git.ozlabs.org Git - ccan/blob - ccan/tdb/tools/replay_trace.c

   1 #include <ccan/tdb/tdb.h>
   2 #include <ccan/grab_file/grab_file.h>
   3 #include <ccan/hash/hash.h>
   4 #include <ccan/talloc/talloc.h>
   5 #include <ccan/str_talloc/str_talloc.h>
   6 #include <ccan/str/str.h>
   7 #include <ccan/list/list.h>
   8 #include <err.h>
   9 #include <ctype.h>
  10 #include <string.h>
  11 #include <unistd.h>
  12 #include <sys/types.h>
  13 #include <sys/wait.h>
  14 #include <sys/time.h>
  15
  16 #define STRINGIFY2(x) #x
  17 #define STRINGIFY(x) STRINGIFY2(x)
  18
  19 /* Avoid mod by zero */
  20 static unsigned int total_keys = 1;
  21
  22 /* #define DEBUG_DEPS 1 */
  23
  24 struct pipe {
  25         int fd[2];
  26 };
  27 static struct pipe *pipes;
  28
  29 static void __attribute__((noreturn)) fail(const char *filename,
  30                                            unsigned int line,
  31                                            const char *fmt, ...)
  32 {
  33         va_list ap;
  34
  35         va_start(ap, fmt);
  36         fprintf(stderr, "%s:%u: FAIL: ", filename, line);
  37         vfprintf(stderr, fmt, ap);
  38         fprintf(stderr, "\n");
  39         va_end(ap);
  40         exit(1);
  41 }
  42
  43 /* Try or die. */
  44 #define try(expr, expect)                                               \
  45         do {                                                            \
  46                 int ret = (expr);                                       \
  47                 if (ret != (expect))                                    \
  48                         fail(filename, i+1, STRINGIFY(expr) "= %i", ret); \
  49         } while (0)
  50
  51 /* Try or imitate results. */
  52 #define unreliable(expr, expect, force, undo)                           \
  53         do {                                                            \
  54                 int ret = expr;                                         \
  55                 if (ret != expect) {                                    \
  56                         fprintf(stderr, "%s:%u: %s gave %i not %i",     \
  57                               filename, i+1, STRINGIFY(expr), ret, expect); \
  58                         if (expect == 0)                                \
  59                                 force;                                  \
  60                         else                                            \
  61                                 undo;                                   \
  62                 }                                                       \
  63         } while (0)
  64
  65 static bool key_eq(TDB_DATA a, TDB_DATA b)
  66 {
  67         if (a.dsize != b.dsize)
  68                 return false;
  69         return memcmp(a.dptr, b.dptr, a.dsize) == 0;
  70 }
  71
  72 /* This is based on the hash algorithm from gdbm */
  73 static unsigned int hash_key(TDB_DATA *key)
  74 {
  75         uint32_t value; /* Used to compute the hash value.  */
  76         uint32_t   i;   /* Used to cycle through random values. */
  77
  78         /* Set the initial value from the key size. */
  79         for (value = 0x238F13AF ^ key->dsize, i=0; i < key->dsize; i++)
  80                 value = (value + (key->dptr[i] << (i*5 % 24)));
  81
  82         return (1103515243 * value + 12345);
  83 }
  84
  85 enum op_type {
  86         OP_TDB_LOCKALL,
  87         OP_TDB_LOCKALL_MARK,
  88         OP_TDB_LOCKALL_UNMARK,
  89         OP_TDB_LOCKALL_NONBLOCK,
  90         OP_TDB_UNLOCKALL,
  91         OP_TDB_LOCKALL_READ,
  92         OP_TDB_LOCKALL_READ_NONBLOCK,
  93         OP_TDB_UNLOCKALL_READ,
  94         OP_TDB_CHAINLOCK,
  95         OP_TDB_CHAINLOCK_NONBLOCK,
  96         OP_TDB_CHAINLOCK_MARK,
  97         OP_TDB_CHAINLOCK_UNMARK,
  98         OP_TDB_CHAINUNLOCK,
  99         OP_TDB_CHAINLOCK_READ,
 100         OP_TDB_CHAINUNLOCK_READ,
 101         OP_TDB_INCREMENT_SEQNUM_NONBLOCK,
 102         OP_TDB_PARSE_RECORD,
 103         OP_TDB_EXISTS,
 104         OP_TDB_STORE,
 105         OP_TDB_APPEND,
 106         OP_TDB_GET_SEQNUM,
 107         OP_TDB_WIPE_ALL,
 108         OP_TDB_TRANSACTION_START,
 109         OP_TDB_TRANSACTION_CANCEL,
 110         OP_TDB_TRANSACTION_COMMIT,
 111         OP_TDB_TRAVERSE_READ_START,
 112         OP_TDB_TRAVERSE_START,
 113         OP_TDB_TRAVERSE_END,
 114         OP_TDB_TRAVERSE,
 115         OP_TDB_FIRSTKEY,
 116         OP_TDB_NEXTKEY,
 117         OP_TDB_FETCH,
 118         OP_TDB_DELETE,
 119 };
 120
 121 struct op {
 122         unsigned int serial;
 123         enum op_type op;
 124         TDB_DATA key;
 125         TDB_DATA data;
 126         int ret;
 127         /* Who is waiting for us? */
 128         struct list_head post;
 129         /* How many are we waiting for? */
 130         unsigned int pre;
 131
 132         union {
 133                 int flag; /* open and store */
