| 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 "constants.h" |
| 13 | #include "merged.h" |
| 14 | #include "reftable-error.h" |
| 15 | #include "reftable-record.h" |
| 16 | #include "reftable-merged.h" |
| 17 | #include "table.h" |
| 18 | #include "writer.h" |
| 19 | |
| 20 | static int stack_filename(struct reftable_buf *dest, struct reftable_stack *st, |
| 21 | const char *name) |
| 22 | { |
| 23 | int err; |
| 24 | reftable_buf_reset(dest); |
| 25 | if ((err = reftable_buf_addstr(dest, st->reftable_dir)) < 0 || |
| 26 | (err = reftable_buf_addstr(dest, "/")) < 0 || |
| 27 | (err = reftable_buf_addstr(dest, name)) < 0) |
| 28 | return err; |
| 29 | return 0; |
| 30 | } |
| 31 | |
| 32 | static ssize_t reftable_write_data(int fd, const void *data, size_t size) |
| 33 | { |
| 34 | size_t total_written = 0; |
| 35 | const char *p = data; |
| 36 | |
| 37 | while (total_written < size) { |
| 38 | ssize_t bytes_written = write(fd, p, size - total_written); |
| 39 | if (bytes_written < 0 && (errno == EAGAIN || errno == EINTR)) |
| 40 | continue; |
| 41 | if (bytes_written < 0) |
| 42 | return REFTABLE_IO_ERROR; |
| 43 | |
| 44 | total_written += bytes_written; |
| 45 | p += bytes_written; |
| 46 | } |
| 47 | |
| 48 | return total_written; |
| 49 | } |
| 50 | |
| 51 | struct fd_writer { |
| 52 | const struct reftable_write_options *opts; |
| 53 | int fd; |
| 54 | }; |
| 55 | |
| 56 | static ssize_t fd_writer_write(void *arg, const void *data, size_t sz) |
| 57 | { |
| 58 | struct fd_writer *writer = arg; |
| 59 | return reftable_write_data(writer->fd, data, sz); |
| 60 | } |
| 61 | |
| 62 | static int fd_writer_flush(void *arg) |
| 63 | { |
| 64 | struct fd_writer *writer = arg; |
| 65 | return fsync(writer->fd); |
| 66 | } |
| 67 | |
| 68 | static int fd_read_lines(int fd, char ***namesp) |
| 69 | { |
| 70 | char *buf = NULL; |
| 71 | int err = 0; |
| 72 | off_t size; |
| 73 | |
| 74 | size = lseek(fd, 0, SEEK_END); |
| 75 | if (size < 0) { |
| 76 | err = REFTABLE_IO_ERROR; |
| 77 | goto done; |
| 78 | } |
| 79 | |
| 80 | err = lseek(fd, 0, SEEK_SET); |
| 81 | if (err < 0) { |
| 82 | err = REFTABLE_IO_ERROR; |
| 83 | goto done; |
| 84 | } |
| 85 | |
| 86 | REFTABLE_ALLOC_ARRAY(buf, size + 1); |
| 87 | if (!buf) { |
| 88 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 89 | goto done; |
| 90 | } |
| 91 | |
| 92 | for (off_t total_read = 0; total_read < size; ) { |
| 93 | ssize_t bytes_read = read(fd, buf + total_read, size - total_read); |
| 94 | if (bytes_read < 0 && (errno == EAGAIN || errno == EINTR)) |
| 95 | continue; |
| 96 | if (bytes_read < 0 || !bytes_read) { |
| 97 | err = REFTABLE_IO_ERROR; |
| 98 | goto done; |
| 99 | } |
| 100 | |
| 101 | total_read += bytes_read; |
| 102 | } |
| 103 | buf[size] = 0; |
| 104 | |
| 105 | err = parse_names(buf, size, namesp); |
| 106 | done: |
| 107 | reftable_free(buf); |
| 108 | return err; |
| 109 | } |
| 110 | |
| 111 | int read_lines(const char *filename, char ***namesp) |
| 112 | { |
| 113 | int fd = open(filename, O_RDONLY); |
| 114 | int err = 0; |
| 115 | if (fd < 0) { |
| 116 | if (errno == ENOENT) { |
| 117 | REFTABLE_CALLOC_ARRAY(*namesp, 1); |
| 118 | if (!*namesp) |
| 119 | return REFTABLE_OUT_OF_MEMORY_ERROR; |
| 120 | return 0; |
| 121 | } |
| 122 | |
| 123 | return REFTABLE_IO_ERROR; |
| 124 | } |
| 125 | err = fd_read_lines(fd, namesp); |
| 126 | close(fd); |
| 127 | return err; |
| 128 | } |
| 129 | |
| 130 | int reftable_stack_init_ref_iterator(struct reftable_stack *st, |
| 131 | struct reftable_iterator *it) |
| 132 | { |
| 133 | return merged_table_init_iter(reftable_stack_merged_table(st), |
| 134 | it, REFTABLE_BLOCK_TYPE_REF); |
| 135 | } |
| 136 | |
| 137 | int reftable_stack_init_log_iterator(struct reftable_stack *st, |
| 138 | struct reftable_iterator *it) |
| 139 | { |
| 140 | return merged_table_init_iter(reftable_stack_merged_table(st), |
| 141 | it, REFTABLE_BLOCK_TYPE_LOG); |
| 142 | } |
| 143 | |
| 144 | struct reftable_merged_table * |
| 145 | reftable_stack_merged_table(struct reftable_stack *st) |
| 146 | { |
| 147 | return st->merged; |
| 148 | } |
| 149 | |
| 150 | static int has_name(char **names, const char *name) |
| 151 | { |
| 152 | while (*names) { |
| 153 | if (!strcmp(*names, name)) |
| 154 | return 1; |
| 155 | names++; |
| 156 | } |
| 157 | return 0; |
| 158 | } |
| 159 | |
| 160 | /* Close and free the stack */ |
| 161 | void reftable_stack_destroy(struct reftable_stack *st) |
| 162 | { |
| 163 | char **names = NULL; |
| 164 | int err = 0; |
| 165 | |
| 166 | if (!st) |
| 167 | return; |
| 168 | |
| 169 | if (st->merged) { |
| 170 | reftable_merged_table_free(st->merged); |
| 171 | st->merged = NULL; |
| 172 | } |
| 173 | |
| 174 | if (st->list_file) |
| 175 | err = read_lines(st->list_file, &names); |
| 176 | if (err < 0) { |
| 177 | REFTABLE_FREE_AND_NULL(names); |
| 178 | } |
| 179 | |
| 180 | if (st->tables) { |
| 181 | struct reftable_buf filename = REFTABLE_BUF_INIT; |
| 182 | |
| 183 | for (size_t i = 0; i < st->tables_len; i++) { |
| 184 | const char *name = reftable_table_name(st->tables[i]); |
| 185 | int try_unlinking = 1; |
| 186 | |
| 187 | reftable_buf_reset(&filename); |
| 188 | if (names && !has_name(names, name)) { |
| 189 | if (stack_filename(&filename, st, name) < 0) |
| 190 | try_unlinking = 0; |
| 191 | } |
| 192 | reftable_table_decref(st->tables[i]); |
| 193 | |
| 194 | if (try_unlinking && filename.len) { |
| 195 | /* On Windows, can only unlink after closing. */ |
| 196 | unlink(filename.buf); |
| 197 | } |
| 198 | } |
| 199 | |
| 200 | reftable_buf_release(&filename); |
| 201 | st->tables_len = 0; |
| 202 | REFTABLE_FREE_AND_NULL(st->tables); |
| 203 | } |
| 204 | |
| 205 | if (st->list_fd >= 0) { |
| 206 | close(st->list_fd); |
| 207 | st->list_fd = -1; |
| 208 | } |
| 209 | |
| 210 | REFTABLE_FREE_AND_NULL(st->list_file); |
| 211 | REFTABLE_FREE_AND_NULL(st->reftable_dir); |
| 212 | reftable_free(st); |
| 213 | free_names(names); |
| 214 | } |
| 215 | |
| 216 | static struct reftable_table **stack_copy_tables(struct reftable_stack *st, |
| 217 | size_t cur_len) |
| 218 | { |
| 219 | struct reftable_table **cur = reftable_calloc(cur_len, sizeof(*cur)); |
| 220 | if (!cur) |
| 221 | return NULL; |
| 222 | for (size_t i = 0; i < cur_len; i++) |
| 223 | cur[i] = st->tables[i]; |
| 224 | return cur; |
| 225 | } |
| 226 | |
| 227 | static int reftable_stack_reload_once(struct reftable_stack *st, |
| 228 | const char **names, |
| 229 | int reuse_open) |
| 230 | { |
| 231 | size_t cur_len = !st->merged ? 0 : st->merged->tables_len; |
| 232 | struct reftable_table **cur = NULL; |
| 233 | struct reftable_table **reused = NULL; |
| 234 | struct reftable_table **new_tables = NULL; |
| 235 | size_t reused_len = 0, reused_alloc = 0, names_len; |
| 236 | size_t new_tables_len = 0; |
| 237 | struct reftable_merged_table *new_merged = NULL; |
| 238 | struct reftable_buf table_path = REFTABLE_BUF_INIT; |
| 239 | int err = 0; |
| 240 | size_t i; |
| 241 | |
| 242 | if (cur_len) { |
| 243 | cur = stack_copy_tables(st, cur_len); |
| 244 | if (!cur) { |
| 245 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 246 | goto done; |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | names_len = names_length(names); |
| 251 | |
| 252 | if (names_len) { |
| 253 | new_tables = reftable_calloc(names_len, sizeof(*new_tables)); |
| 254 | if (!new_tables) { |
| 255 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 256 | goto done; |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | while (*names) { |
| 261 | struct reftable_table *table = NULL; |
| 262 | const char *name = *names++; |
| 263 | |
| 264 | /* this is linear; we assume compaction keeps the number of |
| 265 | tables under control so this is not quadratic. */ |
