Merge branch 'jk/git-dir-lookup'
[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 #include "progress.h"
15
16 #define REACHABLE 0x0001
17 #define SEEN      0x0002
18
19 static int show_root;
20 static int show_tags;
21 static int show_unreachable;
22 static int include_reflogs = 1;
23 static int check_full = 1;
24 static int check_strict;
25 static int keep_cache_objects;
26 static unsigned char head_sha1[20];
27 static const char *head_points_at;
28 static int errors_found;
29 static int write_lost_and_found;
30 static int verbose;
31 static int show_progress = -1;
32 #define ERROR_OBJECT 01
33 #define ERROR_REACHABLE 02
34 #define ERROR_PACK 04
35
36 #ifdef NO_D_INO_IN_DIRENT
37 #define SORT_DIRENT 0
38 #define DIRENT_SORT_HINT(de) 0
39 #else
40 #define SORT_DIRENT 1
41 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
42 #endif
43
44 static void objreport(struct object *obj, const char *severity,
45                       const char *err, va_list params)
46 {
47         fprintf(stderr, "%s in %s %s: ",
48                 severity, typename(obj->type), sha1_to_hex(obj->sha1));
49         vfprintf(stderr, err, params);
50         fputs("\n", stderr);
51 }
52
53 __attribute__((format (printf, 2, 3)))
54 static int objerror(struct object *obj, const char *err, ...)
55 {
56         va_list params;
57         va_start(params, err);
58         errors_found |= ERROR_OBJECT;
59         objreport(obj, "error", err, params);
60         va_end(params);
61         return -1;
62 }
63
64 __attribute__((format (printf, 3, 4)))
65 static int fsck_error_func(struct object *obj, int type, const char *err, ...)
66 {
67         va_list params;
68         va_start(params, err);
69         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
70         va_end(params);
71         return (type == FSCK_WARN) ? 0 : 1;
72 }
73
74 static struct object_array pending;
75
76 static int mark_object(struct object *obj, int type, void *data)
77 {
78         struct object *parent = data;
79
80         /*
81          * The only case data is NULL or type is OBJ_ANY is when
82          * mark_object_reachable() calls us.  All the callers of
83          * that function has non-NULL obj hence ...
84          */
85         if (!obj) {
86                 /* ... these references to parent->fld are safe here */
87                 printf("broken link from %7s %s\n",
88                            typename(parent->type), sha1_to_hex(parent->sha1));
89                 printf("broken link from %7s %s\n",
90                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
91                 errors_found |= ERROR_REACHABLE;
92                 return 1;
93         }
94
95         if (type != OBJ_ANY && obj->type != type)
96                 /* ... and the reference to parent is safe here */
97                 objerror(parent, "wrong object type in link");
98
99         if (obj->flags & REACHABLE)
100                 return 0;
101         obj->flags |= REACHABLE;
102         if (!obj->parsed) {
103                 if (parent && !has_sha1_file(obj->sha1)) {
104                         printf("broken link from %7s %s\n",
105                                  typename(parent->type), sha1_to_hex(parent->sha1));
106                         printf("              to %7s %s\n",
107                                  typename(obj->type), sha1_to_hex(obj->sha1));
108                         errors_found |= ERROR_REACHABLE;
109                 }
110                 return 1;
111         }
112
113         add_object_array(obj, (void *) parent, &pending);
114         return 0;
115 }
116
117 static void mark_object_reachable(struct object *obj)
118 {
119         mark_object(obj, OBJ_ANY, NULL);
120 }
121
122 static int traverse_one_object(struct object *obj)
123 {
124         int result;
125         struct tree *tree = NULL;
126
127         if (obj->type == OBJ_TREE) {
128                 obj->parsed = 0;
129                 tree = (struct tree *)obj;
130                 if (parse_tree(tree) < 0)
131                         return 1; /* error already displayed */
132         }
133         result = fsck_walk(obj, mark_object, obj);
134         if (tree) {
135                 free(tree->buffer);
136                 tree->buffer = NULL;
137         }
138         return result;
139 }
140
141 static int traverse_reachable(void)
142 {
143         struct progress *progress = NULL;
144         unsigned int nr = 0;
145         int result = 0;
146         if (show_progress)
147                 progress = start_progress_delay("Checking connectivity", 0, 0, 2);
