Update draft release notes for 1.6.0
[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
14 #define REACHABLE 0x0001
15 #define SEEN      0x0002
16
17 static int show_root;
18 static int show_tags;
19 static int show_unreachable;
20 static int include_reflogs = 1;
21 static int check_full;
22 static int check_strict;
23 static int keep_cache_objects;
24 static unsigned char head_sha1[20];
25 static int errors_found;
26 static int write_lost_and_found;
27 static int verbose;
28 #define ERROR_OBJECT 01
29 #define ERROR_REACHABLE 02
30
31 #ifdef NO_D_INO_IN_DIRENT
32 #define SORT_DIRENT 0
33 #define DIRENT_SORT_HINT(de) 0
34 #else
35 #define SORT_DIRENT 1
36 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
37 #endif
38
39 static void objreport(struct object *obj, const char *severity,
40                       const char *err, va_list params)
41 {
42         fprintf(stderr, "%s in %s %s: ",
43                 severity, typename(obj->type), sha1_to_hex(obj->sha1));
44         vfprintf(stderr, err, params);
45         fputs("\n", stderr);
46 }
47
48 static int objerror(struct object *obj, const char *err, ...)
49 {
50         va_list params;
51         va_start(params, err);
52         errors_found |= ERROR_OBJECT;
53         objreport(obj, "error", err, params);
54         va_end(params);
55         return -1;
56 }
57
58 static int fsck_error_func(struct object *obj, int type, const char *err, ...)
59 {
60         va_list params;
61         va_start(params, err);
62         objreport(obj, (type == FSCK_WARN) ? "warning" : "error", err, params);
63         va_end(params);
64         return (type == FSCK_WARN) ? 0 : 1;
65 }
66
67 static int mark_object(struct object *obj, int type, void *data)
68 {
69         struct tree *tree = NULL;
70         struct object *parent = data;
71         int result;
72
73         if (!obj) {
74                 printf("broken link from %7s %s\n",
75                            typename(parent->type), sha1_to_hex(parent->sha1));
76                 printf("broken link from %7s %s\n",
77                            (type == OBJ_ANY ? "unknown" : typename(type)), "unknown");
78                 errors_found |= ERROR_REACHABLE;
79                 return 1;
80         }
81
82         if (type != OBJ_ANY && obj->type != type)
83                 objerror(parent, "wrong object type in link");
84
85         if (obj->flags & REACHABLE)
86                 return 0;
87         obj->flags |= REACHABLE;
88         if (!obj->parsed) {
89                 if (parent && !has_sha1_file(obj->sha1)) {
90                         printf("broken link from %7s %s\n",
91                                  typename(parent->type), sha1_to_hex(parent->sha1));
92                         printf("              to %7s %s\n",
93                                  typename(obj->type), sha1_to_hex(obj->sha1));
94                         errors_found |= ERROR_REACHABLE;
95                 }
96                 return 1;
97         }
98
99         if (obj->type == OBJ_TREE) {
100                 obj->parsed = 0;
101                 tree = (struct tree *)obj;
102                 if (parse_tree(tree) < 0)
103                         return 1; /* error already displayed */
104         }
105         result = fsck_walk(obj, mark_object, obj);
106         if (tree) {
107                 free(tree->buffer);
108                 tree->buffer = NULL;
109         }
110         if (result < 0)
111                 result = 1;
112
113         return result;
114 }
115
116 static void mark_object_reachable(struct object *obj)
117 {
118         mark_object(obj, OBJ_ANY, 0);
119 }
120
121 static int mark_used(struct object *obj, int type, void *data)
122 {
123         if (!obj)
124                 return 1;
125         obj->used = 1;
126         return 0;
127 }
128
129 /*
130  * Check a single reachable object
131  */
132 static void check_reachable_object(struct object *obj)
133 {
134         /*
135          * We obviously want the object to be parsed,
136          * except if it was in a pack-file and we didn't
137          * do a full fsck
138          */
139         if (!obj->parsed) {
140                 if (has_sha1_pack(obj->sha1, NULL))
141                         return; /* it is in pack - forget about it */
142                 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
143                 errors_found |= ERROR_REACHABLE;
144                 return;
145         }
146 }
147
148 /*
149  * Check a single unreachable object
150  */
151 static void check_unreachable_object(struct object *obj)
152 {
153         /*
154          * Missing unreachable object? Ignore it. It's not like
155          * we miss it (since it can't be reached), nor do we want
156          * to complain about it being unreachable (since it does
157          * not exist).
