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
21static int show_root;
22static int show_tags;
23static int show_unreachable;
24static int include_reflogs = 1;
25static int check_full = 1;
26static int connectivity_only;
27static int check_strict;
28static int keep_cache_objects;
29static struct fsck_options fsck_walk_options = FSCK_OPTIONS_DEFAULT;
30static struct fsck_options fsck_obj_options = FSCK_OPTIONS_DEFAULT;
31static struct object_id head_oid;
32static const char *head_points_at;
33static int errors_found;
34static int write_lost_and_found;
35static int verbose;
36static int show_progress = -1;
37static int show_dangling = 1;
38#define ERROR_OBJECT 01
39#define ERROR_REACHABLE 02
40#define ERROR_PACK 04
41#define ERROR_REFS 010
42
43static const char *describe_object(struct object *obj)
44{
45 return oid_to_hex(&obj->oid);
46}
47
48static int fsck_config(const char *var, const char *value, void *cb)
49{
50 if (strcmp(var, "fsck.skiplist") == 0) {
51 const char *path;
52 struct strbuf sb = STRBUF_INIT;
53
54 if (git_config_pathname(&path, var, value))
55 return 1;
56 strbuf_addf(&sb, "skiplist=%s", path);
57 free((char *)path);
58 fsck_set_msg_types(&fsck_obj_options, sb.buf);
59 strbuf_release(&sb);
60 return 0;
61 }
62
63 if (skip_prefix(var, "fsck.", &var)) {
64 fsck_set_msg_type(&fsck_obj_options, var, value);
65 return 0;
66 }
67
68 return git_default_config(var, value, cb);
69}
70
71static void objreport(struct object *obj, const char *msg_type,
72 const char *err)
73{
74 fprintf(stderr, "%s in %s %s: %s\n",
75 msg_type, typename(obj->type), describe_object(obj), err);
76}
77
78static int objerror(struct object *obj, const char *err)
79{
80 errors_found |= ERROR_OBJECT;
81 objreport(obj, "error", err);
82 return -1;
83}
84
85static int fsck_error_func(struct object *obj, int type, const char *message)
86{
87 objreport(obj, (type == FSCK_WARN) ? "warning" : "error", message);
88 return (type == FSCK_WARN) ? 0 : 1;
89}
90
91static struct object_array pending;
92
93static int mark_object(struct object *obj, int type, void *data, struct fsck_options *options)
94{
95 struct object *parent = data;
96
97 /*
98 * The only case data is NULL or type is OBJ_ANY is when
99 * mark_object_reachable() calls us. All the callers of
100 * that function has non-NULL obj hence ...
101 */
102 if (!obj) {
103 /* ... these references to parent->fld are safe here */
104 printf("broken link from %7s %s\n",
105 typename(parent->type), describe_object(parent));
106 printf("broken link from %7s %s\n",
107 (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
108 errors_found |= ERROR_REACHABLE;
109 return 1;
110 }
111
112 if (type != OBJ_ANY && obj->type != type)
113 /* ... and the reference to parent is safe here */
114 objerror(parent, "wrong object type in link");
115
116 if (obj->flags & REACHABLE)
117 return 0;
118 obj->flags |= REACHABLE;
119 if (!(obj->flags & HAS_OBJ)) {
120 if (parent && !has_object_file(&obj->oid)) {
121 printf("broken link from %7s %s\n",
122 typename(parent->type), describe_object(parent));
123 printf(" to %7s %s\n",
124 typename(obj->type), describe_object(obj));
125 errors_found |= ERROR_REACHABLE;
126 }
127 return 1;
128 }
129
130 add_object_array(obj, NULL, &pending);
131 return 0;
132}
133
134static void mark_object_reachable(struct object *obj)
135{
136 mark_object(obj, OBJ_ANY, NULL, NULL);
137}
138
139static int traverse_one_object(struct object *obj)
140{
141 int result;
142 struct tree *tree = NULL;
143
144 if (obj->type == OBJ_TREE) {
145 tree = (struct tree *)obj;
146 if (parse_tree(tree) < 0)
147 return 1; /* error already displayed */
