Merge branch 'nd/pack-unreachable-objects-doc'
[git/debian.git] / builtin / fsck.c
blob9d59d7d5a215379b221b02ac34277b542e757df1
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "commit.h"
6 #include "tree.h"
7 #include "blob.h"
8 #include "tag.h"
9 #include "refs.h"
10 #include "pack.h"
11 #include "cache-tree.h"
12 #include "tree-walk.h"
13 #include "fsck.h"
14 #include "parse-options.h"
15 #include "dir.h"
16 #include "progress.h"
17 #include "streaming.h"
18 #include "decorate.h"
19 #include "packfile.h"
20 #include "object-store.h"
22 #define REACHABLE 0x0001
23 #define SEEN 0x0002
24 #define HAS_OBJ 0x0004
25 /* This flag is set if something points to this object. */
26 #define USED 0x0008
28 static int show_root;
29 static int show_tags;
30 static int show_unreachable;
31 static int include_reflogs = 1;
32 static int check_full = 1;
33 static int connectivity_only;
34 static int check_strict;
35 static int keep_cache_objects;
36 static struct fsck_options fsck_walk_options = FSCK_OPTIONS_DEFAULT;
37 static struct fsck_options fsck_obj_options = FSCK_OPTIONS_DEFAULT;
38 static struct object_id head_oid;
39 static const char *head_points_at;
40 static int errors_found;
41 static int write_lost_and_found;
42 static int verbose;
43 static int show_progress = -1;
44 static int show_dangling = 1;
45 static int name_objects;
46 #define ERROR_OBJECT 01
47 #define ERROR_REACHABLE 02
48 #define ERROR_PACK 04
49 #define ERROR_REFS 010
51 static const char *describe_object(struct object *obj)
53 static struct strbuf buf = STRBUF_INIT;
54 char *name = name_objects ?
55 lookup_decoration(fsck_walk_options.object_names, obj) : NULL;
57 strbuf_reset(&buf);
58 strbuf_addstr(&buf, oid_to_hex(&obj->oid));
59 if (name)
60 strbuf_addf(&buf, " (%s)", name);
62 return buf.buf;
65 static const char *printable_type(struct object *obj)
67 const char *ret;
69 if (obj->type == OBJ_NONE) {
70 enum object_type type = oid_object_info(the_repository,
71 &obj->oid, NULL);
72 if (type > 0)
73 object_as_type(obj, type, 0);
76 ret = type_name(obj->type);
77 if (!ret)
78 ret = "unknown";
80 return ret;
83 static int fsck_config(const char *var, const char *value, void *cb)
85 if (strcmp(var, "fsck.skiplist") == 0) {
86 const char *path;
87 struct strbuf sb = STRBUF_INIT;
89 if (git_config_pathname(&path, var, value))
90 return 1;
91 strbuf_addf(&sb, "skiplist=%s", path);
92 free((char *)path);
93 fsck_set_msg_types(&fsck_obj_options, sb.buf);
94 strbuf_release(&sb);
95 return 0;
98 if (skip_prefix(var, "fsck.", &var)) {
99 fsck_set_msg_type(&fsck_obj_options, var, value);
100 return 0;
103 return git_default_config(var, value, cb);
106 static void objreport(struct object *obj, const char *msg_type,
107 const char *err)
109 fprintf(stderr, "%s in %s %s: %s\n",
110 msg_type, printable_type(obj), describe_object(obj), err);
113 static int objerror(struct object *obj, const char *err)
115 errors_found |= ERROR_OBJECT;
116 objreport(obj, "error", err);
117 return -1;
120 static int fsck_error_func(struct fsck_options *o,
121 struct object *obj, int type, const char *message)
123 objreport(obj, (type == FSCK_WARN) ? "warning" : "error", message);
124 return (type == FSCK_WARN) ? 0 : 1;
127 static struct object_array pending;
129 static int mark_object(struct object *obj, int type, void *data, struct fsck_options *options)
131 struct object *parent = data;
134 * The only case data is NULL or type is OBJ_ANY is when
135 * mark_object_reachable() calls us. All the callers of
136 * that function has non-NULL obj hence ...
