fsck: introduce `git fsck --connectivity-only`
[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
42 #ifdef NO_D_INO_IN_DIRENT
43 #define SORT_DIRENT 0
44 #define DIRENT_SORT_HINT(de) 0
45 #else
46 #define SORT_DIRENT 1
47 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
48 #endif
49
50 static int fsck_config(const char *var, const char *value, void *cb)
51 {
52         if (skip_prefix(var, "fsck.", &var)) {
53                 fsck_set_msg_type(&fsck_obj_options, var, value);
54                 return 0;
55         }
56
57         return git_default_config(var, value, cb);
58 }
59
60 static void objreport(struct object *obj, const char *msg_type,
61                         const char *err)
62 {
63         fprintf(stderr, "%s in %s %s: %s\n",
64                 msg_type, typename(obj->type), sha1_to_hex(obj->sha1), err);
65 }
66
67 static int objerror(struct object *obj, const char *err)
68 {
69         errors_found |= ERROR_OBJECT;
70         objreport(obj, "error", err);
71         return -1;
72 }
73
74 static int fsck_error_func(struct object *obj, int type, const char *message)
75 {
76         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", message);
77         return (type == FSCK_WARN) ? 0 : 1;
78 }
79
80 static struct object_array pending;
81
82 static int mark_object(struct object *obj, int type, void *data, struct fsck_options *options)
83 {
84         struct object *parent = data;
85
86         /*
87          * The only case data is NULL or type is OBJ_ANY is when
88          * mark_object_reachable() calls us.  All the callers of
89          * that function has non-NULL obj hence ...
90          */
91         if (!obj) {
92                 /* ... these references to parent->fld are safe here */
93                 printf("broken link from %7s %s\n",
94                            typename(parent->type), sha1_to_hex(parent->sha1));
95                 printf("broken link from %7s %s\n",
96                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
97                 errors_found |= ERROR_REACHABLE;
98                 return 1;
99         }
100
101         if (type != OBJ_ANY && obj->type != type)
102                 /* ... and the reference to parent is safe here */
103                 objerror(parent, "wrong object type in link");
104
105         if (obj->flags & REACHABLE)
106                 return 0;
107         obj->flags |= REACHABLE;
108         if (!(obj->flags & HAS_OBJ)) {
109                 if (parent && !has_sha1_file(obj->sha1)) {
110                         printf("broken link from %7s %s\n",
111                                  typename(parent->type), sha1_to_hex(parent->sha1));
112                         printf("              to %7s %s\n",
113                                  typename(obj->type), sha1_to_hex(obj->sha1));
114                         errors_found |= ERROR_REACHABLE;
115                 }
116                 return 1;
117         }
118
119         add_object_array(obj, NULL, &pending);
120         return 0;
121 }
122
123 static void mark_object_reachable(struct object *obj)
124 {
125         mark_object(obj, OBJ_ANY, NULL, NULL);
126 }
127
128 static int traverse_one_object(struct object *obj)
129 {
130         int result;
131         struct tree *tree = NULL;
132
133         if (obj->type == OBJ_TREE) {
134                 tree = (struct tree *)obj;
135                 if (parse_tree(tree) < 0)
136                         return 1; /* error already displayed */
137         }
138         result = fsck_walk(obj, obj, &fsck_walk_options);
139         if (tree)
140                 free_tree_buffer(tree);
141         return result;
142 }
143
144 static int traverse_reachable(void)
145 {
146         struct progress *progress = NULL;
147         unsigned int nr = 0;
148         int result = 0;
149         if (show_progress)
150                 progress = start_progress_delay(_("Checking connectivity"), 0, 0, 2);
151         while (pending.nr) {
152                 struct object_array_entry *entry;
153                 struct object *obj;
154
155                 entry = pending.objects + --pending.nr;
156                 obj = entry->item;
157                 result |= traverse_one_object(obj);
158                 display_progress(progress, ++nr);
159         }
160         stop_progress(&progress);
161         return !!result;
162 }
163
164 static int mark_used(struct object *obj, int type, void *data, struct fsck_options *options)
165 {
166         if (!obj)
167                 return 1;
168         obj->used = 1;
169         return 0;
170 }
171
172 /*
173  * Check a single reachable object
174  */
175 static void check_reachable_object(struct object *obj)
176 {
177         /*
178          * We obviously want the object to be parsed,
179          * except if it was in a pack-file and we didn't
180          * do a full fsck
181          */
182         if (!(obj->flags & HAS_OBJ)) {
183                 if (has_sha1_pack(obj->sha1))
184                         return; /* it is in pack - forget about it */
185                 if (connectivity_only && has_sha1_file(obj->sha1))
186                         return;
187                 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
188                 errors_found |= ERROR_REACHABLE;
189                 return;
190         }
191 }
192
193 /*
194  * Check a single unreachable object
195  */
196 static void check_unreachable_object(struct object *obj)
197 {
198         /*
199          * Missing unreachable object? Ignore it. It's not like
200          * we miss it (since it can't be reached), nor do we want
201          * to complain about it being unreachable (since it does
202          * not exist).
