| 1 | #define DISABLE_SIGN_COMPARE_WARNINGS |
| 2 | |
| 3 | #include "../git-compat-util.h" |
| 4 | #include "../config.h" |
| 5 | #include "../dir.h" |
| 6 | #include "../fsck.h" |
| 7 | #include "../gettext.h" |
| 8 | #include "../hash.h" |
| 9 | #include "../hex.h" |
| 10 | #include "../refs.h" |
| 11 | #include "refs-internal.h" |
| 12 | #include "packed-backend.h" |
| 13 | #include "../iterator.h" |
| 14 | #include "../lockfile.h" |
| 15 | #include "../chdir-notify.h" |
| 16 | #include "../statinfo.h" |
| 17 | #include "../worktree.h" |
| 18 | #include "../wrapper.h" |
| 19 | #include "../write-or-die.h" |
| 20 | #include "../trace2.h" |
| 21 | |
| 22 | enum mmap_strategy { |
| 23 | /* |
| 24 | * Don't use mmap() at all for reading `packed-refs`. |
| 25 | */ |
| 26 | MMAP_NONE, |
| 27 | |
| 28 | /* |
| 29 | * Can use mmap() for reading `packed-refs`, but the file must |
| 30 | * not remain mmapped. This is the usual option on Windows, |
| 31 | * where you cannot rename a new version of a file onto a file |
| 32 | * that is currently mmapped. |
| 33 | */ |
| 34 | MMAP_TEMPORARY, |
| 35 | |
| 36 | /* |
| 37 | * It is OK to leave the `packed-refs` file mmapped while |
| 38 | * arbitrary other code is running. |
| 39 | */ |
| 40 | MMAP_OK |
| 41 | }; |
| 42 | |
| 43 | #if defined(NO_MMAP) |
| 44 | static enum mmap_strategy mmap_strategy = MMAP_NONE; |
| 45 | #elif defined(MMAP_PREVENTS_DELETE) |
| 46 | static enum mmap_strategy mmap_strategy = MMAP_TEMPORARY; |
| 47 | #else |
| 48 | static enum mmap_strategy mmap_strategy = MMAP_OK; |
| 49 | #endif |
| 50 | |
| 51 | struct packed_ref_store; |
| 52 | |
| 53 | /* |
| 54 | * A `snapshot` represents one snapshot of a `packed-refs` file. |
| 55 | * |
| 56 | * Normally, this will be a mmapped view of the contents of the |
| 57 | * `packed-refs` file at the time the snapshot was created. However, |
| 58 | * if the `packed-refs` file was not sorted, this might point at heap |
| 59 | * memory holding the contents of the `packed-refs` file with its |
| 60 | * records sorted by refname. |
| 61 | * |
| 62 | * `snapshot` instances are reference counted (via |
| 63 | * `acquire_snapshot()` and `release_snapshot()`). This is to prevent |
| 64 | * an instance from disappearing while an iterator is still iterating |
| 65 | * over it. Instances are garbage collected when their `referrers` |
| 66 | * count goes to zero. |
| 67 | * |
| 68 | * The most recent `snapshot`, if available, is referenced by the |
| 69 | * `packed_ref_store`. Its freshness is checked whenever |
| 70 | * `get_snapshot()` is called; if the existing snapshot is obsolete, a |
| 71 | * new snapshot is taken. |
| 72 | */ |
| 73 | struct snapshot { |
| 74 | /* |
| 75 | * A back-pointer to the packed_ref_store with which this |
| 76 | * snapshot is associated: |
| 77 | */ |
| 78 | struct packed_ref_store *refs; |
| 79 | |
| 80 | /* Is the `packed-refs` file currently mmapped? */ |
| 81 | int mmapped; |
| 82 | |
| 83 | /* |
| 84 | * The contents of the `packed-refs` file: |
| 85 | * |
| 86 | * - buf -- a pointer to the start of the memory |
| 87 | * - start -- a pointer to the first byte of actual references |
| 88 | * (i.e., after the header line, if one is present) |
| 89 | * - eof -- a pointer just past the end of the reference |
| 90 | * contents |
| 91 | * |
| 92 | * If the `packed-refs` file was already sorted, `buf` points |
| 93 | * at the mmapped contents of the file. If not, it points at |
| 94 | * heap-allocated memory containing the contents, sorted. If |
| 95 | * there were no contents (e.g., because the file didn't |
| 96 | * exist), `buf`, `start`, and `eof` are all NULL. |
| 97 | */ |
| 98 | char *buf, *start, *eof; |
| 99 | |
| 100 | /* |
| 101 | * What is the peeled state of the `packed-refs` file that |
| 102 | * this snapshot represents? (This is usually determined from |
| 103 | * the file's header.) |
| 104 | */ |
| 105 | enum { PEELED_NONE, PEELED_TAGS, PEELED_FULLY } peeled; |
| 106 | |
| 107 | /* |
| 108 | * Count of references to this instance, including the pointer |
| 109 | * from `packed_ref_store::snapshot`, if any. The instance |
| 110 | * will not be freed as long as the reference count is |
| 111 | * nonzero. |
| 112 | */ |
| 113 | unsigned int referrers; |
| 114 | |
| 115 | /* |
| 116 | * The metadata of the `packed-refs` file from which this |
| 117 | * snapshot was created, used to tell if the file has been |
| 118 | * replaced since we read it. |
| 119 | */ |
| 120 | struct stat_validity validity; |
| 121 | }; |
| 122 | |
| 123 | /* |
| 124 | * A `ref_store` representing references stored in a `packed-refs` |
| 125 | * file. It implements the `ref_store` interface, though it has some |
| 126 | * limitations: |
| 127 | * |
| 128 | * - It cannot store symbolic references. |
| 129 | * |
| 130 | * - It cannot store reflogs. |
| 131 | * |
| 132 | * - It does not support reference renaming (though it could). |
| 133 | * |
| 134 | * On the other hand, it can be locked outside of a reference |
| 135 | * transaction. In that case, it remains locked even after the |
| 136 | * transaction is done and the new `packed-refs` file is activated. |
| 137 | */ |
| 138 | struct packed_ref_store { |
| 139 | struct ref_store base; |
| 140 | |
| 141 | unsigned int store_flags; |
| 142 | |
| 143 | /* The path of the "packed-refs" file: */ |
| 144 | char *path; |
| 145 | |
| 146 | /* |
| 147 | * A snapshot of the values read from the `packed-refs` file, |
| 148 | * if it might still be current; otherwise, NULL. |
| 149 | */ |
| 150 | struct snapshot *snapshot; |
| 151 | |
| 152 | /* |
| 153 | * Lock used for the "packed-refs" file. Note that this (and |
| 154 | * thus the enclosing `packed_ref_store`) must not be freed. |
| 155 | */ |
| 156 | struct lock_file lock; |
| 157 | |
| 158 | /* |
| 159 | * Temporary file used when rewriting new contents to the |
| 160 | * "packed-refs" file. Note that this (and thus the enclosing |
| 161 | * `packed_ref_store`) must not be freed. |
| 162 | */ |
| 163 | struct tempfile *tempfile; |
| 164 | |
| 165 | /* |
| 166 | * Timeout when taking the "packed-refs.lock" file. configurable via |
| 167 | * "core.packedRefsTimeout". |
| 168 | */ |
| 169 | bool timeout_configured; |
| 170 | int timeout_value; |
| 171 | }; |
| 172 | |
| 173 | /* |
| 174 | * Increment the reference count of `*snapshot`. |
| 175 | */ |
| 176 | static void acquire_snapshot(struct snapshot *snapshot) |
| 177 | { |
| 178 | snapshot->referrers++; |
| 179 | } |
| 180 | |
| 181 | /* |
| 182 | * If the buffer in `snapshot` is active, then either munmap the |
| 183 | * memory and close the file, or free the memory. Then set the buffer |
| 184 | * pointers to NULL. |
| 185 | */ |
| 186 | static void clear_snapshot_buffer(struct snapshot *snapshot) |
| 187 | { |
| 188 | if (snapshot->mmapped) { |
| 189 | if (munmap(snapshot->buf, snapshot->eof - snapshot->buf)) |
| 190 | die_errno("error ummapping packed-refs file %s", |
| 191 | snapshot->refs->path); |
| 192 | snapshot->mmapped = 0; |
| 193 | } else { |
| 194 | free(snapshot->buf); |
| 195 | } |
| 196 | snapshot->buf = snapshot->start = snapshot->eof = NULL; |
| 197 | } |
| 198 | |
| 199 | /* |
| 200 | * Decrease the reference count of `*snapshot`. If it goes to zero, |
| 201 | * free `*snapshot` and return true; otherwise return false. |
| 202 | */ |
| 203 | static int release_snapshot(struct snapshot *snapshot) |
| 204 | { |
| 205 | if (!--snapshot->referrers) { |
| 206 | stat_validity_clear(&snapshot->validity); |
| 207 | clear_snapshot_buffer(snapshot); |
| 208 | free(snapshot); |
| 209 | return 1; |
| 210 | } else { |
| 211 | return 0; |
| 212 | } |
| 213 | } |
| 214 | |
| 215 | static size_t snapshot_hexsz(const struct snapshot *snapshot) |
| 216 | { |
| 217 | return snapshot->refs->base.repo->hash_algo->hexsz; |
| 218 | } |
| 219 | |
| 220 | static void packed_ref_store_reparent(const char *name UNUSED, |
| 221 | const char *old_cwd, |
| 222 | const char *new_cwd, |
| 223 | void *payload) |
| 224 | { |
| 225 | struct packed_ref_store *refs = payload; |
| 226 | char *tmp; |
| 227 | |
| 228 | tmp = reparent_relative_path(old_cwd, new_cwd, refs->path); |
| 229 | free(refs->path); |
| 230 | refs->path = tmp; |
| 231 | } |
| 232 | |
| 233 | /* |
| 234 | * Since packed-refs is only stored in the common dir, don't parse the |
| 235 | * payload and rely on the files-backend to set 'gitdir' correctly. |
| 236 | */ |
| 237 | struct ref_store *packed_ref_store_init(struct repository *repo, |
| 238 | const char *payload UNUSED, |
| 239 | const char *gitdir, |
| 240 | const struct ref_store_init_options *opts) |
| 241 | { |
| 242 | struct packed_ref_store *refs = xcalloc(1, sizeof(*refs)); |
| 243 | struct ref_store *ref_store = (struct ref_store *)refs; |
| 244 | struct strbuf sb = STRBUF_INIT; |
| 245 | |
| 246 | base_ref_store_init(ref_store, repo, gitdir, &refs_be_packed); |
| 247 | refs->store_flags = opts->access_flags; |
| 248 | |
| 249 | strbuf_addf(&sb, "%s/packed-refs", gitdir); |
| 250 | refs->path = strbuf_detach(&sb, NULL); |
| 251 | chdir_notify_register(NULL, packed_ref_store_reparent, refs); |
| 252 | return ref_store; |
| 253 | } |
| 254 | |
| 255 | /* |
| 256 | * Downcast `ref_store` to `packed_ref_store`. Die if `ref_store` is |
| 257 | * not a `packed_ref_store`. Also die if `packed_ref_store` doesn't |
| 258 | * support at least the flags specified in `required_flags`. `caller` |
| 259 | * is used in any necessary error messages. |
| 260 | */ |
| 261 | static struct packed_ref_store *packed_downcast(struct ref_store *ref_store, |
| 262 | unsigned int required_flags, |
| 263 | const char *caller) |
| 264 | { |
| 265 | struct packed_ref_store *refs; |
| 266 | |
| 267 | if (ref_store->be != &refs_be_packed) |
