Merge branch 'js/cherry-pick-usability'
[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
15 #define REACHABLE 0x0001
16 #define SEEN      0x0002
17
18 static int show_root;
19 static int show_tags;
20 static int show_unreachable;
21 static int include_reflogs = 1;
22 static int check_full = 1;
23 static int check_strict;
24 static int keep_cache_objects;
25 static unsigned char head_sha1[20];
26 static const char *head_points_at;
27 static int errors_found;
28 static int write_lost_and_found;
29 static int verbose;
30 #define ERROR_OBJECT 01
31 #define ERROR_REACHABLE 02
32
33 #ifdef NO_D_INO_IN_DIRENT
34 #define SORT_DIRENT 0
35 #define DIRENT_SORT_HINT(de) 0
36 #else
37 #define SORT_DIRENT 1
38 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
39 #endif
40
41 static void objreport(struct object *obj, const char *severity,
42                       const char *err, va_list params)
43 {
44         fprintf(stderr, "%s in %s %s: ",
45                 severity, typename(obj->type), sha1_to_hex(obj->sha1));
46         vfprintf(stderr, err, params);
47         fputs("\n", stderr);
48 }
49
50 __attribute__((format (printf, 2, 3)))
51 static int objerror(struct object *obj, const char *err, ...)
52 {
53         va_list params;
54         va_start(params, err);
55         errors_found |= ERROR_OBJECT;
56         objreport(obj, "error", err, params);
57         va_end(params);
58         return -1;
59 }
60
61 __attribute__((format (printf, 3, 4)))
62 static int fsck_error_func(struct object *obj, int type, const char *err, ...)
63 {
64         va_list params;
65         va_start(params, err);
66         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
67         va_end(params);
68         return (type == FSCK_WARN) ? 0 : 1;
69 }
70
71 static struct object_array pending;
72
73 static int mark_object(struct object *obj, int type, void *data)
74 {
75         struct object *parent = data;
76
77         /*
78          * The only case data is NULL or type is OBJ_ANY is when
79          * mark_object_reachable() calls us.  All the callers of
80          * that function has non-NULL obj hence ...
81          */
82         if (!obj) {
83                 /* ... these references to parent->fld are safe here */
84                 printf("broken link from %7s %s\n",
85                            typename(parent->type), sha1_to_hex(parent->sha1));
86                 printf("broken link from %7s %s\n",
87                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
88                 errors_found |= ERROR_REACHABLE;
89                 return 1;
90         }
91
92         if (type != OBJ_ANY && obj->type != type)
93                 /* ... and the reference to parent is safe here */
94                 objerror(parent, "wrong object type in link");
95
96         if (obj->flags & REACHABLE)
97                 return 0;
98         obj->flags |= REACHABLE;
99         if (!obj->parsed) {
100                 if (parent && !has_sha1_file(obj->sha1)) {
101                         printf("broken link from %7s %s\n",
102                                  typename(parent->type), sha1_to_hex(parent->sha1));
103                         printf("              to %7s %s\n",
104                                  typename(obj->type), sha1_to_hex(obj->sha1));
105                         errors_found |= ERROR_REACHABLE;
106                 }
107                 return 1;
108         }
109
110         add_object_array(obj, (void *) parent, &pending);
111         return 0;
112 }
113
114 static void mark_object_reachable(struct object *obj)
115 {
116         mark_object(obj, OBJ_ANY, NULL);
117 }
118
119 static int traverse_one_object(struct object *obj)
120 {
121         int result;
122         struct tree *tree = NULL;
123
124         if (obj->type == OBJ_TREE) {
125                 obj->parsed = 0;
126                 tree = (struct tree *)obj;
127                 if (parse_tree(tree) < 0)
128                         return 1; /* error already displayed */
129         }
130         result = fsck_walk(obj, mark_object, obj);
131         if (tree) {
132                 free(tree->buffer);
133                 tree->buffer = NULL;
134         }
135         return result;
136 }
137
138 static int traverse_reachable(void)
139 {
140         int result = 0;
141         while (pending.nr) {
142                 struct object_array_entry *entry;
143                 struct object *obj, *parent;
144
145                 entry = pending.objects + --pending.nr;
146                 obj = entry->item;
147                 parent = (struct object *) entry->name;
148                 result |= traverse_one_object(obj);
149         }
150         return !!result;
151 }
152
153 static int mark_used(struct object *obj, int type, void *data)
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->parsed) {
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                 printf("dangling %s %s\n", typename(obj->type),
217                        sha1_to_hex(obj->sha1));
218                 if (write_lost_and_found) {
219                         char *filename = git_path("lost-found/%s/%s",
220                                 obj->type == OBJ_COMMIT ? "commit" : "other",
221                                 sha1_to_hex(obj->sha1));
222                         FILE *f;
223
224                         if (safe_create_leading_directories(filename)) {
225                                 error("Could not create lost-found");
226                                 return;
227                         }
228                         if (!(f = fopen(filename, "w")))
229                                 die_errno("Could not open '%s'", filename);
230                         if (obj->type == OBJ_BLOB) {
231                                 enum object_type type;
232                                 unsigned long size;
233                                 char *buf = read_sha1_file(obj->sha1,
234                                                 &type, &size);
235                                 if (buf) {
236                                         if (fwrite(buf, size, 1, f) != 1)
237                                                 die_errno("Could not write '%s'",
238                                                           filename);
239                                         free(buf);
240                                 }
241                         } else
242                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
243                         if (fclose(f))
244                                 die_errno("Could not finish '%s'",
245                                           filename);
246                 }
247                 return;
248         }
249
250         /*
251          * Otherwise? It's there, it's unreachable, and some other unreachable
252          * object points to it. Ignore it - it's not interesting, and we showed
253          * all the interesting cases above.
