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 /* Traversals block transactions in the current implementation. */
  25 #define TRAVERSALS_TAKE_TRANSACTION_LOCK 1
  26
  27 struct pipe {
  28         int fd[2];
  29 };
  30 static struct pipe *pipes;
  31
  32 static void __attribute__((noreturn)) fail(const char *filename,
  33                                            unsigned int line,
  34                                            const char *fmt, ...)
  35 {
  36         va_list ap;
  37
  38         va_start(ap, fmt);
  39         fprintf(stderr, "%s:%u: FAIL: ", filename, line);
  40         vfprintf(stderr, fmt, ap);
  41         fprintf(stderr, "\n");
  42         va_end(ap);
  43         exit(1);
  44 }
  45
  46 /* Try or die. */
  47 #define try(expr, expect)                                               \
  48         do {                                                            \
  49                 int ret = (expr);                                       \
  50                 if (ret != (expect))                                    \
  51                         fail(filename, i+1, STRINGIFY(expr) "= %i", ret); \
  52         } while (0)
  53
  54 /* Try or imitate results. */
  55 #define unreliable(expr, expect, force, undo)                           \
  56         do {                                                            \
  57                 int ret = expr;                                         \
  58                 if (ret != expect) {                                    \
  59                         fprintf(stderr, "%s:%u: %s gave %i not %i",     \
  60                               filename, i+1, STRINGIFY(expr), ret, expect); \
  61                         if (expect == 0)                                \
  62                                 force;                                  \
  63                         else                                            \
  64                                 undo;                                   \
  65                 }                                                       \
  66         } while (0)
  67
  68 static bool key_eq(TDB_DATA a, TDB_DATA b)
  69 {
  70         if (a.dsize != b.dsize)
  71                 return false;
  72         return memcmp(a.dptr, b.dptr, a.dsize) == 0;
  73 }
  74
  75 /* This is based on the hash algorithm from gdbm */
  76 static unsigned int hash_key(TDB_DATA *key)
  77 {
  78         uint32_t value; /* Used to compute the hash value.  */
  79         uint32_t   i;   /* Used to cycle through random values. */
  80
  81         /* Set the initial value from the key size. */
  82         for (value = 0x238F13AF ^ key->dsize, i=0; i < key->dsize; i++)
  83                 value = (value + (key->dptr[i] << (i*5 % 24)));
  84
  85         return (1103515243 * value + 12345);
  86 }
  87
  88 enum op_type {
  89         OP_TDB_LOCKALL,
  90         OP_TDB_LOCKALL_MARK,
  91         OP_TDB_LOCKALL_UNMARK,
  92         OP_TDB_LOCKALL_NONBLOCK,
  93         OP_TDB_UNLOCKALL,
  94         OP_TDB_LOCKALL_READ,
  95         OP_TDB_LOCKALL_READ_NONBLOCK,
  96         OP_TDB_UNLOCKALL_READ,
  97         OP_TDB_CHAINLOCK,
  98         OP_TDB_CHAINLOCK_NONBLOCK,
  99         OP_TDB_CHAINLOCK_MARK,
 100         OP_TDB_CHAINLOCK_UNMARK,
 101         OP_TDB_CHAINUNLOCK,
 102         OP_TDB_CHAINLOCK_READ,
 103         OP_TDB_CHAINUNLOCK_READ,
 104         OP_TDB_PARSE_RECORD,
 105         OP_TDB_EXISTS,
 106         OP_TDB_STORE,
 107         OP_TDB_APPEND,
 108         OP_TDB_GET_SEQNUM,
 109         OP_TDB_WIPE_ALL,
 110         OP_TDB_TRANSACTION_START,
 111         OP_TDB_TRANSACTION_CANCEL,
 112         OP_TDB_TRANSACTION_COMMIT,
 113         OP_TDB_TRAVERSE_READ_START,
 114         OP_TDB_TRAVERSE_START,
 115         OP_TDB_TRAVERSE_END,
 116         OP_TDB_TRAVERSE,
 117         OP_TDB_FIRSTKEY,
 118         OP_TDB_NEXTKEY,
 119         OP_TDB_FETCH,
 120         OP_TDB_DELETE,
 121 };
 122
 123 struct op {
 124         unsigned int serial;
 125         enum op_type op;
 126         TDB_DATA key;
 127         TDB_DATA data;
 128         int ret;
 129
 130         /* Who is waiting for us? */
 131         struct list_head post;
 132         /* How many are we waiting for? */
 133         unsigned int pre;
 134
 135         /* If I'm part of a group (traverse/transaction) where is
 136          * start?  (Otherwise, 0) */
 137         unsigned int group_start;
 138
 139         union {
 140                 int flag; /* open and store */
 141                 struct traverse *trav; /* traverse start */
 142                 TDB_DATA post_append; /* append */
 143                 unsigned int transaction_end; /* transaction start */
 144         };
 145 };
 146
 147 static unsigned char hex_char(const char *filename, unsigned int line, char c)
 148 {
 149         c = toupper(c);
 150         if (c >= 'A' && c <= 'F')
 151                 return c - 'A' + 10;
 152         if (c >= '0' && c <= '9')
 153                 return c - '0';
 154         fail(filename, line, "invalid hex character '%c'", c);
 155 }
 156
 157 /* TDB data is <size>:<%02x>* */
 158 static TDB_DATA make_tdb_data(const void *ctx,
 159                               const char *filename, unsigned int line,
 160                               const char *word)
 161 {