158          */
159         if (!obj->parsed)
160                 return;
161
162         /*
163          * Unreachable object that exists? Show it if asked to,
164          * since this is something that is prunable.
165          */
166         if (show_unreachable) {
167                 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
168                 return;
169         }
170
171         /*
172          * "!used" means that nothing at all points to it, including
173          * other unreachable objects. In other words, it's the "tip"
174          * of some set of unreachable objects, usually a commit that
175          * got dropped.
176          *
177          * Such starting points are more interesting than some random
178          * set of unreachable objects, so we show them even if the user
179          * hasn't asked for _all_ unreachable objects. If you have
180          * deleted a branch by mistake, this is a prime candidate to
181          * start looking at, for example.
182          */
183         if (!obj->used) {
184                 printf("dangling %s %s\n", typename(obj->type),
185                        sha1_to_hex(obj->sha1));
186                 if (write_lost_and_found) {
187                         char *filename = git_path("lost-found/%s/%s",
188                                 obj->type == OBJ_COMMIT ? "commit" : "other",
189                                 sha1_to_hex(obj->sha1));
190                         FILE *f;
191
192                         if (safe_create_leading_directories(filename)) {
193                                 error("Could not create lost-found");
194                                 return;
195                         }
196                         if (!(f = fopen(filename, "w")))
197                                 die("Could not open %s", filename);
198                         if (obj->type == OBJ_BLOB) {
199                                 enum object_type type;
200                                 unsigned long size;
201                                 char *buf = read_sha1_file(obj->sha1,
202                                                 &type, &size);
203                                 if (buf) {
204                                         fwrite(buf, size, 1, f);
205                                         free(buf);
206                                 }
207                         } else
208                                 fprintf(f, "%s\n", sha1_to_hex(obj->sha1));
209                         fclose(f);
210                 }
211                 return;
212         }
213
214         /*
215          * Otherwise? It's there, it's unreachable, and some other unreachable
216          * object points to it. Ignore it - it's not interesting, and we showed
217          * all the interesting cases above.
218          */
219 }
220
221 static void check_object(struct object *obj)
222 {
223         if (verbose)
224                 fprintf(stderr, "Checking %s\n", sha1_to_hex(obj->sha1));
225
226         if (obj->flags & REACHABLE)
227                 check_reachable_object(obj);
228         else
229                 check_unreachable_object(obj);
230 }
231
232 static void check_connectivity(void)
233 {
234         int i, max;
235
236         /* Look up all the requirements, warn about missing objects.. */
237         max = get_max_object_index();
238         if (verbose)
239                 fprintf(stderr, "Checking connectivity (%d objects)\n", max);
240
241         for (i = 0; i < max; i++) {
242                 struct object *obj = get_indexed_object(i);
243
244                 if (obj)
245                         check_object(obj);
246         }
247 }
248
249 static int fsck_sha1(const unsigned char *sha1)
250 {
251         struct object *obj = parse_object(sha1);
252         if (!obj) {
253                 errors_found |= ERROR_OBJECT;
254                 return error("%s: object corrupt or missing",
255                              sha1_to_hex(sha1));
256         }
257         if (obj->flags & SEEN)
258                 return 0;
259         obj->flags |= SEEN;
260
261         if (verbose)
262                 fprintf(stderr, "Checking %s %s\n",
263                         typename(obj->type), sha1_to_hex(obj->sha1));
264
265         if (fsck_walk(obj, mark_used, 0))
266                 objerror(obj, "broken links");
267         if (fsck_object(obj, check_strict, fsck_error_func))
268                 return -1;
269
270         if (obj->type == OBJ_TREE) {
271                 struct tree *item = (struct tree *) obj;
272
273                 free(item->buffer);
274                 item->buffer = NULL;
275         }
276
277         if (obj->type == OBJ_COMMIT) {
278                 struct commit *commit = (struct commit *) obj;
279
280                 free(commit->buffer);
281                 commit->buffer = NULL;
282
283                 if (!commit->parents && show_root)
284                         printf("root %s\n", sha1_to_hex(commit->object.sha1));
285         }
286
287         if (obj->type == OBJ_TAG) {
288                 struct tag *tag = (struct tag *) obj;
289
290                 if (show_tags && tag->tagged) {
291                         printf("tagged %s %s", typename(tag->tagged->type), sha1_to_hex(tag->tagged->sha1));
292                         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
293                 }
294         }
295
296         return 0;
297 }
298
299 /*
300  * This is the sorting chunk size: make it reasonably
301  * big so that we can sort well..
