Merge branch 'maint'
[git/builtin-gsoc.git] / builtin-fsck.c
blob350ec5e1445a3743a4b9a5039d16d9dba505cf4e
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"
12 #define REACHABLE 0x0001
13 #define SEEN 0x0002
15 static int show_root;
16 static int show_tags;
17 static int show_unreachable;
18 static int include_reflogs = 1;
19 static int check_full;
20 static int check_strict;
21 static int keep_cache_objects;
22 static unsigned char head_sha1[20];
23 static int errors_found;
24 static int write_lost_and_found;
25 static int verbose;
26 #define ERROR_OBJECT 01
27 #define ERROR_REACHABLE 02
29 #ifdef NO_D_INO_IN_DIRENT
30 #define SORT_DIRENT 0
31 #define DIRENT_SORT_HINT(de) 0
32 #else
33 #define SORT_DIRENT 1
34 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
35 #endif
37 static void objreport(struct object *obj, const char *severity,
38 const char *err, va_list params)
40 fprintf(stderr, "%s in %s %s: ",
41 severity, typename(obj->type), sha1_to_hex(obj->sha1));
42 vfprintf(stderr, err, params);
43 fputs("\n", stderr);
46 static int objerror(struct object *obj, const char *err, ...)
48 va_list params;
49 va_start(params, err);
50 errors_found |= ERROR_OBJECT;
51 objreport(obj, "error", err, params);
52 va_end(params);
53 return -1;
56 static int objwarning(struct object *obj, const char *err, ...)
58 va_list params;
59 va_start(params, err);
60 objreport(obj, "warning", err, params);
61 va_end(params);
62 return -1;
66 * Check a single reachable object
68 static void check_reachable_object(struct object *obj)
70 const struct object_refs *refs;
73 * We obviously want the object to be parsed,
74 * except if it was in a pack-file and we didn't
75 * do a full fsck
77 if (!obj->parsed) {
78 if (has_sha1_pack(obj->sha1, NULL))
79 return; /* it is in pack - forget about it */
80 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
81 errors_found |= ERROR_REACHABLE;
82 return;
86 * Check that everything that we try to reference is also good.
88 refs = lookup_object_refs(obj);
89 if (refs) {
90 unsigned j;
91 for (j = 0; j < refs->count; j++) {
92 struct object *ref = refs->ref[j];
93 if (ref->parsed ||
94 (has_sha1_file(ref->sha1)))
95 continue;
96 printf("broken link from %7s %s\n",
97 typename(obj->type), sha1_to_hex(obj->sha1));
98 printf(" to %7s %s\n",
99 typename(ref->type), sha1_to_hex(ref->sha1));
100 errors_found |= ERROR_REACHABLE;
106 * Check a single unreachable object
108 static void check_unreachable_object(struct object *obj)
111 * Missing unreachable object? Ignore it. It's not like
112 * we miss it (since it can't be reached), nor do we want
113 * to complain about it being unreachable (since it does
114 * not exist).
116 if (!obj->parsed)
117 return;
120 * Unreachable object that exists? Show it if asked to,
121 * since this is something that is prunable.
123 if (show_unreachable) {
124 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
125 return;
129 * "!used" means that nothing at all points to it, including
130 * other unreachable objects. In other words, it's the "tip"
131 * of some set of unreachable objects, usually a commit that
132 * got dropped.
134 * Such starting points are more interesting than some random
135 * set of unreachable objects, so we show them even if the user
136 * hasn't asked for _all_ unreachable objects. If you have
137 * deleted a branch by mistake, this is a prime candidate to
138 * start looking at, for example.
140 if (!obj->used) {
141 printf("dangling %s %s\n", typename(obj->type),
142 sha1_to_hex(obj->sha1));
143 if (write_lost_and_found) {
144 char *filename = git_path("lost-found/%s/%s",
145 obj->type == OBJ_COMMIT ? "commit" : "other",
146 sha1_to_hex(obj->sha1));
147 FILE *f;
149 if (safe_create_leading_directories(filename)) {
150 error("Could not create lost-found");
151 return;
153 if (!(f = fopen(filename, "w")))
154 die("Could not open %s", filename);
155 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
156 fclose(f);
158 return;
162 * Otherwise? It's there, it's unreachable, and some other unreachable
163 * object points to it. Ignore it - it's not interesting, and we showed
164 * all the interesting cases above.