 162         TDB_DATA data;
 163         unsigned int i;
 164         const char *p;
 165
 166         if (streq(word, "NULL"))
 167                 return tdb_null;
 168
 169         data.dsize = atoi(word);
 170         data.dptr = talloc_array(ctx, unsigned char, data.dsize);
 171         p = strchr(word, ':');
 172         if (!p)
 173                 fail(filename, line, "invalid tdb data '%s'", word);
 174         p++;
 175         for (i = 0; i < data.dsize; i++)
 176                 data.dptr[i] = hex_char(filename, line, p[i*2])*16
 177                         + hex_char(filename, line, p[i*2+1]);
 178
 179         return data;
 180 }
 181
 182 static void add_op(const char *filename, struct op **op, unsigned int i,
 183                    unsigned int serial, enum op_type type)
 184 {
 185         struct op *new;
 186         *op = talloc_realloc(NULL, *op, struct op, i+1);
 187         new = (*op) + i;
 188         new->op = type;
 189         new->serial = serial;
 190         new->pre = 0;
 191         new->ret = 0;
 192         new->group_start = 0;
 193 }
 194
 195 static void op_add_nothing(const char *filename,
 196                            struct op op[], unsigned int op_num, char *words[])
 197 {
 198         if (words[2])
 199                 fail(filename, op_num+1, "Expected no arguments");
 200         op[op_num].key = tdb_null;
 201 }
 202
 203 static void op_add_key(const char *filename,
 204                        struct op op[], unsigned int op_num, char *words[])
 205 {
 206         if (words[2] == NULL || words[3])
 207                 fail(filename, op_num+1, "Expected just a key");
 208
 209         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 210         if (op[op_num].op != OP_TDB_TRAVERSE)
 211                 total_keys++;
 212 }
 213
 214 static void op_add_key_ret(const char *filename,
 215                            struct op op[], unsigned int op_num, char *words[])
 216 {
 217         if (!words[2] || !words[3] || !words[4] || words[5]
 218             || !streq(words[3], "="))
 219                 fail(filename, op_num+1, "Expected <key> = <ret>");
 220         op[op_num].ret = atoi(words[4]);
 221         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 222         /* May only be a unique key if it fails */
 223         if (op[op_num].ret != 0)
 224                 total_keys++;
 225 }
 226
 227 static void op_add_key_data(const char *filename,
 228                             struct op op[], unsigned int op_num, char *words[])
 229 {
 230         if (!words[2] || !words[3] || !words[4] || words[5]
 231             || !streq(words[3], "="))
 232                 fail(filename, op_num+1, "Expected <key> = <data>");
 233         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 234         op[op_num].data = make_tdb_data(op, filename, op_num+1, words[4]);
 235         /* May only be a unique key if it fails */
 236         if (!op[op_num].data.dptr)
 237                 total_keys++;
 238 }
 239
 240 /* <serial> tdb_store <rec> <rec> <flag> = <ret> */
 241 static void op_add_store(const char *filename,
 242                          struct op op[], unsigned int op_num, char *words[])
 243 {
 244         if (!words[2] || !words[3] || !words[4] || !words[5] || !words[6]
 245             || words[7] || !streq(words[5], "="))
 246                 fail(filename, op_num+1, "Expect <key> <data> <flag> = <ret>");
 247
 248         op[op_num].flag = strtoul(words[4], NULL, 0);
 249         op[op_num].ret = atoi(words[6]);
 250         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 251         op[op_num].data = make_tdb_data(op, filename, op_num+1, words[3]);
 252         total_keys++;
 253 }
 254
 255 /* <serial> tdb_append <rec> <rec> = <rec> */
 256 static void op_add_append(const char *filename,
 257                           struct op op[], unsigned int op_num, char *words[])
 258 {
 259         if (!words[2] || !words[3] || !words[4] || !words[5] || words[6]
 260             || !streq(words[4], "="))
 261                 fail(filename, op_num+1, "Expect <key> <data> = <rec>");
 262
 263         op[op_num].key = make_tdb_data(op, filename, op_num+1, words[2]);
 264         op[op_num].data = make_tdb_data(op, filename, op_num+1, words[3]);
 265         op[op_num].post_append
 266                 = make_tdb_data(op, filename, op_num+1, words[5]);
 267         total_keys++;
 268 }
 269
 270 /* <serial> tdb_get_seqnum = <ret> */
 271 static void op_add_seqnum(const char *filename,
 272                           struct op op[], unsigned int op_num, char *words[])
 273 {
 274         if (!words[2] || !words[3] || words[4] || !streq(words[2], "="))
 275                 fail(filename, op_num+1, "Expect = <ret>");
 276
 277         op[op_num].key = tdb_null;
 278         op[op_num].ret = atoi(words[3]);
 279 }
 280
 281 static void op_add_traverse(const char *filename,
 282                             struct op op[], unsigned int op_num, char *words[])
 283 {
 284         if (words[2])
 285                 fail(filename, op_num+1, "Expect no arguments");
 286
 287         op[op_num].key = tdb_null;
 288         op[op_num].trav = NULL;
 289 }
 290
 291 static void op_add_transaction(const char *filename, struct op op[],
 292                                unsigned int op_num, char *words[])
 293 {