138 if (!obj) {
139 /* ... these references to parent->fld are safe here */
140 printf("broken link from %7s %s\n",
141 printable_type(parent), describe_object(parent));
142 printf("broken link from %7s %s\n",
143 (type == OBJ_ANY ? "unknown" : type_name(type)), "unknown");
144 errors_found |= ERROR_REACHABLE;
145 return 1;
148 if (type != OBJ_ANY && obj->type != type)
149 /* ... and the reference to parent is safe here */
150 objerror(parent, "wrong object type in link");
152 if (obj->flags & REACHABLE)
153 return 0;
154 obj->flags |= REACHABLE;
156 if (is_promisor_object(&obj->oid))
158 * Further recursion does not need to be performed on this
159 * object since it is a promisor object (so it does not need to
160 * be added to "pending").
162 return 0;
164 if (!(obj->flags & HAS_OBJ)) {
165 if (parent && !has_object_file(&obj->oid)) {
166 printf("broken link from %7s %s\n",
167 printable_type(parent), describe_object(parent));
168 printf(" to %7s %s\n",
169 printable_type(obj), describe_object(obj));
170 errors_found |= ERROR_REACHABLE;
172 return 1;
175 add_object_array(obj, NULL, &pending);
176 return 0;
179 static void mark_object_reachable(struct object *obj)
181 mark_object(obj, OBJ_ANY, NULL, NULL);
184 static int traverse_one_object(struct object *obj)
186 int result = fsck_walk(obj, obj, &fsck_walk_options);
188 if (obj->type == OBJ_TREE) {
189 struct tree *tree = (struct tree *)obj;
190 free_tree_buffer(tree);
192 return result;
195 static int traverse_reachable(void)
197 struct progress *progress = NULL;
198 unsigned int nr = 0;
199 int result = 0;
200 if (show_progress)
201 progress = start_delayed_progress(_("Checking connectivity"), 0);
202 while (pending.nr) {
203 result |= traverse_one_object(object_array_pop(&pending));
204 display_progress(progress, ++nr);
206 stop_progress(&progress);
207 return !!result;
210 static int mark_used(struct object *obj, int type, void *data, struct fsck_options *options)
212 if (!obj)
213 return 1;
214 obj->flags |= USED;
215 return 0;
219 * Check a single reachable object
221 static void check_reachable_object(struct object *obj)
224 * We obviously want the object to be parsed,
225 * except if it was in a pack-file and we didn't
226 * do a full fsck
228 if (!(obj->flags & HAS_OBJ)) {
229 if (is_promisor_object(&obj->oid))
230 return;
231 if (has_sha1_pack(obj->oid.hash))
232 return; /* it is in pack - forget about it */
233 printf("missing %s %s\n", printable_type(obj),
234 describe_object(obj));
235 errors_found |= ERROR_REACHABLE;
236 return;
241 * Check a single unreachable object
243 static void check_unreachable_object(struct object *obj)
246 * Missing unreachable object? Ignore it. It's not like
247 * we miss it (since it can't be reached), nor do we want
248 * to complain about it being unreachable (since it does
249 * not exist).
251 if (!(obj->flags & HAS_OBJ))
252 return;
255 * Unreachable object that exists? Show it if asked to,
256 * since this is something that is prunable.
258 if (show_unreachable) {
259 printf("unreachable %s %s\n", printable_type(obj),
260 describe_object(obj));
261 return;
265 * "!USED" means that nothing at all points to it, including
266 * other unreachable objects. In other words, it's the "tip"
267 * of some set of unreachable objects, usually a commit that
268 * got dropped.
270 * Such starting points are more interesting than some random
271 * set of unreachable objects, so we show them even if the user
272 * hasn't asked for _all_ unreachable objects. If you have
273 * deleted a branch by mistake, this is a prime candidate to
274 * start looking at, for example.