168 static void check_object(struct object *obj)
170 if (verbose)
171 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
173 if (obj->flags & REACHABLE)
174 check_reachable_object(obj);
175 else
176 check_unreachable_object(obj);
179 static void check_connectivity(void)
181 int i, max;
183 /* Look up all the requirements, warn about missing objects.. */
184 max = get_max_object_index();
185 if (verbose)
186 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
188 for (i = 0; i < max; i++) {
189 struct object *obj = get_indexed_object(i);
191 if (obj)
192 check_object(obj);
197 * The entries in a tree are ordered in the _path_ order,
198 * which means that a directory entry is ordered by adding
199 * a slash to the end of it.
201 * So a directory called "a" is ordered _after_ a file
202 * called "a.c", because "a/" sorts after "a.c".
204 #define TREE_UNORDERED (-1)
205 #define TREE_HAS_DUPS (-2)
207 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
209 int len1 = strlen(name1);
210 int len2 = strlen(name2);
211 int len = len1 < len2 ? len1 : len2;
212 unsigned char c1, c2;
213 int cmp;
215 cmp = memcmp(name1, name2, len);
216 if (cmp < 0)
217 return 0;
218 if (cmp > 0)
219 return TREE_UNORDERED;
222 * Ok, the first <len> characters are the same.
223 * Now we need to order the next one, but turn
224 * a '\0' into a '/' for a directory entry.
226 c1 = name1[len];
227 c2 = name2[len];
228 if (!c1 && !c2)
230 * git-write-tree used to write out a nonsense tree that has
231 * entries with the same name, one blob and one tree. Make
232 * sure we do not have duplicate entries.
234 return TREE_HAS_DUPS;
235 if (!c1 && S_ISDIR(mode1))
236 c1 = '/';
237 if (!c2 && S_ISDIR(mode2))
238 c2 = '/';
239 return c1 < c2 ? 0 : TREE_UNORDERED;
242 static int fsck_tree(struct tree *item)
244 int retval;
245 int has_full_path = 0;
246 int has_empty_name = 0;
247 int has_zero_pad = 0;
248 int has_bad_modes = 0;
249 int has_dup_entries = 0;
250 int not_properly_sorted = 0;
251 struct tree_desc desc;
252 unsigned o_mode;
253 const char *o_name;
254 const unsigned char *o_sha1;
256 if (verbose)
257 fprintf(stderr, "Checking tree %s\n",
258 sha1_to_hex(item->object.sha1));
260 init_tree_desc(&desc, item->buffer, item->size);
262 o_mode = 0;
263 o_name = NULL;
264 o_sha1 = NULL;
265 while (desc.size) {
266 unsigned mode;
267 const char *name;
268 const unsigned char *sha1;
270 sha1 = tree_entry_extract(&desc, &name, &mode);
272 if (strchr(name, '/'))
273 has_full_path = 1;
274 if (!*name)
275 has_empty_name = 1;
276 has_zero_pad |= *(char *)desc.buffer == '0';
277 update_tree_entry(&desc);
279 switch (mode) {
281 * Standard modes..
283 case S_IFREG | 0755:
284 case S_IFREG | 0644:
285 case S_IFLNK:
286 case S_IFDIR:
287 case S_IFGITLINK:
288 break;
290 * This is nonstandard, but we had a few of these
291 * early on when we honored the full set of mode
292 * bits..