 294         if (words[2])
 295                 fail(filename, op_num+1, "Expect no arguments");
 296
 297         op[op_num].key = tdb_null;
 298         op[op_num].transaction_end = 0;
 299 }
 300
 301 static void op_analyze_transaction(const char *filename,
 302                                    struct op op[], unsigned int op_num,
 303                                    char *words[])
 304 {
 305         int i, start;
 306
 307         op[op_num].key = tdb_null;
 308
 309         if (words[2])
 310                 fail(filename, op_num+1, "Expect no arguments");
 311
 312         for (i = op_num-1; i >= 0; i--) {
 313                 if (op[i].op == OP_TDB_TRANSACTION_START &&
 314                     !op[i].transaction_end)
 315                         break;
 316         }
 317
 318         if (i < 0)
 319                 fail(filename, op_num+1, "no transaction start found");
 320
 321         start = i;
 322         op[start].transaction_end = op_num;
 323
 324         /* This rolls in nested transactions.  I think that's right. */
 325         for (i++; i <= op_num; i++)
 326                 op[i].group_start = start;
 327 }
 328
 329 struct traverse_hash {
 330         TDB_DATA key;
 331         unsigned int index;
 332 };
 333
 334 /* A traverse is a hash of keys, each one associated with ops. */
 335 struct traverse {
 336         /* How many traversal callouts should I do? */
 337         unsigned int num;
 338
 339         /* Where is traversal end op? */
 340         unsigned int end;
 341
 342         /* For trivial traversals. */
 343         struct traverse_hash *hash;
 344 };
 345
 346 /* A trivial traversal is one which doesn't terminate early and only
 347  * plays with its own record.  We can reliably replay these even if
 348  * traverse order changes. */
 349 static bool is_trivial_traverse(struct op op[], unsigned int end)
 350 {
 351 #if 0
 352         unsigned int i;
 353         TDB_DATA cur = tdb_null;
 354
 355         if (op[end].ret != 0)
 356                 return false;
 357
 358         for (i = 0; i < end; i++) {
 359                 if (!op[i].key.dptr)
 360                         continue;
 361                 if (op[i].op == OP_TDB_TRAVERSE)
 362                         cur = op[i].key;
 363                 if (!key_eq(cur, op[i].key))
 364                         return false;
 365         }
 366         return true;
 367 #endif
 368         /* With multiple things happening at once, no traverse is trivial. */
 369         return false;
 370 }
 371
 372 static void op_analyze_traverse(const char *filename,
 373                                 struct op op[], unsigned int op_num,
 374                                 char *words[])
 375 {
 376         int i, start;
 377         struct traverse *trav = talloc(op, struct traverse);
 378
 379         op[op_num].key = tdb_null;
 380
 381         /* = %u means traverse function terminated. */
 382         if (words[2]) {
 383                 if (!streq(words[2], "=") || !words[3] || words[4])
 384                         fail(filename, op_num+1, "expect = <num>");
 385                 op[op_num].ret = atoi(words[3]);
 386         } else
 387                 op[op_num].ret = 0;
 388
 389         trav->num = 0;
 390         trav->end = op_num;
 391         for (i = op_num-1; i >= 0; i--) {
 392                 if (op[i].op == OP_TDB_TRAVERSE)
 393                         trav->num++;
 394                 if (op[i].op != OP_TDB_TRAVERSE_READ_START
 395                     && op[i].op != OP_TDB_TRAVERSE_START)
 396                         continue;
 397                 if (op[i].trav)
 398                         continue;
 399                 break;
 400         }
 401
 402         if (i < 0)
 403                 fail(filename, op_num+1, "no traversal start found");
 404
 405         start = i;
 406         op[start].trav = trav;
 407
 408         for (i = start; i <= op_num; i++)
 409                 op[i].group_start = start;
 410
 411         if (is_trivial_traverse(op+i, op_num-i)) {
 412                 /* Fill in a plentiful hash table. */
 413                 op[start].trav->hash = talloc_zero_array(op[i].trav,
 414                                                          struct traverse_hash,
 415                                                          trav->num * 2);
 416                 for (i = start; i < op_num; i++) {
 417                         unsigned int h;
 418                         if (op[i].op != OP_TDB_TRAVERSE)
 419                                 continue;
 420                         h = hash_key(&op[i].key) % (trav->num * 2);
 421                         while (trav->hash[h].index)
 422                                 h = (h + 1) % (trav->num * 2);
 423                         trav->hash[h].index = i+1;
 424                         trav->hash[h].key = op[i].key;
 425                 }
 426         } else
 427                 trav->hash = NULL;
 428 }
 429
 430 /* Keep -Wmissing-declarations happy: */
 431 const struct op_table *
 432 find_keyword (register const char *str, register unsigned int len);
 433
 434 #include "keywords.c"
 435
 436 static int get_len(TDB_DATA key, TDB_DATA data, void *private_data)
 437 {
 438         return data.dsize;
 439 }
 440
 441 static unsigned run_ops(struct tdb_context *tdb,
