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