Make fsck-cache do better tree checking.
[git.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
10 #define REACHABLE 0x0001
11
12 static int show_root = 0;
13 static int show_tags = 0;
14 static int show_unreachable = 0;
15 static unsigned char head_sha1[20];
16
17 static void check_connectivity(void)
18 {
19         int i;
20
21         /* Look up all the requirements, warn about missing objects.. */
22         for (i = 0; i < nr_objs; i++) {
23                 struct object *obj = objs[i];
24                 struct object_list *refs;
25
26                 if (!obj->parsed) {
27                         printf("missing %s %s\n", obj->type, sha1_to_hex(obj->sha1));
28                         continue;
29                 }
30
31                 for (refs = obj->refs; refs; refs = refs->next) {
32                         if (refs->item->parsed)
33                                 continue;
34                         printf("broken link from %7s %s\n",
35                                obj->type, sha1_to_hex(obj->sha1));
36                         printf("              to %7s %s\n",
37                                obj->type, sha1_to_hex(refs->item->sha1));
38                 }
39
40                 /* Don't bother with tag reachability. */
41                 if (obj->type == tag_type)
42                         continue;
43
44                 if (show_unreachable && !(obj->flags & REACHABLE)) {
45                         printf("unreachable %s %s\n", obj->type, sha1_to_hex(obj->sha1));
46                         continue;
47                 }
48
49                 if (!obj->used) {
50                         printf("dangling %s %s\n", obj->type, 
51                                sha1_to_hex(obj->sha1));
52                 }
53         }
54 }
55
56 /*
57  * The entries in a tree are ordered in the _path_ order,
58  * which means that a directory entry is ordered by adding
59  * a slash to the end of it.
60  *
61  * So a directory called "a" is ordered _after_ a file
62  * called "a.c", because "a/" sorts after "a.c".
63  */
64 static int verify_ordered(struct tree_entry_list *a, struct tree_entry_list *b)
65 {
66         int len1 = strlen(a->name);
67         int len2 = strlen(b->name);
68         int len = len1 < len2 ? len1 : len2;
69         unsigned char c1, c2;
70         int cmp;
71
72         cmp = memcmp(a->name, b->name, len);
73         if (cmp < 0)
74                 return 0;
75         if (cmp > 0)
76                 return -1;
77
78         /*
79          * Ok, the first <len> characters are the same.
80          * Now we need to order the next one, but turn
81          * a '\0' into a '/' for a directory entry.
82          */
83         c1 = a->name[len];
84         c2 = b->name[len];
85         if (!c1 && a->directory)
86                 c1 = '/';
87         if (!c2 && b->directory)
88                 c2 = '/';
89         return c1 < c2 ? 0 : -1;
90 }
91
92 static int fsck_tree(struct tree *item)
93 {
94         int has_full_path = 0;
95         struct tree_entry_list *entry, *last;
96
97         last = NULL;
98         for (entry = item->entries; entry; entry = entry->next) {
99                 if (strchr(entry->name, '/'))
100                         has_full_path = 1;
101
102                 if (last) {
103                         if (verify_ordered(last, entry) < 0) {
104                                 fprintf(stderr, "tree %s not ordered\n",
105                                         sha1_to_hex(item->object.sha1));
106                                 return -1;
107                         }
108                 }
109
110                 last = entry;
111         }
112
113         if (has_full_path) {
114                 fprintf(stderr, "warning: fsck-cache: tree %s "
115                         "has full pathnames in it\n", 
116                         sha1_to_hex(item->object.sha1));
117         }
118
119         return 0;
120 }
121
122 static int fsck_commit(struct commit *commit)
123 {
124         if (!commit->tree)
125                 return -1;
126         if (!commit->parents && show_root)
127                 printf("root %s\n", sha1_to_hex(commit->object.sha1));
128         if (!commit->date)
129                 printf("bad commit date in %s\n", 
130                        sha1_to_hex(commit->object.sha1));
131         return 0;
132 }
133
134 static int fsck_tag(struct tag *tag)
135 {
136         if (!show_tags)
137                 return 0;
138
139         printf("tagged %s %s",
140                tag->tagged->type,
141                sha1_to_hex(tag->tagged->sha1));
142         printf(" (%s) in %s\n",
143                tag->tag, sha1_to_hex(tag->object.sha1));
144         return 0;
145 }
146
147 static int fsck_sha1(unsigned char *sha1)
148 {
149         struct object *obj = parse_object(sha1);
150         if (!obj)
151                 return -1;
152         if (obj->type == blob_type)
153                 return 0;
154         if (obj->type == tree_type)
155                 return fsck_tree((struct tree *) obj);
156         if (obj->type == commit_type)
157                 return fsck_commit((struct commit *) obj);
158         if (obj->type == tag_type)
159                 return fsck_tag((struct tag *) obj);
160         return -1;
161 }
162
163 /*
164  * This is the sorting chunk size: make it reasonably
165  * big so that we can sort well..