254          */
255 }
256
257 static void check_object(struct object *obj)
258 {
259         if (verbose)
260                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
261
262         if (obj->flags & REACHABLE)
263                 check_reachable_object(obj);
264         else
265                 check_unreachable_object(obj);
266 }
267
268 static void check_connectivity(void)
269 {
270         int i, max;
271
272         /* Traverse the pending reachable objects */
273         traverse_reachable();
274
275         /* Look up all the requirements, warn about missing objects.. */
276         max = get_max_object_index();
277         if (verbose)
278                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
279
280         for (i = 0; i < max; i++) {
281                 struct object *obj = get_indexed_object(i);
282
283                 if (obj)
284                         check_object(obj);
285         }
286 }
287
288 static int fsck_sha1(const unsigned char *sha1)
289 {
290         struct object *obj = parse_object(sha1);
291         if (!obj) {
292                 errors_found |= ERROR_OBJECT;
293                 return error("%s: object corrupt or missing",
294                              sha1_to_hex(sha1));
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, mark_used, NULL))
305                 objerror(obj, "broken links");
306         if (fsck_object(obj, check_strict, fsck_error_func))
307                 return -1;
308
309         if (obj->type == OBJ_TREE) {
310                 struct tree *item = (struct tree *) obj;
311
312                 free(item->buffer);
313                 item->buffer = NULL;
314         }
315
316         if (obj->type == OBJ_COMMIT) {
317                 struct commit *commit = (struct commit *) obj;
318
319                 free(commit->buffer);
320                 commit->buffer = NULL;
321
322                 if (!commit->parents && show_root)
323                         printf("root %s\n", sha1_to_hex(commit->object.sha1));
324         }
325
326         if (obj->type == OBJ_TAG) {
327                 struct tag *tag = (struct tag *) obj;
328
329                 if (show_tags && tag->tagged) {
330                         printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
331                         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
332                 }
333         }
334
335         return 0;
336 }
337
338 /*
339  * This is the sorting chunk size: make it reasonably
340  * big so that we can sort well..