| 268 | BUG("ref_store is type \"%s\" not \"packed\" in %s", |
| 269 | ref_store->be->name, caller); |
| 270 | |
| 271 | refs = (struct packed_ref_store *)ref_store; |
| 272 | |
| 273 | if ((refs->store_flags & required_flags) != required_flags) |
| 274 | BUG("unallowed operation (%s), requires %x, has %x\n", |
| 275 | caller, required_flags, refs->store_flags); |
| 276 | |
| 277 | return refs; |
| 278 | } |
| 279 | |
| 280 | static void clear_snapshot(struct packed_ref_store *refs) |
| 281 | { |
| 282 | if (refs->snapshot) { |
| 283 | struct snapshot *snapshot = refs->snapshot; |
| 284 | |
| 285 | refs->snapshot = NULL; |
| 286 | release_snapshot(snapshot); |
| 287 | } |
| 288 | } |
| 289 | |
| 290 | static void packed_ref_store_release(struct ref_store *ref_store) |
| 291 | { |
| 292 | struct packed_ref_store *refs = packed_downcast(ref_store, 0, "release"); |
| 293 | clear_snapshot(refs); |
| 294 | rollback_lock_file(&refs->lock); |
| 295 | delete_tempfile(&refs->tempfile); |
| 296 | chdir_notify_unregister(NULL, packed_ref_store_reparent, refs); |
| 297 | free(refs->path); |
| 298 | } |
| 299 | |
| 300 | static NORETURN void die_unterminated_line(const char *path, |
| 301 | const char *p, size_t len) |
| 302 | { |
| 303 | if (len < 80) |
| 304 | die("unterminated line in %s: %.*s", path, (int)len, p); |
| 305 | else |
| 306 | die("unterminated line in %s: %.75s...", path, p); |
| 307 | } |
| 308 | |
| 309 | static NORETURN void die_invalid_line(const char *path, |
| 310 | const char *p, size_t len) |
| 311 | { |
| 312 | const char *eol = memchr(p, '\n', len); |
| 313 | |
| 314 | if (!eol) |
| 315 | die_unterminated_line(path, p, len); |
| 316 | else if (eol - p < 80) |
| 317 | die("unexpected line in %s: %.*s", path, (int)(eol - p), p); |
| 318 | else |
| 319 | die("unexpected line in %s: %.75s...", path, p); |
| 320 | |
| 321 | } |
| 322 | |
| 323 | struct snapshot_record { |
| 324 | const char *start; |
| 325 | size_t len; |
| 326 | }; |
| 327 | |
| 328 | |
| 329 | static int cmp_packed_refname(const char *r1, const char *r2) |
| 330 | { |
| 331 | while (1) { |
| 332 | if (*r1 == '\n') |
| 333 | return *r2 == '\n' ? 0 : -1; |
| 334 | if (*r1 != *r2) { |
| 335 | if (*r2 == '\n') |
| 336 | return 1; |
| 337 | else |
| 338 | return (unsigned char)*r1 < (unsigned char)*r2 ? -1 : +1; |
| 339 | } |
| 340 | r1++; |
| 341 | r2++; |
| 342 | } |
| 343 | } |
| 344 | |
| 345 | static int cmp_packed_ref_records(const void *v1, const void *v2, |
| 346 | void *cb_data) |
| 347 | { |
| 348 | const struct snapshot *snapshot = cb_data; |
| 349 | const struct snapshot_record *e1 = v1, *e2 = v2; |
| 350 | const char *r1 = e1->start + snapshot_hexsz(snapshot) + 1; |
| 351 | const char *r2 = e2->start + snapshot_hexsz(snapshot) + 1; |
| 352 | |
| 353 | return cmp_packed_refname(r1, r2); |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * Compare a snapshot record at `rec` to the specified NUL-terminated |
| 358 | * refname. |
| 359 | */ |
| 360 | static int cmp_record_to_refname(const char *rec, const char *refname, |
| 361 | int start, const struct snapshot *snapshot) |
| 362 | { |
| 363 | const char *r1 = rec + snapshot_hexsz(snapshot) + 1; |
| 364 | const char *r2 = refname; |
| 365 | |
| 366 | while (1) { |
| 367 | if (*r1 == '\n') |
| 368 | return *r2 ? -1 : 0; |
| 369 | if (!*r2) |
| 370 | return start ? 1 : -1; |
| 371 | if (*r1 != *r2) |
| 372 | return (unsigned char)*r1 < (unsigned char)*r2 ? -1 : +1; |
| 373 | r1++; |
| 374 | r2++; |
| 375 | } |
| 376 | } |
| 377 | |
| 378 | /* |
| 379 | * `snapshot->buf` is not known to be sorted. Check whether it is, and |
| 380 | * if not, sort it into new memory and munmap/free the old storage. |
| 381 | */ |
| 382 | static void sort_snapshot(struct snapshot *snapshot) |
| 383 | { |
| 384 | struct snapshot_record *records = NULL; |
| 385 | size_t alloc = 0, nr = 0; |
| 386 | int sorted = 1; |
| 387 | const char *pos, *eof, *eol; |
| 388 | size_t len, i; |
| 389 | char *new_buffer, *dst; |
| 390 | |
| 391 | pos = snapshot->start; |
| 392 | eof = snapshot->eof; |
| 393 | |
| 394 | if (pos == eof) |
| 395 | return; |
| 396 | |
| 397 | len = eof - pos; |
| 398 | |
| 399 | /* |
| 400 | * Initialize records based on a crude estimate of the number |
| 401 | * of references in the file (we'll grow it below if needed): |
| 402 | */ |
| 403 | ALLOC_GROW(records, len / 80 + 20, alloc); |
| 404 | |
| 405 | while (pos < eof) { |
| 406 | eol = memchr(pos, '\n', eof - pos); |
| 407 | if (!eol) |
| 408 | /* The safety check should prevent this. */ |
| 409 | BUG("unterminated line found in packed-refs"); |
| 410 | if (eol - pos < snapshot_hexsz(snapshot) + 2) |
| 411 | die_invalid_line(snapshot->refs->path, |
| 412 | pos, eof - pos); |
| 413 | eol++; |
| 414 | if (eol < eof && *eol == '^') { |
| 415 | /* |
| 416 | * Keep any peeled line together with its |
| 417 | * reference: |
| 418 | */ |
| 419 | const char *peeled_start = eol; |
| 420 | |
| 421 | eol = memchr(peeled_start, '\n', eof - peeled_start); |
| 422 | if (!eol) |
| 423 | /* The safety check should prevent this. */ |
| 424 | BUG("unterminated peeled line found in packed-refs"); |
| 425 | eol++; |
| 426 | } |
| 427 | |
| 428 | ALLOC_GROW(records, nr + 1, alloc); |
| 429 | records[nr].start = pos; |
| 430 | records[nr].len = eol - pos; |
| 431 | nr++; |
| 432 | |
| 433 | if (sorted && |
| 434 | nr > 1 && |
| 435 | cmp_packed_ref_records(&records[nr - 2], |
| 436 | &records[nr - 1], snapshot) >= 0) |
| 437 | sorted = 0; |
| 438 | |
| 439 | pos = eol; |
| 440 | } |
| 441 | |
| 442 | if (sorted) |
| 443 | goto cleanup; |
| 444 | |
| 445 | /* We need to sort the memory. First we sort the records array: */ |
| 446 | QSORT_S(records, nr, cmp_packed_ref_records, snapshot); |
| 447 | |
| 448 | /* |
| 449 | * Allocate a new chunk of memory, and copy the old memory to |
| 450 | * the new in the order indicated by `records` (not bothering |
| 451 | * with the header line): |
| 452 | */ |
| 453 | new_buffer = xmalloc(len); |
| 454 | for (dst = new_buffer, i = 0; i < nr; i++) { |
| 455 | memcpy(dst, records[i].start, records[i].len); |
| 456 | dst += records[i].len; |
| 457 | } |
| 458 | |
| 459 | /* |
| 460 | * Now munmap the old buffer and use the sorted buffer in its |
| 461 | * place: |
| 462 | */ |
| 463 | clear_snapshot_buffer(snapshot); |
| 464 | snapshot->buf = snapshot->start = new_buffer; |
| 465 | snapshot->eof = new_buffer + len; |
| 466 | |
| 467 | cleanup: |
| 468 | free(records); |
| 469 | } |
| 470 | |
| 471 | /* |
| 472 | * Return a pointer to the start of the record that contains the |
| 473 | * character `*p` (which must be within the buffer). If no other |
| 474 | * record start is found, return `buf`. |
| 475 | */ |
| 476 | static const char *find_start_of_record(const char *buf, const char *p) |
| 477 | { |
| 478 | while (p > buf && (p[-1] != '\n' || p[0] == '^')) |
| 479 | p--; |
| 480 | return p; |
| 481 | } |
| 482 | |
| 483 | /* |
| 484 | * Return a pointer to the start of the record following the record |
| 485 | * that contains `*p`. If none is found before `end`, return `end`. |
| 486 | */ |
| 487 | static const char *find_end_of_record(const char *p, const char *end) |
| 488 | { |
| 489 | while (++p < end && (p[-1] != '\n' || p[0] == '^')) |
| 490 | ; |
| 491 | return p; |
| 492 | } |
| 493 | |
| 494 | /* |
| 495 | * We want to be able to compare mmapped reference records quickly, |
| 496 | * without totally parsing them. We can do so because the records are |
| 497 | * LF-terminated, and the refname should start exactly (GIT_SHA1_HEXSZ |
| 498 | * + 1) bytes past the beginning of the record. |
| 499 | * |
| 500 | * But what if the `packed-refs` file contains garbage? We're willing |
| 501 | * to tolerate not detecting the problem, as long as we don't produce |
| 502 | * totally garbled output (we can't afford to check the integrity of |
| 503 | * the whole file during every Git invocation). But we do want to be |
| 504 | * sure that we never read past the end of the buffer in memory and |
| 505 | * perform an illegal memory access. |
| 506 | * |
| 507 | * Guarantee that minimum level of safety by verifying that the last |
| 508 | * record in the file is LF-terminated, and that it has at least |
| 509 | * (GIT_SHA1_HEXSZ + 1) characters before the LF. Die if either of |
| 510 | * these checks fails. |
| 511 | */ |
| 512 | static void verify_buffer_safe(struct snapshot *snapshot) |
| 513 | { |
| 514 | const char *start = snapshot->start; |
| 515 | const char *eof = snapshot->eof; |
| 516 | const char *last_line; |
| 517 | |
| 518 | if (start == eof) |
| 519 | return; |
| 520 | |
| 521 | last_line = find_start_of_record(start, eof - 1); |
| 522 | if (*(eof - 1) != '\n' || |
| 523 | eof - last_line < snapshot_hexsz(snapshot) + 2) |
| 524 | die_invalid_line(snapshot->refs->path, |
| 525 | last_line, eof - last_line); |
| 526 | } |
| 527 | |
| 528 | /* |
| 529 | * When parsing the "packed-refs" file, we will parse it line by line. |
| 530 | * Because we know the start pointer of the refname and the next |
| 531 | * newline pointer, we could calculate the length of the refname by |
| 532 | * subtracting the two pointers. However, there is a corner case where |
| 533 | * the refname contains corrupted embedded NUL characters. And |
| 534 | * `check_refname_format()` will not catch this when the truncated |
| 535 | * refname is still a valid refname. To prevent this, we need to check |
| 536 | * whether the refname contains the NUL characters. |
| 537 | */ |
| 538 | static int refname_contains_nul(struct strbuf *refname) |
| 539 | { |
| 540 | return !!memchr(refname->buf, '\0', refname->len); |
| 541 | } |
| 542 | |
| 543 | #define SMALL_FILE_SIZE (32*1024) |
| 544 | |
| 545 | static int allocate_snapshot_buffer(struct snapshot *snapshot, int fd, struct stat *st) |
