4ead98dcab276b4177fad302ed062a3fb584c9ea
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
15#define REACHABLE 0x0001
16#define SEEN 0x0002
17
18static int show_root;
19static int show_tags;
20static int show_unreachable;
21static int include_reflogs = 1;
22static int check_full = 1;
23static int check_strict;
24static int keep_cache_objects;
25static unsigned char head_sha1[20];
26static const char *head_points_at;
27static int errors_found;
28static int write_lost_and_found;
29static int verbose;
30#define ERROR_OBJECT 01
31#define ERROR_REACHABLE 02
32#define ERROR_PACK 04
33
34#ifdef NO_D_INO_IN_DIRENT
35#define SORT_DIRENT 0
36#define DIRENT_SORT_HINT(de) 0
37#else
38#define SORT_DIRENT 1
39#define DIRENT_SORT_HINT(de) ((de)->d_ino)
40#endif
41
42static void objreport(struct object *obj, const char *severity,
43 const char *err, va_list params)
44{
45 fprintf(stderr, "%s in %s %s: ",
46 severity, typename(obj->type), sha1_to_hex(obj->sha1));
47 vfprintf(stderr, err, params);
48 fputs("\n", stderr);
49}
50
51__attribute__((format (printf, 2, 3)))
52static int objerror(struct object *obj, const char *err, ...)
53{
54 va_list params;
55 va_start(params, err);
56 errors_found |= ERROR_OBJECT;
57 objreport(obj, "error", err, params);
58 va_end(params);
59 return -1;
60}
61
62__attribute__((format (printf, 3, 4)))
63static int fsck_error_func(struct object *obj, int type, const char *err, ...)
64{
65 va_list params;
66 va_start(params, err);
67 objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
68 va_end(params);
69 return (type == FSCK_WARN) ? 0 : 1;
70}
71
72static struct object_array pending;
73
74static int mark_object(struct object *obj, int type, void *data)
75{
76 struct object *parent = data;
77
78 /*
79 * The only case data is NULL or type is OBJ_ANY is when
80 * mark_object_reachable() calls us. All the callers of
81 * that function has non-NULL obj hence ...
82 */
83 if (!obj) {
84 /* ... these references to parent->fld are safe here */
85 printf("broken link from %7s %s\n",
86 typename(parent->type), sha1_to_hex(parent->sha1));
87 printf("broken link from %7s %s\n",
88 (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
89 errors_found |= ERROR_REACHABLE;
90 return 1;
91 }
92
93 if (type != OBJ_ANY && obj->type != type)
94 /* ... and the reference to parent is safe here */
95 objerror(parent, "wrong object type in link");
96
97 if (obj->flags & REACHABLE)
98 return 0;
99 obj->flags |= REACHABLE;
100 if (!obj->parsed) {
101 if (parent && !has_sha1_file(obj->sha1)) {
102 printf("broken link from %7s %s\n",
103 typename(parent->type), sha1_to_hex(parent->sha1));
104 printf(" to %7s %s\n",
105 typename(obj->type), sha1_to_hex(obj->sha1));
106 errors_found |= ERROR_REACHABLE;
107 }
108 return 1;
109 }
110
111 add_object_array(obj, (void *) parent, &pending);
112 return 0;
113}
114
115static void mark_object_reachable(struct object *obj)
116{
117 mark_object(obj, OBJ_ANY, NULL);
118}
119
120static int traverse_one_object(struct object *obj)
121{
122 int result;
123 struct tree *tree = NULL;
124
125 if (obj->type == OBJ_TREE) {
126 obj->parsed = 0;
127 tree = (struct tree *)obj;
128 if (parse_tree(tree) < 0)
129 return 1; /* error already displayed */
130 }
131 result = fsck_walk(obj, mark_object, obj);