 442                         int pre_fd,
 443                         const char *filename,
 444                         struct op op[],
 445                         unsigned int start, unsigned int stop);
 446
 447 struct traverse_info {
 448         struct op *op;
 449         const char *filename;
 450         int pre_fd;
 451         unsigned int start;
 452         unsigned int i;
 453 };
 454
 455 /* Trivial case: do whatever they did for this key. */
 456 static int trivial_traverse(struct tdb_context *tdb,
 457                             TDB_DATA key, TDB_DATA data,
 458                             void *_tinfo)
 459 {
 460         struct traverse_info *tinfo = _tinfo;
 461         struct traverse *trav = tinfo->op[tinfo->start].trav;
 462         unsigned int h = hash_key(&key) % (trav->num * 2);
 463
 464         while (trav->hash[h].index) {
 465                 if (key_eq(trav->hash[h].key, key)) {
 466                         run_ops(tdb, tinfo->pre_fd, tinfo->filename, tinfo->op,
 467                                 trav->hash[h].index, trav->end);
 468                         tinfo->i++;
 469                         return 0;
 470                 }
 471                 h = (h + 1) % (trav->num * 2);
 472         }
 473         fail(tinfo->filename, tinfo->start + 1, "unexpected traverse key");
 474 }
 475
 476 /* More complex.  Just do whatever's they did at the n'th entry. */
 477 static int nontrivial_traverse(struct tdb_context *tdb,
 478                                TDB_DATA key, TDB_DATA data,
 479                                void *_tinfo)
 480 {
 481         struct traverse_info *tinfo = _tinfo;
 482         struct traverse *trav = tinfo->op[tinfo->start].trav;
 483
 484         if (tinfo->i == trav->end) {
 485                 /* This can happen if traverse expects to be empty. */
 486                 if (tinfo->start + 1 == trav->end)
 487                         return 1;
 488                 fail(tinfo->filename, tinfo->start + 1,
 489                      "traverse did not terminate");
 490         }
 491
 492         if (tinfo->op[tinfo->i].op != OP_TDB_TRAVERSE)
 493                 fail(tinfo->filename, tinfo->start + 1,
 494                      "%s:%u:traverse terminated early");
 495
 496         /* Run any normal ops. */
 497         tinfo->i = run_ops(tdb, tinfo->pre_fd, tinfo->filename, tinfo->op,
 498                            tinfo->i+1, trav->end);
 499
 500         if (tinfo->i == trav->end)
 501                 return 1;
 502
 503         return 0;
 504 }
 505
 506 static unsigned op_traverse(struct tdb_context *tdb,
 507                             int pre_fd,
 508                             const char *filename,
 509                             int (*traversefn)(struct tdb_context *,
 510                                               tdb_traverse_func, void *),
 511                             struct op op[],
 512                             unsigned int start)
 513 {
 514         struct traverse *trav = op[start].trav;
 515         struct traverse_info tinfo = { op, filename, pre_fd, start, start+1 };
 516
 517         /* Trivial case. */
 518         if (trav->hash) {
 519                 int ret = traversefn(tdb, trivial_traverse, &tinfo);
 520                 if (ret != trav->num)
 521                         fail(filename, start+1, "short traversal %i", ret);
 522                 return trav->end;
 523         }
 524
 525         traversefn(tdb, nontrivial_traverse, &tinfo);
 526
 527         /* Traversing in wrong order can have strange effects: eg. if
 528          * original traverse went A (delete A), B, we might do B
 529          * (delete A).  So if we have ops left over, we do it now. */
 530         while (tinfo.i != trav->end) {
 531                 if (op[tinfo.i].op == OP_TDB_TRAVERSE)
 532                         tinfo.i++;
 533                 else
 534                         tinfo.i = run_ops(tdb, pre_fd, filename, op,
 535                                           tinfo.i, trav->end);
 536         }
 537         return trav->end;
 538 }
 539
 540 struct depend {
 541         /* We can have more than one */
 542         struct list_node list;
 543         unsigned int file;
 544         unsigned int op;
 545 };
 546
 547 static void do_pre(const char *filename, int pre_fd,
 548                    struct op op[], unsigned int i)
 549 {
 550         while (op[i].pre != 0) {
 551                 unsigned int opnum;
 552
 553 #if DEBUG_DEPS
 554                 printf("%s:%u:waiting for pre\n", filename, i+1);
 555                 fflush(stdout);
 556 #endif
 557                 if (read(pre_fd, &opnum, sizeof(opnum)) != sizeof(opnum))
 558                         errx(1, "Reading from pipe");
 559
 560 #if DEBUG_DEPS
 561                 printf("%s:%u:got pre %u (%u)\n", filename, i+1, opnum+1,
 562                        op[opnum].pre);
 563                 fflush(stdout);
 564 #endif
 565                 /* This could be any op, not just this one. */
 566                 if (op[opnum].pre == 0)
 567                         errx(1, "Got extra notification for op line %u",
 568                              opnum + 1);
 569                 if (opnum < i)
 570                         errx(1, "%s:%u: got late notification for line %u",
 571                              filename, i + 1, opnum + 1);