203          */
204         if (!obj->parsed)
205                 return;
206
207         /*
208          * Unreachable object that exists? Show it if asked to,
209          * since this is something that is prunable.
210          */
211         if (show_unreachable) {
212                 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
213                 return;
214         }
215
216         /*
217          * "!used" means that nothing at all points to it, including
218          * other unreachable objects. In other words, it's the "tip"
219          * of some set of unreachable objects, usually a commit that
220          * got dropped.
221          *
222          * Such starting points are more interesting than some random
223          * set of unreachable objects, so we show them even if the user
224          * hasn't asked for _all_ unreachable objects. If you have
225          * deleted a branch by mistake, this is a prime candidate to
226          * start looking at, for example.
227          */
228         if (!obj->used) {
229                 if (show_dangling)
230                         printf("dangling %s %s\n", typename(obj->type),
231                                sha1_to_hex(obj->sha1));
232                 if (write_lost_and_found) {
233                         const char *filename = git_path("lost-found/%s/%s",
234                                 obj->type == OBJ_COMMIT ? "commit" : "other",
235                                 sha1_to_hex(obj->sha1));
236                         FILE *f;
237
238                         if (safe_create_leading_directories_const(filename)) {
239                                 error("Could not create lost-found");
240                                 return;
241                         }
242                         if (!(f = fopen(filename, "w")))
243                                 die_errno("Could not open '%s'", filename);
244                         if (obj->type == OBJ_BLOB) {
245                                 if (stream_blob_to_fd(fileno(f), obj->sha1, NULL, 1))
246                                         die_errno("Could not write '%s'", filename);
247                         } else
248                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
249                         if (fclose(f))
250                                 die_errno("Could not finish '%s'",
251                                           filename);
252                 }
253                 return;
254         }
255
256         /*
257          * Otherwise? It's there, it's unreachable, and some other unreachable
258          * object points to it. Ignore it - it's not interesting, and we showed
259          * all the interesting cases above.
260          */
261 }
262
263 static void check_object(struct object *obj)
264 {
265         if (verbose)
266                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
267
268         if (obj->flags & REACHABLE)
269                 check_reachable_object(obj);
270         else
271                 check_unreachable_object(obj);
272 }
273
274 static void check_connectivity(void)
275 {
276         int i, max;
277
278         /* Traverse the pending reachable objects */
279         traverse_reachable();
280
281         /* Look up all the requirements, warn about missing objects.. */
282         max = get_max_object_index();
283         if (verbose)
284                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
285
286         for (i = 0; i < max; i++) {
287                 struct object *obj = get_indexed_object(i);
288
289                 if (obj)
290                         check_object(obj);
291         }
292 }
293
294 static int fsck_obj(struct object *obj)
295 {
296         if (obj->flags & SEEN)
297                 return 0;
298         obj->flags |= SEEN;
299
300         if (verbose)
301                 fprintf(stderr, "Checking %s %s\n",
302                         typename(obj->type), sha1_to_hex(obj->sha1));
303
304         if (fsck_walk(obj, NULL, &fsck_obj_options))
305                 objerror(obj, "broken links");
306         if (fsck_object(obj, NULL, 0, &fsck_obj_options))
307                 return -1;
308
309         if (obj->type == OBJ_TREE) {
310                 struct tree *item = (struct tree *) obj;
311
312                 free_tree_buffer(item);
313         }
314
315         if (obj->type == OBJ_COMMIT) {
316                 struct commit *commit = (struct commit *) obj;
317
318                 free_commit_buffer(commit);
319
320                 if (!commit->parents && show_root)
321                         printf("root %s\n", sha1_to_hex(commit->object.sha1));
322         }
323
324         if (obj->type == OBJ_TAG) {
325                 struct tag *tag = (struct tag *) obj;
326
327                 if (show_tags && tag->tagged) {
328                         printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
329                         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
330                 }
331         }
332
333         return 0;
334 }
335
336 static int fsck_sha1(const unsigned char *sha1)
337 {
338         struct object *obj = parse_object(sha1);
339         if (!obj) {
340                 errors_found |= ERROR_OBJECT;
341                 return error("%s: object corrupt or missing",
342                              sha1_to_hex(sha1));
343         }
344         obj->flags |= HAS_OBJ;
345         return fsck_obj(obj);
346 }
347
348 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
349                            unsigned long size, void *buffer, int *eaten)
350 {
351         struct object *obj;
352         obj = parse_object_buffer(sha1, type, size, buffer, eaten);
353         if (!obj) {
354                 errors_found |= ERROR_OBJECT;
355                 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
356         }
357         obj->flags = HAS_OBJ;
358         return fsck_obj(obj);
359 }
360
361 /*
362  * This is the sorting chunk size: make it reasonably
363  * big so that we can sort well..