341  */
342 #define MAX_SHA1_ENTRIES (1024)
343
344 struct sha1_entry {
345         unsigned long ino;
346         unsigned char sha1[20];
347 };
348
349 static struct {
350         unsigned long nr;
351         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
352 } sha1_list;
353
354 static int ino_compare(const void *_a, const void *_b)
355 {
356         const struct sha1_entry *a = _a, *b = _b;
357         unsigned long ino1 = a->ino, ino2 = b->ino;
358         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
359 }
360
361 static void fsck_sha1_list(void)
362 {
363         int i, nr = sha1_list.nr;
364
365         if (SORT_DIRENT)
366                 qsort(sha1_list.entry, nr,
367                       sizeof(struct sha1_entry *), ino_compare);
368         for (i = 0; i < nr; i++) {
369                 struct sha1_entry *entry = sha1_list.entry[i];
370                 unsigned char *sha1 = entry->sha1;
371
372                 sha1_list.entry[i] = NULL;
373                 fsck_sha1(sha1);
374                 free(entry);
375         }
376         sha1_list.nr = 0;
377 }
378
379 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
380 {
381         struct sha1_entry *entry = xmalloc(sizeof(*entry));
382         int nr;
383
384         entry->ino = ino;
385         hashcpy(entry->sha1, sha1);
386         nr = sha1_list.nr;
387         if (nr == MAX_SHA1_ENTRIES) {
388                 fsck_sha1_list();
389                 nr = 0;
390         }
391         sha1_list.entry[nr] = entry;
392         sha1_list.nr = ++nr;
393 }
394
395 static inline int is_loose_object_file(struct dirent *de,
396                                        char *name, unsigned char *sha1)
397 {
398         if (strlen(de->d_name) != 38)
399                 return 0;
400         memcpy(name + 2, de->d_name, 39);
401         return !get_sha1_hex(name, sha1);
402 }
403
404 static void fsck_dir(int i, char *path)
405 {
406         DIR *dir = opendir(path);
407         struct dirent *de;
408         char name[100];
409
410         if (!dir)
411                 return;
412
413         if (verbose)
414                 fprintf(stderr, "Checking directory %s\n", path);
415
416         sprintf(name, "%02x", i);
417         while ((de = readdir(dir)) != NULL) {
418                 unsigned char sha1[20];
419
420                 if (is_dot_or_dotdot(de->d_name))
421                         continue;
422                 if (is_loose_object_file(de, name, sha1)) {
423                         add_sha1_list(sha1, DIRENT_SORT_HINT(de));
424                         continue;
425                 }
426                 if (!prefixcmp(de->d_name, "tmp_obj_"))
427                         continue;
428                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
429         }
430         closedir(dir);
431 }
432
433 static int default_refs;
434
435 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
436                 const char *email, unsigned long timestamp, int tz,
437                 const char *message, void *cb_data)
438 {
439         struct object *obj;
440
441         if (verbose)
442                 fprintf(stderr, "Checking reflog %s->%s\n",
443                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
444
445         if (!is_null_sha1(osha1)) {
446                 obj = lookup_object(osha1);
447                 if (obj) {
448                         obj->used = 1;
449                         mark_object_reachable(obj);
450                 }
451         }
452         obj = lookup_object(nsha1);
453         if (obj) {
454                 obj->used = 1;
455                 mark_object_reachable(obj);
456         }
457         return 0;
458 }
459
460 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
461 {
462         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
463         return 0;
464 }
465
466 static int is_branch(const char *refname)
467 {
468         return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
469 }
470
471 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
472 {
473         struct object *obj;
474
475         obj = parse_object(sha1);
476         if (!obj) {
477                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
478                 /* We'll continue with the rest despite the error.. */
479                 return 0;
480         }
481         if (obj->type != OBJ_COMMIT && is_branch(refname))
482                 error("%s: not a commit", refname);
483         default_refs++;
484         obj->used = 1;
485         mark_object_reachable(obj);
486
487         return 0;
488 }
489
490 static void get_default_heads(void)
491 {
492         if (head_points_at && !is_null_sha1(head_sha1))
493                 fsck_handle_ref("HEAD", head_sha1, 0, NULL);
494         for_each_ref(fsck_handle_ref, NULL);
495         if (include_reflogs)
496                 for_each_reflog(fsck_handle_reflog, NULL);
497
498         /*
499          * Not having any default heads isn't really fatal, but
500          * it does mean that "--unreachable" no longer makes any
501          * sense (since in this case everything will obviously
502          * be unreachable by definition.
503          *
504          * Showing dangling objects is valid, though (as those
505          * dangling objects are likely lost heads).
506          *
507          * So we just print a warning about it, and clear the
508          * "show_unreachable" flag.