148 }
149 result = fsck_walk(obj, obj, &fsck_walk_options);
150 if (tree)
151 free_tree_buffer(tree);
152 return result;
153}
154
155static int traverse_reachable(void)
156{
157 struct progress *progress = NULL;
158 unsigned int nr = 0;
159 int result = 0;
160 if (show_progress)
161 progress = start_progress_delay(_("Checking connectivity"), 0, 0, 2);
162 while (pending.nr) {
163 struct object_array_entry *entry;
164 struct object *obj;
165
166 entry = pending.objects + --pending.nr;
167 obj = entry->item;
168 result |= traverse_one_object(obj);
169 display_progress(progress, ++nr);
170 }
171 stop_progress(&progress);
172 return !!result;
173}
174
175static int mark_used(struct object *obj, int type, void *data, struct fsck_options *options)
176{
177 if (!obj)
178 return 1;
179 obj->used = 1;
180 return 0;
181}
182
183/*
184 * Check a single reachable object
185 */
186static void check_reachable_object(struct object *obj)
187{
188 /*
189 * We obviously want the object to be parsed,
190 * except if it was in a pack-file and we didn't
191 * do a full fsck
192 */
193 if (!(obj->flags & HAS_OBJ)) {
194 if (has_sha1_pack(obj->oid.hash))
195 return; /* it is in pack - forget about it */
196 if (connectivity_only && has_object_file(&obj->oid))
197 return;
198 printf("missing %s %s\n", typename(obj->type),
199 describe_object(obj));
200 errors_found |= ERROR_REACHABLE;
201 return;
202 }
203}
204
205/*
206 * Check a single unreachable object
207 */
208static void check_unreachable_object(struct object *obj)
209{
210 /*
211 * Missing unreachable object? Ignore it. It's not like
212 * we miss it (since it can't be reached), nor do we want
213 * to complain about it being unreachable (since it does
214 * not exist).
215 */
216 if (!obj->parsed)
217 return;
218
219 /*
220 * Unreachable object that exists? Show it if asked to,
221 * since this is something that is prunable.
222 */
223 if (show_unreachable) {
224 printf("unreachable %s %s\n", typename(obj->type),
225 describe_object(obj));
226 return;
227 }
228
229 /*
230 * "!used" means that nothing at all points to it, including
231 * other unreachable objects. In other words, it's the "tip"
232 * of some set of unreachable objects, usually a commit that
233 * got dropped.
234 *
235 * Such starting points are more interesting than some random
236 * set of unreachable objects, so we show them even if the user
237 * hasn't asked for _all_ unreachable objects. If you have
238 * deleted a branch by mistake, this is a prime candidate to
239 * start looking at, for example.
240 */
241 if (!obj->used) {
242 if (show_dangling)
243 printf("dangling %s %s\n", typename(obj->type),
244 describe_object(obj));
245 if (write_lost_and_found) {
246 char *filename = git_pathdup("lost-found/%s/%s",
247 obj->type == OBJ_COMMIT ? "commit" : "other",
248 describe_object(obj));
249 FILE *f;
250
251 if (safe_create_leading_directories_const(filename)) {
252 error("Could not create lost-found");
253 free(filename);
254 return;
255 }
256 if (!(f = fopen(filename, "w")))
257 die_errno("Could not open '%s'", filename);
258 if (obj->type == OBJ_BLOB) {
259 if (stream_blob_to_fd(fileno(f), obj->oid.hash, NULL, 1))
260 die_errno("Could not write '%s'", filename);
261 } else
262 fprintf(f, "%s\n", describe_object(obj));
263 if (fclose(f))
264 die_errno("Could not finish '%s'",
265 filename);
266 free(filename);
267 }
268 return;
269 }
270
271 /*
272 * Otherwise? It's there, it's unreachable, and some other unreachable
273 * object points to it. Ignore it - it's not interesting, and we showed
274 * all the interesting cases above.