364  */
365 #define MAX_SHA1_ENTRIES (1024)
366
367 struct sha1_entry {
368         unsigned long ino;
369         unsigned char sha1[20];
370 };
371
372 static struct {
373         unsigned long nr;
374         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
375 } sha1_list;
376
377 static int ino_compare(const void *_a, const void *_b)
378 {
379         const struct sha1_entry *a = _a, *b = _b;
380         unsigned long ino1 = a->ino, ino2 = b->ino;
381         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
382 }
383
384 static void fsck_sha1_list(void)
385 {
386         int i, nr = sha1_list.nr;
387
388         if (SORT_DIRENT)
389                 qsort(sha1_list.entry, nr,
390                       sizeof(struct sha1_entry *), ino_compare);
391         for (i = 0; i < nr; i++) {
392                 struct sha1_entry *entry = sha1_list.entry[i];
393                 unsigned char *sha1 = entry->sha1;
394
395                 sha1_list.entry[i] = NULL;
396                 if (fsck_sha1(sha1))
397                         errors_found |= ERROR_OBJECT;
398                 free(entry);
399         }
400         sha1_list.nr = 0;
401 }
402
403 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
404 {
405         struct sha1_entry *entry = xmalloc(sizeof(*entry));
406         int nr;
407
408         entry->ino = ino;
409         hashcpy(entry->sha1, sha1);
410         nr = sha1_list.nr;
411         if (nr == MAX_SHA1_ENTRIES) {
412                 fsck_sha1_list();
413                 nr = 0;
414         }
415         sha1_list.entry[nr] = entry;
416         sha1_list.nr = ++nr;
417 }
418
419 static inline int is_loose_object_file(struct dirent *de,
420                                        char *name, unsigned char *sha1)
421 {
422         if (strlen(de->d_name) != 38)
423                 return 0;
424         memcpy(name + 2, de->d_name, 39);
425         return !get_sha1_hex(name, sha1);
426 }
427
428 static void fsck_dir(int i, char *path)
429 {
430         DIR *dir = opendir(path);
431         struct dirent *de;
432         char name[100];
433
434         if (!dir)
435                 return;
436
437         if (verbose)
438                 fprintf(stderr, "Checking directory %s\n", path);
439
440         sprintf(name, "%02x", i);
441         while ((de = readdir(dir)) != NULL) {
442                 unsigned char sha1[20];
443
444                 if (is_dot_or_dotdot(de->d_name))
445                         continue;
446                 if (is_loose_object_file(de, name, sha1)) {
447                         add_sha1_list(sha1, DIRENT_SORT_HINT(de));
448                         continue;
449                 }
450                 if (starts_with(de->d_name, "tmp_obj_"))
451                         continue;
452                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
453         }
454         closedir(dir);
455 }
456
457 static int default_refs;
458
459 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
460                 const char *email, unsigned long timestamp, int tz,
461                 const char *message, void *cb_data)
462 {
463         struct object *obj;
464
465         if (verbose)
466                 fprintf(stderr, "Checking reflog %s->%s\n",
467                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
468
469         if (!is_null_sha1(osha1)) {
470                 obj = lookup_object(osha1);
471                 if (obj) {
472                         obj->used = 1;
473                         mark_object_reachable(obj);
474                 }
475         }
476         obj = lookup_object(nsha1);
477         if (obj) {
478                 obj->used = 1;
479                 mark_object_reachable(obj);
480         }
481         return 0;
482 }
483
484 static int fsck_handle_reflog(const char *logname, const struct object_id *oid,
485                               int flag, void *cb_data)
486 {
487         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
488         return 0;
489 }
490
491 static int fsck_handle_ref(const char *refname, const struct object_id *oid,
492                            int flag, void *cb_data)
493 {
494         struct object *obj;
495
496         obj = parse_object(oid->hash);
497         if (!obj) {
498                 error("%s: invalid sha1 pointer %s", refname, oid_to_hex(oid));
499                 errors_found |= ERROR_REACHABLE;
500                 /* We'll continue with the rest despite the error.. */
501                 return 0;
502         }
503         if (obj->type != OBJ_COMMIT && is_branch(refname))
504                 error("%s: not a commit", refname);
505         default_refs++;
506         obj->used = 1;
507         mark_object_reachable(obj);
508
509         return 0;
510 }
511
512 static void get_default_heads(void)
513 {
514         if (head_points_at && !is_null_oid(&head_oid))
515                 fsck_handle_ref("HEAD", &head_oid, 0, NULL);
516         for_each_rawref(fsck_handle_ref, NULL);
517         if (include_reflogs)
518                 for_each_reflog(fsck_handle_reflog, NULL);
519
520         /*
521          * Not having any default heads isn't really fatal, but
522          * it does mean that "--unreachable" no longer makes any
523          * sense (since in this case everything will obviously
524          * be unreachable by definition.
