diff --color-words: bit of clean-up
[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;
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
33 #ifdef NO_D_INO_IN_DIRENT
34 #define SORT_DIRENT 0
35 #define DIRENT_SORT_HINT(de) 0
36 #else
37 #define SORT_DIRENT 1
38 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
39 #endif
40
41 static void objreport(struct object *obj, const char *severity,
42                       const char *err, va_list params)
43 {
44         fprintf(stderr, "%s in %s %s: ",
45                 severity, typename(obj->type), sha1_to_hex(obj->sha1));
46         vfprintf(stderr, err, params);
47         fputs("\n", stderr);
48 }
49
50 static int objerror(struct object *obj, const char *err, ...)
51 {
52         va_list params;
53         va_start(params, err);
54         errors_found |= ERROR_OBJECT;
55         objreport(obj, "error", err, params);
56         va_end(params);
57         return -1;
58 }
59
60 static int fsck_error_func(struct object *obj, int type, const char *err, ...)
61 {
62         va_list params;
63         va_start(params, err);
64         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
65         va_end(params);
66         return (type == FSCK_WARN) ? 0 : 1;
67 }
68
69 static struct object_array pending;
70
71 static int mark_object(struct object *obj, int type, void *data)
72 {
73         struct object *parent = data;
74
75         if (!obj) {
76                 printf("broken link from %7s %s\n",
77                            typename(parent->type), sha1_to_hex(parent->sha1));
78                 printf("broken link from %7s %s\n",
79                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
80                 errors_found |= ERROR_REACHABLE;
81                 return 1;
82         }
83
84         if (type != OBJ_ANY && obj->type != type)
85                 objerror(parent, "wrong object type in link");
86
87         if (obj->flags & REACHABLE)
88                 return 0;
89         obj->flags |= REACHABLE;
90         if (!obj->parsed) {
91                 if (parent && !has_sha1_file(obj->sha1)) {
92                         printf("broken link from %7s %s\n",
93                                  typename(parent->type), sha1_to_hex(parent->sha1));
94                         printf("              to %7s %s\n",
95                                  typename(obj->type), sha1_to_hex(obj->sha1));
96                         errors_found |= ERROR_REACHABLE;
97                 }
98                 return 1;
99         }
100
101         add_object_array(obj, (void *) parent, &pending);
102         return 0;
103 }
104
105 static void mark_object_reachable(struct object *obj)
106 {
107         mark_object(obj, OBJ_ANY, NULL);
108 }
109
110 static int traverse_one_object(struct object *obj, struct object *parent)
111 {
112         int result;
113         struct tree *tree = NULL;
114
115         if (obj->type == OBJ_TREE) {
116                 obj->parsed = 0;
117                 tree = (struct tree *)obj;
118                 if (parse_tree(tree) < 0)
119                         return 1; /* error already displayed */
120         }
121         result = fsck_walk(obj, mark_object, obj);
122         if (tree) {
123                 free(tree->buffer);
124                 tree->buffer = NULL;
125         }
126         return result;
127 }
128
129 static int traverse_reachable(void)
130 {
131         int result = 0;
132         while (pending.nr) {
133                 struct object_array_entry *entry;
134                 struct object *obj, *parent;
135
136                 entry = pending.objects + --pending.nr;
137                 obj = entry->item;
138                 parent = (struct object *) entry->name;
139                 result |= traverse_one_object(obj, parent);
140         }
141         return !!result;
142 }
143
144 static int mark_used(struct object *obj, int type, void *data)
145 {
146         if (!obj)
147                 return 1;
148         obj->used = 1;
149         return 0;
150 }
151
152 /*
153  * Check a single reachable object
154  */
155 static void check_reachable_object(struct object *obj)
156 {
157         /*
158          * We obviously want the object to be parsed,
159          * except if it was in a pack-file and we didn't
160          * do a full fsck
161          */
162         if (!obj->parsed) {
163                 if (has_sha1_pack(obj->sha1))
164                         return; /* it is in pack - forget about it */
165                 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
166                 errors_found |= ERROR_REACHABLE;
167                 return;
168         }
169 }
170
171 /*
172  * Check a single unreachable object
173  */
174 static void check_unreachable_object(struct object *obj)
175 {
176         /*
177          * Missing unreachable object? Ignore it. It's not like
178          * we miss it (since it can't be reached), nor do we want
179          * to complain about it being unreachable (since it does
180          * not exist).
