Merge branch 'jc/commit-hook-authorship' into maint
[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
16 #define REACHABLE 0x0001
17 #define SEEN      0x0002
18
19 static int show_root;
20 static int show_tags;
21 static int show_unreachable;
22 static int include_reflogs = 1;
23 static int check_full = 1;
24 static int check_strict;
25 static int keep_cache_objects;
26 static unsigned char head_sha1[20];
27 static const char *head_points_at;
28 static int errors_found;
29 static int write_lost_and_found;
30 static int verbose;
31 static int show_progress = -1;
32 static int show_dangling = 1;
33 #define ERROR_OBJECT 01
34 #define ERROR_REACHABLE 02
35 #define ERROR_PACK 04
36
37 #ifdef NO_D_INO_IN_DIRENT
38 #define SORT_DIRENT 0
39 #define DIRENT_SORT_HINT(de) 0
40 #else
41 #define SORT_DIRENT 1
42 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
43 #endif
44
45 static void objreport(struct object *obj, const char *severity,
46                       const char *err, va_list params)
47 {
48         fprintf(stderr, "%s in %s %s: ",
49                 severity, typename(obj->type), sha1_to_hex(obj->sha1));
50         vfprintf(stderr, err, params);
51         fputs("\n", stderr);
52 }
53
54 __attribute__((format (printf, 2, 3)))
55 static int objerror(struct object *obj, const char *err, ...)
56 {
57         va_list params;
58         va_start(params, err);
59         errors_found |= ERROR_OBJECT;
60         objreport(obj, "error", err, params);
61         va_end(params);
62         return -1;
63 }
64
65 __attribute__((format (printf, 3, 4)))
66 static int fsck_error_func(struct object *obj, int type, const char *err, ...)
67 {
68         va_list params;
69         va_start(params, err);
70         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
71         va_end(params);
72         return (type == FSCK_WARN) ? 0 : 1;
73 }
74
75 static struct object_array pending;
76
77 static int mark_object(struct object *obj, int type, void *data)
78 {
79         struct object *parent = data;
80
81         /*
82          * The only case data is NULL or type is OBJ_ANY is when
83          * mark_object_reachable() calls us.  All the callers of
84          * that function has non-NULL obj hence ...
85          */
86         if (!obj) {
87                 /* ... these references to parent->fld are safe here */
88                 printf("broken link from %7s %s\n",
89                            typename(parent->type), sha1_to_hex(parent->sha1));
90                 printf("broken link from %7s %s\n",
91                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
92                 errors_found |= ERROR_REACHABLE;
93                 return 1;
94         }
95
96         if (type != OBJ_ANY && obj->type != type)
97                 /* ... and the reference to parent is safe here */
98                 objerror(parent, "wrong object type in link");
99
100         if (obj->flags & REACHABLE)
101                 return 0;
102         obj->flags |= REACHABLE;
103         if (!obj->parsed) {
104                 if (parent && !has_sha1_file(obj->sha1)) {
105                         printf("broken link from %7s %s\n",
106                                  typename(parent->type), sha1_to_hex(parent->sha1));
107                         printf("              to %7s %s\n",
108                                  typename(obj->type), sha1_to_hex(obj->sha1));
109                         errors_found |= ERROR_REACHABLE;
110                 }
111                 return 1;
112         }
113
114         add_object_array(obj, (void *) parent, &pending);
115         return 0;
116 }
117
118 static void mark_object_reachable(struct object *obj)
119 {
120         mark_object(obj, OBJ_ANY, NULL);
121 }
122
123 static int traverse_one_object(struct object *obj)
124 {
125         int result;
126         struct tree *tree = NULL;
127
128         if (obj->type == OBJ_TREE) {
129                 obj->parsed = 0;
130                 tree = (struct tree *)obj;
131                 if (parse_tree(tree) < 0)
132                         return 1; /* error already displayed */
133         }
134         result = fsck_walk(obj, mark_object, obj);
135         if (tree) {
136                 free(tree->buffer);
137                 tree->buffer = NULL;
138         }
139         return result;
140 }
141
142 static int traverse_reachable(void)
143 {
144         struct progress *progress = NULL;
145         unsigned int nr = 0;
146         int result = 0;
147         if (show_progress)
148                 progress = start_progress_delay("Checking connectivity", 0, 0, 2);
