diff-tree: fix "whole sub-tree disappeared or appeared" case
[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 int keep_cache_objects = 0; 
16 static unsigned char head_sha1[20];
17
18 static void check_connectivity(void)
19 {
20         int i;
21
22         /* Look up all the requirements, warn about missing objects.. */
23         for (i = 0; i < nr_objs; i++) {
24                 struct object *obj = objs[i];
25                 struct object_list *refs;
26
27                 if (!obj->parsed) {
28                         printf("missing %s %s\n", obj->type, sha1_to_hex(obj->sha1));
29                         continue;
30                 }
31
32                 for (refs = obj->refs; refs; refs = refs->next) {
33                         if (refs->item->parsed)
34                                 continue;
35                         printf("broken link from %7s %s\n",
36                                obj->type, sha1_to_hex(obj->sha1));
37                         printf("              to %7s %s\n",
38                                refs->item->type, sha1_to_hex(refs->item->sha1));
39                 }
40
41                 /* Don't bother with tag reachability. */
42                 if (obj->type == tag_type)
43                         continue;
44
45                 if (show_unreachable && !(obj->flags & REACHABLE)) {
46                         printf("unreachable %s %s\n", obj->type, sha1_to_hex(obj->sha1));
47                         continue;
48                 }
49
50                 if (!obj->used) {
51                         printf("dangling %s %s\n", obj->type, 
52                                sha1_to_hex(obj->sha1));
53                 }
54         }
55 }
56
57 /*
58  * The entries in a tree are ordered in the _path_ order,
59  * which means that a directory entry is ordered by adding
60  * a slash to the end of it.
61  *
62  * So a directory called "a" is ordered _after_ a file
63  * called "a.c", because "a/" sorts after "a.c".
64  */
65 #define TREE_UNORDERED (-1)
66 #define TREE_HAS_DUPS  (-2)
67
68 static int verify_ordered(struct tree_entry_list *a, struct tree_entry_list *b)
69 {
70         int len1 = strlen(a->name);
71         int len2 = strlen(b->name);
72         int len = len1 < len2 ? len1 : len2;
73         unsigned char c1, c2;
74         int cmp;
75
76         cmp = memcmp(a->name, b->name, len);
77         if (cmp < 0)
78                 return 0;
79         if (cmp > 0)
80                 return TREE_UNORDERED;
81
82         /*
83          * Ok, the first <len> characters are the same.
84          * Now we need to order the next one, but turn
85          * a '\0' into a '/' for a directory entry.
86          */
87         c1 = a->name[len];
88         c2 = b->name[len];
89         if (!c1 && !c2)
90                 /*
91                  * git-write-tree used to write out a nonsense tree that has
92                  * entries with the same name, one blob and one tree.  Make
93                  * sure we do not have duplicate entries.
94                  */
95                 return TREE_HAS_DUPS;
96         if (!c1 && a->directory)
97                 c1 = '/';
98         if (!c2 && b->directory)
99                 c2 = '/';
100         return c1 < c2 ? 0 : TREE_UNORDERED;
101 }
102
103 static int fsck_tree(struct tree *item)
104 {
105         int has_full_path = 0;
106         struct tree_entry_list *entry, *last;
107
108         last = NULL;
109         for (entry = item->entries; entry; entry = entry->next) {
110                 if (strchr(entry->name, '/'))
111                         has_full_path = 1;
112
113                 switch (entry->mode) {
114                 /*
115                  * Standard modes.. 
116                  */
117                 case S_IFREG | 0755:
118                 case S_IFREG | 0644:
119                 case S_IFLNK:
120                 case S_IFDIR:
121                         break;
122                 /*
123                  * This is nonstandard, but we had a few of these
124                  * early on when we honored the full set of mode
125                  * bits..
