Makefile: allow generating git.o for debugging purposes
[git/spearce.git] / builtin-fsck.c
blob944a496650b6e13d2684bb221ee8bc7aadd78d6f
1 #include "cache.h"
2 #include "commit.h"
3 #include "tree.h"
4 #include "blob.h"
5 #include "tag.h"
6 #include "refs.h"
7 #include "pack.h"
8 #include "cache-tree.h"
9 #include "tree-walk.h"
11 #define REACHABLE 0x0001
12 #define SEEN 0x0002
14 static int show_root;
15 static int show_tags;
16 static int show_unreachable;
17 static int include_reflogs = 1;
18 static int check_full;
19 static int check_strict;
20 static int keep_cache_objects;
21 static unsigned char head_sha1[20];
22 static int errors_found;
23 static int verbose;
24 #define ERROR_OBJECT 01
25 #define ERROR_REACHABLE 02
27 #ifdef NO_D_INO_IN_DIRENT
28 #define SORT_DIRENT 0
29 #define DIRENT_SORT_HINT(de) 0
30 #else
31 #define SORT_DIRENT 1
32 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
33 #endif
35 static void objreport(struct object *obj, const char *severity,
36 const char *err, va_list params)
38 fprintf(stderr, "%s in %s %s: ",
39 severity, typename(obj->type), sha1_to_hex(obj->sha1));
40 vfprintf(stderr, err, params);
41 fputs("\n", stderr);
44 static int objerror(struct object *obj, const char *err, ...)
46 va_list params;
47 va_start(params, err);
48 errors_found |= ERROR_OBJECT;
49 objreport(obj, "error", err, params);
50 va_end(params);
51 return -1;
54 static int objwarning(struct object *obj, const char *err, ...)
56 va_list params;
57 va_start(params, err);
58 objreport(obj, "warning", err, params);
59 va_end(params);
60 return -1;
64 * Check a single reachable object
66 static void check_reachable_object(struct object *obj)
68 const struct object_refs *refs;
71 * We obviously want the object to be parsed,
72 * except if it was in a pack-file and we didn't
73 * do a full fsck
75 if (!obj->parsed) {
76 if (has_sha1_pack(obj->sha1, NULL))
77 return; /* it is in pack - forget about it */
78 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
79 errors_found |= ERROR_REACHABLE;
80 return;
84 * Check that everything that we try to reference is also good.
86 refs = lookup_object_refs(obj);
87 if (refs) {
88 unsigned j;
89 for (j = 0; j < refs->count; j++) {
90 struct object *ref = refs->ref[j];
91 if (ref->parsed ||
92 (has_sha1_file(ref->sha1)))
93 continue;
94 printf("broken link from %7s %s\n",
95 typename(obj->type), sha1_to_hex(obj->sha1));
96 printf(" to %7s %s\n",
97 typename(ref->type), sha1_to_hex(ref->sha1));
98 errors_found |= ERROR_REACHABLE;
104 * Check a single unreachable object
106 static void check_unreachable_object(struct object *obj)
109 * Missing unreachable object? Ignore it. It's not like
110 * we miss it (since it can't be reached), nor do we want
111 * to complain about it being unreachable (since it does
112 * not exist).
114 if (!obj->parsed)
115 return;
118 * Unreachable object that exists? Show it if asked to,
119 * since this is something that is prunable.
121 if (show_unreachable) {
122 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
123 return;
127 * "!used" means that nothing at all points to it, including
128 * other unreachable objects. In other words, it's the "tip"
129 * of some set of unreachable objects, usually a commit that
130 * got dropped.
132 * Such starting points are more interesting than some random
133 * set of unreachable objects, so we show them even if the user
134 * hasn't asked for _all_ unreachable objects. If you have
135 * deleted a branch by mistake, this is a prime candidate to
136 * start looking at, for example.
138 if (!obj->used) {
139 printf("dangling %s %s\n", typename(obj->type),
140 sha1_to_hex(obj->sha1));
141 return;
145 * Otherwise? It's there, it's unreachable, and some other unreachable
146 * object points to it. Ignore it - it's not interesting, and we showed
147 * all the interesting cases above.
