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