294 case S_IFREG | 0664:
295 if (!check_strict)
296 break;
297 default:
298 has_bad_modes = 1;
301 if (o_name) {
302 switch (verify_ordered(o_mode, o_name, mode, name)) {
303 case TREE_UNORDERED:
304 not_properly_sorted = 1;
305 break;
306 case TREE_HAS_DUPS:
307 has_dup_entries = 1;
308 break;
309 default:
310 break;
314 o_mode = mode;
315 o_name = name;
316 o_sha1 = sha1;
318 free(item->buffer);
319 item->buffer = NULL;
321 retval = 0;
322 if (has_full_path) {
323 objwarning(&item->object, "contains full pathnames");
325 if (has_empty_name) {
326 objwarning(&item->object, "contains empty pathname");
328 if (has_zero_pad) {
329 objwarning(&item->object, "contains zero-padded file modes");
331 if (has_bad_modes) {
332 objwarning(&item->object, "contains bad file modes");
334 if (has_dup_entries) {
335 retval = objerror(&item->object, "contains duplicate file entries");
337 if (not_properly_sorted) {
338 retval = objerror(&item->object, "not properly sorted");
340 return retval;
343 static int fsck_commit(struct commit *commit)
345 char *buffer = commit->buffer;
346 unsigned char tree_sha1[20], sha1[20];
348 if (verbose)
349 fprintf(stderr, "Checking commit %s\n",
350 sha1_to_hex(commit->object.sha1));
352 if (memcmp(buffer, "tree ", 5))
353 return objerror(&commit->object, "invalid format - expected 'tree' line");
354 if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
355 return objerror(&commit->object, "invalid 'tree' line format - bad sha1");
356 buffer += 46;
357 while (!memcmp(buffer, "parent ", 7)) {
358 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
359 return objerror(&commit->object, "invalid 'parent' line format - bad sha1");
360 buffer += 48;
362 if (memcmp(buffer, "author ", 7))
363 return objerror(&commit->object, "invalid format - expected 'author' line");
364 free(commit->buffer);
365 commit->buffer = NULL;
366 if (!commit->tree)
367 return objerror(&commit->object, "could not load commit's tree %s", tree_sha1);
368 if (!commit->parents && show_root)
369 printf("root %s\n", sha1_to_hex(commit->object.sha1));
370 if (!commit->date)
371 printf("bad commit date in %s\n",
372 sha1_to_hex(commit->object.sha1));
373 return 0;
376 static int fsck_tag(struct tag *tag)
378 struct object *tagged = tag->tagged;
380 if (verbose)
381 fprintf(stderr, "Checking tag %s\n",
382 sha1_to_hex(tag->object.sha1));
384 if (!tagged) {
385 return objerror(&tag->object, "could not load tagged object");
387 if (!show_tags)
388 return 0;
390 printf("tagged %s %s", typename(tagged->type), sha1_to_hex(tagged->sha1));
391 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
392 return 0;
395 static int fsck_sha1(const unsigned char *sha1)
397 struct object *obj = parse_object(sha1);
398 if (!obj) {
399 errors_found |= ERROR_OBJECT;
400 return error("%s: object corrupt or missing",
401 sha1_to_hex(sha1));
403 if (obj->flags & SEEN)
404 return 0;
405 obj->flags |= SEEN;
406 if (obj->type == OBJ_BLOB)
407 return 0;
408 if (obj->type == OBJ_TREE)
409 return fsck_tree((struct tree *) obj);
410 if (obj->type == OBJ_COMMIT)
411 return fsck_commit((struct commit *) obj);
412 if (obj->type == OBJ_TAG)
413 return fsck_tag((struct tag *) obj);
415 /* By now, parse_object() would've returned NULL instead. */
416 return objerror(obj, "unknown type '%d' (internal fsck error)",
417 obj->type);
421 * This is the sorting chunk size: make it reasonably
422 * big so that we can sort well..
424 #define MAX_SHA1_ENTRIES (1024)
426 struct sha1_entry {
427 unsigned long ino;
428 unsigned char sha1[20];
431 static struct {
432 unsigned long nr;
433 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
434 } sha1_list;
436 static int ino_compare(const void *_a, const void *_b)
438 const struct sha1_entry *a = _a, *b = _b;
439 unsigned long ino1 = a->ino, ino2 = b->ino;
440 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
443 static void fsck_sha1_list(void)
445 int i, nr = sha1_list.nr;
447 if (SORT_DIRENT)
448 qsort(sha1_list.entry, nr,
449 sizeof(struct sha1_entry *), ino_compare);
450 for (i = 0; i < nr; i++) {
451 struct sha1_entry *entry = sha1_list.entry[i];
452 unsigned char *sha1 = entry->sha1;
454 sha1_list.entry[i] = NULL;
455 fsck_sha1(sha1);
456 free(entry);
458 sha1_list.nr = 0;
461 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
463 struct sha1_entry *entry = xmalloc(sizeof(*entry));
464 int nr;
466 entry->ino = ino;
467 hashcpy(entry->sha1, sha1);
468 nr = sha1_list.nr;
469 if (nr == MAX_SHA1_ENTRIES) {
470 fsck_sha1_list();
471 nr = 0;
473 sha1_list.entry[nr] = entry;
474 sha1_list.nr = ++nr;
477 static void fsck_dir(int i, char *path)
479 DIR *dir = opendir(path);
480 struct dirent *de;
482 if (!dir)
483 return;
485 if (verbose)
486 fprintf(stderr, "Checking directory %s\n", path);
488 while ((de = readdir(dir)) != NULL) {
489 char name[100];
490 unsigned char sha1[20];
491 int len = strlen(de->d_name);
493 switch (len) {
494 case 2:
495 if (de->d_name[1] != '.')