181          */
182         if (!obj->parsed)
183                 return;
184
185         /*
186          * Unreachable object that exists? Show it if asked to,
187          * since this is something that is prunable.
188          */
189         if (show_unreachable) {
190                 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
191                 return;
192         }
193
194         /*
195          * "!used" means that nothing at all points to it, including
196          * other unreachable objects. In other words, it's the "tip"
197          * of some set of unreachable objects, usually a commit that
198          * got dropped.
199          *
200          * Such starting points are more interesting than some random
201          * set of unreachable objects, so we show them even if the user
202          * hasn't asked for _all_ unreachable objects. If you have
203          * deleted a branch by mistake, this is a prime candidate to
204          * start looking at, for example.
205          */
206         if (!obj->used) {
207                 printf("dangling %s %s\n", typename(obj->type),
208                        sha1_to_hex(obj->sha1));
209                 if (write_lost_and_found) {
210                         char *filename = git_path("lost-found/%s/%s",
211                                 obj->type == OBJ_COMMIT ? "commit" : "other",
212                                 sha1_to_hex(obj->sha1));
213                         FILE *f;
214
215                         if (safe_create_leading_directories(filename)) {
216                                 error("Could not create lost-found");
217                                 return;
218                         }
219                         if (!(f = fopen(filename, "w")))
220                                 die_errno("Could not open '%s'", filename);
221                         if (obj->type == OBJ_BLOB) {
222                                 enum object_type type;
223                                 unsigned long size;
224                                 char *buf = read_sha1_file(obj->sha1,
225                                                 &type, &size);
226                                 if (buf) {
227                                         if (fwrite(buf, size, 1, f) != 1)
228                                                 die_errno("Could not write '%s'",
229                                                           filename);
230                                         free(buf);
231                                 }
232                         } else
233                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
234                         if (fclose(f))
235                                 die_errno("Could not finish '%s'",
236                                           filename);
237                 }
238                 return;
239         }
240
241         /*
242          * Otherwise? It's there, it's unreachable, and some other unreachable
243          * object points to it. Ignore it - it's not interesting, and we showed
244          * all the interesting cases above.
245          */
246 }
247
248 static void check_object(struct object *obj)
249 {
250         if (verbose)
251                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
252
253         if (obj->flags & REACHABLE)
254                 check_reachable_object(obj);
255         else
256                 check_unreachable_object(obj);
257 }
258
259 static void check_connectivity(void)
260 {
261         int i, max;
262
263         /* Traverse the pending reachable objects */
264         traverse_reachable();
265
266         /* Look up all the requirements, warn about missing objects.. */
267         max = get_max_object_index();
268         if (verbose)
269                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
270
271         for (i = 0; i < max; i++) {
272                 struct object *obj = get_indexed_object(i);
273
274                 if (obj)
275                         check_object(obj);
276         }
277 }
278
279 static int fsck_sha1(const unsigned char *sha1)
280 {
281         struct object *obj = parse_object(sha1);
282         if (!obj) {
283                 errors_found |= ERROR_OBJECT;
284                 return error("%s: object corrupt or missing",
285                              sha1_to_hex(sha1));
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 /*
330  * This is the sorting chunk size: make it reasonably
331  * big so that we can sort well..
