Drop strbuf's 'eof' marker, and make read_line a first class citizen.
[git/dscho.git] / builtin-fsck.c
blob8d12287f037c499acad26ea81acab73490c38d5c
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"
12 #define REACHABLE 0x0001
13 #define SEEN      0x0002
15 static int show_root;
16 static int show_tags;
17 static int show_unreachable;
18 static int include_reflogs = 1;
19 static int check_full;
20 static int check_strict;
21 static int keep_cache_objects;
22 static unsigned char head_sha1[20];
23 static int errors_found;
24 static int write_lost_and_found;
25 static int verbose;
26 #define ERROR_OBJECT 01
27 #define ERROR_REACHABLE 02
29 #ifdef NO_D_INO_IN_DIRENT
30 #define SORT_DIRENT 0
31 #define DIRENT_SORT_HINT(de) 0
32 #else
33 #define SORT_DIRENT 1
34 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
35 #endif
37 static void objreport(struct object *obj, const char *severity,
38                       const char *err, va_list params)
40         fprintf(stderr, "%s in %s %s: ",
41                 severity, typename(obj->type), sha1_to_hex(obj->sha1));
42         vfprintf(stderr, err, params);
43         fputs("\n", stderr);
46 static int objerror(struct object *obj, const char *err, ...)
48         va_list params;
49         va_start(params, err);
50         errors_found |= ERROR_OBJECT;
51         objreport(obj, "error", err, params);
52         va_end(params);
53         return -1;
56 static int objwarning(struct object *obj, const char *err, ...)
58         va_list params;
59         va_start(params, err);
60         objreport(obj, "warning", err, params);
61         va_end(params);
62         return -1;
66  * Check a single reachable object
67  */
68 static void check_reachable_object(struct object *obj)
70         const struct object_refs *refs;
72         /*
73          * We obviously want the object to be parsed,
74          * except if it was in a pack-file and we didn't
75          * do a full fsck
76          */
77         if (!obj->parsed) {
78                 if (has_sha1_pack(obj->sha1, NULL))
79                         return; /* it is in pack - forget about it */
80                 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
81                 errors_found |= ERROR_REACHABLE;
82                 return;
83         }
85         /*
86          * Check that everything that we try to reference is also good.
87          */
88         refs = lookup_object_refs(obj);
89         if (refs) {
90                 unsigned j;
91                 for (j = 0; j < refs->count; j++) {
92                         struct object *ref = refs->ref[j];
93                         if (ref->parsed ||
94                             (has_sha1_file(ref->sha1)))
95                                 continue;
96                         printf("broken link from %7s %s\n",
97                                typename(obj->type), sha1_to_hex(obj->sha1));
98                         printf("              to %7s %s\n",
99                                typename(ref->type), sha1_to_hex(ref->sha1));
100                         errors_found |= ERROR_REACHABLE;
101                 }
102         }
106  * Check a single unreachable object
107  */
108 static void check_unreachable_object(struct object *obj)
110         /*
111          * Missing unreachable object? Ignore it. It's not like
112          * we miss it (since it can't be reached), nor do we want
113          * to complain about it being unreachable (since it does
114          * not exist).
115          */
116         if (!obj->parsed)
117                 return;
119         /*
120          * Unreachable object that exists? Show it if asked to,
121          * since this is something that is prunable.
122          */
123         if (show_unreachable) {
124                 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
125                 return;
126         }
128         /*
129          * "!used" means that nothing at all points to it, including
130          * other unreachable objects. In other words, it's the "tip"
131          * of some set of unreachable objects, usually a commit that
132          * got dropped.
133          *
134          * Such starting points are more interesting than some random
135          * set of unreachable objects, so we show them even if the user
136          * hasn't asked for _all_ unreachable objects. If you have
137          * deleted a branch by mistake, this is a prime candidate to
138          * start looking at, for example.