126                  */
127                 case S_IFREG | 0664:
128                         break;
129                 default:
130                         printf("tree %s has entry %o %s\n",
131                                 sha1_to_hex(item->object.sha1),
132                                 entry->mode, entry->name);
133                 }
134
135                 if (last) {
136                         switch (verify_ordered(last, entry)) {
137                         case TREE_UNORDERED:
138                                 fprintf(stderr, "tree %s not ordered\n",
139                                         sha1_to_hex(item->object.sha1));
140                                 return -1;
141                         case TREE_HAS_DUPS:
142                                 fprintf(stderr, "tree %s has duplicate entries for '%s'\n",
143                                         sha1_to_hex(item->object.sha1),
144                                         entry->name);
145                                 return -1;
146                         default:
147                                 break;
148                         }
149                 }
150
151                 last = entry;
152         }
153
154         if (has_full_path) {
155                 fprintf(stderr, "warning: fsck-cache: tree %s "
156                         "has full pathnames in it\n", 
157                         sha1_to_hex(item->object.sha1));
158         }
159
160         return 0;
161 }
162
163 static int fsck_commit(struct commit *commit)
164 {
165         if (!commit->tree)
166                 return -1;
167         if (!commit->parents && show_root)
168                 printf("root %s\n", sha1_to_hex(commit->object.sha1));
169         if (!commit->date)
170                 printf("bad commit date in %s\n", 
171                        sha1_to_hex(commit->object.sha1));
172         return 0;
173 }
174
175 static int fsck_tag(struct tag *tag)
176 {
177         struct object *tagged = tag->tagged;
178
179         if (!tagged) {
180                 printf("bad object in tag %s\n", sha1_to_hex(tag->object.sha1));
181                 return -1;
182         }
183         if (!show_tags)
184                 return 0;
185
186         printf("tagged %s %s", tagged->type, sha1_to_hex(tagged->sha1));
187         printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
188         return 0;
189 }
190
191 static int fsck_sha1(unsigned char *sha1)
192 {
193         struct object *obj = parse_object(sha1);
194         if (!obj)
195                 return -1;
196         if (obj->type == blob_type)
197                 return 0;
198         if (obj->type == tree_type)
199                 return fsck_tree((struct tree *) obj);
200         if (obj->type == commit_type)
201                 return fsck_commit((struct commit *) obj);
202         if (obj->type == tag_type)
203                 return fsck_tag((struct tag *) obj);
204         return -1;
205 }
206
207 /*
208  * This is the sorting chunk size: make it reasonably
209  * big so that we can sort well..
210  */
211 #define MAX_SHA1_ENTRIES (1024)
212
213 struct sha1_entry {
214         unsigned long ino;
215         unsigned char sha1[20];
216 };
217
218 static struct {
219         unsigned long nr;
220         struct sha1_entry *entry[MAX_SHA1_ENTRIES];
221 } sha1_list;
222
223 static int ino_compare(const void *_a, const void *_b)
224 {
225         const struct sha1_entry *a = _a, *b = _b;
226         unsigned long ino1 = a->ino, ino2 = b->ino;
227         return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
228 }
229
230 static void fsck_sha1_list(void)
231 {
232         int i, nr = sha1_list.nr;
233
234         qsort(sha1_list.entry, nr, sizeof(struct sha1_entry *), ino_compare);
235         for (i = 0; i < nr; i++) {
236                 struct sha1_entry *entry = sha1_list.entry[i];
237                 unsigned char *sha1 = entry->sha1;
238
239                 sha1_list.entry[i] = NULL;
240                 if (fsck_sha1(sha1) < 0)
241                         fprintf(stderr, "bad sha1 entry '%s'\n", sha1_to_hex(sha1));
242                 free(entry);
243         }
244         sha1_list.nr = 0;
245 }
246
247 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
248 {
249         struct sha1_entry *entry = xmalloc(sizeof(*entry));
250         int nr;
251
252         entry->ino = ino;
253         memcpy(entry->sha1, sha1, 20);
254         nr = sha1_list.nr;
255         if (nr == MAX_SHA1_ENTRIES) {
256                 fsck_sha1_list();
257                 nr = 0;
258         }
259         sha1_list.entry[nr] = entry;
260         sha1_list.nr = ++nr;
261 }
262
263 static int fsck_dir(int i, char *path)
264 {
265         DIR *dir = opendir(path);
266         struct dirent *de;
267
268         if (!dir) {
269                 return error("missing sha1 directory '%s'", path);
270         }
271
272         while ((de = readdir(dir)) != NULL) {
273                 char name[100];
274                 unsigned char sha1[20];
275                 int len = strlen(de->d_name);
276
277                 switch (len) {
278                 case 2:
279                         if (de->d_name[1] != '.')
280                                 break;
281                 case 1:
282                         if (de->d_name[0] != '.')