276 if (!(obj->flags & USED)) {
277 if (show_dangling)
278 printf("dangling %s %s\n", printable_type(obj),
279 describe_object(obj));
280 if (write_lost_and_found) {
281 char *filename = git_pathdup("lost-found/%s/%s",
282 obj->type == OBJ_COMMIT ? "commit" : "other",
283 describe_object(obj));
284 FILE *f;
286 if (safe_create_leading_directories_const(filename)) {
287 error("Could not create lost-found");
288 free(filename);
289 return;
291 f = xfopen(filename, "w");
292 if (obj->type == OBJ_BLOB) {
293 if (stream_blob_to_fd(fileno(f), &obj->oid, NULL, 1))
294 die_errno("Could not write '%s'", filename);
295 } else
296 fprintf(f, "%s\n", describe_object(obj));
297 if (fclose(f))
298 die_errno("Could not finish '%s'",
299 filename);
300 free(filename);
302 return;
306 * Otherwise? It's there, it's unreachable, and some other unreachable
307 * object points to it. Ignore it - it's not interesting, and we showed
308 * all the interesting cases above.
312 static void check_object(struct object *obj)
314 if (verbose)
315 fprintf(stderr, "Checking %s\n", describe_object(obj));
317 if (obj->flags & REACHABLE)
318 check_reachable_object(obj);
319 else
320 check_unreachable_object(obj);
323 static void check_connectivity(void)
325 int i, max;
327 /* Traverse the pending reachable objects */
328 traverse_reachable();
330 /* Look up all the requirements, warn about missing objects.. */
331 max = get_max_object_index();
332 if (verbose)
333 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
335 for (i = 0; i < max; i++) {
336 struct object *obj = get_indexed_object(i);
338 if (obj)
339 check_object(obj);
343 static int fsck_obj(struct object *obj)
345 int err;
347 if (obj->flags & SEEN)
348 return 0;
349 obj->flags |= SEEN;
351 if (verbose)
352 fprintf(stderr, "Checking %s %s\n",
353 printable_type(obj), describe_object(obj));
355 if (fsck_walk(obj, NULL, &fsck_obj_options))
356 objerror(obj, "broken links");
357 err = fsck_object(obj, NULL, 0, &fsck_obj_options);
358 if (err)
359 goto out;
361 if (obj->type == OBJ_COMMIT) {
362 struct commit *commit = (struct commit *) obj;
364 if (!commit->parents && show_root)
365 printf("root %s\n", describe_object(&commit->object));
368 if (obj->type == OBJ_TAG) {
369 struct tag *tag = (struct tag *) obj;
371 if (show_tags && tag->tagged) {
372 printf("tagged %s %s", printable_type(tag->tagged),
373 describe_object(tag->tagged));
374 printf(" (%s) in %s\n", tag->tag,
375 describe_object(&tag->object));
379 out:
380 if (obj->type == OBJ_TREE)
381 free_tree_buffer((struct tree *)obj);
382 if (obj->type == OBJ_COMMIT)
383 free_commit_buffer((struct commit *)obj);
384 return err;
387 static int fsck_obj_buffer(const struct object_id *oid, enum object_type type,
388 unsigned long size, void *buffer, int *eaten)
391 * Note, buffer may be NULL if type is OBJ_BLOB. See
392 * verify_packfile(), data_valid variable for details.