151 static void check_object(struct object *obj)
153 if (verbose)
154 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
156 if (obj->flags & REACHABLE)
157 check_reachable_object(obj);
158 else
159 check_unreachable_object(obj);
162 static void check_connectivity(void)
164 int i, max;
166 /* Look up all the requirements, warn about missing objects.. */
167 max = get_max_object_index();
168 if (verbose)
169 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
171 for (i = 0; i < max; i++) {
172 struct object *obj = get_indexed_object(i);
174 if (obj)
175 check_object(obj);
180 * The entries in a tree are ordered in the _path_ order,
181 * which means that a directory entry is ordered by adding
182 * a slash to the end of it.
184 * So a directory called "a" is ordered _after_ a file
185 * called "a.c", because "a/" sorts after "a.c".
187 #define TREE_UNORDERED (-1)
188 #define TREE_HAS_DUPS (-2)
190 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
192 int len1 = strlen(name1);
193 int len2 = strlen(name2);
194 int len = len1 < len2 ? len1 : len2;
195 unsigned char c1, c2;
196 int cmp;
198 cmp = memcmp(name1, name2, len);
199 if (cmp < 0)
200 return 0;
201 if (cmp > 0)
202 return TREE_UNORDERED;
205 * Ok, the first <len> characters are the same.
206 * Now we need to order the next one, but turn
207 * a '\0' into a '/' for a directory entry.
209 c1 = name1[len];
210 c2 = name2[len];
211 if (!c1 && !c2)
213 * git-write-tree used to write out a nonsense tree that has
214 * entries with the same name, one blob and one tree. Make
215 * sure we do not have duplicate entries.
217 return TREE_HAS_DUPS;
218 if (!c1 && S_ISDIR(mode1))
219 c1 = '/';
220 if (!c2 && S_ISDIR(mode2))
221 c2 = '/';
222 return c1 < c2 ? 0 : TREE_UNORDERED;
225 static int fsck_tree(struct tree *item)
227 int retval;
228 int has_full_path = 0;
229 int has_empty_name = 0;
230 int has_zero_pad = 0;
231 int has_bad_modes = 0;
232 int has_dup_entries = 0;
233 int not_properly_sorted = 0;
234 struct tree_desc desc;
235 unsigned o_mode;
236 const char *o_name;
237 const unsigned char *o_sha1;
239 if (verbose)
240 fprintf(stderr, "Checking tree %s\n",
241 sha1_to_hex(item->object.sha1));
243 init_tree_desc(&desc, item->buffer, item->size);
245 o_mode = 0;
246 o_name = NULL;
247 o_sha1 = NULL;
248 while (desc.size) {
249 unsigned mode;
250 const char *name;
251 const unsigned char *sha1;
253 sha1 = tree_entry_extract(&desc, &name, &mode);
255 if (strchr(name, '/'))
256 has_full_path = 1;
257 if (!*name)
258 has_empty_name = 1;
259 has_zero_pad |= *(char *)desc.buffer == '0';
260 update_tree_entry(&desc);
262 switch (mode) {
264 * Standard modes..
266 case S_IFREG | 0755:
267 case S_IFREG | 0644:
268 case S_IFLNK:
269 case S_IFDIR:
270 case S_IFGITLINK:
271 break;
273 * This is nonstandard, but we had a few of these
274 * early on when we honored the full set of mode
275 * bits..
