reftable/stack.c

   1 /*
   2 Copyright 2020 Google LLC
   3
   4 Use of this source code is governed by a BSD-style
   5 license that can be found in the LICENSE file or at
   6 https://developers.google.com/open-source/licenses/bsd
   7 */
   8
   9 #include "stack.h"
  10
  11 #include "system.h"
  12 #include "merged.h"
  13 #include "reader.h"
  14 #include "refname.h"
  15 #include "reftable-error.h"
  16 #include "reftable-record.h"
  17 #include "reftable-merged.h"
  18 #include "writer.h"
  19
  20 static int stack_try_add(struct reftable_stack *st,
  21                          int (*write_table)(struct reftable_writer *wr,
  22                                             void *arg),
  23                          void *arg);
  24 static int stack_write_compact(struct reftable_stack *st,
  25                                struct reftable_writer *wr, int first, int last,
  26                                struct reftable_log_expiry_config *config);
  27 static int stack_check_addition(struct reftable_stack *st,
  28                                 const char *new_tab_name);
  29 static void reftable_addition_close(struct reftable_addition *add);
  30 static int reftable_stack_reload_maybe_reuse(struct reftable_stack *st,
  31                                              int reuse_open);
  32
  33 static void stack_filename(struct strbuf *dest, struct reftable_stack *st,
  34                            const char *name)
  35 {
  36         strbuf_reset(dest);
  37         strbuf_addstr(dest, st->reftable_dir);
  38         strbuf_addstr(dest, "/");
  39         strbuf_addstr(dest, name);
  40 }
  41
  42 static ssize_t reftable_fd_write(void *arg, const void *data, size_t sz)
  43 {
  44         int *fdp = (int *)arg;
  45         return write(*fdp, data, sz);
  46 }
  47
  48 int reftable_new_stack(struct reftable_stack **dest, const char *dir,
  49                        struct reftable_write_options config)
  50 {
  51         struct reftable_stack *p =
  52                 reftable_calloc(sizeof(struct reftable_stack));
  53         struct strbuf list_file_name = STRBUF_INIT;
  54         int err = 0;
  55
  56         if (config.hash_id == 0) {
  57                 config.hash_id = GIT_SHA1_FORMAT_ID;
  58         }
  59
  60         *dest = NULL;
  61
  62         strbuf_reset(&list_file_name);
  63         strbuf_addstr(&list_file_name, dir);
  64         strbuf_addstr(&list_file_name, "/tables.list");
  65
  66         p->list_file = strbuf_detach(&list_file_name, NULL);
  67         p->reftable_dir = xstrdup(dir);
  68         p->config = config;
  69
  70         err = reftable_stack_reload_maybe_reuse(p, 1);
  71         if (err < 0) {
  72                 reftable_stack_destroy(p);
  73         } else {
  74                 *dest = p;
  75         }
  76         return err;
  77 }
  78
  79 static int fd_read_lines(int fd, char ***namesp)
  80 {
  81         off_t size = lseek(fd, 0, SEEK_END);
  82         char *buf = NULL;
  83         int err = 0;
  84         if (size < 0) {
  85                 err = REFTABLE_IO_ERROR;
  86                 goto done;
  87         }
  88         err = lseek(fd, 0, SEEK_SET);
  89         if (err < 0) {
  90                 err = REFTABLE_IO_ERROR;
  91                 goto done;
  92         }
  93
  94         buf = reftable_malloc(size + 1);
  95         if (read(fd, buf, size) != size) {
  96                 err = REFTABLE_IO_ERROR;
  97                 goto done;
  98         }
  99         buf[size] = 0;
 100
 101         parse_names(buf, size, namesp);
 102
 103 done:
 104         reftable_free(buf);
 105         return err;
 106 }
 107
 108 int read_lines(const char *filename, char ***namesp)
 109 {
 110         int fd = open(filename, O_RDONLY);
 111         int err = 0;
 112         if (fd < 0) {
 113                 if (errno == ENOENT) {
 114                         *namesp = reftable_calloc(sizeof(char *));
 115                         return 0;
 116                 }
 117
 118                 return REFTABLE_IO_ERROR;
 119         }
 120         err = fd_read_lines(fd, namesp);
 121         close(fd);
 122         return err;
 123 }
 124
 125 struct reftable_merged_table *
 126 reftable_stack_merged_table(struct reftable_stack *st)
 127 {
 128         return st->merged;
 129 }
 130
 131 static int has_name(char **names, const char *name)
 132 {
 133         while (*names) {
 134                 if (!strcmp(*names, name))
 135                         return 1;
 136                 names++;
 137         }
 138         return 0;
 139 }
 140
 141 /* Close and free the stack */
 142 void reftable_stack_destroy(struct reftable_stack *st)
 143 {
 144         char **names = NULL;
 145         int err = 0;
 146         if (st->merged) {
 147                 reftable_merged_table_free(st->merged);
 148                 st->merged = NULL;
 149         }
 150
 151         err = read_lines(st->list_file, &names);
 152         if (err < 0) {
 153                 FREE_AND_NULL(names);
 154         }
 155
 156         if (st->readers) {
 157                 int i = 0;
 158                 struct strbuf filename = STRBUF_INIT;
 159                 for (i = 0; i < st->readers_len; i++) {
 160                         const char *name = reader_name(st->readers[i]);
 161                         strbuf_reset(&filename);
 162                         if (names && !has_name(names, name)) {
 163                                 stack_filename(&filename, st, name);
 164                         }
 165                         reftable_reader_free(st->readers[i]);
 166
 167                         if (filename.len) {
 168                                 /* On Windows, can only unlink after closing. */
 169                                 unlink(filename.buf);
 170                         }
 171                 }
 172                 strbuf_release(&filename);
 173                 st->readers_len = 0;
 174                 FREE_AND_NULL(st->readers);
 175         }
 176         FREE_AND_NULL(st->list_file);
 177         FREE_AND_NULL(st->reftable_dir);
 178         reftable_free(st);
 179         free_names(names);
 180 }
 181
 182 static struct reftable_reader **stack_copy_readers(struct reftable_stack *st,
 183                                                    int cur_len)
 184 {
 185         struct reftable_reader **cur =
 186                 reftable_calloc(sizeof(struct reftable_reader *) * cur_len);
 187         int i = 0;