275 */
276}
277
278static void check_object(struct object *obj)
279{
280 if (verbose)
281 fprintf(stderr, "Checking %s\n", describe_object(obj));
282
283 if (obj->flags & REACHABLE)
284 check_reachable_object(obj);
285 else
286 check_unreachable_object(obj);
287}
288
289static void check_connectivity(void)
290{
291 int i, max;
292
293 /* Traverse the pending reachable objects */
294 traverse_reachable();
295
296 /* Look up all the requirements, warn about missing objects.. */
297 max = get_max_object_index();
298 if (verbose)
299 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
300
301 for (i = 0; i < max; i++) {
302 struct object *obj = get_indexed_object(i);
303
304 if (obj)
305 check_object(obj);
306 }
307}
308
309static int fsck_obj(struct object *obj)
310{
311 if (obj->flags & SEEN)
312 return 0;
313 obj->flags |= SEEN;
314
315 if (verbose)
316 fprintf(stderr, "Checking %s %s\n",
317 typename(obj->type), describe_object(obj));
318
319 if (fsck_walk(obj, NULL, &fsck_obj_options))
320 objerror(obj, "broken links");
321 if (fsck_object(obj, NULL, 0, &fsck_obj_options))
322 return -1;
323
324 if (obj->type == OBJ_TREE) {
325 struct tree *item = (struct tree *) obj;
326
327 free_tree_buffer(item);
328 }
329
330 if (obj->type == OBJ_COMMIT) {
331 struct commit *commit = (struct commit *) obj;
332
333 free_commit_buffer(commit);
334
335 if (!commit->parents && show_root)
336 printf("root %s\n", describe_object(&commit->object));
337 }
338
339 if (obj->type == OBJ_TAG) {
340 struct tag *tag = (struct tag *) obj;
341
342 if (show_tags && tag->tagged) {
343 printf("tagged %s %s", typename(tag->tagged->type),
344 describe_object(tag->tagged));
345 printf(" (%s) in %s\n", tag->tag,
346 describe_object(&tag->object));
347 }
348 }
349
350 return 0;
351}
352
353static int fsck_sha1(const unsigned char *sha1)
354{
355 struct object *obj = parse_object(sha1);
356 if (!obj) {
357 errors_found |= ERROR_OBJECT;
358 return error("%s: object corrupt or missing",
359 sha1_to_hex(sha1));
360 }
361 obj->flags |= HAS_OBJ;
362 return fsck_obj(obj);
363}
364
365static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
366 unsigned long size, void *buffer, int *eaten)
367{
368 struct object *obj;
369 obj = parse_object_buffer(sha1, type, size, buffer, eaten);
370 if (!obj) {
371 errors_found |= ERROR_OBJECT;
372 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
373 }
374 obj->flags = HAS_OBJ;
375 return fsck_obj(obj);
376}
377
378static int default_refs;
379
380static void fsck_handle_reflog_sha1(const char *refname, unsigned char *sha1)
381{
382 struct object *obj;
383
384 if (!is_null_sha1(sha1)) {
385 obj = lookup_object(sha1);
386 if (obj) {
387 obj->used = 1;
388 mark_object_reachable(obj);
389 } else {
390 error("%s: invalid reflog entry %s", refname, sha1_to_hex(sha1));
391 errors_found |= ERROR_REACHABLE;
392 }
393 }
394}
395
396static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
397 const char *email, unsigned long timestamp, int tz,
398 const char *message, void *cb_data)
399{
400 const char *refname = cb_data;
401
402 if (verbose)
403 fprintf(stderr, "Checking reflog %s->%s\n",
404 sha1_to_hex(osha1), sha1_to_hex(nsha1));
405
406 fsck_handle_reflog_sha1(refname, osha1);
407 fsck_handle_reflog_sha1(refname, nsha1);
408 return 0;
409}
410
411static int fsck_handle_reflog(const char *logname, const struct object_id *oid,
412 int flag, void *cb_data)
413{
414 for_each_reflog_ent(logname, fsck_handle_reflog_ent, (void *)logname);
415 return 0;
416}
417
418static int fsck_handle_ref(const char *refname, const struct object_id *oid,
419 int flag, void *cb_data)
420{
421 struct object *obj;
422
423 obj = parse_object(oid->hash);
424 if (!obj) {
425 error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
426 errors_found |= ERROR_REACHABLE;
427 /* We'll continue with the rest despite the error.. */
428 return 0;
429 }
430 if (obj->type != OBJ_COMMIT && is_branch(refname)) {
431 error("%s: not a commit", refname);
432 errors_found |= ERROR_REFS;
433 }
434 default_refs++;
435 obj->used = 1;
436 mark_object_reachable(obj);
437
438 return 0;
439}
440
441static void get_default_heads(void)
442{
443 if (head_points_at && !is_null_oid(&head_oid))
444 fsck_handle_ref("HEAD", &head_oid, 0, NULL);
445 for_each_rawref(fsck_handle_ref, NULL);
446 if (include_reflogs)
447 for_each_reflog(fsck_handle_reflog, NULL);
448
449 /*
450 * Not having any default heads isn't really fatal, but
451 * it does mean that "--unreachable" no longer makes any
452 * sense (since in this case everything will obviously
453 * be unreachable by definition.