| 266 | for (i = 0; reuse_open && i < cur_len; i++) { |
| 267 | if (cur[i] && 0 == strcmp(cur[i]->name, name)) { |
| 268 | table = cur[i]; |
| 269 | cur[i] = NULL; |
| 270 | |
| 271 | /* |
| 272 | * When reloading the stack fails, we end up |
| 273 | * releasing all new tables. This also |
| 274 | * includes the reused tables, even though |
| 275 | * they are still in used by the old stack. We |
| 276 | * thus need to keep them alive here, which we |
| 277 | * do by bumping their refcount. |
| 278 | */ |
| 279 | REFTABLE_ALLOC_GROW_OR_NULL(reused, |
| 280 | reused_len + 1, |
| 281 | reused_alloc); |
| 282 | if (!reused) { |
| 283 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 284 | goto done; |
| 285 | } |
| 286 | reused[reused_len++] = table; |
| 287 | reftable_table_incref(table); |
| 288 | break; |
| 289 | } |
| 290 | } |
| 291 | |
| 292 | if (!table) { |
| 293 | struct reftable_block_source src = { NULL }; |
| 294 | |
| 295 | err = stack_filename(&table_path, st, name); |
| 296 | if (err < 0) |
| 297 | goto done; |
| 298 | |
| 299 | err = reftable_block_source_from_file(&src, |
| 300 | table_path.buf); |
| 301 | if (err < 0) |
| 302 | goto done; |
| 303 | |
| 304 | err = reftable_table_new(&table, &src, name); |
| 305 | if (err < 0) |
| 306 | goto done; |
| 307 | } |
| 308 | |
| 309 | new_tables[new_tables_len] = table; |
| 310 | new_tables_len++; |
| 311 | } |
| 312 | |
| 313 | /* success! */ |
| 314 | err = reftable_merged_table_new(&new_merged, new_tables, |
| 315 | new_tables_len, st->opts.hash_id); |
| 316 | if (err < 0) |
| 317 | goto done; |
| 318 | |
| 319 | /* |
| 320 | * Close the old, non-reused tables and proactively try to unlink |
| 321 | * them. This is done for systems like Windows, where the underlying |
| 322 | * file of such an open table wouldn't have been possible to be |
| 323 | * unlinked by the compacting process. |
| 324 | */ |
| 325 | for (i = 0; i < cur_len; i++) { |
| 326 | if (cur[i]) { |
| 327 | const char *name = reftable_table_name(cur[i]); |
| 328 | |
| 329 | err = stack_filename(&table_path, st, name); |
| 330 | if (err < 0) |
| 331 | goto done; |
| 332 | |
| 333 | reftable_table_decref(cur[i]); |
| 334 | unlink(table_path.buf); |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | /* Update the stack to point to the new tables. */ |
| 339 | if (st->merged) |
| 340 | reftable_merged_table_free(st->merged); |
| 341 | new_merged->suppress_deletions = st->opts.suppress_deletions; |
| 342 | st->merged = new_merged; |
| 343 | |
| 344 | if (st->tables) |
| 345 | reftable_free(st->tables); |
| 346 | st->tables = new_tables; |
| 347 | st->tables_len = new_tables_len; |
| 348 | new_tables = NULL; |
| 349 | new_tables_len = 0; |
| 350 | |
| 351 | /* |
| 352 | * Decrement the refcount of reused tables again. This only needs to |
| 353 | * happen on the successful case, because on the unsuccessful one we |
| 354 | * decrement their refcount via `new_tables`. |
| 355 | */ |
| 356 | for (i = 0; i < reused_len; i++) |
| 357 | reftable_table_decref(reused[i]); |
| 358 | |
| 359 | done: |
| 360 | for (i = 0; i < new_tables_len; i++) |
| 361 | reftable_table_decref(new_tables[i]); |
| 362 | reftable_free(new_tables); |
| 363 | reftable_free(reused); |
| 364 | reftable_free(cur); |
| 365 | reftable_buf_release(&table_path); |
| 366 | return err; |
| 367 | } |
| 368 | |
| 369 | static int reftable_stack_reload_maybe_reuse(struct reftable_stack *st, |
| 370 | int reuse_open) |
| 371 | { |
| 372 | char **names = NULL, **names_after = NULL; |
| 373 | uint64_t deadline; |
| 374 | int64_t delay = 0; |
| 375 | int tries = 0, err; |
| 376 | int fd = -1; |
| 377 | |
| 378 | deadline = reftable_time_ms() + 3000; |
| 379 | |
| 380 | while (1) { |
| 381 | uint64_t now = reftable_time_ms(); |
| 382 | |
| 383 | /* |
| 384 | * Only look at deadlines after the first few times. This |
| 385 | * simplifies debugging in GDB. |
| 386 | */ |
| 387 | tries++; |
| 388 | if (tries > 3 && now >= deadline) |
| 389 | goto out; |
| 390 | |
| 391 | fd = open(st->list_file, O_RDONLY); |
| 392 | if (fd < 0) { |
| 393 | if (errno != ENOENT) { |
| 394 | err = REFTABLE_IO_ERROR; |
| 395 | goto out; |
| 396 | } |
| 397 | |
| 398 | REFTABLE_CALLOC_ARRAY(names, 1); |
| 399 | if (!names) { |
| 400 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 401 | goto out; |
| 402 | } |
| 403 | } else { |
| 404 | err = fd_read_lines(fd, &names); |
| 405 | if (err < 0) |
| 406 | goto out; |
| 407 | } |
| 408 | |
| 409 | err = reftable_stack_reload_once(st, (const char **) names, reuse_open); |
| 410 | if (!err) |
| 411 | break; |
| 412 | if (err != REFTABLE_NOT_EXIST_ERROR) |
| 413 | goto out; |
| 414 | |
| 415 | /* |
| 416 | * REFTABLE_NOT_EXIST_ERROR can be caused by a concurrent |
| 417 | * writer. Check if there was one by checking if the name list |
| 418 | * changed. |
| 419 | */ |
| 420 | err = read_lines(st->list_file, &names_after); |
| 421 | if (err < 0) |
| 422 | goto out; |
| 423 | if (names_equal((const char **) names_after, |
| 424 | (const char **) names)) { |
| 425 | err = REFTABLE_NOT_EXIST_ERROR; |
| 426 | goto out; |
| 427 | } |
| 428 | |
| 429 | free_names(names); |
| 430 | names = NULL; |
| 431 | free_names(names_after); |
| 432 | names_after = NULL; |
| 433 | close(fd); |
| 434 | fd = -1; |
| 435 | |
| 436 | delay = delay + (delay * reftable_rand()) / UINT32_MAX + 1; |
| 437 | poll(NULL, 0, delay); |
| 438 | } |
| 439 | |
| 440 | out: |
| 441 | /* |
| 442 | * Invalidate the stat cache. It is sufficient to only close the file |
| 443 | * descriptor and keep the cached stat info because we never use the |
| 444 | * latter when the former is negative. |
| 445 | */ |
| 446 | if (st->list_fd >= 0) { |
| 447 | close(st->list_fd); |
| 448 | st->list_fd = -1; |
| 449 | } |
| 450 | |
| 451 | /* |
| 452 | * Cache stat information in case it provides a useful signal to us. |
| 453 | * According to POSIX, "The st_ino and st_dev fields taken together |
| 454 | * uniquely identify the file within the system." That being said, |
| 455 | * Windows is not POSIX compliant and we do not have these fields |
| 456 | * available. So the information we have there is insufficient to |
| 457 | * determine whether two file descriptors point to the same file. |
| 458 | * |
| 459 | * While we could fall back to using other signals like the file's |
| 460 | * mtime, those are not sufficient to avoid races. We thus refrain from |
| 461 | * using the stat cache on such systems and fall back to the secondary |
| 462 | * caching mechanism, which is to check whether contents of the file |
| 463 | * have changed. |
| 464 | * |
| 465 | * On other systems which are POSIX compliant we must keep the file |
| 466 | * descriptor open. This is to avoid a race condition where two |
| 467 | * processes access the reftable stack at the same point in time: |
| 468 | * |
| 469 | * 1. A reads the reftable stack and caches its stat info. |
| 470 | * |
| 471 | * 2. B updates the stack, appending a new table to "tables.list". |
| 472 | * This will both use a new inode and result in a different file |
| 473 | * size, thus invalidating A's cache in theory. |
| 474 | * |
| 475 | * 3. B decides to auto-compact the stack and merges two tables. The |
| 476 | * file size now matches what A has cached again. Furthermore, the |
| 477 | * filesystem may decide to recycle the inode number of the file |
| 478 | * we have replaced in (2) because it is not in use anymore. |
| 479 | * |
| 480 | * 4. A reloads the reftable stack. Neither the inode number nor the |
| 481 | * file size changed. If the timestamps did not change either then |
| 482 | * we think the cached copy of our stack is up-to-date. |
| 483 | * |
| 484 | * By keeping the file descriptor open the inode number cannot be |
| 485 | * recycled, mitigating the race. |
