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