132 if (tree) {
133 free(tree->buffer);
134 tree->buffer = NULL;
135 }
136 return result;
137}
138
139static int traverse_reachable(void)
140{
141 int result = 0;
142 while (pending.nr) {
143 struct object_array_entry *entry;
144 struct object *obj;
145
146 entry = pending.objects + --pending.nr;
147 obj = entry->item;
148 result |= traverse_one_object(obj);
149 }
150 return !!result;
151}
152
153static int mark_used(struct object *obj, int type, void *data)
154{
155 if (!obj)
156 return 1;
157 obj->used = 1;
158 return 0;
159}
160
161/*
162 * Check a single reachable object
163 */
164static void check_reachable_object(struct object *obj)
165{
166 /*
167 * We obviously want the object to be parsed,
168 * except if it was in a pack-file and we didn't
169 * do a full fsck
170 */
171 if (!obj->parsed) {
172 if (has_sha1_pack(obj->sha1))
173 return; /* it is in pack - forget about it */
174 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
175 errors_found |= ERROR_REACHABLE;
176 return;
177 }
178}
179
180/*
181 * Check a single unreachable object
182 */
183static void check_unreachable_object(struct object *obj)
184{
185 /*
186 * Missing unreachable object? Ignore it. It's not like
187 * we miss it (since it can't be reached), nor do we want
188 * to complain about it being unreachable (since it does
189 * not exist).
190 */
191 if (!obj->parsed)
192 return;
193
194 /*
195 * Unreachable object that exists? Show it if asked to,
196 * since this is something that is prunable.
197 */
198 if (show_unreachable) {
199 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
200 return;
201 }
202
203 /*
204 * "!used" means that nothing at all points to it, including
205 * other unreachable objects. In other words, it's the "tip"
206 * of some set of unreachable objects, usually a commit that
207 * got dropped.
208 *
209 * Such starting points are more interesting than some random
210 * set of unreachable objects, so we show them even if the user
211 * hasn't asked for _all_ unreachable objects. If you have
212 * deleted a branch by mistake, this is a prime candidate to
213 * start looking at, for example.
214 */
215 if (!obj->used) {
216 printf("dangling %s %s\n", typename(obj->type),
217 sha1_to_hex(obj->sha1));
218 if (write_lost_and_found) {
219 char *filename = git_path("lost-found/%s/%s",
220 obj->type == OBJ_COMMIT ? "commit" : "other",
221 sha1_to_hex(obj->sha1));
222 FILE *f;
223
224 if (safe_create_leading_directories(filename)) {
225 error("Could not create lost-found");
226 return;
227 }
228 if (!(f = fopen(filename, "w")))
229 die_errno("Could not open '%s'", filename);
230 if (obj->type == OBJ_BLOB) {
231 enum object_type type;
232 unsigned long size;
233 char *buf = read_sha1_file(obj->sha1,
234 &type, &size);
235 if (buf && fwrite(buf, 1, size, f) != size)
236 die_errno("Could not write '%s'", filename);
237 free(buf);
238 } else
239 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
240 if (fclose(f))
241 die_errno("Could not finish '%s'",
242 filename);
243 }
244 return;
245 }
246
247 /*
248 * Otherwise? It's there, it's unreachable, and some other unreachable
249 * object points to it. Ignore it - it's not interesting, and we showed
250 * all the interesting cases above.