149         while (pending.nr) {
150                 struct object_array_entry *entry;
151                 struct object *obj;
152
153                 entry = pending.objects + --pending.nr;
154                 obj = entry->item;
155                 result |= traverse_one_object(obj);
156                 display_progress(progress, ++nr);
157         }
158         stop_progress(&progress);
159         return !!result;
160 }
161
162 static int mark_used(struct object *obj, int type, void *data)
163 {
164         if (!obj)
165                 return 1;
166         obj->used = 1;
167         return 0;
168 }
169
170 /*
171  * Check a single reachable object
172  */
173 static void check_reachable_object(struct object *obj)
174 {
175         /*
176          * We obviously want the object to be parsed,
177          * except if it was in a pack-file and we didn't
178          * do a full fsck
179          */
180         if (!obj->parsed) {
181                 if (has_sha1_pack(obj->sha1))
182                         return; /* it is in pack - forget about it */
183                 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
184                 errors_found |= ERROR_REACHABLE;
185                 return;
186         }
187 }
188
189 /*
190  * Check a single unreachable object
191  */
192 static void check_unreachable_object(struct object *obj)
193 {
194         /*
195          * Missing unreachable object? Ignore it. It's not like
196          * we miss it (since it can't be reached), nor do we want
197          * to complain about it being unreachable (since it does
198          * not exist).
199          */
200         if (!obj->parsed)
201                 return;
202
203         /*
204          * Unreachable object that exists? Show it if asked to,
205          * since this is something that is prunable.
206          */
207         if (show_unreachable) {
208                 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
209                 return;
210         }
211
212         /*
213          * "!used" means that nothing at all points to it, including
214          * other unreachable objects. In other words, it's the "tip"
215          * of some set of unreachable objects, usually a commit that
216          * got dropped.
217          *
218          * Such starting points are more interesting than some random
219          * set of unreachable objects, so we show them even if the user
220          * hasn't asked for _all_ unreachable objects. If you have
221          * deleted a branch by mistake, this is a prime candidate to
222          * start looking at, for example.
223          */
224         if (!obj->used) {
225                 if (show_dangling)
226                         printf("dangling %s %s\n", typename(obj->type),
227                                sha1_to_hex(obj->sha1));
228                 if (write_lost_and_found) {
229                         char *filename = git_path("lost-found/%s/%s",
230                                 obj->type == OBJ_COMMIT ? "commit" : "other",
231                                 sha1_to_hex(obj->sha1));
232                         FILE *f;
233
234                         if (safe_create_leading_directories(filename)) {
235                                 error("Could not create lost-found");
236                                 return;
237                         }
238                         if (!(f = fopen(filename, "w")))
239                                 die_errno("Could not open '%s'", filename);
240                         if (obj->type == OBJ_BLOB) {
241                                 enum object_type type;
242                                 unsigned long size;
243                                 char *buf = read_sha1_file(obj->sha1,
244                                                 &type, &size);
245                                 if (buf && fwrite(buf, 1, size, f) != size)
246                                         die_errno("Could not write '%s'", filename);
247                                 free(buf);
248                         } else
249                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
250                         if (fclose(f))
251                                 die_errno("Could not finish '%s'",
252                                           filename);
253                 }
254                 return;
255         }
256
257         /*
258          * Otherwise? It's there, it's unreachable, and some other unreachable
259          * object points to it. Ignore it - it's not interesting, and we showed
260          * all the interesting cases above.