 572                 op[opnum].pre--;
 573         }
 574 }
 575
 576 static void do_post(const char *filename, const struct op op[], unsigned int i)
 577 {
 578         struct depend *dep;
 579
 580         list_for_each(&op[i].post, dep, list) {
 581 #if DEBUG_DEPS
 582                 printf("%s:%u:sending %u to file %u\n", filename, i+1,
 583                        dep->op+1, dep->file);
 584 #endif
 585                 if (write(pipes[dep->file].fd[1], &dep->op, sizeof(dep->op))
 586                     != sizeof(dep->op))
 587                         err(1, "Failed to tell file %u", dep->file);
 588         }
 589 }
 590
 591 static __attribute__((noinline))
 592 unsigned run_ops(struct tdb_context *tdb,
 593                  int pre_fd,
 594                  const char *filename,
 595                  struct op op[], unsigned int start, unsigned int stop)
 596 {
 597         unsigned int i;
 598
 599         for (i = start; i < stop; i++) {
 600                 do_pre(filename, pre_fd, op, i);
 601
 602                 switch (op[i].op) {
 603                 case OP_TDB_LOCKALL:
 604                         try(tdb_lockall(tdb), op[i].ret);
 605                         break;
 606                 case OP_TDB_LOCKALL_MARK:
 607                         try(tdb_lockall_mark(tdb), op[i].ret);
 608                         break;
 609                 case OP_TDB_LOCKALL_UNMARK:
 610                         try(tdb_lockall_unmark(tdb), op[i].ret);
 611                         break;
 612                 case OP_TDB_LOCKALL_NONBLOCK:
 613                         unreliable(tdb_lockall_nonblock(tdb), op[i].ret,
 614                                    tdb_lockall(tdb), tdb_unlockall(tdb));
 615                         break;
 616                 case OP_TDB_UNLOCKALL:
 617                         try(tdb_unlockall(tdb), op[i].ret);
 618                         break;
 619                 case OP_TDB_LOCKALL_READ:
 620                         try(tdb_lockall_read(tdb), op[i].ret);
 621                         break;
 622                 case OP_TDB_LOCKALL_READ_NONBLOCK:
 623                         unreliable(tdb_lockall_read_nonblock(tdb), op[i].ret,
 624                                    tdb_lockall_read(tdb),
 625                                    tdb_unlockall_read(tdb));
 626                         break;
 627                 case OP_TDB_UNLOCKALL_READ:
 628                         try(tdb_unlockall_read(tdb), op[i].ret);
 629                         break;
 630                 case OP_TDB_CHAINLOCK:
 631                         try(tdb_chainlock(tdb, op[i].key), op[i].ret);
 632                         break;
 633                 case OP_TDB_CHAINLOCK_NONBLOCK:
 634                         unreliable(tdb_chainlock_nonblock(tdb, op[i].key),
 635                                    op[i].ret,
 636                                    tdb_chainlock(tdb, op[i].key),
 637                                    tdb_chainunlock(tdb, op[i].key));
 638                         break;
 639                 case OP_TDB_CHAINLOCK_MARK:
 640                         try(tdb_chainlock_mark(tdb, op[i].key), op[i].ret);
 641                         break;
 642                 case OP_TDB_CHAINLOCK_UNMARK:
 643                         try(tdb_chainlock_unmark(tdb, op[i].key), op[i].ret);
 644                         break;
 645                 case OP_TDB_CHAINUNLOCK:
 646                         try(tdb_chainunlock(tdb, op[i].key), op[i].ret);
 647                         break;
 648                 case OP_TDB_CHAINLOCK_READ:
 649                         try(tdb_chainlock_read(tdb, op[i].key), op[i].ret);
 650                         break;
 651                 case OP_TDB_CHAINUNLOCK_READ:
 652                         try(tdb_chainunlock_read(tdb, op[i].key), op[i].ret);
 653                         break;
 654                 case OP_TDB_PARSE_RECORD:
 655                         try(tdb_parse_record(tdb, op[i].key, get_len, NULL),
 656                             op[i].ret);
 657                         break;
 658                 case OP_TDB_EXISTS:
 659                         try(tdb_exists(tdb, op[i].key), op[i].ret);
 660                         break;
 661                 case OP_TDB_STORE:
 662                         try(tdb_store(tdb, op[i].key, op[i].data, op[i].flag),
 663                             op[i].ret < 0 ? op[i].ret : 0);
 664                         break;
 665                 case OP_TDB_APPEND:
 666                         try(tdb_append(tdb, op[i].key, op[i].data),
 667                             op[i].ret < 0 ? op[i].ret : 0);
 668                         break;
 669                 case OP_TDB_GET_SEQNUM:
 670                         try(tdb_get_seqnum(tdb), op[i].ret);
 671                         break;
 672                 case OP_TDB_WIPE_ALL:
 673                         try(tdb_wipe_all(tdb), op[i].ret);
 674                         break;
 675                 case OP_TDB_TRANSACTION_START:
 676                         try(tdb_transaction_start(tdb), op[i].ret);
 677                         break;
 678                 case OP_TDB_TRANSACTION_CANCEL:
 679                         try(tdb_transaction_cancel(tdb), op[i].ret);
 680                         break;
 681                 case OP_TDB_TRANSACTION_COMMIT:
 682                         try(tdb_transaction_commit(tdb), op[i].ret);
 683                         break;
 684                 case OP_TDB_TRAVERSE_READ_START:
 685                         i = op_traverse(tdb, pre_fd, filename,
 686                                         tdb_traverse_read, op, i);
 687                         break;
 688                 case OP_TDB_TRAVERSE_START:
 689                         i = op_traverse(tdb, pre_fd, filename,
 690                                         tdb_traverse, op, i);
 691                         break;
 692                 case OP_TDB_TRAVERSE:
 693                         /* Terminate: we're in a traverse, and we've
 694                          * done our ops. */
 695                         return i;
 696                 case OP_TDB_TRAVERSE_END:
 697                         fail(filename, i+1, "unepxected end traverse");
 698                 /* FIXME: These must be treated like traverse. */
 699                 case OP_TDB_FIRSTKEY:
 700                         if (!key_eq(tdb_firstkey(tdb), op[i].data))
 701                                 fail(filename, i+1, "bad firstkey");
 702                         break;
 703                 case OP_TDB_NEXTKEY:
 704                         if (!key_eq(tdb_nextkey(tdb, op[i].key), op[i].data))
 705                                 fail(filename, i+1, "bad nextkey");
 706                         break;
 707                 case OP_TDB_FETCH: {
 708                         TDB_DATA f = tdb_fetch(tdb, op[i].key);
 709                         if (!key_eq(f, op[i].data))
 710                                 fail(filename, i+1, "bad fetch %u", f.dsize);
 711                         break;
 712                 }
 713                 case OP_TDB_DELETE:
 714                         try(tdb_delete(tdb, op[i].key), op[i].ret);
 715                         break;
 716                 }
 717                 do_post(filename, op, i);
 718         }
 719         return i;
 720 }
 721
 722 static struct op *load_tracefile(const char *filename, unsigned int *num,
 723                                  unsigned int *hashsize,
 724                                  unsigned int *tdb_flags,
 725                                  unsigned int *open_flags)
 726 {
 727         unsigned int i;
 728         struct op *op = talloc_array(NULL, struct op, 1);
 729         char **words;
 730         char **lines;
 731         char *file;
 732
 733         file = grab_file(NULL, filename, NULL);
 734         if (!file)
 735                 err(1, "Reading %s", filename);
 736
 737         lines = strsplit(file, file, "\n", NULL);
 738         if (!lines[0])
 739                 errx(1, "%s is empty", filename);
 740
 741         words = strsplit(lines, lines[0], " ", NULL);
 742         if (!streq(words[1], "tdb_open"))
 743                 fail(filename, 1, "does not start with tdb_open");
 744
 745         *hashsize = atoi(words[2]);
 746         *tdb_flags = strtoul(words[3], NULL, 0);
 747         *open_flags = strtoul(words[4], NULL, 0);
 748
 749         for (i = 1; lines[i]; i++) {
 750                 const struct op_table *opt;
 751
 752                 words = strsplit(lines, lines[i], " ", NULL);
 753                 if (!words[0] || !words[1])
 754                         fail(filename, i+1, "Expected serial number and op");
 755
 756                 opt = find_keyword(words[1], strlen(words[1]));
 757                 if (!opt) {
 758                         if (streq(words[1], "tdb_close")) {
 759                                 if (lines[i+1])
 760                                         fail(filename, i+2,
 761                                              "lines after tdb_close");
 762                                 *num = i;
 763                                 talloc_free(lines);
 764                                 return op;
 765                         }
 766                         fail(filename, i+1, "Unknown operation '%s'", words[1]);
 767                 }
 768
 769                 add_op(filename, &op, i, atoi(words[0]), opt->type);
 770                 opt->enhance_op(filename, op, i, words);
 771         }
 772
 773         fprintf(stderr, "%s:%u:last operation is not tdb_close: incomplete?",
 774               filename, i);
 775         talloc_free(lines);
 776         *num = i - 1;
 777         return op;
 778 }
 779
 780 /* We remember all the keys we've ever seen, and who has them. */
 781 struct key_user {
 782         unsigned int file;
 783         unsigned int op_num;
 784 };
 785
 786 struct keyinfo {
 787         TDB_DATA key;
 788         unsigned int num_users;
 789         struct key_user *user;
 790 };
 791
 792 static bool changes_db(const struct op *op)
 793 {
 794         if (op->ret != 0)
 795                 return false;
 796
 797         return op->op == OP_TDB_STORE
 798                 || op->op == OP_TDB_APPEND
 799                 || op->op == OP_TDB_WIPE_ALL
 800                 || op->op == OP_TDB_TRANSACTION_COMMIT
 801                 || op->op == OP_TDB_DELETE;
 802 }
 803
 804 static struct keyinfo *hash_ops(struct op *op[], unsigned int num_ops[],
 805                                 unsigned int num)
 806 {
 807         unsigned int i, j, h;
 808         struct keyinfo *hash;
 809
 810         /* Gcc nexted function extension.  How cool is this? */
