Merge git://ozlabs.org/~paulus/gitk
[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 #include "decorate.h"
17
18 #define REACHABLE 0x0001
19 #define SEEN      0x0002
20 #define HAS_OBJ   0x0004
21
22 static int show_root;
23 static int show_tags;
24 static int show_unreachable;
25 static int include_reflogs = 1;
26 static int check_full = 1;
27 static int connectivity_only;
28 static int check_strict;
29 static int keep_cache_objects;
30 static struct fsck_options fsck_walk_options = FSCK_OPTIONS_DEFAULT;
31 static struct fsck_options fsck_obj_options = FSCK_OPTIONS_DEFAULT;
32 static struct object_id head_oid;
33 static const char *head_points_at;
34 static int errors_found;
35 static int write_lost_and_found;
36 static int verbose;
37 static int show_progress = -1;
38 static int show_dangling = 1;
39 static int name_objects;
40 #define ERROR_OBJECT 01
41 #define ERROR_REACHABLE 02
42 #define ERROR_PACK 04
43 #define ERROR_REFS 010
44
45 static const char *describe_object(struct object *obj)
46 {
47         static struct strbuf buf = STRBUF_INIT;
48         char *name = name_objects ?
49                 lookup_decoration(fsck_walk_options.object_names, obj) : NULL;
50
51         strbuf_reset(&buf);
52         strbuf_addstr(&buf, oid_to_hex(&obj->oid));
53         if (name)
54                 strbuf_addf(&buf, " (%s)", name);
55
56         return buf.buf;
57 }
58
59 static int fsck_config(const char *var, const char *value, void *cb)
60 {
61         if (strcmp(var, "fsck.skiplist") == 0) {
62                 const char *path;
63                 struct strbuf sb = STRBUF_INIT;
64
65                 if (git_config_pathname(&path, var, value))
66                         return 1;
67                 strbuf_addf(&sb, "skiplist=%s", path);
68                 free((char *)path);
69                 fsck_set_msg_types(&fsck_obj_options, sb.buf);
70                 strbuf_release(&sb);
71                 return 0;
72         }
73
74         if (skip_prefix(var, "fsck.", &var)) {
75                 fsck_set_msg_type(&fsck_obj_options, var, value);
76                 return 0;
77         }
78
79         return git_default_config(var, value, cb);
80 }
81
82 static void objreport(struct object *obj, const char *msg_type,
83                         const char *err)
84 {
85         fprintf(stderr, "%s in %s %s: %s\n",
86                 msg_type, typename(obj->type), describe_object(obj), err);
87 }
88
89 static int objerror(struct object *obj, const char *err)
90 {
91         errors_found |= ERROR_OBJECT;
92         objreport(obj, "error", err);
93         return -1;
94 }
95
96 static int fsck_error_func(struct fsck_options *o,
97         struct object *obj, int type, const char *message)
98 {
99         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", message);
100         return (type == FSCK_WARN) ? 0 : 1;
101 }
102
103 static struct object_array pending;
104
105 static int mark_object(struct object *obj, int type, void *data, struct fsck_options *options)
106 {
107         struct object *parent = data;
108
109         /*
110          * The only case data is NULL or type is OBJ_ANY is when
111          * mark_object_reachable() calls us.  All the callers of
112          * that function has non-NULL obj hence ...