496 break;
497 case 1:
498 if (de->d_name[0] != '.')
499 break;
500 continue;
501 case 38:
502 sprintf(name, "%02x", i);
503 memcpy(name+2, de->d_name, len+1);
504 if (get_sha1_hex(name, sha1) < 0)
505 break;
506 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
507 continue;
509 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
511 closedir(dir);
514 static int default_refs;
516 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
517 const char *email, unsigned long timestamp, int tz,
518 const char *message, void *cb_data)
520 struct object *obj;
522 if (verbose)
523 fprintf(stderr, "Checking reflog %s->%s\n",
524 sha1_to_hex(osha1), sha1_to_hex(nsha1));
526 if (!is_null_sha1(osha1)) {
527 obj = lookup_object(osha1);
528 if (obj) {
529 obj->used = 1;
530 mark_reachable(obj, REACHABLE);
533 obj = lookup_object(nsha1);
534 if (obj) {
535 obj->used = 1;
536 mark_reachable(obj, REACHABLE);
538 return 0;
541 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
543 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
544 return 0;
547 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
549 struct object *obj;
551 obj = lookup_object(sha1);
552 if (!obj) {
553 if (has_sha1_file(sha1)) {
554 default_refs++;
555 return 0; /* it is in a pack */
557 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
558 /* We'll continue with the rest despite the error.. */
559 return 0;
561 default_refs++;
562 obj->used = 1;
563 mark_reachable(obj, REACHABLE);
565 return 0;
568 static void get_default_heads(void)
570 for_each_ref(fsck_handle_ref, NULL);
571 if (include_reflogs)
572 for_each_reflog(fsck_handle_reflog, NULL);
575 * Not having any default heads isn't really fatal, but
576 * it does mean that "--unreachable" no longer makes any
577 * sense (since in this case everything will obviously
578 * be unreachable by definition.
580 * Showing dangling objects is valid, though (as those
581 * dangling objects are likely lost heads).
583 * So we just print a warning about it, and clear the
584 * "show_unreachable" flag.
586 if (!default_refs) {
587 fprintf(stderr, "notice: No default references\n");
588 show_unreachable = 0;
592 static void fsck_object_dir(const char *path)
594 int i;
596 if (verbose)
597 fprintf(stderr, "Checking object directory\n");
599 for (i = 0; i < 256; i++) {
600 static char dir[4096];
601 sprintf(dir, "%s/%02x", path, i);
602 fsck_dir(i, dir);
604 fsck_sha1_list();
607 static int fsck_head_link(void)
609 unsigned char sha1[20];
610 int flag;
611 int null_is_error = 0;
612 const char *head_points_at = resolve_ref("HEAD", sha1, 0, &flag);
614 if (verbose)
615 fprintf(stderr, "Checking HEAD link\n");
617 if (!head_points_at)
618 return error("Invalid HEAD");
619 if (!strcmp(head_points_at, "HEAD"))
620 /* detached HEAD */
621 null_is_error = 1;
622 else if (prefixcmp(head_points_at, "refs/heads/"))
623 return error("HEAD points to something strange (%s)",
624 head_points_at);
625 if (is_null_sha1(sha1)) {
626 if (null_is_error)
627 return error("HEAD: detached HEAD points at nothing");
628 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
629 head_points_at + 11);
631 return 0;
634 static int fsck_cache_tree(struct cache_tree *it)
636 int i;
637 int err = 0;
639 if (verbose)
640 fprintf(stderr, "Checking cache tree\n");
642 if (0 <= it->entry_count) {
643 struct object *obj = parse_object(it->sha1);
644 if (!obj) {
