaction || random() % TRANSACTION_PROB == 0)) {
 151                 if (tdb_transaction_start(db) != 0) {
 152                         fatal("tdb_transaction_start failed");
 153                 }
 154                 in_transaction++;
 155                 goto next;
 156         }
 157         if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 158                 if (random() % TRANSACTION_PREPARE_PROB == 0) {
 159                         if (tdb_transaction_prepare_commit(db) != 0) {
 160                                 fatal("tdb_transaction_prepare_commit failed");
 161                         }
 162                 }
 163                 if (tdb_transaction_commit(db) != 0) {
 164                         fatal("tdb_transaction_commit failed");
 165                 }
 166                 in_transaction--;
 167                 goto next;
 168         }
 169
 170         if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 171                 tdb_transaction_cancel(db);
 172                 in_transaction--;
 173                 goto next;
 174         }
 175 #endif
 176
 177 #if DELETE_PROB
 178         if (random() % DELETE_PROB == 0) {
 179                 tdb_delete(db, key);
 180                 goto next;
 181         }
 182 #endif
 183
 184 #if STORE_PROB
 185         if (random() % STORE_PROB == 0) {
 186                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 187                         fatal("tdb_store failed");
 188                 }
 189                 goto next;
 190         }
 191 #endif
 192
 193 #if APPEND_PROB
 194         if (random() % APPEND_PROB == 0) {
 195                 if (tdb_append(db, key, data) != 0) {
 196                         fatal("tdb_append failed");
 197                 }
 198                 goto next;
 199         }
 200 #endif
 201
 202 #if LOCKSTORE_PROB
 203         if (random() % LOCKSTORE_PROB == 0) {
 204                 tdb_chainlock(db, key);
 205                 data = tdb_fetch(db, key);
 206                 if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
 207                         fatal("tdb_store failed");
 208                 }
 209                 if (data.dptr) free(data.dptr);
 210                 tdb_chainunlock(db, key);
 211                 goto next;
 212         }
 213 #endif
 214
 215 #if TRAVERSE_PROB
 216         /* FIXME: recursive traverses break transactions? */
 217         if (in_traverse == 0 && random() % TRAVERSE_PROB == 0) {
 218                 in_traverse++;
 219                 tdb_traverse(db, modify_traverse, NULL);
 220                 in_traverse--;
 221                 goto next;
 222         }
 223 #endif
 224
 225 #if TRAVERSE_READ_PROB
 226         if (in_traverse == 0 && random() % TRAVERSE_READ_PROB == 0) {
 227                 in_traverse++;
 228                 tdb_traverse_read(db, NULL, NULL);
 229                 in_traverse--;
 230                 goto next;
 231         }
 232 #endif
 233
 234         data = tdb_fetch(db, key);
 235         if (data.dptr) 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]\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         write(count_pipe, &loopnum, sizeof(loopnum));
 263         /* This gives a unique signature. */
 264         kill(getpid(), SIGUSR2);
 265 }
 266
 267 static int run_child(int i, int seed, unsigned num_loops, unsigned start)
 268 {
 269         struct sigaction act = { .sa_sigaction = segv_handler,
 270                                  .sa_flags = SA_SIGINFO };
 271         sigaction(11, &act, NULL);
 272
 273         db = tdb_open("torture.tdb", TDB_DEFAULT, O_RDWR | O_CREAT, 0600,
 274                       &log_attr);
 275         if (!db) {
 276                 fatal("db open failed");
 277         }
 278
 279 #if 0
 280         if (i == 0) {
 281                 printf("pid %i\n", getpid());
 282                 sleep(9);
 283         } else
 284                 sleep(10);
 285 #endif
 286
 287         srand(seed + i);
 288         srandom(seed + i);
 289
 290         /* Set global, then we're ready to handle being killed. */
 291         loopnum = start;
 292         signal(SIGUSR1, send_count_and_suicide);
 293
 294         for (;loopnum<num_loops && error_count == 0;loopnum++) {
 295                 addrec_db();
 296         }
 297
 298         if (error_count == 0) {
 299                 tdb_traverse_read(db, NULL, NULL);
 300 #if TRANSACTION_PROB
 301                 if (always_transaction) {
 302                         while (in_transaction) {
 303                                 tdb_transaction_cancel(db);
 304                                 in_transaction--;
 305                         }
 306                         if (tdb_transaction_start(db) != 0)
 307                                 fatal("tdb_transaction_start failed");
 308                 }
 309 #endif
 310                 tdb_traverse(db, traverse_fn, NULL);
 311                 tdb_traverse(db, traverse_fn, NULL);
 312
 313 #if TRANSACTION_PROB
 314                 if (always_transaction) {
 315                         if (tdb_transaction_commit(db) != 0)
 316                                 fatal("tdb_transaction_commit failed");
 317                 }
 318 #endif
 319         }
 320
 321         tdb_close(db);
 322
 323         return (error_count < 100 ? error_count : 100);
 324 }
 325
 326 int main(int argc, char * const *argv)
 327 {
 328         int i, seed = -1;
 329         int num_loops = 5000;
 330         int num_procs = 3;
 331         int c, pfds[2];
 332         extern char *optarg;
 333         pid_t *pids;
 334         int kill_random = 0;
 335         int *done;
 336
 337         log_attr.base.attr = TDB_ATTRIBUTE_LOG;
 338         log_attr.base.next = &seed_attr;
 339         log_attr.log.log_fn = tdb_log;
 340         seed_attr.base.attr = TDB_ATTRIBUTE_SEED;
 341
 342         while ((c = getopt(argc, argv, "n:l:s:thk")) != -1) {
 343                 switch (c) {
 344                 case 'n':
 345                         num_procs = strtol(optarg, NULL, 0);
 346                         break;
 347                 case 'l':
 348                         num_loops = strtol(optarg, NULL, 0);
 349                         break;
 350                 case 's':
 351                         seed = strtol(optarg, NULL, 0);
 352                         break;
 353                 case 't':
 354 #if TRANSACTION_PROB
 355                         always_transaction = 1;
 356 #else
 357                         fprintf(stderr, "Transactions not supported\n");
 358                         usage();
 359 #endif
 360                         break;
 361                 case 'k':
 362                         kill_random = 1;
 363                         break;
 364                 default:
 365                         usage();
 366                 }
 367         }
 368
 369         unlink("torture.tdb");
 370
 371         if (seed == -1) {
 372                 seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
 373         }
 374         seed_attr.seed.seed = (((uint64_t)seed) << 32) | seed;
 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 }