148         while (pending.nr) {
149                 struct object_array_entry *entry;
150                 struct object *obj;
151
152                 entry = pending.objects + --pending.nr;
153                 obj = entry->item;
154                 result |= traverse_one_object(obj);
155                 display_progress(progress, ++nr);
156         }
157         stop_progress(&progress);
158         return !!result;
159 }
160
161 static int mark_used(struct object *obj, int type, void *data)
162 {
163         if (!obj)
164                 return 1;
165         obj->used = 1;
166         return 0;
167 }
168
169 /*
170  * Check a single reachable object
171  */
172 static void check_reachable_object(struct object *obj)
173 {
174         /*
175          * We obviously want the object to be parsed,
176          * except if it was in a pack-file and we didn't
177          * do a full fsck
178          */
179         if (!obj->parsed) {
180                 if (has_sha1_pack(obj->sha1))
181                         return; /* it is in pack - forget about it */
182                 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
183                 errors_found |= ERROR_REACHABLE;
184                 return;
185         }
186 }
187
188 /*
189  * Check a single unreachable object
190  */
191 static void check_unreachable_object(struct object *obj)
192 {
193         /*
194          * Missing unreachable object? Ignore it. It's not like
195          * we miss it (since it can't be reached), nor do we want
196          * to complain about it being unreachable (since it does
197          * not exist).
198          */
199         if (!obj->parsed)
200                 return;
201
202         /*
203          * Unreachable object that exists? Show it if asked to,
204          * since this is something that is prunable.
205          */
206         if (show_unreachable) {
207                 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
208                 return;
209         }
210
211         /*
212          * "!used" means that nothing at all points to it, including
213          * other unreachable objects. In other words, it's the "tip"
214          * of some set of unreachable objects, usually a commit that
215          * got dropped.
216          *
217          * Such starting points are more interesting than some random
218          * set of unreachable objects, so we show them even if the user
219          * hasn't asked for _all_ unreachable objects. If you have
220          * deleted a branch by mistake, this is a prime candidate to
221          * start looking at, for example.
222          */
223         if (!obj->used) {
224                 printf("dangling %s %s\n", typename(obj->type),
225                        sha1_to_hex(obj->sha1));
226                 if (write_lost_and_found) {
227                         char *filename = git_path("lost-found/%s/%s",
228                                 obj->type == OBJ_COMMIT ? "commit" : "other",
229                                 sha1_to_hex(obj->sha1));
230                         FILE *f;
231
232                         if (safe_create_leading_directories(filename)) {
233                                 error("Could not create lost-found");
234                                 return;
235                         }
236                         if (!(f = fopen(filename, "w")))
237                                 die_errno("Could not open '%s'", filename);
238                         if (obj->type == OBJ_BLOB) {
239                                 enum object_type type;
240                                 unsigned long size;
241                                 char *buf = read_sha1_file(obj->sha1,
242                                                 &type, &size);
243                                 if (buf && fwrite(buf, 1, size, f) != size)
244                                         die_errno("Could not write '%s'", filename);
245                                 free(buf);
246                         } else
247                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
248                         if (fclose(f))
249                                 die_errno("Could not finish '%s'",
250                                           filename);
251                 }
252                 return;
253         }
254
255         /*
256          * Otherwise? It's there, it's unreachable, and some other unreachable
257          * object points to it. Ignore it - it's not interesting, and we showed
258          * all the interesting cases above.