113          */
114         if (!obj) {
115                 /* ... these references to parent->fld are safe here */
116                 printf("broken link from %7s %s\n",
117                            typename(parent->type), describe_object(parent));
118                 printf("broken link from %7s %s\n",
119                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
120                 errors_found |= ERROR_REACHABLE;
121                 return 1;
122         }
123
124         if (type != OBJ_ANY && obj->type != type)
125                 /* ... and the reference to parent is safe here */
126                 objerror(parent, "wrong object type in link");
127
128         if (obj->flags & REACHABLE)
129                 return 0;
130         obj->flags |= REACHABLE;
131         if (!(obj->flags & HAS_OBJ)) {
132                 if (parent && !has_object_file(&obj->oid)) {
133                         printf("broken link from %7s %s\n",
134                                  typename(parent->type), describe_object(parent));
135                         printf("              to %7s %s\n",
136                                  typename(obj->type), describe_object(obj));
137                         errors_found |= ERROR_REACHABLE;
138                 }
139                 return 1;
140         }
141
142         add_object_array(obj, NULL, &pending);
143         return 0;
144 }
145
146 static void mark_object_reachable(struct object *obj)
147 {
148         mark_object(obj, OBJ_ANY, NULL, NULL);
149 }
150
151 static int traverse_one_object(struct object *obj)
152 {
153         int result;
154         struct tree *tree = NULL;
155
156         if (obj->type == OBJ_TREE) {
157                 tree = (struct tree *)obj;
158                 if (parse_tree(tree) < 0)
159                         return 1; /* error already displayed */
160         }
161         result = fsck_walk(obj, obj, &fsck_walk_options);
162         if (tree)
163                 free_tree_buffer(tree);
164         return result;
165 }
166
167 static int traverse_reachable(void)
168 {
169         struct progress *progress = NULL;
170         unsigned int nr = 0;
171         int result = 0;
172         if (show_progress)
173                 progress = start_progress_delay(_("Checking connectivity"), 0, 0, 2);
174         while (pending.nr) {
175                 struct object_array_entry *entry;
176                 struct object *obj;
177
178                 entry = pending.objects + --pending.nr;
179                 obj = entry->item;
180                 result |= traverse_one_object(obj);
181                 display_progress(progress, ++nr);
182         }
183         stop_progress(&progress);
184         return !!result;
185 }
186
187 static int mark_used(struct object *obj, int type, void *data, struct fsck_options *options)
188 {
189         if (!obj)
190                 return 1;
191         obj->used = 1;
192         return 0;
193 }
194
195 /*
196  * Check a single reachable object
197  */
198 static void check_reachable_object(struct object *obj)
199 {
200         /*
201          * We obviously want the object to be parsed,
202          * except if it was in a pack-file and we didn't
203          * do a full fsck
204          */
205         if (!(obj->flags & HAS_OBJ)) {
206                 if (has_sha1_pack(obj->oid.hash))
207                         return; /* it is in pack - forget about it */
208                 if (connectivity_only && has_object_file(&obj->oid))
209                         return;
210                 printf("missing %s %s\n", typename(obj->type),
211                         describe_object(obj));
212                 errors_found |= ERROR_REACHABLE;
213                 return;
214         }
215 }
216
217 /*
218  * Check a single unreachable object
219  */
220 static void check_unreachable_object(struct object *obj)
221 {
222         /*
223          * Missing unreachable object? Ignore it. It's not like
224          * we miss it (since it can't be reached), nor do we want
225          * to complain about it being unreachable (since it does
226          * not exist).
227          */
228         if (!obj->parsed)
229                 return;
230
231         /*
232          * Unreachable object that exists? Show it if asked to,
233          * since this is something that is prunable.
234          */
235         if (show_unreachable) {
236                 printf("unreachable %s %s\n", typename(obj->type),
237                         describe_object(obj));
238                 return;
239         }
240
241         /*
242          * "!used" means that nothing at all points to it, including
243          * other unreachable objects. In other words, it's the "tip"
244          * of some set of unreachable objects, usually a commit that
245          * got dropped.
246          *
247          * Such starting points are more interesting than some random
248          * set of unreachable objects, so we show them even if the user
249          * hasn't asked for _all_ unreachable objects. If you have
250          * deleted a branch by mistake, this is a prime candidate to
251          * start looking at, for example.