 188         for (i = 0; i < cur_len; i++) {
 189                 cur[i] = st->readers[i];
 190         }
 191         return cur;
 192 }
 193
 194 static int reftable_stack_reload_once(struct reftable_stack *st, char **names,
 195                                       int reuse_open)
 196 {
 197         int cur_len = !st->merged ? 0 : st->merged->stack_len;
 198         struct reftable_reader **cur = stack_copy_readers(st, cur_len);
 199         int err = 0;
 200         int names_len = names_length(names);
 201         struct reftable_reader **new_readers =
 202                 reftable_calloc(sizeof(struct reftable_reader *) * names_len);
 203         struct reftable_table *new_tables =
 204                 reftable_calloc(sizeof(struct reftable_table) * names_len);
 205         int new_readers_len = 0;
 206         struct reftable_merged_table *new_merged = NULL;
 207         int i;
 208
 209         while (*names) {
 210                 struct reftable_reader *rd = NULL;
 211                 char *name = *names++;
 212
 213                 /* this is linear; we assume compaction keeps the number of
 214                    tables under control so this is not quadratic. */
 215                 int j = 0;
 216                 for (j = 0; reuse_open && j < cur_len; j++) {
 217                         if (cur[j] && 0 == strcmp(cur[j]->name, name)) {
 218                                 rd = cur[j];
 219                                 cur[j] = NULL;
 220                                 break;
 221                         }
 222                 }
 223
 224                 if (!rd) {
 225                         struct reftable_block_source src = { NULL };
 226                         struct strbuf table_path = STRBUF_INIT;
 227                         stack_filename(&table_path, st, name);
 228
 229                         err = reftable_block_source_from_file(&src,
 230                                                               table_path.buf);
 231                         strbuf_release(&table_path);
 232
 233                         if (err < 0)
 234                                 goto done;
 235
 236                         err = reftable_new_reader(&rd, &src, name);
 237                         if (err < 0)
 238                                 goto done;
 239                 }
 240
 241                 new_readers[new_readers_len] = rd;
 242                 reftable_table_from_reader(&new_tables[new_readers_len], rd);
 243                 new_readers_len++;
 244         }
 245
 246         /* success! */
 247         err = reftable_new_merged_table(&new_merged, new_tables,
 248                                         new_readers_len, st->config.hash_id);
 249         if (err < 0)
 250                 goto done;
 251
 252         new_tables = NULL;
 253         st->readers_len = new_readers_len;
 254         if (st->merged) {
 255                 merged_table_release(st->merged);
 256                 reftable_merged_table_free(st->merged);
 257         }
 258         if (st->readers) {
 259                 reftable_free(st->readers);
 260         }
 261         st->readers = new_readers;
 262         new_readers = NULL;
 263         new_readers_len = 0;
 264
 265         new_merged->suppress_deletions = 1;
 266         st->merged = new_merged;
 267         for (i = 0; i < cur_len; i++) {
 268                 if (cur[i]) {
 269                         const char *name = reader_name(cur[i]);
 270                         struct strbuf filename = STRBUF_INIT;
 271                         stack_filename(&filename, st, name);
 272
 273                         reader_close(cur[i]);
 274                         reftable_reader_free(cur[i]);
 275
 276                         /* On Windows, can only unlink after closing. */
 277                         unlink(filename.buf);
 278
 279                         strbuf_release(&filename);
 280                 }
 281         }
 282
 283 done:
 284         for (i = 0; i < new_readers_len; i++) {
 285                 reader_close(new_readers[i]);
 286                 reftable_reader_free(new_readers[i]);
 287         }
 288         reftable_free(new_readers);
 289         reftable_free(new_tables);
 290         reftable_free(cur);
 291         return err;
 292 }
 293
 294 /* return negative if a before b. */
 295 static int tv_cmp(struct timeval *a, struct timeval *b)
 296 {
 297         time_t diff = a->tv_sec - b->tv_sec;
 298         int udiff = a->tv_usec - b->tv_usec;
 299
 300         if (diff != 0)
 301                 return diff;
 302
 303         return udiff;
 304 }
 305
 306 static int reftable_stack_reload_maybe_reuse(struct reftable_stack *st,
 307                                              int reuse_open)
 308 {
 309         struct timeval deadline = { 0 };
 310         int err = gettimeofday(&deadline, NULL);
 311         int64_t delay = 0;
 312         int tries = 0;
 313         if (err < 0)
 314                 return err;
 315
 316         deadline.tv_sec += 3;
 317         while (1) {
 318                 char **names = NULL;
 319                 char **names_after = NULL;
 320                 struct timeval now = { 0 };
 321                 int err = gettimeofday(&now, NULL);
 322                 int err2 = 0;
 323                 if (err < 0) {
 324                         return err;
 325                 }
 326
 327                 /* Only look at deadlines after the first few times. This
 328                    simplifies debugging in GDB */
 329                 tries++;
 330                 if (tries > 3 && tv_cmp(&now, &deadline) >= 0) {
 331                         break;
 332                 }
 333
 334                 err = read_lines(st->list_file, &names);
 335                 if (err < 0) {
 336                         free_names(names);
 337                         return err;
 338                 }
 339                 err = reftable_stack_reload_once(st, names, reuse_open);
 340                 if (err == 0) {
 341                         free_names(names);
 342                         break;
 343                 }
 344                 if (err != REFTABLE_NOT_EXIST_ERROR) {
 345                         free_names(names);
 346                         return err;
 347                 }
 348
 349                 /* err == REFTABLE_NOT_EXIST_ERROR can be caused by a concurrent
 350                    writer. Check if there was one by checking if the name list
 351                    changed.
