Initialize tree descriptors with a helper function rather than by hand.
[git/git.git] / builtin-fsck.c
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"
10
11 #define REACHABLE 0x0001
12 #define SEEN 0x0002
13
14 static int show_root;
15 static int show_tags;
16 static int show_unreachable;
17 static int check_full;
18 static int check_strict;
19 static int keep_cache_objects;
20 static unsigned char head_sha1[20];
21 static int errors_found;
22 #define ERROR_OBJECT 01
23 #define ERROR_REACHABLE 02
24
25 #ifdef NO_D_INO_IN_DIRENT
26 #define SORT_DIRENT 0
27 #define DIRENT_SORT_HINT(de) 0
28 #else
29 #define SORT_DIRENT 1
30 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
31 #endif
32
33 static void objreport(struct object *obj, const char *severity,
34 const char *err, va_list params)
35 {
36 fprintf(stderr, "%s in %s %s: ",
37 severity, typename(obj->type), sha1_to_hex(obj->sha1));
38 vfprintf(stderr, err, params);
39 fputs("\n", stderr);
40 }
41
42 static int objerror(struct object *obj, const char *err, ...)
43 {
44 va_list params;
45 va_start(params, err);
46 errors_found |= ERROR_OBJECT;
47 objreport(obj, "error", err, params);
48 va_end(params);
49 return -1;
50 }
51
52 static int objwarning(struct object *obj, const char *err, ...)
53 {
54 va_list params;
55 va_start(params, err);
56 objreport(obj, "warning", err, params);
57 va_end(params);
58 return -1;
59 }
60
61 /*
62 * Check a single reachable object
63 */
64 static void check_reachable_object(struct object *obj)
65 {
66 const struct object_refs *refs;
67
68 /*
69 * We obviously want the object to be parsed,
70 * except if it was in a pack-file and we didn't
71 * do a full fsck
72 */
73 if (!obj->parsed) {
74 if (has_sha1_pack(obj->sha1, NULL))
75 return; /* it is in pack - forget about it */
76 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
77 errors_found |= ERROR_REACHABLE;
78 return;
79 }
80
81 /*
82 * Check that everything that we try to reference is also good.
83 */
84 refs = lookup_object_refs(obj);
85 if (refs) {
86 unsigned j;
87 for (j = 0; j < refs->count; j++) {
88 struct object *ref = refs->ref[j];
89 if (ref->parsed ||
90 (has_sha1_file(ref->sha1)))
91 continue;
92 printf("broken link from %7s %s\n",
93 typename(obj->type), sha1_to_hex(obj->sha1));
94 printf(" to %7s %s\n",
95 typename(ref->type), sha1_to_hex(ref->sha1));
96 errors_found |= ERROR_REACHABLE;
97 }
98 }
99 }
100
101 /*
102 * Check a single unreachable object
103 */
104 static void check_unreachable_object(struct object *obj)
105 {
106 /*
107 * Missing unreachable object? Ignore it. It's not like
108 * we miss it (since it can't be reached), nor do we want
109 * to complain about it being unreachable (since it does
110 * not exist).
111 */
112 if (!obj->parsed)
113 return;
114
115 /*
116 * Unreachable object that exists? Show it if asked to,
117 * since this is something that is prunable.
118 */
119 if (show_unreachable) {
120 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
121 return;
122 }
123
124 /*
125 * "!used" means that nothing at all points to it, including
126 * other unreachable objects. In other words, it's the "tip"
127 * of some set of unreachable objects, usually a commit that
128 * got dropped.
129 *
130 * Such starting points are more interesting than some random
131 * set of unreachable objects, so we show them even if the user
132 * hasn't asked for _all_ unreachable objects. If you have
133 * deleted a branch by mistake, this is a prime candidate to
134 * start looking at, for example.
135 */
136 if (!obj->used) {
137 printf("dangling %s %s\n", typename(obj->type),
138 sha1_to_hex(obj->sha1));
139 return;
140 }
141
142 /*
143 * Otherwise? It's there, it's unreachable, and some other unreachable
144 * object points to it. Ignore it - it's not interesting, and we showed
145 * all the interesting cases above.