252          */
253         if (!obj->used) {
254                 if (show_dangling)
255                         printf("dangling %s %s\n", typename(obj->type),
256                                describe_object(obj));
257                 if (write_lost_and_found) {
258                         char *filename = git_pathdup("lost-found/%s/%s",
259                                 obj->type == OBJ_COMMIT ? "commit" : "other",
260                                 describe_object(obj));
261                         FILE *f;
262
263                         if (safe_create_leading_directories_const(filename)) {
264                                 error("Could not create lost-found");
265                                 free(filename);
266                                 return;
267                         }
268                         if (!(f = fopen(filename, "w")))
269                                 die_errno("Could not open '%s'", filename);
270                         if (obj->type == OBJ_BLOB) {
271                                 if (stream_blob_to_fd(fileno(f), &obj->oid, NULL, 1))
272                                         die_errno("Could not write '%s'", filename);
273                         } else
274                                 fprintf(f, "%s\n", describe_object(obj));
275                         if (fclose(f))
276                                 die_errno("Could not finish '%s'",
277                                           filename);
278                         free(filename);
279                 }
280                 return;
281         }
282
283         /*
284          * Otherwise? It's there, it's unreachable, and some other unreachable
285          * object points to it. Ignore it - it's not interesting, and we showed
286          * all the interesting cases above.
287          */
288 }
289
290 static void check_object(struct object *obj)
291 {
292         if (verbose)
293                 fprintf(stderr, "Checking %s\n", describe_object(obj));
294
295         if (obj->flags & REACHABLE)
296                 check_reachable_object(obj);
297         else
298                 check_unreachable_object(obj);
299 }
300
301 static void check_connectivity(void)
302 {
303         int i, max;
304
305         /* Traverse the pending reachable objects */
306         traverse_reachable();
307
308         /* Look up all the requirements, warn about missing objects.. */
309         max = get_max_object_index();
310         if (verbose)
311                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
312
313         for (i = 0; i < max; i++) {
314                 struct object *obj = get_indexed_object(i);
315
316                 if (obj)
317                         check_object(obj);
318         }
319 }
320
321 static int fsck_obj(struct object *obj)
322 {
323         if (obj->flags & SEEN)
324                 return 0;
325         obj->flags |= SEEN;
326
327         if (verbose)
328                 fprintf(stderr, "Checking %s %s\n",
329                         typename(obj->type), describe_object(obj));
330
331         if (fsck_walk(obj, NULL, &fsck_obj_options))
332                 objerror(obj, "broken links");
333         if (fsck_object(obj, NULL, 0, &fsck_obj_options))
334                 return -1;
335
336         if (obj->type == OBJ_TREE) {
337                 struct tree *item = (struct tree *) obj;
338
339                 free_tree_buffer(item);
340         }
341
342         if (obj->type == OBJ_COMMIT) {
343                 struct commit *commit = (struct commit *) obj;
344
345                 free_commit_buffer(commit);
346
347                 if (!commit->parents && show_root)
348                         printf("root %s\n", describe_object(&commit->object));
349         }
350
351         if (obj->type == OBJ_TAG) {
352                 struct tag *tag = (struct tag *) obj;
353
354                 if (show_tags && tag->tagged) {
355                         printf("tagged %s %s", typename(tag->tagged->type),
356                                 describe_object(tag->tagged));
357                         printf(" (%s) in %s\n", tag->tag,
358                                 describe_object(&tag->object));
359                 }
360         }
361
362         return 0;
363 }
364
365 static int fsck_sha1(const unsigned char *sha1)
366 {
367         struct object *obj = parse_object(sha1);
368         if (!obj) {
369                 errors_found |= ERROR_OBJECT;
370                 return error("%s: object corrupt or missing",
371                              sha1_to_hex(sha1));
372         }
373         obj->flags |= HAS_OBJ;
374         return fsck_obj(obj);
375 }
376
377 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
378                            unsigned long size, void *buffer, int *eaten)
379 {
380         /*
381          * Note, buffer may be NULL if type is OBJ_BLOB. See
382          * verify_packfile(), data_valid variable for details.
