tests: fix cleanup after tests in t1509-root-worktree
[git/git.git] / builtin / fsck.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "commit.h"
4 #include "tree.h"
5 #include "blob.h"
6 #include "tag.h"
7 #include "refs.h"
8 #include "pack.h"
9 #include "cache-tree.h"
10 #include "tree-walk.h"
11 #include "fsck.h"
12 #include "parse-options.h"
13 #include "dir.h"
14 #include "progress.h"
15 #include "streaming.h"
16
17 #define REACHABLE 0x0001
18 #define SEEN 0x0002
19 #define HAS_OBJ 0x0004
20
21 static int show_root;
22 static int show_tags;
23 static int show_unreachable;
24 static int include_reflogs = 1;
25 static int check_full = 1;
26 static int check_strict;
27 static int keep_cache_objects;
28 static struct object_id head_oid;
29 static const char *head_points_at;
30 static int errors_found;
31 static int write_lost_and_found;
32 static int verbose;
33 static int show_progress = -1;
34 static int show_dangling = 1;
35 #define ERROR_OBJECT 01
36 #define ERROR_REACHABLE 02
37 #define ERROR_PACK 04
38
39 #ifdef NO_D_INO_IN_DIRENT
40 #define SORT_DIRENT 0
41 #define DIRENT_SORT_HINT(de) 0
42 #else
43 #define SORT_DIRENT 1
44 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
45 #endif
46
47 static void objreport(struct object *obj, const char *severity,
48 const char *err, va_list params)
49 {
50 fprintf(stderr, "%s in %s %s: ",
51 severity, typename(obj->type), sha1_to_hex(obj->sha1));
52 vfprintf(stderr, err, params);
53 fputs("\n", stderr);
54 }
55
56 __attribute__((format (printf, 2, 3)))
57 static int objerror(struct object *obj, const char *err, ...)
58 {
59 va_list params;
60 va_start(params, err);
61 errors_found |= ERROR_OBJECT;
62 objreport(obj, "error", err, params);
63 va_end(params);
64 return -1;
65 }
66
67 __attribute__((format (printf, 3, 4)))
68 static int fsck_error_func(struct object *obj, int type, const char *err, ...)
69 {
70 va_list params;
71 va_start(params, err);
72 objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
73 va_end(params);
74 return (type == FSCK_WARN) ? 0 : 1;
75 }
76
77 static struct object_array pending;
78
79 static int mark_object(struct object *obj, int type, void *data)
80 {
81 struct object *parent = data;
82
83 /*
84 * The only case data is NULL or type is OBJ_ANY is when
85 * mark_object_reachable() calls us. All the callers of
86 * that function has non-NULL obj hence ...
87 */
88 if (!obj) {
89 /* ... these references to parent->fld are safe here */
90 printf("broken link from %7s %s\n",
91 typename(parent->type), sha1_to_hex(parent->sha1));
92 printf("broken link from %7s %s\n",
93 (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
94 errors_found |= ERROR_REACHABLE;
95 return 1;
96 }
97
98 if (type != OBJ_ANY && obj->type != type)
99 /* ... and the reference to parent is safe here */
100 objerror(parent, "wrong object type in link");
101
102 if (obj->flags & REACHABLE)
103 return 0;
104 obj->flags |= REACHABLE;
105 if (!(obj->flags & HAS_OBJ)) {
106 if (parent && !has_sha1_file(obj->sha1)) {
107 printf("broken link from %7s %s\n",
108 typename(parent->type), sha1_to_hex(parent->sha1));
109 printf(" to %7s %s\n",
110 typename(obj->type), sha1_to_hex(obj->sha1));
111 errors_found |= ERROR_REACHABLE;
112 }
113 return 1;
114 }
115
116 add_object_array(obj, NULL, &pending);
117 return 0;
118 }
119
120 static void mark_object_reachable(struct object *obj)
121 {
122 mark_object(obj, OBJ_ANY, NULL);
123 }
124
125 static int traverse_one_object(struct object *obj)
126 {
127 int result;
128 struct tree *tree = NULL;
129