146 */
147 }
148
149 static void check_object(struct object *obj)
150 {
151 if (obj->flags & REACHABLE)
152 check_reachable_object(obj);
153 else
154 check_unreachable_object(obj);
155 }
156
157 static void check_connectivity(void)
158 {
159 int i, max;
160
161 /* Look up all the requirements, warn about missing objects.. */
162 max = get_max_object_index();
163 for (i = 0; i < max; i++) {
164 struct object *obj = get_indexed_object(i);
165
166 if (obj)
167 check_object(obj);
168 }
169 }
170
171 /*
172 * The entries in a tree are ordered in the _path_ order,
173 * which means that a directory entry is ordered by adding
174 * a slash to the end of it.
175 *
176 * So a directory called "a" is ordered _after_ a file
177 * called "a.c", because "a/" sorts after "a.c".
178 */
179 #define TREE_UNORDERED (-1)
180 #define TREE_HAS_DUPS (-2)
181
182 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
183 {
184 int len1 = strlen(name1);
185 int len2 = strlen(name2);
186 int len = len1 < len2 ? len1 : len2;
187 unsigned char c1, c2;
188 int cmp;
189
190 cmp = memcmp(name1, name2, len);
191 if (cmp < 0)
192 return 0;
193 if (cmp > 0)
194 return TREE_UNORDERED;
195
196 /*
197 * Ok, the first <len> characters are the same.
198 * Now we need to order the next one, but turn
199 * a '\0' into a '/' for a directory entry.
200 */
201 c1 = name1[len];
202 c2 = name2[len];
203 if (!c1 && !c2)
204 /*
205 * git-write-tree used to write out a nonsense tree that has
206 * entries with the same name, one blob and one tree. Make
207 * sure we do not have duplicate entries.
208 */
209 return TREE_HAS_DUPS;
210 if (!c1 && S_ISDIR(mode1))
211 c1 = '/';
212 if (!c2 && S_ISDIR(mode2))
213 c2 = '/';
214 return c1 < c2 ? 0 : TREE_UNORDERED;
215 }
216
217 static int fsck_tree(struct tree *item)
218 {
219 int retval;
220 int has_full_path = 0;
221 int has_zero_pad = 0;
222 int has_bad_modes = 0;
223 int has_dup_entries = 0;
224 int not_properly_sorted = 0;
225 struct tree_desc desc;
226 unsigned o_mode;
227 const char *o_name;
228 const unsigned char *o_sha1;
229
230 init_tree_desc(&desc, item->buffer, item->size);
231
232 o_mode = 0;
233 o_name = NULL;
234 o_sha1 = NULL;
235 while (desc.size) {
236 unsigned mode;
237 const char *name;
238 const unsigned char *sha1;
239
240 sha1 = tree_entry_extract(&desc, &name, &mode);
241
242 if (strchr(name, '/'))
243 has_full_path = 1;
244 has_zero_pad |= *(char *)desc.buffer == '0';
245 update_tree_entry(&desc);
246
247 switch (mode) {
248 /*
249 * Standard modes..
250 */
251 case S_IFREG | 0755:
252 case S_IFREG | 0644:
253 case S_IFLNK:
254 case S_IFDIR:
255 break;
256 /*
257 * This is nonstandard, but we had a few of these
258 * early on when we honored the full set of mode
259 * bits..