302  */
303 #define MAX_SHA1_ENTRIES (1024)
304
305 struct sha1_entry {
306         unsigned long ino;
307         unsigned char sha1[20];
308 };
309
310 static struct {
311         unsigned long nr;
312         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
313 } sha1_list;
314
315 static int ino_compare(const void *_a, const void *_b)
316 {
317         const struct sha1_entry *a = _a, *b = _b;
318         unsigned long ino1 = a->ino, ino2 = b->ino;
319         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
320 }
321
322 static void fsck_sha1_list(void)
323 {
324         int i, nr = sha1_list.nr;
325
326         if (SORT_DIRENT)
327                 qsort(sha1_list.entry, nr,
328                       sizeof(struct sha1_entry *), ino_compare);
329         for (i = 0; i < nr; i++) {
330                 struct sha1_entry *entry = sha1_list.entry[i];
331                 unsigned char *sha1 = entry->sha1;
332
333                 sha1_list.entry[i] = NULL;
334                 fsck_sha1(sha1);
335                 free(entry);
336         }
337         sha1_list.nr = 0;
338 }
339
340 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
341 {
342         struct sha1_entry *entry = xmalloc(sizeof(*entry));
343         int nr;
344
345         entry->ino = ino;
346         hashcpy(entry->sha1, sha1);
347         nr = sha1_list.nr;
348         if (nr == MAX_SHA1_ENTRIES) {
349                 fsck_sha1_list();
350                 nr = 0;
351         }
352         sha1_list.entry[nr] = entry;
353         sha1_list.nr = ++nr;
354 }
355
356 static void fsck_dir(int i, char *path)
357 {
358         DIR *dir = opendir(path);
359         struct dirent *de;
360
361         if (!dir)
362                 return;
363
364         if (verbose)
365                 fprintf(stderr, "Checking directory %s\n", path);
366
367         while ((de = readdir(dir)) != NULL) {
368                 char name[100];
369                 unsigned char sha1[20];
370                 int len = strlen(de->d_name);
371
372                 switch (len) {
373                 case 2:
374                         if (de->d_name[1] != '.')
375                                 break;
376                 case 1:
377                         if (de->d_name[0] != '.')
