xdiff: give up scanning similar lines early
[git] / fsck.c
1 #include "cache.h"
2 #include "object.h"
3 #include "blob.h"
4 #include "tree.h"
5 #include "tree-walk.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "fsck.h"
9
10 static int fsck_walk_tree(struct tree *tree, fsck_walk_func walk, void *data)
11 {
12         struct tree_desc desc;
13         struct name_entry entry;
14         int res = 0;
15
16         if (parse_tree(tree))
17                 return -1;
18
19         init_tree_desc(&desc, tree->buffer, tree->size);
20         while (tree_entry(&desc, &entry)) {
21                 int result;
22
23                 if (S_ISGITLINK(entry.mode))
24                         continue;
25                 if (S_ISDIR(entry.mode))
26                         result = walk(&lookup_tree(entry.sha1)->object, OBJ_TREE, data);
27                 else if (S_ISREG(entry.mode) || S_ISLNK(entry.mode))
28                         result = walk(&lookup_blob(entry.sha1)->object, OBJ_BLOB, data);
29                 else {
30                         result = error("in tree %s: entry %s has bad mode %.6o\n",
31                                         sha1_to_hex(tree->object.sha1), entry.path, entry.mode);
32                 }
33                 if (result < 0)
34                         return result;
35                 if (!res)
36                         res = result;
37         }
38         return res;
39 }
40
41 static int fsck_walk_commit(struct commit *commit, fsck_walk_func walk, void *data)
42 {
43         struct commit_list *parents;
44         int res;
45         int result;
46
47         if (parse_commit(commit))
48                 return -1;
49
50         result = walk((struct object *)commit->tree, OBJ_TREE, data);
51         if (result < 0)
52                 return result;
53         res = result;
54
55         parents = commit->parents;
56         while (parents) {
57                 result = walk((struct object *)parents->item, OBJ_COMMIT, data);
58                 if (result < 0)
59                         return result;
60                 if (!res)
61                         res = result;
62                 parents = parents->next;
63         }
64         return res;
65 }
66
67 static int fsck_walk_tag(struct tag *tag, fsck_walk_func walk, void *data)
68 {
69         if (parse_tag(tag))
70                 return -1;
71         return walk(tag->tagged, OBJ_ANY, data);
72 }
73
74 int fsck_walk(struct object *obj, fsck_walk_func walk, void *data)
75 {
76         if (!obj)
77                 return -1;
78         switch (obj->type) {
79         case OBJ_BLOB:
80                 return 0;
81         case OBJ_TREE:
82                 return fsck_walk_tree((struct tree *)obj, walk, data);
83         case OBJ_COMMIT:
84                 return fsck_walk_commit((struct commit *)obj, walk, data);
85         case OBJ_TAG:
86                 return fsck_walk_tag((struct tag *)obj, walk, data);
87         default:
88                 error("Unknown object type for %s", sha1_to_hex(obj->sha1));
89                 return -1;
90         }
91 }
92
93 /*
94  * The entries in a tree are ordered in the _path_ order,
95  * which means that a directory entry is ordered by adding
96  * a slash to the end of it.
97  *
98  * So a directory called "a" is ordered _after_ a file
99  * called "a.c", because "a/" sorts after "a.c".
100  */
101 #define TREE_UNORDERED (-1)
102 #define TREE_HAS_DUPS  (-2)
103
104 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
105 {
106         int len1 = strlen(name1);
107         int len2 = strlen(name2);
108         int len = len1 < len2 ? len1 : len2;
109         unsigned char c1, c2;
110         int cmp;
111
112         cmp = memcmp(name1, name2, len);
113         if (cmp < 0)
114                 return 0;
115         if (cmp > 0)
116                 return TREE_UNORDERED;
117
118         /*
119          * Ok, the first <len> characters are the same.
120          * Now we need to order the next one, but turn
121          * a '\0' into a '/' for a directory entry.
122          */
123         c1 = name1[len];
124         c2 = name2[len];
125         if (!c1 && !c2)
126                 /*
127                  * git-write-tree used to write out a nonsense tree that has
128                  * entries with the same name, one blob and one tree.  Make
129                  * sure we do not have duplicate entries.
130                  */
131                 return TREE_HAS_DUPS;
132         if (!c1 && S_ISDIR(mode1))
133                 c1 = '/';
134         if (!c2 && S_ISDIR(mode2))
135                 c2 = '/';
136         return c1 < c2 ? 0 : TREE_UNORDERED;
137 }
138
139 static int fsck_tree(struct tree *item, int strict, fsck_error error_func)
140 {
141         int retval;
142         int has_full_path = 0;
143         int has_empty_name = 0;
144         int has_zero_pad = 0;
145         int has_bad_modes = 0;
146         int has_dup_entries = 0;
147         int not_properly_sorted = 0;
148         struct tree_desc desc;
149         unsigned o_mode;
150         const char *o_name;
151         const unsigned char *o_sha1;
152
153         init_tree_desc(&desc, item->buffer, item->size);
154
155         o_mode = 0;
156         o_name = NULL;
157         o_sha1 = NULL;
158
159         while (desc.size) {
160                 unsigned mode;
161                 const char *name;
162                 const unsigned char *sha1;
163
164                 sha1 = tree_entry_extract(&desc, &name, &mode);
165
166                 if (strchr(name, '/'))
167                         has_full_path = 1;
168                 if (!*name)
169                         has_empty_name = 1;
170                 has_zero_pad |= *(char *)desc.buffer == '0';
171                 update_tree_entry(&desc);
172
173                 switch (mode) {
174                 /*
175                  * Standard modes..