454 *
455 * Showing dangling objects is valid, though (as those
456 * dangling objects are likely lost heads).
457 *
458 * So we just print a warning about it, and clear the
459 * "show_unreachable" flag.
460 */
461 if (!default_refs) {
462 fprintf(stderr, "notice: No default references\n");
463 show_unreachable = 0;
464 }
465}
466
467static int fsck_loose(const unsigned char *sha1, const char *path, void *data)
468{
469 if (fsck_sha1(sha1))
470 errors_found |= ERROR_OBJECT;
471 return 0;
472}
473
474static int fsck_cruft(const char *basename, const char *path, void *data)
475{
476 if (!starts_with(basename, "tmp_obj_"))
477 fprintf(stderr, "bad sha1 file: %s\n", path);
478 return 0;
479}
480
481static int fsck_subdir(int nr, const char *path, void *progress)
482{
483 display_progress(progress, nr + 1);
484 return 0;
485}
486
487static void fsck_object_dir(const char *path)
488{
489 struct progress *progress = NULL;
490
491 if (verbose)
492 fprintf(stderr, "Checking object directory\n");
493
494 if (show_progress)
495 progress = start_progress(_("Checking object directories"), 256);
496
497 for_each_loose_file_in_objdir(path, fsck_loose, fsck_cruft, fsck_subdir,
498 progress);
499 display_progress(progress, 256);
500 stop_progress(&progress);
501}
502
503static int fsck_head_link(void)
504{
505 int null_is_error = 0;
506
507 if (verbose)
508 fprintf(stderr, "Checking HEAD link\n");
509
510 head_points_at = resolve_ref_unsafe("HEAD", 0, head_oid.hash, NULL);
511 if (!head_points_at) {
512 errors_found |= ERROR_REFS;
513 return error("Invalid HEAD");
514 }
515 if (!strcmp(head_points_at, "HEAD"))
516 /* detached HEAD */
517 null_is_error = 1;
518 else if (!starts_with(head_points_at, "refs/heads/")) {
519 errors_found |= ERROR_REFS;
520 return error("HEAD points to something strange (%s)",
521 head_points_at);
522 }
523 if (is_null_oid(&head_oid)) {
524 if (null_is_error) {
525 errors_found |= ERROR_REFS;
526 return error("HEAD: detached HEAD points at nothing");
527 }
528 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
529 head_points_at + 11);
530 }
531 return 0;
532}
533
534static int fsck_cache_tree(struct cache_tree *it)
535{
536 int i;
537 int err = 0;
538
539 if (verbose)
540 fprintf(stderr, "Checking cache tree\n");
541
542 if (0 <= it->entry_count) {
543 struct object *obj = parse_object(it->sha1);
544 if (!obj) {
545 error("%s: invalid sha1 pointer in cache-tree",
546 sha1_to_hex(it->sha1));
547 errors_found |= ERROR_REFS;
548 return 1;
549 }
550 obj->used = 1;
551 mark_object_reachable(obj);
552 if (obj->type != OBJ_TREE)
553 err |= objerror(obj, "non-tree in cache-tree");
554 }
555 for (i = 0; i < it->subtree_nr; i++)
556 err |= fsck_cache_tree(it->down[i]->cache_tree);
557 return err;
558}
559
560static char const * const fsck_usage[] = {
561 N_("git fsck [<options>] [<object>...]"),
562 NULL
563};
564
565static struct option fsck_opts[] = {
566 OPT__VERBOSE(&verbose, N_("be verbose")),
567 OPT_BOOL(0, "unreachable", &show_unreachable, N_("show unreachable objects")),
568 OPT_BOOL(0, "dangling", &show_dangling, N_("show dangling objects")),
569 OPT_BOOL(0, "tags", &show_tags, N_("report tags")),
570 OPT_BOOL(0, "root", &show_root, N_("report root nodes")),
571 OPT_BOOL(0, "cache", &keep_cache_objects, N_("make index objects head nodes")),
572 OPT_BOOL(0, "reflogs", &include_reflogs, N_("make reflogs head nodes (default)")),