277 case S_IFREG | 0664:
278 if (!check_strict)
279 break;
280 default:
281 has_bad_modes = 1;
284 if (o_name) {
285 switch (verify_ordered(o_mode, o_name, mode, name)) {
286 case TREE_UNORDERED:
287 not_properly_sorted = 1;
288 break;
289 case TREE_HAS_DUPS:
290 has_dup_entries = 1;
291 break;
292 default:
293 break;
297 o_mode = mode;
298 o_name = name;
299 o_sha1 = sha1;
301 free(item->buffer);
302 item->buffer = NULL;
304 retval = 0;
305 if (has_full_path) {
306 objwarning(&item->object, "contains full pathnames");
308 if (has_empty_name) {
309 objwarning(&item->object, "contains empty pathname");
311 if (has_zero_pad) {
312 objwarning(&item->object, "contains zero-padded file modes");
314 if (has_bad_modes) {
315 objwarning(&item->object, "contains bad file modes");
317 if (has_dup_entries) {
318 retval = objerror(&item->object, "contains duplicate file entries");
320 if (not_properly_sorted) {
321 retval = objerror(&item->object, "not properly sorted");
323 return retval;
326 static int fsck_commit(struct commit *commit)
328 char *buffer = commit->buffer;
329 unsigned char tree_sha1[20], sha1[20];
331 if (verbose)
332 fprintf(stderr, "Checking commit %s\n",
333 sha1_to_hex(commit->object.sha1));
335 if (memcmp(buffer, "tree ", 5))
336 return objerror(&commit->object, "invalid format - expected 'tree' line");
337 if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
338 return objerror(&commit->object, "invalid 'tree' line format - bad sha1");
339 buffer += 46;
340 while (!memcmp(buffer, "parent ", 7)) {
341 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
342 return objerror(&commit->object, "invalid 'parent' line format - bad sha1");
343 buffer += 48;
345 if (memcmp(buffer, "author ", 7))
346 return objerror(&commit->object, "invalid format - expected 'author' line");
347 free(commit->buffer);
348 commit->buffer = NULL;
349 if (!commit->tree)
350 return objerror(&commit->object, "could not load commit's tree %s", tree_sha1);
351 if (!commit->parents && show_root)
352 printf("root %s\n", sha1_to_hex(commit->object.sha1));
353 if (!commit->date)
354 printf("bad commit date in %s\n",
355 sha1_to_hex(commit->object.sha1));
356 return 0;
359 static int fsck_tag(struct tag *tag)
361 struct object *tagged = tag->tagged;
363 if (verbose)
364 fprintf(stderr, "Checking tag %s\n",
365 sha1_to_hex(tag->object.sha1));
367 if (!tagged) {
368 return objerror(&tag->object, "could not load tagged object");
370 if (!show_tags)
371 return 0;
373 printf("tagged %s %s", typename(tagged->type), sha1_to_hex(tagged->sha1));
374 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
375 return 0;
378 static int fsck_sha1(const unsigned char *sha1)
380 struct object *obj = parse_object(sha1);
381 if (!obj) {
382 errors_found |= ERROR_OBJECT;
383 return error("%s: object corrupt or missing",
384 sha1_to_hex(sha1));
386 if (obj->flags & SEEN)
387 return 0;
388 obj->flags |= SEEN;
389 if (obj->type == OBJ_BLOB)
390 return 0;
391 if (obj->type == OBJ_TREE)
392 return fsck_tree((struct tree *) obj);
393 if (obj->type == OBJ_COMMIT)
394 return fsck_commit((struct commit *) obj);
395 if (obj->type == OBJ_TAG)
396 return fsck_tag((struct tag *) obj);
398 /* By now, parse_object() would've returned NULL instead. */
399 return objerror(obj, "unknown type '%d' (internal fsck error)",
400 obj->type);
404 * This is the sorting chunk size: make it reasonably
405 * big so that we can sort well..
407 #define MAX_SHA1_ENTRIES (1024)
409 struct sha1_entry {
410 unsigned long ino;
411 unsigned char sha1[20];
414 static struct {
415 unsigned long nr;
416 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
417 } sha1_list;
419 static int ino_compare(const void *_a, const void *_b)
421 const struct sha1_entry *a = _a, *b = _b;
422 unsigned long ino1 = a->ino, ino2 = b->ino;
423 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
426 static void fsck_sha1_list(void)
428 int i, nr = sha1_list.nr;
430 if (SORT_DIRENT)
431 qsort(sha1_list.entry, nr,
432 sizeof(struct sha1_entry *), ino_compare);
433 for (i = 0; i < nr; i++) {
434 struct sha1_entry *entry = sha1_list.entry[i];
435 unsigned char *sha1 = entry->sha1;
437 sha1_list.entry[i] = NULL;
438 fsck_sha1(sha1);
439 free(entry);
441 sha1_list.nr = 0;
444 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
446 struct sha1_entry *entry = xmalloc(sizeof(*entry));
447 int nr;
449 entry->ino = ino;
450 hashcpy(entry->sha1, sha1);
451 nr = sha1_list.nr;
452 if (nr == MAX_SHA1_ENTRIES) {
453 fsck_sha1_list();
454 nr = 0;
456 sha1_list.entry[nr] = entry;
457 sha1_list.nr = ++nr;
460 static void fsck_dir(int i, char *path)
462 DIR *dir = opendir(path);
463 struct dirent *de;
465 if (!dir)
466 return;
468 if (verbose)
469 fprintf(stderr, "Checking directory %s\n", path);
471 while ((de = readdir(dir)) != NULL) {
472 char name[100];
473 unsigned char sha1[20];
474 int len = strlen(de->d_name);
476 switch (len) {
477 case 2:
478 if (de->d_name[1] != '.')