| 486 | */ |
| 487 | if (!err && fd >= 0 && !fstat(fd, &st->list_st) && |
| 488 | st->list_st.st_dev && st->list_st.st_ino) { |
| 489 | st->list_fd = fd; |
| 490 | fd = -1; |
| 491 | } |
| 492 | |
| 493 | if (fd >= 0) |
| 494 | close(fd); |
| 495 | free_names(names); |
| 496 | free_names(names_after); |
| 497 | |
| 498 | if (st->opts.on_reload) |
| 499 | st->opts.on_reload(st->opts.on_reload_payload); |
| 500 | |
| 501 | return err; |
| 502 | } |
| 503 | |
| 504 | int reftable_new_stack(struct reftable_stack **dest, const char *dir, |
| 505 | const struct reftable_stack_options *_opts) |
| 506 | { |
| 507 | struct reftable_buf list_file_name = REFTABLE_BUF_INIT; |
| 508 | struct reftable_stack_options opts = { 0 }; |
| 509 | struct reftable_stack *p; |
| 510 | int err; |
| 511 | |
| 512 | p = reftable_calloc(1, sizeof(*p)); |
| 513 | if (!p) { |
| 514 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 515 | goto out; |
| 516 | } |
| 517 | |
| 518 | if (_opts) |
| 519 | opts = *_opts; |
| 520 | if (opts.hash_id == 0) |
| 521 | opts.hash_id = REFTABLE_HASH_SHA1; |
| 522 | |
| 523 | *dest = NULL; |
| 524 | |
| 525 | reftable_buf_reset(&list_file_name); |
| 526 | if ((err = reftable_buf_addstr(&list_file_name, dir)) < 0 || |
| 527 | (err = reftable_buf_addstr(&list_file_name, "/tables.list")) < 0) |
| 528 | goto out; |
| 529 | |
| 530 | p->list_file = reftable_buf_detach(&list_file_name); |
| 531 | p->list_fd = -1; |
| 532 | p->opts = opts; |
| 533 | p->reftable_dir = reftable_strdup(dir); |
| 534 | if (!p->reftable_dir) { |
| 535 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 536 | goto out; |
| 537 | } |
| 538 | |
| 539 | err = reftable_stack_reload_maybe_reuse(p, 1); |
| 540 | if (err < 0) |
| 541 | goto out; |
| 542 | |
| 543 | *dest = p; |
| 544 | err = 0; |
| 545 | |
| 546 | out: |
| 547 | if (err < 0) |
| 548 | reftable_stack_destroy(p); |
| 549 | return err; |
| 550 | } |
| 551 | |
| 552 | /* |
| 553 | * Check whether the given stack is up-to-date with what we have in memory. |
| 554 | * Returns 0 if so, 1 if the stack is out-of-date or a negative error code |
| 555 | * otherwise. |
| 556 | */ |
| 557 | static int stack_uptodate(struct reftable_stack *st) |
| 558 | { |
| 559 | char **names = NULL; |
| 560 | int err; |
| 561 | |
| 562 | /* |
| 563 | * When we have cached stat information available then we use it to |
| 564 | * verify whether the file has been rewritten. |
| 565 | * |
| 566 | * Note that we explicitly do not want to use `stat_validity_check()` |
| 567 | * and friends here because they may end up not comparing the `st_dev` |
| 568 | * and `st_ino` fields. These functions thus cannot guarantee that we |
| 569 | * indeed still have the same file. |
| 570 | */ |
| 571 | if (st->list_fd >= 0) { |
| 572 | struct stat list_st; |
| 573 | |
| 574 | if (stat(st->list_file, &list_st) < 0) { |
| 575 | /* |
| 576 | * It's fine for "tables.list" to not exist. In that |
| 577 | * case, we have to refresh when the loaded stack has |
| 578 | * any tables. |
| 579 | */ |
| 580 | if (errno == ENOENT) |
| 581 | return !!st->tables_len; |
| 582 | return REFTABLE_IO_ERROR; |
| 583 | } |
| 584 | |
| 585 | /* |
| 586 | * When "tables.list" refers to the same file we can assume |
| 587 | * that it didn't change. This is because we always use |
| 588 | * rename(3P) to update the file and never write to it |
| 589 | * directly. |
| 590 | */ |
| 591 | if (st->list_st.st_dev == list_st.st_dev && |
| 592 | st->list_st.st_ino == list_st.st_ino) |
| 593 | return 0; |
| 594 | } |
| 595 | |
| 596 | err = read_lines(st->list_file, &names); |
| 597 | if (err < 0) |
| 598 | return err; |
| 599 | |
| 600 | for (size_t i = 0; i < st->tables_len; i++) { |
| 601 | if (!names[i]) { |
| 602 | err = 1; |
| 603 | goto done; |
| 604 | } |
| 605 | |
| 606 | if (strcmp(st->tables[i]->name, names[i])) { |
| 607 | err = 1; |
| 608 | goto done; |
| 609 | } |
| 610 | } |
| 611 | |
| 612 | if (names[st->merged->tables_len]) { |
| 613 | err = 1; |
| 614 | goto done; |
| 615 | } |
| 616 | |
| 617 | done: |
| 618 | free_names(names); |
| 619 | return err; |
| 620 | } |
| 621 | |
| 622 | int reftable_stack_reload(struct reftable_stack *st) |
| 623 | { |
| 624 | int err = stack_uptodate(st); |
| 625 | if (err > 0) |
| 626 | return reftable_stack_reload_maybe_reuse(st, 1); |
| 627 | return err; |
| 628 | } |
| 629 | |
| 630 | struct reftable_addition { |
| 631 | struct reftable_flock tables_list_lock; |
| 632 | struct reftable_stack *stack; |
| 633 | struct reftable_write_options opts; |
| 634 | |
| 635 | char **new_tables; |
| 636 | size_t new_tables_len, new_tables_cap; |
| 637 | uint64_t next_update_index; |
| 638 | }; |
| 639 | |
| 640 | static void reftable_addition_close(struct reftable_addition *add) |
| 641 | { |
| 642 | struct reftable_buf nm = REFTABLE_BUF_INIT; |
| 643 | size_t i; |
| 644 | |
| 645 | for (i = 0; i < add->new_tables_len; i++) { |
| 646 | if (!stack_filename(&nm, add->stack, add->new_tables[i])) |
| 647 | unlink(nm.buf); |
| 648 | reftable_free(add->new_tables[i]); |
| 649 | add->new_tables[i] = NULL; |
| 650 | } |
| 651 | reftable_free(add->new_tables); |
| 652 | add->new_tables = NULL; |
| 653 | add->new_tables_len = 0; |
| 654 | add->new_tables_cap = 0; |
| 655 | |
| 656 | flock_release(&add->tables_list_lock); |
| 657 | reftable_buf_release(&nm); |
| 658 | } |
| 659 | |
| 660 | static int reftable_stack_init_addition(struct reftable_addition *add, |
| 661 | struct reftable_stack *st, |
| 662 | const struct reftable_write_options *opts, |
| 663 | unsigned int flags) |
| 664 | { |
| 665 | struct reftable_buf lock_file_name = REFTABLE_BUF_INIT; |
| 666 | int err; |
| 667 | |
| 668 | memset(add, 0, sizeof(*add)); |
| 669 | add->stack = st; |
| 670 | if (opts) |
| 671 | add->opts = *opts; |
| 672 | |
| 673 | err = flock_acquire(&add->tables_list_lock, st->list_file, |
| 674 | add->opts.lock_timeout_ms); |
| 675 | if (err < 0) |
| 676 | goto done; |
| 677 | |
| 678 | if (add->opts.default_permissions) { |
| 679 | if (chmod(add->tables_list_lock.path, |
| 680 | add->opts.default_permissions) < 0) { |
| 681 | err = REFTABLE_IO_ERROR; |
| 682 | goto done; |
| 683 | } |
| 684 | } |
| 685 | |
| 686 | err = stack_uptodate(st); |
| 687 | if (err < 0) |
| 688 | goto done; |
| 689 | if (err > 0 && flags & REFTABLE_STACK_NEW_ADDITION_RELOAD) { |
| 690 | err = reftable_stack_reload_maybe_reuse(add->stack, 1); |
| 691 | if (err) |
| 692 | goto done; |
| 693 | } |
| 694 | if (err > 0) { |
| 695 | err = REFTABLE_OUTDATED_ERROR; |
| 696 | goto done; |
| 697 | } |
| 698 | |
| 699 | add->next_update_index = reftable_stack_next_update_index(st); |
| 700 | done: |
| 701 | if (err) |
| 702 | reftable_addition_close(add); |
| 703 | reftable_buf_release(&lock_file_name); |
| 704 | return err; |
| 705 | } |
| 706 | |
| 707 | static int stack_try_add(struct reftable_stack *st, |
| 708 | int (*write_table)(struct reftable_writer *wr, |
| 709 | void *arg), |
| 710 | void *arg, |
| 711 | const struct reftable_write_options *opts, |
| 712 | unsigned flags) |
| 713 | { |
| 714 | struct reftable_addition add; |
| 715 | int err; |
| 716 | |
| 717 | err = reftable_stack_init_addition(&add, st, opts, flags); |
| 718 | if (err < 0) |
| 719 | goto done; |
| 720 | |
| 721 | err = reftable_addition_add(&add, write_table, arg); |
| 722 | if (err < 0) |
| 723 | goto done; |
| 724 | |
| 725 | err = reftable_addition_commit(&add); |
| 726 | done: |
| 727 | reftable_addition_close(&add); |
| 728 | return err; |
| 729 | } |
| 730 | |
| 731 | int reftable_stack_add(struct reftable_stack *st, |
| 732 | int (*write)(struct reftable_writer *wr, void *arg), |
| 733 | void *arg, |
| 734 | const struct reftable_write_options *opts, |
| 735 | unsigned flags) |
| 736 | { |
| 737 | int err = stack_try_add(st, write, arg, opts, flags); |