525          *
526          * Showing dangling objects is valid, though (as those
527          * dangling objects are likely lost heads).
528          *
529          * So we just print a warning about it, and clear the
530          * "show_unreachable" flag.
531          */
532         if (!default_refs) {
533                 fprintf(stderr, "notice: No default references\n");
534                 show_unreachable = 0;
535         }
536 }
537
538 static void fsck_object_dir(const char *path)
539 {
540         int i;
541         struct progress *progress = NULL;
542
543         if (verbose)
544                 fprintf(stderr, "Checking object directory\n");
545
546         if (show_progress)
547                 progress = start_progress(_("Checking object directories"), 256);
548         for (i = 0; i < 256; i++) {
549                 static char dir[4096];
550                 sprintf(dir, "%s/%02x", path, i);
551                 fsck_dir(i, dir);
552                 display_progress(progress, i+1);
553         }
554         stop_progress(&progress);
555         fsck_sha1_list();
556 }
557
558 static int fsck_head_link(void)
559 {
560         int flag;
561         int null_is_error = 0;
562
563         if (verbose)
564                 fprintf(stderr, "Checking HEAD link\n");
565
566         head_points_at = resolve_ref_unsafe("HEAD", 0, head_oid.hash, &flag);
567         if (!head_points_at)
568                 return error("Invalid HEAD");
569         if (!strcmp(head_points_at, "HEAD"))
570                 /* detached HEAD */
571                 null_is_error = 1;
572         else if (!starts_with(head_points_at, "refs/heads/"))
573                 return error("HEAD points to something strange (%s)",
574                              head_points_at);
575         if (is_null_oid(&head_oid)) {
576                 if (null_is_error)
577                         return error("HEAD: detached HEAD points at nothing");
578                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
579                         head_points_at + 11);
580         }
581         return 0;
582 }
583
584 static int fsck_cache_tree(struct cache_tree *it)
585 {
586         int i;
587         int err = 0;
588
589         if (verbose)
590                 fprintf(stderr, "Checking cache tree\n");
591
592         if (0 <= it->entry_count) {
593                 struct object *obj = parse_object(it->sha1);
594                 if (!obj) {
595                         error("%s: invalid sha1 pointer in cache-tree",
596                               sha1_to_hex(it->sha1));
597                         return 1;
598                 }
599                 obj->used = 1;
600                 mark_object_reachable(obj);
601                 if (obj->type != OBJ_TREE)
602                         err |= objerror(obj, "non-tree in cache-tree");
603         }
604         for (i = 0; i < it->subtree_nr; i++)
605                 err |= fsck_cache_tree(it->down[i]->cache_tree);
606         return err;
607 }
608
609 static char const * const fsck_usage[] = {
610         N_("git fsck [<options>] [<object>...]"),
611         NULL
612 };
613
614 static struct option fsck_opts[] = {
615         OPT__VERBOSE(&verbose, N_("be verbose")),
616         OPT_BOOL(0, "unreachable", &show_unreachable, N_("show unreachable objects")),
617         OPT_BOOL(0, "dangling", &show_dangling, N_("show dangling objects")),
618         OPT_BOOL(0, "tags", &show_tags, N_("report tags")),
619         OPT_BOOL(0, "root", &show_root, N_("report root nodes")),
620         OPT_BOOL(0, "cache", &keep_cache_objects, N_("make index objects head nodes")),
621         OPT_BOOL(0, "reflogs", &include_reflogs, N_("make reflogs head nodes (default)")),
622         OPT_BOOL(0, "full", &check_full, N_("also consider packs and alternate objects")),
623         OPT_BOOL(0, "connectivity-only", &connectivity_only, N_("check only connectivity")),
624         OPT_BOOL(0, "strict", &check_strict, N_("enable more strict checking")),
625         OPT_BOOL(0, "lost-found", &write_lost_and_found,
626                                 N_("write dangling objects in .git/lost-found")),
627         OPT_BOOL(0, "progress", &show_progress, N_("show progress")),
628         OPT_END(),
629 };
630
631 int cmd_fsck(int argc, const char **argv, const char *prefix)
632 {
633         int i, heads;
634         struct alternate_object_database *alt;