479 break;
480 case 1:
481 if (de->d_name[0] != '.')
482 break;
483 continue;
484 case 38:
485 sprintf(name, "%02x", i);
486 memcpy(name+2, de->d_name, len+1);
487 if (get_sha1_hex(name, sha1) < 0)
488 break;
489 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
490 continue;
492 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
494 closedir(dir);
497 static int default_refs;
499 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
500 const char *email, unsigned long timestamp, int tz,
501 const char *message, void *cb_data)
503 struct object *obj;
505 if (verbose)
506 fprintf(stderr, "Checking reflog %s->%s\n",
507 sha1_to_hex(osha1), sha1_to_hex(nsha1));
509 if (!is_null_sha1(osha1)) {
510 obj = lookup_object(osha1);
511 if (obj) {
512 obj->used = 1;
513 mark_reachable(obj, REACHABLE);
516 obj = lookup_object(nsha1);
517 if (obj) {
518 obj->used = 1;
519 mark_reachable(obj, REACHABLE);
521 return 0;
524 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
526 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
527 return 0;
530 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
532 struct object *obj;
534 obj = lookup_object(sha1);
535 if (!obj) {
536 if (has_sha1_file(sha1)) {
537 default_refs++;
538 return 0; /* it is in a pack */
540 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
541 /* We'll continue with the rest despite the error.. */
542 return 0;
544 default_refs++;
545 obj->used = 1;
546 mark_reachable(obj, REACHABLE);
548 return 0;
551 static void get_default_heads(void)
553 for_each_ref(fsck_handle_ref, NULL);
554 if (include_reflogs)
555 for_each_reflog(fsck_handle_reflog, NULL);
558 * Not having any default heads isn't really fatal, but
559 * it does mean that "--unreachable" no longer makes any
560 * sense (since in this case everything will obviously
561 * be unreachable by definition.
563 * Showing dangling objects is valid, though (as those
564 * dangling objects are likely lost heads).
566 * So we just print a warning about it, and clear the
567 * "show_unreachable" flag.
569 if (!default_refs) {
570 fprintf(stderr, "notice: No default references\n");
571 show_unreachable = 0;
575 static void fsck_object_dir(const char *path)
577 int i;
579 if (verbose)
580 fprintf(stderr, "Checking object directory\n");
582 for (i = 0; i < 256; i++) {
583 static char dir[4096];
584 sprintf(dir, "%s/%02x", path, i);
585 fsck_dir(i, dir);
587 fsck_sha1_list();
590 static int fsck_head_link(void)
592 unsigned char sha1[20];
593 int flag;
594 int null_is_error = 0;
595 const char *head_points_at = resolve_ref("HEAD", sha1, 0, &flag);
597 if (verbose)
598 fprintf(stderr, "Checking HEAD link\n");
600 if (!head_points_at)
601 return error("Invalid HEAD");
602 if (!strcmp(head_points_at, "HEAD"))
603 /* detached HEAD */
604 null_is_error = 1;
605 else if (prefixcmp(head_points_at, "refs/heads/"))
606 return error("HEAD points to something strange (%s)",
607 head_points_at);
608 if (is_null_sha1(sha1)) {
609 if (null_is_error)
610 return error("HEAD: detached HEAD points at nothing");
611 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
612 head_points_at + 11);
614 return 0;
617 static int fsck_cache_tree(struct cache_tree *it)
619 int i;
620 int err = 0;
622 if (verbose)
623 fprintf(stderr, "Checking cache tree\n");
625 if (0 <= it->entry_count) {
