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