Make git-fsck-cache check HEAD integrity
[git] / fsck-cache.c
1 #include <sys/types.h>
2 #include <dirent.h>
3
4 #include "cache.h"
5 #include "commit.h"
6 #include "tree.h"
7 #include "blob.h"
8 #include "tag.h"
9 #include "refs.h"
10 #include "pack.h"
11
12 #define REACHABLE 0x0001
13
14 static int show_root = 0;
15 static int show_tags = 0;
16 static int show_unreachable = 0;
17 static int standalone = 0;
18 static int check_full = 0;
19 static int keep_cache_objects = 0; 
20 static unsigned char head_sha1[20];
21
22 static void check_connectivity(void)
23 {
24         int i;
25
26         /* Look up all the requirements, warn about missing objects.. */
27         for (i = 0; i < nr_objs; i++) {
28                 struct object *obj = objs[i];
29                 struct object_list *refs;
30
31                 if (!obj->parsed) {
32                         if (!standalone && has_sha1_file(obj->sha1))
33                                 ; /* it is in pack */
34                         else
35                                 printf("missing %s %s\n",
36                                        obj->type, sha1_to_hex(obj->sha1));
37                         continue;
38                 }
39
40                 for (refs = obj->refs; refs; refs = refs->next) {
41                         if (refs->item->parsed ||
42                             (!standalone && has_sha1_file(refs->item->sha1)))
43                                 continue;
44                         printf("broken link from %7s %s\n",
45                                obj->type, sha1_to_hex(obj->sha1));
46                         printf("              to %7s %s\n",
47                                refs->item->type, sha1_to_hex(refs->item->sha1));
48                 }
49
50                 if (show_unreachable && !(obj->flags & REACHABLE)) {
51                         printf("unreachable %s %s\n",
52                                obj->type, sha1_to_hex(obj->sha1));
53                         continue;
54                 }
55
56                 if (!obj->used) {
57                         printf("dangling %s %s\n", obj->type, 
58                                sha1_to_hex(obj->sha1));
59                 }
60         }
61 }
62
63 /*
64  * The entries in a tree are ordered in the _path_ order,
65  * which means that a directory entry is ordered by adding
66  * a slash to the end of it.
67  *
68  * So a directory called "a" is ordered _after_ a file
69  * called "a.c", because "a/" sorts after "a.c".
70  */
71 #define TREE_UNORDERED (-1)
72 #define TREE_HAS_DUPS  (-2)
73
74 static int verify_ordered(struct tree_entry_list *a, struct tree_entry_list *b)
75 {
76         int len1 = strlen(a->name);
77         int len2 = strlen(b->name);
78         int len = len1 < len2 ? len1 : len2;
79         unsigned char c1, c2;
80         int cmp;
81
82         cmp = memcmp(a->name, b->name, len);
83         if (cmp < 0)
84                 return 0;
85         if (cmp > 0)
86                 return TREE_UNORDERED;
87
88         /*
89          * Ok, the first <len> characters are the same.
90          * Now we need to order the next one, but turn
91          * a '\0' into a '/' for a directory entry.
92          */
93         c1 = a->name[len];
94         c2 = b->name[len];
95         if (!c1 && !c2)
96                 /*
97                  * git-write-tree used to write out a nonsense tree that has
98                  * entries with the same name, one blob and one tree.  Make
99                  * sure we do not have duplicate entries.
100                  */
101                 return TREE_HAS_DUPS;
102         if (!c1 && a->directory)
103                 c1 = '/';
104         if (!c2 && b->directory)
105                 c2 = '/';
106         return c1 < c2 ? 0 : TREE_UNORDERED;
107 }
108
109 static int fsck_tree(struct tree *item)
110 {
111         int has_full_path = 0;
112         struct tree_entry_list *entry, *last;
113
114         last = NULL;
115         for (entry = item->entries; entry; entry = entry->next) {
116                 if (strchr(entry->name, '/'))
117                         has_full_path = 1;
118
119                 switch (entry->mode) {
120                 /*
121                  * Standard modes.. 
122                  */
123                 case S_IFREG | 0755:
124                 case S_IFREG | 0644:
125                 case S_IFLNK:
126                 case S_IFDIR:
127                         break;
128                 /*
129                  * This is nonstandard, but we had a few of these
130                  * early on when we honored the full set of mode
131                  * bits..