176                  */
177                 case S_IFREG | 0755:
178                 case S_IFREG | 0644:
179                 case S_IFLNK:
180                 case S_IFDIR:
181                 case S_IFGITLINK:
182                         break;
183                 /*
184                  * This is nonstandard, but we had a few of these
185                  * early on when we honored the full set of mode
186                  * bits..
187                  */
188                 case S_IFREG | 0664:
189                         if (!strict)
190                                 break;
191                 default:
192                         has_bad_modes = 1;
193                 }
194
195                 if (o_name) {
196                         switch (verify_ordered(o_mode, o_name, mode, name)) {
197                         case TREE_UNORDERED:
198                                 not_properly_sorted = 1;
199                                 break;
200                         case TREE_HAS_DUPS:
201                                 has_dup_entries = 1;
202                                 break;
203                         default:
204                                 break;
205                         }
206                 }
207
208                 o_mode = mode;
209                 o_name = name;
210                 o_sha1 = sha1;
211         }
212
213         retval = 0;
214         if (has_full_path)
215                 retval += error_func(&item->object, FSCK_WARN, "contains full pathnames");
216         if (has_empty_name)
217                 retval += error_func(&item->object, FSCK_WARN, "contains empty pathname");
218         if (has_zero_pad)
219                 retval += error_func(&item->object, FSCK_WARN, "contains zero-padded file modes");
220         if (has_bad_modes)
221                 retval += error_func(&item->object, FSCK_WARN, "contains bad file modes");
222         if (has_dup_entries)
223                 retval += error_func(&item->object, FSCK_ERROR, "contains duplicate file entries");
224         if (not_properly_sorted)
225                 retval += error_func(&item->object, FSCK_ERROR, "not properly sorted");
226         return retval;
227 }
228
229 static int fsck_commit(struct commit *commit, fsck_error error_func)
230 {
231         char *buffer = commit->buffer;
232         unsigned char tree_sha1[20], sha1[20];
233         struct commit_graft *graft;
234         int parents = 0;
235
236         if (!commit->date)
237                 return error_func(&commit->object, FSCK_ERROR, "invalid author/committer line");
238
239         if (memcmp(buffer, "tree ", 5))
240                 return error_func(&commit->object, FSCK_ERROR, "invalid format - expected 'tree' line");
241         if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
242                 return error_func(&commit->object, FSCK_ERROR, "invalid 'tree' line format - bad sha1");
243         buffer += 46;
244         while (!memcmp(buffer, "parent ", 7)) {
245                 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
246                         return error_func(&commit->object, FSCK_ERROR, "invalid 'parent' line format - bad sha1");
247                 buffer += 48;
248                 parents++;
249         }
250         graft = lookup_commit_graft(commit->object.sha1);
251         if (graft) {
252                 struct commit_list *p = commit->parents;
253                 parents = 0;
254                 while (p) {
255                         p = p->next;
256                         parents++;
257                 }
258                 if (graft->nr_parent == -1 && !parents)
259                         ; /* shallow commit */
260                 else if (graft->nr_parent != parents)
261                         return error_func(&commit->object, FSCK_ERROR, "graft objects missing");
262         } else {
263                 struct commit_list *p = commit->parents;
264                 while (p && parents) {
265                         p = p->next;
266                         parents--;
267                 }
268                 if (p || parents)
269                         return error_func(&commit->object, FSCK_ERROR, "parent objects missing");
270         }
271         if (memcmp(buffer, "author ", 7))
272                 return error_func(&commit->object, FSCK_ERROR, "invalid format - expected 'author' line");
273         if (!commit->tree)
274                 return error_func(&commit->object, FSCK_ERROR, "could not load commit's tree %s", sha1_to_hex(tree_sha1));
275
276         return 0;
277 }
278
279 static int fsck_tag(struct tag *tag, fsck_error error_func)
280 {
281         struct object *tagged = tag->tagged;
282
283         if (!tagged)
284                 return error_func(&tag->object, FSCK_ERROR, "could not load tagged object");
285         return 0;
286 }
287
288 int fsck_object(struct object *obj, int strict, fsck_error error_func)
289 {
290         if (!obj)
291                 return error_func(obj, FSCK_ERROR, "no valid object to fsck");
292
293         if (obj->type == OBJ_BLOB)
294                 return 0;
295         if (obj->type == OBJ_TREE)
296                 return fsck_tree((struct tree *) obj, strict, error_func);
297         if (obj->type == OBJ_COMMIT)
298                 return fsck_commit((struct commit *) obj, error_func);
299         if (obj->type == OBJ_TAG)
300                 return fsck_tag((struct tag *) obj, error_func);
301
302         return error_func(obj, FSCK_ERROR, "unknown type '%d' (internal fsck error)",
303                           obj->type);
304 }
305
306 int fsck_error_function(struct object *obj, int type, const char *fmt, ...)
307 {
308         va_list ap;
309         int len;
310         struct strbuf sb = STRBUF_INIT;
311
312         strbuf_addf(&sb, "object %s:", obj->sha1?sha1_to_hex(obj->sha1):"(null)");
313
314         va_start(ap, fmt);
315         len = vsnprintf(sb.buf + sb.len, strbuf_avail(&sb), fmt, ap);
316         va_end(ap);
317
318         if (len < 0)
319                 len = 0;
320         if (len >= strbuf_avail(&sb)) {
321                 strbuf_grow(&sb, len + 2);
322                 va_start(ap, fmt);
323                 len = vsnprintf(sb.buf + sb.len, strbuf_avail(&sb), fmt, ap);
324                 va_end(ap);
325                 if (len >= strbuf_avail(&sb))
326                         die("this should not happen, your snprintf is broken");
327         }
328
329         error(sb.buf);
330         strbuf_release(&sb);
331         return 1;
332 }