fsck: avoid reading every object twice
[git/git.git] / builtin / fsck.c
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
18 static int show_root;
19 static int show_tags;
20 static int show_unreachable;
21 static int include_reflogs = 1;
22 static int check_full = 1;
23 static int check_strict;
24 static int keep_cache_objects;
25 static unsigned char head_sha1[20];
26 static const char *head_points_at;
27 static int errors_found;
28 static int write_lost_and_found;
29 static 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
42 static 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)))
52 static 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)))
63 static 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
72 static struct object_array pending;
73
74 static 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
115 static void mark_object_reachable(struct object *obj)
116 {
117 mark_object(obj, OBJ_ANY, NULL);
118 }
119
120 static 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
139 static 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
153 static 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 */
164 static 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 */
183 static 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
254 static 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
265 static 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
285 static int fsck_obj(struct object *obj)
286 {
287 if (obj->flags & SEEN)
288 return 0;
289 obj->flags |= SEEN;
290
291 if (verbose)
292 fprintf(stderr, "Checking %s %s\n",
293 typename(obj->type), sha1_to_hex(obj->sha1));
294
295 if (fsck_walk(obj, mark_used, NULL))
296 objerror(obj, "broken links");
297 if (fsck_object(obj, check_strict, fsck_error_func))
298 return -1;
299
300 if (obj->type == OBJ_TREE) {
301 struct tree *item = (struct tree *) obj;
302
303 free(item->buffer);
304 item->buffer = NULL;
305 }
306
307 if (obj->type == OBJ_COMMIT) {
308 struct commit *commit = (struct commit *) obj;
309
310 free(commit->buffer);
311 commit->buffer = NULL;
312
313 if (!commit->parents && show_root)
314 printf("root %s\n", sha1_to_hex(commit->object.sha1));
315 }
316
317 if (obj->type == OBJ_TAG) {
318 struct tag *tag = (struct tag *) obj;
319
320 if (show_tags && tag->tagged) {
321 printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
322 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
323 }
324 }
325
326 return 0;
327 }
328
329 static int fsck_sha1(const unsigned char *sha1)
330 {
331 struct object *obj = parse_object(sha1);
332 if (!obj) {
333 errors_found |= ERROR_OBJECT;
334 return error("%s: object corrupt or missing",
335 sha1_to_hex(sha1));
336 }
337 return fsck_obj(obj);
338 }
339
340 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
341 unsigned long size, void *buffer, int *eaten)
342 {
343 struct object *obj;
344 obj = parse_object_buffer(sha1, type, size, buffer, eaten);
345 if (!obj) {
346 errors_found |= ERROR_OBJECT;
347 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
348 }
349 return fsck_obj(obj);
350 }
351
352 /*
353 * This is the sorting chunk size: make it reasonably
354 * big so that we can sort well..
355 */
356 #define MAX_SHA1_ENTRIES (1024)
357
358 struct sha1_entry {
359 unsigned long ino;
360 unsigned char sha1[20];
361 };
362
363 static struct {
364 unsigned long nr;
365 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
366 } sha1_list;
367
368 static int ino_compare(const void *_a, const void *_b)
369 {
370 const struct sha1_entry *a = _a, *b = _b;
371 unsigned long ino1 = a->ino, ino2 = b->ino;
372 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
373 }
374
375 static void fsck_sha1_list(void)
376 {
377 int i, nr = sha1_list.nr;
378
379 if (SORT_DIRENT)
380 qsort(sha1_list.entry, nr,
381 sizeof(struct sha1_entry *), ino_compare);
382 for (i = 0; i < nr; i++) {
383 struct sha1_entry *entry = sha1_list.entry[i];
384 unsigned char *sha1 = entry->sha1;
385
386 sha1_list.entry[i] = NULL;
387 fsck_sha1(sha1);
388 free(entry);
389 }
390 sha1_list.nr = 0;
391 }
392
393 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
394 {
395 struct sha1_entry *entry = xmalloc(sizeof(*entry));
396 int nr;
397
398 entry->ino = ino;
399 hashcpy(entry->sha1, sha1);
400 nr = sha1_list.nr;
401 if (nr == MAX_SHA1_ENTRIES) {
402 fsck_sha1_list();
403 nr = 0;
404 }
405 sha1_list.entry[nr] = entry;
406 sha1_list.nr = ++nr;
407 }
408
409 static inline int is_loose_object_file(struct dirent *de,
410 char *name, unsigned char *sha1)
411 {
412 if (strlen(de->d_name) != 38)
413 return 0;
414 memcpy(name + 2, de->d_name, 39);
415 return !get_sha1_hex(name, sha1);
416 }
417
418 static void fsck_dir(int i, char *path)
419 {
420 DIR *dir = opendir(path);
421 struct dirent *de;
422 char name[100];
423
424 if (!dir)
425 return;
426
427 if (verbose)
428 fprintf(stderr, "Checking directory %s\n", path);
429
430 sprintf(name, "%02x", i);
431 while ((de = readdir(dir)) != NULL) {
432 unsigned char sha1[20];
433
434 if (is_dot_or_dotdot(de->d_name))
435 continue;
436 if (is_loose_object_file(de, name, sha1)) {
437 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
438 continue;
439 }
440 if (!prefixcmp(de->d_name, "tmp_obj_"))
441 continue;
442 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
443 }
444 closedir(dir);
445 }
446
447 static int default_refs;
448