| 546 | { |
| 547 | ssize_t bytes_read; |
| 548 | size_t size; |
| 549 | |
| 550 | size = xsize_t(st->st_size); |
| 551 | if (!size) |
| 552 | return 0; |
| 553 | |
| 554 | if (mmap_strategy == MMAP_NONE || size <= SMALL_FILE_SIZE) { |
| 555 | snapshot->buf = xmalloc(size); |
| 556 | bytes_read = read_in_full(fd, snapshot->buf, size); |
| 557 | if (bytes_read < 0 || bytes_read != size) |
| 558 | die_errno("couldn't read %s", snapshot->refs->path); |
| 559 | snapshot->mmapped = 0; |
| 560 | } else { |
| 561 | snapshot->buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0); |
| 562 | snapshot->mmapped = 1; |
| 563 | } |
| 564 | |
| 565 | snapshot->start = snapshot->buf; |
| 566 | snapshot->eof = snapshot->buf + size; |
| 567 | |
| 568 | return 1; |
| 569 | } |
| 570 | |
| 571 | /* |
| 572 | * Depending on `mmap_strategy`, either mmap or read the contents of |
| 573 | * the `packed-refs` file into the snapshot. Return 1 if the file |
| 574 | * existed and was read, or 0 if the file was absent or empty. Die on |
| 575 | * errors. |
| 576 | */ |
| 577 | static int load_contents(struct snapshot *snapshot) |
| 578 | { |
| 579 | struct stat st; |
| 580 | int ret; |
| 581 | int fd; |
| 582 | |
| 583 | fd = open(snapshot->refs->path, O_RDONLY); |
| 584 | if (fd < 0) { |
| 585 | if (errno == ENOENT) { |
| 586 | /* |
| 587 | * This is OK; it just means that no |
| 588 | * "packed-refs" file has been written yet, |
| 589 | * which is equivalent to it being empty, |
| 590 | * which is its state when initialized with |
| 591 | * zeros. |
| 592 | */ |
| 593 | return 0; |
| 594 | } else { |
| 595 | die_errno("couldn't read %s", snapshot->refs->path); |
| 596 | } |
| 597 | } |
| 598 | |
| 599 | stat_validity_update(&snapshot->validity, fd); |
| 600 | |
| 601 | if (fstat(fd, &st) < 0) |
| 602 | die_errno("couldn't stat %s", snapshot->refs->path); |
| 603 | |
| 604 | ret = allocate_snapshot_buffer(snapshot, fd, &st); |
| 605 | |
| 606 | close(fd); |
| 607 | return ret; |
| 608 | } |
| 609 | |
| 610 | static const char *find_reference_location_1(struct snapshot *snapshot, |
| 611 | const char *refname, int mustexist, |
| 612 | int start) |
| 613 | { |
| 614 | /* |
| 615 | * This is not *quite* a garden-variety binary search, because |
| 616 | * the data we're searching is made up of records, and we |
| 617 | * always need to find the beginning of a record to do a |
| 618 | * comparison. A "record" here is one line for the reference |
| 619 | * itself and zero or one peel lines that start with '^'. Our |
| 620 | * loop invariant is described in the next two comments. |
| 621 | */ |
| 622 | |
| 623 | /* |
| 624 | * A pointer to the character at the start of a record whose |
| 625 | * preceding records all have reference names that come |
| 626 | * *before* `refname`. |
| 627 | */ |
| 628 | const char *lo = snapshot->start; |
| 629 | |
| 630 | /* |
| 631 | * A pointer to a the first character of a record whose |
| 632 | * reference name comes *after* `refname`. |
| 633 | */ |
| 634 | const char *hi = snapshot->eof; |
| 635 | |
| 636 | while (lo != hi) { |
| 637 | const char *mid, *rec; |
| 638 | int cmp; |
| 639 | |
| 640 | mid = lo + (hi - lo) / 2; |
| 641 | rec = find_start_of_record(lo, mid); |
| 642 | cmp = cmp_record_to_refname(rec, refname, start, snapshot); |
| 643 | if (cmp < 0) { |
| 644 | lo = find_end_of_record(mid, hi); |
| 645 | } else if (cmp > 0) { |
| 646 | hi = rec; |
| 647 | } else { |
| 648 | return rec; |
| 649 | } |
| 650 | } |
| 651 | |
| 652 | if (mustexist) |
| 653 | return NULL; |
| 654 | else |
| 655 | return lo; |
| 656 | } |
| 657 | |
| 658 | /* |
| 659 | * Find the place in `snapshot->buf` where the start of the record for |
| 660 | * `refname` starts. If `mustexist` is true and the reference doesn't |
| 661 | * exist, then return NULL. If `mustexist` is false and the reference |
| 662 | * doesn't exist, then return the point where that reference would be |
| 663 | * inserted, or `snapshot->eof` (which might be NULL) if it would be |
| 664 | * inserted at the end of the file. In the latter mode, `refname` |
| 665 | * doesn't have to be a proper reference name; for example, one could |
| 666 | * search for "refs/replace/" to find the start of any replace |
| 667 | * references. |
| 668 | * |
| 669 | * The record is sought using a binary search, so `snapshot->buf` must |
| 670 | * be sorted. |
| 671 | */ |
| 672 | static const char *find_reference_location(struct snapshot *snapshot, |
| 673 | const char *refname, int mustexist) |
| 674 | { |
| 675 | return find_reference_location_1(snapshot, refname, mustexist, 1); |
| 676 | } |
| 677 | |
| 678 | /* |
| 679 | * Find the place in `snapshot->buf` after the end of the record for |
| 680 | * `refname`. In other words, find the location of first thing *after* |
| 681 | * `refname`. |
| 682 | * |
| 683 | * Other semantics are identical to the ones in |
| 684 | * `find_reference_location()`. |
| 685 | */ |
| 686 | static const char *find_reference_location_end(struct snapshot *snapshot, |
| 687 | const char *refname, |
| 688 | int mustexist) |
| 689 | { |
| 690 | return find_reference_location_1(snapshot, refname, mustexist, 0); |
| 691 | } |
| 692 | |
| 693 | /* |
| 694 | * Create a newly-allocated `snapshot` of the `packed-refs` file in |
| 695 | * its current state and return it. The return value will already have |
| 696 | * its reference count incremented. |
| 697 | * |
| 698 | * A comment line of the form "# pack-refs with: " may contain zero or |
| 699 | * more traits. We interpret the traits as follows: |
| 700 | * |
| 701 | * Neither `peeled` nor `fully-peeled`: |
| 702 | * |
| 703 | * Probably no references are peeled. But if the file contains a |
| 704 | * peeled value for a reference, we will use it. |
| 705 | * |
| 706 | * `peeled`: |
| 707 | * |
| 708 | * References under "refs/tags/", if they *can* be peeled, *are* |
| 709 | * peeled in this file. References outside of "refs/tags/" are |
| 710 | * probably not peeled even if they could have been, but if we find |
| 711 | * a peeled value for such a reference we will use it. |
| 712 | * |
| 713 | * `fully-peeled`: |
| 714 | * |
| 715 | * All references in the file that can be peeled are peeled. |
| 716 | * Inversely (and this is more important), any references in the |
| 717 | * file for which no peeled value is recorded is not peelable. This |
| 718 | * trait should typically be written alongside "peeled" for |
| 719 | * compatibility with older clients, but we do not require it |
| 720 | * (i.e., "peeled" is a no-op if "fully-peeled" is set). |
| 721 | * |
| 722 | * `sorted`: |
| 723 | * |
| 724 | * The references in this file are known to be sorted by refname. |
| 725 | */ |
| 726 | static struct snapshot *create_snapshot(struct packed_ref_store *refs) |
| 727 | { |
| 728 | struct snapshot *snapshot = xcalloc(1, sizeof(*snapshot)); |
| 729 | int sorted = 0; |
| 730 | |
| 731 | snapshot->refs = refs; |
| 732 | acquire_snapshot(snapshot); |
| 733 | snapshot->peeled = PEELED_NONE; |
| 734 | |
| 735 | if (!load_contents(snapshot)) |
| 736 | return snapshot; |
| 737 | |
| 738 | /* If the file has a header line, process it: */ |
| 739 | if (snapshot->buf < snapshot->eof && *snapshot->buf == '#') { |
| 740 | char *tmp, *p, *eol; |
| 741 | struct string_list traits = STRING_LIST_INIT_NODUP; |
| 742 | |
| 743 | eol = memchr(snapshot->buf, '\n', |
| 744 | snapshot->eof - snapshot->buf); |
| 745 | if (!eol) |
| 746 | die_unterminated_line(refs->path, |
| 747 | snapshot->buf, |
| 748 | snapshot->eof - snapshot->buf); |
| 749 | |
| 750 | tmp = xmemdupz(snapshot->buf, eol - snapshot->buf); |
| 751 | |
| 752 | if (!skip_prefix(tmp, "# pack-refs with: ", (const char **)&p)) |
| 753 | die_invalid_line(refs->path, |
| 754 | snapshot->buf, |
| 755 | snapshot->eof - snapshot->buf); |
| 756 | |
| 757 | string_list_split_in_place(&traits, p, " ", -1); |
| 758 | |
| 759 | if (unsorted_string_list_has_string(&traits, "fully-peeled")) |
| 760 | snapshot->peeled = PEELED_FULLY; |
| 761 | else if (unsorted_string_list_has_string(&traits, "peeled")) |
| 762 | snapshot->peeled = PEELED_TAGS; |
| 763 | |
| 764 | sorted = unsorted_string_list_has_string(&traits, "sorted"); |
| 765 | |
| 766 | /* perhaps other traits later as well */ |
| 767 | |
| 768 | /* The "+ 1" is for the LF character. */ |
| 769 | snapshot->start = eol + 1; |
| 770 | |
| 771 | string_list_clear(&traits, 0); |
| 772 | free(tmp); |
| 773 | } |
| 774 | |
| 775 | verify_buffer_safe(snapshot); |
| 776 | |
| 777 | if (!sorted) { |
| 778 | sort_snapshot(snapshot); |
| 779 | |
| 780 | /* |
| 781 | * Reordering the records might have moved a short one |
| 782 | * to the end of the buffer, so verify the buffer's |
| 783 | * safety again: |
| 784 | */ |
| 785 | verify_buffer_safe(snapshot); |
| 786 | } |
| 787 | |
| 788 | if (mmap_strategy != MMAP_OK && snapshot->mmapped) { |
| 789 | /* |
| 790 | * We don't want to leave the file mmapped, so we are |
| 791 | * forced to make a copy now: |
| 792 | */ |
| 793 | size_t size = snapshot->eof - snapshot->start; |
| 794 | char *buf_copy = xmalloc(size); |
| 795 | |
| 796 | memcpy(buf_copy, snapshot->start, size); |
| 797 | clear_snapshot_buffer(snapshot); |
| 798 | snapshot->buf = snapshot->start = buf_copy; |
| 799 | snapshot->eof = buf_copy + size; |
| 800 | } |
| 801 | |
| 802 | return snapshot; |
| 803 | } |
| 804 | |
| 805 | /* |
| 806 | * Check that `refs->snapshot` (if present) still reflects the |
| 807 | * contents of the `packed-refs` file. If not, clear the snapshot. |
| 808 | */ |
| 809 | static void validate_snapshot(struct packed_ref_store *refs) |
| 810 | { |
| 811 | if (refs->snapshot && |
| 812 | !stat_validity_check(&refs->snapshot->validity, refs->path)) |
| 813 | clear_snapshot(refs); |