 352                 */
 353                 err2 = read_lines(st->list_file, &names_after);
 354                 if (err2 < 0) {
 355                         free_names(names);
 356                         return err2;
 357                 }
 358
 359                 if (names_equal(names_after, names)) {
 360                         free_names(names);
 361                         free_names(names_after);
 362                         return err;
 363                 }
 364                 free_names(names);
 365                 free_names(names_after);
 366
 367                 delay = delay + (delay * rand()) / RAND_MAX + 1;
 368                 sleep_millisec(delay);
 369         }
 370
 371         return 0;
 372 }
 373
 374 /* -1 = error
 375  0 = up to date
 376  1 = changed. */
 377 static int stack_uptodate(struct reftable_stack *st)
 378 {
 379         char **names = NULL;
 380         int err = read_lines(st->list_file, &names);
 381         int i = 0;
 382         if (err < 0)
 383                 return err;
 384
 385         for (i = 0; i < st->readers_len; i++) {
 386                 if (!names[i]) {
 387                         err = 1;
 388                         goto done;
 389                 }
 390
 391                 if (strcmp(st->readers[i]->name, names[i])) {
 392                         err = 1;
 393                         goto done;
 394                 }
 395         }
 396
 397         if (names[st->merged->stack_len]) {
 398                 err = 1;
 399                 goto done;
 400         }
 401
 402 done:
 403         free_names(names);
 404         return err;
 405 }
 406
 407 int reftable_stack_reload(struct reftable_stack *st)
 408 {
 409         int err = stack_uptodate(st);
 410         if (err > 0)
 411                 return reftable_stack_reload_maybe_reuse(st, 1);
 412         return err;
 413 }
 414
 415 int reftable_stack_add(struct reftable_stack *st,
 416                        int (*write)(struct reftable_writer *wr, void *arg),
 417                        void *arg)
 418 {
 419         int err = stack_try_add(st, write, arg);
 420         if (err < 0) {
 421                 if (err == REFTABLE_LOCK_ERROR) {
 422                         /* Ignore error return, we want to propagate
 423                            REFTABLE_LOCK_ERROR.
 424                         */
 425                         reftable_stack_reload(st);
 426                 }
 427                 return err;
 428         }
 429
 430         if (!st->disable_auto_compact)
 431                 return reftable_stack_auto_compact(st);
 432
 433         return 0;
 434 }
 435
 436 static void format_name(struct strbuf *dest, uint64_t min, uint64_t max)
 437 {
 438         char buf[100];
 439         uint32_t rnd = (uint32_t)rand();
 440         snprintf(buf, sizeof(buf), "0x%012" PRIx64 "-0x%012" PRIx64 "-%08x",
 441                  min, max, rnd);
 442         strbuf_reset(dest);
 443         strbuf_addstr(dest, buf);
 444 }
 445
 446 struct reftable_addition {
 447         int lock_file_fd;
 448         struct strbuf lock_file_name;
 449         struct reftable_stack *stack;
 450
 451         char **new_tables;
 452         int new_tables_len;
 453         uint64_t next_update_index;
 454 };
 455
 456 #define REFTABLE_ADDITION_INIT                \
 457         {                                     \
 458                 .lock_file_name = STRBUF_INIT \
 459         }
 460
 461 static int reftable_stack_init_addition(struct reftable_addition *add,
 462                                         struct reftable_stack *st)
 463 {
 464         int err = 0;
 465         add->stack = st;
 466
 467         strbuf_reset(&add->lock_file_name);
 468         strbuf_addstr(&add->lock_file_name, st->list_file);
 469         strbuf_addstr(&add->lock_file_name, ".lock");
 470
 471         add->lock_file_fd = open(add->lock_file_name.buf,
 472                                  O_EXCL | O_CREAT | O_WRONLY, 0644);
 473         if (add->lock_file_fd < 0) {
 474                 if (errno == EEXIST) {
 475                         err = REFTABLE_LOCK_ERROR;
 476                 } else {
 477                         err = REFTABLE_IO_ERROR;
 478                 }
 479                 goto done;
 480         }
 481         err = stack_uptodate(st);
 482         if (err < 0)
 483                 goto done;
 484
 485         if (err > 1) {
 486                 err = REFTABLE_LOCK_ERROR;
 487                 goto done;
 488         }
 489
 490         add->next_update_index = reftable_stack_next_update_index(st);
 491 done:
 492         if (err) {
 493                 reftable_addition_close(add);
 494         }
 495         return err;
 496 }
 497
 498 static void reftable_addition_close(struct reftable_addition *add)
 499 {
 500         int i = 0;
 501         struct strbuf nm = STRBUF_INIT;
 502         for (i = 0; i < add->new_tables_len; i++) {
 503                 stack_filename(&nm, add->stack, add->new_tables[i]);
 504                 unlink(nm.buf);
 505                 reftable_free(add->new_tables[i]);
 506                 add->new_tables[i] = NULL;
 507         }
 508         reftable_free(add->new_tables);
 509         add->new_tables = NULL;
 510         add->new_tables_len = 0;
 511
 512         if (add->lock_file_fd > 0) {
 513                 close(add->lock_file_fd);
 514                 add->lock_file_fd = 0;
 515         }
 516         if (add->lock_file_name.len > 0) {
 517                 unlink(add->lock_file_name.buf);
 518                 strbuf_release(&add->lock_file_name);
 519         }
 520
