Merge branch 'st/verify-tag'
[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 #define HAS_OBJ   0x0004
20
21 static int show_root;
22 static int show_tags;
23 static int show_unreachable;
24 static int include_reflogs = 1;
25 static int check_full = 1;
26 static int connectivity_only;
27 static int check_strict;
28 static int keep_cache_objects;
29 static struct fsck_options fsck_walk_options = FSCK_OPTIONS_DEFAULT;
30 static struct fsck_options fsck_obj_options = FSCK_OPTIONS_DEFAULT;
31 static struct object_id head_oid;
32 static const char *head_points_at;
33 static int errors_found;
34 static int write_lost_and_found;
35 static int verbose;
36 static int show_progress = -1;
37 static int show_dangling = 1;
38 #define ERROR_OBJECT 01
39 #define ERROR_REACHABLE 02
40 #define ERROR_PACK 04
41 #define ERROR_REFS 010
42
43 static int fsck_config(const char *var, const char *value, void *cb)
44 {
45         if (strcmp(var, "fsck.skiplist") == 0) {
46                 const char *path;
47                 struct strbuf sb = STRBUF_INIT;
48
49                 if (git_config_pathname(&path, var, value))
50                         return 1;
51                 strbuf_addf(&sb, "skiplist=%s", path);
52                 free((char *)path);
53                 fsck_set_msg_types(&fsck_obj_options, sb.buf);
54                 strbuf_release(&sb);
55                 return 0;
56         }
57
58         if (skip_prefix(var, "fsck.", &var)) {
59                 fsck_set_msg_type(&fsck_obj_options, var, value);
60                 return 0;
61         }
62
63         return git_default_config(var, value, cb);
64 }
65
66 static void objreport(struct object *obj, const char *msg_type,
67                         const char *err)
68 {
69         fprintf(stderr, "%s in %s %s: %s\n",
70                 msg_type, typename(obj->type), oid_to_hex(&obj->oid), err);
71 }
72
73 static int objerror(struct object *obj, const char *err)
74 {
75         errors_found |= ERROR_OBJECT;
76         objreport(obj, "error", err);
77         return -1;
78 }
79
80 static int fsck_error_func(struct object *obj, int type, const char *message)
81 {
82         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", message);
83         return (type == FSCK_WARN) ? 0 : 1;
84 }
85
86 static struct object_array pending;
87
88 static int mark_object(struct object *obj, int type, void *data, struct fsck_options *options)
89 {
90         struct object *parent = data;
91
92         /*
93          * The only case data is NULL or type is OBJ_ANY is when
94          * mark_object_reachable() calls us.  All the callers of
95          * that function has non-NULL obj hence ...
96          */
97         if (!obj) {
98                 /* ... these references to parent->fld are safe here */
99                 printf("broken link from %7s %s\n",
100                            typename(parent->type), oid_to_hex(&parent->oid));
101                 printf("broken link from %7s %s\n",
102                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
103                 errors_found |= ERROR_REACHABLE;
104                 return 1;
105         }
106
107         if (type != OBJ_ANY && obj->type != type)
108                 /* ... and the reference to parent is safe here */
109                 objerror(parent, "wrong object type in link");
110
111         if (obj->flags & REACHABLE)
112                 return 0;
113         obj->flags |= REACHABLE;
114         if (!(obj->flags & HAS_OBJ)) {
115                 if (parent && !has_object_file(&obj->oid)) {
116                         printf("broken link from %7s %s\n",
117                                  typename(parent->type), oid_to_hex(&parent->oid));
118                         printf("              to %7s %s\n",
119                                  typename(obj->type), oid_to_hex(&obj->oid));
120                         errors_found |= ERROR_REACHABLE;
121                 }
122                 return 1;
123         }
124
125         add_object_array(obj, NULL, &pending);
126         return 0;
127 }
128
129 static void mark_object_reachable(struct object *obj)
130 {
131         mark_object(obj, OBJ_ANY, NULL, NULL);
132 }
133
134 static int traverse_one_object(struct object *obj)
135 {
136         int result;
137         struct tree *tree = NULL;
138
139         if (obj->type == OBJ_TREE) {
140                 tree = (struct tree *)obj;
141                 if (parse_tree(tree) < 0)
142                         return 1; /* error already displayed */
143         }
144         result = fsck_walk(obj, obj, &fsck_walk_options);
145         if (tree)
146                 free_tree_buffer(tree);
147         return result;
148 }
149
150 static int traverse_reachable(void)
151 {
152         struct progress *progress = NULL;