378                                 break;
379                         continue;
380                 case 38:
381                         sprintf(name, "%02x", i);
382                         memcpy(name+2, de->d_name, len+1);
383                         if (get_sha1_hex(name, sha1) < 0)
384                                 break;
385                         add_sha1_list(sha1, DIRENT_SORT_HINT(de));
386                         continue;
387                 }
388                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
389         }
390         closedir(dir);
391 }
392
393 static int default_refs;
394
395 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
396                 const char *email, unsigned long timestamp, int tz,
397                 const char *message, void *cb_data)
398 {
399         struct object *obj;
400
401         if (verbose)
402                 fprintf(stderr, "Checking reflog %s->%s\n",
403                         sha1_to_hex(osha1), sha1_to_hex(nsha1));
404
405         if (!is_null_sha1(osha1)) {
406                 obj = lookup_object(osha1);
407                 if (obj) {
408                         obj->used = 1;
409                         mark_object_reachable(obj);
410                 }
411         }
412         obj = lookup_object(nsha1);
413         if (obj) {
414                 obj->used = 1;
415                 mark_object_reachable(obj);
416         }
417         return 0;
418 }
419
420 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
421 {
422         for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
423         return 0;
424 }
425
426 static int is_branch(const char *refname)
427 {
428         return !strcmp(refname, "HEAD") || !prefixcmp(refname, "refs/heads/");
429 }
430
431 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
432 {
433         struct object *obj;
434
435         obj = parse_object(sha1);
436         if (!obj) {
437                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
438                 /* We'll continue with the rest despite the error.. */
439                 return 0;
440         }
441         if (obj->type != OBJ_COMMIT && is_branch(refname))
442                 error("%s: not a commit", refname);
443         default_refs++;
444         obj->used = 1;
445         mark_object_reachable(obj);
446
447         return 0;
448 }
449
450 static void get_default_heads(void)
451 {
452         for_each_ref(fsck_handle_ref, NULL);
453         if (include_reflogs)
454                 for_each_reflog(fsck_handle_reflog, NULL);
455
456         /*
457          * Not having any default heads isn't really fatal, but
458          * it does mean that "--unreachable" no longer makes any
459          * sense (since in this case everything will obviously
460          * be unreachable by definition.
461          *
462          * Showing dangling objects is valid, though (as those
463          * dangling objects are likely lost heads).
464          *
465          * So we just print a warning about it, and clear the
466          * "show_unreachable" flag.
467          */
468         if (!default_refs) {
469                 fprintf(stderr, "notice: No default references\n");
470                 show_unreachable = 0;
471         }
472 }
473
474 static void fsck_object_dir(const char *path)
475 {
476         int i;
477
478         if (verbose)
479                 fprintf(stderr, "Checking object directory\n");
480
481         for (i = 0; i < 256; i++) {
482                 static char dir[4096];
483                 sprintf(dir, "%s/%02x", path, i);
484                 fsck_dir(i, dir);
485         }
486         fsck_sha1_list();
487 }
488
489 static int fsck_head_link(void)
490 {
491         unsigned char sha1[20];
492         int flag;
493         int null_is_error = 0;
494         const char *head_points_at = resolve_ref("HEAD", sha1, 0, &flag);
495
496         if (verbose)
497                 fprintf(stderr, "Checking HEAD link\n");
498
499         if (!head_points_at)
500                 return error("Invalid HEAD");
501         if (!strcmp(head_points_at, "HEAD"))
502                 /* detached HEAD */
503                 null_is_error = 1;
504         else if (prefixcmp(head_points_at, "refs/heads/"))
505                 return error("HEAD points to something strange (%s)",
506                              head_points_at);
507         if (is_null_sha1(sha1)) {
508                 if (null_is_error)
509                         return error("HEAD: detached HEAD points at nothing");
510                 fprintf(stderr, "notice: HEAD points to an unborn branch (%s)\n",
511                         head_points_at + 11);
512         }
513         return 0;
514 }
515
516 static int fsck_cache_tree(struct cache_tree *it)
517 {
518         int i;
519         int err = 0;
520
521         if (verbose)
522                 fprintf(stderr, "Checking cache tree\n");
523
524         if (0 <= it->entry_count) {
525                 struct object *obj = parse_object(it->sha1);
526                 if (!obj) {
527                         error("%s: invalid sha1 pointer in cache-tree",