 521         strbuf_release(&nm);
 522 }
 523
 524 void reftable_addition_destroy(struct reftable_addition *add)
 525 {
 526         if (!add) {
 527                 return;
 528         }
 529         reftable_addition_close(add);
 530         reftable_free(add);
 531 }
 532
 533 int reftable_addition_commit(struct reftable_addition *add)
 534 {
 535         struct strbuf table_list = STRBUF_INIT;
 536         int i = 0;
 537         int err = 0;
 538         if (add->new_tables_len == 0)
 539                 goto done;
 540
 541         for (i = 0; i < add->stack->merged->stack_len; i++) {
 542                 strbuf_addstr(&table_list, add->stack->readers[i]->name);
 543                 strbuf_addstr(&table_list, "\n");
 544         }
 545         for (i = 0; i < add->new_tables_len; i++) {
 546                 strbuf_addstr(&table_list, add->new_tables[i]);
 547                 strbuf_addstr(&table_list, "\n");
 548         }
 549
 550         err = write(add->lock_file_fd, table_list.buf, table_list.len);
 551         strbuf_release(&table_list);
 552         if (err < 0) {
 553                 err = REFTABLE_IO_ERROR;
 554                 goto done;
 555         }
 556
 557         err = close(add->lock_file_fd);
 558         add->lock_file_fd = 0;
 559         if (err < 0) {
 560                 err = REFTABLE_IO_ERROR;
 561                 goto done;
 562         }
 563
 564         err = rename(add->lock_file_name.buf, add->stack->list_file);
 565         if (err < 0) {
 566                 err = REFTABLE_IO_ERROR;
 567                 goto done;
 568         }
 569
 570         /* success, no more state to clean up. */
 571         strbuf_release(&add->lock_file_name);
 572         for (i = 0; i < add->new_tables_len; i++) {
 573                 reftable_free(add->new_tables[i]);
 574         }
 575         reftable_free(add->new_tables);
 576         add->new_tables = NULL;
 577         add->new_tables_len = 0;
 578
 579         err = reftable_stack_reload(add->stack);
 580 done:
 581         reftable_addition_close(add);
 582         return err;
 583 }
 584
 585 int reftable_stack_new_addition(struct reftable_addition **dest,
 586                                 struct reftable_stack *st)
 587 {
 588         int err = 0;
 589         struct reftable_addition empty = REFTABLE_ADDITION_INIT;
 590         *dest = reftable_calloc(sizeof(**dest));
 591         **dest = empty;
 592         err = reftable_stack_init_addition(*dest, st);
 593         if (err) {
 594                 reftable_free(*dest);
 595                 *dest = NULL;
 596         }
 597         return err;
 598 }
 599
 600 static int stack_try_add(struct reftable_stack *st,
 601                          int (*write_table)(struct reftable_writer *wr,
 602                                             void *arg),
 603                          void *arg)
 604 {
 605         struct reftable_addition add = REFTABLE_ADDITION_INIT;
 606         int err = reftable_stack_init_addition(&add, st);
 607         if (err < 0)
 608                 goto done;
 609         if (err > 0) {
 610                 err = REFTABLE_LOCK_ERROR;
 611                 goto done;
 612         }
 613
 614         err = reftable_addition_add(&add, write_table, arg);
 615         if (err < 0)
 616                 goto done;
 617
 618         err = reftable_addition_commit(&add);
 619 done:
 620         reftable_addition_close(&add);
 621         return err;
 622 }
 623
 624 int reftable_addition_add(struct reftable_addition *add,
 625                           int (*write_table)(struct reftable_writer *wr,
 626                                              void *arg),
 627                           void *arg)
 628 {
 629         struct strbuf temp_tab_file_name = STRBUF_INIT;
 630         struct strbuf tab_file_name = STRBUF_INIT;
 631         struct strbuf next_name = STRBUF_INIT;
 632         struct reftable_writer *wr = NULL;
 633         int err = 0;
 634         int tab_fd = 0;
 635
 636         strbuf_reset(&next_name);
 637         format_name(&next_name, add->next_update_index, add->next_update_index);
 638
 639         stack_filename(&temp_tab_file_name, add->stack, next_name.buf);
 640         strbuf_addstr(&temp_tab_file_name, ".temp.XXXXXX");
 641
 642         tab_fd = mkstemp(temp_tab_file_name.buf);
 643         if (tab_fd < 0) {
 644                 err = REFTABLE_IO_ERROR;
 645                 goto done;
 646         }
 647
 648         wr = reftable_new_writer(reftable_fd_write, &tab_fd,
 649                                  &add->stack->config);
 650         err = write_table(wr, arg);
 651         if (err < 0)
 652                 goto done;
 653
 654         err = reftable_writer_close(wr);
 655         if (err == REFTABLE_EMPTY_TABLE_ERROR) {
 656                 err = 0;
 657                 goto done;
 658         }
 659         if (err < 0)
 660                 goto done;
 661
 662         err = close(tab_fd);
 663         tab_fd = 0;
 664         if (err < 0) {
 665                 err = REFTABLE_IO_ERROR;
 666                 goto done;
 667         }
 668
 669         err = stack_check_addition(add->stack, temp_tab_file_name.buf);
 670         if (err < 0)
 671                 goto done;
 672
 673         if (wr->min_update_index < add->next_update_index) {
 674                 err = REFTABLE_API_ERROR;
 675                 goto done;
 676         }
 677
 678         format_name(&next_name, wr->min_update_index, wr->max_update_index);
 679         strbuf_addstr(&next_name, ".ref");
 680
 681         stack_filename(&tab_file_name, add->stack, next_name.buf);