| 738 | if (err < 0) { |
| 739 | if (err == REFTABLE_OUTDATED_ERROR) { |
| 740 | /* Ignore error return, we want to propagate |
| 741 | REFTABLE_OUTDATED_ERROR. |
| 742 | */ |
| 743 | reftable_stack_reload(st); |
| 744 | } |
| 745 | return err; |
| 746 | } |
| 747 | |
| 748 | return 0; |
| 749 | } |
| 750 | |
| 751 | static int format_name(struct reftable_buf *dest, uint64_t min, uint64_t max) |
| 752 | { |
| 753 | char buf[100]; |
| 754 | uint32_t rnd = reftable_rand(); |
| 755 | snprintf(buf, sizeof(buf), "0x%012" PRIx64 "-0x%012" PRIx64 "-%08x", |
| 756 | min, max, rnd); |
| 757 | reftable_buf_reset(dest); |
| 758 | return reftable_buf_addstr(dest, buf); |
| 759 | } |
| 760 | |
| 761 | void reftable_addition_destroy(struct reftable_addition *add) |
| 762 | { |
| 763 | if (!add) { |
| 764 | return; |
| 765 | } |
| 766 | reftable_addition_close(add); |
| 767 | reftable_free(add); |
| 768 | } |
| 769 | |
| 770 | int reftable_addition_commit(struct reftable_addition *add) |
| 771 | { |
| 772 | struct reftable_buf table_list = REFTABLE_BUF_INIT; |
| 773 | int err = 0; |
| 774 | size_t i; |
| 775 | |
| 776 | if (add->new_tables_len == 0) |
| 777 | goto done; |
| 778 | |
| 779 | for (i = 0; i < add->stack->merged->tables_len; i++) { |
| 780 | if ((err = reftable_buf_addstr(&table_list, add->stack->tables[i]->name)) < 0 || |
| 781 | (err = reftable_buf_addstr(&table_list, "\n")) < 0) |
| 782 | goto done; |
| 783 | } |
| 784 | for (i = 0; i < add->new_tables_len; i++) { |
| 785 | if ((err = reftable_buf_addstr(&table_list, add->new_tables[i])) < 0 || |
| 786 | (err = reftable_buf_addstr(&table_list, "\n")) < 0) |
| 787 | goto done; |
| 788 | } |
| 789 | |
| 790 | err = reftable_write_data(add->tables_list_lock.fd, |
| 791 | table_list.buf, table_list.len); |
| 792 | reftable_buf_release(&table_list); |
| 793 | if (err < 0) { |
| 794 | err = REFTABLE_IO_ERROR; |
| 795 | goto done; |
| 796 | } |
| 797 | |
| 798 | err = fsync(add->tables_list_lock.fd); |
| 799 | if (err < 0) { |
| 800 | err = REFTABLE_IO_ERROR; |
| 801 | goto done; |
| 802 | } |
| 803 | |
| 804 | err = flock_commit(&add->tables_list_lock); |
| 805 | if (err < 0) { |
| 806 | err = REFTABLE_IO_ERROR; |
| 807 | goto done; |
| 808 | } |
| 809 | |
| 810 | /* success, no more state to clean up. */ |
| 811 | for (i = 0; i < add->new_tables_len; i++) |
| 812 | reftable_free(add->new_tables[i]); |
| 813 | reftable_free(add->new_tables); |
| 814 | add->new_tables = NULL; |
| 815 | add->new_tables_len = 0; |
| 816 | add->new_tables_cap = 0; |
| 817 | |
| 818 | err = reftable_stack_reload_maybe_reuse(add->stack, 1); |
| 819 | if (err) |
| 820 | goto done; |
| 821 | |
| 822 | if (!add->opts.disable_auto_compact) { |
| 823 | /* |
| 824 | * Auto-compact the stack to keep the number of tables in |
| 825 | * control. It is possible that a concurrent writer is already |
| 826 | * trying to compact parts of the stack, which would lead to a |
| 827 | * `REFTABLE_LOCK_ERROR` because parts of the stack are locked |
| 828 | * already. Similarly, the stack may have been rewritten by a |
| 829 | * concurrent writer, which causes `REFTABLE_OUTDATED_ERROR`. |
| 830 | * Both of these errors are benign, so we simply ignore them. |
| 831 | */ |
| 832 | err = reftable_stack_auto_compact(add->stack, &add->opts); |
| 833 | if (err < 0 && err != REFTABLE_LOCK_ERROR && |
| 834 | err != REFTABLE_OUTDATED_ERROR) |
| 835 | goto done; |
| 836 | err = 0; |
| 837 | } |
| 838 | |
| 839 | done: |
| 840 | reftable_addition_close(add); |
| 841 | return err; |
| 842 | } |
| 843 | |
| 844 | int reftable_stack_new_addition(struct reftable_addition **dest, |
| 845 | struct reftable_stack *st, |
| 846 | const struct reftable_write_options *opts, |
| 847 | unsigned int flags) |
| 848 | { |
| 849 | int err; |
| 850 | |
| 851 | REFTABLE_CALLOC_ARRAY(*dest, 1); |
| 852 | if (!*dest) |
| 853 | return REFTABLE_OUT_OF_MEMORY_ERROR; |
| 854 | |
| 855 | err = reftable_stack_init_addition(*dest, st, opts, flags); |
| 856 | if (err) { |
| 857 | reftable_free(*dest); |
| 858 | *dest = NULL; |
| 859 | } |
| 860 | |
| 861 | return err; |
| 862 | } |
| 863 | |
| 864 | int reftable_addition_add(struct reftable_addition *add, |
| 865 | int (*write_table)(struct reftable_writer *wr, |
| 866 | void *arg), |
| 867 | void *arg) |
| 868 | { |
| 869 | struct reftable_buf temp_tab_file_name = REFTABLE_BUF_INIT; |
| 870 | struct reftable_buf tab_file_name = REFTABLE_BUF_INIT; |
| 871 | struct reftable_buf next_name = REFTABLE_BUF_INIT; |
| 872 | struct reftable_writer *wr = NULL; |
| 873 | struct reftable_tmpfile tab_file = REFTABLE_TMPFILE_INIT; |
| 874 | struct fd_writer writer = { |
| 875 | .opts = &add->opts, |
| 876 | }; |
| 877 | int err = 0; |
| 878 | |
| 879 | reftable_buf_reset(&next_name); |
| 880 | |
| 881 | err = format_name(&next_name, add->next_update_index, add->next_update_index); |
| 882 | if (err < 0) |
| 883 | goto done; |
| 884 | |
| 885 | err = stack_filename(&temp_tab_file_name, add->stack, next_name.buf); |
| 886 | if (err < 0) |
| 887 | goto done; |
| 888 | |
| 889 | err = reftable_buf_addstr(&temp_tab_file_name, ".temp.XXXXXX"); |
| 890 | if (err < 0) |
| 891 | goto done; |
| 892 | |
| 893 | err = tmpfile_from_pattern(&tab_file, temp_tab_file_name.buf); |
| 894 | if (err < 0) |
| 895 | goto done; |
| 896 | if (add->opts.default_permissions) { |
| 897 | if (chmod(tab_file.path, |
| 898 | add->opts.default_permissions)) { |
| 899 | err = REFTABLE_IO_ERROR; |
| 900 | goto done; |
| 901 | } |
| 902 | } |
| 903 | |
| 904 | writer.fd = tab_file.fd; |
| 905 | err = reftable_writer_new(&wr, fd_writer_write, fd_writer_flush, |
| 906 | &writer, add->stack->opts.hash_id, &add->opts); |
| 907 | if (err < 0) |
| 908 | goto done; |
| 909 | |
| 910 | err = write_table(wr, arg); |
| 911 | if (err < 0) |
| 912 | goto done; |
| 913 | |
| 914 | err = reftable_writer_close(wr); |
| 915 | if (err == REFTABLE_EMPTY_TABLE_ERROR) { |
| 916 | err = 0; |
| 917 | goto done; |
| 918 | } |
| 919 | if (err < 0) |
| 920 | goto done; |
| 921 | |
| 922 | err = tmpfile_close(&tab_file); |
| 923 | if (err < 0) |
| 924 | goto done; |
| 925 | |
| 926 | if (wr->min_update_index < add->next_update_index) { |
| 927 | err = REFTABLE_API_ERROR; |
| 928 | goto done; |
| 929 | } |
| 930 | |
| 931 | err = format_name(&next_name, wr->min_update_index, wr->max_update_index); |
| 932 | if (err < 0) |
| 933 | goto done; |
| 934 | |
| 935 | err = reftable_buf_addstr(&next_name, ".ref"); |
| 936 | if (err < 0) |
| 937 | goto done; |
| 938 | |
| 939 | err = stack_filename(&tab_file_name, add->stack, next_name.buf); |
| 940 | if (err < 0) |
| 941 | goto done; |
| 942 | |
| 943 | /* |
| 944 | On windows, this relies on rand() picking a unique destination name. |
| 945 | Maybe we should do retry loop as well? |
| 946 | */ |
| 947 | err = tmpfile_rename(&tab_file, tab_file_name.buf); |
| 948 | if (err < 0) |
| 949 | goto done; |
| 950 | |
| 951 | REFTABLE_ALLOC_GROW_OR_NULL(add->new_tables, add->new_tables_len + 1, |
| 952 | add->new_tables_cap); |
| 953 | if (!add->new_tables) { |
| 954 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 955 | goto done; |
| 956 | } |
| 957 | add->new_tables[add->new_tables_len++] = reftable_buf_detach(&next_name); |
| 958 | |
| 959 | done: |
| 960 | tmpfile_delete(&tab_file); |
| 961 | reftable_buf_release(&temp_tab_file_name); |
| 962 | reftable_buf_release(&tab_file_name); |
| 963 | reftable_buf_release(&next_name); |
| 964 | reftable_writer_free(wr); |
| 965 | return err; |
| 966 | } |
| 967 | |
| 968 | uint64_t reftable_stack_next_update_index(struct reftable_stack *st) |
| 969 | { |
| 970 | int sz = st->merged->tables_len; |
| 971 | if (sz > 0) |
| 972 | return reftable_table_max_update_index(st->tables[sz - 1]) + |
| 973 | 1; |