261          */
262 }
263
264 static void check_object(struct object *obj)
265 {
266         if (verbose)
267                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
268
269         if (obj->flags & REACHABLE)
270                 check_reachable_object(obj);
271         else
272                 check_unreachable_object(obj);
273 }
274
275 static void check_connectivity(void)
276 {
277         int i, max;
278
279         /* Traverse the pending reachable objects */
280         traverse_reachable();
281
282         /* Look up all the requirements, warn about missing objects.. */
283         max = get_max_object_index();
284         if (verbose)
285                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
286
287         for (i = 0; i < max; i++) {
288                 struct object *obj = get_indexed_object(i);
289
290                 if (obj)
291                         check_object(obj);
292         }
293 }
294
295 static int fsck_obj(struct object *obj)
296 {
297         if (obj->flags & SEEN)
298                 return 0;
299         obj->flags |= SEEN;
300
301         if (verbose)
302                 fprintf(stderr, "Checking %s %s\n",
303                         typename(obj->type), sha1_to_hex(obj->sha1));
304
305         if (fsck_walk(obj, mark_used, NULL))
306                 objerror(obj, "broken links");
307         if (fsck_object(obj, check_strict, fsck_error_func))
308                 return -1;
309
310         if (obj->type == OBJ_TREE) {
311                 struct tree *item = (struct tree *) obj;
312
313                 free(item->buffer);
314                 item->buffer = NULL;
315         }
316
317         if (obj->type == OBJ_COMMIT) {
318                 struct commit *commit = (struct commit *) obj;
319
320                 free(commit->buffer);
321                 commit->buffer = NULL;
322
323                 if (!commit->parents && show_root)
324                         printf("root %s\n", sha1_to_hex(commit->object.sha1));
325         }
326
327         if (obj->type == OBJ_TAG) {
328                 struct tag *tag = (struct tag *) obj;
329
330                 if (show_tags && tag->tagged) {
331                         printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
332                         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
333                 }
334         }
335
336         return 0;
337 }
338
339 static int fsck_sha1(const unsigned char *sha1)
340 {
341         struct object *obj = parse_object(sha1);
342         if (!obj) {
343                 errors_found |= ERROR_OBJECT;
344                 return error("%s: object corrupt or missing",
345                              sha1_to_hex(sha1));
346         }
347         return fsck_obj(obj);
348 }
349
350 static int fsck_obj_buffer(const unsigned char *sha1, enum object_type type,
351                            unsigned long size, void *buffer, int *eaten)
352 {
353         struct object *obj;
354         obj = parse_object_buffer(sha1, type, size, buffer, eaten);
355         if (!obj) {
356                 errors_found |= ERROR_OBJECT;
357                 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
358         }
359         return fsck_obj(obj);
360 }
361
362 /*
363  * This is the sorting chunk size: make it reasonably
364  * big so that we can sort well..