 811         int compare_user_serial(const void *_a, const void *_b)
 812         {
 813                 const struct key_user *a = _a, *b = _b;
 814                 int ret = op[a->file][a->op_num].serial
 815                         - op[b->file][b->op_num].serial;
 816
 817                 /* Fetches don't inc serial, so we put changes first. */
 818                 if (ret == 0) {
 819                         if (changes_db(&op[a->file][a->op_num])
 820                             && !changes_db(&op[b->file][b->op_num]))
 821                                 return -1;
 822                         if (changes_db(&op[b->file][b->op_num])
 823                             && !changes_db(&op[a->file][a->op_num]))
 824                                 return 1;
 825                 }
 826                 return ret;
 827         }
 828
 829         hash = talloc_zero_array(op[0], struct keyinfo, total_keys*2);
 830         for (i = 0; i < num; i++) {
 831                 for (j = 1; j < num_ops[i]; j++) {
 832                         /* We can't do this on allocation, due to realloc. */
 833                         list_head_init(&op[i][j].post);
 834
 835                         if (!op[i][j].key.dptr)
 836                                 continue;
 837
 838                         /* We don't wait for traverse keys */
 839                         /* FIXME: We should, for trivial traversals. */
 840                         if (op[i][j].op == OP_TDB_TRAVERSE)
 841                                 continue;
 842
 843                         h = hash_key(&op[i][j].key) % (total_keys * 2);
 844                         while (!key_eq(hash[h].key, op[i][j].key)) {
 845                                 if (!hash[h].key.dptr) {
 846                                         hash[h].key = op[i][j].key;
 847                                         break;
 848                                 }
 849                                 h = (h + 1) % (total_keys * 2);
 850                         }
 851                         /* Might as well save some memory if we can. */
 852                         if (op[i][j].key.dptr != hash[h].key.dptr) {
 853                                 talloc_free(op[i][j].key.dptr);
 854                                 op[i][j].key.dptr = hash[h].key.dptr;
 855                         }
 856                         hash[h].user = talloc_realloc(hash, hash[h].user,
 857                                                      struct key_user,
 858                                                      hash[h].num_users+1);
 859                         hash[h].user[hash[h].num_users].op_num = j;
 860                         hash[h].user[hash[h].num_users].file = i;
 861                         hash[h].num_users++;
 862                 }
 863         }
 864
 865         /* Now sort into seqnum order. */
 866         for (h = 0; h < total_keys * 2; h++)
 867                 qsort(hash[h].user, hash[h].num_users, sizeof(hash[h].user[0]),
 868                       compare_user_serial);
 869
 870         return hash;
 871 }
 872
 873 static void add_dependency(void *ctx,
 874                            struct op *op[],
 875                            unsigned int needs_file,
 876                            unsigned int needs_opnum,
 877                            unsigned int satisfies_file,
 878                            unsigned int satisfies_opnum)
 879 {
 880         struct depend *post;
 881         unsigned int needs_start, sat_start;
 882
 883         needs_start = op[needs_file][needs_opnum].group_start;
 884         sat_start = op[satisfies_file][satisfies_opnum].group_start;
 885
 886         /* If needs is in a transaction, we need it before start. */
 887         if (needs_start) {
 888                 switch (op[needs_file][needs_start].op) {
 889                 case OP_TDB_TRANSACTION_START:
 890 #if TRAVERSALS_TAKE_TRANSACTION_LOCK
 891                 case OP_TDB_TRAVERSE_START:
 892                 case OP_TDB_TRAVERSE_READ_START:
 893 #endif
 894                         needs_opnum = needs_start;
 895 #ifdef DEBUG_DEPS
 896                         printf("  -> Back to %u\n", needs_start+1);
 897                         fflush(stdout);
 898 #endif
 899                         break;
 900                 default:
 901                         break;
 902                 }
 903         }
 904
 905         /* If satisfies is in a transaction, we wait until after commit. */
 906         /* FIXME: If transaction is cancelled, don't need dependency. */
 907         if (sat_start) {
 908                 if (op[satisfies_file][sat_start].op
 909                     == OP_TDB_TRANSACTION_START) {
 910                         satisfies_opnum
 911                                 = op[satisfies_file][sat_start].transaction_end;
 912 #ifdef DEBUG_DEPS
 913                         printf("  -> Depends on %u\n", satisfies_opnum+1);
 914                         fflush(stdout);
 915 #endif
 916                 }
 917         }
 918
 919         post = talloc(ctx, struct depend);
 920         post->file = needs_file;
 921         post->op = needs_opnum;
 922         list_add(&op[satisfies_file][satisfies_opnum].post, &post->list);
 923
 924         op[needs_file][needs_opnum].pre++;
 925 #ifdef DEBUG_DEPS