139          */
140         if (!obj->used) {
141                 printf("dangling %s %s\n", typename(obj->type),
142                        sha1_to_hex(obj->sha1));
143                 if (write_lost_and_found) {
144                         char *filename = git_path("lost-found/%s/%s",
145                                 obj->type == OBJ_COMMIT ? "commit" : "other",
146                                 sha1_to_hex(obj->sha1));
147                         FILE *f;
149                         if (safe_create_leading_directories(filename)) {
150                                 error("Could not create lost-found");
151                                 return;
152                         }
153                         if (!(f = fopen(filename, "w")))
154                                 die("Could not open %s", filename);
155                         if (obj->type == OBJ_BLOB) {
156                                 enum object_type type;
157                                 unsigned long size;
158                                 char *buf = read_sha1_file(obj->sha1,
159                                                 &type, &size);
160                                 if (buf) {
161                                         fwrite(buf, size, 1, f);
162                                         free(buf);
163                                 }
164                         } else
165                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
166                         fclose(f);
167                 }
168                 return;
169         }
171         /*
172          * Otherwise? It's there, it's unreachable, and some other unreachable
173          * object points to it. Ignore it - it's not interesting, and we showed
174          * all the interesting cases above.
175          */
178 static void check_object(struct object *obj)
180         if (verbose)
181                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
183         if (obj->flags & REACHABLE)
184                 check_reachable_object(obj);
185         else
186                 check_unreachable_object(obj);
189 static void check_connectivity(void)
191         int i, max;
193         /* Look up all the requirements, warn about missing objects.. */
194         max = get_max_object_index();
195         if (verbose)
196                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
198         for (i = 0; i < max; i++) {
199                 struct object *obj = get_indexed_object(i);
201                 if (obj)
202                         check_object(obj);
203         }
207  * The entries in a tree are ordered in the _path_ order,
208  * which means that a directory entry is ordered by adding
209  * a slash to the end of it.
211  * So a directory called "a" is ordered _after_ a file
212  * called "a.c", because "a/" sorts after "a.c".
213  */
214 #define TREE_UNORDERED (-1)
215 #define TREE_HAS_DUPS  (-2)
217 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
219         int len1 = strlen(name1);
220         int len2 = strlen(name2);
221         int len = len1 < len2 ? len1 : len2;
222         unsigned char c1, c2;
223         int cmp;
225         cmp = memcmp(name1, name2, len);
226         if (cmp < 0)
227                 return 0;
228         if (cmp > 0)
229                 return TREE_UNORDERED;
231         /*
232          * Ok, the first <len> characters are the same.
233          * Now we need to order the next one, but turn
234          * a '\0' into a '/' for a directory entry.
235          */
236         c1 = name1[len];
237         c2 = name2[len];
238         if (!c1 && !c2)
239                 /*
240                  * git-write-tree used to write out a nonsense tree that has
241                  * entries with the same name, one blob and one tree.  Make
242                  * sure we do not have duplicate entries.
243                  */
244                 return TREE_HAS_DUPS;
245         if (!c1 && S_ISDIR(mode1))
246                 c1 = '/';
247         if (!c2 && S_ISDIR(mode2))
248                 c2 = '/';
249         return c1 < c2 ? 0 : TREE_UNORDERED;
252 static int fsck_tree(struct tree *item)
254         int retval;
255         int has_full_path = 0;
256         int has_empty_name = 0;
257         int has_zero_pad = 0;
258         int has_bad_modes = 0;
259         int has_dup_entries = 0;
260         int not_properly_sorted = 0;
261         struct tree_desc desc;
262         unsigned o_mode;
263         const char *o_name;
264         const unsigned char *o_sha1;
266         if (verbose)
267                 fprintf(stderr, "Checking tree %s\n",
268                                 sha1_to_hex(item->object.sha1));
270         init_tree_desc(&desc, item->buffer, item->size);
272         o_mode = 0;
273         o_name = NULL;
274         o_sha1 = NULL;
275         while (desc.size) {
276                 unsigned mode;
277                 const char *name;
278                 const unsigned char *sha1;
280                 sha1 = tree_entry_extract(&desc, &name, &mode);
282                 if (strchr(name, '/'))
283                         has_full_path = 1;
284                 if (!*name)
285                         has_empty_name = 1;
286                 has_zero_pad |= *(char *)desc.buffer == '0';
287                 update_tree_entry(&desc);
289                 switch (mode) {
290                 /*
291                  * Standard modes..