 134                 struct traverse *trav; /* traverse start */
 135                 TDB_DATA post_append; /* append */
 136         };
 137 };
 138
 139 static unsigned char hex_char(const char *filename, unsigned int line, char c)
 140 {
 141         c = toupper(c);
 142         if (c >= 'A' && c <= 'F')
 143                 return c - 'A' + 10;
 144         if (c >= '0' && c <= '9')
 145                 return c - '0';
 146         fail(filename, line, "invalid hex character '%c'", c);
 147 }
 148
 149 /* TDB data is <size>:<%02x>* */
 150 static TDB_DATA make_tdb_data(const void *ctx,
 151                               const char *filename, unsigned int line,
 152                               const char *word)
 153 {
 154         TDB_DATA data;
 155         unsigned int i;
 156         const char *p;
 157
 158         if (streq(word, "NULL"))
 159                 return tdb_null;
 160
 161         data.dsize = atoi(word);
 162         data.dptr = talloc_array(ctx, unsigned char, data.dsize);
 163         p = strchr(word, ':');
 164         if (!p)
 165                 fail(filename, line, "invalid tdb data '%s'", word);
 166         p++;
 167         for (i = 0; i < data.dsize; i++)
 168                 data.dptr[i] = hex_char(filename, line, p[i*2])*16
 169                         + hex_char(filename, line, p[i*2+1]);
 170
 171         return data;
 172 }
 173
 174 static void add_op(const char *filename, struct op **op, unsigned int i,
 175                    unsigned int serial, enum op_type type)
 176 {
 177         struct op *new;
 178         *op = talloc_realloc(NULL, *op, struct op, i+1);
 179         new = (*op) + i;
 180         new->op = type;
 181         new->serial = serial;
 182         new->pre = 0;
 183         new->ret = 0;
 184 }
 185
 186 static void op_add_nothing(const char *filename,
 187                            struct op op[], unsigned int op_num, char *words[])
 188 {
 189         if (words[2])
 190                 fail(filename, op_num+1, "Expected no arguments");
 191         op[op_num].key = tdb_null;
 192 }
 193
 194 static void op_add_key(const char *filename,
 195                        struct op op[], unsigned int op_num, char *words[])
 196 {
 197         if (words[2] == NULL || words[3])
 198                 fail(filename, op_num+1, "Expected just a key");
 199
 200         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 201         if (op[op_num].op != OP_TDB_TRAVERSE)
 202                 total_keys++;
 203 }
 204
 205 static void op_add_key_ret(const char *filename,
 206                            struct op op[], unsigned int op_num, char *words[])
 207 {
 208         if (!words[2] || !words[3] || !words[4] || words[5]
 209             || !streq(words[3], "="))
 210                 fail(filename, op_num+1, "Expected <key> = <ret>");
 211         op[op_num].ret = atoi(words[4]);
 212         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 213         /* May only be a unique key if it fails */
 214         if (op[op_num].ret != 0)
 215                 total_keys++;
 216 }
 217
 218 static void op_add_key_data(const char *filename,
 219                             struct op op[], unsigned int op_num, char *words[])
 220 {
 221         if (!words[2] || !words[3] || !words[4] || words[5]
 222             || !streq(words[3], "="))
 223                 fail(filename, op_num+1, "Expected <key> = <data>");
 224         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 225         op[op_num].data = make_tdb_data(op, filename, op_num+1, words[4]);
 226         /* May only be a unique key if it fails */
 227         if (!op[op_num].data.dptr)
 228                 total_keys++;
 229 }
 230
 231 /* <serial> tdb_store <rec> <rec> <flag> = <ret> */
 232 static void op_add_store(const char *filename,
 233                          struct op op[], unsigned int op_num, char *words[])
 234 {
 235         if (!words[2] || !words[3] || !words[4] || !words[5] || !words[6]
 236             || words[7] || !streq(words[5], "="))
 237                 fail(filename, op_num+1, "Expect <key> <data> <flag> = <ret>");
 238
 239         op[op_num].flag = strtoul(words[4], NULL, 0);
 240         op[op_num].ret = atoi(words[6]);
 241         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 242         op[op_num].data = make_tdb_data(op, filename, op_num+1, words[3]);
 243         total_keys++;
 244 }
 245
 246 /* <serial> tdb_append <rec> <rec> = <rec> */
 247 static void op_add_append(const char *filename,
 248                           struct op op[], unsigned int op_num, char *words[])
 249 {
 250         if (!words[2] || !words[3] || !words[4] || !words[5] || words[6]
 251             || !streq(words[4], "="))
 252                 fail(filename, op_num+1, "Expect <key> <data> = <rec>");
 253
 254         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 255         op[op_num].data = make_tdb_data(op, filename, op_num+1, words[3]);
