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