132                  */
133                 case S_IFREG | 0664:
134                         break;
135                 default:
136                         printf("tree %s has entry %o %s\n",
137                                 sha1_to_hex(item->object.sha1),
138                                 entry->mode, entry->name);
139                 }
140
141                 if (last) {
142                         switch (verify_ordered(last, entry)) {
143                         case TREE_UNORDERED:
144                                 fprintf(stderr, "tree %s not ordered\n",
145                                         sha1_to_hex(item->object.sha1));
146                                 return -1;
147                         case TREE_HAS_DUPS:
148                                 fprintf(stderr, "tree %s has duplicate entries for '%s'\n",
149                                         sha1_to_hex(item->object.sha1),
150                                         entry->name);
151                                 return -1;
152                         default:
153                                 break;
154                         }
155                 }
156
157                 last = entry;
158         }
159
160         if (has_full_path) {
161                 fprintf(stderr, "warning: git-fsck-cache: tree %s "
162                         "has full pathnames in it\n", 
163                         sha1_to_hex(item->object.sha1));
164         }
165
166         return 0;
167 }
168
169 static int fsck_commit(struct commit *commit)
170 {
171         free(commit->buffer);
172         commit->buffer = NULL;
173         if (!commit->tree)
174                 return -1;
175         if (!commit->parents && show_root)
176                 printf("root %s\n", sha1_to_hex(commit->object.sha1));
177         if (!commit->date)
178                 printf("bad commit date in %s\n", 
179                        sha1_to_hex(commit->object.sha1));
180         return 0;
181 }
182
183 static int fsck_tag(struct tag *tag)
184 {
185         struct object *tagged = tag->tagged;
186
187         if (!tagged) {
188                 printf("bad object in tag %s\n", sha1_to_hex(tag->object.sha1));
189                 return -1;
190         }
191         if (!show_tags)
192                 return 0;
193
194         printf("tagged %s %s", tagged->type, sha1_to_hex(tagged->sha1));
195         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
196         return 0;
197 }
198
199 static int fsck_sha1(unsigned char *sha1)
200 {
201         struct object *obj = parse_object(sha1);
202         if (!obj)
203                 return -1;
204         if (obj->type == blob_type)
205                 return 0;
206         if (obj->type == tree_type)
207                 return fsck_tree((struct tree *) obj);
208         if (obj->type == commit_type)
209                 return fsck_commit((struct commit *) obj);
210         if (obj->type == tag_type)
211                 return fsck_tag((struct tag *) obj);
212         return -1;
213 }
214
215 /*
216  * This is the sorting chunk size: make it reasonably
217  * big so that we can sort well..
218  */
219 #define MAX_SHA1_ENTRIES (1024)
220
221 struct sha1_entry {
222         unsigned long ino;
223         unsigned char sha1[20];
224 };
225
226 static struct {
227         unsigned long nr;
228         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
229 } sha1_list;
230
231 static int ino_compare(const void *_a, const void *_b)
232 {
233         const struct sha1_entry *a = _a, *b = _b;
234         unsigned long ino1 = a->ino, ino2 = b->ino;
235         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
236 }
237
238 static void fsck_sha1_list(void)
239 {
240         int i, nr = sha1_list.nr;
241
242         qsort(sha1_list.entry, nr, sizeof(struct sha1_entry *), ino_compare);
243         for (i = 0; i < nr; i++) {
244                 struct sha1_entry *entry = sha1_list.entry[i];
245                 unsigned char *sha1 = entry->sha1;
246
247                 sha1_list.entry[i] = NULL;
248                 if (fsck_sha1(sha1) < 0)
249                         fprintf(stderr, "bad sha1 entry '%s'\n", sha1_to_hex(sha1));
250                 free(entry);
251         }
252         sha1_list.nr = 0;
253 }
254
255 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
256 {
257         struct sha1_entry *entry = xmalloc(sizeof(*entry));
258         int nr;
259
260         entry->ino = ino;
261         memcpy(entry->sha1, sha1, 20);
262         nr = sha1_list.nr;
263         if (nr == MAX_SHA1_ENTRIES) {
264                 fsck_sha1_list();
265                 nr = 0;
266         }
267         sha1_list.entry[nr] = entry;
268         sha1_list.nr = ++nr;
269 }
270
271 static int fsck_dir(int i, char *path)
272 {
273         DIR *dir = opendir(path);
274         struct dirent *de;
275
276         if (!dir) {
277                 return error("missing sha1 directory '%s'", path);
278         }
279
280         while ((de = readdir(dir)) != NULL) {
281                 char name[100];
282                 unsigned char sha1[20];
283                 int len = strlen(de->d_name);
284
285                 switch (len) {
286                 case 2:
287                         if (de->d_name[1] != '.')
288                                 break;
289                 case 1:
290                         if (de->d_name[0] != '.')