509          */
510         if (!default_refs) {
511                 fprintf(stderr, "notice: No default references\n");
512                 show_unreachable = 0;
513         }
514 }
515
516 static void fsck_object_dir(const char *path)
517 {
518         int i;
519
520         if (verbose)
521                 fprintf(stderr, "Checking object directory\n");
522
523         for (i = 0; i < 256; i++) {
524                 static char dir[4096];
525                 sprintf(dir, "%s/%02x", path, i);
526                 fsck_dir(i, dir);
527         }
528         fsck_sha1_list();
529 }
530
531 static int fsck_head_link(void)
532 {
533         int flag;
534         int null_is_error = 0;
535
536         if (verbose)
537                 fprintf(stderr, "Checking HEAD link\n");
538
539         head_points_at = resolve_ref("HEAD", head_sha1, 0, &flag);
540         if (!head_points_at)
541                 return error("Invalid HEAD");
542         if (!strcmp(head_points_at, "HEAD"))
543                 /* detached HEAD */
544                 null_is_error = 1;
545         else if (prefixcmp(head_points_at, "refs/heads/"))
546                 return error("HEAD points to something strange (%s)",
547                              head_points_at);
548         if (is_null_sha1(head_sha1)) {
549                 if (null_is_error)
550                         return error("HEAD: detached HEAD points at nothing");
551                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
552                         head_points_at + 11);
553         }
554         return 0;
555 }
556
557 static int fsck_cache_tree(struct cache_tree *it)
558 {
559         int i;
560         int err = 0;
561
562         if (verbose)
563                 fprintf(stderr, "Checking cache tree\n");
564
565         if (0 <= it->entry_count) {
566                 struct object *obj = parse_object(it->sha1);
567                 if (!obj) {
568                         error("%s: invalid sha1 pointer in cache-tree",
569                               sha1_to_hex(it->sha1));
570                         return 1;
571                 }
572                 obj->used = 1;
573                 mark_object_reachable(obj);
574                 if (obj->type != OBJ_TREE)
575                         err |= objerror(obj, "non-tree in cache-tree");
576         }
577         for (i = 0; i < it->subtree_nr; i++)
578                 err |= fsck_cache_tree(it->down[i]->cache_tree);
579         return err;
580 }
581
582 static char const * const fsck_usage[] = {
583         "git fsck [options] [<object>...]",
584         NULL
585 };
586
587 static struct option fsck_opts[] = {
588         OPT__VERBOSE(&verbose, "be verbose"),
589         OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
590         OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
591         OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
592         OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
593         OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
594         OPT_BOOLEAN(0, "full", &check_full, "also consider packs and alternate objects"),
595         OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
596         OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
597                                 "write dangling objects in .git/lost-found"),
598         OPT_END(),
599 };
600
601 int cmd_fsck(int argc, const char **argv, const char *prefix)
602 {
603         int i, heads;
604         struct alternate_object_database *alt;
605
606         errors_found = 0;
607         read_replace_refs = 0;
608
609         argc = parse_options(argc, argv, prefix, fsck_opts, fsck_usage, 0);
610         if (write_lost_and_found) {
611                 check_full = 1;
612                 include_reflogs = 0;
613         }
614
615         fsck_head_link();
616         fsck_object_dir(get_object_directory());
617
618         prepare_alt_odb();
619         for (alt = alt_odb_list; alt; alt = alt->next) {
620                 char namebuf[PATH_MAX];
621                 int namelen = alt->name - alt->base;
622                 memcpy(namebuf, alt->base, namelen);
623                 namebuf[namelen - 1] = 0;
624                 fsck_object_dir(namebuf);
625         }
626
627         if (check_full) {
628                 struct packed_git *p;
629
630                 prepare_packed_git();
631                 for (p = packed_git; p; p = p->next)
632                         /* verify gives error messages itself */
633                         verify_pack(p);
634
635                 for (p = packed_git; p; p = p->next) {
636                         uint32_t j, num;
637                         if (open_pack_index(p))
638                                 continue;
639                         num = p->num_objects;
640                         for (j = 0; j < num; j++)
641                                 fsck_sha1(nth_packed_object_sha1(p, j));
642                 }
643         }
644
645         heads = 0;
646         for (i = 0; i < argc; i++) {
647                 const char *arg = argv[i];
648                 unsigned char sha1[20];
649                 if (!get_sha1(arg, sha1)) {
650                         struct object *obj = lookup_object(sha1);
651
652                         /* Error is printed by lookup_object(). */
653                         if (!obj)
654                                 continue;
655
656                         obj->used = 1;
657                         mark_object_reachable(obj);
658                         heads++;
659                         continue;
660                 }
661                 error("invalid parameter: expected sha1, got '%s'", arg);
662         }
663
664         /*
665          * If we've not been given any explicit head information, do the
666          * default ones from .git/refs. We also consider the index file
667          * in this case (ie this implies --cache).
668          */
669         if (!heads) {
670                 get_default_heads();
671                 keep_cache_objects = 1;
672         }
673
674         if (keep_cache_objects) {
675                 read_cache();
676                 for (i = 0; i < active_nr; i++) {
677                         unsigned int mode;
678                         struct blob *blob;
679                         struct object *obj;
680
681                         mode = active_cache[i]->ce_mode;
682                         if (S_ISGITLINK(mode))
683                                 continue;
684                         blob = lookup_blob(active_cache[i]->sha1);
685                         if (!blob)
686                                 continue;
687                         obj = &blob->object;
688                         obj->used = 1;
689                         mark_object_reachable(obj);
690                 }
691                 if (active_cache_tree)
692                         fsck_cache_tree(active_cache_tree);
693         }
694
695         check_connectivity();
696         return errors_found;
697 }