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