 256         op[op_num].post_append
 257                 = make_tdb_data(op, filename, op_num+1, words[5]);
 258         total_keys++;
 259 }
 260
 261 /* <serial> tdb_get_seqnum = <ret> */
 262 static void op_add_seqnum(const char *filename,
 263                           struct op op[], unsigned int op_num, char *words[])
 264 {
 265         if (!words[2] || !words[3] || words[4] || !streq(words[2], "="))
 266                 fail(filename, op_num+1, "Expect = <ret>");
 267
 268         op[op_num].key = tdb_null;
 269         op[op_num].ret = atoi(words[3]);
 270 }
 271
 272 static void op_add_traverse(const char *filename,
 273                             struct op op[], unsigned int op_num, char *words[])
 274 {
 275         if (words[2])
 276                 fail(filename, op_num+1, "Expect no arguments");
 277
 278         op[op_num].key = tdb_null;
 279         op[op_num].trav = NULL;
 280 }
 281
 282 struct traverse_hash {
 283         TDB_DATA key;
 284         unsigned int index;
 285 };
 286
 287 /* A traverse is a hash of keys, each one associated with ops. */
 288 struct traverse {
 289         /* How many traversal callouts should I do? */
 290         unsigned int num;
 291
 292         /* Where is traversal end op? */
 293         unsigned int end;
 294
 295         /* For trivial traversals. */
 296         struct traverse_hash *hash;
 297 };
 298
 299 /* A trivial traversal is one which doesn't terminate early and only
 300  * plays with its own record.  We can reliably replay these even if
 301  * traverse order changes. */
 302 static bool is_trivial_traverse(struct op op[], unsigned int end)
 303 {
 304 #if 0
 305         unsigned int i;
 306         TDB_DATA cur = tdb_null;
 307
 308         if (op[end].ret != 0)
 309                 return false;
 310
 311         for (i = 0; i < end; i++) {
 312                 if (!op[i].key.dptr)
 313                         continue;
 314                 if (op[i].op == OP_TDB_TRAVERSE)
 315                         cur = op[i].key;
 316                 if (!key_eq(cur, op[i].key))
 317                         return false;
 318         }
 319         return true;
 320 #endif
 321         /* With multiple things happening at once, no traverse is trivial. */
 322         return false;
 323 }
 324
 325 static void op_analyze_traverse(const char *filename,
 326                                 struct op op[], unsigned int op_num,
 327                                 char *words[])
 328 {
 329         int i;
 330         struct traverse *trav = talloc(op, struct traverse);
 331
 332         /* = %u means traverse function terminated. */
 333         if (words[2]) {
 334                 if (!streq(words[2], "=") || !words[3] || words[4])
 335                         fail(filename, op_num+1, "expect = <num>");
 336                 op[op_num].ret = atoi(words[3]);
 337         } else
 338                 op[op_num].ret = 0;
 339
 340         trav->num = 0;
 341         trav->end = op_num;
 342         for (i = op_num-1; i >= 0; i--) {
 343                 if (op[i].op == OP_TDB_TRAVERSE)
 344                         trav->num++;
 345                 if (op[i].op != OP_TDB_TRAVERSE_READ_START
 346                     && op[i].op != OP_TDB_TRAVERSE_START)
 347                         continue;
 348                 if (op[i].trav)
 349                         continue;
 350                 break;
 351         }
 352
 353         if (i < 0)
 354                 fail(filename, op_num+1, "no traversal start found");
 355
 356         op[i].trav = trav;
 357
 358         if (is_trivial_traverse(op+i, op_num-i)) {
 359                 /* Fill in a plentiful hash table. */
 360                 op[i].trav->hash = talloc_zero_array(op[i].trav,
 361                                                      struct traverse_hash,
 362                                                      trav->num * 2);
 363                 for (; i < op_num; i++) {
 364                         unsigned int h;
 365                         if (op[i].op != OP_TDB_TRAVERSE)
 366                                 continue;
 367                         h = hash_key(&op[i].key) % (trav->num * 2);
 368                         while (trav->hash[h].index)
 369                                 h = (h + 1) % (trav->num * 2);
 370                         trav->hash[h].index = i+1;
 371                         trav->hash[h].key = op[i].key;
 372                 }
 373         } else
 374                 trav->hash = NULL;
 375 }
 376
 377 /* Keep -Wmissing-declarations happy: */
 378 const struct op_table *
 379 find_keyword (register const char *str, register unsigned int len);
 380
 381 #include "keywords.c"
 382
 383 static int get_len(TDB_DATA key, TDB_DATA data, void *private_data)
 384 {
 385         return data.dsize;
 386 }
 387
 388 static unsigned run_ops(struct tdb_context *tdb,
