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65 | * `get_snapshot()` is called; if the existing snapshot is obsolete, a | |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