291                                 break;
292                         continue;
293                 case 38:
294                         sprintf(name, "%02x", i);
295                         memcpy(name+2, de->d_name, len+1);
296                         if (get_sha1_hex(name, sha1) < 0)
297                                 break;
298                         add_sha1_list(sha1, de->d_ino);
299                         continue;
300                 }
301                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
302         }
303         closedir(dir);
304         return 0;
305 }
306
307 static int default_refs = 0;
308
309 static int fsck_handle_ref(const char *refname, const unsigned char *sha1)
310 {
311         struct object *obj;
312
313         obj = lookup_object(sha1);
314         if (!obj) {
315                 if (!standalone && has_sha1_file(sha1))
316                         return 0; /* it is in a pack */
317                 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
318                 /* We'll continue with the rest despite the error.. */
319                 return 0;
320         }
321         default_refs++;
322         obj->used = 1;
323         mark_reachable(obj, REACHABLE);
324         return 0;
325 }
326
327 static void get_default_heads(void)
328 {
329         for_each_ref(fsck_handle_ref);
330         if (!default_refs)
331                 die("No default references");
332 }
333
334 static void fsck_object_dir(const char *path)
335 {
336         int i;
337         for (i = 0; i < 256; i++) {
338                 static char dir[4096];
339                 sprintf(dir, "%s/%02x", path, i);
340                 fsck_dir(i, dir);
341         }
342         fsck_sha1_list();
343 }
344
345 static int fsck_head_link(void)
346 {
347         int fd, count;
348         char hex[40];
349         unsigned char sha1[20];
350         static char path[PATH_MAX], link[PATH_MAX];
351         const char *git_dir = gitenv(GIT_DIR_ENVIRONMENT) ? : DEFAULT_GIT_DIR_ENVIRONMENT;
352
353         snprintf(path, sizeof(path), "%s/HEAD", git_dir);
354         if (readlink(path, link, sizeof(link)) < 0)
355                 return error("HEAD is not a symlink");
356         if (strncmp("refs/heads/", link, 11))
357                 return error("HEAD points to something strange (%s)", link);
358         fd = open(path, O_RDONLY);
359         if (fd < 0)
360                 return error("HEAD: %s", strerror(errno));
361         count = read(fd, hex, sizeof(hex));
362         close(fd);
363         if (count < 0)
364                 return error("HEAD: %s", strerror(errno));
365         if (count < 40 || get_sha1_hex(hex, sha1))
366                 return error("HEAD: not a valid git pointer");
367         return 0;
368 }
369
370 int main(int argc, char **argv)
371 {
372         int i, heads;
373
374         for (i = 1; i < argc; i++) {
375                 const char *arg = argv[i];
376
377                 if (!strcmp(arg, "--unreachable")) {
378                         show_unreachable = 1;
379                         continue;
380                 }
381                 if (!strcmp(arg, "--tags")) {
382                         show_tags = 1;
383                         continue;
384                 }
385                 if (!strcmp(arg, "--root")) {
386                         show_root = 1;
387                         continue;
388                 }
389                 if (!strcmp(arg, "--cache")) {
390                         keep_cache_objects = 1;
391                         continue;
392                 }
393                 if (!strcmp(arg, "--standalone")) {
394                         standalone = 1;
395                         continue;
396                 }
397                 if (!strcmp(arg, "--full")) {
398                         check_full = 1;
399                         continue;
400                 }
401                 if (*arg == '-')
402                         usage("git-fsck-cache [--tags] [[--unreachable] [--cache] [--standalone | --full] <head-sha1>*]");
403         }
404
405         if (standalone && check_full)
406                 die("Only one of --standalone or --full can be used.");
407         if (standalone)
408                 unsetenv("GIT_ALTERNATE_OBJECT_DIRECTORIES");
409
410         fsck_head_link();
411         fsck_object_dir(get_object_directory());
412         if (check_full) {
413                 int j;
414                 struct packed_git *p;
415                 prepare_alt_odb();
416                 for (j = 0; alt_odb[j].base; j++) {
417                         alt_odb[j].name[-1] = 0; /* was slash */
418                         fsck_object_dir(alt_odb[j].base);
419                         alt_odb[j].name[-1] = '/';
420                 }
421                 prepare_packed_git();
422                 for (p = packed_git; p; p = p->next)
423                         /* verify gives error messages itself */
424                         verify_pack(p, 0);
425
426                 for (p = packed_git; p; p = p->next) {
427                         int num = num_packed_objects(p);
428                         for (i = 0; i < num; i++) {
429                                 unsigned char sha1[20];
430                                 nth_packed_object_sha1(p, i, sha1);
431                                 if (fsck_sha1(sha1) < 0)
432                                         fprintf(stderr, "bad sha1 entry '%s'\n", sha1_to_hex(sha1));
433
434                         }
435                 }
436         }
437
438         heads = 0;
439         for (i = 1; i < argc; i++) {
440                 const char *arg = argv[i]; 
441
442                 if (*arg == '-')
443                         continue;
444
445                 if (!get_sha1(arg, head_sha1)) {
446                         struct object *obj = lookup_object(head_sha1);
447
448                         /* Error is printed by lookup_object(). */
449                         if (!obj)
450                                 continue;
451
452                         obj->used = 1;
453                         mark_reachable(obj, REACHABLE);
454                         heads++;
455                         continue;
456                 }
457                 error("expected sha1, got %s", arg);
458         }
459
460         /*
461          * If we've not been given any explicit head information, do the
462          * default ones from .git/refs. We also consider the index file
463          * in this case (ie this implies --cache).
464          */
465         if (!heads) {
466                 get_default_heads();
467                 keep_cache_objects = 1;
468         }
469
470         if (keep_cache_objects) {
471                 int i;
472                 read_cache();
473                 for (i = 0; i < active_nr; i++) {
474                         struct blob *blob = lookup_blob(active_cache[i]->sha1);
475                         struct object *obj;
476                         if (!blob)
477                                 continue;
478                         obj = &blob->object;
479                         obj->used = 1;
480                         mark_reachable(obj, REACHABLE);
481                 }
482         }
483
484         check_connectivity();
485         return 0;
486 }