 389                         int pre_fd,
 390                         const char *filename,
 391                         struct op op[],
 392                         unsigned int start, unsigned int stop);
 393
 394 struct traverse_info {
 395         struct op *op;
 396         const char *filename;
 397         int pre_fd;
 398         unsigned int start;
 399         unsigned int i;
 400 };
 401
 402 /* Trivial case: do whatever they did for this key. */
 403 static int trivial_traverse(struct tdb_context *tdb,
 404                             TDB_DATA key, TDB_DATA data,
 405                             void *_tinfo)
 406 {
 407         struct traverse_info *tinfo = _tinfo;
 408         struct traverse *trav = tinfo->op[tinfo->start].trav;
 409         unsigned int h = hash_key(&key) % (trav->num * 2);
 410
 411         while (trav->hash[h].index) {
 412                 if (key_eq(trav->hash[h].key, key)) {
 413                         run_ops(tdb, tinfo->pre_fd, tinfo->filename, tinfo->op,
 414                                 trav->hash[h].index, trav->end);
 415                         tinfo->i++;
 416                         return 0;
 417                 }
 418                 h = (h + 1) % (trav->num * 2);
 419         }
 420         fail(tinfo->filename, tinfo->start + 1, "unexpected traverse key");
 421 }
 422
 423 /* More complex.  Just do whatever's they did at the n'th entry. */
 424 static int nontrivial_traverse(struct tdb_context *tdb,
 425                                TDB_DATA key, TDB_DATA data,
 426                                void *_tinfo)
 427 {
 428         struct traverse_info *tinfo = _tinfo;
 429         struct traverse *trav = tinfo->op[tinfo->start].trav;
 430
 431         if (tinfo->i == trav->end) {
 432                 /* This can happen if traverse expects to be empty. */
 433                 if (tinfo->start + 1 == trav->end)
 434                         return 1;
 435                 fail(tinfo->filename, tinfo->start + 1,
 436                      "traverse did not terminate");
 437         }
 438
 439         if (tinfo->op[tinfo->i].op != OP_TDB_TRAVERSE)
 440                 fail(tinfo->filename, tinfo->start + 1,
 441                      "%s:%u:traverse terminated early");
 442
 443         /* Run any normal ops. */
 444         tinfo->i = run_ops(tdb, tinfo->pre_fd, tinfo->filename, tinfo->op,
 445                            tinfo->i+1, trav->end);
 446
 447         if (tinfo->i == trav->end)
 448                 return 1;
 449
 450         return 0;
 451 }
 452
 453 static unsigned op_traverse(struct tdb_context *tdb,
 454                             int pre_fd,
 455                             const char *filename,
 456                             int (*traversefn)(struct tdb_context *,
 457                                               tdb_traverse_func, void *),
 458                             struct op op[],
 459                             unsigned int start)
 460 {
 461         struct traverse *trav = op[start].trav;
 462         struct traverse_info tinfo = { op, filename, pre_fd, start, start+1 };
 463
 464         /* Trivial case. */
 465         if (trav->hash) {
 466                 int ret = traversefn(tdb, trivial_traverse, &tinfo);
 467                 if (ret != trav->num)
 468                         fail(filename, start+1, "short traversal %i", ret);
 469                 return trav->end;
 470         }
 471
 472         traversefn(tdb, nontrivial_traverse, &tinfo);
 473
 474         /* Traversing in wrong order can have strange effects: eg. if
 475          * original traverse went A (delete A), B, we might do B
 476          * (delete A).  So if we have ops left over, we do it now. */
 477         while (tinfo.i != trav->end) {
 478                 if (op[tinfo.i].op == OP_TDB_TRAVERSE)
 479                         tinfo.i++;
 480                 else
 481                         tinfo.i = run_ops(tdb, pre_fd, filename, op,
 482                                           tinfo.i, trav->end);
 483         }
 484         return trav->end;
 485 }
 486
 487 struct depend {
 488         /* We can have more than one */
 489         struct list_node list;
 490         unsigned int file;
 491         unsigned int op;
 492 };
 493
 494 static void do_pre(const char *filename, int pre_fd,
 495                    struct op op[], unsigned int i)
 496 {
 497         while (op[i].pre != 0) {
 498                 unsigned int opnum;
 499
 500 #if DEBUG_DEPS
 501                 printf("%s:%u:waiting for pre\n", filename, i+1);
 502 #endif
 503                 if (read(pre_fd, &opnum, sizeof(opnum)) != sizeof(opnum))
 504                         errx(1, "Reading from pipe");
 505
 506 #if DEBUG_DEPS
 507                 printf("%s:%u:got pre %u\n",
 508                        filename, i+1, opnum);
 509 #endif
 510                 /* This could be any op, not just this one. */
 511                 if (op[opnum].pre == 0)
