diff: honor binariness specified in attributes
[git/jnareb-git.git] / builtin-fsck.c
bloba6ef65ea321c614c7616862a44ce0af7998e5749
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 write_lost_and_found;
24 static int verbose;
25 #define ERROR_OBJECT 01
26 #define ERROR_REACHABLE 02
28 #ifdef NO_D_INO_IN_DIRENT
29 #define SORT_DIRENT 0
30 #define DIRENT_SORT_HINT(de) 0
31 #else
32 #define SORT_DIRENT 1
33 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
34 #endif
36 static void objreport(struct object *obj, const char *severity,
37 const char *err, va_list params)
39 fprintf(stderr, "%s in %s %s: ",
40 severity, typename(obj->type), sha1_to_hex(obj->sha1));
41 vfprintf(stderr, err, params);
42 fputs("\n", stderr);
45 static int objerror(struct object *obj, const char *err, ...)
47 va_list params;
48 va_start(params, err);
49 errors_found |= ERROR_OBJECT;
50 objreport(obj, "error", err, params);
51 va_end(params);
52 return -1;
55 static int objwarning(struct object *obj, const char *err, ...)
57 va_list params;
58 va_start(params, err);
59 objreport(obj, "warning", err, params);
60 va_end(params);
61 return -1;
65 * Check a single reachable object
67 static void check_reachable_object(struct object *obj)
69 const struct object_refs *refs;
72 * We obviously want the object to be parsed,
73 * except if it was in a pack-file and we didn't
74 * do a full fsck
76 if (!obj->parsed) {
77 if (has_sha1_pack(obj->sha1, NULL))
78 return; /* it is in pack - forget about it */
79 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
80 errors_found |= ERROR_REACHABLE;
81 return;
85 * Check that everything that we try to reference is also good.
87 refs = lookup_object_refs(obj);
88 if (refs) {
89 unsigned j;
90 for (j = 0; j < refs->count; j++) {
91 struct object *ref = refs->ref[j];
92 if (ref->parsed ||
93 (has_sha1_file(ref->sha1)))
94 continue;
95 printf("broken link from %7s %s\n",
96 typename(obj->type), sha1_to_hex(obj->sha1));
97 printf(" to %7s %s\n",
98 typename(ref->type), sha1_to_hex(ref->sha1));
99 errors_found |= ERROR_REACHABLE;
105 * Check a single unreachable object
107 static void check_unreachable_object(struct object *obj)
110 * Missing unreachable object? Ignore it. It's not like
111 * we miss it (since it can't be reached), nor do we want
112 * to complain about it being unreachable (since it does
113 * not exist).
115 if (!obj->parsed)
116 return;
119 * Unreachable object that exists? Show it if asked to,
120 * since this is something that is prunable.
122 if (show_unreachable) {
123 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
124 return;
128 * "!used" means that nothing at all points to it, including
129 * other unreachable objects. In other words, it's the "tip"
130 * of some set of unreachable objects, usually a commit that
131 * got dropped.
133 * Such starting points are more interesting than some random
134 * set of unreachable objects, so we show them even if the user
135 * hasn't asked for _all_ unreachable objects. If you have
136 * deleted a branch by mistake, this is a prime candidate to
137 * start looking at, for example.
139 if (!obj->used) {
140 printf("dangling %s %s\n", typename(obj->type),
141 sha1_to_hex(obj->sha1));
142 if (write_lost_and_found) {
143 char *filename = git_path("lost-found/%s/%s",
144 obj->type == OBJ_COMMIT ? "commit" : "other",
145 sha1_to_hex(obj->sha1));
146 FILE *f;
148 if (safe_create_leading_directories(filename)) {
149 error("Could not create lost-found");
150 return;
152 if (!(f = fopen(filename, "w")))
153 die("Could not open %s", filename);
154 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
155 fclose(f);
157 return;
161 * Otherwise? It's there, it's unreachable, and some other unreachable
162 * object points to it. Ignore it - it's not interesting, and we showed
163 * all the interesting cases above.