| 814 | } |
| 815 | |
| 816 | /* |
| 817 | * Get the `snapshot` for the specified packed_ref_store, creating and |
| 818 | * populating it if it hasn't been read before or if the file has been |
| 819 | * changed (according to its `validity` field) since it was last read. |
| 820 | * On the other hand, if we hold the lock, then assume that the file |
| 821 | * hasn't been changed out from under us, so skip the extra `stat()` |
| 822 | * call in `stat_validity_check()`. This function does *not* increase |
| 823 | * the snapshot's reference count on behalf of the caller. |
| 824 | */ |
| 825 | static struct snapshot *get_snapshot(struct packed_ref_store *refs) |
| 826 | { |
| 827 | if (!is_lock_file_locked(&refs->lock)) |
| 828 | validate_snapshot(refs); |
| 829 | |
| 830 | if (!refs->snapshot) |
| 831 | refs->snapshot = create_snapshot(refs); |
| 832 | |
| 833 | return refs->snapshot; |
| 834 | } |
| 835 | |
| 836 | static int packed_read_raw_ref(struct ref_store *ref_store, const char *refname, |
| 837 | struct object_id *oid, struct strbuf *referent UNUSED, |
| 838 | unsigned int *type, int *failure_errno) |
| 839 | { |
| 840 | struct packed_ref_store *refs = |
| 841 | packed_downcast(ref_store, REF_STORE_READ, "read_raw_ref"); |
| 842 | struct snapshot *snapshot = get_snapshot(refs); |
| 843 | const char *rec; |
| 844 | |
| 845 | *type = 0; |
| 846 | |
| 847 | rec = find_reference_location(snapshot, refname, 1); |
| 848 | |
| 849 | if (!rec) { |
| 850 | /* refname is not a packed reference. */ |
| 851 | *failure_errno = ENOENT; |
| 852 | return -1; |
| 853 | } |
| 854 | |
| 855 | if (get_oid_hex_algop(rec, oid, ref_store->repo->hash_algo)) |
| 856 | die_invalid_line(refs->path, rec, snapshot->eof - rec); |
| 857 | |
| 858 | *type = REF_ISPACKED; |
| 859 | return 0; |
| 860 | } |
| 861 | |
| 862 | /* |
| 863 | * This value is set in `base.flags` if the peeled value of the |
| 864 | * current reference is known. In that case, `peeled` contains the |
| 865 | * correct peeled value for the reference, which might be `null_oid` |
| 866 | * if the reference is not a tag or if it is broken. |
| 867 | */ |
| 868 | #define REF_KNOWS_PEELED 0x40 |
| 869 | |
| 870 | /* |
| 871 | * An iterator over a snapshot of a `packed-refs` file. |
| 872 | */ |
| 873 | struct packed_ref_iterator { |
| 874 | struct ref_iterator base; |
| 875 | |
| 876 | struct snapshot *snapshot; |
| 877 | |
| 878 | char *prefix; |
| 879 | |
| 880 | /* The current position in the snapshot's buffer: */ |
| 881 | const char *pos; |
| 882 | |
| 883 | /* The end of the part of the buffer that will be iterated over: */ |
| 884 | const char *eof; |
| 885 | |
| 886 | struct jump_list_entry { |
| 887 | const char *start; |
| 888 | const char *end; |
| 889 | } *jump; |
| 890 | size_t jump_nr, jump_alloc; |
| 891 | size_t jump_cur; |
| 892 | |
| 893 | /* Scratch space for current values: */ |
| 894 | struct object_id oid, peeled; |
| 895 | struct strbuf refname_buf; |
| 896 | |
| 897 | struct repository *repo; |
| 898 | unsigned int flags; |
| 899 | }; |
| 900 | |
| 901 | /* |
| 902 | * Move the iterator to the next record in the snapshot. Adjust the fields in |
| 903 | * `iter` and return `ITER_OK` or `ITER_DONE`. This function does not free the |
| 904 | * iterator in the case of `ITER_DONE`. |
| 905 | */ |
| 906 | static int next_record(struct packed_ref_iterator *iter) |
| 907 | { |
| 908 | const char *p, *eol; |
| 909 | |
| 910 | memset(&iter->base.ref, 0, sizeof(iter->base.ref)); |
| 911 | strbuf_reset(&iter->refname_buf); |
| 912 | |
| 913 | /* |
| 914 | * If iter->pos is contained within a skipped region, jump past |
| 915 | * it. |
| 916 | * |
| 917 | * Note that each skipped region is considered at most once, |
| 918 | * since they are ordered based on their starting position. |
| 919 | */ |
| 920 | while (iter->jump_cur < iter->jump_nr) { |
| 921 | struct jump_list_entry *curr = &iter->jump[iter->jump_cur]; |
| 922 | if (iter->pos < curr->start) |
| 923 | break; /* not to the next jump yet */ |
| 924 | |
| 925 | iter->jump_cur++; |
| 926 | if (iter->pos < curr->end) { |
| 927 | iter->pos = curr->end; |
| 928 | trace2_counter_add(TRACE2_COUNTER_ID_PACKED_REFS_JUMPS, 1); |
| 929 | /* jumps are coalesced, so only one jump is necessary */ |
| 930 | break; |
| 931 | } |
| 932 | } |
| 933 | |
| 934 | if (iter->pos == iter->eof) |
| 935 | return ITER_DONE; |
| 936 | |
| 937 | iter->base.ref.flags = REF_ISPACKED; |
| 938 | p = iter->pos; |
| 939 | |
| 940 | if (iter->eof - p < snapshot_hexsz(iter->snapshot) + 2 || |
| 941 | parse_oid_hex_algop(p, &iter->oid, &p, iter->repo->hash_algo) || |
| 942 | !isspace(*p++)) |
| 943 | die_invalid_line(iter->snapshot->refs->path, |
| 944 | iter->pos, iter->eof - iter->pos); |
| 945 | iter->base.ref.oid = &iter->oid; |
| 946 | |
| 947 | eol = memchr(p, '\n', iter->eof - p); |
| 948 | if (!eol) |
| 949 | die_unterminated_line(iter->snapshot->refs->path, |
| 950 | iter->pos, iter->eof - iter->pos); |
| 951 | |
| 952 | strbuf_add(&iter->refname_buf, p, eol - p); |
| 953 | iter->base.ref.name = iter->refname_buf.buf; |
| 954 | |
| 955 | if (refname_contains_nul(&iter->refname_buf)) |
| 956 | die("packed refname contains embedded NULL: %s", iter->base.ref.name); |
| 957 | |
| 958 | if (check_refname_format(iter->base.ref.name, REFNAME_ALLOW_ONELEVEL)) { |
| 959 | if (!refname_is_safe(iter->base.ref.name)) |
| 960 | die("packed refname is dangerous: %s", |
| 961 | iter->base.ref.name); |
| 962 | oidclr(&iter->oid, iter->repo->hash_algo); |
| 963 | iter->base.ref.flags |= REF_BAD_NAME | REF_ISBROKEN; |
| 964 | } |
| 965 | if (iter->snapshot->peeled == PEELED_FULLY || |
| 966 | (iter->snapshot->peeled == PEELED_TAGS && |
| 967 | starts_with(iter->base.ref.name, "refs/tags/"))) |
| 968 | iter->base.ref.flags |= REF_KNOWS_PEELED; |
| 969 | |
| 970 | iter->pos = eol + 1; |
| 971 | |
| 972 | if (iter->pos < iter->eof && *iter->pos == '^') { |
| 973 | p = iter->pos + 1; |
| 974 | if (iter->eof - p < snapshot_hexsz(iter->snapshot) + 1 || |
| 975 | parse_oid_hex_algop(p, &iter->peeled, &p, iter->repo->hash_algo) || |
| 976 | *p++ != '\n') |
| 977 | die_invalid_line(iter->snapshot->refs->path, |
| 978 | iter->pos, iter->eof - iter->pos); |
| 979 | iter->pos = p; |
| 980 | |
| 981 | /* |
| 982 | * Regardless of what the file header said, we |
| 983 | * definitely know the value of *this* reference. But |
| 984 | * we suppress it if the reference is broken: |
| 985 | */ |
| 986 | if ((iter->base.ref.flags & REF_ISBROKEN)) { |
| 987 | oidclr(&iter->peeled, iter->repo->hash_algo); |
| 988 | iter->base.ref.flags &= ~REF_KNOWS_PEELED; |
| 989 | } else { |
| 990 | iter->base.ref.flags |= REF_KNOWS_PEELED; |
| 991 | iter->base.ref.peeled_oid = &iter->peeled; |
| 992 | } |
| 993 | } else { |
| 994 | oidclr(&iter->peeled, iter->repo->hash_algo); |
| 995 | } |
| 996 | |
| 997 | return ITER_OK; |
| 998 | } |
| 999 | |
| 1000 | static int packed_ref_iterator_advance(struct ref_iterator *ref_iterator) |
| 1001 | { |
| 1002 | struct packed_ref_iterator *iter = |
| 1003 | (struct packed_ref_iterator *)ref_iterator; |
| 1004 | int ok; |
| 1005 | |
| 1006 | while ((ok = next_record(iter)) == ITER_OK) { |
| 1007 | const char *refname = iter->base.ref.name; |
| 1008 | const char *prefix = iter->prefix; |
| 1009 | |
| 1010 | if (iter->flags & REFS_FOR_EACH_PER_WORKTREE_ONLY && |
| 1011 | !is_per_worktree_ref(iter->base.ref.name)) |
| 1012 | continue; |
| 1013 | |
| 1014 | if (!(iter->flags & REFS_FOR_EACH_INCLUDE_BROKEN) && |
| 1015 | !ref_resolves_to_object(iter->base.ref.name, iter->repo, |
| 1016 | &iter->oid, iter->flags)) |
| 1017 | continue; |
| 1018 | |
| 1019 | while (prefix && *prefix) { |
| 1020 | if ((unsigned char)*refname < (unsigned char)*prefix) |
| 1021 | BUG("packed-refs backend yielded reference preceding its prefix"); |
| 1022 | else if ((unsigned char)*refname > (unsigned char)*prefix) |
| 1023 | return ITER_DONE; |
| 1024 | prefix++; |
| 1025 | refname++; |
| 1026 | } |
| 1027 | |
| 1028 | return ITER_OK; |
| 1029 | } |
| 1030 | |
| 1031 | return ok; |
| 1032 | } |
| 1033 | |
| 1034 | static int packed_ref_iterator_seek(struct ref_iterator *ref_iterator, |
| 1035 | const char *refname, unsigned int flags) |
| 1036 | { |
| 1037 | struct packed_ref_iterator *iter = |
| 1038 | (struct packed_ref_iterator *)ref_iterator; |
| 1039 | const char *start; |
| 1040 | |
| 1041 | if (refname && *refname) |
| 1042 | start = find_reference_location(iter->snapshot, refname, 0); |
| 1043 | else |
| 1044 | start = iter->snapshot->start; |
| 1045 | |
| 1046 | /* Unset any previously set prefix */ |
| 1047 | FREE_AND_NULL(iter->prefix); |
| 1048 | |
| 1049 | if (flags & REF_ITERATOR_SEEK_SET_PREFIX) |
| 1050 | iter->prefix = xstrdup_or_null(refname); |
| 1051 | |
| 1052 | iter->pos = start; |
| 1053 | iter->eof = iter->snapshot->eof; |
| 1054 | |
| 1055 | return 0; |
| 1056 | } |
| 1057 | |
| 1058 | static void packed_ref_iterator_release(struct ref_iterator *ref_iterator) |
| 1059 | { |
| 1060 | struct packed_ref_iterator *iter = |
| 1061 | (struct packed_ref_iterator *)ref_iterator; |
| 1062 | strbuf_release(&iter->refname_buf); |
| 1063 | free(iter->jump); |
| 1064 | free(iter->prefix); |
| 1065 | release_snapshot(iter->snapshot); |
| 1066 | } |
| 1067 | |
| 1068 | static struct ref_iterator_vtable packed_ref_iterator_vtable = { |
| 1069 | .advance = packed_ref_iterator_advance, |
| 1070 | .seek = packed_ref_iterator_seek, |
| 1071 | .release = packed_ref_iterator_release, |
| 1072 | }; |
| 1073 | |
| 1074 | static int jump_list_entry_cmp(const void *va, const void *vb) |
| 1075 | { |
| 1076 | const struct jump_list_entry *a = va; |
| 1077 | const struct jump_list_entry *b = vb; |
| 1078 | |
| 1079 | if (a->start < b->start) |