292                  */
293                 case S_IFREG | 0755:
294                 case S_IFREG | 0644:
295                 case S_IFLNK:
296                 case S_IFDIR:
297                 case S_IFGITLINK:
298                         break;
299                 /*
300                  * This is nonstandard, but we had a few of these
301                  * early on when we honored the full set of mode
302                  * bits..
303                  */
304                 case S_IFREG | 0664:
305                         if (!check_strict)
306                                 break;
307                 default:
308                         has_bad_modes = 1;
309                 }
311                 if (o_name) {
312                         switch (verify_ordered(o_mode, o_name, mode, name)) {
313                         case TREE_UNORDERED:
314                                 not_properly_sorted = 1;
315                                 break;
316                         case TREE_HAS_DUPS:
317                                 has_dup_entries = 1;
318                                 break;
319                         default:
320                                 break;
321                         }
322                 }
324                 o_mode = mode;
325                 o_name = name;
326                 o_sha1 = sha1;
327         }
328         free(item->buffer);
329         item->buffer = NULL;
331         retval = 0;
332         if (has_full_path) {
333                 objwarning(&item->object, "contains full pathnames");
334         }
335         if (has_empty_name) {
336                 objwarning(&item->object, "contains empty pathname");
337         }
338         if (has_zero_pad) {
339                 objwarning(&item->object, "contains zero-padded file modes");
340         }
341         if (has_bad_modes) {
342                 objwarning(&item->object, "contains bad file modes");
343         }
344         if (has_dup_entries) {
345                 retval = objerror(&item->object, "contains duplicate file entries");
346         }
347         if (not_properly_sorted) {
348                 retval = objerror(&item->object, "not properly sorted");
349         }
350         return retval;
353 static int fsck_commit(struct commit *commit)
355         char *buffer = commit->buffer;
356         unsigned char tree_sha1[20], sha1[20];
358         if (verbose)
359                 fprintf(stderr, "Checking commit %s\n",
360                         sha1_to_hex(commit->object.sha1));
362         if (memcmp(buffer, "tree ", 5))
363                 return objerror(&commit->object, "invalid format - expected 'tree' line");
364         if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
365                 return objerror(&commit->object, "invalid 'tree' line format - bad sha1");
366         buffer += 46;
367         while (!memcmp(buffer, "parent ", 7)) {
368                 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
369                         return objerror(&commit->object, "invalid 'parent' line format - bad sha1");
370                 buffer += 48;
371         }
372         if (memcmp(buffer, "author ", 7))
373                 return objerror(&commit->object, "invalid format - expected 'author' line");
374         free(commit->buffer);
375         commit->buffer = NULL;
376         if (!commit->tree)
377                 return objerror(&commit->object, "could not load commit's tree %s", tree_sha1);
378         if (!commit->parents && show_root)
379                 printf("root %s\n", sha1_to_hex(commit->object.sha1));
380         if (!commit->date)
381                 printf("bad commit date in %s\n",
382                        sha1_to_hex(commit->object.sha1));
383         return 0;
386 static int fsck_tag(struct tag *tag)
388         struct object *tagged = tag->tagged;
390         if (verbose)
391                 fprintf(stderr, "Checking tag %s\n",
392                         sha1_to_hex(tag->object.sha1));
394         if (!tagged) {
395                 return objerror(&tag->object, "could not load tagged object");
396         }
397         if (!show_tags)
398                 return 0;
400         printf("tagged %s %s", typename(tagged->type), sha1_to_hex(tagged->sha1));
401         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
402         return 0;
405 static int fsck_sha1(const unsigned char *sha1)
407         struct object *obj = parse_object(sha1);
408         if (!obj) {
409                 errors_found |= ERROR_OBJECT;
410                 return error("%s: object corrupt or missing",
411                              sha1_to_hex(sha1));
412         }
413         if (obj->flags & SEEN)
414                 return 0;
415         obj->flags |= SEEN;
416         if (obj->type == OBJ_BLOB)
417                 return 0;
418         if (obj->type == OBJ_TREE)
419                 return fsck_tree((struct tree *) obj);
420         if (obj->type == OBJ_COMMIT)
421                 return fsck_commit((struct commit *) obj);
422         if (obj->type == OBJ_TAG)
423                 return fsck_tag((struct tag *) obj);
425         /* By now, parse_object() would've returned NULL instead. */
426         return objerror(obj, "unknown type '%d' (internal fsck error)",
427                         obj->type);
431  * This is the sorting chunk size: make it reasonably
432  * big so that we can sort well..