394 struct object *obj;
395 obj = parse_object_buffer(oid, type, size, buffer, eaten);
396 if (!obj) {
397 errors_found |= ERROR_OBJECT;
398 return error("%s: object corrupt or missing", oid_to_hex(oid));
400 obj->flags &= ~(REACHABLE | SEEN);
401 obj->flags |= HAS_OBJ;
402 return fsck_obj(obj);
405 static int default_refs;
407 static void fsck_handle_reflog_oid(const char *refname, struct object_id *oid,
408 timestamp_t timestamp)
410 struct object *obj;
412 if (!is_null_oid(oid)) {
413 obj = lookup_object(oid->hash);
414 if (obj && (obj->flags & HAS_OBJ)) {
415 if (timestamp && name_objects)
416 add_decoration(fsck_walk_options.object_names,
417 obj,
418 xstrfmt("%s@{%"PRItime"}", refname, timestamp));
419 obj->flags |= USED;
420 mark_object_reachable(obj);
421 } else if (!is_promisor_object(oid)) {
422 error("%s: invalid reflog entry %s", refname, oid_to_hex(oid));
423 errors_found |= ERROR_REACHABLE;
428 static int fsck_handle_reflog_ent(struct object_id *ooid, struct object_id *noid,
429 const char *email, timestamp_t timestamp, int tz,
430 const char *message, void *cb_data)
432 const char *refname = cb_data;
434 if (verbose)
435 fprintf(stderr, "Checking reflog %s->%s\n",
436 oid_to_hex(ooid), oid_to_hex(noid));
438 fsck_handle_reflog_oid(refname, ooid, 0);
439 fsck_handle_reflog_oid(refname, noid, timestamp);
440 return 0;
443 static int fsck_handle_reflog(const char *logname, const struct object_id *oid,
444 int flag, void *cb_data)
446 for_each_reflog_ent(logname, fsck_handle_reflog_ent, (void *)logname);
447 return 0;
450 static int fsck_handle_ref(const char *refname, const struct object_id *oid,
451 int flag, void *cb_data)
453 struct object *obj;
455 obj = parse_object(oid);
456 if (!obj) {
457 if (is_promisor_object(oid)) {
459 * Increment default_refs anyway, because this is a
460 * valid ref.
462 default_refs++;
463 return 0;
465 error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
466 errors_found |= ERROR_REACHABLE;
467 /* We'll continue with the rest despite the error.. */
468 return 0;
470 if (obj->type != OBJ_COMMIT && is_branch(refname)) {
471 error("%s: not a commit", refname);
472 errors_found |= ERROR_REFS;
474 default_refs++;
475 obj->flags |= USED;
476 if (name_objects)
477 add_decoration(fsck_walk_options.object_names,
478 obj, xstrdup(refname));
479 mark_object_reachable(obj);
481 return 0;
484 static void get_default_heads(void)
486 if (head_points_at && !is_null_oid(&head_oid))
487 fsck_handle_ref("HEAD", &head_oid, 0, NULL);
488 for_each_rawref(fsck_handle_ref, NULL);
489 if (include_reflogs)
490 for_each_reflog(fsck_handle_reflog, NULL);
493 * Not having any default heads isn't really fatal, but
494 * it does mean that "--unreachable" no longer makes any
495 * sense (since in this case everything will obviously
496 * be unreachable by definition.
498 * Showing dangling objects is valid, though (as those
499 * dangling objects are likely lost heads).
501 * So we just print a warning about it, and clear the
502 * "show_unreachable" flag.
504 if (!default_refs) {
505 fprintf(stderr, "notice: No default references\n");
506 show_unreachable = 0;
510 static struct object *parse_loose_object(const struct object_id *oid,
511 const char *path)
513 struct object *obj;
514 void *contents;
515 enum object_type type;
516 unsigned long size;
517 int eaten;
519 if (read_loose_object(path, oid, &type, &size, &contents) < 0)
520 return NULL;
522 if (!contents && type != OBJ_BLOB)
523 die("BUG: read_loose_object streamed a non-blob");
525 obj = parse_object_buffer(oid, type, size, contents, &eaten);
527 if (!eaten)
528 free(contents);
529 return obj;
532 static int fsck_loose(const struct object_id *oid, const char *path, void *data)
534 struct object *obj = parse_loose_object(oid, path);
536 if (!obj) {
537 errors_found |= ERROR_OBJECT;
