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