 512                         errx(1, "Got unexpected notification for op line %u",
 513                              opnum + 1);
 514                 op[opnum].pre--;
 515         }
 516 }
 517
 518 static void do_post(const char *filename, const struct op op[], unsigned int i)
 519 {
 520         struct depend *dep;
 521
 522         list_for_each(&op[i].post, dep, list) {
 523 #if DEBUG_DEPS
 524                 printf("%s:%u:sending %u to file %u\n", filename, i+1,
 525                        dep->op, dep->file);
 526 #endif
 527                 if (write(pipes[dep->file].fd[1], &dep->op, sizeof(dep->op))
 528                     != sizeof(dep->op))
 529                         err(1, "Failed to tell file %u", dep->file);
 530         }
 531 }
 532
 533 static __attribute__((noinline))
 534 unsigned run_ops(struct tdb_context *tdb,
 535                  int pre_fd,
 536                  const char *filename,
 537                  struct op op[], unsigned int start, unsigned int stop)
 538 {
 539         unsigned int i;
 540
 541         for (i = start; i < stop; i++) {
 542                 do_pre(filename, pre_fd, op, i);
 543
 544                 switch (op[i].op) {
 545                 case OP_TDB_LOCKALL:
 546                         try(tdb_lockall(tdb), op[i].ret);
 547                         break;
 548                 case OP_TDB_LOCKALL_MARK:
 549                         try(tdb_lockall_mark(tdb), op[i].ret);
 550                         break;
 551                 case OP_TDB_LOCKALL_UNMARK:
 552                         try(tdb_lockall_unmark(tdb), op[i].ret);
 553                         break;
 554                 case OP_TDB_LOCKALL_NONBLOCK:
 555                         unreliable(tdb_lockall_nonblock(tdb), op[i].ret,
 556                                    tdb_lockall(tdb), tdb_unlockall(tdb));
 557                         break;
 558                 case OP_TDB_UNLOCKALL:
 559                         try(tdb_unlockall(tdb), op[i].ret);
 560                         break;
 561                 case OP_TDB_LOCKALL_READ:
 562                         try(tdb_lockall_read(tdb), op[i].ret);
 563                         break;
 564                 case OP_TDB_LOCKALL_READ_NONBLOCK:
 565                         unreliable(tdb_lockall_read_nonblock(tdb), op[i].ret,
 566                                    tdb_lockall_read(tdb),
 567                                    tdb_unlockall_read(tdb));
 568                         break;
 569                 case OP_TDB_UNLOCKALL_READ:
 570                         try(tdb_unlockall_read(tdb), op[i].ret);
 571                         break;
 572                 case OP_TDB_CHAINLOCK:
 573                         try(tdb_chainlock(tdb, op[i].key), op[i].ret);
 574                         break;
 575                 case OP_TDB_CHAINLOCK_NONBLOCK:
 576                         unreliable(tdb_chainlock_nonblock(tdb, op[i].key),
 577                                    op[i].ret,
 578                                    tdb_chainlock(tdb, op[i].key),
 579                                    tdb_chainunlock(tdb, op[i].key));
 580                         break;
 581                 case OP_TDB_CHAINLOCK_MARK:
 582                         try(tdb_chainlock_mark(tdb, op[i].key), op[i].ret);
 583                         break;
 584                 case OP_TDB_CHAINLOCK_UNMARK:
 585                         try(tdb_chainlock_unmark(tdb, op[i].key), op[i].ret);
 586                         break;
 587                 case OP_TDB_CHAINUNLOCK:
 588                         try(tdb_chainunlock(tdb, op[i].key), op[i].ret);
 589                         break;
 590                 case OP_TDB_CHAINLOCK_READ:
 591                         try(tdb_chainlock_read(tdb, op[i].key), op[i].ret);
 592                         break;
 593                 case OP_TDB_CHAINUNLOCK_READ:
 594                         try(tdb_chainunlock_read(tdb, op[i].key), op[i].ret);
 595                         break;
 596                 case OP_TDB_INCREMENT_SEQNUM_NONBLOCK:
 597                         tdb_increment_seqnum_nonblock(tdb);
 598                         break;
 599                 case OP_TDB_PARSE_RECORD:
 600                         try(tdb_parse_record(tdb, op[i].key, get_len, NULL),
 601                             op[i].ret);
 602                         break;
 603                 case OP_TDB_EXISTS:
 604                         try(tdb_exists(tdb, op[i].key), op[i].ret);
 605                         break;
 606                 case OP_TDB_STORE:
 607                         try(tdb_store(tdb, op[i].key, op[i].data, op[i].flag),
 608                             op[i].ret < 0 ? op[i].ret : 0);
 609                         break;
 610                 case OP_TDB_APPEND:
 611                         try(tdb_append(tdb, op[i].key, op[i].data),
 612                             op[i].ret < 0 ? op[i].ret : 0);