| 974 | return 1; |
| 975 | } |
| 976 | |
| 977 | static int stack_write_compact(struct reftable_stack *st, |
| 978 | struct reftable_writer *wr, |
| 979 | size_t first, size_t last, |
| 980 | struct reftable_log_expiry_config *config) |
| 981 | { |
| 982 | struct reftable_merged_table *mt = NULL; |
| 983 | struct reftable_iterator it = { NULL }; |
| 984 | struct reftable_ref_record ref = { NULL }; |
| 985 | struct reftable_log_record log = { NULL }; |
| 986 | size_t subtabs_len = last - first + 1; |
| 987 | uint64_t entries = 0; |
| 988 | int err = 0; |
| 989 | |
| 990 | for (size_t i = first; i <= last; i++) |
| 991 | st->stats.bytes += st->tables[i]->size; |
| 992 | err = reftable_writer_set_limits(wr, st->tables[first]->min_update_index, |
| 993 | st->tables[last]->max_update_index); |
| 994 | if (err < 0) |
| 995 | goto done; |
| 996 | |
| 997 | err = reftable_merged_table_new(&mt, st->tables + first, subtabs_len, |
| 998 | st->opts.hash_id); |
| 999 | if (err < 0) |
| 1000 | goto done; |
| 1001 | |
| 1002 | err = merged_table_init_iter(mt, &it, REFTABLE_BLOCK_TYPE_REF); |
| 1003 | if (err < 0) |
| 1004 | goto done; |
| 1005 | |
| 1006 | err = reftable_iterator_seek_ref(&it, ""); |
| 1007 | if (err < 0) |
| 1008 | goto done; |
| 1009 | |
| 1010 | while (1) { |
| 1011 | err = reftable_iterator_next_ref(&it, &ref); |
| 1012 | if (err > 0) { |
| 1013 | err = 0; |
| 1014 | break; |
| 1015 | } |
| 1016 | if (err < 0) |
| 1017 | goto done; |
| 1018 | |
| 1019 | if (first == 0 && reftable_ref_record_is_deletion(&ref)) { |
| 1020 | continue; |
| 1021 | } |
| 1022 | |
| 1023 | err = reftable_writer_add_ref(wr, &ref); |
| 1024 | if (err < 0) |
| 1025 | goto done; |
| 1026 | entries++; |
| 1027 | } |
| 1028 | reftable_iterator_destroy(&it); |
| 1029 | |
| 1030 | err = merged_table_init_iter(mt, &it, REFTABLE_BLOCK_TYPE_LOG); |
| 1031 | if (err < 0) |
| 1032 | goto done; |
| 1033 | |
| 1034 | err = reftable_iterator_seek_log(&it, ""); |
| 1035 | if (err < 0) |
| 1036 | goto done; |
| 1037 | |
| 1038 | while (1) { |
| 1039 | err = reftable_iterator_next_log(&it, &log); |
| 1040 | if (err > 0) { |
| 1041 | err = 0; |
| 1042 | break; |
| 1043 | } |
| 1044 | if (err < 0) |
| 1045 | goto done; |
| 1046 | if (first == 0 && reftable_log_record_is_deletion(&log)) { |
| 1047 | continue; |
| 1048 | } |
| 1049 | |
| 1050 | if (config && config->min_update_index > 0 && |
| 1051 | log.update_index < config->min_update_index) { |
| 1052 | continue; |
| 1053 | } |
| 1054 | |
| 1055 | if (config && config->time > 0 && |
| 1056 | log.value.update.time < config->time) { |
| 1057 | continue; |
| 1058 | } |
| 1059 | |
| 1060 | err = reftable_writer_add_log(wr, &log); |
| 1061 | if (err < 0) |
| 1062 | goto done; |
| 1063 | entries++; |
| 1064 | } |
| 1065 | |
| 1066 | done: |
| 1067 | reftable_iterator_destroy(&it); |
| 1068 | if (mt) |
| 1069 | reftable_merged_table_free(mt); |
| 1070 | reftable_ref_record_release(&ref); |
| 1071 | reftable_log_record_release(&log); |
| 1072 | st->stats.entries_written += entries; |
| 1073 | return err; |
| 1074 | } |
| 1075 | |
| 1076 | static int stack_compact_locked(struct reftable_stack *st, |
| 1077 | size_t first, size_t last, |
| 1078 | struct reftable_log_expiry_config *config, |
| 1079 | const struct reftable_write_options *opts, |
| 1080 | struct reftable_tmpfile *tab_file_out) |
| 1081 | { |
| 1082 | struct reftable_buf next_name = REFTABLE_BUF_INIT; |
| 1083 | struct reftable_buf tab_file_path = REFTABLE_BUF_INIT; |
| 1084 | struct reftable_writer *wr = NULL; |
| 1085 | struct fd_writer writer= { |
| 1086 | .opts = opts, |
| 1087 | }; |
| 1088 | struct reftable_tmpfile tab_file = REFTABLE_TMPFILE_INIT; |
| 1089 | int err = 0; |
| 1090 | |
| 1091 | err = format_name(&next_name, reftable_table_min_update_index(st->tables[first]), |
| 1092 | reftable_table_max_update_index(st->tables[last])); |
| 1093 | if (err < 0) |
| 1094 | goto done; |
| 1095 | |
| 1096 | err = stack_filename(&tab_file_path, st, next_name.buf); |
| 1097 | if (err < 0) |
| 1098 | goto done; |
| 1099 | |
| 1100 | err = reftable_buf_addstr(&tab_file_path, ".temp.XXXXXX"); |
| 1101 | if (err < 0) |
| 1102 | goto done; |
| 1103 | |
| 1104 | err = tmpfile_from_pattern(&tab_file, tab_file_path.buf); |
| 1105 | if (err < 0) |
| 1106 | goto done; |
| 1107 | |
| 1108 | if (opts->default_permissions && |
| 1109 | chmod(tab_file.path, opts->default_permissions) < 0) { |
| 1110 | err = REFTABLE_IO_ERROR; |
| 1111 | goto done; |
| 1112 | } |
| 1113 | |
| 1114 | writer.fd = tab_file.fd; |
| 1115 | err = reftable_writer_new(&wr, fd_writer_write, fd_writer_flush, |
| 1116 | &writer, st->opts.hash_id, opts); |
| 1117 | if (err < 0) |
| 1118 | goto done; |
| 1119 | |
| 1120 | err = stack_write_compact(st, wr, first, last, config); |
| 1121 | if (err < 0) |
| 1122 | goto done; |
| 1123 | |
| 1124 | err = reftable_writer_close(wr); |
| 1125 | if (err < 0) |
| 1126 | goto done; |
| 1127 | |
| 1128 | err = tmpfile_close(&tab_file); |
| 1129 | if (err < 0) |
| 1130 | goto done; |
| 1131 | |
| 1132 | *tab_file_out = tab_file; |
| 1133 | tab_file = REFTABLE_TMPFILE_INIT; |
| 1134 | |
| 1135 | done: |
| 1136 | tmpfile_delete(&tab_file); |
| 1137 | reftable_writer_free(wr); |
| 1138 | reftable_buf_release(&next_name); |
| 1139 | reftable_buf_release(&tab_file_path); |
| 1140 | return err; |
| 1141 | } |
| 1142 | |
| 1143 | enum stack_compact_range_flags { |
| 1144 | /* |
| 1145 | * Perform a best-effort compaction. That is, even if we cannot lock |
| 1146 | * all tables in the specified range, we will try to compact the |
| 1147 | * remaining slice. |
| 1148 | */ |
| 1149 | STACK_COMPACT_RANGE_BEST_EFFORT = (1 << 0), |
| 1150 | }; |
| 1151 | |
| 1152 | /* |
| 1153 | * Compact all tables in the range `[first, last)` into a single new table. |
| 1154 | * |
| 1155 | * This function returns `0` on success or a code `< 0` on failure. When the |
| 1156 | * stack or any of the tables in the specified range are already locked then |
| 1157 | * this function returns `REFTABLE_LOCK_ERROR`. This is a benign error that |
| 1158 | * callers can either ignore, or they may choose to retry compaction after some |
| 1159 | * amount of time. |
| 1160 | */ |
| 1161 | static int stack_compact_range(struct reftable_stack *st, |
| 1162 | size_t first, size_t last, |
| 1163 | struct reftable_log_expiry_config *expiry, |
| 1164 | const struct reftable_write_options *opts, |
| 1165 | unsigned int flags) |
| 1166 | { |
| 1167 | struct reftable_buf tables_list_buf = REFTABLE_BUF_INIT; |
| 1168 | struct reftable_buf new_table_name = REFTABLE_BUF_INIT; |
| 1169 | struct reftable_buf new_table_path = REFTABLE_BUF_INIT; |
| 1170 | struct reftable_buf table_name = REFTABLE_BUF_INIT; |
| 1171 | struct reftable_flock tables_list_lock = REFTABLE_FLOCK_INIT; |
| 1172 | struct reftable_flock *table_locks = NULL; |
| 1173 | struct reftable_tmpfile new_table = REFTABLE_TMPFILE_INIT; |
| 1174 | int is_empty_table = 0, err = 0; |
| 1175 | size_t first_to_replace, last_to_replace; |
| 1176 | size_t i, nlocks = 0; |
| 1177 | char **names = NULL; |
| 1178 | |
| 1179 | if (first > last || (!expiry && first == last)) { |
| 1180 | err = 0; |
| 1181 | goto done; |
| 1182 | } |
| 1183 | |
| 1184 | st->stats.attempts++; |
| 1185 | |
| 1186 | /* |
| 1187 | * Hold the lock so that we can read "tables.list" and lock all tables |
| 1188 | * which are part of the user-specified range. |
| 1189 | */ |
| 1190 | err = flock_acquire(&tables_list_lock, st->list_file, opts->lock_timeout_ms); |
| 1191 | if (err < 0) |
| 1192 | goto done; |
| 1193 | |
| 1194 | /* |
| 1195 | * Check whether the stack is up-to-date. We unfortunately cannot |
| 1196 | * handle the situation gracefully in case it's _not_ up-to-date |
| 1197 | * because the range of tables that the user has requested us to |