538 error("%s: object corrupt or missing: %s",
539 oid_to_hex(oid), path);
540 return 0; /* keep checking other objects */
543 obj->flags &= ~(REACHABLE | SEEN);
544 obj->flags |= HAS_OBJ;
545 if (fsck_obj(obj))
546 errors_found |= ERROR_OBJECT;
547 return 0;
550 static int fsck_cruft(const char *basename, const char *path, void *data)
552 if (!starts_with(basename, "tmp_obj_"))
553 fprintf(stderr, "bad sha1 file: %s\n", path);
554 return 0;
557 static int fsck_subdir(unsigned int nr, const char *path, void *progress)
559 display_progress(progress, nr + 1);
560 return 0;
563 static void fsck_object_dir(const char *path)
565 struct progress *progress = NULL;
567 if (verbose)
568 fprintf(stderr, "Checking object directory\n");
570 if (show_progress)
571 progress = start_progress(_("Checking object directories"), 256);
573 for_each_loose_file_in_objdir(path, fsck_loose, fsck_cruft, fsck_subdir,
574 progress);
575 display_progress(progress, 256);
576 stop_progress(&progress);
579 static int fsck_head_link(void)
581 int null_is_error = 0;
583 if (verbose)
584 fprintf(stderr, "Checking HEAD link\n");
586 head_points_at = resolve_ref_unsafe("HEAD", 0, &head_oid, NULL);
587 if (!head_points_at) {
588 errors_found |= ERROR_REFS;
589 return error("Invalid HEAD");
591 if (!strcmp(head_points_at, "HEAD"))
592 /* detached HEAD */
593 null_is_error = 1;
594 else if (!starts_with(head_points_at, "refs/heads/")) {
595 errors_found |= ERROR_REFS;
596 return error("HEAD points to something strange (%s)",
597 head_points_at);
599 if (is_null_oid(&head_oid)) {
600 if (null_is_error) {
601 errors_found |= ERROR_REFS;
602 return error("HEAD: detached HEAD points at nothing");
604 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
605 head_points_at + 11);
607 return 0;
610 static int fsck_cache_tree(struct cache_tree *it)
612 int i;
613 int err = 0;
615 if (verbose)
616 fprintf(stderr, "Checking cache tree\n");
618 if (0 <= it->entry_count) {
619 struct object *obj = parse_object(&it->oid);
620 if (!obj) {
621 error("%s: invalid sha1 pointer in cache-tree",
622 oid_to_hex(&it->oid));
623 errors_found |= ERROR_REFS;
624 return 1;
626 obj->flags |= USED;
627 if (name_objects)
628 add_decoration(fsck_walk_options.object_names,
629 obj, xstrdup(":"));
630 mark_object_reachable(obj);
631 if (obj->type != OBJ_TREE)
632 err |= objerror(obj, "non-tree in cache-tree");
634 for (i = 0; i < it->subtree_nr; i++)
635 err |= fsck_cache_tree(it->down[i]->cache_tree);
636 return err;
639 static void mark_object_for_connectivity(const struct object_id *oid)
641 struct object *obj = lookup_unknown_object(oid->hash);
642 obj->flags |= HAS_OBJ;
645 static int mark_loose_for_connectivity(const struct object_id *oid,
646 const char *path,
647 void *data)
649 mark_object_for_connectivity(oid);
650 return 0;
653 static int mark_packed_for_connectivity(const struct object_id *oid,
654 struct packed_git *pack,
655 uint32_t pos,
656 void *data)
658 mark_object_for_connectivity(oid);
659 return 0;
662 static char const * const fsck_usage[] = {
663 N_("git fsck [<options>] [<object>...]"),
664 NULL
667 static struct option fsck_opts[] = {
668 OPT__VERBOSE(&verbose, N_("be verbose")),
669 OPT_BOOL(0, "unreachable", &show_unreachable, N_("show unreachable objects")),
670 OPT_BOOL(0, "dangling", &show_dangling, N_("show dangling objects")),
671 OPT_BOOL(0, "tags", &show_tags, N_("report tags")),
672 OPT_BOOL(0, "root", &show_root, N_("report root nodes")),
673 OPT_BOOL(0, "cache", &keep_cache_objects, N_("make index objects head nodes")),
674 OPT_BOOL(0, "reflogs", &include_reflogs, N_("make reflogs head nodes (default)")),
675 OPT_BOOL(0, "full", &check_full, N_("also consider packs and alternate objects")),
676 OPT_BOOL(0, "connectivity-only", &connectivity_only, N_("check only connectivity")),