260 */
261 case S_IFREG | 0664:
262 if (!check_strict)
263 break;
264 default:
265 has_bad_modes = 1;
266 }
267
268 if (o_name) {
269 switch (verify_ordered(o_mode, o_name, mode, name)) {
270 case TREE_UNORDERED:
271 not_properly_sorted = 1;
272 break;
273 case TREE_HAS_DUPS:
274 has_dup_entries = 1;
275 break;
276 default:
277 break;
278 }
279 }
280
281 o_mode = mode;
282 o_name = name;
283 o_sha1 = sha1;
284 }
285 free(item->buffer);
286 item->buffer = NULL;
287
288 retval = 0;
289 if (has_full_path) {
290 objwarning(&item->object, "contains full pathnames");
291 }
292 if (has_zero_pad) {
293 objwarning(&item->object, "contains zero-padded file modes");
294 }
295 if (has_bad_modes) {
296 objwarning(&item->object, "contains bad file modes");
297 }
298 if (has_dup_entries) {
299 retval = objerror(&item->object, "contains duplicate file entries");
300 }
301 if (not_properly_sorted) {
302 retval = objerror(&item->object, "not properly sorted");
303 }
304 return retval;
305 }
306
307 static int fsck_commit(struct commit *commit)
308 {
309 char *buffer = commit->buffer;
310 unsigned char tree_sha1[20], sha1[20];
311
312 if (memcmp(buffer, "tree ", 5))
313 return objerror(&commit->object, "invalid format - expected 'tree' line");
314 if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
315 return objerror(&commit->object, "invalid 'tree' line format - bad sha1");
316 buffer += 46;
317 while (!memcmp(buffer, "parent ", 7)) {
318 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
319 return objerror(&commit->object, "invalid 'parent' line format - bad sha1");
320 buffer += 48;
321 }
322 if (memcmp(buffer, "author ", 7))
323 return objerror(&commit->object, "invalid format - expected 'author' line");
324 free(commit->buffer);
325 commit->buffer = NULL;
326 if (!commit->tree)
327 return objerror(&commit->object, "could not load commit's tree %s", tree_sha1);
328 if (!commit->parents && show_root)
329 printf("root %s\n", sha1_to_hex(commit->object.sha1));
330 if (!commit->date)
331 printf("bad commit date in %s\n",
332 sha1_to_hex(commit->object.sha1));
333 return 0;
334 }
335
336 static int fsck_tag(struct tag *tag)
337 {
338 struct object *tagged = tag->tagged;
339
340 if (!tagged) {
341 return objerror(&tag->object, "could not load tagged object");
342 }
343 if (!show_tags)
344 return 0;
345
346 printf("tagged %s %s", typename(tagged->type), sha1_to_hex(tagged->sha1));
347 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
348 return 0;
349 }
350
351 static int fsck_sha1(unsigned char *sha1)
352 {
353 struct object *obj = parse_object(sha1);
354 if (!obj) {
355 errors_found |= ERROR_OBJECT;
356 return error("%s: object corrupt or missing",
357 sha1_to_hex(sha1));
358 }
359 if (obj->flags & SEEN)
360 return 0;
361 obj->flags |= SEEN;
362 if (obj->type == OBJ_BLOB)
363 return 0;
364 if (obj->type == OBJ_TREE)
365 return fsck_tree((struct tree *) obj);
366 if (obj->type == OBJ_COMMIT)
367 return fsck_commit((struct commit *) obj);
368 if (obj->type == OBJ_TAG)
369 return fsck_tag((struct tag *) obj);
370
371 /* By now, parse_object() would've returned NULL instead. */
372 return objerror(obj, "unknown type '%d' (internal fsck error)",
373 obj->type);
374 }
375
376 /*
377 * This is the sorting chunk size: make it reasonably
378 * big so that we can sort well..
379 */
380 #define MAX_SHA1_ENTRIES (1024)
381
382 struct sha1_entry {
383 unsigned long ino;
384 unsigned char sha1[20];
385 };
386
387 static struct {
388 unsigned long nr;
389 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
390 } sha1_list;
391
392 static int ino_compare(const void *_a, const void *_b)
393 {
394 const struct sha1_entry *a = _a, *b = _b;
395 unsigned long ino1 = a->ino, ino2 = b->ino;
396 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
397 }
398
399 static void fsck_sha1_list(void)
400 {
401 int i, nr = sha1_list.nr;
402
403 if (SORT_DIRENT)
404 qsort(sha1_list.entry, nr,
405 sizeof(struct sha1_entry *), ino_compare);
406 for (i = 0; i < nr; i++) {
407 struct sha1_entry *entry = sha1_list.entry[i];
408 unsigned char *sha1 = entry->sha1;
409
410 sha1_list.entry[i] = NULL;
411 fsck_sha1(sha1);
412 free(entry);
413 }
414 sha1_list.nr = 0;
415 }
416
417 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
418 {
419 struct sha1_entry *entry = xmalloc(sizeof(*entry));
420 int nr;
421
422 entry->ino = ino;
423 hashcpy(entry->sha1, sha1);
424 nr = sha1_list.nr;
425 if (nr == MAX_SHA1_ENTRIES) {
426 fsck_sha1_list();
427 nr = 0;
428 }
429 sha1_list.entry[nr] = entry;
430 sha1_list.nr = ++nr;
431 }
432
433 static void fsck_dir(int i, char *path)
434 {
435 DIR *dir = opendir(path);
436 struct dirent *de;
437
438 if (!dir)
439 return;
440
441 while ((de = readdir(dir)) != NULL) {
442 char name[100];
443 unsigned char sha1[20];
444 int len = strlen(de->d_name);
445
446 switch (len) {
447 case 2:
448 if (de->d_name[1] != '.')