| 1198 | * compact may have been changed. So instead we abort. |
| 1199 | * |
| 1200 | * We could in theory improve the situation by having the caller not |
| 1201 | * pass in a range, but instead the list of tables to compact. If so, |
| 1202 | * we could check that relevant tables still exist. But for now it's |
| 1203 | * good enough to just abort. |
| 1204 | */ |
| 1205 | err = stack_uptodate(st); |
| 1206 | if (err < 0) |
| 1207 | goto done; |
| 1208 | if (err > 0) { |
| 1209 | err = REFTABLE_OUTDATED_ERROR; |
| 1210 | goto done; |
| 1211 | } |
| 1212 | |
| 1213 | /* |
| 1214 | * Lock all tables in the user-provided range. This is the slice of our |
| 1215 | * stack which we'll compact. |
| 1216 | * |
| 1217 | * Note that we lock tables in reverse order from last to first. The |
| 1218 | * intent behind this is to allow a newer process to perform best |
| 1219 | * effort compaction of tables that it has added in the case where an |
| 1220 | * older process is still busy compacting tables which are preexisting |
| 1221 | * from the point of view of the newer process. |
| 1222 | */ |
| 1223 | REFTABLE_ALLOC_ARRAY(table_locks, last - first + 1); |
| 1224 | if (!table_locks) { |
| 1225 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 1226 | goto done; |
| 1227 | } |
| 1228 | for (i = 0; i < last - first + 1; i++) |
| 1229 | table_locks[i] = REFTABLE_FLOCK_INIT; |
| 1230 | |
| 1231 | for (i = last + 1; i > first; i--) { |
| 1232 | err = stack_filename(&table_name, st, reftable_table_name(st->tables[i - 1])); |
| 1233 | if (err < 0) |
| 1234 | goto done; |
| 1235 | |
| 1236 | err = flock_acquire(&table_locks[nlocks], table_name.buf, 0); |
| 1237 | if (err < 0) { |
| 1238 | /* |
| 1239 | * When the table is locked already we may do a |
| 1240 | * best-effort compaction and compact only the tables |
| 1241 | * that we have managed to lock so far. This of course |
| 1242 | * requires that we have been able to lock at least two |
| 1243 | * tables, otherwise there would be nothing to compact. |
| 1244 | * In that case, we return a lock error to our caller. |
| 1245 | */ |
| 1246 | if (err == REFTABLE_LOCK_ERROR && last - (i - 1) >= 2 && |
| 1247 | flags & STACK_COMPACT_RANGE_BEST_EFFORT) { |
| 1248 | err = 0; |
| 1249 | /* |
| 1250 | * The subtraction is to offset the index, the |
| 1251 | * addition is to only compact up to the table |
| 1252 | * of the preceding iteration. They obviously |
| 1253 | * cancel each other out, but that may be |
| 1254 | * non-obvious when it was omitted. |
| 1255 | */ |
| 1256 | first = (i - 1) + 1; |
| 1257 | break; |
| 1258 | } |
| 1259 | |
| 1260 | goto done; |
| 1261 | } |
| 1262 | |
| 1263 | /* |
| 1264 | * We need to close the lockfiles as we might otherwise easily |
| 1265 | * run into file descriptor exhaustion when we compress a lot |
| 1266 | * of tables. |
| 1267 | */ |
| 1268 | err = flock_close(&table_locks[nlocks++]); |
| 1269 | if (err < 0) |
| 1270 | goto done; |
| 1271 | } |
| 1272 | |
| 1273 | /* |
| 1274 | * We have locked all tables in our range and can thus release the |
| 1275 | * "tables.list" lock while compacting the locked tables. This allows |
| 1276 | * concurrent updates to the stack to proceed. |
| 1277 | */ |
| 1278 | err = flock_release(&tables_list_lock); |
| 1279 | if (err < 0) { |
| 1280 | err = REFTABLE_IO_ERROR; |
| 1281 | goto done; |
| 1282 | } |
| 1283 | |
| 1284 | /* |
| 1285 | * Compact the now-locked tables into a new table. Note that compacting |
| 1286 | * these tables may end up with an empty new table in case tombstones |
| 1287 | * end up cancelling out all refs in that range. |
| 1288 | */ |
| 1289 | err = stack_compact_locked(st, first, last, expiry, opts, &new_table); |
| 1290 | if (err < 0) { |
| 1291 | if (err != REFTABLE_EMPTY_TABLE_ERROR) |
| 1292 | goto done; |
| 1293 | is_empty_table = 1; |
| 1294 | } |
| 1295 | |
| 1296 | /* |
| 1297 | * Now that we have written the new, compacted table we need to re-lock |
| 1298 | * "tables.list". We'll then replace the compacted range of tables with |
| 1299 | * the new table. |
| 1300 | */ |
| 1301 | err = flock_acquire(&tables_list_lock, st->list_file, opts->lock_timeout_ms); |
| 1302 | if (err < 0) |
| 1303 | goto done; |
| 1304 | |
| 1305 | if (opts->default_permissions) { |
| 1306 | if (chmod(tables_list_lock.path, |
| 1307 | opts->default_permissions) < 0) { |
| 1308 | err = REFTABLE_IO_ERROR; |
| 1309 | goto done; |
| 1310 | } |
| 1311 | } |
| 1312 | |
| 1313 | /* |
| 1314 | * As we have unlocked the stack while compacting our slice of tables |
| 1315 | * it may have happened that a concurrently running process has updated |
| 1316 | * the stack while we were compacting. In that case, we need to check |
| 1317 | * whether the tables that we have just compacted still exist in the |
| 1318 | * stack in the exact same order as we have compacted them. |
| 1319 | * |
| 1320 | * If they do exist, then it is fine to continue and replace those |
| 1321 | * tables with our compacted version. If they don't, then we need to |
| 1322 | * abort. |
| 1323 | */ |
| 1324 | err = stack_uptodate(st); |
| 1325 | if (err < 0) |
| 1326 | goto done; |
| 1327 | if (err > 0) { |
| 1328 | ssize_t new_offset = -1; |
| 1329 | int fd; |
| 1330 | |
| 1331 | fd = open(st->list_file, O_RDONLY); |
| 1332 | if (fd < 0) { |
| 1333 | err = REFTABLE_IO_ERROR; |
| 1334 | goto done; |
| 1335 | } |
| 1336 | |
| 1337 | err = fd_read_lines(fd, &names); |
| 1338 | close(fd); |
| 1339 | if (err < 0) |
| 1340 | goto done; |
| 1341 | |
| 1342 | /* |
| 1343 | * Search for the offset of the first table that we have |
| 1344 | * compacted in the updated "tables.list" file. |
| 1345 | */ |
| 1346 | for (size_t i = 0; names[i]; i++) { |
| 1347 | if (strcmp(names[i], st->tables[first]->name)) |
| 1348 | continue; |
| 1349 | |
| 1350 | /* |
| 1351 | * We have found the first entry. Verify that all the |
| 1352 | * subsequent tables we have compacted still exist in |
| 1353 | * the modified stack in the exact same order as we |
| 1354 | * have compacted them. |
| 1355 | */ |
| 1356 | for (size_t j = 1; j < last - first + 1; j++) { |
| 1357 | const char *old = first + j < st->merged->tables_len ? |
| 1358 | st->tables[first + j]->name : NULL; |
| 1359 | const char *new = names[i + j]; |
| 1360 | |
| 1361 | /* |
| 1362 | * If some entries are missing or in case the tables |
| 1363 | * have changed then we need to bail out. Again, this |
| 1364 | * shouldn't ever happen because we have locked the |
| 1365 | * tables we are compacting. |
| 1366 | */ |
| 1367 | if (!old || !new || strcmp(old, new)) { |
| 1368 | err = REFTABLE_OUTDATED_ERROR; |
| 1369 | goto done; |
| 1370 | } |
| 1371 | } |
| 1372 | |
| 1373 | new_offset = i; |
| 1374 | break; |
| 1375 | } |
| 1376 | |
| 1377 | /* |
| 1378 | * In case we didn't find our compacted tables in the stack we |
| 1379 | * need to bail out. In theory, this should have never happened |
| 1380 | * because we locked the tables we are compacting. |
| 1381 | */ |
| 1382 | if (new_offset < 0) { |
| 1383 | err = REFTABLE_OUTDATED_ERROR; |
| 1384 | goto done; |
| 1385 | } |
| 1386 | |
| 1387 | /* |
| 1388 | * We have found the new range that we want to replace, so |
| 1389 | * let's update the range of tables that we want to replace. |
| 1390 | */ |
| 1391 | first_to_replace = new_offset; |
| 1392 | last_to_replace = last + (new_offset - first); |
| 1393 | } else { |
| 1394 | /* |
| 1395 | * `fd_read_lines()` uses a `NULL` sentinel to indicate that |
| 1396 | * the array is at its end. As we use `free_names()` to free |
| 1397 | * the array, we need to include this sentinel value here and |
| 1398 | * thus have to allocate `tables_len + 1` many entries. |
| 1399 | */ |
| 1400 | REFTABLE_CALLOC_ARRAY(names, st->merged->tables_len + 1); |