| 1080 | return -1; |
| 1081 | if (a->start > b->start) |
| 1082 | return 1; |
| 1083 | return 0; |
| 1084 | } |
| 1085 | |
| 1086 | static int has_glob_special(const char *str) |
| 1087 | { |
| 1088 | const char *p; |
| 1089 | for (p = str; *p; p++) { |
| 1090 | if (is_glob_special(*p)) |
| 1091 | return 1; |
| 1092 | } |
| 1093 | return 0; |
| 1094 | } |
| 1095 | |
| 1096 | static void populate_excluded_jump_list(struct packed_ref_iterator *iter, |
| 1097 | struct snapshot *snapshot, |
| 1098 | const char **excluded_patterns) |
| 1099 | { |
| 1100 | size_t i, j; |
| 1101 | const char **pattern; |
| 1102 | struct jump_list_entry *last_disjoint; |
| 1103 | |
| 1104 | if (!excluded_patterns) |
| 1105 | return; |
| 1106 | |
| 1107 | for (pattern = excluded_patterns; *pattern; pattern++) { |
| 1108 | struct jump_list_entry *e; |
| 1109 | const char *start, *end; |
| 1110 | |
| 1111 | /* |
| 1112 | * We can't feed any excludes with globs in them to the |
| 1113 | * refs machinery. It only understands prefix matching. |
| 1114 | * We likewise can't even feed the string leading up to |
| 1115 | * the first meta-character, as something like "foo[a]" |
| 1116 | * should not exclude "foobar" (but the prefix "foo" |
| 1117 | * would match that and mark it for exclusion). |
| 1118 | */ |
| 1119 | if (has_glob_special(*pattern)) |
| 1120 | continue; |
| 1121 | |
| 1122 | start = find_reference_location(snapshot, *pattern, 0); |
| 1123 | end = find_reference_location_end(snapshot, *pattern, 0); |
| 1124 | |
| 1125 | if (start == end) |
| 1126 | continue; /* nothing to jump over */ |
| 1127 | |
| 1128 | ALLOC_GROW(iter->jump, iter->jump_nr + 1, iter->jump_alloc); |
| 1129 | |
| 1130 | e = &iter->jump[iter->jump_nr++]; |
| 1131 | e->start = start; |
| 1132 | e->end = end; |
| 1133 | } |
| 1134 | |
| 1135 | if (!iter->jump_nr) { |
| 1136 | /* |
| 1137 | * Every entry in exclude_patterns has a meta-character, |
| 1138 | * nothing to do here. |
| 1139 | */ |
| 1140 | return; |
| 1141 | } |
| 1142 | |
| 1143 | QSORT(iter->jump, iter->jump_nr, jump_list_entry_cmp); |
| 1144 | |
| 1145 | /* |
| 1146 | * As an optimization, merge adjacent entries in the jump list |
| 1147 | * to jump forwards as far as possible when entering a skipped |
| 1148 | * region. |
| 1149 | * |
| 1150 | * For example, if we have two skipped regions: |
| 1151 | * |
| 1152 | * [[A, B], [B, C]] |
| 1153 | * |
| 1154 | * we want to combine that into a single entry jumping from A to |
| 1155 | * C. |
| 1156 | */ |
| 1157 | last_disjoint = iter->jump; |
| 1158 | |
| 1159 | for (i = 1, j = 1; i < iter->jump_nr; i++) { |
| 1160 | struct jump_list_entry *ours = &iter->jump[i]; |
| 1161 | if (ours->start <= last_disjoint->end) { |
| 1162 | /* overlapping regions extend the previous one */ |
| 1163 | last_disjoint->end = last_disjoint->end > ours->end |
| 1164 | ? last_disjoint->end : ours->end; |
| 1165 | } else { |
| 1166 | /* otherwise, insert a new region */ |
| 1167 | iter->jump[j++] = *ours; |
| 1168 | last_disjoint = ours; |
| 1169 | } |
| 1170 | } |
| 1171 | |
| 1172 | iter->jump_nr = j; |
| 1173 | iter->jump_cur = 0; |
| 1174 | } |
| 1175 | |
| 1176 | static struct ref_iterator *packed_ref_iterator_begin( |
| 1177 | struct ref_store *ref_store, |
| 1178 | const char *prefix, const char **exclude_patterns, |
| 1179 | unsigned int flags) |
| 1180 | { |
| 1181 | struct packed_ref_store *refs; |
| 1182 | struct snapshot *snapshot; |
| 1183 | struct packed_ref_iterator *iter; |
| 1184 | struct ref_iterator *ref_iterator; |
| 1185 | unsigned int required_flags = REF_STORE_READ; |
| 1186 | |
| 1187 | if (!(flags & REFS_FOR_EACH_INCLUDE_BROKEN)) |
| 1188 | required_flags |= REF_STORE_ODB; |
| 1189 | refs = packed_downcast(ref_store, required_flags, "ref_iterator_begin"); |
| 1190 | |
| 1191 | /* |
| 1192 | * Note that `get_snapshot()` internally checks whether the |
| 1193 | * snapshot is up to date with what is on disk, and re-reads |
| 1194 | * it if not. |
| 1195 | */ |
| 1196 | snapshot = get_snapshot(refs); |
| 1197 | |
| 1198 | CALLOC_ARRAY(iter, 1); |
| 1199 | ref_iterator = &iter->base; |
| 1200 | base_ref_iterator_init(ref_iterator, &packed_ref_iterator_vtable); |
| 1201 | |
| 1202 | if (exclude_patterns) |
| 1203 | populate_excluded_jump_list(iter, snapshot, exclude_patterns); |
| 1204 | |
| 1205 | iter->snapshot = snapshot; |
| 1206 | acquire_snapshot(snapshot); |
| 1207 | strbuf_init(&iter->refname_buf, 0); |
| 1208 | iter->repo = ref_store->repo; |
| 1209 | iter->flags = flags; |
| 1210 | |
| 1211 | if (packed_ref_iterator_seek(&iter->base, prefix, |
| 1212 | REF_ITERATOR_SEEK_SET_PREFIX) < 0) { |
| 1213 | ref_iterator_free(&iter->base); |
| 1214 | return NULL; |
| 1215 | } |
| 1216 | |
| 1217 | return ref_iterator; |
| 1218 | } |
| 1219 | |
| 1220 | /* |
| 1221 | * Write an entry to the packed-refs file for the specified refname. |
| 1222 | * If peeled is non-NULL, write it as the entry's peeled value. On |
| 1223 | * error, return a nonzero value and leave errno set at the value left |
| 1224 | * by the failing call to `fprintf()`. |
| 1225 | */ |
| 1226 | static int write_packed_entry(FILE *fh, const char *refname, |
| 1227 | const struct object_id *oid, |
| 1228 | const struct object_id *peeled) |
| 1229 | { |
| 1230 | if (fprintf(fh, "%s %s\n", oid_to_hex(oid), refname) < 0 || |
| 1231 | (peeled && fprintf(fh, "^%s\n", oid_to_hex(peeled)) < 0)) |
| 1232 | return -1; |
| 1233 | |
| 1234 | return 0; |
| 1235 | } |
| 1236 | |
| 1237 | int packed_refs_lock(struct ref_store *ref_store, int flags, struct strbuf *err) |
| 1238 | { |
| 1239 | struct packed_ref_store *refs = |
| 1240 | packed_downcast(ref_store, REF_STORE_WRITE | REF_STORE_MAIN, |
| 1241 | "packed_refs_lock"); |
| 1242 | |
| 1243 | if (!refs->timeout_configured) { |
| 1244 | if (repo_config_get_int(ref_store->repo, "core.packedrefstimeout", |
| 1245 | &refs->timeout_value)) |
| 1246 | refs->timeout_value = 1000; |
| 1247 | refs->timeout_configured = true; |
| 1248 | } |
| 1249 | |
| 1250 | /* |
| 1251 | * Note that we close the lockfile immediately because we |
| 1252 | * don't write new content to it, but rather to a separate |
| 1253 | * tempfile. |
| 1254 | */ |
| 1255 | if (repo_hold_lock_file_for_update_timeout(ref_store->repo, &refs->lock, |
| 1256 | refs->path, flags, |
| 1257 | refs->timeout_value) < 0) { |
| 1258 | unable_to_lock_message(refs->path, errno, err); |
| 1259 | return -1; |
| 1260 | } |
| 1261 | |
| 1262 | if (close_lock_file_gently(&refs->lock)) { |
| 1263 | strbuf_addf(err, "unable to close %s: %s", refs->path, strerror(errno)); |
| 1264 | rollback_lock_file(&refs->lock); |
| 1265 | return -1; |
| 1266 | } |
| 1267 | |
| 1268 | /* |
| 1269 | * There is a stat-validity problem might cause `update-ref -d` |
| 1270 | * lost the newly commit of a ref, because a new `packed-refs` |
| 1271 | * file might has the same on-disk file attributes such as |
| 1272 | * timestamp, file size and inode value, but has a changed |
| 1273 | * ref value. |
| 1274 | * |
| 1275 | * This could happen with a very small chance when |
| 1276 | * `update-ref -d` is called and at the same time another |
| 1277 | * `pack-refs --all` process is running. |
| 1278 | * |
| 1279 | * Now that we hold the `packed-refs` lock, it is important |
| 1280 | * to make sure we could read the latest version of |
| 1281 | * `packed-refs` file no matter we have just mmap it or not. |
| 1282 | * So what need to do is clear the snapshot if we hold it |
| 1283 | * already. |
| 1284 | */ |
| 1285 | clear_snapshot(refs); |
| 1286 | |
| 1287 | /* |
| 1288 | * Now make sure that the packed-refs file as it exists in the |
| 1289 | * locked state is loaded into the snapshot: |
| 1290 | */ |
| 1291 | get_snapshot(refs); |
| 1292 | return 0; |
| 1293 | } |
| 1294 | |
| 1295 | void packed_refs_unlock(struct ref_store *ref_store) |
| 1296 | { |
| 1297 | struct packed_ref_store *refs = packed_downcast( |
| 1298 | ref_store, |
| 1299 | REF_STORE_READ | REF_STORE_WRITE, |
| 1300 | "packed_refs_unlock"); |
| 1301 | |
| 1302 | if (!is_lock_file_locked(&refs->lock)) |
| 1303 | BUG("packed_refs_unlock() called when not locked"); |
| 1304 | rollback_lock_file(&refs->lock); |
| 1305 | } |
| 1306 | |
| 1307 | int packed_refs_is_locked(struct ref_store *ref_store) |
| 1308 | { |
| 1309 | struct packed_ref_store *refs = packed_downcast( |
| 1310 | ref_store, |
| 1311 | REF_STORE_READ | REF_STORE_WRITE, |
| 1312 | "packed_refs_is_locked"); |
| 1313 | |
| 1314 | return is_lock_file_locked(&refs->lock); |
| 1315 | } |
| 1316 | |
| 1317 | int packed_refs_size(struct ref_store *ref_store, |
| 1318 | size_t *out) |
| 1319 | { |
| 1320 | struct packed_ref_store *refs = packed_downcast(ref_store, REF_STORE_READ, |
| 1321 | "packed_refs_size"); |
| 1322 | struct stat st; |
| 1323 | |
| 1324 | if (stat(refs->path, &st) < 0) { |
| 1325 | if (errno != ENOENT) |
| 1326 | return -1; |
| 1327 | *out = 0; |
| 1328 | return 0; |
| 1329 | } |
| 1330 | |
| 1331 | *out = st.st_size; |
| 1332 | return 0; |
| 1333 | } |
| 1334 | |
| 1335 | /* |
| 1336 | * The packed-refs header line that we write out. Perhaps other traits |
| 1337 | * will be added later. |
| 1338 | * |
| 1339 | * Note that earlier versions of Git used to parse these traits by |
| 1340 | * looking for " trait " in the line. For this reason, the space after |
| 1341 | * the colon and the trailing space are required. |
| 1342 | */ |
| 1343 | static const char PACKED_REFS_HEADER[] = |
| 1344 | "# pack-refs with: peeled fully-peeled sorted \n"; |
| 1345 | |
| 1346 | static int packed_ref_store_create_on_disk(struct ref_store *ref_store UNUSED, |
| 1347 | int flags UNUSED, |
| 1348 | struct strbuf *err UNUSED) |
| 1349 | { |
| 1350 | /* Nothing to do. */ |
| 1351 | return 0; |
| 1352 | } |
| 1353 | |