153         unsigned int nr = 0;
154         int result = 0;
155         if (show_progress)
156                 progress = start_progress_delay(_("Checking connectivity"), 0, 0, 2);
157         while (pending.nr) {
158                 struct object_array_entry *entry;
159                 struct object *obj;
160
161                 entry = pending.objects + --pending.nr;
162                 obj = entry->item;
163                 result |= traverse_one_object(obj);
164                 display_progress(progress, ++nr);
165         }
166         stop_progress(&progress);
167         return !!result;
168 }
169
170 static int mark_used(struct object *obj, int type, void *data, struct fsck_options *options)
171 {
172         if (!obj)
173                 return 1;
174         obj->used = 1;
175         return 0;
176 }
177
178 /*
179  * Check a single reachable object
180  */
181 static void check_reachable_object(struct object *obj)
182 {
183         /*
184          * We obviously want the object to be parsed,
185          * except if it was in a pack-file and we didn't
186          * do a full fsck
187          */
188         if (!(obj->flags & HAS_OBJ)) {
189                 if (has_sha1_pack(obj->oid.hash))
190                         return; /* it is in pack - forget about it */
191                 if (connectivity_only && has_object_file(&obj->oid))
192                         return;
193                 printf("missing %s %s\n", typename(obj->type), oid_to_hex(&obj->oid));
194                 errors_found |= ERROR_REACHABLE;
195                 return;
196         }
197 }
198
199 /*
200  * Check a single unreachable object
201  */
202 static void check_unreachable_object(struct object *obj)
203 {
204         /*
205          * Missing unreachable object? Ignore it. It's not like
206          * we miss it (since it can't be reached), nor do we want
207          * to complain about it being unreachable (since it does
208          * not exist).
209          */
210         if (!obj->parsed)
211                 return;
212
213         /*
214          * Unreachable object that exists? Show it if asked to,
215          * since this is something that is prunable.
216          */
217         if (show_unreachable) {
218                 printf("unreachable %s %s\n", typename(obj->type), oid_to_hex(&obj->oid));
219                 return;
220         }
221
222         /*
223          * "!used" means that nothing at all points to it, including
224          * other unreachable objects. In other words, it's the "tip"
225          * of some set of unreachable objects, usually a commit that
226          * got dropped.
227          *
228          * Such starting points are more interesting than some random
229          * set of unreachable objects, so we show them even if the user
230          * hasn't asked for _all_ unreachable objects. If you have
231          * deleted a branch by mistake, this is a prime candidate to
232          * start looking at, for example.
233          */
234         if (!obj->used) {
235                 if (show_dangling)
236                         printf("dangling %s %s\n", typename(obj->type),
237                                oid_to_hex(&obj->oid));
238                 if (write_lost_and_found) {
239                         char *filename = git_pathdup("lost-found/%s/%s",
240                                 obj->type == OBJ_COMMIT ? "commit" : "other",
241                                 oid_to_hex(&obj->oid));
242                         FILE *f;
243
244                         if (safe_create_leading_directories_const(filename)) {
245                                 error("Could not create lost-found");
246                                 free(filename);
247                                 return;
248                         }
249                         if (!(f = fopen(filename, "w")))
250                                 die_errno("Could not open '%s'", filename);
251                         if (obj->type == OBJ_BLOB) {
252                                 if (stream_blob_to_fd(fileno(f), obj->oid.hash, NULL, 1))
253                                         die_errno("Could not write '%s'", filename);
254                         } else
255                                 fprintf(f, "%s\n", oid_to_hex(&obj->oid));
256                         if (fclose(f))
257                                 die_errno("Could not finish '%s'",
258                                           filename);
259                         free(filename);
260                 }
261                 return;
262         }
263
264         /*
265          * Otherwise? It's there, it's unreachable, and some other unreachable
266          * object points to it. Ignore it - it's not interesting, and we showed
267          * all the interesting cases above.