332  */
333 #define MAX_SHA1_ENTRIES (1024)
334
335 struct sha1_entry {
336         unsigned long ino;
337         unsigned char sha1[20];
338 };
339
340 static struct {
341         unsigned long nr;
342         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
343 } sha1_list;
344
345 static int ino_compare(const void *_a, const void *_b)
346 {
347         const struct sha1_entry *a = _a, *b = _b;
348         unsigned long ino1 = a->ino, ino2 = b->ino;
349         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
350 }
351
352 static void fsck_sha1_list(void)
353 {
354         int i, nr = sha1_list.nr;
355
356         if (SORT_DIRENT)
357                 qsort(sha1_list.entry, nr,
358                       sizeof(struct sha1_entry *), ino_compare);
359         for (i = 0; i < nr; i++) {
360                 struct sha1_entry *entry = sha1_list.entry[i];
361                 unsigned char *sha1 = entry->sha1;
362
363                 sha1_list.entry[i] = NULL;
364                 fsck_sha1(sha1);
365                 free(entry);
366         }
367         sha1_list.nr = 0;
368 }
369
370 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
371 {
372         struct sha1_entry *entry = xmalloc(sizeof(*entry));
373         int nr;
374
375         entry->ino = ino;
376         hashcpy(entry->sha1, sha1);
377         nr = sha1_list.nr;
378         if (nr == MAX_SHA1_ENTRIES) {
379                 fsck_sha1_list();
380                 nr = 0;
381         }
382         sha1_list.entry[nr] = entry;
383         sha1_list.nr = ++nr;
384 }
385
386 static void fsck_dir(int i, char *path)
387 {
388         DIR *dir = opendir(path);
389         struct dirent *de;
390
391         if (!dir)
392                 return;
393
394         if (verbose)
395                 fprintf(stderr, "Checking directory %s\n", path);
396
397         while ((de = readdir(dir)) != NULL) {
398                 char name[100];
399                 unsigned char sha1[20];
400
401                 if (is_dot_or_dotdot(de->d_name))
402                         continue;
403                 if (strlen(de->d_name) == 38) {
404                         sprintf(name, "%02x", i);
405                         memcpy(name+2, de->d_name, 39);
406                         if (get_sha1_hex(name, sha1) < 0)
407                                 break;
408                         add_sha1_list(sha1, DIRENT_SORT_HINT(de));
409                         continue;
410                 }
411                 if (!prefixcmp(de->d_name, "tmp_obj_"))
412                         continue;
413                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
414         }
415         closedir(dir);
416 }
417
418 static int default_refs;
419
420 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
421                 const char *email, unsigned long timestamp, int tz,
422                 const char *message, void *cb_data)
423 {
424         struct object *obj;
425
426         if (verbose)
427                 fprintf(stderr, "Checking reflog %s->%s\n",
428                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
429
430         if (!is_null_sha1(osha1)) {
431                 obj = lookup_object(osha1);
432                 if (obj) {
433                         obj->used = 1;
434                         mark_object_reachable(obj);
435                 }
436         }
437         obj = lookup_object(nsha1);
438         if (obj) {
439                 obj->used = 1;
440                 mark_object_reachable(obj);
441         }
442         return 0;
443 }
444
445 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
446 {
447         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
448         return 0;
449 }
450
451 static int is_branch(const char *refname)
452 {
453         return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
454 }
455
456 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
457 {
458         struct object *obj;
459
460         obj = parse_object(sha1);
461         if (!obj) {
462                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
463                 /* We'll continue with the rest despite the error.. */
464                 return 0;
465         }
466         if (obj->type != OBJ_COMMIT && is_branch(refname))
467                 error("%s: not a commit", refname);
468         default_refs++;
469         obj->used = 1;
470         mark_object_reachable(obj);
471
472         return 0;
473 }
474
475 static void get_default_heads(void)
476 {
477         if (head_points_at && !is_null_sha1(head_sha1))
478                 fsck_handle_ref("HEAD", head_sha1, 0, NULL);
479         for_each_ref(fsck_handle_ref, NULL);
480         if (include_reflogs)
481                 for_each_reflog(fsck_handle_reflog, NULL);
482
483         /*
484          * Not having any default heads isn't really fatal, but
485          * it does mean that "--unreachable" no longer makes any
486          * sense (since in this case everything will obviously
487          * be unreachable by definition.
488          *
489          * Showing dangling objects is valid, though (as those
490          * dangling objects are likely lost heads).
491          *
492          * So we just print a warning about it, and clear the
493          * "show_unreachable" flag.