130 if (obj->type == OBJ_TREE) {
131 tree = (struct tree *)obj;
132 if (parse_tree(tree) < 0)
133 return 1; /* error already displayed */
134 }
135 result = fsck_walk(obj, mark_object, obj);
136 if (tree)
137 free_tree_buffer(tree);
138 return result;
139 }
140
141 static int traverse_reachable(void)
142 {
143 struct progress *progress = NULL;
144 unsigned int nr = 0;
145 int result = 0;
146 if (show_progress)
147 progress = start_progress_delay(_("Checking connectivity"), 0, 0, 2);
148 while (pending.nr) {
149 struct object_array_entry *entry;
150 struct object *obj;
151
152 entry = pending.objects + --pending.nr;
153 obj = entry->item;
154 result |= traverse_one_object(obj);
155 display_progress(progress, ++nr);
156 }
157 stop_progress(&progress);
158 return !!result;
159 }
160
161 static int mark_used(struct object *obj, int type, void *data)
162 {
163 if (!obj)
164 return 1;
165 obj->used = 1;
166 return 0;
167 }
168
169 /*
170 * Check a single reachable object
171 */
172 static void check_reachable_object(struct object *obj)
173 {
174 /*
175 * We obviously want the object to be parsed,
176 * except if it was in a pack-file and we didn't
177 * do a full fsck
178 */
179 if (!(obj->flags & HAS_OBJ)) {
180 if (has_sha1_pack(obj->sha1))
181 return; /* it is in pack - forget about it */
182 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
183 errors_found |= ERROR_REACHABLE;
184 return;
185 }
186 }
187
188 /*
189 * Check a single unreachable object
190 */
191 static void check_unreachable_object(struct object *obj)
192 {
193 /*
194 * Missing unreachable object? Ignore it. It's not like
195 * we miss it (since it can't be reached), nor do we want
196 * to complain about it being unreachable (since it does
197 * not exist).
198 */
199 if (!obj->parsed)
200 return;
201
202 /*
203 * Unreachable object that exists? Show it if asked to,
204 * since this is something that is prunable.
205 */
206 if (show_unreachable) {
207 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
208 return;
209 }
210
211 /*
212 * "!used" means that nothing at all points to it, including
213 * other unreachable objects. In other words, it's the "tip"
214 * of some set of unreachable objects, usually a commit that
215 * got dropped.
216 *
217 * Such starting points are more interesting than some random
218 * set of unreachable objects, so we show them even if the user
219 * hasn't asked for _all_ unreachable objects. If you have
220 * deleted a branch by mistake, this is a prime candidate to
221 * start looking at, for example.
222 */
223 if (!obj->used) {
224 if (show_dangling)
225 printf("dangling %s %s\n", typename(obj->type),
226 sha1_to_hex(obj->sha1));
227 if (write_lost_and_found) {
228 const char *filename = git_path("lost-found/%s/%s",
229 obj->type == OBJ_COMMIT ? "commit" : "other",
230 sha1_to_hex(obj->sha1));
231 FILE *f;
232
233 if (safe_create_leading_directories_const(filename)) {
234 error("Could not create lost-found");
235 return;
236 }
237 if (!(f = fopen(filename, "w")))
238 die_errno("Could not open '%s'", filename);
239 if (obj->type == OBJ_BLOB) {
240 if (stream_blob_to_fd(fileno(f), obj->sha1, NULL, 1))
241 die_errno("Could not write '%s'", filename);
242 } else
243 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
244 if (fclose(f))
245 die_errno("Could not finish '%s'",
246 filename);
247 }
248 return;
249 }
250
251 /*
252 * Otherwise? It's there, it's unreachable, and some other unreachable
253 * object points to it. Ignore it - it's not interesting, and we showed
254 * all the interesting cases above.