626 struct object *obj = parse_object(it->sha1);
627 if (!obj) {
628 error("%s: invalid sha1 pointer in cache-tree",
629 sha1_to_hex(it->sha1));
630 return 1;
632 mark_reachable(obj, REACHABLE);
633 obj->used = 1;
634 if (obj->type != OBJ_TREE)
635 err |= objerror(obj, "non-tree in cache-tree");
637 for (i = 0; i < it->subtree_nr; i++)
638 err |= fsck_cache_tree(it->down[i]->cache_tree);
639 return err;
642 static const char fsck_usage[] =
643 "git-fsck [--tags] [--root] [[--unreachable] [--cache] [--full] "
644 "[--strict] [--verbose] <head-sha1>*]";
646 int cmd_fsck(int argc, char **argv, const char *prefix)
648 int i, heads;
650 track_object_refs = 1;
651 errors_found = 0;
653 for (i = 1; i < argc; i++) {
654 const char *arg = argv[i];
656 if (!strcmp(arg, "--unreachable")) {
657 show_unreachable = 1;
658 continue;
660 if (!strcmp(arg, "--tags")) {
661 show_tags = 1;
662 continue;
664 if (!strcmp(arg, "--root")) {
665 show_root = 1;
666 continue;
668 if (!strcmp(arg, "--cache")) {
669 keep_cache_objects = 1;
670 continue;
672 if (!strcmp(arg, "--no-reflogs")) {
673 include_reflogs = 0;
674 continue;
676 if (!strcmp(arg, "--full")) {
677 check_full = 1;
678 continue;
680 if (!strcmp(arg, "--strict")) {
681 check_strict = 1;
682 continue;
684 if (!strcmp(arg, "--verbose")) {
685 verbose = 1;
686 continue;
688 if (*arg == '-')
689 usage(fsck_usage);
692 fsck_head_link();
693 fsck_object_dir(get_object_directory());
694 if (check_full) {
695 struct alternate_object_database *alt;
696 struct packed_git *p;
697 prepare_alt_odb();
698 for (alt = alt_odb_list; alt; alt = alt->next) {
699 char namebuf[PATH_MAX];
700 int namelen = alt->name - alt->base;
701 memcpy(namebuf, alt->base, namelen);
702 namebuf[namelen - 1] = 0;
703 fsck_object_dir(namebuf);
705 prepare_packed_git();
706 for (p = packed_git; p; p = p->next)
707 /* verify gives error messages itself */
708 verify_pack(p, 0);
710 for (p = packed_git; p; p = p->next) {
711 uint32_t i, num;
712 if (open_pack_index(p))
713 continue;
714 num = p->num_objects;
715 for (i = 0; i < num; i++)
716 fsck_sha1(nth_packed_object_sha1(p, i));
720 heads = 0;
721 for (i = 1; i < argc; i++) {
722 const char *arg = argv[i];
724 if (*arg == '-')
725 continue;
727 if (!get_sha1(arg, head_sha1)) {
728 struct object *obj = lookup_object(head_sha1);
730 /* Error is printed by lookup_object(). */
731 if (!obj)
732 continue;
734 obj->used = 1;
735 mark_reachable(obj, REACHABLE);
736 heads++;
737 continue;
739 error("invalid parameter: expected sha1, got '%s'", arg);
743 * If we've not been given any explicit head information, do the
744 * default ones from .git/refs. We also consider the index file
745 * in this case (ie this implies --cache).
747 if (!heads) {
748 get_default_heads();
749 keep_cache_objects = 1;
752 if (keep_cache_objects) {
753 int i;
754 read_cache();
755 for (i = 0; i < active_nr; i++) {
756 unsigned int mode;
757 struct blob *blob;
758 struct object *obj;
760 mode = ntohl(active_cache[i]->ce_mode);
761 if (S_ISGITLINK(mode))
762 continue;
763 blob = lookup_blob(active_cache[i]->sha1);
764 if (!blob)
765 continue;
766 obj = &blob->object;
767 obj->used = 1;
768 mark_reachable(obj, REACHABLE);
770 if (active_cache_tree)
771 fsck_cache_tree(active_cache_tree);
774 check_connectivity();
775 return errors_found;