449 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
450 const char *email, unsigned long timestamp, int tz,
451 const char *message, void *cb_data)
452 {
453 struct object *obj;
454
455 if (verbose)
456 fprintf(stderr, "Checking reflog %s->%s\n",
457 sha1_to_hex(osha1), sha1_to_hex(nsha1));
458
459 if (!is_null_sha1(osha1)) {
460 obj = lookup_object(osha1);
461 if (obj) {
462 obj->used = 1;
463 mark_object_reachable(obj);
464 }
465 }
466 obj = lookup_object(nsha1);
467 if (obj) {
468 obj->used = 1;
469 mark_object_reachable(obj);
470 }
471 return 0;
472 }
473
474 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
475 {
476 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
477 return 0;
478 }
479
480 static int is_branch(const char *refname)
481 {
482 return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
483 }
484
485 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
486 {
487 struct object *obj;
488
489 obj = parse_object(sha1);
490 if (!obj) {
491 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
492 /* We'll continue with the rest despite the error.. */
493 return 0;
494 }
495 if (obj->type != OBJ_COMMIT && is_branch(refname))
496 error("%s: not a commit", refname);
497 default_refs++;
498 obj->used = 1;
499 mark_object_reachable(obj);
500
501 return 0;
502 }
503
504 static void get_default_heads(void)
505 {
506 if (head_points_at && !is_null_sha1(head_sha1))
507 fsck_handle_ref("HEAD", head_sha1, 0, NULL);
508 for_each_ref(fsck_handle_ref, NULL);
509 if (include_reflogs)
510 for_each_reflog(fsck_handle_reflog, NULL);
511
512 /*
513 * Not having any default heads isn't really fatal, but
514 * it does mean that "--unreachable" no longer makes any
515 * sense (since in this case everything will obviously
516 * be unreachable by definition.
517 *
518 * Showing dangling objects is valid, though (as those
519 * dangling objects are likely lost heads).
520 *
521 * So we just print a warning about it, and clear the
522 * "show_unreachable" flag.
523 */
524 if (!default_refs) {
525 fprintf(stderr, "notice: No default references\n");
526 show_unreachable = 0;
527 }
528 }
529
530 static void fsck_object_dir(const char *path)
531 {
532 int i;
533
534 if (verbose)
535 fprintf(stderr, "Checking object directory\n");
536
537 for (i = 0; i < 256; i++) {
538 static char dir[4096];
539 sprintf(dir, "%s/%02x", path, i);
540 fsck_dir(i, dir);
541 }
542 fsck_sha1_list();
543 }
544
545 static int fsck_head_link(void)
546 {
547 int flag;
548 int null_is_error = 0;
549
550 if (verbose)
551 fprintf(stderr, "Checking HEAD link\n");
552
553 head_points_at = resolve_ref("HEAD", head_sha1, 0, &flag);
554 if (!head_points_at)
555 return error("Invalid HEAD");
556 if (!strcmp(head_points_at, "HEAD"))
557 /* detached HEAD */
558 null_is_error = 1;
559 else 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(head_sha1)) {
563 if (null_is_error)
564 return error("HEAD: detached HEAD points at nothing");
565 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
566 head_points_at + 11);
567 }
568 return 0;
569 }
570
571 static int fsck_cache_tree(struct cache_tree *it)
572 {
573 int i;
574 int err = 0;
575
576 if (verbose)
577 fprintf(stderr, "Checking cache tree\n");
578
579 if (0 <= it->entry_count) {
580 struct object *obj = parse_object(it->sha1);
581 if (!obj) {
582 error("%s: invalid sha1 pointer in cache-tree",
583 sha1_to_hex(it->sha1));
584 return 1;
585 }
586 obj->used = 1;
587 mark_object_reachable(obj);
588 if (obj->type != OBJ_TREE)
589 err |= objerror(obj, "non-tree in cache-tree");
590 }
591 for (i = 0; i < it->subtree_nr; i++)
592 err |= fsck_cache_tree(it->down[i]->cache_tree);
593 return err;
594 }
595
596 static char const * const fsck_usage[] = {
597 "git fsck [options] [<object>...]",
598 NULL
599 };
600
601 static struct option fsck_opts[] = {
602 OPT__VERBOSE(&verbose, "be verbose"),
603 OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
604 OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
605 OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
606 OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
607 OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
608 OPT_BOOLEAN(0, "full", &check_full, "also consider packs and alternate objects"),
609 OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
610 OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
611 "write dangling objects in .git/lost-found"),
612 OPT_END(),
613 };
614
615 int cmd_fsck(int argc, const char **argv, const char *prefix)
616 {
617 int i, heads;
618 struct alternate_object_database *alt;
619
620 errors_found = 0;
621 read_replace_refs = 0;
622
623 argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
624 if (write_lost_and_found) {
625 check_full = 1;
626 include_reflogs = 0;
627 }
628
629 fsck_head_link();
630 fsck_object_dir(get_object_directory());
631
632 prepare_alt_odb();
633 for (alt = alt_odb_list; alt; alt = alt->next) {
634 char namebuf[PATH_MAX];
635 int namelen = alt->name - alt->base;
636 memcpy(namebuf, alt->base, namelen);
637 namebuf[namelen - 1] = 0;
638 fsck_object_dir(namebuf);
639 }
640
641 if (check_full) {
642 struct packed_git *p;
643
644 prepare_packed_git();
645 for (p = packed_git; p; p = p->next)
646 /* verify gives error messages itself */
647 if (verify_pack(p, fsck_obj_buffer))
648 errors_found |= ERROR_PACK;
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 }