 682
 683         /*
 684           On windows, this relies on rand() picking a unique destination name.
 685           Maybe we should do retry loop as well?
 686          */
 687         err = rename(temp_tab_file_name.buf, tab_file_name.buf);
 688         if (err < 0) {
 689                 err = REFTABLE_IO_ERROR;
 690                 goto done;
 691         }
 692
 693         add->new_tables = reftable_realloc(add->new_tables,
 694                                            sizeof(*add->new_tables) *
 695                                                    (add->new_tables_len + 1));
 696         add->new_tables[add->new_tables_len] = strbuf_detach(&next_name, NULL);
 697         add->new_tables_len++;
 698 done:
 699         if (tab_fd > 0) {
 700                 close(tab_fd);
 701                 tab_fd = 0;
 702         }
 703         if (temp_tab_file_name.len > 0) {
 704                 unlink(temp_tab_file_name.buf);
 705         }
 706
 707         strbuf_release(&temp_tab_file_name);
 708         strbuf_release(&tab_file_name);
 709         strbuf_release(&next_name);
 710         reftable_writer_free(wr);
 711         return err;
 712 }
 713
 714 uint64_t reftable_stack_next_update_index(struct reftable_stack *st)
 715 {
 716         int sz = st->merged->stack_len;
 717         if (sz > 0)
 718                 return reftable_reader_max_update_index(st->readers[sz - 1]) +
 719                        1;
 720         return 1;
 721 }
 722
 723 static int stack_compact_locked(struct reftable_stack *st, int first, int last,
 724                                 struct strbuf *temp_tab,
 725                                 struct reftable_log_expiry_config *config)
 726 {
 727         struct strbuf next_name = STRBUF_INIT;
 728         int tab_fd = -1;
 729         struct reftable_writer *wr = NULL;
 730         int err = 0;
 731
 732         format_name(&next_name,
 733                     reftable_reader_min_update_index(st->readers[first]),
 734                     reftable_reader_max_update_index(st->readers[last]));
 735
 736         stack_filename(temp_tab, st, next_name.buf);
 737         strbuf_addstr(temp_tab, ".temp.XXXXXX");
 738
 739         tab_fd = mkstemp(temp_tab->buf);
 740         wr = reftable_new_writer(reftable_fd_write, &tab_fd, &st->config);
 741
 742         err = stack_write_compact(st, wr, first, last, config);
 743         if (err < 0)
 744                 goto done;
 745         err = reftable_writer_close(wr);
 746         if (err < 0)
 747                 goto done;
 748
 749         err = close(tab_fd);
 750         tab_fd = 0;
 751
 752 done:
 753         reftable_writer_free(wr);
 754         if (tab_fd > 0) {
 755                 close(tab_fd);
 756                 tab_fd = 0;
 757         }
 758         if (err != 0 && temp_tab->len > 0) {
 759                 unlink(temp_tab->buf);
 760                 strbuf_release(temp_tab);
 761         }
 762         strbuf_release(&next_name);
 763         return err;
 764 }
 765
 766 static int stack_write_compact(struct reftable_stack *st,
 767                                struct reftable_writer *wr, int first, int last,
 768                                struct reftable_log_expiry_config *config)
 769 {
 770         int subtabs_len = last - first + 1;
 771         struct reftable_table *subtabs = reftable_calloc(
 772                 sizeof(struct reftable_table) * (last - first + 1));
 773         struct reftable_merged_table *mt = NULL;
 774         int err = 0;
 775         struct reftable_iterator it = { NULL };
 776         struct reftable_ref_record ref = { NULL };
 777         struct reftable_log_record log = { NULL };
 778
 779         uint64_t entries = 0;
 780
 781         int i = 0, j = 0;
 782         for (i = first, j = 0; i <= last; i++) {
 783                 struct reftable_reader *t = st->readers[i];
 784                 reftable_table_from_reader(&subtabs[j++], t);
 785                 st->stats.bytes += t->size;
 786         }
 787         reftable_writer_set_limits(wr, st->readers[first]->min_update_index,
 788                                    st->readers[last]->max_update_index);
 789
 790         err = reftable_new_merged_table(&mt, subtabs, subtabs_len,
 791                                         st->config.hash_id);
 792         if (err < 0) {
 793                 reftable_free(subtabs);
 794                 goto done;
 795         }
 796
 797         err = reftable_merged_table_seek_ref(mt, &it, "");
 798         if (err < 0)
 799                 goto done;
 800
 801         while (1) {
 802                 err = reftable_iterator_next_ref(&it, &ref);
 803                 if (err > 0) {
 804                         err = 0;
 805                         break;
 806                 }
 807                 if (err < 0) {
 808                         break;
 809                 }
 810
 811                 if (first == 0 && reftable_ref_record_is_deletion(&ref)) {
 812                         continue;
 813                 }
 814
 815                 err = reftable_writer_add_ref(wr, &ref);
 816                 if (err < 0) {
 817                         break;
 818                 }
 819                 entries++;
 820         }
 821         reftable_iterator_destroy(&it);
 822
 823         err = reftable_merged_table_seek_log(mt, &it, "");
 824         if (err < 0)
 825                 goto done;
 826
 827         while (1) {
 828                 err = reftable_iterator_next_log(&it, &log);
 829                 if (err > 0) {
 830                         err = 0;
 831                         break;
 832                 }
 833                 if (err < 0) {
 834                         break;
 835                 }
 836                 if (first == 0 && reftable_log_record_is_deletion(&log)) {
 837                         continue;
 838                 }
 839
 840                 if (config && config->min_update_index > 0 &&
 841                     log.update_index < config->min_update_index) {
 842                         continue;
 843                 }
 844
 845                 if (config && config->time > 0 &&
 846                     log.value.update.time < config->time) {
 847                         continue;
 848                 }
 849
 850                 err = reftable_writer_add_log(wr, &log);
 851                 if (err < 0) {
 852                         break;
 853                 }
 854                 entries++;
 855         }
 856
 857 done:
 858         reftable_iterator_destroy(&it);
 859         if (mt) {
 860                 merged_table_release(mt);
 861                 reftable_merged_table_free(mt);