166  */
167 #define MAX_SHA1_ENTRIES (1024)
168
169 struct sha1_entry {
170         unsigned long ino;
171         unsigned char sha1[20];
172 };
173
174 static struct {
175         unsigned long nr;
176         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
177 } sha1_list;
178
179 static int ino_compare(const void *_a, const void *_b)
180 {
181         const struct sha1_entry *a = _a, *b = _b;
182         unsigned long ino1 = a->ino, ino2 = b->ino;
183         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
184 }
185
186 static void fsck_sha1_list(void)
187 {
188         int i, nr = sha1_list.nr;
189
190         qsort(sha1_list.entry, nr, sizeof(struct sha1_entry *), ino_compare);
191         for (i = 0; i < nr; i++) {
192                 struct sha1_entry *entry = sha1_list.entry[i];
193                 unsigned char *sha1 = entry->sha1;
194
195                 sha1_list.entry[i] = NULL;
196                 if (fsck_sha1(sha1) < 0)
197                         fprintf(stderr, "bad sha1 entry '%s'\n", sha1_to_hex(sha1));
198                 free(entry);
199         }
200         sha1_list.nr = 0;
201 }
202
203 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
204 {
205         struct sha1_entry *entry = xmalloc(sizeof(*entry));
206         int nr;
207
208         entry->ino = ino;
209         memcpy(entry->sha1, sha1, 20);
210         nr = sha1_list.nr;
211         if (nr == MAX_SHA1_ENTRIES) {
212                 fsck_sha1_list();
213                 nr = 0;
214         }
215         sha1_list.entry[nr] = entry;
216         sha1_list.nr = ++nr;
217 }
218
219 static int fsck_dir(int i, char *path)
220 {
221         DIR *dir = opendir(path);
222         struct dirent *de;
223
224         if (!dir) {
225                 return error("missing sha1 directory '%s'", path);
226         }
227
228         while ((de = readdir(dir)) != NULL) {
229                 char name[100];
230                 unsigned char sha1[20];
231                 int len = strlen(de->d_name);
232
233                 switch (len) {
234                 case 2:
235                         if (de->d_name[1] != '.')
236                                 break;
237                 case 1:
238                         if (de->d_name[0] != '.')
239                                 break;
240                         continue;
241                 case 38:
242                         sprintf(name, "%02x", i);
243                         memcpy(name+2, de->d_name, len+1);
244                         if (get_sha1_hex(name, sha1) < 0)
245                                 break;
246                         add_sha1_list(sha1, de->d_ino);
247                         continue;
248                 }
249                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
250         }
251         closedir(dir);
252         return 0;
253 }
254
255 int main(int argc, char **argv)
256 {
257         int i, heads;
258         char *sha1_dir;
259
260         for (i = 1; i < argc; i++) {
261                 const char *arg = argv[i];
262
263                 if (!strcmp(arg, "--unreachable")) {
264                         show_unreachable = 1;
265                         continue;
266                 }
267                 if (!strcmp(arg, "--tags")) {
268                         show_tags = 1;
269                         continue;
270                 }
271                 if (!strcmp(arg, "--root")) {
272                         show_root = 1;
273                         continue;
274                 }
275                 if (*arg == '-')
276                         usage("fsck-cache [--tags] [[--unreachable] <head-sha1>*]");
277         }
278
279         sha1_dir = getenv(DB_ENVIRONMENT) ? : DEFAULT_DB_ENVIRONMENT;
280         for (i = 0; i < 256; i++) {
281                 static char dir[4096];
282                 sprintf(dir, "%s/%02x", sha1_dir, i);
283                 fsck_dir(i, dir);
284         }
285         fsck_sha1_list();
286
287         heads = 0;
288         for (i = 1; i < argc; i++) {
289                 const char *arg = argv[i]; 
290
291                 if (*arg == '-')
292                         continue;
293
294                 if (!get_sha1(arg, head_sha1)) {
295                         struct commit *commit = lookup_commit(head_sha1);
296                         struct object *obj;
297
298                         /* Error is printed by lookup_commit(). */
299                         if (!commit)
300                                 continue;
301
302                         obj = &commit->object;
303                         obj->used = 1;
304                         mark_reachable(obj, REACHABLE);
305                         heads++;
306                         continue;
307                 }
308                 error("expected sha1, got %s", arg);
309         }
310
311         if (!heads) {
312                 if (show_unreachable) {
313                         fprintf(stderr, "unable to do reachability without a head\n");
314                         show_unreachable = 0; 
315                 }
316                 fprintf(stderr, "expect dangling commits - potential heads - due to lack of head information\n");
317         }
318
319         check_connectivity();
320         return 0;
321 }