| 1354 | static int packed_ref_store_remove_on_disk(struct ref_store *ref_store, |
| 1355 | struct strbuf *err) |
| 1356 | { |
| 1357 | struct packed_ref_store *refs = packed_downcast(ref_store, 0, "remove"); |
| 1358 | |
| 1359 | if (remove_path(refs->path) < 0) { |
| 1360 | strbuf_addstr(err, "could not delete packed-refs"); |
| 1361 | return -1; |
| 1362 | } |
| 1363 | |
| 1364 | return 0; |
| 1365 | } |
| 1366 | |
| 1367 | /* |
| 1368 | * Write the packed refs from the current snapshot to the packed-refs |
| 1369 | * tempfile, incorporating any changes from `updates`. `updates` must |
| 1370 | * be a sorted string list whose keys are the refnames and whose util |
| 1371 | * values are `struct ref_update *`. On error, rollback the tempfile, |
| 1372 | * write an error message to `err`, and return a nonzero value. |
| 1373 | * |
| 1374 | * The packfile must be locked before calling this function and will |
| 1375 | * remain locked when it is done. |
| 1376 | */ |
| 1377 | static enum ref_transaction_error write_with_updates(struct packed_ref_store *refs, |
| 1378 | struct ref_transaction *transaction, |
| 1379 | struct strbuf *err) |
| 1380 | { |
| 1381 | enum ref_transaction_error ret = REF_TRANSACTION_ERROR_GENERIC; |
| 1382 | struct string_list *updates = &transaction->refnames; |
| 1383 | struct ref_iterator *iter = NULL; |
| 1384 | size_t i; |
| 1385 | int ok; |
| 1386 | FILE *out; |
| 1387 | struct strbuf sb = STRBUF_INIT; |
| 1388 | char *packed_refs_path; |
| 1389 | |
| 1390 | if (!is_lock_file_locked(&refs->lock)) |
| 1391 | BUG("write_with_updates() called while unlocked"); |
| 1392 | |
| 1393 | /* |
| 1394 | * If packed-refs is a symlink, we want to overwrite the |
| 1395 | * symlinked-to file, not the symlink itself. Also, put the |
| 1396 | * staging file next to it: |
| 1397 | */ |
| 1398 | packed_refs_path = get_locked_file_path(&refs->lock); |
| 1399 | strbuf_addf(&sb, "%s.new", packed_refs_path); |
| 1400 | free(packed_refs_path); |
| 1401 | refs->tempfile = repo_create_tempfile(refs->base.repo, sb.buf); |
| 1402 | if (!refs->tempfile) { |
| 1403 | strbuf_addf(err, "unable to create file %s: %s", |
| 1404 | sb.buf, strerror(errno)); |
| 1405 | strbuf_release(&sb); |
| 1406 | return REF_TRANSACTION_ERROR_GENERIC; |
| 1407 | } |
| 1408 | strbuf_release(&sb); |
| 1409 | |
| 1410 | out = fdopen_tempfile(refs->tempfile, "w"); |
| 1411 | if (!out) { |
| 1412 | strbuf_addf(err, "unable to fdopen packed-refs tempfile: %s", |
| 1413 | strerror(errno)); |
| 1414 | goto error; |
| 1415 | } |
| 1416 | |
| 1417 | if (fprintf(out, "%s", PACKED_REFS_HEADER) < 0) |
| 1418 | goto write_error; |
| 1419 | |
| 1420 | /* |
| 1421 | * We iterate in parallel through the current list of refs and |
| 1422 | * the list of updates, processing an entry from at least one |
| 1423 | * of the lists each time through the loop. When the current |
| 1424 | * list of refs is exhausted, set iter to NULL. When the list |
| 1425 | * of updates is exhausted, leave i set to updates->nr. |
| 1426 | */ |
| 1427 | iter = packed_ref_iterator_begin(&refs->base, "", NULL, |
| 1428 | REFS_FOR_EACH_INCLUDE_BROKEN); |
| 1429 | if ((ok = ref_iterator_advance(iter)) != ITER_OK) { |
| 1430 | ref_iterator_free(iter); |
| 1431 | iter = NULL; |
| 1432 | } |
| 1433 | |
| 1434 | i = 0; |
| 1435 | |
| 1436 | while (iter || i < updates->nr) { |
| 1437 | struct ref_update *update = NULL; |
| 1438 | int cmp; |
| 1439 | |
| 1440 | if (i >= updates->nr) { |
| 1441 | cmp = -1; |
| 1442 | } else { |
| 1443 | update = updates->items[i].util; |
| 1444 | |
| 1445 | if (!iter) |
| 1446 | cmp = +1; |
| 1447 | else |
| 1448 | cmp = strcmp(iter->ref.name, update->refname); |
| 1449 | } |
| 1450 | |
| 1451 | if (!cmp) { |
| 1452 | /* |
| 1453 | * There is both an old value and an update |
| 1454 | * for this reference. Check the old value if |
| 1455 | * necessary: |
| 1456 | */ |
| 1457 | if ((update->flags & REF_HAVE_OLD)) { |
| 1458 | if (is_null_oid(&update->old_oid)) { |
| 1459 | strbuf_addf(err, "cannot update ref '%s': " |
| 1460 | "reference already exists", |
| 1461 | update->refname); |
| 1462 | ret = REF_TRANSACTION_ERROR_CREATE_EXISTS; |
| 1463 | |
| 1464 | if (ref_transaction_maybe_set_rejected(transaction, i, |
| 1465 | ret, err)) { |
| 1466 | ret = 0; |
| 1467 | continue; |
| 1468 | } |
| 1469 | |
| 1470 | goto error; |
| 1471 | } else if (!oideq(&update->old_oid, iter->ref.oid)) { |
| 1472 | strbuf_addf(err, "cannot update ref '%s': " |
| 1473 | "is at %s but expected %s", |
| 1474 | update->refname, |
| 1475 | oid_to_hex(iter->ref.oid), |
| 1476 | oid_to_hex(&update->old_oid)); |
| 1477 | ret = REF_TRANSACTION_ERROR_INCORRECT_OLD_VALUE; |
| 1478 | |
| 1479 | if (ref_transaction_maybe_set_rejected(transaction, i, |
| 1480 | ret, err)) { |
| 1481 | ret = 0; |
| 1482 | continue; |
| 1483 | } |
| 1484 | |
| 1485 | goto error; |
| 1486 | } |
| 1487 | } |
| 1488 | |
| 1489 | /* Now figure out what to use for the new value: */ |
| 1490 | if ((update->flags & REF_HAVE_NEW)) { |
| 1491 | /* |
| 1492 | * The update takes precedence. Skip |
| 1493 | * the iterator over the unneeded |
| 1494 | * value. |
| 1495 | */ |
| 1496 | if ((ok = ref_iterator_advance(iter)) != ITER_OK) { |
| 1497 | ref_iterator_free(iter); |
| 1498 | iter = NULL; |
| 1499 | } |
| 1500 | cmp = +1; |
| 1501 | } else { |
| 1502 | /* |
| 1503 | * The update doesn't actually want to |
| 1504 | * change anything. We're done with it. |
| 1505 | */ |
| 1506 | i++; |
| 1507 | cmp = -1; |
| 1508 | } |
| 1509 | } else if (cmp > 0) { |
| 1510 | /* |
| 1511 | * There is no old value but there is an |
| 1512 | * update for this reference. Make sure that |
| 1513 | * the update didn't expect an existing value: |
| 1514 | */ |
| 1515 | if ((update->flags & REF_HAVE_OLD) && |
| 1516 | !is_null_oid(&update->old_oid)) { |
| 1517 | strbuf_addf(err, "cannot update ref '%s': " |
| 1518 | "reference is missing but expected %s", |
| 1519 | update->refname, |
| 1520 | oid_to_hex(&update->old_oid)); |
| 1521 | ret = REF_TRANSACTION_ERROR_NONEXISTENT_REF; |
| 1522 | |
| 1523 | if (ref_transaction_maybe_set_rejected(transaction, i, |
| 1524 | ret, err)) { |
| 1525 | ret = 0; |
| 1526 | continue; |
| 1527 | } |
| 1528 | |
| 1529 | goto error; |
| 1530 | } |
| 1531 | } |
| 1532 | |
| 1533 | if (cmp < 0) { |
| 1534 | /* Pass the old reference through. */ |
| 1535 | if (write_packed_entry(out, iter->ref.name, |
| 1536 | iter->ref.oid, iter->ref.peeled_oid)) |
| 1537 | goto write_error; |
| 1538 | |
| 1539 | if ((ok = ref_iterator_advance(iter)) != ITER_OK) { |
| 1540 | ref_iterator_free(iter); |
| 1541 | iter = NULL; |
| 1542 | } |
| 1543 | } else if (is_null_oid(&update->new_oid)) { |
| 1544 | /* |
| 1545 | * The update wants to delete the reference, |
| 1546 | * and the reference either didn't exist or we |
| 1547 | * have already skipped it. So we're done with |
| 1548 | * the update (and don't have to write |
| 1549 | * anything). |
| 1550 | */ |
| 1551 | i++; |
| 1552 | } else { |
| 1553 | bool peeled = update->flags & REF_HAVE_PEELED; |
| 1554 | |
| 1555 | if (write_packed_entry(out, update->refname, |
| 1556 | &update->new_oid, |
| 1557 | peeled ? &update->peeled : NULL)) |
| 1558 | goto write_error; |
| 1559 | |
| 1560 | i++; |
| 1561 | } |
| 1562 | } |
| 1563 | |
| 1564 | if (ok != ITER_DONE) { |
| 1565 | strbuf_addstr(err, "unable to write packed-refs file: " |
| 1566 | "error iterating over old contents"); |
| 1567 | goto error; |
| 1568 | } |
| 1569 | |
| 1570 | if (fflush(out) || |
| 1571 | fsync_component(FSYNC_COMPONENT_REFERENCE, get_tempfile_fd(refs->tempfile)) || |
| 1572 | close_tempfile_gently(refs->tempfile)) { |
| 1573 | strbuf_addf(err, "error closing file %s: %s", |
| 1574 | get_tempfile_path(refs->tempfile), |
| 1575 | strerror(errno)); |
| 1576 | strbuf_release(&sb); |
| 1577 | delete_tempfile(&refs->tempfile); |
| 1578 | return REF_TRANSACTION_ERROR_GENERIC; |
| 1579 | } |
| 1580 | |
| 1581 | return 0; |
| 1582 | |
| 1583 | write_error: |
| 1584 | strbuf_addf(err, "error writing to %s: %s", |
| 1585 | get_tempfile_path(refs->tempfile), strerror(errno)); |
| 1586 | ret = REF_TRANSACTION_ERROR_GENERIC; |
| 1587 | |
| 1588 | error: |
| 1589 | ref_iterator_free(iter); |
| 1590 | delete_tempfile(&refs->tempfile); |
| 1591 | return ret; |
| 1592 | } |
| 1593 | |
| 1594 | int is_packed_transaction_needed(struct ref_store *ref_store, |
| 1595 | struct ref_transaction *transaction) |
| 1596 | { |
| 1597 | struct packed_ref_store *refs = packed_downcast( |
| 1598 | ref_store, |
| 1599 | REF_STORE_READ, |
| 1600 | "is_packed_transaction_needed"); |
| 1601 | struct strbuf referent = STRBUF_INIT; |
| 1602 | size_t i; |
| 1603 | int ret; |
| 1604 | |
| 1605 | if (!is_lock_file_locked(&refs->lock)) |
| 1606 | BUG("is_packed_transaction_needed() called while unlocked"); |
| 1607 | |
| 1608 | /* |
| 1609 | * We're only going to bother returning false for the common, |
| 1610 | * trivial case that references are only being deleted, their |
| 1611 | * old values are not being checked, and the old `packed-refs` |
| 1612 | * file doesn't contain any of those reference(s). This gives |
| 1613 | * false positives for some other cases that could |
| 1614 | * theoretically be optimized away: |
| 1615 | * |
| 1616 | * 1. It could be that the old value is being verified without |
| 1617 | * setting a new value. In this case, we could verify the |
| 1618 | * old value here and skip the update if it agrees. If it |
| 1619 | * disagrees, we could either let the update go through |
| 1620 | * (the actual commit would re-detect and report the |
| 1621 | * problem), or come up with a way of reporting such an |
| 1622 | * error to *our* caller. |
| 1623 | * |
| 1624 | * 2. It could be that a new value is being set, but that it |