 862         }
 863         reftable_ref_record_release(&ref);
 864         reftable_log_record_release(&log);
 865         st->stats.entries_written += entries;
 866         return err;
 867 }
 868
 869 /* <  0: error. 0 == OK, > 0 attempt failed; could retry. */
 870 static int stack_compact_range(struct reftable_stack *st, int first, int last,
 871                                struct reftable_log_expiry_config *expiry)
 872 {
 873         struct strbuf temp_tab_file_name = STRBUF_INIT;
 874         struct strbuf new_table_name = STRBUF_INIT;
 875         struct strbuf lock_file_name = STRBUF_INIT;
 876         struct strbuf ref_list_contents = STRBUF_INIT;
 877         struct strbuf new_table_path = STRBUF_INIT;
 878         int err = 0;
 879         int have_lock = 0;
 880         int lock_file_fd = 0;
 881         int compact_count = last - first + 1;
 882         char **listp = NULL;
 883         char **delete_on_success =
 884                 reftable_calloc(sizeof(char *) * (compact_count + 1));
 885         char **subtable_locks =
 886                 reftable_calloc(sizeof(char *) * (compact_count + 1));
 887         int i = 0;
 888         int j = 0;
 889         int is_empty_table = 0;
 890
 891         if (first > last || (!expiry && first == last)) {
 892                 err = 0;
 893                 goto done;
 894         }
 895
 896         st->stats.attempts++;
 897
 898         strbuf_reset(&lock_file_name);
 899         strbuf_addstr(&lock_file_name, st->list_file);
 900         strbuf_addstr(&lock_file_name, ".lock");
 901
 902         lock_file_fd =
 903                 open(lock_file_name.buf, O_EXCL | O_CREAT | O_WRONLY, 0644);
 904         if (lock_file_fd < 0) {
 905                 if (errno == EEXIST) {
 906                         err = 1;
 907                 } else {
 908                         err = REFTABLE_IO_ERROR;
 909                 }
 910                 goto done;
 911         }
 912         /* Don't want to write to the lock for now.  */
 913         close(lock_file_fd);
 914         lock_file_fd = 0;
 915
 916         have_lock = 1;
 917         err = stack_uptodate(st);
 918         if (err != 0)
 919                 goto done;
 920
 921         for (i = first, j = 0; i <= last; i++) {
 922                 struct strbuf subtab_file_name = STRBUF_INIT;
 923                 struct strbuf subtab_lock = STRBUF_INIT;
 924                 int sublock_file_fd = -1;
 925
 926                 stack_filename(&subtab_file_name, st,
 927                                reader_name(st->readers[i]));
 928
 929                 strbuf_reset(&subtab_lock);
 930                 strbuf_addbuf(&subtab_lock, &subtab_file_name);
 931                 strbuf_addstr(&subtab_lock, ".lock");
 932
 933                 sublock_file_fd = open(subtab_lock.buf,
 934                                        O_EXCL | O_CREAT | O_WRONLY, 0644);
 935                 if (sublock_file_fd > 0) {
 936                         close(sublock_file_fd);
 937                 } else if (sublock_file_fd < 0) {
 938                         if (errno == EEXIST) {
 939                                 err = 1;
 940                         } else {
 941                                 err = REFTABLE_IO_ERROR;
 942                         }
 943                 }
 944
 945                 subtable_locks[j] = subtab_lock.buf;
 946                 delete_on_success[j] = subtab_file_name.buf;
 947                 j++;
 948
 949                 if (err != 0)
 950                         goto done;
 951         }
 952
 953         err = unlink(lock_file_name.buf);
 954         if (err < 0)
 955                 goto done;
 956         have_lock = 0;
 957
 958         err = stack_compact_locked(st, first, last, &temp_tab_file_name,
 959                                    expiry);
 960         /* Compaction + tombstones can create an empty table out of non-empty
 961          * tables. */
 962         is_empty_table = (err == REFTABLE_EMPTY_TABLE_ERROR);
 963         if (is_empty_table) {
 964                 err = 0;
 965         }
 966         if (err < 0)
 967                 goto done;
 968
 969         lock_file_fd =
 970                 open(lock_file_name.buf, O_EXCL | O_CREAT | O_WRONLY, 0644);
 971         if (lock_file_fd < 0) {
 972                 if (errno == EEXIST) {
 973                         err = 1;
 974                 } else {
 975                         err = REFTABLE_IO_ERROR;
 976                 }
 977                 goto done;
 978         }
 979         have_lock = 1;
 980
 981         format_name(&new_table_name, st->readers[first]->min_update_index,
 982                     st->readers[last]->max_update_index);
 983         strbuf_addstr(&new_table_name, ".ref");
 984
 985         stack_filename(&new_table_path, st, new_table_name.buf);
 986
 987         if (!is_empty_table) {
 988                 /* retry? */
 989                 err = rename(temp_tab_file_name.buf, new_table_path.buf);
 990                 if (err < 0) {
 991                         err = REFTABLE_IO_ERROR;
 992                         goto done;
 993                 }
 994         }
 995
 996         for (i = 0; i < first; i++) {
 997                 strbuf_addstr(&ref_list_contents, st->readers[i]->name);
 998                 strbuf_addstr(&ref_list_contents, "\n");
 999         }
1000         if (!is_empty_table) {
1001                 strbuf_addbuf(&ref_list_contents, &new_table_name);
1002                 strbuf_addstr(&ref_list_contents, "\n");
1003         }
1004         for (i = last + 1; i < st->merged->stack_len; i++) {
1005                 strbuf_addstr(&ref_list_contents, st->readers[i]->name);
1006                 strbuf_addstr(&ref_list_contents, "\n");
1007         }
1008
1009         err = write(lock_file_fd, ref_list_contents.buf, ref_list_contents.len);
1010         if (err < 0) {
1011                 err = REFTABLE_IO_ERROR;
1012                 unlink(new_table_path.buf);
1013                 goto done;
1014         }
1015         err = close(lock_file_fd);
1016         lock_file_fd = 0;
1017         if (err < 0) {
1018                 err = REFTABLE_IO_ERROR;
1019                 unlink(new_table_path.buf);
1020                 goto done;
1021         }
1022
1023         err = rename(lock_file_name.buf, st->list_file);
1024         if (err < 0) {
1025                 err = REFTABLE_IO_ERROR;
1026                 unlink(new_table_path.buf);
1027                 goto done;
1028         }
1029         have_lock = 0;
1030
1031         /* Reload the stack before deleting. On windows, we can only delete the
1032            files after we closed them.
