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