251 */
252}
253
254static void check_object(struct object *obj)
255{
256 if (verbose)
257 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
258
259 if (obj->flags & REACHABLE)
260 check_reachable_object(obj);
261 else
262 check_unreachable_object(obj);
263}
264
265static void check_connectivity(void)
266{
267 int i, max;
268
269 /* Traverse the pending reachable objects */
270 traverse_reachable();
271
272 /* Look up all the requirements, warn about missing objects.. */
273 max = get_max_object_index();
274 if (verbose)
275 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
276
277 for (i = 0; i < max; i++) {
278 struct object *obj = get_indexed_object(i);
279
280 if (obj)
281 check_object(obj);
282 }
283}
284
285static int fsck_sha1(const unsigned char *sha1)
286{
287 struct object *obj = parse_object(sha1);
288 if (!obj) {
289 errors_found |= ERROR_OBJECT;
290 return error("%s: object corrupt or missing",
291 sha1_to_hex(sha1));
292 }
293 if (obj->flags & SEEN)
294 return 0;
295 obj->flags |= SEEN;
296
297 if (verbose)
298 fprintf(stderr, "Checking %s %s\n",
299 typename(obj->type), sha1_to_hex(obj->sha1));
300
301 if (fsck_walk(obj, mark_used, NULL))
302 objerror(obj, "broken links");
303 if (fsck_object(obj, check_strict, fsck_error_func))
304 return -1;
305
306 if (obj->type == OBJ_TREE) {
307 struct tree *item = (struct tree *) obj;
308
309 free(item->buffer);
310 item->buffer = NULL;
311 }
312
313 if (obj->type == OBJ_COMMIT) {
314 struct commit *commit = (struct commit *) obj;
315
316 free(commit->buffer);
317 commit->buffer = NULL;
318
319 if (!commit->parents && show_root)
320 printf("root %s\n", sha1_to_hex(commit->object.sha1));
321 }
322
323 if (obj->type == OBJ_TAG) {
324 struct tag *tag = (struct tag *) obj;
325
326 if (show_tags && tag->tagged) {
327 printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
328 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
329 }
330 }
331
332 return 0;
333}
334
335/*
336 * This is the sorting chunk size: make it reasonably
337 * big so that we can sort well..
338 */
339#define MAX_SHA1_ENTRIES (1024)
340
341struct sha1_entry {
342 unsigned long ino;
343 unsigned char sha1[20];
344};
345
346static struct {
347 unsigned long nr;
348 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
349} sha1_list;
350
351static int ino_compare(const void *_a, const void *_b)
352{
353 const struct sha1_entry *a = _a, *b = _b;
354 unsigned long ino1 = a->ino, ino2 = b->ino;
355 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
356}
357
358static void fsck_sha1_list(void)
359{
360 int i, nr = sha1_list.nr;
361
362 if (SORT_DIRENT)
363 qsort(sha1_list.entry, nr,
364 sizeof(struct sha1_entry *), ino_compare);
365 for (i = 0; i < nr; i++) {
366 struct sha1_entry *entry = sha1_list.entry[i];
367 unsigned char *sha1 = entry->sha1;
368
369 sha1_list.entry[i] = NULL;
370 fsck_sha1(sha1);
371 free(entry);
372 }
373 sha1_list.nr = 0;
374}
375
376static void add_sha1_list(unsigned char *sha1, unsigned long ino)
377{
378 struct sha1_entry *entry = xmalloc(sizeof(*entry));
379 int nr;
380
381 entry->ino = ino;
382 hashcpy(entry->sha1, sha1);
383 nr = sha1_list.nr;
384 if (nr == MAX_SHA1_ENTRIES) {
385 fsck_sha1_list();
386 nr = 0;
387 }
388 sha1_list.entry[nr] = entry;
389 sha1_list.nr = ++nr;
390}
391
392static inline int is_loose_object_file(struct dirent *de,
393 char *name, unsigned char *sha1)
394{
395 if (strlen(de->d_name) != 38)
396 return 0;
397 memcpy(name + 2, de->d_name, 39);
398 return !get_sha1_hex(name, sha1);
399}
400
401static void fsck_dir(int i, char *path)
402{
403 DIR *dir = opendir(path);
404 struct dirent *de;
405 char name[100];
406
407 if (!dir)
408 return;
409