| 1625 | * is identical to the current packed value of the |
| 1626 | * reference. |
| 1627 | * |
| 1628 | * Neither of these cases will come up in the current code, |
| 1629 | * because the only caller of this function passes to it a |
| 1630 | * transaction that only includes `delete` updates with no |
| 1631 | * `old_id`. Even if that ever changes, false positives only |
| 1632 | * cause an optimization to be missed; they do not affect |
| 1633 | * correctness. |
| 1634 | */ |
| 1635 | |
| 1636 | /* |
| 1637 | * Start with the cheap checks that don't require old |
| 1638 | * reference values to be read: |
| 1639 | */ |
| 1640 | for (i = 0; i < transaction->nr; i++) { |
| 1641 | struct ref_update *update = transaction->updates[i]; |
| 1642 | |
| 1643 | if (update->flags & REF_HAVE_OLD) |
| 1644 | /* Have to check the old value -> needed. */ |
| 1645 | return 1; |
| 1646 | |
| 1647 | if ((update->flags & REF_HAVE_NEW) && !is_null_oid(&update->new_oid)) |
| 1648 | /* Have to set a new value -> needed. */ |
| 1649 | return 1; |
| 1650 | } |
| 1651 | |
| 1652 | /* |
| 1653 | * The transaction isn't checking any old values nor is it |
| 1654 | * setting any nonzero new values, so it still might be able |
| 1655 | * to be skipped. Now do the more expensive check: the update |
| 1656 | * is needed if any of the updates is a delete, and the old |
| 1657 | * `packed-refs` file contains a value for that reference. |
| 1658 | */ |
| 1659 | ret = 0; |
| 1660 | for (i = 0; i < transaction->nr; i++) { |
| 1661 | struct ref_update *update = transaction->updates[i]; |
| 1662 | int failure_errno; |
| 1663 | unsigned int type; |
| 1664 | struct object_id oid; |
| 1665 | |
| 1666 | if (!(update->flags & REF_HAVE_NEW)) |
| 1667 | /* |
| 1668 | * This reference isn't being deleted -> not |
| 1669 | * needed. |
| 1670 | */ |
| 1671 | continue; |
| 1672 | |
| 1673 | if (!refs_read_raw_ref(ref_store, update->refname, &oid, |
| 1674 | &referent, &type, &failure_errno) || |
| 1675 | failure_errno != ENOENT) { |
| 1676 | /* |
| 1677 | * We have to actually delete that reference |
| 1678 | * -> this transaction is needed. |
| 1679 | */ |
| 1680 | ret = 1; |
| 1681 | break; |
| 1682 | } |
| 1683 | } |
| 1684 | |
| 1685 | strbuf_release(&referent); |
| 1686 | return ret; |
| 1687 | } |
| 1688 | |
| 1689 | struct packed_transaction_backend_data { |
| 1690 | /* True iff the transaction owns the packed-refs lock. */ |
| 1691 | int own_lock; |
| 1692 | }; |
| 1693 | |
| 1694 | static void packed_transaction_cleanup(struct packed_ref_store *refs, |
| 1695 | struct ref_transaction *transaction) |
| 1696 | { |
| 1697 | struct packed_transaction_backend_data *data = transaction->backend_data; |
| 1698 | |
| 1699 | if (data) { |
| 1700 | if (is_tempfile_active(refs->tempfile)) |
| 1701 | delete_tempfile(&refs->tempfile); |
| 1702 | |
| 1703 | if (data->own_lock && is_lock_file_locked(&refs->lock)) { |
| 1704 | packed_refs_unlock(&refs->base); |
| 1705 | data->own_lock = 0; |
| 1706 | } |
| 1707 | |
| 1708 | free(data); |
| 1709 | transaction->backend_data = NULL; |
| 1710 | } |
| 1711 | |
| 1712 | transaction->state = REF_TRANSACTION_CLOSED; |
| 1713 | } |
| 1714 | |
| 1715 | static int packed_transaction_prepare(struct ref_store *ref_store, |
| 1716 | struct ref_transaction *transaction, |
| 1717 | struct strbuf *err) |
| 1718 | { |
| 1719 | struct packed_ref_store *refs = packed_downcast( |
| 1720 | ref_store, |
| 1721 | REF_STORE_READ | REF_STORE_WRITE | REF_STORE_ODB, |
| 1722 | "ref_transaction_prepare"); |
| 1723 | struct packed_transaction_backend_data *data; |
| 1724 | enum ref_transaction_error ret = REF_TRANSACTION_ERROR_GENERIC; |
| 1725 | |
| 1726 | /* |
| 1727 | * Note that we *don't* skip transactions with zero updates, |
| 1728 | * because such a transaction might be executed for the side |
| 1729 | * effect of ensuring that all of the references are peeled or |
| 1730 | * ensuring that the `packed-refs` file is sorted. If the |
| 1731 | * caller wants to optimize away empty transactions, it should |
| 1732 | * do so itself. |
| 1733 | */ |
| 1734 | |
| 1735 | CALLOC_ARRAY(data, 1); |
| 1736 | |
| 1737 | transaction->backend_data = data; |
| 1738 | |
| 1739 | if (!is_lock_file_locked(&refs->lock)) { |
| 1740 | if (packed_refs_lock(ref_store, 0, err)) |
| 1741 | goto failure; |
| 1742 | data->own_lock = 1; |
| 1743 | } |
| 1744 | |
| 1745 | ret = write_with_updates(refs, transaction, err); |
| 1746 | if (ret) |
| 1747 | goto failure; |
| 1748 | |
| 1749 | transaction->state = REF_TRANSACTION_PREPARED; |
| 1750 | return 0; |
| 1751 | |
| 1752 | failure: |
| 1753 | packed_transaction_cleanup(refs, transaction); |
| 1754 | return ret; |
| 1755 | } |
| 1756 | |
| 1757 | static int packed_transaction_abort(struct ref_store *ref_store, |
| 1758 | struct ref_transaction *transaction, |
| 1759 | struct strbuf *err UNUSED) |
| 1760 | { |
| 1761 | struct packed_ref_store *refs = packed_downcast( |
| 1762 | ref_store, |
| 1763 | REF_STORE_READ | REF_STORE_WRITE | REF_STORE_ODB, |
| 1764 | "ref_transaction_abort"); |
| 1765 | |
| 1766 | packed_transaction_cleanup(refs, transaction); |
| 1767 | return 0; |
| 1768 | } |
| 1769 | |
| 1770 | static int packed_transaction_finish(struct ref_store *ref_store, |
| 1771 | struct ref_transaction *transaction, |
| 1772 | struct strbuf *err) |
| 1773 | { |
| 1774 | struct packed_ref_store *refs = packed_downcast( |
| 1775 | ref_store, |
| 1776 | REF_STORE_READ | REF_STORE_WRITE | REF_STORE_ODB, |
| 1777 | "ref_transaction_finish"); |
| 1778 | int ret = REF_TRANSACTION_ERROR_GENERIC; |
| 1779 | char *packed_refs_path; |
| 1780 | |
| 1781 | clear_snapshot(refs); |
| 1782 | |
| 1783 | packed_refs_path = get_locked_file_path(&refs->lock); |
| 1784 | if (rename_tempfile(&refs->tempfile, packed_refs_path)) { |
| 1785 | strbuf_addf(err, "error replacing %s: %s", |
| 1786 | refs->path, strerror(errno)); |
| 1787 | goto cleanup; |
| 1788 | } |
| 1789 | |
| 1790 | ret = 0; |
| 1791 | |
| 1792 | cleanup: |
| 1793 | free(packed_refs_path); |
| 1794 | packed_transaction_cleanup(refs, transaction); |
| 1795 | return ret; |
| 1796 | } |
| 1797 | |
| 1798 | static int packed_optimize(struct ref_store *ref_store UNUSED, |
| 1799 | struct refs_optimize_opts *opts UNUSED) |
| 1800 | { |
| 1801 | /* |
| 1802 | * Packed refs are already packed. It might be that loose refs |
| 1803 | * are packed *into* a packed refs store, but that is done by |
| 1804 | * updating the packed references via a transaction. |
| 1805 | */ |
| 1806 | return 0; |
| 1807 | } |
| 1808 | |
| 1809 | static int packed_optimize_required(struct ref_store *ref_store UNUSED, |
| 1810 | struct refs_optimize_opts *opts UNUSED, |
| 1811 | bool *required) |
| 1812 | { |
| 1813 | /* |
| 1814 | * Packed refs are already optimized. |
| 1815 | */ |
| 1816 | *required = false; |
| 1817 | return 0; |
| 1818 | } |
| 1819 | |
| 1820 | static struct ref_iterator *packed_reflog_iterator_begin(struct ref_store *ref_store UNUSED) |
| 1821 | { |
| 1822 | return empty_ref_iterator_begin(); |
| 1823 | } |
| 1824 | |
| 1825 | static int packed_fsck_ref_next_line(struct fsck_options *o, |
| 1826 | unsigned long line_number, const char *start, |
| 1827 | const char *eof, const char **eol) |
| 1828 | { |
| 1829 | int ret = 0; |
| 1830 | |
| 1831 | *eol = memchr(start, '\n', eof - start); |
| 1832 | if (!*eol) { |
| 1833 | struct strbuf packed_entry = STRBUF_INIT; |
| 1834 | struct fsck_ref_report report = { 0 }; |
| 1835 | |
| 1836 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1837 | report.path = packed_entry.buf; |
| 1838 | ret = fsck_report_ref(o, &report, |
| 1839 | FSCK_MSG_PACKED_REF_ENTRY_NOT_TERMINATED, |
| 1840 | "'%.*s' is not terminated with a newline", |
| 1841 | (int)(eof - start), start); |
| 1842 | |
| 1843 | /* |
| 1844 | * There is no newline but we still want to parse it to the end of |
| 1845 | * the buffer. |
| 1846 | */ |
| 1847 | *eol = eof; |
| 1848 | strbuf_release(&packed_entry); |
| 1849 | } |
| 1850 | |
| 1851 | return ret; |
| 1852 | } |
| 1853 | |
| 1854 | static int packed_fsck_ref_header(struct fsck_options *o, |
| 1855 | const char *start, const char *eol, |
| 1856 | unsigned int *sorted) |
| 1857 | { |
| 1858 | struct string_list traits = STRING_LIST_INIT_NODUP; |
| 1859 | char *tmp_line; |
| 1860 | int ret = 0; |
| 1861 | char *p; |
| 1862 | |
| 1863 | tmp_line = xmemdupz(start, eol - start); |
| 1864 | if (!skip_prefix(tmp_line, "# pack-refs with: ", (const char **)&p)) { |
| 1865 | struct fsck_ref_report report = { 0 }; |
| 1866 | report.path = "packed-refs.header"; |
| 1867 | |
| 1868 | ret = fsck_report_ref(o, &report, |
| 1869 | FSCK_MSG_BAD_PACKED_REF_HEADER, |
| 1870 | "'%.*s' does not start with '# pack-refs with: '", |
| 1871 | (int)(eol - start), start); |
| 1872 | goto cleanup; |
| 1873 | } |
| 1874 | |
| 1875 | string_list_split_in_place(&traits, p, " ", -1); |
| 1876 | *sorted = unsorted_string_list_has_string(&traits, "sorted"); |
| 1877 | |
| 1878 | cleanup: |
| 1879 | free(tmp_line); |
| 1880 | string_list_clear(&traits, 0); |
| 1881 | return ret; |
| 1882 | } |
| 1883 | |
| 1884 | static int packed_fsck_ref_peeled_line(struct fsck_options *o, |
| 1885 | struct ref_store *ref_store, |
| 1886 | unsigned long line_number, |
| 1887 | const char *start, const char *eol) |
| 1888 | { |
| 1889 | struct strbuf packed_entry = STRBUF_INIT; |
| 1890 | struct fsck_ref_report report = { 0 }; |
| 1891 | struct object_id peeled; |
| 1892 | const char *p; |
| 1893 | int ret = 0; |
| 1894 | |
| 1895 | /* |
| 1896 | * Skip the '^' and parse the peeled oid. |
| 1897 | */ |
| 1898 | start++; |
| 1899 | if (parse_oid_hex_algop(start, &peeled, &p, ref_store->repo->hash_algo)) { |