255 */
256 }
257
258 static void check_object(struct object *obj)
259 {
260 if (verbose)
261 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
262
263 if (obj->flags & REACHABLE)
264 check_reachable_object(obj);
265 else
266 check_unreachable_object(obj);
267 }
268
269 static void check_connectivity(void)
270 {
271 int i, max;
272
273 /* Traverse the pending reachable objects */
274 traverse_reachable();
275
276 /* Look up all the requirements, warn about missing objects.. */
277 max = get_max_object_index();
278 if (verbose)
279 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
280
281 for (i = 0; i < max; i++) {
282 struct object *obj = get_indexed_object(i);
283
284 if (obj)
285 check_object(obj);
286 }
287 }
288
289 static int fsck_obj(struct object *obj)
290 {
291 if (obj->flags & SEEN)
292 return 0;
293 obj->flags |= SEEN;
294
295 if (verbose)
296 fprintf(stderr, "Checking %s %s\n",
297 typename(obj->type), sha1_to_hex(obj->sha1));
298
299 if (fsck_walk(obj, mark_used, NULL))
300 objerror(obj, "broken links");
301 if (fsck_object(obj, NULL, 0, check_strict, fsck_error_func))
302 return -1;
303
304 if (obj->type == OBJ_TREE) {
305 struct tree *item = (struct tree *) obj;
306
307 free_tree_buffer(item);
308 }
309
310 if (obj->type == OBJ_COMMIT) {
311 struct commit *commit = (struct commit *) obj;
312
313 free_commit_buffer(commit);
314
315 if (!commit->parents && show_root)
316 printf("root %s\n", sha1_to_hex(commit->object.sha1));
317 }
318
319 if (obj->type == OBJ_TAG) {
320 struct tag *tag = (struct tag *) obj;
321
322 if (show_tags && tag->tagged) {
323 printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
324 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
325 }
326 }
327
328 return 0;
329 }
330
331 static int fsck_sha1(const unsigned char *sha1)
332 {
333 struct object *obj = parse_object(sha1);
334 if (!obj) {
335 errors_found |= ERROR_OBJECT;
336 return error("%s: object corrupt or missing",
337 sha1_to_hex(sha1));
338 }
339 obj->flags |= HAS_OBJ;
340 return fsck_obj(obj);
341 }
342
343 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
344 unsigned long size, void *buffer, int *eaten)
345 {
346 struct object *obj;
347 obj = parse_object_buffer(sha1, type, size, buffer, eaten);
348 if (!obj) {
349 errors_found |= ERROR_OBJECT;
350 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
351 }
352 obj->flags = HAS_OBJ;
353 return fsck_obj(obj);
354 }
355
356 /*
357 * This is the sorting chunk size: make it reasonably
358 * big so that we can sort well..
359 */
360 #define MAX_SHA1_ENTRIES (1024)
361
362 struct sha1_entry {
363 unsigned long ino;
364 unsigned char sha1[20];
365 };
366
367 static struct {
368 unsigned long nr;
369 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
370 } sha1_list;
371
372 static int ino_compare(const void *_a, const void *_b)
373 {
374 const struct sha1_entry *a = _a, *b = _b;
375 unsigned long ino1 = a->ino, ino2 = b->ino;
376 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
377 }
378
379 static void fsck_sha1_list(void)
380 {
381 int i, nr = sha1_list.nr;
382
383 if (SORT_DIRENT)
384 qsort(sha1_list.entry, nr,
385 sizeof(struct sha1_entry *), ino_compare);
386 for (i = 0; i < nr; i++) {
387 struct sha1_entry *entry = sha1_list.entry[i];
388 unsigned char *sha1 = entry->sha1;
389
390 sha1_list.entry[i] = NULL;
391 if (fsck_sha1(sha1))
392 errors_found |= ERROR_OBJECT;
393 free(entry);
394 }
395 sha1_list.nr = 0;
396 }
397
398 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
399 {
400 struct sha1_entry *entry = xmalloc(sizeof(*entry));
401 int nr;
402
403 entry->ino = ino;
404 hashcpy(entry->sha1, sha1);
405 nr = sha1_list.nr;
406 if (nr == MAX_SHA1_ENTRIES) {
407 fsck_sha1_list();
408 nr = 0;
409 }
410 sha1_list.entry[nr] = entry;
411 sha1_list.nr = ++nr;
412 }
413
414 static inline int is_loose_object_file(struct dirent *de,
415 char *name, unsigned char *sha1)
416 {
417 if (strlen(de->d_name) != 38)
418 return 0;
419 memcpy(name + 2, de->d_name, 39);
420 return !get_sha1_hex(name, sha1);
421 }
422
423 static void fsck_dir(int i, char *path)
424 {
425 DIR *dir = opendir(path);
426 struct dirent *de;
427 char name[100];
428
429 if (!dir)
430 return;
431
432 if (verbose)