 926         if (op[needs_file][needs_opnum].pre > 1) {
 927                 printf("   (File %u opnum %u hash %u needs)\n",
 928                        needs_file, needs_opnum+1,
 929                        op[needs_file][needs_opnum].pre);
 930                 fflush(stdout);
 931         }
 932 #endif
 933 }
 934
 935 static void derive_dependencies(char *filename[],
 936                                 struct op *op[], unsigned int num_ops[],
 937                                 unsigned int num)
 938 {
 939         struct keyinfo *hash;
 940         unsigned int i;
 941
 942         /* Create hash table for faster key lookup. */
 943         hash = hash_ops(op, num_ops, num);
 944
 945         /* We make the naive assumption that two ops on the same key
 946          * have to be ordered; it's overkill. */
 947         for (i = 0; i < total_keys * 2; i++) {
 948                 unsigned int j;
 949
 950                 for (j = 1; j < hash[i].num_users; j++) {
 951                         /* We don't depend on ourselves. */
 952                         if (hash[i].user[j].file == hash[i].user[j-1].file)
 953                                 continue;
 954 #if DEBUG_DEPS
 955                         printf("%s:%u: depends on %s:%u\n",
 956                                filename[hash[i].user[j].file],
 957                                hash[i].user[j].op_num+1,
 958                                filename[hash[i].user[j-1].file],
 959                                hash[i].user[j-1].op_num+1);
 960                         fflush(stdout);
 961 #endif
 962                         add_dependency(hash, op,
 963                                        hash[i].user[j].file,
 964                                        hash[i].user[j].op_num,
 965                                        hash[i].user[j-1].file,
 966                                        hash[i].user[j-1].op_num);
 967                 }
 968         }
 969 }
 970
 971 int main(int argc, char *argv[])
 972 {
 973         struct timeval start, end;
 974         unsigned int i, num_ops[argc], hashsize[argc], tdb_flags[argc], open_flags[argc];
 975         struct op *op[argc];
 976         int fds[2];
 977         char c;
 978
 979         if (argc < 3)
 980                 errx(1, "Usage: %s <tdbfile> <tracefile>...", argv[0]);
 981
 982         pipes = talloc_array(NULL, struct pipe, argc - 2);
 983         for (i = 0; i < argc - 2; i++) {
 984                 op[i] = load_tracefile(argv[2+i], &num_ops[i], &hashsize[i],
 985                                        &tdb_flags[i], &open_flags[i]);
 986                 if (pipe(pipes[i].fd) != 0)
 987                         err(1, "creating pipe");
 988         }
 989
 990         derive_dependencies(argv+2, op, num_ops, i);
 991
 992         /* Don't fork for single arg case: simple debugging. */
 993         if (argc == 3) {
 994                 struct tdb_context *tdb;
 995                 tdb = tdb_open_ex(argv[1], hashsize[0], tdb_flags[0],
 996                                   open_flags[0], 0600,
 997                                   NULL, hash_key);
 998                 run_ops(tdb, pipes[0].fd[0], argv[2],
 999                         op[0], 1, num_ops[0]);
1000                 exit(0);
1001         }
1002
1003         if (pipe(fds) != 0)
1004                 err(1, "creating pipe");
1005
1006         for (i = 0; i < argc - 2; i++) {
1007                 struct tdb_context *tdb;
1008
1009                 switch (fork()) {
1010                 case -1:
1011                         err(1, "fork failed");
1012                 case 0:
1013                         close(fds[1]);
1014                         tdb = tdb_open_ex(argv[1], hashsize[i], tdb_flags[i],
1015                                           open_flags[i], 0600,
1016                                           NULL, hash_key);
1017                         if (!tdb)
1018                                 err(1, "Opening tdb %s", argv[1]);
1019
1020                         /* This catches parent exiting. */
1021                         if (read(fds[0], &c, 1) != 1)
1022                                 exit(1);
1023                         run_ops(tdb, pipes[i].fd[0], argv[2+i],
1024                                 op[i], 1, num_ops[i]);
1025                         exit(0);
1026                 default:
1027                         break;
1028                 }
1029         }
1030
1031         /* Let everything settle. */
1032         sleep(1);
1033
1034         gettimeofday(&start, NULL);
1035         /* Tell them all to go!  Any write of sufficient length will do. */
1036         if (write(fds[1], hashsize, i) != i)
1037                 err(1, "Writing to wakeup pipe");
1038
1039         for (i = 0; i < argc - 2; i++) {
1040                 int status;
1041                 wait(&status);
1042                 if (!WIFEXITED(status))
1043                         errx(1, "Child died with signal %i", WTERMSIG(status));
1044                 if (WEXITSTATUS(status) != 0)
1045                         errx(1, "Child died with error code");
1046         }
1047         gettimeofday(&end, NULL);
1048
1049         end.tv_sec -= start.tv_sec;
1050         printf("Time replaying: %lu usec\n",
1051                end.tv_sec * 1000000UL + (end.tv_usec - start.tv_usec));
1052
1053         exit(0);
1054 }