383          */
384         struct object *obj;
385         obj = parse_object_buffer(sha1, type, size, buffer, eaten);
386         if (!obj) {
387                 errors_found |= ERROR_OBJECT;
388                 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
389         }
390         obj->flags = HAS_OBJ;
391         return fsck_obj(obj);
392 }
393
394 static int default_refs;
395
396 static void fsck_handle_reflog_sha1(const char *refname, unsigned char *sha1,
397         unsigned long timestamp)
398 {
399         struct object *obj;
400
401         if (!is_null_sha1(sha1)) {
402                 obj = lookup_object(sha1);
403                 if (obj) {
404                         if (timestamp && name_objects)
405                                 add_decoration(fsck_walk_options.object_names,
406                                         obj,
407                                         xstrfmt("%s@{%ld}", refname, timestamp));
408                         obj->used = 1;
409                         mark_object_reachable(obj);
410                 } else {
411                         error("%s: invalid reflog entry %s", refname, sha1_to_hex(sha1));
412                         errors_found |= ERROR_REACHABLE;
413                 }
414         }
415 }
416
417 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
418                 const char *email, unsigned long timestamp, int tz,
419                 const char *message, void *cb_data)
420 {
421         const char *refname = cb_data;
422
423         if (verbose)
424                 fprintf(stderr, "Checking reflog %s->%s\n",
425                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
426
427         fsck_handle_reflog_sha1(refname, osha1, 0);
428         fsck_handle_reflog_sha1(refname, nsha1, timestamp);
429         return 0;
430 }
431
432 static int fsck_handle_reflog(const char *logname, const struct object_id *oid,
433                               int flag, void *cb_data)
434 {
435         for_each_reflog_ent(logname, fsck_handle_reflog_ent, (void *)logname);
436         return 0;
437 }
438
439 static int fsck_handle_ref(const char *refname, const struct object_id *oid,
440                            int flag, void *cb_data)
441 {
442         struct object *obj;
443
444         obj = parse_object(oid->hash);
445         if (!obj) {
446                 error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
447                 errors_found |= ERROR_REACHABLE;
448                 /* We'll continue with the rest despite the error.. */
449                 return 0;
450         }
451         if (obj->type != OBJ_COMMIT && is_branch(refname)) {
452                 error("%s: not a commit", refname);
453                 errors_found |= ERROR_REFS;
454         }
455         default_refs++;
456         obj->used = 1;
457         if (name_objects)
458                 add_decoration(fsck_walk_options.object_names,
459                         obj, xstrdup(refname));
460         mark_object_reachable(obj);
461
462         return 0;
463 }
464
465 static void get_default_heads(void)
466 {
467         if (head_points_at && !is_null_oid(&head_oid))
468                 fsck_handle_ref("HEAD", &head_oid, 0, NULL);
469         for_each_rawref(fsck_handle_ref, NULL);
470         if (include_reflogs)
471                 for_each_reflog(fsck_handle_reflog, NULL);
472
473         /*
474          * Not having any default heads isn't really fatal, but
475          * it does mean that "--unreachable" no longer makes any
476          * sense (since in this case everything will obviously
477          * be unreachable by definition.
478          *
479          * Showing dangling objects is valid, though (as those
480          * dangling objects are likely lost heads).
481          *
482          * So we just print a warning about it, and clear the
483          * "show_unreachable" flag.