283                                 break;
284                         continue;
285                 case 38:
286                         sprintf(name, "%02x", i);
287                         memcpy(name+2, de->d_name, len+1);
288                         if (get_sha1_hex(name, sha1) < 0)
289                                 break;
290                         add_sha1_list(sha1, de->d_ino);
291                         continue;
292                 }
293                 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
294         }
295         closedir(dir);
296         return 0;
297 }
298
299 static void read_sha1_reference(const char *path)
300 {
301         char hexname[60];
302         unsigned char sha1[20];
303         int fd = open(path, O_RDONLY), len;
304         struct object *obj;
305
306         if (fd < 0)
307                 return;
308
309         len = read(fd, hexname, sizeof(hexname));
310         close(fd);
311         if (len < 40)
312                 return;
313
314         if (get_sha1_hex(hexname, sha1) < 0)
315                 return;
316
317         obj = lookup_object(sha1);
318         obj->used = 1;
319         mark_reachable(obj, REACHABLE);
320 }
321
322 static void find_file_objects(const char *base, const char *name)
323 {
324         int baselen = strlen(base);
325         int namelen = strlen(name);
326         char *path = xmalloc(baselen + namelen + 2);
327         struct stat st;
328
329         memcpy(path, base, baselen);
330         path[baselen] = '/';
331         memcpy(path + baselen + 1, name, namelen+1);
332         if (stat(path, &st) < 0)
333                 return;
334
335         /*
336          * Recurse into directories
337          */
338         if (S_ISDIR(st.st_mode)) {
339                 DIR *dir = opendir(path);
340                 if (dir) {
341                         struct dirent *de;
342                         while ((de = readdir(dir)) != NULL) {
343                                 if (de->d_name[0] == '.')
344                                         continue;
345                                 find_file_objects(path, de->d_name);
346                         }
347                         closedir(dir);
348                 }
349                 return;
350         }
351         if (S_ISREG(st.st_mode)) {
352                 read_sha1_reference(path);
353                 return;
354         }
355 }
356
357 static void get_default_heads(void)
358 {
359         char *git_dir = gitenv(GIT_DIR_ENVIRONMENT) ? : DEFAULT_GIT_DIR_ENVIRONMENT;
360         find_file_objects(git_dir, "refs");
361 }
362
363 int main(int argc, char **argv)
364 {
365         int i, heads;
366         char *sha1_dir;
367
368         for (i = 1; i < argc; i++) {
369                 const char *arg = argv[i];
370
371                 if (!strcmp(arg, "--unreachable")) {
372                         show_unreachable = 1;
373                         continue;
374                 }
375                 if (!strcmp(arg, "--tags")) {
376                         show_tags = 1;
377                         continue;
378                 }
379                 if (!strcmp(arg, "--root")) {
380                         show_root = 1;
381                         continue;
382                 }
383                 if (!strcmp(arg, "--cache")) {
384                         keep_cache_objects = 1;
385                         continue;
386                 }
387                 if (*arg == '-')
388                         usage("fsck-cache [--tags] [[--unreachable] [--cache] <head-sha1>*]");
389         }
390
391         sha1_dir = get_object_directory();
392         for (i = 0; i < 256; i++) {
393                 static char dir[4096];
394                 sprintf(dir, "%s/%02x", sha1_dir, i);
395                 fsck_dir(i, dir);
396         }
397         fsck_sha1_list();
398
399         heads = 0;
400         for (i = 1; i < argc; i++) {
401                 const char *arg = argv[i]; 
402
403                 if (*arg == '-')
404                         continue;
405
406                 if (!get_sha1(arg, head_sha1)) {
407                         struct object *obj = lookup_object(head_sha1);
408
409                         /* Error is printed by lookup_object(). */
410                         if (!obj)
411                                 continue;
412
413                         obj->used = 1;
414                         mark_reachable(obj, REACHABLE);
415                         heads++;
416                         continue;
417                 }
418                 error("expected sha1, got %s", arg);
419         }
420
421         /*
422          * If we've not been gived any explicit head information, do the
423          * default ones from .git/refs. We also consider the index file
424          * in this case (ie this implies --cache).
425          */
426         if (!heads) {
427                 get_default_heads();
428                 keep_cache_objects = 1;
429         }
430
431         if (keep_cache_objects) {
432                 int i;
433                 read_cache();
434                 for (i = 0; i < active_nr; i++) {
435                         struct blob *blob = lookup_blob(active_cache[i]->sha1);
436                         struct object *obj;
437                         if (!blob)
438                                 continue;
439                         obj = &blob->object;
440                         obj->used = 1;
441                         mark_reachable(obj, REACHABLE);
442                 }
443         }
444
445         check_connectivity();
446         return 0;
447 }