167 static void check_object(struct object *obj)
169 if (verbose)
170 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
172 if (obj->flags & REACHABLE)
173 check_reachable_object(obj);
174 else
175 check_unreachable_object(obj);
178 static void check_connectivity(void)
180 int i, max;
182 /* Look up all the requirements, warn about missing objects.. */
183 max = get_max_object_index();
184 if (verbose)
185 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
187 for (i = 0; i < max; i++) {
188 struct object *obj = get_indexed_object(i);
190 if (obj)
191 check_object(obj);
196 * The entries in a tree are ordered in the _path_ order,
197 * which means that a directory entry is ordered by adding
198 * a slash to the end of it.
200 * So a directory called "a" is ordered _after_ a file
201 * called "a.c", because "a/" sorts after "a.c".
203 #define TREE_UNORDERED (-1)
204 #define TREE_HAS_DUPS (-2)
206 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
208 int len1 = strlen(name1);
209 int len2 = strlen(name2);
210 int len = len1 < len2 ? len1 : len2;
211 unsigned char c1, c2;
212 int cmp;
214 cmp = memcmp(name1, name2, len);
215 if (cmp < 0)
216 return 0;
217 if (cmp > 0)
218 return TREE_UNORDERED;
221 * Ok, the first <len> characters are the same.
222 * Now we need to order the next one, but turn
223 * a '\0' into a '/' for a directory entry.
225 c1 = name1[len];
226 c2 = name2[len];
227 if (!c1 && !c2)
229 * git-write-tree used to write out a nonsense tree that has
230 * entries with the same name, one blob and one tree. Make
231 * sure we do not have duplicate entries.
233 return TREE_HAS_DUPS;
234 if (!c1 && S_ISDIR(mode1))
235 c1 = '/';
236 if (!c2 && S_ISDIR(mode2))
237 c2 = '/';
238 return c1 < c2 ? 0 : TREE_UNORDERED;
241 static int fsck_tree(struct tree *item)
243 int retval;
244 int has_full_path = 0;
245 int has_empty_name = 0;
246 int has_zero_pad = 0;
247 int has_bad_modes = 0;
248 int has_dup_entries = 0;
249 int not_properly_sorted = 0;
250 struct tree_desc desc;
251 unsigned o_mode;
252 const char *o_name;
253 const unsigned char *o_sha1;
255 if (verbose)
256 fprintf(stderr, "Checking tree %s\n",
257 sha1_to_hex(item->object.sha1));
259 init_tree_desc(&desc, item->buffer, item->size);
261 o_mode = 0;
262 o_name = NULL;
263 o_sha1 = NULL;
264 while (desc.size) {
265 unsigned mode;
266 const char *name;
267 const unsigned char *sha1;
269 sha1 = tree_entry_extract(&desc, &name, &mode);
271 if (strchr(name, '/'))
272 has_full_path = 1;
273 if (!*name)
274 has_empty_name = 1;
275 has_zero_pad |= *(char *)desc.buffer == '0';
276 update_tree_entry(&desc);
278 switch (mode) {
280 * Standard modes..
282 case S_IFREG | 0755:
283 case S_IFREG | 0644:
284 case S_IFLNK:
285 case S_IFDIR:
286 case S_IFGITLINK:
287 break;
289 * This is nonstandard, but we had a few of these
290 * early on when we honored the full set of mode
291 * bits..