268          */
269 }
270
271 static void check_object(struct object *obj)
272 {
273         if (verbose)
274                 fprintf(stderr, "Checking %s\n", oid_to_hex(&obj->oid));
275
276         if (obj->flags & REACHABLE)
277                 check_reachable_object(obj);
278         else
279                 check_unreachable_object(obj);
280 }
281
282 static void check_connectivity(void)
283 {
284         int i, max;
285
286         /* Traverse the pending reachable objects */
287         traverse_reachable();
288
289         /* Look up all the requirements, warn about missing objects.. */
290         max = get_max_object_index();
291         if (verbose)
292                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
293
294         for (i = 0; i < max; i++) {
295                 struct object *obj = get_indexed_object(i);
296
297                 if (obj)
298                         check_object(obj);
299         }
300 }
301
302 static int fsck_obj(struct object *obj)
303 {
304         if (obj->flags & SEEN)
305                 return 0;
306         obj->flags |= SEEN;
307
308         if (verbose)
309                 fprintf(stderr, "Checking %s %s\n",
310                         typename(obj->type), oid_to_hex(&obj->oid));
311
312         if (fsck_walk(obj, NULL, &fsck_obj_options))
313                 objerror(obj, "broken links");
314         if (fsck_object(obj, NULL, 0, &fsck_obj_options))
315                 return -1;
316
317         if (obj->type == OBJ_TREE) {
318                 struct tree *item = (struct tree *) obj;
319
320                 free_tree_buffer(item);
321         }
322
323         if (obj->type == OBJ_COMMIT) {
324                 struct commit *commit = (struct commit *) obj;
325
326                 free_commit_buffer(commit);
327
328                 if (!commit->parents && show_root)
329                         printf("root %s\n", oid_to_hex(&commit->object.oid));
330         }
331
332         if (obj->type == OBJ_TAG) {
333                 struct tag *tag = (struct tag *) obj;
334
335                 if (show_tags && tag->tagged) {
336                         printf("tagged %s %s", typename(tag->tagged->type), oid_to_hex(&tag->tagged->oid));
337                         printf(" (%s) in %s\n", tag->tag, oid_to_hex(&tag->object.oid));
338                 }
339         }
340
341         return 0;
342 }
343
344 static int fsck_sha1(const unsigned char *sha1)
345 {
346         struct object *obj = parse_object(sha1);
347         if (!obj) {
348                 errors_found |= ERROR_OBJECT;
349                 return error("%s: object corrupt or missing",
350                              sha1_to_hex(sha1));
351         }
352         obj->flags |= HAS_OBJ;
353         return fsck_obj(obj);
354 }
355
356 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
357                            unsigned long size, void *buffer, int *eaten)
358 {
359         struct object *obj;
360         obj = parse_object_buffer(sha1, type, size, buffer, eaten);
361         if (!obj) {
362                 errors_found |= ERROR_OBJECT;
363                 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
364         }
365         obj->flags = HAS_OBJ;
366         return fsck_obj(obj);
367 }
368
369 static int default_refs;
370
371 static void fsck_handle_reflog_sha1(const char *refname, unsigned char *sha1)
372 {
373         struct object *obj;
374
375         if (!is_null_sha1(sha1)) {
376                 obj = lookup_object(sha1);
377                 if (obj) {
378                         obj->used = 1;
379                         mark_object_reachable(obj);
380                 } else {
381                         error("%s: invalid reflog entry %s", refname, sha1_to_hex(sha1));
382                         errors_found |= ERROR_REACHABLE;
383                 }
384         }
385 }
386
387 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
388                 const char *email, unsigned long timestamp, int tz,
389                 const char *message, void *cb_data)
390 {
391         const char *refname = cb_data;
392
393         if (verbose)
394                 fprintf(stderr, "Checking reflog %s->%s\n",
395                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
396
397         fsck_handle_reflog_sha1(refname, osha1);
398         fsck_handle_reflog_sha1(refname, nsha1);
399         return 0;
400 }
401
402 static int fsck_handle_reflog(const char *logname, const struct object_id *oid,
403                               int flag, void *cb_data)
404 {
405         for_each_reflog_ent(logname, fsck_handle_reflog_ent, (void *)logname);
406         return 0;
407 }
408
409 static int fsck_handle_ref(const char *refname, const struct object_id *oid,
410                            int flag, void *cb_data)
411 {
412         struct object *obj;
413
414         obj = parse_object(oid->hash);
415         if (!obj) {
416                 error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
417                 errors_found |= ERROR_REACHABLE;
418                 /* We'll continue with the rest despite the error.. */
419                 return 0;
420         }
421         if (obj->type != OBJ_COMMIT && is_branch(refname)) {
422                 error("%s: not a commit", refname);
423                 errors_found |= ERROR_REFS;
424         }
425         default_refs++;
426         obj->used = 1;
427         mark_object_reachable(obj);
428
429         return 0;
430 }
431
432 static void get_default_heads(void)
433 {
434         if (head_points_at && !is_null_oid(&head_oid))
435                 fsck_handle_ref("HEAD", &head_oid, 0, NULL);
436         for_each_rawref(fsck_handle_ref, NULL);
437         if (include_reflogs)
438                 for_each_reflog(fsck_handle_reflog, NULL);
439
440         /*
441          * Not having any default heads isn't really fatal, but
442          * it does mean that "--unreachable" no longer makes any
443          * sense (since in this case everything will obviously
444          * be unreachable by definition.