| 1900 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1901 | report.path = packed_entry.buf; |
| 1902 | |
| 1903 | ret = fsck_report_ref(o, &report, |
| 1904 | FSCK_MSG_BAD_PACKED_REF_ENTRY, |
| 1905 | "'%.*s' has invalid peeled oid", |
| 1906 | (int)(eol - start), start); |
| 1907 | goto cleanup; |
| 1908 | } |
| 1909 | |
| 1910 | if (p != eol) { |
| 1911 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1912 | report.path = packed_entry.buf; |
| 1913 | |
| 1914 | ret = fsck_report_ref(o, &report, |
| 1915 | FSCK_MSG_BAD_PACKED_REF_ENTRY, |
| 1916 | "has trailing garbage after peeled oid '%.*s'", |
| 1917 | (int)(eol - p), p); |
| 1918 | goto cleanup; |
| 1919 | } |
| 1920 | |
| 1921 | cleanup: |
| 1922 | strbuf_release(&packed_entry); |
| 1923 | return ret; |
| 1924 | } |
| 1925 | |
| 1926 | static int packed_fsck_ref_main_line(struct fsck_options *o, |
| 1927 | struct ref_store *ref_store, |
| 1928 | unsigned long line_number, |
| 1929 | struct strbuf *refname, |
| 1930 | const char *start, const char *eol) |
| 1931 | { |
| 1932 | struct strbuf packed_entry = STRBUF_INIT; |
| 1933 | struct fsck_ref_report report = { 0 }; |
| 1934 | struct object_id oid; |
| 1935 | const char *p; |
| 1936 | int ret = 0; |
| 1937 | |
| 1938 | if (parse_oid_hex_algop(start, &oid, &p, ref_store->repo->hash_algo)) { |
| 1939 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1940 | report.path = packed_entry.buf; |
| 1941 | |
| 1942 | ret = fsck_report_ref(o, &report, |
| 1943 | FSCK_MSG_BAD_PACKED_REF_ENTRY, |
| 1944 | "'%.*s' has invalid oid", |
| 1945 | (int)(eol - start), start); |
| 1946 | goto cleanup; |
| 1947 | } |
| 1948 | |
| 1949 | if (p == eol || !isspace(*p)) { |
| 1950 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1951 | report.path = packed_entry.buf; |
| 1952 | |
| 1953 | ret = fsck_report_ref(o, &report, |
| 1954 | FSCK_MSG_BAD_PACKED_REF_ENTRY, |
| 1955 | "has no space after oid '%s' but with '%.*s'", |
| 1956 | oid_to_hex(&oid), (int)(eol - p), p); |
| 1957 | goto cleanup; |
| 1958 | } |
| 1959 | |
| 1960 | p++; |
| 1961 | strbuf_reset(refname); |
| 1962 | strbuf_add(refname, p, eol - p); |
| 1963 | if (refname_contains_nul(refname)) { |
| 1964 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1965 | report.path = packed_entry.buf; |
| 1966 | |
| 1967 | ret = fsck_report_ref(o, &report, |
| 1968 | FSCK_MSG_BAD_PACKED_REF_ENTRY, |
| 1969 | "refname '%s' contains NULL binaries", |
| 1970 | refname->buf); |
| 1971 | } |
| 1972 | |
| 1973 | if (check_refname_format(refname->buf, 0)) { |
| 1974 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 1975 | report.path = packed_entry.buf; |
| 1976 | |
| 1977 | ret = fsck_report_ref(o, &report, |
| 1978 | FSCK_MSG_BAD_REF_NAME, |
| 1979 | "has bad refname '%s'", refname->buf); |
| 1980 | } |
| 1981 | |
| 1982 | cleanup: |
| 1983 | strbuf_release(&packed_entry); |
| 1984 | return ret; |
| 1985 | } |
| 1986 | |
| 1987 | static int packed_fsck_ref_sorted(struct fsck_options *o, |
| 1988 | struct ref_store *ref_store, |
| 1989 | const char *start, const char *eof) |
| 1990 | { |
| 1991 | size_t hexsz = ref_store->repo->hash_algo->hexsz; |
| 1992 | struct strbuf packed_entry = STRBUF_INIT; |
| 1993 | struct fsck_ref_report report = { 0 }; |
| 1994 | struct strbuf refname1 = STRBUF_INIT; |
| 1995 | struct strbuf refname2 = STRBUF_INIT; |
| 1996 | unsigned long line_number = 1; |
| 1997 | const char *former = NULL; |
| 1998 | const char *current; |
| 1999 | const char *eol; |
| 2000 | int ret = 0; |
| 2001 | |
| 2002 | if (*start == '#') { |
| 2003 | eol = memchr(start, '\n', eof - start); |
| 2004 | start = eol + 1; |
| 2005 | line_number++; |
| 2006 | } |
| 2007 | |
| 2008 | for (; start < eof; line_number++, start = eol + 1) { |
| 2009 | eol = memchr(start, '\n', eof - start); |
| 2010 | |
| 2011 | if (*start == '^') |
| 2012 | continue; |
| 2013 | |
| 2014 | if (!former) { |
| 2015 | former = start + hexsz + 1; |
| 2016 | continue; |
| 2017 | } |
| 2018 | |
| 2019 | current = start + hexsz + 1; |
| 2020 | if (cmp_packed_refname(former, current) >= 0) { |
| 2021 | const char *err_fmt = |
| 2022 | "refname '%s' is less than previous refname '%s'"; |
| 2023 | |
| 2024 | eol = memchr(former, '\n', eof - former); |
| 2025 | strbuf_add(&refname1, former, eol - former); |
| 2026 | eol = memchr(current, '\n', eof - current); |
| 2027 | strbuf_add(&refname2, current, eol - current); |
| 2028 | |
| 2029 | strbuf_addf(&packed_entry, "packed-refs line %lu", line_number); |
| 2030 | report.path = packed_entry.buf; |
| 2031 | ret = fsck_report_ref(o, &report, |
| 2032 | FSCK_MSG_PACKED_REF_UNSORTED, |
| 2033 | err_fmt, refname2.buf, refname1.buf); |
| 2034 | goto cleanup; |
| 2035 | } |
| 2036 | former = current; |
| 2037 | } |
| 2038 | |
| 2039 | cleanup: |
| 2040 | strbuf_release(&packed_entry); |
| 2041 | strbuf_release(&refname1); |
| 2042 | strbuf_release(&refname2); |
| 2043 | return ret; |
| 2044 | } |
| 2045 | |
| 2046 | static int packed_fsck_ref_content(struct fsck_options *o, |
| 2047 | struct ref_store *ref_store, |
| 2048 | unsigned int *sorted, |
| 2049 | const char *start, const char *eof) |
| 2050 | { |
| 2051 | struct strbuf refname = STRBUF_INIT; |
| 2052 | unsigned long line_number = 1; |
| 2053 | const char *eol; |
| 2054 | int ret = 0; |
| 2055 | |
| 2056 | ret |= packed_fsck_ref_next_line(o, line_number, start, eof, &eol); |
| 2057 | if (*start == '#') { |
| 2058 | ret |= packed_fsck_ref_header(o, start, eol, sorted); |
| 2059 | |
| 2060 | start = eol + 1; |
| 2061 | line_number++; |
| 2062 | } |
| 2063 | |
| 2064 | while (start < eof) { |
| 2065 | ret |= packed_fsck_ref_next_line(o, line_number, start, eof, &eol); |
| 2066 | ret |= packed_fsck_ref_main_line(o, ref_store, line_number, &refname, start, eol); |
| 2067 | start = eol + 1; |
| 2068 | line_number++; |
| 2069 | if (start < eof && *start == '^') { |
| 2070 | ret |= packed_fsck_ref_next_line(o, line_number, start, eof, &eol); |
| 2071 | ret |= packed_fsck_ref_peeled_line(o, ref_store, line_number, |
| 2072 | start, eol); |
| 2073 | start = eol + 1; |
| 2074 | line_number++; |
| 2075 | } |
| 2076 | } |
| 2077 | |
| 2078 | strbuf_release(&refname); |
| 2079 | return ret; |
| 2080 | } |
| 2081 | |
| 2082 | static int packed_fsck(struct ref_store *ref_store, |
| 2083 | struct fsck_options *o, |
| 2084 | struct worktree *wt) |
| 2085 | { |
| 2086 | struct packed_ref_store *refs = packed_downcast(ref_store, |
| 2087 | REF_STORE_READ, "fsck"); |
| 2088 | struct snapshot snapshot = { 0 }; |
| 2089 | unsigned int sorted = 0; |
| 2090 | struct stat st; |
| 2091 | int ret = 0; |
| 2092 | int fd = -1; |
| 2093 | |
| 2094 | if (!is_main_worktree(wt)) |
| 2095 | goto cleanup; |
| 2096 | |
| 2097 | if (o->verbose) |
| 2098 | fprintf_ln(stderr, "Checking packed-refs file %s", refs->path); |
| 2099 | |
| 2100 | fd = open_nofollow(refs->path, O_RDONLY); |
| 2101 | if (fd < 0) { |
| 2102 | /* |
| 2103 | * If the packed-refs file doesn't exist, there's nothing |
| 2104 | * to check. |
| 2105 | */ |
| 2106 | if (errno == ENOENT) |
| 2107 | goto cleanup; |
| 2108 | |
| 2109 | if (errno == ELOOP) { |
| 2110 | struct fsck_ref_report report = { 0 }; |
| 2111 | report.path = "packed-refs"; |
| 2112 | ret = fsck_report_ref(o, &report, |
| 2113 | FSCK_MSG_BAD_REF_FILETYPE, |
| 2114 | "not a regular file but a symlink"); |
| 2115 | goto cleanup; |
| 2116 | } |
| 2117 | |
| 2118 | ret = error_errno(_("unable to open '%s'"), refs->path); |
| 2119 | goto cleanup; |
| 2120 | } else if (fstat(fd, &st) < 0) { |
| 2121 | ret = error_errno(_("unable to stat '%s'"), refs->path); |
| 2122 | goto cleanup; |
| 2123 | } else if (!S_ISREG(st.st_mode)) { |
| 2124 | struct fsck_ref_report report = { 0 }; |
| 2125 | report.path = "packed-refs"; |
| 2126 | ret = fsck_report_ref(o, &report, |
| 2127 | FSCK_MSG_BAD_REF_FILETYPE, |
| 2128 | "not a regular file"); |
| 2129 | goto cleanup; |
| 2130 | } |
| 2131 | |
| 2132 | if (!allocate_snapshot_buffer(&snapshot, fd, &st)) { |
| 2133 | struct fsck_ref_report report = { 0 }; |
| 2134 | report.path = "packed-refs"; |
| 2135 | ret = fsck_report_ref(o, &report, |
| 2136 | FSCK_MSG_EMPTY_PACKED_REFS_FILE, |
| 2137 | "file is empty"); |
| 2138 | goto cleanup; |
| 2139 | } |
| 2140 | |
| 2141 | ret = packed_fsck_ref_content(o, ref_store, &sorted, snapshot.start, |
| 2142 | snapshot.eof); |
| 2143 | if (!ret && sorted) |
| 2144 | ret = packed_fsck_ref_sorted(o, ref_store, snapshot.start, |
| 2145 | snapshot.eof); |
| 2146 | |
| 2147 | cleanup: |
| 2148 | if (fd >= 0) |
| 2149 | close(fd); |
| 2150 | clear_snapshot_buffer(&snapshot); |
| 2151 | return ret; |
| 2152 | } |
| 2153 | |
| 2154 | struct ref_storage_be refs_be_packed = { |
| 2155 | .name = "packed", |
| 2156 | .init = packed_ref_store_init, |
| 2157 | .release = packed_ref_store_release, |
| 2158 | .create_on_disk = packed_ref_store_create_on_disk, |
| 2159 | .remove_on_disk = packed_ref_store_remove_on_disk, |
| 2160 | |
| 2161 | .transaction_prepare = packed_transaction_prepare, |
| 2162 | .transaction_finish = packed_transaction_finish, |
| 2163 | .transaction_abort = packed_transaction_abort, |
| 2164 | |
| 2165 | .optimize = packed_optimize, |
| 2166 | .optimize_required = packed_optimize_required, |
| 2167 | |
| 2168 | .rename_ref = NULL, |
| 2169 | .copy_ref = NULL, |
| 2170 | |
| 2171 | .iterator_begin = packed_ref_iterator_begin, |
| 2172 | .read_raw_ref = packed_read_raw_ref, |
| 2173 | .read_symbolic_ref = NULL, |
| 2174 | |
| 2175 | .reflog_iterator_begin = packed_reflog_iterator_begin, |
| 2176 | .for_each_reflog_ent = NULL, |
| 2177 | .for_each_reflog_ent_reverse = NULL, |
| 2178 | .reflog_exists = NULL, |
| 2179 | .create_reflog = NULL, |
| 2180 | .delete_reflog = NULL, |
| 2181 | .reflog_expire = NULL, |
| 2182 | |
| 2183 | .fsck = packed_fsck, |
| 2184 | }; |