1033         */
1034         err = reftable_stack_reload_maybe_reuse(st, first < last);
1035
1036         listp = delete_on_success;
1037         while (*listp) {
1038                 if (strcmp(*listp, new_table_path.buf)) {
1039                         unlink(*listp);
1040                 }
1041                 listp++;
1042         }
1043
1044 done:
1045         free_names(delete_on_success);
1046
1047         listp = subtable_locks;
1048         while (*listp) {
1049                 unlink(*listp);
1050                 listp++;
1051         }
1052         free_names(subtable_locks);
1053         if (lock_file_fd > 0) {
1054                 close(lock_file_fd);
1055                 lock_file_fd = 0;
1056         }
1057         if (have_lock) {
1058                 unlink(lock_file_name.buf);
1059         }
1060         strbuf_release(&new_table_name);
1061         strbuf_release(&new_table_path);
1062         strbuf_release(&ref_list_contents);
1063         strbuf_release(&temp_tab_file_name);
1064         strbuf_release(&lock_file_name);
1065         return err;
1066 }
1067
1068 int reftable_stack_compact_all(struct reftable_stack *st,
1069                                struct reftable_log_expiry_config *config)
1070 {
1071         return stack_compact_range(st, 0, st->merged->stack_len - 1, config);
1072 }
1073
1074 static int stack_compact_range_stats(struct reftable_stack *st, int first,
1075                                      int last,
1076                                      struct reftable_log_expiry_config *config)
1077 {
1078         int err = stack_compact_range(st, first, last, config);
1079         if (err > 0) {
1080                 st->stats.failures++;
1081         }
1082         return err;
1083 }
1084
1085 static int segment_size(struct segment *s)
1086 {
1087         return s->end - s->start;
1088 }
1089
1090 int fastlog2(uint64_t sz)
1091 {
1092         int l = 0;
1093         if (sz == 0)
1094                 return 0;
1095         for (; sz; sz /= 2) {
1096                 l++;
1097         }
1098         return l - 1;
1099 }
1100
1101 struct segment *sizes_to_segments(int *seglen, uint64_t *sizes, int n)
1102 {
1103         struct segment *segs = reftable_calloc(sizeof(struct segment) * n);
1104         int next = 0;
1105         struct segment cur = { 0 };
1106         int i = 0;
1107
1108         if (n == 0) {
1109                 *seglen = 0;
1110                 return segs;
1111         }
1112         for (i = 0; i < n; i++) {
1113                 int log = fastlog2(sizes[i]);
1114                 if (cur.log != log && cur.bytes > 0) {
1115                         struct segment fresh = {
1116                                 .start = i,
1117                         };
1118
1119                         segs[next++] = cur;
1120                         cur = fresh;
1121                 }
1122
1123                 cur.log = log;
1124                 cur.end = i + 1;
1125                 cur.bytes += sizes[i];
1126         }
1127         segs[next++] = cur;
1128         *seglen = next;
1129         return segs;
1130 }
1131
1132 struct segment suggest_compaction_segment(uint64_t *sizes, int n)
1133 {
1134         int seglen = 0;
1135         struct segment *segs = sizes_to_segments(&seglen, sizes, n);
1136         struct segment min_seg = {
1137                 .log = 64,
1138         };
1139         int i = 0;
1140         for (i = 0; i < seglen; i++) {
1141                 if (segment_size(&segs[i]) == 1) {
1142                         continue;
1143                 }
1144
1145                 if (segs[i].log < min_seg.log) {
1146                         min_seg = segs[i];
1147                 }
1148         }
1149
1150         while (min_seg.start > 0) {
1151                 int prev = min_seg.start - 1;
1152                 if (fastlog2(min_seg.bytes) < fastlog2(sizes[prev])) {
1153                         break;
1154                 }
1155
1156                 min_seg.start = prev;
1157                 min_seg.bytes += sizes[prev];
1158         }
1159
1160         reftable_free(segs);
1161         return min_seg;
1162 }
1163
1164 static uint64_t *stack_table_sizes_for_compaction(struct reftable_stack *st)
1165 {
1166         uint64_t *sizes =
1167                 reftable_calloc(sizeof(uint64_t) * st->merged->stack_len);
1168         int version = (st->config.hash_id == GIT_SHA1_FORMAT_ID) ? 1 : 2;
1169         int overhead = header_size(version) - 1;
1170         int i = 0;
1171         for (i = 0; i < st->merged->stack_len; i++) {
1172                 sizes[i] = st->readers[i]->size - overhead;
1173         }
1174         return sizes;
1175 }
1176
1177 int reftable_stack_auto_compact(struct reftable_stack *st)
1178 {
1179         uint64_t *sizes = stack_table_sizes_for_compaction(st);
1180         struct segment seg =
1181                 suggest_compaction_segment(sizes, st->merged->stack_len);
1182         reftable_free(sizes);
1183         if (segment_size(&seg) > 0)
1184                 return stack_compact_range_stats(st, seg.start, seg.end - 1,
1185                                                  NULL);
1186
1187         return 0;
1188 }
1189
1190 struct reftable_compaction_stats *
1191 reftable_stack_compaction_stats(struct reftable_stack *st)
1192 {
1193         return &st->stats;
1194 }
1195
1196 int reftable_stack_read_ref(struct reftable_stack *st, const char *refname,
1197                             struct reftable_ref_record *ref)
1198 {
1199         struct reftable_table tab = { NULL };
1200         reftable_table_from_merged_table(&tab, reftable_stack_merged_table(st));
1201         return reftable_table_read_ref(&tab, refname, ref);
1202 }
1203
1204 int reftable_stack_read_log(struct reftable_stack *st, const char *refname,
1205                             struct reftable_log_record *log)
1206 {
1207         struct reftable_iterator it = { NULL };
1208         struct reftable_merged_table *mt = reftable_stack_merged_table(st);
1209         int err = reftable_merged_table_seek_log(mt, &it, refname);