635
636         errors_found = 0;
637         check_replace_refs = 0;
638
639         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
640
641         fsck_walk_options.walk = mark_object;
642         fsck_obj_options.walk = mark_used;
643         fsck_obj_options.error_func = fsck_error_func;
644         if (check_strict)
645                 fsck_obj_options.strict = 1;
646
647         if (show_progress == -1)
648                 show_progress = isatty(2);
649         if (verbose)
650                 show_progress = 0;
651
652         if (write_lost_and_found) {
653                 check_full = 1;
654                 include_reflogs = 0;
655         }
656
657         git_config(fsck_config, NULL);
658
659         fsck_head_link();
660         if (!connectivity_only)
661                 fsck_object_dir(get_object_directory());
662
663         prepare_alt_odb();
664         for (alt = alt_odb_list; alt; alt = alt->next) {
665                 char namebuf[PATH_MAX];
666                 int namelen = alt->name - alt->base;
667                 memcpy(namebuf, alt->base, namelen);
668                 namebuf[namelen - 1] = 0;
669                 fsck_object_dir(namebuf);
670         }
671
672         if (check_full) {
673                 struct packed_git *p;
674                 uint32_t total = 0, count = 0;
675                 struct progress *progress = NULL;
676
677                 prepare_packed_git();
678
679                 if (show_progress) {
680                         for (p = packed_git; p; p = p->next) {
681                                 if (open_pack_index(p))
682                                         continue;
683                                 total += p->num_objects;
684                         }
685
686                         progress = start_progress(_("Checking objects"), total);
687                 }
688                 for (p = packed_git; p; p = p->next) {
689                         /* verify gives error messages itself */
690                         if (verify_pack(p, fsck_obj_buffer,
691                                         progress, count))
692                                 errors_found |= ERROR_PACK;
693                         count += p->num_objects;
694                 }
695                 stop_progress(&progress);
696         }
697
698         heads = 0;
699         for (i = 0; i < argc; i++) {
700                 const char *arg = argv[i];
701                 unsigned char sha1[20];
702                 if (!get_sha1(arg, sha1)) {
703                         struct object *obj = lookup_object(sha1);
704
705                         /* Error is printed by lookup_object(). */
706                         if (!obj)
707                                 continue;
708
709                         obj->used = 1;
710                         mark_object_reachable(obj);
711                         heads++;
712                         continue;
713                 }
714                 error("invalid parameter: expected sha1, got '%s'", arg);
715         }
716
717         /*
718          * If we've not been given any explicit head information, do the
719          * default ones from .git/refs. We also consider the index file
720          * in this case (ie this implies --cache).
721          */
722         if (!heads) {
723                 get_default_heads();
724                 keep_cache_objects = 1;
725         }
726
727         if (keep_cache_objects) {
728                 read_cache();
729                 for (i = 0; i < active_nr; i++) {
730                         unsigned int mode;
731                         struct blob *blob;
732                         struct object *obj;
733
734                         mode = active_cache[i]->ce_mode;
735                         if (S_ISGITLINK(mode))
736                                 continue;
737                         blob = lookup_blob(active_cache[i]->sha1);
738                         if (!blob)
739                                 continue;
740                         obj = &blob->object;
741                         obj->used = 1;
742                         mark_object_reachable(obj);
743                 }
744                 if (active_cache_tree)
745                         fsck_cache_tree(active_cache_tree);
746         }
747
748         check_connectivity();
749         return errors_found;
750 }