293 case S_IFREG | 0664:
294 if (!check_strict)
295 break;
296 default:
297 has_bad_modes = 1;
300 if (o_name) {
301 switch (verify_ordered(o_mode, o_name, mode, name)) {
302 case TREE_UNORDERED:
303 not_properly_sorted = 1;
304 break;
305 case TREE_HAS_DUPS:
306 has_dup_entries = 1;
307 break;
308 default:
309 break;
313 o_mode = mode;
314 o_name = name;
315 o_sha1 = sha1;
317 free(item->buffer);
318 item->buffer = NULL;
320 retval = 0;
321 if (has_full_path) {
322 objwarning(&item->object, "contains full pathnames");
324 if (has_empty_name) {
325 objwarning(&item->object, "contains empty pathname");
327 if (has_zero_pad) {
328 objwarning(&item->object, "contains zero-padded file modes");
330 if (has_bad_modes) {
331 objwarning(&item->object, "contains bad file modes");
333 if (has_dup_entries) {
334 retval = objerror(&item->object, "contains duplicate file entries");
336 if (not_properly_sorted) {
337 retval = objerror(&item->object, "not properly sorted");
339 return retval;
342 static int fsck_commit(struct commit *commit)
344 char *buffer = commit->buffer;
345 unsigned char tree_sha1[20], sha1[20];
347 if (verbose)
348 fprintf(stderr, "Checking commit %s\n",
349 sha1_to_hex(commit->object.sha1));
351 if (memcmp(buffer, "tree ", 5))
352 return objerror(&commit->object, "invalid format - expected 'tree' line");
353 if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
354 return objerror(&commit->object, "invalid 'tree' line format - bad sha1");
355 buffer += 46;
356 while (!memcmp(buffer, "parent ", 7)) {
357 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
358 return objerror(&commit->object, "invalid 'parent' line format - bad sha1");
359 buffer += 48;
361 if (memcmp(buffer, "author ", 7))
362 return objerror(&commit->object, "invalid format - expected 'author' line");
363 free(commit->buffer);
364 commit->buffer = NULL;
365 if (!commit->tree)
366 return objerror(&commit->object, "could not load commit's tree %s", tree_sha1);
367 if (!commit->parents && show_root)
368 printf("root %s\n", sha1_to_hex(commit->object.sha1));
369 if (!commit->date)
370 printf("bad commit date in %s\n",
371 sha1_to_hex(commit->object.sha1));
372 return 0;
375 static int fsck_tag(struct tag *tag)
377 struct object *tagged = tag->tagged;
379 if (verbose)
380 fprintf(stderr, "Checking tag %s\n",
381 sha1_to_hex(tag->object.sha1));
383 if (!tagged) {
384 return objerror(&tag->object, "could not load tagged object");
386 if (!show_tags)
387 return 0;
389 printf("tagged %s %s", typename(tagged->type), sha1_to_hex(tagged->sha1));
390 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
391 return 0;
394 static int fsck_sha1(const unsigned char *sha1)
396 struct object *obj = parse_object(sha1);
397 if (!obj) {
398 errors_found |= ERROR_OBJECT;
399 return error("%s: object corrupt or missing",
400 sha1_to_hex(sha1));
402 if (obj->flags & SEEN)
403 return 0;
404 obj->flags |= SEEN;
405 if (obj->type == OBJ_BLOB)
406 return 0;
407 if (obj->type == OBJ_TREE)
408 return fsck_tree((struct tree *) obj);
409 if (obj->type == OBJ_COMMIT)
410 return fsck_commit((struct commit *) obj);
411 if (obj->type == OBJ_TAG)
412 return fsck_tag((struct tag *) obj);
414 /* By now, parse_object() would've returned NULL instead. */
415 return objerror(obj, "unknown type '%d' (internal fsck error)",
416 obj->type);
420 * This is the sorting chunk size: make it reasonably
421 * big so that we can sort well..
423 #define MAX_SHA1_ENTRIES (1024)
425 struct sha1_entry {
426 unsigned long ino;
427 unsigned char sha1[20];
430 static struct {
431 unsigned long nr;
432 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
433 } sha1_list;
435 static int ino_compare(const void *_a, const void *_b)
437 const struct sha1_entry *a = _a, *b = _b;
438 unsigned long ino1 = a->ino, ino2 = b->ino;
439 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
442 static void fsck_sha1_list(void)
444 int i, nr = sha1_list.nr;
446 if (SORT_DIRENT)
447 qsort(sha1_list.entry, nr,
448 sizeof(struct sha1_entry *), ino_compare);
449 for (i = 0; i < nr; i++) {
450 struct sha1_entry *entry = sha1_list.entry[i];
451 unsigned char *sha1 = entry->sha1;
453 sha1_list.entry[i] = NULL;
454 fsck_sha1(sha1);
455 free(entry);
457 sha1_list.nr = 0;
460 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
462 struct sha1_entry *entry = xmalloc(sizeof(*entry));
463 int nr;
465 entry->ino = ino;
466 hashcpy(entry->sha1, sha1);
467 nr = sha1_list.nr;
468 if (nr == MAX_SHA1_ENTRIES) {
469 fsck_sha1_list();
470 nr = 0;
472 sha1_list.entry[nr] = entry;
473 sha1_list.nr = ++nr;
476 static void fsck_dir(int i, char *path)
478 DIR *dir = opendir(path);
479 struct dirent *de;
481 if (!dir)
482 return;
484 if (verbose)
485 fprintf(stderr, "Checking directory %s\n", path);
487 while ((de = readdir(dir)) != NULL) {
488 char name[100];
489 unsigned char sha1[20];
490 int len = strlen(de->d_name);
492 switch (len) {
493 case 2:
494 if (de->d_name[1] != '.')