259          */
260 }
261
262 static void check_object(struct object *obj)
263 {
264         if (verbose)
265                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
266
267         if (obj->flags & REACHABLE)
268                 check_reachable_object(obj);
269         else
270                 check_unreachable_object(obj);
271 }
272
273 static void check_connectivity(void)
274 {
275         int i, max;
276
277         /* Traverse the pending reachable objects */
278         traverse_reachable();
279
280         /* Look up all the requirements, warn about missing objects.. */
281         max = get_max_object_index();
282         if (verbose)
283                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
284
285         for (i = 0; i < max; i++) {
286                 struct object *obj = get_indexed_object(i);
287
288                 if (obj)
289                         check_object(obj);
290         }
291 }
292
293 static int fsck_obj(struct object *obj)
294 {
295         if (obj->flags & SEEN)
296                 return 0;
297         obj->flags |= SEEN;
298
299         if (verbose)
300                 fprintf(stderr, "Checking %s %s\n",
301                         typename(obj->type), sha1_to_hex(obj->sha1));
302
303         if (fsck_walk(obj, mark_used, NULL))
304                 objerror(obj, "broken links");
305         if (fsck_object(obj, check_strict, fsck_error_func))
306                 return -1;
307
308         if (obj->type == OBJ_TREE) {
309                 struct tree *item = (struct tree *) obj;
310
311                 free(item->buffer);
312                 item->buffer = NULL;
313         }
314
315         if (obj->type == OBJ_COMMIT) {
316                 struct commit *commit = (struct commit *) obj;
317
318                 free(commit->buffer);
319                 commit->buffer = NULL;
320
321                 if (!commit->parents && show_root)
322                         printf("root %s\n", sha1_to_hex(commit->object.sha1));
323         }
324
325         if (obj->type == OBJ_TAG) {
326                 struct tag *tag = (struct tag *) obj;
327
328                 if (show_tags && tag->tagged) {
329                         printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
330                         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
331                 }
332         }
333
334         return 0;
335 }
336
337 static int fsck_sha1(const unsigned char *sha1)
338 {
339         struct object *obj = parse_object(sha1);
340         if (!obj) {
341                 errors_found |= ERROR_OBJECT;
342                 return error("%s: object corrupt or missing",
343                              sha1_to_hex(sha1));
344         }
345         return fsck_obj(obj);
346 }
347
348 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
349                            unsigned long size, void *buffer, int *eaten)
350 {
351         struct object *obj;
352         obj = parse_object_buffer(sha1, type, size, buffer, eaten);
353         if (!obj) {
354                 errors_found |= ERROR_OBJECT;
355                 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
356         }
357         return fsck_obj(obj);
358 }
359
360 /*
361  * This is the sorting chunk size: make it reasonably
362  * big so that we can sort well..
363  */
364 #define MAX_SHA1_ENTRIES (1024)
365
366 struct sha1_entry {
367         unsigned long ino;
368         unsigned char sha1[20];
369 };
370
371 static struct {
372         unsigned long nr;
373         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
374 } sha1_list;
375
376 static int ino_compare(const void *_a, const void *_b)
377 {
378         const struct sha1_entry *a = _a, *b = _b;
379         unsigned long ino1 = a->ino, ino2 = b->ino;
380         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
381 }
382
383 static void fsck_sha1_list(void)
384 {
385         int i, nr = sha1_list.nr;
386
387         if (SORT_DIRENT)
388                 qsort(sha1_list.entry, nr,
389                       sizeof(struct sha1_entry *), ino_compare);
390         for (i = 0; i < nr; i++) {
391                 struct sha1_entry *entry = sha1_list.entry[i];
392                 unsigned char *sha1 = entry->sha1;
393
394                 sha1_list.entry[i] = NULL;
395                 fsck_sha1(sha1);
396                 free(entry);
397         }
398         sha1_list.nr = 0;
399 }
400
401 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
402 {
403         struct sha1_entry *entry = xmalloc(sizeof(*entry));
404         int nr;
405
406         entry->ino = ino;
407         hashcpy(entry->sha1, sha1);
408         nr = sha1_list.nr;
409         if (nr == MAX_SHA1_ENTRIES) {
410                 fsck_sha1_list();
411                 nr = 0;
412         }
413         sha1_list.entry[nr] = entry;
414         sha1_list.nr = ++nr;
415 }
416
417 static inline int is_loose_object_file(struct dirent *de,
418                                        char *name, unsigned char *sha1)
419 {
420         if (strlen(de->d_name) != 38)
421                 return 0;
422         memcpy(name + 2, de->d_name, 39);
423         return !get_sha1_hex(name, sha1);
424 }
425
426 static void fsck_dir(int i, char *path)
427 {
428         DIR *dir = opendir(path);
429         struct dirent *de;
430         char name[100];
431
432         if (!dir)
433                 return;
434
435         if (verbose)
436                 fprintf(stderr, "Checking directory %s\n", path);
437
438         sprintf(name, "%02x", i);