445          *
446          * Showing dangling objects is valid, though (as those
447          * dangling objects are likely lost heads).
448          *
449          * So we just print a warning about it, and clear the
450          * "show_unreachable" flag.
451          */
452         if (!default_refs) {
453                 fprintf(stderr, "notice: No default references\n");
454                 show_unreachable = 0;
455         }
456 }
457
458 static int fsck_loose(const unsigned char *sha1, const char *path, void *data)
459 {
460         if (fsck_sha1(sha1))
461                 errors_found |= ERROR_OBJECT;
462         return 0;
463 }
464
465 static int fsck_cruft(const char *basename, const char *path, void *data)
466 {
467         if (!starts_with(basename, "tmp_obj_"))
468                 fprintf(stderr, "bad sha1 file: %s\n", path);
469         return 0;
470 }
471
472 static int fsck_subdir(int nr, const char *path, void *progress)
473 {
474         display_progress(progress, nr + 1);
475         return 0;
476 }
477
478 static void fsck_object_dir(const char *path)
479 {
480         struct progress *progress = NULL;
481
482         if (verbose)
483                 fprintf(stderr, "Checking object directory\n");
484
485         if (show_progress)
486                 progress = start_progress(_("Checking object directories"), 256);
487
488         for_each_loose_file_in_objdir(path, fsck_loose, fsck_cruft, fsck_subdir,
489                                       progress);
490         display_progress(progress, 256);
491         stop_progress(&progress);
492 }
493
494 static int fsck_head_link(void)
495 {
496         int null_is_error = 0;
497
498         if (verbose)
499                 fprintf(stderr, "Checking HEAD link\n");
500
501         head_points_at = resolve_ref_unsafe("HEAD", 0, head_oid.hash, NULL);
502         if (!head_points_at) {
503                 errors_found |= ERROR_REFS;
504                 return error("Invalid HEAD");
505         }
506         if (!strcmp(head_points_at, "HEAD"))
507                 /* detached HEAD */
508                 null_is_error = 1;
509         else if (!starts_with(head_points_at, "refs/heads/")) {
510                 errors_found |= ERROR_REFS;
511                 return error("HEAD points to something strange (%s)",
512                              head_points_at);
513         }
514         if (is_null_oid(&head_oid)) {
515                 if (null_is_error) {
516                         errors_found |= ERROR_REFS;
517                         return error("HEAD: detached HEAD points at nothing");
518                 }
519                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
520                         head_points_at + 11);
521         }
522         return 0;
523 }
524
525 static int fsck_cache_tree(struct cache_tree *it)
526 {
527         int i;
528         int err = 0;
529
530         if (verbose)
531                 fprintf(stderr, "Checking cache tree\n");
532
533         if (0 <= it->entry_count) {
534                 struct object *obj = parse_object(it->sha1);
535                 if (!obj) {
536                         error("%s: invalid sha1 pointer in cache-tree",
537                               sha1_to_hex(it->sha1));
538                         errors_found |= ERROR_REFS;
539                         return 1;
540                 }
541                 obj->used = 1;
542                 mark_object_reachable(obj);
543                 if (obj->type != OBJ_TREE)
544                         err |= objerror(obj, "non-tree in cache-tree");
545         }
546         for (i = 0; i < it->subtree_nr; i++)
547                 err |= fsck_cache_tree(it->down[i]->cache_tree);
548         return err;
549 }
550
551 static char const * const fsck_usage[] = {
552         N_("git fsck [<options>] [<object>...]"),
553         NULL
554 };
555
556 static struct option fsck_opts[] = {
557         OPT__VERBOSE(&verbose, N_("be verbose")),
558         OPT_BOOL(0, "unreachable", &show_unreachable, N_("show unreachable objects")),
559         OPT_BOOL(0, "dangling", &show_dangling, N_("show dangling objects")),
560         OPT_BOOL(0, "tags", &show_tags, N_("report tags")),
561         OPT_BOOL(0, "root", &show_root, N_("report root nodes")),
562         OPT_BOOL(0, "cache", &keep_cache_objects, N_("make index objects head nodes")),
563         OPT_BOOL(0, "reflogs", &include_reflogs, N_("make reflogs head nodes (default)")),
564         OPT_BOOL(0, "full", &check_full, N_("also consider packs and alternate objects")),
565         OPT_BOOL(0, "connectivity-only", &connectivity_only, N_("check only connectivity")),
566         OPT_BOOL(0, "strict", &check_strict, N_("enable more strict checking")),
567         OPT_BOOL(0, "lost-found", &write_lost_and_found,