573 OPT_BOOL(0, "full", &check_full, N_("also consider packs and alternate objects")),
574 OPT_BOOL(0, "connectivity-only", &connectivity_only, N_("check only connectivity")),
575 OPT_BOOL(0, "strict", &check_strict, N_("enable more strict checking")),
576 OPT_BOOL(0, "lost-found", &write_lost_and_found,
577 N_("write dangling objects in .git/lost-found")),
578 OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
579 OPT_END(),
580};
581
582int cmd_fsck(int argc, const char **argv, const char *prefix)
583{
584 int i, heads;
585 struct alternate_object_database *alt;
586
587 errors_found = 0;
588 check_replace_refs = 0;
589
590 argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
591
592 fsck_walk_options.walk = mark_object;
593 fsck_obj_options.walk = mark_used;
594 fsck_obj_options.error_func = fsck_error_func;
595 if (check_strict)
596 fsck_obj_options.strict = 1;
597
598 if (show_progress == -1)
599 show_progress = isatty(2);
600 if (verbose)
601 show_progress = 0;
602
603 if (write_lost_and_found) {
604 check_full = 1;
605 include_reflogs = 0;
606 }
607
608 git_config(fsck_config, NULL);
609
610 fsck_head_link();
611 if (!connectivity_only) {
612 fsck_object_dir(get_object_directory());
613
614 prepare_alt_odb();
615 for (alt = alt_odb_list; alt; alt = alt->next) {
616 /* directory name, minus trailing slash */
617 size_t namelen = alt->name - alt->base - 1;
618 struct strbuf name = STRBUF_INIT;
619 strbuf_add(&name, alt->base, namelen);
620 fsck_object_dir(name.buf);
621 strbuf_release(&name);
622 }
623 }
624
625 if (check_full) {
626 struct packed_git *p;
627 uint32_t total = 0, count = 0;
628 struct progress *progress = NULL;
629
630 prepare_packed_git();
631
632 if (show_progress) {
633 for (p = packed_git; p; p = p->next) {
634 if (open_pack_index(p))
635 continue;
636 total += p->num_objects;
637 }
638
639 progress = start_progress(_("Checking objects"), total);
640 }
641 for (p = packed_git; p; p = p->next) {
642 /* verify gives error messages itself */
643 if (verify_pack(p, fsck_obj_buffer,
644 progress, count))
645 errors_found |= ERROR_PACK;
646 count += p->num_objects;
647 }
648 stop_progress(&progress);
649 }
650
651 heads = 0;
652 for (i = 0; i < argc; i++) {
653 const char *arg = argv[i];
654 unsigned char sha1[20];
655 if (!get_sha1(arg, sha1)) {
656 struct object *obj = lookup_object(sha1);
657
658 /* Error is printed by lookup_object(). */
659 if (!obj)
660 continue;
661
662 obj->used = 1;
663 mark_object_reachable(obj);
664 heads++;
665 continue;
666 }
667 error("invalid parameter: expected sha1, got '%s'", arg);
668 }
669
670 /*
671 * If we've not been given any explicit head information, do the
672 * default ones from .git/refs. We also consider the index file
673 * in this case (ie this implies --cache).
674 */
675 if (!heads) {
676 get_default_heads();
677 keep_cache_objects = 1;
678 }
679
680 if (keep_cache_objects) {
681 read_cache();
682 for (i = 0; i < active_nr; i++) {
683 unsigned int mode;
684 struct blob *blob;
685 struct object *obj;
686
687 mode = active_cache[i]->ce_mode;
688 if (S_ISGITLINK(mode))
689 continue;
690 blob = lookup_blob(active_cache[i]->sha1);
691 if (!blob)
692 continue;
693 obj = &blob->object;
694 obj->used = 1;
695 mark_object_reachable(obj);
696 }
697 if (active_cache_tree)
698 fsck_cache_tree(active_cache_tree);
699 }
700
701 check_connectivity();
702 return errors_found;
703}