410 if (verbose)
411 fprintf(stderr, "Checking directory %s\n", path);
412
413 sprintf(name, "%02x", i);
414 while ((de = readdir(dir)) != NULL) {
415 unsigned char sha1[20];
416
417 if (is_dot_or_dotdot(de->d_name))
418 continue;
419 if (is_loose_object_file(de, name, sha1)) {
420 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
421 continue;
422 }
423 if (!prefixcmp(de->d_name, "tmp_obj_"))
424 continue;
425 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
426 }
427 closedir(dir);
428}
429
430static int default_refs;
431
432static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
433 const char *email, unsigned long timestamp, int tz,
434 const char *message, void *cb_data)
435{
436 struct object *obj;
437
438 if (verbose)
439 fprintf(stderr, "Checking reflog %s->%s\n",
440 sha1_to_hex(osha1), sha1_to_hex(nsha1));
441
442 if (!is_null_sha1(osha1)) {
443 obj = lookup_object(osha1);
444 if (obj) {
445 obj->used = 1;
446 mark_object_reachable(obj);
447 }
448 }
449 obj = lookup_object(nsha1);
450 if (obj) {
451 obj->used = 1;
452 mark_object_reachable(obj);
453 }
454 return 0;
455}
456
457static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
458{
459 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
460 return 0;
461}
462
463static int is_branch(const char *refname)
464{
465 return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
466}
467
468static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
469{
470 struct object *obj;
471
472 obj = parse_object(sha1);
473 if (!obj) {
474 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
475 /* We'll continue with the rest despite the error.. */
476 return 0;
477 }
478 if (obj->type != OBJ_COMMIT && is_branch(refname))
479 error("%s: not a commit", refname);
480 default_refs++;
481 obj->used = 1;
482 mark_object_reachable(obj);
483
484 return 0;
485}
486
487static void get_default_heads(void)
488{
489 if (head_points_at && !is_null_sha1(head_sha1))
490 fsck_handle_ref("HEAD", head_sha1, 0, NULL);
491 for_each_ref(fsck_handle_ref, NULL);
492 if (include_reflogs)
493 for_each_reflog(fsck_handle_reflog, NULL);
494
495 /*
496 * Not having any default heads isn't really fatal, but
497 * it does mean that "--unreachable" no longer makes any
498 * sense (since in this case everything will obviously
499 * be unreachable by definition.
500 *
501 * Showing dangling objects is valid, though (as those
502 * dangling objects are likely lost heads).
503 *
504 * So we just print a warning about it, and clear the
505 * "show_unreachable" flag.
506 */
507 if (!default_refs) {
508 fprintf(stderr, "notice: No default references\n");
509 show_unreachable = 0;
510 }
511}
512
513static void fsck_object_dir(const char *path)
514{
515 int i;
516
517 if (verbose)
518 fprintf(stderr, "Checking object directory\n");
519
520 for (i = 0; i < 256; i++) {
521 static char dir[4096];
522 sprintf(dir, "%s/%02x", path, i);
523 fsck_dir(i, dir);
524 }
525 fsck_sha1_list();
526}
527
528static int fsck_head_link(void)
529{
530 int flag;
531 int null_is_error = 0;
532
533 if (verbose)
534 fprintf(stderr, "Checking HEAD link\n");
535
536 head_points_at = resolve_ref("HEAD", head_sha1, 0, &flag);
537 if (!head_points_at)
538 return error("Invalid HEAD");
539 if (!strcmp(head_points_at, "HEAD"))
540 /* detached HEAD */
541 null_is_error = 1;
542 else if (prefixcmp(head_points_at, "refs/heads/"))
543 return error("HEAD points to something strange (%s)",
544 head_points_at);
545 if (is_null_sha1(head_sha1)) {
546 if (null_is_error)
547 return error("HEAD: detached HEAD points at nothing");
548 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
549 head_points_at + 11);
550 }
551 return 0;
552}
553