439         while ((de = readdir(dir)) != NULL) {
440                 unsigned char sha1[20];
441
442                 if (is_dot_or_dotdot(de->d_name))
443                         continue;
444                 if (is_loose_object_file(de, name, sha1)) {
445                         add_sha1_list(sha1, DIRENT_SORT_HINT(de));
446                         continue;
447                 }
448                 if (!prefixcmp(de->d_name, "tmp_obj_"))
449                         continue;
450                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
451         }
452         closedir(dir);
453 }
454
455 static int default_refs;
456
457 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
458                 const char *email, unsigned long timestamp, int tz,
459                 const char *message, void *cb_data)
460 {
461         struct object *obj;
462
463         if (verbose)
464                 fprintf(stderr, "Checking reflog %s->%s\n",
465                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
466
467         if (!is_null_sha1(osha1)) {
468                 obj = lookup_object(osha1);
469                 if (obj) {
470                         obj->used = 1;
471                         mark_object_reachable(obj);
472                 }
473         }
474         obj = lookup_object(nsha1);
475         if (obj) {
476                 obj->used = 1;
477                 mark_object_reachable(obj);
478         }
479         return 0;
480 }
481
482 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
483 {
484         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
485         return 0;
486 }
487
488 static int is_branch(const char *refname)
489 {
490         return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
491 }
492
493 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
494 {
495         struct object *obj;
496
497         obj = parse_object(sha1);
498         if (!obj) {
499                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
500                 /* We'll continue with the rest despite the error.. */
501                 return 0;
502         }
503         if (obj->type != OBJ_COMMIT && is_branch(refname))
504                 error("%s: not a commit", refname);
505         default_refs++;
506         obj->used = 1;
507         mark_object_reachable(obj);
508
509         return 0;
510 }
511
512 static void get_default_heads(void)
513 {
514         if (head_points_at && !is_null_sha1(head_sha1))
515                 fsck_handle_ref("HEAD", head_sha1, 0, NULL);
516         for_each_ref(fsck_handle_ref, NULL);
517         if (include_reflogs)
518                 for_each_reflog(fsck_handle_reflog, NULL);
519
520         /*
521          * Not having any default heads isn't really fatal, but
522          * it does mean that "--unreachable" no longer makes any
523          * sense (since in this case everything will obviously
524          * be unreachable by definition.
525          *
526          * Showing dangling objects is valid, though (as those
527          * dangling objects are likely lost heads).
528          *
529          * So we just print a warning about it, and clear the
530          * "show_unreachable" flag.
531          */
532         if (!default_refs) {
533                 fprintf(stderr, "notice: No default references\n");
534                 show_unreachable = 0;
535         }
536 }
537
538 static void fsck_object_dir(const char *path)
539 {
540         int i;
541         struct progress *progress = NULL;
542
543         if (verbose)
544                 fprintf(stderr, "Checking object directory\n");
545
546         if (show_progress)
547                 progress = start_progress("Checking object directories", 256);
548         for (i = 0; i < 256; i++) {
549                 static char dir[4096];
550                 sprintf(dir, "%s/%02x", path, i);
551                 fsck_dir(i, dir);
552                 display_progress(progress, i+1);
553         }
554         stop_progress(&progress);
555         fsck_sha1_list();
556 }
557
558 static int fsck_head_link(void)
559 {
560         int flag;
561         int null_is_error = 0;
562
563         if (verbose)
564                 fprintf(stderr, "Checking HEAD link\n");
565
566         head_points_at = resolve_ref_unsafe("HEAD", head_sha1, 0, &flag);
567         if (!head_points_at)
568                 return error("Invalid HEAD");
569         if (!strcmp(head_points_at, "HEAD"))
570                 /* detached HEAD */
571                 null_is_error = 1;
572         else if (prefixcmp(head_points_at, "refs/heads/"))
573                 return error("HEAD points to something strange (%s)",
574                              head_points_at);
575         if (is_null_sha1(head_sha1)) {
576                 if (null_is_error)
577                         return error("HEAD: detached HEAD points at nothing");
578                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
579                         head_points_at + 11);
580         }
581         return 0;
582 }
583
584 static int fsck_cache_tree(struct cache_tree *it)
585 {
586         int i;
587         int err = 0;
588
589         if (verbose)
590                 fprintf(stderr, "Checking cache tree\n");
591
592         if (0 <= it->entry_count) {
593                 struct object *obj = parse_object(it->sha1);
594                 if (!obj) {
595                         error("%s: invalid sha1 pointer in cache-tree",
596                               sha1_to_hex(it->sha1));