| 1401 | if (!names) { |
| 1402 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 1403 | goto done; |
| 1404 | } |
| 1405 | |
| 1406 | for (size_t i = 0; i < st->merged->tables_len; i++) { |
| 1407 | names[i] = reftable_strdup(st->tables[i]->name); |
| 1408 | if (!names[i]) { |
| 1409 | err = REFTABLE_OUT_OF_MEMORY_ERROR; |
| 1410 | goto done; |
| 1411 | } |
| 1412 | } |
| 1413 | first_to_replace = first; |
| 1414 | last_to_replace = last; |
| 1415 | } |
| 1416 | |
| 1417 | /* |
| 1418 | * If the resulting compacted table is not empty, then we need to move |
| 1419 | * it into place now. |
| 1420 | */ |
| 1421 | if (!is_empty_table) { |
| 1422 | err = format_name(&new_table_name, st->tables[first]->min_update_index, |
| 1423 | st->tables[last]->max_update_index); |
| 1424 | if (err < 0) |
| 1425 | goto done; |
| 1426 | |
| 1427 | err = reftable_buf_addstr(&new_table_name, ".ref"); |
| 1428 | if (err < 0) |
| 1429 | goto done; |
| 1430 | |
| 1431 | err = stack_filename(&new_table_path, st, new_table_name.buf); |
| 1432 | if (err < 0) |
| 1433 | goto done; |
| 1434 | |
| 1435 | err = tmpfile_rename(&new_table, new_table_path.buf); |
| 1436 | if (err < 0) |
| 1437 | goto done; |
| 1438 | } |
| 1439 | |
| 1440 | /* |
| 1441 | * Write the new "tables.list" contents with the compacted table we |
| 1442 | * have just written. In case the compacted table became empty we |
| 1443 | * simply skip writing it. |
| 1444 | */ |
| 1445 | for (i = 0; i < first_to_replace; i++) { |
| 1446 | if ((err = reftable_buf_addstr(&tables_list_buf, names[i])) < 0 || |
| 1447 | (err = reftable_buf_addstr(&tables_list_buf, "\n")) < 0) |
| 1448 | goto done; |
| 1449 | } |
| 1450 | if (!is_empty_table) { |
| 1451 | if ((err = reftable_buf_addstr(&tables_list_buf, new_table_name.buf)) < 0 || |
| 1452 | (err = reftable_buf_addstr(&tables_list_buf, "\n")) < 0) |
| 1453 | goto done; |
| 1454 | } |
| 1455 | for (i = last_to_replace + 1; names[i]; i++) { |
| 1456 | if ((err = reftable_buf_addstr(&tables_list_buf, names[i])) < 0 || |
| 1457 | (err = reftable_buf_addstr(&tables_list_buf, "\n")) < 0) |
| 1458 | goto done; |
| 1459 | } |
| 1460 | |
| 1461 | err = reftable_write_data(tables_list_lock.fd, |
| 1462 | tables_list_buf.buf, tables_list_buf.len); |
| 1463 | if (err < 0) { |
| 1464 | err = REFTABLE_IO_ERROR; |
| 1465 | unlink(new_table_path.buf); |
| 1466 | goto done; |
| 1467 | } |
| 1468 | |
| 1469 | err = fsync(tables_list_lock.fd); |
| 1470 | if (err < 0) { |
| 1471 | err = REFTABLE_IO_ERROR; |
| 1472 | unlink(new_table_path.buf); |
| 1473 | goto done; |
| 1474 | } |
| 1475 | |
| 1476 | err = flock_commit(&tables_list_lock); |
| 1477 | if (err < 0) { |
| 1478 | err = REFTABLE_IO_ERROR; |
| 1479 | unlink(new_table_path.buf); |
| 1480 | goto done; |
| 1481 | } |
| 1482 | |
| 1483 | /* |
| 1484 | * Reload the stack before deleting the compacted tables. We can only |
| 1485 | * delete the files after we closed them on Windows, so this needs to |
| 1486 | * happen first. |
| 1487 | */ |
| 1488 | err = reftable_stack_reload_maybe_reuse(st, first < last); |
| 1489 | if (err < 0) |
| 1490 | goto done; |
| 1491 | |
| 1492 | /* |
| 1493 | * Delete the old tables. They may still be in use by concurrent |
| 1494 | * readers, so it is expected that unlinking tables may fail. |
| 1495 | */ |
| 1496 | for (i = 0; i < nlocks; i++) { |
| 1497 | struct reftable_flock *table_lock = &table_locks[i]; |
| 1498 | |
| 1499 | reftable_buf_reset(&table_name); |
| 1500 | err = reftable_buf_add(&table_name, table_lock->path, |
| 1501 | strlen(table_lock->path) - strlen(".lock")); |
| 1502 | if (err) |
| 1503 | continue; |
| 1504 | |
| 1505 | unlink(table_name.buf); |
| 1506 | } |
| 1507 | |
| 1508 | done: |
| 1509 | flock_release(&tables_list_lock); |
| 1510 | for (i = 0; table_locks && i < nlocks; i++) |
| 1511 | flock_release(&table_locks[i]); |
| 1512 | reftable_free(table_locks); |
| 1513 | |
| 1514 | tmpfile_delete(&new_table); |
| 1515 | reftable_buf_release(&new_table_name); |
| 1516 | reftable_buf_release(&new_table_path); |
| 1517 | reftable_buf_release(&tables_list_buf); |
| 1518 | reftable_buf_release(&table_name); |
| 1519 | free_names(names); |
| 1520 | |
| 1521 | if (err == REFTABLE_LOCK_ERROR) |
| 1522 | st->stats.failures++; |
| 1523 | |
| 1524 | return err; |
| 1525 | } |
| 1526 | |
| 1527 | int reftable_stack_compact_all(struct reftable_stack *st, |
| 1528 | const struct reftable_write_options *opts, |
| 1529 | struct reftable_log_expiry_config *config) |
| 1530 | { |
| 1531 | struct reftable_write_options opts_default = { 0 }; |
| 1532 | size_t last = st->merged->tables_len ? st->merged->tables_len - 1 : 0; |
| 1533 | |
| 1534 | if (!opts) |
| 1535 | opts = &opts_default; |
| 1536 | |
| 1537 | return stack_compact_range(st, 0, last, config, opts, 0); |
| 1538 | } |
| 1539 | |
| 1540 | static int segment_size(struct segment *s) |
| 1541 | { |
| 1542 | return s->end - s->start; |
| 1543 | } |
| 1544 | |
| 1545 | struct segment suggest_compaction_segment(uint64_t *sizes, size_t n, |
| 1546 | uint8_t factor) |
| 1547 | { |
| 1548 | struct segment seg = { 0 }; |
| 1549 | uint64_t bytes; |
| 1550 | size_t i; |
| 1551 | |
| 1552 | if (!factor) |
| 1553 | factor = DEFAULT_GEOMETRIC_FACTOR; |
| 1554 | |
| 1555 | /* |
| 1556 | * If there are no tables or only a single one then we don't have to |
| 1557 | * compact anything. The sequence is geometric by definition already. |
| 1558 | */ |
| 1559 | if (n <= 1) |
| 1560 | return seg; |
| 1561 | |
| 1562 | /* |
| 1563 | * Find the ending table of the compaction segment needed to restore the |
| 1564 | * geometric sequence. Note that the segment end is exclusive. |
| 1565 | * |
| 1566 | * To do so, we iterate backwards starting from the most recent table |
| 1567 | * until a valid segment end is found. If the preceding table is smaller |
| 1568 | * than the current table multiplied by the geometric factor (2), the |
| 1569 | * compaction segment end has been identified. |
| 1570 | * |
| 1571 | * Tables after the ending point are not added to the byte count because |
| 1572 | * they are already valid members of the geometric sequence. Due to the |
| 1573 | * properties of a geometric sequence, it is not possible for the sum of |
| 1574 | * these tables to exceed the value of the ending point table. |
| 1575 | * |
| 1576 | * Example table size sequence requiring no compaction: |
| 1577 | * 64, 32, 16, 8, 4, 2, 1 |
| 1578 | * |
| 1579 | * Example table size sequence where compaction segment end is set to |
| 1580 | * the last table. Since the segment end is exclusive, the last table is |
| 1581 | * excluded during subsequent compaction and the table with size 3 is |
| 1582 | * the final table included: |
| 1583 | * 64, 32, 16, 8, 4, 3, 1 |
| 1584 | */ |
| 1585 | for (i = n - 1; i > 0; i--) { |
| 1586 | if (sizes[i - 1] < sizes[i] * factor) { |
| 1587 | seg.end = i + 1; |
| 1588 | bytes = sizes[i]; |
| 1589 | break; |
| 1590 | } |
| 1591 | } |
| 1592 | |
| 1593 | /* |
| 1594 | * Find the starting table of the compaction segment by iterating |
| 1595 | * through the remaining tables and keeping track of the accumulated |
| 1596 | * size of all tables seen from the segment end table. The previous |
| 1597 | * table is compared to the accumulated size because the tables from the |
| 1598 | * segment end are merged backwards recursively. |
| 1599 | * |
| 1600 | * Note that we keep iterating even after we have found the first |
| 1601 | * starting point. This is because there may be tables in the stack |
| 1602 | * preceding that first starting point which violate the geometric |
| 1603 | * sequence. |
| 1604 | * |
| 1605 | * Example compaction segment start set to table with size 32: |
| 1606 | * 128, 32, 16, 8, 4, 3, 1 |
| 1607 | */ |