554static int fsck_cache_tree(struct cache_tree *it)
555{
556 int i;
557 int err = 0;
558
559 if (verbose)
560 fprintf(stderr, "Checking cache tree\n");
561
562 if (0 <= it->entry_count) {
563 struct object *obj = parse_object(it->sha1);
564 if (!obj) {
565 error("%s: invalid sha1 pointer in cache-tree",
566 sha1_to_hex(it->sha1));
567 return 1;
568 }
569 obj->used = 1;
570 mark_object_reachable(obj);
571 if (obj->type != OBJ_TREE)
572 err |= objerror(obj, "non-tree in cache-tree");
573 }
574 for (i = 0; i < it->subtree_nr; i++)
575 err |= fsck_cache_tree(it->down[i]->cache_tree);
576 return err;
577}
578
579static char const * const fsck_usage[] = {
580 "git fsck [options] [<object>...]",
581 NULL
582};
583
584static struct option fsck_opts[] = {
585 OPT__VERBOSE(&verbose, "be verbose"),
586 OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
587 OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
588 OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
589 OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
590 OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
591 OPT_BOOLEAN(0, "full", &check_full, "also consider packs and alternate objects"),
592 OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
593 OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
594 "write dangling objects in .git/lost-found"),
595 OPT_END(),
596};
597
598int cmd_fsck(int argc, const char **argv, const char *prefix)
599{
600 int i, heads;
601 struct alternate_object_database *alt;
602
603 errors_found = 0;
604 read_replace_refs = 0;
605
606 argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
607 if (write_lost_and_found) {
608 check_full = 1;
609 include_reflogs = 0;
610 }
611
612 fsck_head_link();
613 fsck_object_dir(get_object_directory());
614
615 prepare_alt_odb();
616 for (alt = alt_odb_list; alt; alt = alt->next) {
617 char namebuf[PATH_MAX];
618 int namelen = alt->name - alt->base;
619 memcpy(namebuf, alt->base, namelen);
620 namebuf[namelen - 1] = 0;
621 fsck_object_dir(namebuf);
622 }
623
624 if (check_full) {
625 struct packed_git *p;
626
627 prepare_packed_git();
628 for (p = packed_git; p; p = p->next)
629 /* verify gives error messages itself */
630 if (verify_pack(p))
631 errors_found |= ERROR_PACK;
632
633 for (p = packed_git; p; p = p->next) {
634 uint32_t j, num;
635 if (open_pack_index(p))
636 continue;
637 num = p->num_objects;
638 for (j = 0; j < num; j++)
639 fsck_sha1(nth_packed_object_sha1(p, j));
640 }
641 }
642
643 heads = 0;
644 for (i = 0; i < argc; i++) {
645 const char *arg = argv[i];
646 unsigned char sha1[20];
647 if (!get_sha1(arg, sha1)) {
648 struct object *obj = lookup_object(sha1);
649
650 /* Error is printed by lookup_object(). */
651 if (!obj)
652 continue;
653
654 obj->used = 1;
655 mark_object_reachable(obj);
656 heads++;
657 continue;
658 }
659 error("invalid parameter: expected sha1, got '%s'", arg);
660 }
661
662 /*
663 * If we've not been given any explicit head information, do the
664 * default ones from .git/refs. We also consider the index file
665 * in this case (ie this implies --cache).
666 */
667 if (!heads) {
668 get_default_heads();
669 keep_cache_objects = 1;
670 }
671
672 if (keep_cache_objects) {
673 read_cache();
674 for (i = 0; i < active_nr; i++) {
675 unsigned int mode;
676 struct blob *blob;
677 struct object *obj;
678
679 mode = active_cache[i]->ce_mode;
680 if (S_ISGITLINK(mode))
681 continue;
682 blob = lookup_blob(active_cache[i]->sha1);
683 if (!blob)
684 continue;
685 obj = &blob->object;
686 obj->used = 1;
687 mark_object_reachable(obj);
688 }
689 if (active_cache_tree)
690 fsck_cache_tree(active_cache_tree);
691 }
692
693 check_connectivity();
694 return errors_found;
695}