597                         return 1;
598                 }
599                 obj->used = 1;
600                 mark_object_reachable(obj);
601                 if (obj->type != OBJ_TREE)
602                         err |= objerror(obj, "non-tree in cache-tree");
603         }
604         for (i = 0; i < it->subtree_nr; i++)
605                 err |= fsck_cache_tree(it->down[i]->cache_tree);
606         return err;
607 }
608
609 static char const * const fsck_usage[] = {
610         "git fsck [options] [<object>...]",
611         NULL
612 };
613
614 static struct option fsck_opts[] = {
615         OPT__VERBOSE(&verbose, "be verbose"),
616         OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
617         OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
618         OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
619         OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
620         OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
621         OPT_BOOLEAN(0, "full", &check_full, "also consider packs and alternate objects"),
622         OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
623         OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
624                                 "write dangling objects in .git/lost-found"),
625         OPT_BOOL(0, "progress", &show_progress, "show progress"),
626         OPT_END(),
627 };
628
629 int cmd_fsck(int argc, const char **argv, const char *prefix)
630 {
631         int i, heads;
632         struct alternate_object_database *alt;
633
634         errors_found = 0;
635         read_replace_refs = 0;
636
637         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
638
639         if (show_progress == -1)
640                 show_progress = isatty(2);
641         if (verbose)
642                 show_progress = 0;
643
644         if (write_lost_and_found) {
645                 check_full = 1;
646                 include_reflogs = 0;
647         }
648
649         fsck_head_link();
650         fsck_object_dir(get_object_directory());
651
652         prepare_alt_odb();
653         for (alt = alt_odb_list; alt; alt = alt->next) {
654                 char namebuf[PATH_MAX];
655                 int namelen = alt->name - alt->base;
656                 memcpy(namebuf, alt->base, namelen);
657                 namebuf[namelen - 1] = 0;
658                 fsck_object_dir(namebuf);
659         }
660
661         if (check_full) {
662                 struct packed_git *p;
663                 uint32_t total = 0, count = 0;
664                 struct progress *progress = NULL;
665
666                 prepare_packed_git();
667
668                 if (show_progress) {
669                         for (p = packed_git; p; p = p->next) {
670                                 if (open_pack_index(p))
671                                         continue;
672                                 total += p->num_objects;
673                         }
674
675                         progress = start_progress("Checking objects", total);
676                 }
677                 for (p = packed_git; p; p = p->next) {
678                         /* verify gives error messages itself */
679                         if (verify_pack(p, fsck_obj_buffer,
680                                         progress, count))
681                                 errors_found |= ERROR_PACK;
682                         count += p->num_objects;
683                 }
684                 stop_progress(&progress);
685         }
686
687         heads = 0;
688         for (i = 0; i < argc; i++) {
689                 const char *arg = argv[i];
690                 unsigned char sha1[20];
691                 if (!get_sha1(arg, sha1)) {
692                         struct object *obj = lookup_object(sha1);
693
694                         /* Error is printed by lookup_object(). */
695                         if (!obj)
696                                 continue;
697
698                         obj->used = 1;
699                         mark_object_reachable(obj);
700                         heads++;
701                         continue;
702                 }
703                 error("invalid parameter: expected sha1, got '%s'", arg);
704         }
705
706         /*
707          * If we've not been given any explicit head information, do the
708          * default ones from .git/refs. We also consider the index file
709          * in this case (ie this implies --cache).
710          */
711         if (!heads) {
712                 get_default_heads();
713                 keep_cache_objects = 1;
714         }
715
716         if (keep_cache_objects) {
717                 read_cache();
718                 for (i = 0; i < active_nr; i++) {
719                         unsigned int mode;
720                         struct blob *blob;
721                         struct object *obj;
722
723                         mode = active_cache[i]->ce_mode;
724                         if (S_ISGITLINK(mode))
725                                 continue;
726                         blob = lookup_blob(active_cache[i]->sha1);
727                         if (!blob)
728                                 continue;
729                         obj = &blob->object;
730                         obj->used = 1;
731                         mark_object_reachable(obj);
732                 }
733                 if (active_cache_tree)
734                         fsck_cache_tree(active_cache_tree);
735         }
736
737         check_connectivity();
738         return errors_found;
739 }