433 fprintf(stderr, "Checking directory %s\n", path);
434
435 sprintf(name, "%02x", i);
436 while ((de = readdir(dir)) != NULL) {
437 unsigned char sha1[20];
438
439 if (is_dot_or_dotdot(de->d_name))
440 continue;
441 if (is_loose_object_file(de, name, sha1)) {
442 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
443 continue;
444 }
445 if (starts_with(de->d_name, "tmp_obj_"))
446 continue;
447 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
448 }
449 closedir(dir);
450 }
451
452 static int default_refs;
453
454 static void fsck_handle_reflog_sha1(const char *refname, unsigned char *sha1)
455 {
456 struct object *obj;
457
458 if (!is_null_sha1(sha1)) {
459 obj = lookup_object(sha1);
460 if (obj) {
461 obj->used = 1;
462 mark_object_reachable(obj);
463 } else {
464 error("%s: invalid reflog entry %s", refname, sha1_to_hex(sha1));
465 errors_found |= ERROR_REACHABLE;
466 }
467 }
468 }
469
470 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
471 const char *email, unsigned long timestamp, int tz,
472 const char *message, void *cb_data)
473 {
474 const char *refname = cb_data;
475
476 if (verbose)
477 fprintf(stderr, "Checking reflog %s->%s\n",
478 sha1_to_hex(osha1), sha1_to_hex(nsha1));
479
480 fsck_handle_reflog_sha1(refname, osha1);
481 fsck_handle_reflog_sha1(refname, nsha1);
482 return 0;
483 }
484
485 static int fsck_handle_reflog(const char *logname, const struct object_id *oid,
486 int flag, void *cb_data)
487 {
488 for_each_reflog_ent(logname, fsck_handle_reflog_ent, (void *)logname);
489 return 0;
490 }
491
492 static int fsck_handle_ref(const char *refname, const struct object_id *oid,
493 int flag, void *cb_data)
494 {
495 struct object *obj;
496
497 obj = parse_object(oid->hash);
498 if (!obj) {
499 error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
500 errors_found |= ERROR_REACHABLE;
501 /* We'll continue with the rest despite the error.. */
502 return 0;
503 }
504 if (obj->type != OBJ_COMMIT && is_branch(refname))
505 error("%s: not a commit", refname);
506 default_refs++;
507 obj->used = 1;
508 mark_object_reachable(obj);
509
510 return 0;
511 }
512
513 static void get_default_heads(void)
514 {
515 if (head_points_at && !is_null_oid(&head_oid))
516 fsck_handle_ref("HEAD", &head_oid, 0, NULL);
517 for_each_rawref(fsck_handle_ref, NULL);
518 if (include_reflogs)
519 for_each_reflog(fsck_handle_reflog, NULL);
520
521 /*
522 * Not having any default heads isn't really fatal, but
523 * it does mean that "--unreachable" no longer makes any
524 * sense (since in this case everything will obviously
525 * be unreachable by definition.
526 *
527 * Showing dangling objects is valid, though (as those
528 * dangling objects are likely lost heads).
529 *
530 * So we just print a warning about it, and clear the
531 * "show_unreachable" flag.
532 */
533 if (!default_refs) {
534 fprintf(stderr, "notice: No default references\n");
535 show_unreachable = 0;
536 }
537 }
538
539 static void fsck_object_dir(const char *path)
540 {
541 int i;
542 struct progress *progress = NULL;
543
544 if (verbose)
545 fprintf(stderr, "Checking object directory\n");
546
547 if (show_progress)
548 progress = start_progress(_("Checking object directories"), 256);
549 for (i = 0; i < 256; i++) {
550 static char dir[4096];
551 sprintf(dir, "%s/%02x", path, i);
552 fsck_dir(i, dir);
553 display_progress(progress, i+1);
554 }
555 stop_progress(&progress);
556 fsck_sha1_list();
557 }
558
559 static int fsck_head_link(void)
560 {
561 int flag;
562 int null_is_error = 0;
563
564 if (verbose)
565 fprintf(stderr, "Checking HEAD link\n");
566
567 head_points_at = resolve_ref_unsafe("HEAD", 0, head_oid.hash, &flag);
568 if (!head_points_at)
569 return error("Invalid HEAD");
570 if (!strcmp(head_points_at, "HEAD"))
571 /* detached HEAD */
572 null_is_error = 1;
573 else if (!starts_with(head_points_at, "refs/heads/"))
574 return error("HEAD points to something strange (%s)",
575 head_points_at);
576 if (is_null_oid(&head_oid)) {
577 if (null_is_error)
578 return error("HEAD: detached HEAD points at nothing");
579 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
580 head_points_at + 11);
581 }
582 return 0;
583 }
584
585 static int fsck_cache_tree(struct cache_tree *it)