449 break;
450 case 1:
451 if (de->d_name[0] != '.')
452 break;
453 continue;
454 case 38:
455 sprintf(name, "%02x", i);
456 memcpy(name+2, de->d_name, len+1);
457 if (get_sha1_hex(name, sha1) < 0)
458 break;
459 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
460 continue;
461 }
462 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
463 }
464 closedir(dir);
465 }
466
467 static int default_refs;
468
469 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
470 const char *email, unsigned long timestamp, int tz,
471 const char *message, void *cb_data)
472 {
473 struct object *obj;
474
475 if (!is_null_sha1(osha1)) {
476 obj = lookup_object(osha1);
477 if (obj) {
478 obj->used = 1;
479 mark_reachable(obj, REACHABLE);
480 }
481 }
482 obj = lookup_object(nsha1);
483 if (obj) {
484 obj->used = 1;
485 mark_reachable(obj, REACHABLE);
486 }
487 return 0;
488 }
489
490 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
491 {
492 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
493 return 0;
494 }
495
496 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
497 {
498 struct object *obj;
499
500 obj = lookup_object(sha1);
501 if (!obj) {
502 if (has_sha1_file(sha1)) {
503 default_refs++;
504 return 0; /* it is in a pack */
505 }
506 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
507 /* We'll continue with the rest despite the error.. */
508 return 0;
509 }
510 default_refs++;
511 obj->used = 1;
512 mark_reachable(obj, REACHABLE);
513
514 return 0;
515 }
516
517 static void get_default_heads(void)
518 {
519 for_each_ref(fsck_handle_ref, NULL);
520 for_each_reflog(fsck_handle_reflog, NULL);
521
522 /*
523 * Not having any default heads isn't really fatal, but
524 * it does mean that "--unreachable" no longer makes any
525 * sense (since in this case everything will obviously
526 * be unreachable by definition.
527 *
528 * Showing dangling objects is valid, though (as those
529 * dangling objects are likely lost heads).
530 *
531 * So we just print a warning about it, and clear the
532 * "show_unreachable" flag.