568                                 N_("write dangling objects in .git/lost-found")),
569         OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
570         OPT_END(),
571 };
572
573 int cmd_fsck(int argc, const char **argv, const char *prefix)
574 {
575         int i, heads;
576         struct alternate_object_database *alt;
577
578         errors_found = 0;
579         check_replace_refs = 0;
580
581         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
582
583         fsck_walk_options.walk = mark_object;
584         fsck_obj_options.walk = mark_used;
585         fsck_obj_options.error_func = fsck_error_func;
586         if (check_strict)
587                 fsck_obj_options.strict = 1;
588
589         if (show_progress == -1)
590                 show_progress = isatty(2);
591         if (verbose)
592                 show_progress = 0;
593
594         if (write_lost_and_found) {
595                 check_full = 1;
596                 include_reflogs = 0;
597         }
598
599         git_config(fsck_config, NULL);
600
601         fsck_head_link();
602         if (!connectivity_only) {
603                 fsck_object_dir(get_object_directory());
604
605                 prepare_alt_odb();
606                 for (alt = alt_odb_list; alt; alt = alt->next) {
607                         /* directory name, minus trailing slash */
608                         size_t namelen = alt->name - alt->base - 1;
609                         struct strbuf name = STRBUF_INIT;
610                         strbuf_add(&name, alt->base, namelen);
611                         fsck_object_dir(name.buf);
612                         strbuf_release(&name);
613                 }
614         }
615
616         if (check_full) {
617                 struct packed_git *p;
618                 uint32_t total = 0, count = 0;
619                 struct progress *progress = NULL;
620
621                 prepare_packed_git();
622
623                 if (show_progress) {
624                         for (p = packed_git; p; p = p->next) {
625                                 if (open_pack_index(p))
626                                         continue;
627                                 total += p->num_objects;
628                         }
629
630                         progress = start_progress(_("Checking objects"), total);
631                 }
632                 for (p = packed_git; p; p = p->next) {
633                         /* verify gives error messages itself */
634                         if (verify_pack(p, fsck_obj_buffer,
635                                         progress, count))
636                                 errors_found |= ERROR_PACK;
637                         count += p->num_objects;
638                 }
639                 stop_progress(&progress);
640         }
641
642         heads = 0;
643         for (i = 0; i < argc; i++) {
644                 const char *arg = argv[i];
645                 unsigned char sha1[20];
646                 if (!get_sha1(arg, sha1)) {
647                         struct object *obj = lookup_object(sha1);
648
649                         /* Error is printed by lookup_object(). */
650                         if (!obj)
651                                 continue;
652
653                         obj->used = 1;
654                         mark_object_reachable(obj);
655                         heads++;
656                         continue;
657                 }
658                 error("invalid parameter: expected sha1, got '%s'", arg);
659         }
660
661         /*
662          * If we've not been given any explicit head information, do the
663          * default ones from .git/refs. We also consider the index file
664          * in this case (ie this implies --cache).
665          */
666         if (!heads) {
667                 get_default_heads();
668                 keep_cache_objects = 1;
669         }
670
671         if (keep_cache_objects) {
672                 read_cache();
673                 for (i = 0; i < active_nr; i++) {
674                         unsigned int mode;
675                         struct blob *blob;
676                         struct object *obj;
677
678                         mode = active_cache[i]->ce_mode;
679                         if (S_ISGITLINK(mode))
680                                 continue;
681                         blob = lookup_blob(active_cache[i]->sha1);
682                         if (!blob)
683                                 continue;
684                         obj = &blob->object;
685                         obj->used = 1;
686                         mark_object_reachable(obj);
687                 }
688                 if (active_cache_tree)
689                         fsck_cache_tree(active_cache_tree);
690         }
691
692         check_connectivity();
693         return errors_found;
694 }