484          */
485         if (!default_refs) {
486                 fprintf(stderr, "notice: No default references\n");
487                 show_unreachable = 0;
488         }
489 }
490
491 static int fsck_loose(const unsigned char *sha1, const char *path, void *data)
492 {
493         if (fsck_sha1(sha1))
494                 errors_found |= ERROR_OBJECT;
495         return 0;
496 }
497
498 static int fsck_cruft(const char *basename, const char *path, void *data)
499 {
500         if (!starts_with(basename, "tmp_obj_"))
501                 fprintf(stderr, "bad sha1 file: %s\n", path);
502         return 0;
503 }
504
505 static int fsck_subdir(int nr, const char *path, void *progress)
506 {
507         display_progress(progress, nr + 1);
508         return 0;
509 }
510
511 static void fsck_object_dir(const char *path)
512 {
513         struct progress *progress = NULL;
514
515         if (verbose)
516                 fprintf(stderr, "Checking object directory\n");
517
518         if (show_progress)
519                 progress = start_progress(_("Checking object directories"), 256);
520
521         for_each_loose_file_in_objdir(path, fsck_loose, fsck_cruft, fsck_subdir,
522                                       progress);
523         display_progress(progress, 256);
524         stop_progress(&progress);
525 }
526
527 static int fsck_head_link(void)
528 {
529         int null_is_error = 0;
530
531         if (verbose)
532                 fprintf(stderr, "Checking HEAD link\n");
533
534         head_points_at = resolve_ref_unsafe("HEAD", 0, head_oid.hash, NULL);
535         if (!head_points_at) {
536                 errors_found |= ERROR_REFS;
537                 return error("Invalid HEAD");
538         }
539         if (!strcmp(head_points_at, "HEAD"))
540                 /* detached HEAD */
541                 null_is_error = 1;
542         else if (!starts_with(head_points_at, "refs/heads/")) {
543                 errors_found |= ERROR_REFS;
544                 return error("HEAD points to something strange (%s)",
545                              head_points_at);
546         }
547         if (is_null_oid(&head_oid)) {
548                 if (null_is_error) {
549                         errors_found |= ERROR_REFS;
550                         return error("HEAD: detached HEAD points at nothing");
551                 }
552                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
553                         head_points_at + 11);
554         }
555         return 0;
556 }
557
558 static int fsck_cache_tree(struct cache_tree *it)
559 {
560         int i;
561         int err = 0;
562
563         if (verbose)
564                 fprintf(stderr, "Checking cache tree\n");
565
566         if (0 <= it->entry_count) {
567                 struct object *obj = parse_object(it->sha1);
568                 if (!obj) {
569                         error("%s: invalid sha1 pointer in cache-tree",
570                               sha1_to_hex(it->sha1));
571                         errors_found |= ERROR_REFS;
572                         return 1;
573                 }
574                 obj->used = 1;
575                 if (name_objects)
576                         add_decoration(fsck_walk_options.object_names,
577                                 obj, xstrdup(":"));
578                 mark_object_reachable(obj);
579                 if (obj->type != OBJ_TREE)
580                         err |= objerror(obj, "non-tree in cache-tree");
581         }
582         for (i = 0; i < it->subtree_nr; i++)
583                 err |= fsck_cache_tree(it->down[i]->cache_tree);
584         return err;
585 }
586
587 static char const * const fsck_usage[] = {
588         N_("git fsck [<options>] [<object>...]"),
589         NULL
590 };
591
592 static struct option fsck_opts[] = {
593         OPT__VERBOSE(&verbose, N_("be verbose")),
594         OPT_BOOL(0, "unreachable", &show_unreachable, N_("show unreachable objects")),
595         OPT_BOOL(0, "dangling", &show_dangling, N_("show dangling objects")),
596         OPT_BOOL(0, "tags", &show_tags, N_("report tags")),
597         OPT_BOOL(0, "root", &show_root, N_("report root nodes")),
598         OPT_BOOL(0, "cache", &keep_cache_objects, N_("make index objects head nodes")),
599         OPT_BOOL(0, "reflogs", &include_reflogs, N_("make reflogs head nodes (default)")),
600         OPT_BOOL(0, "full", &check_full, N_("also consider packs and alternate objects")),
601         OPT_BOOL(0, "connectivity-only", &connectivity_only, N_("check only connectivity")),
602         OPT_BOOL(0, "strict", &check_strict, N_("enable more strict checking")),
603         OPT_BOOL(0, "lost-found", &write_lost_and_found,
604                                 N_("write dangling objects in .git/lost-found")),
605         OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
606         OPT_BOOL(0, "name-objects", &name_objects, N_("show verbose names for reachable objects")),