| 1608 | for (; i > 0; i--) { |
| 1609 | uint64_t curr = bytes; |
| 1610 | bytes += sizes[i - 1]; |
| 1611 | |
| 1612 | if (sizes[i - 1] < curr * factor) { |
| 1613 | seg.start = i - 1; |
| 1614 | seg.bytes = bytes; |
| 1615 | } |
| 1616 | } |
| 1617 | |
| 1618 | return seg; |
| 1619 | } |
| 1620 | |
| 1621 | static int stack_segments_for_compaction(struct reftable_stack *st, |
| 1622 | const struct reftable_write_options *opts, |
| 1623 | struct segment *seg) |
| 1624 | { |
| 1625 | int version = (st->opts.hash_id == REFTABLE_HASH_SHA1) ? 1 : 2; |
| 1626 | int overhead = header_size(version) - 1; |
| 1627 | uint64_t *sizes; |
| 1628 | |
| 1629 | REFTABLE_CALLOC_ARRAY(sizes, st->merged->tables_len); |
| 1630 | if (!sizes) |
| 1631 | return REFTABLE_OUT_OF_MEMORY_ERROR; |
| 1632 | |
| 1633 | for (size_t i = 0; i < st->merged->tables_len; i++) |
| 1634 | sizes[i] = st->tables[i]->size - overhead; |
| 1635 | |
| 1636 | *seg = suggest_compaction_segment(sizes, st->merged->tables_len, |
| 1637 | opts->auto_compaction_factor); |
| 1638 | reftable_free(sizes); |
| 1639 | |
| 1640 | return 0; |
| 1641 | } |
| 1642 | |
| 1643 | static int update_segment_if_compaction_required(struct reftable_stack *st, |
| 1644 | const struct reftable_write_options *opts, |
| 1645 | struct segment *seg, |
| 1646 | bool use_geometric, |
| 1647 | bool *required) |
| 1648 | { |
| 1649 | int err; |
| 1650 | |
| 1651 | if (st->merged->tables_len < 2) { |
| 1652 | *required = false; |
| 1653 | return 0; |
| 1654 | } |
| 1655 | |
| 1656 | if (!use_geometric) { |
| 1657 | *required = true; |
| 1658 | return 0; |
| 1659 | } |
| 1660 | |
| 1661 | err = stack_segments_for_compaction(st, opts, seg); |
| 1662 | if (err) |
| 1663 | return err; |
| 1664 | |
| 1665 | *required = segment_size(seg) > 0; |
| 1666 | return 0; |
| 1667 | } |
| 1668 | |
| 1669 | int reftable_stack_compaction_required(struct reftable_stack *st, |
| 1670 | const struct reftable_write_options *opts, |
| 1671 | bool use_heuristics, |
| 1672 | bool *required) |
| 1673 | { |
| 1674 | struct reftable_write_options opts_default = { 0 }; |
| 1675 | struct segment seg; |
| 1676 | |
| 1677 | if (!opts) |
| 1678 | opts = &opts_default; |
| 1679 | |
| 1680 | return update_segment_if_compaction_required(st, opts, &seg, |
| 1681 | use_heuristics, required); |
| 1682 | } |
| 1683 | |
| 1684 | int reftable_stack_auto_compact(struct reftable_stack *st, |
| 1685 | const struct reftable_write_options *opts) |
| 1686 | { |
| 1687 | struct reftable_write_options opts_default = { 0 }; |
| 1688 | struct segment seg; |
| 1689 | bool required; |
| 1690 | int err; |
| 1691 | |
| 1692 | if (!opts) |
| 1693 | opts = &opts_default; |
| 1694 | |
| 1695 | err = update_segment_if_compaction_required(st, opts, &seg, true, |
| 1696 | &required); |
| 1697 | if (err) |
| 1698 | return err; |
| 1699 | |
| 1700 | if (required) |
| 1701 | return stack_compact_range(st, seg.start, seg.end - 1, |
| 1702 | NULL, opts, |
| 1703 | STACK_COMPACT_RANGE_BEST_EFFORT); |
| 1704 | |
| 1705 | return 0; |
| 1706 | } |
| 1707 | |
| 1708 | struct reftable_compaction_stats * |
| 1709 | reftable_stack_compaction_stats(struct reftable_stack *st) |
| 1710 | { |
| 1711 | return &st->stats; |
| 1712 | } |
| 1713 | |
| 1714 | int reftable_stack_read_ref(struct reftable_stack *st, const char *refname, |
| 1715 | struct reftable_ref_record *ref) |
| 1716 | { |
| 1717 | struct reftable_iterator it = { 0 }; |
| 1718 | int ret; |
| 1719 | |
| 1720 | ret = reftable_merged_table_init_ref_iterator(st->merged, &it); |
| 1721 | if (ret) |
| 1722 | goto out; |
| 1723 | |
| 1724 | ret = reftable_iterator_seek_ref(&it, refname); |
| 1725 | if (ret) |
| 1726 | goto out; |
| 1727 | |
| 1728 | ret = reftable_iterator_next_ref(&it, ref); |
| 1729 | if (ret) |
| 1730 | goto out; |
| 1731 | |
| 1732 | if (strcmp(ref->refname, refname) || |
| 1733 | reftable_ref_record_is_deletion(ref)) { |
| 1734 | reftable_ref_record_release(ref); |
| 1735 | ret = 1; |
| 1736 | goto out; |
| 1737 | } |
| 1738 | |
| 1739 | out: |
| 1740 | reftable_iterator_destroy(&it); |
| 1741 | return ret; |
| 1742 | } |
| 1743 | |
| 1744 | int reftable_stack_read_log(struct reftable_stack *st, const char *refname, |
| 1745 | struct reftable_log_record *log) |
| 1746 | { |
| 1747 | struct reftable_iterator it = {0}; |
| 1748 | int err; |
| 1749 | |
| 1750 | err = reftable_stack_init_log_iterator(st, &it); |
| 1751 | if (err) |
| 1752 | goto done; |
| 1753 | |
| 1754 | err = reftable_iterator_seek_log(&it, refname); |
| 1755 | if (err) |
| 1756 | goto done; |
| 1757 | |
| 1758 | err = reftable_iterator_next_log(&it, log); |
| 1759 | if (err) |
| 1760 | goto done; |
| 1761 | |
| 1762 | if (strcmp(log->refname, refname) || |
| 1763 | reftable_log_record_is_deletion(log)) { |
| 1764 | err = 1; |
| 1765 | goto done; |
| 1766 | } |
| 1767 | |
| 1768 | done: |
| 1769 | if (err) { |
| 1770 | reftable_log_record_release(log); |
| 1771 | } |
| 1772 | reftable_iterator_destroy(&it); |
| 1773 | return err; |
| 1774 | } |
| 1775 | |
| 1776 | static int is_table_name(const char *s) |
| 1777 | { |
| 1778 | const char *dot = strrchr(s, '.'); |
| 1779 | return dot && !strcmp(dot, ".ref"); |
| 1780 | } |
| 1781 | |
| 1782 | static void remove_maybe_stale_table(struct reftable_stack *st, uint64_t max, |
| 1783 | const char *name) |
| 1784 | { |
| 1785 | int err = 0; |
| 1786 | uint64_t update_idx = 0; |
| 1787 | struct reftable_block_source src = { NULL }; |
| 1788 | struct reftable_table *table = NULL; |
| 1789 | struct reftable_buf table_path = REFTABLE_BUF_INIT; |
| 1790 | |
| 1791 | err = stack_filename(&table_path, st, name); |
| 1792 | if (err < 0) |
| 1793 | goto done; |
| 1794 | |
| 1795 | err = reftable_block_source_from_file(&src, table_path.buf); |
| 1796 | if (err < 0) |
| 1797 | goto done; |
| 1798 | |
| 1799 | err = reftable_table_new(&table, &src, name); |
| 1800 | if (err < 0) |
| 1801 | goto done; |
| 1802 | |
| 1803 | update_idx = reftable_table_max_update_index(table); |
| 1804 | reftable_table_decref(table); |
| 1805 | |
| 1806 | if (update_idx <= max) { |
| 1807 | unlink(table_path.buf); |
| 1808 | } |
| 1809 | done: |
| 1810 | reftable_buf_release(&table_path); |
| 1811 | } |
| 1812 | |
| 1813 | static int reftable_stack_clean_locked(struct reftable_stack *st) |
| 1814 | { |
| 1815 | uint64_t max = reftable_merged_table_max_update_index( |
| 1816 | reftable_stack_merged_table(st)); |
| 1817 | DIR *dir = opendir(st->reftable_dir); |
| 1818 | struct dirent *d = NULL; |
| 1819 | if (!dir) { |
| 1820 | return REFTABLE_IO_ERROR; |
| 1821 | } |
| 1822 | |
| 1823 | while ((d = readdir(dir))) { |
| 1824 | int found = 0; |
| 1825 | if (!is_table_name(d->d_name)) |
| 1826 | continue; |
| 1827 | |
| 1828 | for (size_t i = 0; !found && i < st->tables_len; i++) |
| 1829 | found = !strcmp(reftable_table_name(st->tables[i]), d->d_name); |
| 1830 | if (found) |
| 1831 | continue; |
| 1832 | |
| 1833 | remove_maybe_stale_table(st, max, d->d_name); |
| 1834 | } |
| 1835 | |
| 1836 | closedir(dir); |
| 1837 | return 0; |
| 1838 | } |
| 1839 | |
| 1840 | int reftable_stack_clean(struct reftable_stack *st) |
| 1841 | { |
| 1842 | struct reftable_addition *add = NULL; |
| 1843 | int err = reftable_stack_new_addition(&add, st, NULL, 0); |
| 1844 | if (err < 0) { |
| 1845 | goto done; |
| 1846 | } |
| 1847 | |
| 1848 | err = reftable_stack_reload(st); |
| 1849 | if (err < 0) { |
| 1850 | goto done; |
| 1851 | } |
| 1852 | |
| 1853 | err = reftable_stack_clean_locked(st); |
| 1854 | |
| 1855 | done: |
| 1856 | reftable_addition_destroy(add); |
| 1857 | return err; |
| 1858 | } |
| 1859 | |
| 1860 | enum reftable_hash reftable_stack_hash_id(struct reftable_stack *st) |
| 1861 | { |
| 1862 | return reftable_merged_table_hash_id(st->merged); |
| 1863 | } |