433  */
434 #define MAX_SHA1_ENTRIES (1024)
436 struct sha1_entry {
437         unsigned long ino;
438         unsigned char sha1[20];
441 static struct {
442         unsigned long nr;
443         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
444 } sha1_list;
446 static int ino_compare(const void *_a, const void *_b)
448         const struct sha1_entry *a = _a, *b = _b;
449         unsigned long ino1 = a->ino, ino2 = b->ino;
450         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
453 static void fsck_sha1_list(void)
455         int i, nr = sha1_list.nr;
457         if (SORT_DIRENT)
458                 qsort(sha1_list.entry, nr,
459                       sizeof(struct sha1_entry *), ino_compare);
460         for (i = 0; i < nr; i++) {
461                 struct sha1_entry *entry = sha1_list.entry[i];
462                 unsigned char *sha1 = entry->sha1;
464                 sha1_list.entry[i] = NULL;
465                 fsck_sha1(sha1);
466                 free(entry);
467         }
468         sha1_list.nr = 0;
471 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
473         struct sha1_entry *entry = xmalloc(sizeof(*entry));
474         int nr;
476         entry->ino = ino;
477         hashcpy(entry->sha1, sha1);
478         nr = sha1_list.nr;
479         if (nr == MAX_SHA1_ENTRIES) {
480                 fsck_sha1_list();
481                 nr = 0;
482         }
483         sha1_list.entry[nr] = entry;
484         sha1_list.nr = ++nr;
487 static void fsck_dir(int i, char *path)
489         DIR *dir = opendir(path);
490         struct dirent *de;
492         if (!dir)
493                 return;
495         if (verbose)
496                 fprintf(stderr, "Checking directory %s\n", path);
498         while ((de = readdir(dir)) != NULL) {
499                 char name[100];
500                 unsigned char sha1[20];
501                 int len = strlen(de->d_name);
503                 switch (len) {
504                 case 2:
505                         if (de->d_name[1] != '.')
506                                 break;
507                 case 1:
508                         if (de->d_name[0] != '.')