1210         if (err)
1211                 goto done;
1212
1213         err = reftable_iterator_next_log(&it, log);
1214         if (err)
1215                 goto done;
1216
1217         if (strcmp(log->refname, refname) ||
1218             reftable_log_record_is_deletion(log)) {
1219                 err = 1;
1220                 goto done;
1221         }
1222
1223 done:
1224         if (err) {
1225                 reftable_log_record_release(log);
1226         }
1227         reftable_iterator_destroy(&it);
1228         return err;
1229 }
1230
1231 static int stack_check_addition(struct reftable_stack *st,
1232                                 const char *new_tab_name)
1233 {
1234         int err = 0;
1235         struct reftable_block_source src = { NULL };
1236         struct reftable_reader *rd = NULL;
1237         struct reftable_table tab = { NULL };
1238         struct reftable_ref_record *refs = NULL;
1239         struct reftable_iterator it = { NULL };
1240         int cap = 0;
1241         int len = 0;
1242         int i = 0;
1243
1244         if (st->config.skip_name_check)
1245                 return 0;
1246
1247         err = reftable_block_source_from_file(&src, new_tab_name);
1248         if (err < 0)
1249                 goto done;
1250
1251         err = reftable_new_reader(&rd, &src, new_tab_name);
1252         if (err < 0)
1253                 goto done;
1254
1255         err = reftable_reader_seek_ref(rd, &it, "");
1256         if (err > 0) {
1257                 err = 0;
1258                 goto done;
1259         }
1260         if (err < 0)
1261                 goto done;
1262
1263         while (1) {
1264                 struct reftable_ref_record ref = { NULL };
1265                 err = reftable_iterator_next_ref(&it, &ref);
1266                 if (err > 0) {
1267                         break;
1268                 }
1269                 if (err < 0)
1270                         goto done;
1271
1272                 if (len >= cap) {
1273                         cap = 2 * cap + 1;
1274                         refs = reftable_realloc(refs, cap * sizeof(refs[0]));
1275                 }
1276
1277                 refs[len++] = ref;
1278         }
1279
1280         reftable_table_from_merged_table(&tab, reftable_stack_merged_table(st));
1281
1282         err = validate_ref_record_addition(tab, refs, len);
1283
1284 done:
1285         for (i = 0; i < len; i++) {
1286                 reftable_ref_record_release(&refs[i]);
1287         }
1288
1289         free(refs);
1290         reftable_iterator_destroy(&it);
1291         reftable_reader_free(rd);
1292         return err;
1293 }
1294
1295 static int is_table_name(const char *s)
1296 {
1297         const char *dot = strrchr(s, '.');
1298         return dot && !strcmp(dot, ".ref");
1299 }
1300
1301 static void remove_maybe_stale_table(struct reftable_stack *st, uint64_t max,
1302                                      const char *name)
1303 {
1304         int err = 0;
1305         uint64_t update_idx = 0;
1306         struct reftable_block_source src = { NULL };
1307         struct reftable_reader *rd = NULL;
1308         struct strbuf table_path = STRBUF_INIT;
1309         stack_filename(&table_path, st, name);
1310
1311         err = reftable_block_source_from_file(&src, table_path.buf);
1312         if (err < 0)
1313                 goto done;
1314
1315         err = reftable_new_reader(&rd, &src, name);
1316         if (err < 0)
1317                 goto done;
1318
1319         update_idx = reftable_reader_max_update_index(rd);
1320         reftable_reader_free(rd);
1321
1322         if (update_idx <= max) {
1323                 unlink(table_path.buf);
1324         }
1325 done:
1326         strbuf_release(&table_path);
1327 }
1328
1329 static int reftable_stack_clean_locked(struct reftable_stack *st)
1330 {
1331         uint64_t max = reftable_merged_table_max_update_index(
1332                 reftable_stack_merged_table(st));
1333         DIR *dir = opendir(st->reftable_dir);
1334         struct dirent *d = NULL;
1335         if (!dir) {
1336                 return REFTABLE_IO_ERROR;
1337         }
1338
1339         while ((d = readdir(dir))) {
1340                 int i = 0;
1341                 int found = 0;
1342                 if (!is_table_name(d->d_name))
1343                         continue;
1344
1345                 for (i = 0; !found && i < st->readers_len; i++) {
1346                         found = !strcmp(reader_name(st->readers[i]), d->d_name);
1347                 }
1348                 if (found)
1349                         continue;
1350
1351                 remove_maybe_stale_table(st, max, d->d_name);
1352         }
1353
1354         closedir(dir);
1355         return 0;
1356 }
1357
1358 int reftable_stack_clean(struct reftable_stack *st)
1359 {
1360         struct reftable_addition *add = NULL;
1361         int err = reftable_stack_new_addition(&add, st);
1362         if (err < 0) {
1363                 goto done;
1364         }
1365
1366         err = reftable_stack_reload(st);
1367         if (err < 0) {
1368                 goto done;
1369         }
1370
1371         err = reftable_stack_clean_locked(st);
1372
1373 done:
1374         reftable_addition_destroy(add);
1375         return err;
1376 }
1377
1378 int reftable_stack_print_directory(const char *stackdir, uint32_t hash_id)
1379 {
1380         struct reftable_stack *stack = NULL;
1381         struct reftable_write_options cfg = { .hash_id = hash_id };
1382         struct reftable_merged_table *merged = NULL;
1383         struct reftable_table table = { NULL };
1384
1385         int err = reftable_new_stack(&stack, stackdir, cfg);
1386         if (err < 0)
1387                 goto done;
1388
1389         merged = reftable_stack_merged_table(stack);
1390         reftable_table_from_merged_table(&table, merged);
1391         err = reftable_table_print(&table);
1392 done:
1393         if (stack)
1394                 reftable_stack_destroy(stack);
1395         return err;
1396 }