365  */
366 #define MAX_SHA1_ENTRIES (1024)
367
368 struct sha1_entry {
369         unsigned long ino;
370         unsigned char sha1[20];
371 };
372
373 static struct {
374         unsigned long nr;
375         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
376 } sha1_list;
377
378 static int ino_compare(const void *_a, const void *_b)
379 {
380         const struct sha1_entry *a = _a, *b = _b;
381         unsigned long ino1 = a->ino, ino2 = b->ino;
382         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
383 }
384
385 static void fsck_sha1_list(void)
386 {
387         int i, nr = sha1_list.nr;
388
389         if (SORT_DIRENT)
390                 qsort(sha1_list.entry, nr,
391                       sizeof(struct sha1_entry *), ino_compare);
392         for (i = 0; i < nr; i++) {
393                 struct sha1_entry *entry = sha1_list.entry[i];
394                 unsigned char *sha1 = entry->sha1;
395
396                 sha1_list.entry[i] = NULL;
397                 fsck_sha1(sha1);
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 (!prefixcmp(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 unsigned char *sha1, int flag, void *cb_data)
485 {
486         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
487         return 0;
488 }
489
490 static int is_branch(const char *refname)
491 {
492         return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
493 }
494
495 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
496 {
497         struct object *obj;
498
499         obj = parse_object(sha1);
500         if (!obj) {
501                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
502                 /* We'll continue with the rest despite the error.. */
503                 return 0;
504         }
505         if (obj->type != OBJ_COMMIT && is_branch(refname))
506                 error("%s: not a commit", refname);
507         default_refs++;
508         obj->used = 1;
509         mark_object_reachable(obj);
510
511         return 0;
512 }
513
514 static void get_default_heads(void)
515 {
516         if (head_points_at && !is_null_sha1(head_sha1))
517                 fsck_handle_ref("HEAD", head_sha1, 0, NULL);
518         for_each_ref(fsck_handle_ref, NULL);
519         if (include_reflogs)
520                 for_each_reflog(fsck_handle_reflog, NULL);
521
522         /*
523          * Not having any default heads isn't really fatal, but
524          * it does mean that "--unreachable" no longer makes any
525          * sense (since in this case everything will obviously
526          * be unreachable by definition.
527          *
528          * Showing dangling objects is valid, though (as those
529          * dangling objects are likely lost heads).
530          *
531          * So we just print a warning about it, and clear the
532          * "show_unreachable" flag.
533          */
534         if (!default_refs) {
535                 fprintf(stderr, "notice: No default references\n");
536                 show_unreachable = 0;
537         }
538 }
539
540 static void fsck_object_dir(const char *path)
541 {
542         int i;
543         struct progress *progress = NULL;
544
545         if (verbose)
546                 fprintf(stderr, "Checking object directory\n");
547
548         if (show_progress)
549                 progress = start_progress("Checking object directories", 256);
550         for (i = 0; i < 256; i++) {
551                 static char dir[4096];
552                 sprintf(dir, "%s/%02x", path, i);
553                 fsck_dir(i, dir);
554                 display_progress(progress, i+1);
555         }
556         stop_progress(&progress);
557         fsck_sha1_list();
558 }
559
560 static int fsck_head_link(void)
561 {
562         int flag;
563         int null_is_error = 0;
564
565         if (verbose)
566                 fprintf(stderr, "Checking HEAD link\n");
567
568         head_points_at = resolve_ref_unsafe("HEAD", head_sha1, 0, &flag);
569         if (!head_points_at)
570                 return error("Invalid HEAD");
571         if (!strcmp(head_points_at, "HEAD"))
572                 /* detached HEAD */
573                 null_is_error = 1;
574         else if (prefixcmp(head_points_at, "refs/heads/"))
575                 return error("HEAD points to something strange (%s)",
576                              head_points_at);
577         if (is_null_sha1(head_sha1)) {
578                 if (null_is_error)
579                         return error("HEAD: detached HEAD points at nothing");
580                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
581                         head_points_at + 11);
582         }
583         return 0;
584 }
585
586 static int fsck_cache_tree(struct cache_tree *it)
587 {
588         int i;
589         int err = 0;
590
591         if (verbose)
592                 fprintf(stderr, "Checking cache tree\n");
593
594         if (0 <= it->entry_count) {
595                 struct object *obj = parse_object(it->sha1);
596                 if (!obj) {
597                         error("%s: invalid sha1 pointer in cache-tree",
598                               sha1_to_hex(it->sha1));
599                         return 1;
600                 }
601                 obj->used = 1;
602                 mark_object_reachable(obj);
603                 if (obj->type != OBJ_TREE)
604                         err |= objerror(obj, "non-tree in cache-tree");