509                                 break;
510                         continue;
511                 case 38:
512                         sprintf(name, "%02x", i);
513                         memcpy(name+2, de->d_name, len+1);
514                         if (get_sha1_hex(name, sha1) < 0)
515                                 break;
516                         add_sha1_list(sha1, DIRENT_SORT_HINT(de));
517                         continue;
518                 }
519                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
520         }
521         closedir(dir);
524 static int default_refs;
526 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
527                 const char *email, unsigned long timestamp, int tz,
528                 const char *message, void *cb_data)
530         struct object *obj;
532         if (verbose)
533                 fprintf(stderr, "Checking reflog %s->%s\n",
534                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
536         if (!is_null_sha1(osha1)) {
537                 obj = lookup_object(osha1);
538                 if (obj) {
539                         obj->used = 1;
540                         mark_reachable(obj, REACHABLE);
541                 }
542         }
543         obj = lookup_object(nsha1);
544         if (obj) {
545                 obj->used = 1;
546                 mark_reachable(obj, REACHABLE);
547         }
548         return 0;
551 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
553         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
554         return 0;
557 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
559         struct object *obj;
561         obj = lookup_object(sha1);
562         if (!obj) {
563                 if (has_sha1_file(sha1)) {
564                         default_refs++;
565                         return 0; /* it is in a pack */
566                 }
567                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
568                 /* We'll continue with the rest despite the error.. */
569                 return 0;
570         }
571         default_refs++;
572         obj->used = 1;
573         mark_reachable(obj, REACHABLE);
575         return 0;
578 static void get_default_heads(void)
580         for_each_ref(fsck_handle_ref, NULL);
581         if (include_reflogs)
582                 for_each_reflog(fsck_handle_reflog, NULL);
584         /*
585          * Not having any default heads isn't really fatal, but
586          * it does mean that "--unreachable" no longer makes any
587          * sense (since in this case everything will obviously
588          * be unreachable by definition.
589          *
590          * Showing dangling objects is valid, though (as those
591          * dangling objects are likely lost heads).
592          *
593          * So we just print a warning about it, and clear the
594          * "show_unreachable" flag.
595          */
596         if (!default_refs) {
597                 fprintf(stderr, "notice: No default references\n");
598                 show_unreachable = 0;
599         }
602 static void fsck_object_dir(const char *path)
604         int i;
606         if (verbose)
607                 fprintf(stderr, "Checking object directory\n");
609         for (i = 0; i < 256; i++) {
610                 static char dir[4096];
611                 sprintf(dir, "%s/%02x", path, i);
612                 fsck_dir(i, dir);
613         }
614         fsck_sha1_list();
617 static int fsck_head_link(void)
619         unsigned char sha1[20];
620         int flag;
621         int null_is_error = 0;
622         const char *head_points_at = resolve_ref("HEAD", sha1, 0, &flag);
624         if (verbose)
625                 fprintf(stderr, "Checking HEAD link\n");
627         if (!head_points_at)
628                 return error("Invalid HEAD");
629         if (!strcmp(head_points_at, "HEAD"))
630                 /* detached HEAD */
631                 null_is_error = 1;
632         else if (prefixcmp(head_points_at, "refs/heads/"))
633                 return error("HEAD points to something strange (%s)",
634                              head_points_at);
635         if (is_null_sha1(sha1)) {
636                 if (null_is_error)
637                         return error("HEAD: detached HEAD points at nothing");
638                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
639                         head_points_at + 11);
640         }
641         return 0;
644 static int fsck_cache_tree(struct cache_tree *it)
646         int i;
647         int err = 0;
649         if (verbose)
650                 fprintf(stderr, "Checking cache tree\n");
652         if (0 <= it->entry_count) {
653                 struct object *obj = parse_object(it->sha1);
654                 if (!obj) {
655                         error("%s: invalid sha1 pointer in cache-tree",
656                               sha1_to_hex(it->sha1));
657                         return 1;
658                 }
659                 mark_reachable(obj, REACHABLE);
660                 obj->used = 1;
661                 if (obj->type != OBJ_TREE)
662                         err |= objerror(obj, "non-tree in cache-tree");
663         }
664         for (i = 0; i < it->subtree_nr; i++)
665                 err |= fsck_cache_tree(it->down[i]->cache_tree);
666         return err;