533 */
534 if (!default_refs) {
535 error("No default references");
536 show_unreachable = 0;
537 }
538 }
539
540 static void fsck_object_dir(const char *path)
541 {
542 int i;
543 for (i = 0; i < 256; i++) {
544 static char dir[4096];
545 sprintf(dir, "%s/%02x", path, i);
546 fsck_dir(i, dir);
547 }
548 fsck_sha1_list();
549 }
550
551 static int fsck_head_link(void)
552 {
553 unsigned char sha1[20];
554 int flag;
555 const char *head_points_at = resolve_ref("HEAD", sha1, 1, &flag);
556
557 if (!head_points_at || !(flag & REF_ISSYMREF))
558 return error("HEAD is not a symbolic ref");
559 if (prefixcmp(head_points_at, "refs/heads/"))
560 return error("HEAD points to something strange (%s)",
561 head_points_at);
562 if (is_null_sha1(sha1))
563 return error("HEAD: not a valid git pointer");
564 return 0;
565 }
566
567 static int fsck_cache_tree(struct cache_tree *it)
568 {
569 int i;
570 int err = 0;
571
572 if (0 <= it->entry_count) {
573 struct object *obj = parse_object(it->sha1);
574 if (!obj) {
575 error("%s: invalid sha1 pointer in cache-tree",
576 sha1_to_hex(it->sha1));
577 return 1;
578 }
579 mark_reachable(obj, REACHABLE);
580 obj->used = 1;
581 if (obj->type != OBJ_TREE)
582 err |= objerror(obj, "non-tree in cache-tree");
583 }
584 for (i = 0; i < it->subtree_nr; i++)
585 err |= fsck_cache_tree(it->down[i]->cache_tree);
586 return err;
587 }
588
589 static const char fsck_usage[] =
590 "git-fsck [--tags] [--root] [[--unreachable] [--cache] [--full] "
591 "[--strict] <head-sha1>*]";
592
593 int cmd_fsck(int argc, char **argv, const char *prefix)
594 {
595 int i, heads;
596
597 track_object_refs = 1;
598 errors_found = 0;
599
600 for (i = 1; i < argc; i++) {
601 const char *arg = argv[i];
602
603 if (!strcmp(arg, "--unreachable")) {
604 show_unreachable = 1;
605 continue;
606 }
607 if (!strcmp(arg, "--tags")) {
608 show_tags = 1;
609 continue;
610 }
611 if (!strcmp(arg, "--root")) {
612 show_root = 1;
613 continue;
614 }
615 if (!strcmp(arg, "--cache")) {
616 keep_cache_objects = 1;
617 continue;
618 }
619 if (!strcmp(arg, "--full")) {
620 check_full = 1;
621 continue;
622 }
623 if (!strcmp(arg, "--strict")) {
624 check_strict = 1;
625 continue;
626 }
627 if (*arg == '-')
628 usage(fsck_usage);
629 }
630
631 fsck_head_link();
632 fsck_object_dir(get_object_directory());
633 if (check_full) {
634 struct alternate_object_database *alt;
635 struct packed_git *p;
636 prepare_alt_odb();
637 for (alt = alt_odb_list; alt; alt = alt->next) {
638 char namebuf[PATH_MAX];
639 int namelen = alt->name - alt->base;
640 memcpy(namebuf, alt->base, namelen);
641 namebuf[namelen - 1] = 0;
642 fsck_object_dir(namebuf);
643 }
644 prepare_packed_git();
645 for (p = packed_git; p; p = p->next)
646 /* verify gives error messages itself */
647 verify_pack(p, 0);
648
649 for (p = packed_git; p; p = p->next) {
650 uint32_t i, num = num_packed_objects(p);
651 for (i = 0; i < num; i++) {
652 unsigned char sha1[20];
653 nth_packed_object_sha1(p, i, sha1);
654 fsck_sha1(sha1);
655 }
656 }
657 }
658
659 heads = 0;
660 for (i = 1; i < argc; i++) {
661 const char *arg = argv[i];
662
663 if (*arg == '-')
664 continue;
665
666 if (!get_sha1(arg, head_sha1)) {
667 struct object *obj = lookup_object(head_sha1);
668
669 /* Error is printed by lookup_object(). */
670 if (!obj)
671 continue;
672
673 obj->used = 1;
674 mark_reachable(obj, REACHABLE);
675 heads++;
676 continue;
677 }
678 error("invalid parameter: expected sha1, got '%s'", arg);
679 }
680
681 /*
682 * If we've not been given any explicit head information, do the
683 * default ones from .git/refs. We also consider the index file
684 * in this case (ie this implies --cache).
685 */
686 if (!heads) {
687 get_default_heads();
688 keep_cache_objects = 1;
689 }
690
691 if (keep_cache_objects) {
692 int i;
693 read_cache();
694 for (i = 0; i < active_nr; i++) {
695 struct blob *blob = lookup_blob(active_cache[i]->sha1);
696 struct object *obj;
697 if (!blob)
698 continue;
699 obj = &blob->object;
700 obj->used = 1;
701 mark_reachable(obj, REACHABLE);
702 }
703 if (active_cache_tree)
704 fsck_cache_tree(active_cache_tree);
705 }
706
707 check_connectivity();
708 return errors_found;
709 }