494          */
495         if (!default_refs) {
496                 fprintf(stderr, "notice: No default references\n");
497                 show_unreachable = 0;
498         }
499 }
500
501 static void fsck_object_dir(const char *path)
502 {
503         int i;
504
505         if (verbose)
506                 fprintf(stderr, "Checking object directory\n");
507
508         for (i = 0; i < 256; i++) {
509                 static char dir[4096];
510                 sprintf(dir, "%s/%02x", path, i);
511                 fsck_dir(i, dir);
512         }
513         fsck_sha1_list();
514 }
515
516 static int fsck_head_link(void)
517 {
518         int flag;
519         int null_is_error = 0;
520
521         if (verbose)
522                 fprintf(stderr, "Checking HEAD link\n");
523
524         head_points_at = resolve_ref("HEAD", head_sha1, 0, &flag);
525         if (!head_points_at)
526                 return error("Invalid HEAD");
527         if (!strcmp(head_points_at, "HEAD"))
528                 /* detached HEAD */
529                 null_is_error = 1;
530         else if (prefixcmp(head_points_at, "refs/heads/"))
531                 return error("HEAD points to something strange (%s)",
532                              head_points_at);
533         if (is_null_sha1(head_sha1)) {
534                 if (null_is_error)
535                         return error("HEAD: detached HEAD points at nothing");
536                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
537                         head_points_at + 11);
538         }
539         return 0;
540 }
541
542 static int fsck_cache_tree(struct cache_tree *it)
543 {
544         int i;
545         int err = 0;
546
547         if (verbose)
548                 fprintf(stderr, "Checking cache tree\n");
549
550         if (0 <= it->entry_count) {
551                 struct object *obj = parse_object(it->sha1);
552                 if (!obj) {
553                         error("%s: invalid sha1 pointer in cache-tree",
554                               sha1_to_hex(it->sha1));
555                         return 1;
556                 }
557                 mark_object_reachable(obj);
558                 obj->used = 1;
559                 if (obj->type != OBJ_TREE)
560                         err |= objerror(obj, "non-tree in cache-tree");
561         }
562         for (i = 0; i < it->subtree_nr; i++)
563                 err |= fsck_cache_tree(it->down[i]->cache_tree);
564         return err;
565 }
566
567 static char const * const fsck_usage[] = {
568         "git fsck [options] [<object>...]",
569         NULL
570 };
571
572 static struct option fsck_opts[] = {
573         OPT__VERBOSE(&verbose),
574         OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
575         OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
576         OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
577         OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
578         OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
579         OPT_BOOLEAN(0, "full", &check_full, "also consider alternate objects"),
580         OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
581         OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
582                                 "write dangling objects in .git/lost-found"),
583         OPT_END(),
584 };
585
586 int cmd_fsck(int argc, const char **argv, const char *prefix)
587 {
588         int i, heads;
589         struct alternate_object_database *alt;
590
591         errors_found = 0;
592         read_replace_refs = 0;
593
594         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
595         if (write_lost_and_found) {
596                 check_full = 1;
597                 include_reflogs = 0;
598         }
599
600         fsck_head_link();
601         fsck_object_dir(get_object_directory());
602
603         prepare_alt_odb();
604         for (alt = alt_odb_list; alt; alt = alt->next) {
605                 char namebuf[PATH_MAX];
606                 int namelen = alt->name - alt->base;
607                 memcpy(namebuf, alt->base, namelen);
608                 namebuf[namelen - 1] = 0;
609                 fsck_object_dir(namebuf);
610         }
611
612         if (check_full) {
613                 struct packed_git *p;
614
615                 prepare_packed_git();
616                 for (p = packed_git; p; p = p->next)
617                         /* verify gives error messages itself */
618                         verify_pack(p);
619
620                 for (p = packed_git; p; p = p->next) {
621                         uint32_t j, num;
622                         if (open_pack_index(p))
623                                 continue;
624                         num = p->num_objects;
625                         for (j = 0; j < num; j++)
626                                 fsck_sha1(nth_packed_object_sha1(p, j));
627                 }
628         }
629
630         heads = 0;
631         for (i = 0; i < argc; i++) {
632                 const char *arg = argv[i];
633                 unsigned char sha1[20];
634                 if (!get_sha1(arg, sha1)) {
635                         struct object *obj = lookup_object(sha1);
636
637                         /* Error is printed by lookup_object(). */
638                         if (!obj)
639                                 continue;
640
641                         obj->used = 1;
642                         mark_object_reachable(obj);
643                         heads++;
644                         continue;
645                 }
646                 error("invalid parameter: expected sha1, got '%s'", arg);
647         }
648
649         /*
650          * If we've not been given any explicit head information, do the
651          * default ones from .git/refs. We also consider the index file
652          * in this case (ie this implies --cache).
653          */
654         if (!heads) {
655                 get_default_heads();
656                 keep_cache_objects = 1;
657         }
658
659         if (keep_cache_objects) {
660                 read_cache();
661                 for (i = 0; i < active_nr; i++) {
662                         unsigned int mode;
663                         struct blob *blob;
664                         struct object *obj;
665
666                         mode = active_cache[i]->ce_mode;
667                         if (S_ISGITLINK(mode))
668                                 continue;
669                         blob = lookup_blob(active_cache[i]->sha1);
670                         if (!blob)
671                                 continue;
672                         obj = &blob->object;
673                         obj->used = 1;
674                         mark_object_reachable(obj);
675                 }
676                 if (active_cache_tree)
677                         fsck_cache_tree(active_cache_tree);
678         }
679
680         check_connectivity();
681         return errors_found;
682 }