 613                         break;
 614                 case OP_TDB_GET_SEQNUM:
 615                         try(tdb_get_seqnum(tdb), op[i].ret);
 616                         break;
 617                 case OP_TDB_WIPE_ALL:
 618                         try(tdb_wipe_all(tdb), op[i].ret);
 619                         break;
 620                 case OP_TDB_TRANSACTION_START:
 621                         try(tdb_transaction_start(tdb), op[i].ret);
 622                         break;
 623                 case OP_TDB_TRANSACTION_CANCEL:
 624                         try(tdb_transaction_cancel(tdb), op[i].ret);
 625                         break;
 626                 case OP_TDB_TRANSACTION_COMMIT:
 627                         try(tdb_transaction_commit(tdb), op[i].ret);
 628                         break;
 629                 case OP_TDB_TRAVERSE_READ_START:
 630                         i = op_traverse(tdb, pre_fd, filename,
 631                                         tdb_traverse_read, op, i);
 632                         break;
 633                 case OP_TDB_TRAVERSE_START:
 634                         i = op_traverse(tdb, pre_fd, filename,
 635                                         tdb_traverse, op, i);
 636                         break;
 637                 case OP_TDB_TRAVERSE:
 638                         /* Terminate: we're in a traverse, and we've
 639                          * done our ops. */
 640                         return i;
 641                 case OP_TDB_TRAVERSE_END:
 642                         fail(filename, i+1, "unepxected end traverse");
 643                 /* FIXME: These must be treated like traverse. */
 644                 case OP_TDB_FIRSTKEY:
 645                         if (!key_eq(tdb_firstkey(tdb), op[i].data))
 646                                 fail(filename, i+1, "bad firstkey");
 647                         break;
 648                 case OP_TDB_NEXTKEY:
 649                         if (!key_eq(tdb_nextkey(tdb, op[i].key), op[i].data))
 650                                 fail(filename, i+1, "bad nextkey");
 651                         break;
 652                 case OP_TDB_FETCH: {
 653                         TDB_DATA f = tdb_fetch(tdb, op[i].key);
 654                         if (!key_eq(f, op[i].data))
 655                                 fail(filename, i+1, "bad fetch %u", f.dsize);
 656                         break;
 657                 }
 658                 case OP_TDB_DELETE:
 659                         try(tdb_delete(tdb, op[i].key), op[i].ret);
 660                         break;
 661                 }
 662                 do_post(filename, op, i);
 663         }
 664         return i;
 665 }
 666
 667 static struct op *load_tracefile(const char *filename, unsigned int *num,
 668                                  unsigned int *hashsize,
 669                                  unsigned int *tdb_flags,
 670                                  unsigned int *open_flags)
 671 {
 672         unsigned int i;
 673         struct op *op = talloc_array(NULL, struct op, 1);
 674         char **words;
 675         char **lines;
 676         char *file;
 677
 678         file = grab_file(NULL, filename, NULL);
 679         if (!file)
 680                 err(1, "Reading %s", filename);
 681
 682         lines = strsplit(file, file, "\n", NULL);
 683         if (!lines[0])
 684                 errx(1, "%s is empty", filename);
 685
 686         words = strsplit(lines, lines[0], " ", NULL);
 687         if (!streq(words[1], "tdb_open"))
 688                 fail(filename, 1, "does not start with tdb_open");
 689
 690         *hashsize = atoi(words[2]);
 691         *tdb_flags = strtoul(words[3], NULL, 0);
 692         *open_flags = strtoul(words[4], NULL, 0);
 693
 694         for (i = 1; lines[i]; i++) {
 695                 const struct op_table *opt;
 696
 697                 words = strsplit(lines, lines[i], " ", NULL);
 698                 if (!words[0] || !words[1])
 699                         fail(filename, i+1, "Expected serial number and op");
 700
 701                 opt = find_keyword(words[1], strlen(words[1]));
 702                 if (!opt) {
 703                         if (streq(words[1], "tdb_close")) {
 704                                 if (lines[i+1])
 705                                         fail(filename, i+2,
 706                                              "lines after tdb_close");
 707                                 *num = i;
 708                                 talloc_free(lines);
 709                                 return op;
 710                         }
 711                         fail(filename, i+1, "Unknown operation '%s'", words[1]);
 712                 }
 713
 714                 add_op(filename, &op, i, atoi(words[0]), opt->type);
 715                 opt->enhance_op(filename, op, i, words);
 716         }
 717