677 OPT_BOOL(0, "strict", &check_strict, N_("enable more strict checking")),
678 OPT_BOOL(0, "lost-found", &write_lost_and_found,
679 N_("write dangling objects in .git/lost-found")),
680 OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
681 OPT_BOOL(0, "name-objects", &name_objects, N_("show verbose names for reachable objects")),
682 OPT_END(),
685 int cmd_fsck(int argc, const char **argv, const char *prefix)
687 int i;
688 struct alternate_object_database *alt;
690 /* fsck knows how to handle missing promisor objects */
691 fetch_if_missing = 0;
693 errors_found = 0;
694 check_replace_refs = 0;
696 argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
698 fsck_walk_options.walk = mark_object;
699 fsck_obj_options.walk = mark_used;
700 fsck_obj_options.error_func = fsck_error_func;
701 if (check_strict)
702 fsck_obj_options.strict = 1;
704 if (show_progress == -1)
705 show_progress = isatty(2);
706 if (verbose)
707 show_progress = 0;
709 if (write_lost_and_found) {
710 check_full = 1;
711 include_reflogs = 0;
714 if (name_objects)
715 fsck_walk_options.object_names =
716 xcalloc(1, sizeof(struct decoration));
718 git_config(fsck_config, NULL);
720 fsck_head_link();
721 if (connectivity_only) {
722 for_each_loose_object(mark_loose_for_connectivity, NULL, 0);
723 for_each_packed_object(mark_packed_for_connectivity, NULL, 0);
724 } else {
725 struct alternate_object_database *alt_odb_list;
727 fsck_object_dir(get_object_directory());
729 prepare_alt_odb(the_repository);
730 alt_odb_list = the_repository->objects->alt_odb_list;
731 for (alt = alt_odb_list; alt; alt = alt->next)
732 fsck_object_dir(alt->path);
734 if (check_full) {
735 struct packed_git *p;
736 uint32_t total = 0, count = 0;
737 struct progress *progress = NULL;
739 if (show_progress) {
740 for (p = get_packed_git(the_repository); p;
741 p = p->next) {
742 if (open_pack_index(p))
743 continue;
744 total += p->num_objects;
747 progress = start_progress(_("Checking objects"), total);
749 for (p = get_packed_git(the_repository); p;
750 p = p->next) {
751 /* verify gives error messages itself */
752 if (verify_pack(p, fsck_obj_buffer,
753 progress, count))
754 errors_found |= ERROR_PACK;
755 count += p->num_objects;
757 stop_progress(&progress);
761 for (i = 0; i < argc; i++) {
762 const char *arg = argv[i];
763 struct object_id oid;
764 if (!get_oid(arg, &oid)) {
765 struct object *obj = lookup_object(oid.hash);
767 if (!obj || !(obj->flags & HAS_OBJ)) {
768 if (is_promisor_object(&oid))
769 continue;
770 error("%s: object missing", oid_to_hex(&oid));
771 errors_found |= ERROR_OBJECT;
772 continue;
775 obj->flags |= USED;
776 if (name_objects)
777 add_decoration(fsck_walk_options.object_names,
778 obj, xstrdup(arg));
779 mark_object_reachable(obj);
780 continue;
782 error("invalid parameter: expected sha1, got '%s'", arg);
783 errors_found |= ERROR_OBJECT;
787 * If we've not been given any explicit head information, do the
788 * default ones from .git/refs. We also consider the index file
789 * in this case (ie this implies --cache).
791 if (!argc) {
792 get_default_heads();
793 keep_cache_objects = 1;
796 if (keep_cache_objects) {
797 verify_index_checksum = 1;
798 verify_ce_order = 1;
799 read_cache();
800 for (i = 0; i < active_nr; i++) {
801 unsigned int mode;
802 struct blob *blob;
803 struct object *obj;
805 mode = active_cache[i]->ce_mode;
806 if (S_ISGITLINK(mode))
807 continue;
808 blob = lookup_blob(&active_cache[i]->oid);
809 if (!blob)
810 continue;
811 obj = &blob->object;
812 obj->flags |= USED;
813 if (name_objects)
814 add_decoration(fsck_walk_options.object_names,
815 obj,
816 xstrfmt(":%s", active_cache[i]->name));
817 mark_object_reachable(obj);
819 if (active_cache_tree)
820 fsck_cache_tree(active_cache_tree);
823 check_connectivity();
824 return errors_found;