645 error("%s: invalid sha1 pointer in cache-tree",
646 sha1_to_hex(it->sha1));
647 return 1;
649 mark_reachable(obj, REACHABLE);
650 obj->used = 1;
651 if (obj->type != OBJ_TREE)
652 err |= objerror(obj, "non-tree in cache-tree");
654 for (i = 0; i < it->subtree_nr; i++)
655 err |= fsck_cache_tree(it->down[i]->cache_tree);
656 return err;
659 static const char fsck_usage[] =
660 "git-fsck [--tags] [--root] [[--unreachable] [--cache] [--full] "
661 "[--strict] [--verbose] <head-sha1>*]";
663 int cmd_fsck(int argc, const char **argv, const char *prefix)
665 int i, heads;
667 track_object_refs = 1;
668 errors_found = 0;
670 for (i = 1; i < argc; i++) {
671 const char *arg = argv[i];
673 if (!strcmp(arg, "--unreachable")) {
674 show_unreachable = 1;
675 continue;
677 if (!strcmp(arg, "--tags")) {
678 show_tags = 1;
679 continue;
681 if (!strcmp(arg, "--root")) {
682 show_root = 1;
683 continue;
685 if (!strcmp(arg, "--cache")) {
686 keep_cache_objects = 1;
687 continue;
689 if (!strcmp(arg, "--no-reflogs")) {
690 include_reflogs = 0;
691 continue;
693 if (!strcmp(arg, "--full")) {
694 check_full = 1;
695 continue;
697 if (!strcmp(arg, "--strict")) {
698 check_strict = 1;
699 continue;
701 if (!strcmp(arg, "--verbose")) {
702 verbose = 1;
703 continue;
705 if (!strcmp(arg, "--lost-found")) {
706 check_full = 1;
707 include_reflogs = 0;
708 write_lost_and_found = 1;
709 continue;
711 if (*arg == '-')
712 usage(fsck_usage);
715 fsck_head_link();
716 fsck_object_dir(get_object_directory());
717 if (check_full) {
718 struct alternate_object_database *alt;
719 struct packed_git *p;
720 prepare_alt_odb();
721 for (alt = alt_odb_list; alt; alt = alt->next) {
722 char namebuf[PATH_MAX];
723 int namelen = alt->name - alt->base;
724 memcpy(namebuf, alt->base, namelen);
725 namebuf[namelen - 1] = 0;
726 fsck_object_dir(namebuf);
728 prepare_packed_git();
729 for (p = packed_git; p; p = p->next)
730 /* verify gives error messages itself */
731 verify_pack(p, 0);
733 for (p = packed_git; p; p = p->next) {
734 uint32_t i, num;
735 if (open_pack_index(p))
736 continue;
737 num = p->num_objects;
738 for (i = 0; i < num; i++)
739 fsck_sha1(nth_packed_object_sha1(p, i));
743 heads = 0;
744 for (i = 1; i < argc; i++) {
745 const char *arg = argv[i];
747 if (*arg == '-')
748 continue;
750 if (!get_sha1(arg, head_sha1)) {
751 struct object *obj = lookup_object(head_sha1);
753 /* Error is printed by lookup_object(). */
754 if (!obj)
755 continue;
757 obj->used = 1;
758 mark_reachable(obj, REACHABLE);
759 heads++;
760 continue;
762 error("invalid parameter: expected sha1, got '%s'", arg);
766 * If we've not been given any explicit head information, do the
767 * default ones from .git/refs. We also consider the index file
768 * in this case (ie this implies --cache).
770 if (!heads) {
771 get_default_heads();
772 keep_cache_objects = 1;
775 if (keep_cache_objects) {
776 int i;
777 read_cache();
778 for (i = 0; i < active_nr; i++) {
779 unsigned int mode;
780 struct blob *blob;
781 struct object *obj;
783 mode = ntohl(active_cache[i]->ce_mode);
784 if (S_ISGITLINK(mode))
785 continue;
786 blob = lookup_blob(active_cache[i]->sha1);
787 if (!blob)
788 continue;
789 obj = &blob->object;
790 obj->used = 1;
791 mark_reachable(obj, REACHABLE);
793 if (active_cache_tree)
794 fsck_cache_tree(active_cache_tree);
797 check_connectivity();
798 return errors_found;