605         }
606         for (i = 0; i < it->subtree_nr; i++)
607                 err |= fsck_cache_tree(it->down[i]->cache_tree);
608         return err;
609 }
610
611 static char const * const fsck_usage[] = {
612         "git fsck [options] [<object>...]",
613         NULL
614 };
615
616 static struct option fsck_opts[] = {
617         OPT__VERBOSE(&verbose, "be verbose"),
618         OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
619         OPT_BOOL(0, "dangling", &show_dangling, "show dangling objects"),
620         OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
621         OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
622         OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
623         OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
624         OPT_BOOLEAN(0, "full", &check_full, "also consider packs and alternate objects"),
625         OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
626         OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
627                                 "write dangling objects in .git/lost-found"),
628         OPT_BOOL(0, "progress", &show_progress, "show progress"),
629         OPT_END(),
630 };
631
632 int cmd_fsck(int argc, const char **argv, const char *prefix)
633 {
634         int i, heads;
635         struct alternate_object_database *alt;
636
637         errors_found = 0;
638         read_replace_refs = 0;
639
640         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
641
642         if (show_progress == -1)
643                 show_progress = isatty(2);
644         if (verbose)
645                 show_progress = 0;
646
647         if (write_lost_and_found) {
648                 check_full = 1;
649                 include_reflogs = 0;
650         }
651
652         fsck_head_link();
653         fsck_object_dir(get_object_directory());
654
655         prepare_alt_odb();
656         for (alt = alt_odb_list; alt; alt = alt->next) {
657                 char namebuf[PATH_MAX];
658                 int namelen = alt->name - alt->base;
659                 memcpy(namebuf, alt->base, namelen);
660                 namebuf[namelen - 1] = 0;
661                 fsck_object_dir(namebuf);
662         }
663
664         if (check_full) {
665                 struct packed_git *p;
666                 uint32_t total = 0, count = 0;
667                 struct progress *progress = NULL;
668
669                 prepare_packed_git();
670
671                 if (show_progress) {
672                         for (p = packed_git; p; p = p->next) {
673                                 if (open_pack_index(p))
674                                         continue;
675                                 total += p->num_objects;
676                         }
677
678                         progress = start_progress("Checking objects", total);
679                 }
680                 for (p = packed_git; p; p = p->next) {
681                         /* verify gives error messages itself */
682                         if (verify_pack(p, fsck_obj_buffer,
683                                         progress, count))
684                                 errors_found |= ERROR_PACK;
685                         count += p->num_objects;
686                 }
687                 stop_progress(&progress);
688         }
689
690         heads = 0;
691         for (i = 0; i < argc; i++) {
692                 const char *arg = argv[i];
693                 unsigned char sha1[20];
694                 if (!get_sha1(arg, sha1)) {
695                         struct object *obj = lookup_object(sha1);
696
697                         /* Error is printed by lookup_object(). */
698                         if (!obj)
699                                 continue;
700
701                         obj->used = 1;
702                         mark_object_reachable(obj);
703                         heads++;
704                         continue;
705                 }
706                 error("invalid parameter: expected sha1, got '%s'", arg);
707         }
708
709         /*
710          * If we've not been given any explicit head information, do the
711          * default ones from .git/refs. We also consider the index file
712          * in this case (ie this implies --cache).
713          */
714         if (!heads) {
715                 get_default_heads();
716                 keep_cache_objects = 1;
717         }
718
719         if (keep_cache_objects) {
720                 read_cache();
721                 for (i = 0; i < active_nr; i++) {
722                         unsigned int mode;
723                         struct blob *blob;
724                         struct object *obj;
725
726                         mode = active_cache[i]->ce_mode;
727                         if (S_ISGITLINK(mode))
728                                 continue;
729                         blob = lookup_blob(active_cache[i]->sha1);
730                         if (!blob)
731                                 continue;
732                         obj = &blob->object;
733                         obj->used = 1;
734                         mark_object_reachable(obj);
735                 }
736                 if (active_cache_tree)
737                         fsck_cache_tree(active_cache_tree);
738         }
739
740         check_connectivity();
741         return errors_found;
742 }