528                               sha1_to_hex(it->sha1));
529                         return 1;
530                 }
531                 mark_object_reachable(obj);
532                 obj->used = 1;
533                 if (obj->type != OBJ_TREE)
534                         err |= objerror(obj, "non-tree in cache-tree");
535         }
536         for (i = 0; i < it->subtree_nr; i++)
537                 err |= fsck_cache_tree(it->down[i]->cache_tree);
538         return err;
539 }
540
541 static char const * const fsck_usage[] = {
542         "git fsck [options] [<object>...]",
543         NULL
544 };
545
546 static struct option fsck_opts[] = {
547         OPT__VERBOSE(&verbose),
548         OPT_BOOLEAN(0, "unreachable", &show_unreachable, "show unreachable objects"),
549         OPT_BOOLEAN(0, "tags", &show_tags, "report tags"),
550         OPT_BOOLEAN(0, "root", &show_root, "report root nodes"),
551         OPT_BOOLEAN(0, "cache", &keep_cache_objects, "make index objects head nodes"),
552         OPT_BOOLEAN(0, "reflogs", &include_reflogs, "make reflogs head nodes (default)"),
553         OPT_BOOLEAN(0, "full", &check_full, "also consider alternate objects"),
554         OPT_BOOLEAN(0, "strict", &check_strict, "enable more strict checking"),
555         OPT_BOOLEAN(0, "lost-found", &write_lost_and_found,
556                                 "write dangling objects in .git/lost-found"),
557         OPT_END(),
558 };
559
560 int cmd_fsck(int argc, const char **argv, const char *prefix)
561 {
562         int i, heads;
563
564         errors_found = 0;
565
566         argc = parse_options(argc, argv, fsck_opts, fsck_usage, 0);
567         if (write_lost_and_found) {
568                 check_full = 1;
569                 include_reflogs = 0;
570         }
571
572         fsck_head_link();
573         fsck_object_dir(get_object_directory());
574         if (check_full) {
575                 struct alternate_object_database *alt;
576                 struct packed_git *p;
577                 prepare_alt_odb();
578                 for (alt = alt_odb_list; alt; alt = alt->next) {
579                         char namebuf[PATH_MAX];
580                         int namelen = alt->name - alt->base;
581                         memcpy(namebuf, alt->base, namelen);
582                         namebuf[namelen - 1] = 0;
583                         fsck_object_dir(namebuf);
584                 }
585                 prepare_packed_git();
586                 for (p = packed_git; p; p = p->next)
587                         /* verify gives error messages itself */
588                         verify_pack(p);
589
590                 for (p = packed_git; p; p = p->next) {
591                         uint32_t j, num;
592                         if (open_pack_index(p))
593                                 continue;
594                         num = p->num_objects;
595                         for (j = 0; j < num; j++)
596                                 fsck_sha1(nth_packed_object_sha1(p, j));
597                 }
598         }
599
600         heads = 0;
601         for (i = 1; i < argc; i++) {
602                 const char *arg = argv[i];
603                 if (!get_sha1(arg, head_sha1)) {
604                         struct object *obj = lookup_object(head_sha1);
605
606                         /* Error is printed by lookup_object(). */
607                         if (!obj)
608                                 continue;
609
610                         obj->used = 1;
611                         mark_object_reachable(obj);
612                         heads++;
613                         continue;
614                 }
615                 error("invalid parameter: expected sha1, got '%s'", arg);
616         }
617
618         /*
619          * If we've not been given any explicit head information, do the
620          * default ones from .git/refs. We also consider the index file
621          * in this case (ie this implies --cache).
622          */
623         if (!heads) {
624                 get_default_heads();
625                 keep_cache_objects = 1;
626         }
627
628         if (keep_cache_objects) {
629                 read_cache();
630                 for (i = 0; i < active_nr; i++) {
631                         unsigned int mode;
632                         struct blob *blob;
633                         struct object *obj;
634
635                         mode = active_cache[i]->ce_mode;
636                         if (S_ISGITLINK(mode))
637                                 continue;
638                         blob = lookup_blob(active_cache[i]->sha1);
639                         if (!blob)
640                                 continue;
641                         obj = &blob->object;
642                         obj->used = 1;
643                         mark_object_reachable(obj);
644                 }
645                 if (active_cache_tree)
646                         fsck_cache_tree(active_cache_tree);
647         }
648
649         check_connectivity();
650         return errors_found;
651 }