495 break;
496 case 1:
497 if (de->d_name[0] != '.')
498 break;
499 continue;
500 case 38:
501 sprintf(name, "%02x", i);
502 memcpy(name+2, de->d_name, len+1);
503 if (get_sha1_hex(name, sha1) < 0)
504 break;
505 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
506 continue;
508 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
510 closedir(dir);
513 static int default_refs;
515 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
516 const char *email, unsigned long timestamp, int tz,
517 const char *message, void *cb_data)
519 struct object *obj;
521 if (verbose)
522 fprintf(stderr, "Checking reflog %s->%s\n",
523 sha1_to_hex(osha1), sha1_to_hex(nsha1));
525 if (!is_null_sha1(osha1)) {
526 obj = lookup_object(osha1);
527 if (obj) {
528 obj->used = 1;
529 mark_reachable(obj, REACHABLE);
532 obj = lookup_object(nsha1);
533 if (obj) {
534 obj->used = 1;
535 mark_reachable(obj, REACHABLE);
537 return 0;
540 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
542 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
543 return 0;
546 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
548 struct object *obj;
550 obj = lookup_object(sha1);
551 if (!obj) {
552 if (has_sha1_file(sha1)) {
553 default_refs++;
554 return 0; /* it is in a pack */
556 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
557 /* We'll continue with the rest despite the error.. */
558 return 0;
560 default_refs++;
561 obj->used = 1;
562 mark_reachable(obj, REACHABLE);
564 return 0;
567 static void get_default_heads(void)
569 for_each_ref(fsck_handle_ref, NULL);
570 if (include_reflogs)
571 for_each_reflog(fsck_handle_reflog, NULL);
574 * Not having any default heads isn't really fatal, but
575 * it does mean that "--unreachable" no longer makes any
576 * sense (since in this case everything will obviously
577 * be unreachable by definition.
579 * Showing dangling objects is valid, though (as those
580 * dangling objects are likely lost heads).
582 * So we just print a warning about it, and clear the
583 * "show_unreachable" flag.
585 if (!default_refs) {
586 fprintf(stderr, "notice: No default references\n");
587 show_unreachable = 0;
591 static void fsck_object_dir(const char *path)
593 int i;
595 if (verbose)
596 fprintf(stderr, "Checking object directory\n");
598 for (i = 0; i < 256; i++) {
599 static char dir[4096];
600 sprintf(dir, "%s/%02x", path, i);
601 fsck_dir(i, dir);
603 fsck_sha1_list();
606 static int fsck_head_link(void)
608 unsigned char sha1[20];
609 int flag;
610 int null_is_error = 0;
611 const char *head_points_at = resolve_ref("HEAD", sha1, 0, &flag);
613 if (verbose)
614 fprintf(stderr, "Checking HEAD link\n");
616 if (!head_points_at)
617 return error("Invalid HEAD");
618 if (!strcmp(head_points_at, "HEAD"))
619 /* detached HEAD */
620 null_is_error = 1;
621 else if (prefixcmp(head_points_at, "refs/heads/"))
622 return error("HEAD points to something strange (%s)",
623 head_points_at);
624 if (is_null_sha1(sha1)) {
625 if (null_is_error)
626 return error("HEAD: detached HEAD points at nothing");
627 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
628 head_points_at + 11);
630 return 0;
633 static int fsck_cache_tree(struct cache_tree *it)
635 int i;
636 int err = 0;
638 if (verbose)
639 fprintf(stderr, "Checking cache tree\n");
641 if (0 <= it->entry_count) {
642 struct object *obj = parse_object(it->sha1);