586 {
587 int i;
588 int err = 0;
589
590 if (verbose)
591 fprintf(stderr, "Checking cache tree\n");
592
593 if (0 <= it->entry_count) {
594 struct object *obj = parse_object(it->sha1);
595 if (!obj) {
596 error("%s: invalid sha1 pointer in cache-tree",
597 sha1_to_hex(it->sha1));
598 return 1;
599 }
600 obj->used = 1;
601 mark_object_reachable(obj);
602 if (obj->type != OBJ_TREE)
603 err |= objerror(obj, "non-tree in cache-tree");
604 }
605 for (i = 0; i < it->subtree_nr; i++)
606 err |= fsck_cache_tree(it->down[i]->cache_tree);
607 return err;
608 }
609
610 static char const * const fsck_usage[] = {
611 N_("git fsck [<options>] [<object>...]"),
612 NULL
613 };
614
615 static struct option fsck_opts[] = {
616 OPT__VERBOSE(&verbose, N_("be verbose")),
617 OPT_BOOL(0, "unreachable", &show_unreachable, N_("show unreachable objects")),
618 OPT_BOOL(0, "dangling", &show_dangling, N_("show dangling objects")),
619 OPT_BOOL(0, "tags", &show_tags, N_("report tags")),
620 OPT_BOOL(0, "root", &show_root, N_("report root nodes")),
621 OPT_BOOL(0, "cache", &keep_cache_objects, N_("make index objects head nodes")),
622 OPT_BOOL(0, "reflogs", &include_reflogs, N_("make reflogs head nodes (default)")),
623 OPT_BOOL(0, "full", &check_full, N_("also consider packs and alternate objects")),
624 OPT_BOOL(0, "strict", &check_strict, N_("enable more strict checking")),
625 OPT_BOOL(0, "lost-found", &write_lost_and_found,
626 N_("write dangling objects in .git/lost-found")),
627 OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
628 OPT_END(),
629 };
630
631 int cmd_fsck(int argc, const char **argv, const char *prefix)
632 {
633 int i, heads;
634 struct alternate_object_database *alt;
635
636 errors_found = 0;
637 check_replace_refs = 0;
638
639 argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
640
641 if (show_progress == -1)
642 show_progress = isatty(2);
643 if (verbose)
644 show_progress = 0;
645
646 if (write_lost_and_found) {
647 check_full = 1;
648 include_reflogs = 0;
649 }
650
651 fsck_head_link();
652 fsck_object_dir(get_object_directory());
653
654 prepare_alt_odb();
655 for (alt = alt_odb_list; alt; alt = alt->next) {
656 char namebuf[PATH_MAX];
657 int namelen = alt->name - alt->base;
658 memcpy(namebuf, alt->base, namelen);
659 namebuf[namelen - 1] = 0;
660 fsck_object_dir(namebuf);
661 }
662
663 if (check_full) {
664 struct packed_git *p;
665 uint32_t total = 0, count = 0;
666 struct progress *progress = NULL;
667
668 prepare_packed_git();
669
670 if (show_progress) {
671 for (p = packed_git; p; p = p->next) {
672 if (open_pack_index(p))
673 continue;
674 total += p->num_objects;
675 }
676
677 progress = start_progress(_("Checking objects"), total);
678 }
679 for (p = packed_git; p; p = p->next) {
680 /* verify gives error messages itself */
681 if (verify_pack(p, fsck_obj_buffer,
682 progress, count))
683 errors_found |= ERROR_PACK;
684 count += p->num_objects;
685 }
686 stop_progress(&progress);
687 }
688
689 heads = 0;
690 for (i = 0; i < argc; i++) {
691 const char *arg = argv[i];
692 unsigned char sha1[20];
693 if (!get_sha1(arg, sha1)) {
694 struct object *obj = lookup_object(sha1);
695
696 /* Error is printed by lookup_object(). */
697 if (!obj)
698 continue;
699
700 obj->used = 1;
701 mark_object_reachable(obj);
702 heads++;
703 continue;
704 }
705 error("invalid parameter: expected sha1, got '%s'", arg);
706 }
707
708 /*
709 * If we've not been given any explicit head information, do the
710 * default ones from .git/refs. We also consider the index file
711 * in this case (ie this implies --cache).
712 */
713 if (!heads) {
714 get_default_heads();
715 keep_cache_objects = 1;
716 }
717
718 if (keep_cache_objects) {
719 read_cache();
720 for (i = 0; i < active_nr; i++) {
721 unsigned int mode;
722 struct blob *blob;
723 struct object *obj;
724
725 mode = active_cache[i]->ce_mode;
726 if (S_ISGITLINK(mode))
727 continue;
728 blob = lookup_blob(active_cache[i]->sha1);
729 if (!blob)
730 continue;
731 obj = &blob->object;
732 obj->used = 1;
733 mark_object_reachable(obj);
734 }
735 if (active_cache_tree)
736 fsck_cache_tree(active_cache_tree);
737 }
738
739 check_connectivity();
740 return errors_found;
741 }