669 static const char fsck_usage[] =
670 "git-fsck [--tags] [--root] [[--unreachable] [--cache] [--full] "
671 "[--strict] [--verbose] <head-sha1>*]";
673 int cmd_fsck(int argc, const char **argv, const char *prefix)
675         int i, heads;
677         track_object_refs = 1;
678         errors_found = 0;
680         for (i = 1; i < argc; i++) {
681                 const char *arg = argv[i];
683                 if (!strcmp(arg, "--unreachable")) {
684                         show_unreachable = 1;
685                         continue;
686                 }
687                 if (!strcmp(arg, "--tags")) {
688                         show_tags = 1;
689                         continue;
690                 }
691                 if (!strcmp(arg, "--root")) {
692                         show_root = 1;
693                         continue;
694                 }
695                 if (!strcmp(arg, "--cache")) {
696                         keep_cache_objects = 1;
697                         continue;
698                 }
699                 if (!strcmp(arg, "--no-reflogs")) {
700                         include_reflogs = 0;
701                         continue;
702                 }
703                 if (!strcmp(arg, "--full")) {
704                         check_full = 1;
705                         continue;
706                 }
707                 if (!strcmp(arg, "--strict")) {
708                         check_strict = 1;
709                         continue;
710                 }
711                 if (!strcmp(arg, "--verbose")) {
712                         verbose = 1;
713                         continue;
714                 }
715                 if (!strcmp(arg, "--lost-found")) {
716                         check_full = 1;
717                         include_reflogs = 0;
718                         write_lost_and_found = 1;
719                         continue;
720                 }
721                 if (*arg == '-')
722                         usage(fsck_usage);
723         }
725         fsck_head_link();
726         fsck_object_dir(get_object_directory());
727         if (check_full) {
728                 struct alternate_object_database *alt;
729                 struct packed_git *p;
730                 prepare_alt_odb();
731                 for (alt = alt_odb_list; alt; alt = alt->next) {
732                         char namebuf[PATH_MAX];
733                         int namelen = alt->name - alt->base;
734                         memcpy(namebuf, alt->base, namelen);
735                         namebuf[namelen - 1] = 0;
736                         fsck_object_dir(namebuf);
737                 }
738                 prepare_packed_git();
739                 for (p = packed_git; p; p = p->next)
740                         /* verify gives error messages itself */
741                         verify_pack(p, 0);
743                 for (p = packed_git; p; p = p->next) {
744                         uint32_t i, num;
745                         if (open_pack_index(p))
746                                 continue;
747                         num = p->num_objects;
748                         for (i = 0; i < num; i++)
749                                 fsck_sha1(nth_packed_object_sha1(p, i));
750                 }
751         }
753         heads = 0;
754         for (i = 1; i < argc; i++) {
755                 const char *arg = argv[i];
757                 if (*arg == '-')
758                         continue;
760                 if (!get_sha1(arg, head_sha1)) {
761                         struct object *obj = lookup_object(head_sha1);
763                         /* Error is printed by lookup_object(). */
764                         if (!obj)
765                                 continue;
767                         obj->used = 1;
768                         mark_reachable(obj, REACHABLE);
769                         heads++;
770                         continue;
771                 }
772                 error("invalid parameter: expected sha1, got '%s'", arg);
773         }
775         /*
776          * If we've not been given any explicit head information, do the
777          * default ones from .git/refs. We also consider the index file
778          * in this case (ie this implies --cache).
779          */
780         if (!heads) {
781                 get_default_heads();
782                 keep_cache_objects = 1;
783         }
785         if (keep_cache_objects) {
786                 int i;
787                 read_cache();
788                 for (i = 0; i < active_nr; i++) {
789                         unsigned int mode;
790                         struct blob *blob;
791                         struct object *obj;
793                         mode = ntohl(active_cache[i]->ce_mode);
794                         if (S_ISGITLINK(mode))
795                                 continue;
796                         blob = lookup_blob(active_cache[i]->sha1);
797                         if (!blob)
798                                 continue;
799                         obj = &blob->object;
800                         obj->used = 1;
801                         mark_reachable(obj, REACHABLE);
802                 }
803                 if (active_cache_tree)
804                         fsck_cache_tree(active_cache_tree);
805         }
807         check_connectivity();
808         return errors_found;