643 if (!obj) {
644 error("%s: invalid sha1 pointer in cache-tree",
645 sha1_to_hex(it->sha1));
646 return 1;
648 mark_reachable(obj, REACHABLE);
649 obj->used = 1;
650 if (obj->type != OBJ_TREE)
651 err |= objerror(obj, "non-tree in cache-tree");
653 for (i = 0; i < it->subtree_nr; i++)
654 err |= fsck_cache_tree(it->down[i]->cache_tree);
655 return err;
658 static const char fsck_usage[] =
659 "git-fsck [--tags] [--root] [[--unreachable] [--cache] [--full] "
660 "[--strict] [--verbose] <head-sha1>*]";
662 int cmd_fsck(int argc, char **argv, const char *prefix)
664 int i, heads;
666 track_object_refs = 1;
667 errors_found = 0;
669 for (i = 1; i < argc; i++) {
670 const char *arg = argv[i];
672 if (!strcmp(arg, "--unreachable")) {
673 show_unreachable = 1;
674 continue;
676 if (!strcmp(arg, "--tags")) {
677 show_tags = 1;
678 continue;
680 if (!strcmp(arg, "--root")) {
681 show_root = 1;
682 continue;
684 if (!strcmp(arg, "--cache")) {
685 keep_cache_objects = 1;
686 continue;
688 if (!strcmp(arg, "--no-reflogs")) {
689 include_reflogs = 0;
690 continue;
692 if (!strcmp(arg, "--full")) {
693 check_full = 1;
694 continue;
696 if (!strcmp(arg, "--strict")) {
697 check_strict = 1;
698 continue;
700 if (!strcmp(arg, "--verbose")) {
701 verbose = 1;
702 continue;
704 if (!strcmp(arg, "--lost-found")) {
705 check_full = 1;
706 include_reflogs = 0;
707 write_lost_and_found = 1;
708 continue;
710 if (*arg == '-')
711 usage(fsck_usage);
714 fsck_head_link();
715 fsck_object_dir(get_object_directory());
716 if (check_full) {
717 struct alternate_object_database *alt;
718 struct packed_git *p;
719 prepare_alt_odb();
720 for (alt = alt_odb_list; alt; alt = alt->next) {
721 char namebuf[PATH_MAX];
722 int namelen = alt->name - alt->base;
723 memcpy(namebuf, alt->base, namelen);
724 namebuf[namelen - 1] = 0;
725 fsck_object_dir(namebuf);
727 prepare_packed_git();
728 for (p = packed_git; p; p = p->next)
729 /* verify gives error messages itself */
730 verify_pack(p, 0);
732 for (p = packed_git; p; p = p->next) {
733 uint32_t i, num;
734 if (open_pack_index(p))
735 continue;
736 num = p->num_objects;
737 for (i = 0; i < num; i++)
738 fsck_sha1(nth_packed_object_sha1(p, i));
742 heads = 0;
743 for (i = 1; i < argc; i++) {
744 const char *arg = argv[i];
746 if (*arg == '-')
747 continue;
749 if (!get_sha1(arg, head_sha1)) {
750 struct object *obj = lookup_object(head_sha1);
752 /* Error is printed by lookup_object(). */
753 if (!obj)
754 continue;
756 obj->used = 1;
757 mark_reachable(obj, REACHABLE);
758 heads++;
759 continue;
761 error("invalid parameter: expected sha1, got '%s'", arg);
765 * If we've not been given any explicit head information, do the
766 * default ones from .git/refs. We also consider the index file
767 * in this case (ie this implies --cache).
769 if (!heads) {
770 get_default_heads();
771 keep_cache_objects = 1;
774 if (keep_cache_objects) {
775 int i;
776 read_cache();
777 for (i = 0; i < active_nr; i++) {
778 unsigned int mode;
779 struct blob *blob;
780 struct object *obj;
782 mode = ntohl(active_cache[i]->ce_mode);
783 if (S_ISGITLINK(mode))
784 continue;
785 blob = lookup_blob(active_cache[i]->sha1);
786 if (!blob)
787 continue;
788 obj = &blob->object;
789 obj->used = 1;
790 mark_reachable(obj, REACHABLE);
792 if (active_cache_tree)
793 fsck_cache_tree(active_cache_tree);
796 check_connectivity();
797 return errors_found;