 718         fprintf(stderr, "%s:%u:last operation is not tdb_close: incomplete?",
 719               filename, i);
 720         talloc_free(lines);
 721         *num = i - 1;
 722         return op;
 723 }
 724
 725 /* We remember all the keys we've ever seen, and who has them. */
 726 struct key_user {
 727         unsigned int file;
 728         unsigned int op_num;
 729 };
 730
 731 struct keyinfo {
 732         TDB_DATA key;
 733         unsigned int num_users;
 734         struct key_user *user;
 735 };
 736
 737 static bool changes_db(const struct op *op)
 738 {
 739         if (op->ret != 0)
 740                 return false;
 741
 742         return op->op == OP_TDB_STORE
 743                 || op->op == OP_TDB_APPEND
 744                 || op->op == OP_TDB_WIPE_ALL
 745                 || op->op == OP_TDB_TRANSACTION_COMMIT
 746                 || op->op == OP_TDB_DELETE;
 747 }
 748
 749 static struct keyinfo *hash_ops(struct op *op[], unsigned int num_ops[],
 750                                 unsigned int num)
 751 {
 752         unsigned int i, j, h;
 753         struct keyinfo *hash;
 754
 755         /* Gcc nexted function extension.  How cool is this? */
 756         int compare_user_serial(const void *_a, const void *_b)
 757         {
 758                 const struct key_user *a = _a, *b = _b;
 759                 int ret = op[a->file][a->op_num].serial
 760                         - op[b->file][b->op_num].serial;
 761
 762                 /* Fetches don't inc serial, so we put changes first. */
 763                 if (ret == 0) {
 764                         if (changes_db(&op[a->file][a->op_num])
 765                             && !changes_db(&op[b->file][b->op_num]))
 766                                 return -1;
 767                         if (changes_db(&op[b->file][b->op_num])
 768                             && !changes_db(&op[a->file][a->op_num]))
 769                                 return 1;
 770                 }
 771                 return ret;
 772         }
 773
 774         hash = talloc_zero_array(op[0], struct keyinfo, total_keys*2);
 775         for (i = 0; i < num; i++) {
 776                 for (j = 1; j < num_ops[i]; j++) {
 777                         /* We can't do this on allocation, due to realloc. */
 778                         list_head_init(&op[i][j].post);
 779
 780                         if (!op[i][j].key.dptr)
 781                                 continue;
 782
 783                         /* We don't wait for traverse keys */
 784                         /* FIXME: We should, for trivial traversals. */
 785                         if (op[i][j].op == OP_TDB_TRAVERSE)
 786                                 continue;
 787
 788                         h = hash_key(&op[i][j].key) % (total_keys * 2);
 789                         while (!key_eq(hash[h].key, op[i][j].key)) {
 790                                 if (!hash[h].key.dptr) {
 791                                         hash[h].key = op[i][j].key;
 792                                         break;
 793                                 }
 794                                 h = (h + 1) % (total_keys * 2);
 795                         }
 796                         /* Might as well save some memory if we can. */
 797                         if (op[i][j].key.dptr != hash[h].key.dptr) {
 798                                 talloc_free(op[i][j].key.dptr);
 799                                 op[i][j].key.dptr = hash[h].key.dptr;
 800                         }
 801                         hash[h].user = talloc_realloc(hash, hash[h].user,
 802                                                      struct key_user,
 803                                                      hash[h].num_users+1);
 804                         hash[h].user[hash[h].num_users].op_num = j;
 805                         hash[h].user[hash[h].num_users].file = i;
 806                         hash[h].num_users++;
 807                 }
 808         }
 809
 810         /* Now sort into seqnum order. */
 811         for (h = 0; h < total_keys * 2; h++)
 812                 qsort(hash[h].user, hash[h].num_users, sizeof(hash[h].user[0]),
 813                       compare_user_serial);
 814
 815         return hash;
 816 }
 817
 818 static void add_dependency(void *ctx,
 819                            struct op *op[],
 820                            unsigned int needs_file,
 821                            unsigned int needs_opnum,
 822                            unsigned int satisfies_file,
 823                            unsigned int satisfies_opnum)
 824 {
 825         struct depend *post;
 826
 827         post = talloc(ctx, struct depend);
 828         post->file = needs_file;
 829         post->op = needs_opnum;
 830         list_add(&op[satisfies_file][satisfies_opnum].post, &post->list);
 831
 832         op[needs_file][needs_opnum].pre++;
 833 }
 834
 835 static void derive_dependencies(char *filename[],