607         OPT_END(),
608 };
609
610 int cmd_fsck(int argc, const char **argv, const char *prefix)
611 {
612         int i, heads;
613         struct alternate_object_database *alt;
614
615         errors_found = 0;
616         check_replace_refs = 0;
617
618         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
619
620         fsck_walk_options.walk = mark_object;
621         fsck_obj_options.walk = mark_used;
622         fsck_obj_options.error_func = fsck_error_func;
623         if (check_strict)
624                 fsck_obj_options.strict = 1;
625
626         if (show_progress == -1)
627                 show_progress = isatty(2);
628         if (verbose)
629                 show_progress = 0;
630
631         if (write_lost_and_found) {
632                 check_full = 1;
633                 include_reflogs = 0;
634         }
635
636         if (name_objects)
637                 fsck_walk_options.object_names =
638                         xcalloc(1, sizeof(struct decoration));
639
640         git_config(fsck_config, NULL);
641
642         fsck_head_link();
643         if (!connectivity_only) {
644                 fsck_object_dir(get_object_directory());
645
646                 prepare_alt_odb();
647                 for (alt = alt_odb_list; alt; alt = alt->next)
648                         fsck_object_dir(alt->path);
649         }
650
651         if (check_full) {
652                 struct packed_git *p;
653                 uint32_t total = 0, count = 0;
654                 struct progress *progress = NULL;
655
656                 prepare_packed_git();
657
658                 if (show_progress) {
659                         for (p = packed_git; p; p = p->next) {
660                                 if (open_pack_index(p))
661                                         continue;
662                                 total += p->num_objects;
663                         }
664
665                         progress = start_progress(_("Checking objects"), total);
666                 }
667                 for (p = packed_git; p; p = p->next) {
668                         /* verify gives error messages itself */
669                         if (verify_pack(p, fsck_obj_buffer,
670                                         progress, count))
671                                 errors_found |= ERROR_PACK;
672                         count += p->num_objects;
673                 }
674                 stop_progress(&progress);
675         }
676
677         heads = 0;
678         for (i = 0; i < argc; i++) {
679                 const char *arg = argv[i];
680                 unsigned char sha1[20];
681                 if (!get_sha1(arg, sha1)) {
682                         struct object *obj = lookup_object(sha1);
683
684                         /* Error is printed by lookup_object(). */
685                         if (!obj)
686                                 continue;
687
688                         obj->used = 1;
689                         if (name_objects)
690                                 add_decoration(fsck_walk_options.object_names,
691                                         obj, xstrdup(arg));
692                         mark_object_reachable(obj);
693                         heads++;
694                         continue;
695                 }
696                 error("invalid parameter: expected sha1, got '%s'", arg);
697         }
698
699         /*
700          * If we've not been given any explicit head information, do the
701          * default ones from .git/refs. We also consider the index file
702          * in this case (ie this implies --cache).
703          */
704         if (!heads) {
705                 get_default_heads();
706                 keep_cache_objects = 1;
707         }
708
709         if (keep_cache_objects) {
710                 read_cache();
711                 for (i = 0; i < active_nr; i++) {
712                         unsigned int mode;
713                         struct blob *blob;
714                         struct object *obj;
715
716                         mode = active_cache[i]->ce_mode;
717                         if (S_ISGITLINK(mode))
718                                 continue;
719                         blob = lookup_blob(active_cache[i]->oid.hash);
720                         if (!blob)
721                                 continue;
722                         obj = &blob->object;
723                         obj->used = 1;
724                         if (name_objects)
725                                 add_decoration(fsck_walk_options.object_names,
726                                         obj,
727                                         xstrfmt(":%s", active_cache[i]->name));
728                         mark_object_reachable(obj);
729                 }
730                 if (active_cache_tree)
731                         fsck_cache_tree(active_cache_tree);
732         }
733
734         check_connectivity();
735         return errors_found;
736 }