 836                                 struct op *op[], unsigned int num_ops[],
 837                                 unsigned int num)
 838 {
 839         struct keyinfo *hash;
 840         unsigned int i;
 841
 842         /* Create hash table for faster key lookup. */
 843         hash = hash_ops(op, num_ops, num);
 844
 845         /* We make the naive assumption that two ops on the same key
 846          * have to be ordered; it's overkill. */
 847         for (i = 0; i < total_keys * 2; i++) {
 848                 unsigned int j;
 849
 850                 for (j = 1; j < hash[i].num_users; j++) {
 851                         /* We don't depend on ourselves. */
 852                         if (hash[i].user[j].file == hash[i].user[j-1].file)
 853                                 continue;
 854 #if DEBUG_DEPS
 855                         printf("%s:%u: depends on %s:%u\n",
 856                                filename[hash[i].user[j].file],
 857                                hash[i].user[j].op_num+1,
 858                                filename[hash[i].user[j-1].file],
 859                                hash[i].user[j-1].op_num+1);
 860 #endif
 861                         add_dependency(hash, op,
 862                                        hash[i].user[j].file,
 863                                        hash[i].user[j].op_num,
 864                                        hash[i].user[j-1].file,
 865                                        hash[i].user[j-1].op_num);
 866                 }
 867         }
 868 }
 869
 870 int main(int argc, char *argv[])
 871 {
 872         struct timeval start, end;
 873         unsigned int i, num_ops[argc], hashsize[argc], tdb_flags[argc], open_flags[argc];
 874         struct op *op[argc];
 875         int fds[2];
 876         char c;
 877
 878         if (argc < 3)
 879                 errx(1, "Usage: %s <tdbfile> <tracefile>...", argv[0]);
 880
 881         pipes = talloc_array(NULL, struct pipe, argc - 2);
 882         for (i = 0; i < argc - 2; i++) {
 883                 op[i] = load_tracefile(argv[2+i], &num_ops[i], &hashsize[i],
 884                                        &tdb_flags[i], &open_flags[i]);
 885                 if (pipe(pipes[i].fd) != 0)
 886                         err(1, "creating pipe");
 887         }
 888
 889         derive_dependencies(argv+2, op, num_ops, i);
 890
 891         /* Don't fork for single arg case: simple debugging. */
 892         if (argc == 3) {
 893                 struct tdb_context *tdb;
 894                 tdb = tdb_open_ex(argv[1], hashsize[0], tdb_flags[0],
 895                                   open_flags[0], 0600,
 896                                   NULL, hash_key);
 897                 run_ops(tdb, pipes[0].fd[0], argv[2],
 898                         op[0], 1, num_ops[0]);
 899                 exit(0);
 900         }
 901
 902         if (pipe(fds) != 0)
 903                 err(1, "creating pipe");
 904
 905         for (i = 0; i < argc - 2; i++) {
 906                 struct tdb_context *tdb;
 907
 908                 switch (fork()) {
 909                 case -1:
 910                         err(1, "fork failed");
 911                 case 0:
 912                         close(fds[1]);
 913                         tdb = tdb_open_ex(argv[1], hashsize[i], tdb_flags[i],
 914                                           open_flags[i], 0600,
 915                                           NULL, hash_key);
 916                         if (!tdb)
 917                                 err(1, "Opening tdb %s", argv[1]);
 918
 919                         /* This catches parent exiting. */
 920                         if (read(fds[0], &c, 1) != 1)
 921                                 exit(1);
 922                         run_ops(tdb, pipes[i].fd[0], argv[2+i],
 923                                 op[i], 1, num_ops[i]);
 924                         exit(0);
 925                 default:
 926                         break;
 927                 }
 928         }
 929
 930         /* Let everything settle. */
 931         sleep(1);
 932
 933         gettimeofday(&start, NULL);
 934         /* Tell them all to go!  Any write of sufficient length will do. */
 935         if (write(fds[1], hashsize, i) != 1)
 936                 err(1, "Writing to wakeup pipe");
 937
 938         for (i = 0; i < argc - 2; i++) {
 939                 int status;
 940                 wait(&status);
 941                 if (!WIFEXITED(status))
 942                         errx(1, "Child died with signal");
 943                 if (WEXITSTATUS(status) != 0)
 944                         errx(1, "Child died with error code");
 945         }
 946         gettimeofday(&end, NULL);
 947
 948         end.tv_sec -= start.tv_sec;
 949         printf("Time replaying: %lu usec\n",
 950                end.tv_sec * 1000000UL + (end.tv_usec - start.tv_usec));
 951
 952         exit(0);
 953 }