git-svn: remember to check for clean indices on globbed refs, too
[git/fastimport.git] / builtin-fsck.c
blob6abf498d2b391c97e9efcdb32da1cf3b5d566397
1 #include "cache.h"
2 #include "commit.h"
3 #include "tree.h"
4 #include "blob.h"
5 #include "tag.h"
6 #include "refs.h"
7 #include "pack.h"
8 #include "cache-tree.h"
9 #include "tree-walk.h"
11 #define REACHABLE 0x0001
12 #define SEEN 0x0002
14 static int show_root;
15 static int show_tags;
16 static int show_unreachable;
17 static int check_full;
18 static int check_strict;
19 static int keep_cache_objects;
20 static unsigned char head_sha1[20];
22 #ifdef NO_D_INO_IN_DIRENT
23 #define SORT_DIRENT 0
24 #define DIRENT_SORT_HINT(de) 0
25 #else
26 #define SORT_DIRENT 1
27 #define DIRENT_SORT_HINT(de) ((de)->d_ino)
28 #endif
30 static void objreport(struct object *obj, const char *severity,
31 const char *err, va_list params)
33 fprintf(stderr, "%s in %s %s: ",
34 severity, typename(obj->type), sha1_to_hex(obj->sha1));
35 vfprintf(stderr, err, params);
36 fputs("\n", stderr);
39 static int objerror(struct object *obj, const char *err, ...)
41 va_list params;
42 va_start(params, err);
43 objreport(obj, "error", err, params);
44 va_end(params);
45 return -1;
48 static int objwarning(struct object *obj, const char *err, ...)
50 va_list params;
51 va_start(params, err);
52 objreport(obj, "warning", err, params);
53 va_end(params);
54 return -1;
58 * Check a single reachable object
60 static void check_reachable_object(struct object *obj)
62 const struct object_refs *refs;
65 * We obviously want the object to be parsed,
66 * except if it was in a pack-file and we didn't
67 * do a full fsck
69 if (!obj->parsed) {
70 if (has_sha1_file(obj->sha1))
71 return; /* it is in pack - forget about it */
72 printf("missing %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
73 return;
77 * Check that everything that we try to reference is also good.
79 refs = lookup_object_refs(obj);
80 if (refs) {
81 unsigned j;
82 for (j = 0; j < refs->count; j++) {
83 struct object *ref = refs->ref[j];
84 if (ref->parsed ||
85 (has_sha1_file(ref->sha1)))
86 continue;
87 printf("broken link from %7s %s\n",
88 typename(obj->type), sha1_to_hex(obj->sha1));
89 printf(" to %7s %s\n",
90 typename(ref->type), sha1_to_hex(ref->sha1));
96 * Check a single unreachable object
98 static void check_unreachable_object(struct object *obj)
101 * Missing unreachable object? Ignore it. It's not like
102 * we miss it (since it can't be reached), nor do we want
103 * to complain about it being unreachable (since it does
104 * not exist).
106 if (!obj->parsed)
107 return;
110 * Unreachable object that exists? Show it if asked to,
111 * since this is something that is prunable.
113 if (show_unreachable) {
114 printf("unreachable %s %s\n", typename(obj->type), sha1_to_hex(obj->sha1));
115 return;
119 * "!used" means that nothing at all points to it, including
120 * other unreachable objects. In other words, it's the "tip"
121 * of some set of unreachable objects, usually a commit that
122 * got dropped.
124 * Such starting points are more interesting than some random
125 * set of unreachable objects, so we show them even if the user
126 * hasn't asked for _all_ unreachable objects. If you have
127 * deleted a branch by mistake, this is a prime candidate to
128 * start looking at, for example.
130 if (!obj->used) {
131 printf("dangling %s %s\n", typename(obj->type),
132 sha1_to_hex(obj->sha1));
133 return;
137 * Otherwise? It's there, it's unreachable, and some other unreachable
138 * object points to it. Ignore it - it's not interesting, and we showed
139 * all the interesting cases above.
143 static void check_object(struct object *obj)
145 if (obj->flags & REACHABLE)
146 check_reachable_object(obj);
147 else
148 check_unreachable_object(obj);
151 static void check_connectivity(void)
153 int i, max;
155 /* Look up all the requirements, warn about missing objects.. */
156 max = get_max_object_index();
157 for (i = 0; i < max; i++) {
158 struct object *obj = get_indexed_object(i);
160 if (obj)
161 check_object(obj);
166 * The entries in a tree are ordered in the _path_ order,
167 * which means that a directory entry is ordered by adding
168 * a slash to the end of it.
170 * So a directory called "a" is ordered _after_ a file
171 * called "a.c", because "a/" sorts after "a.c".
173 #define TREE_UNORDERED (-1)
174 #define TREE_HAS_DUPS (-2)
176 static int verify_ordered(unsigned mode1, const char *name1, unsigned mode2, const char *name2)
178 int len1 = strlen(name1);
179 int len2 = strlen(name2);
180 int len = len1 < len2 ? len1 : len2;
181 unsigned char c1, c2;
182 int cmp;
184 cmp = memcmp(name1, name2, len);
185 if (cmp < 0)
186 return 0;
187 if (cmp > 0)
188 return TREE_UNORDERED;
191 * Ok, the first <len> characters are the same.
192 * Now we need to order the next one, but turn
193 * a '\0' into a '/' for a directory entry.
195 c1 = name1[len];
196 c2 = name2[len];
197 if (!c1 && !c2)
199 * git-write-tree used to write out a nonsense tree that has
200 * entries with the same name, one blob and one tree. Make
201 * sure we do not have duplicate entries.
203 return TREE_HAS_DUPS;
204 if (!c1 && S_ISDIR(mode1))
205 c1 = '/';
206 if (!c2 && S_ISDIR(mode2))
207 c2 = '/';
208 return c1 < c2 ? 0 : TREE_UNORDERED;
211 static int fsck_tree(struct tree *item)
213 int retval;
214 int has_full_path = 0;
215 int has_zero_pad = 0;
216 int has_bad_modes = 0;
217 int has_dup_entries = 0;
218 int not_properly_sorted = 0;
219 struct tree_desc desc;
220 unsigned o_mode;
221 const char *o_name;
222 const unsigned char *o_sha1;
224 desc.buf = item->buffer;
225 desc.size = item->size;
227 o_mode = 0;
228 o_name = NULL;
229 o_sha1 = NULL;
230 while (desc.size) {
231 unsigned mode;
232 const char *name;
233 const unsigned char *sha1;
235 sha1 = tree_entry_extract(&desc, &name, &mode);
237 if (strchr(name, '/'))
238 has_full_path = 1;
239 has_zero_pad |= *(char *)desc.buf == '0';
240 update_tree_entry(&desc);
242 switch (mode) {
244 * Standard modes..
246 case S_IFREG | 0755:
247 case S_IFREG | 0644:
248 case S_IFLNK:
249 case S_IFDIR:
250 break;
252 * This is nonstandard, but we had a few of these
253 * early on when we honored the full set of mode
254 * bits..
256 case S_IFREG | 0664:
257 if (!check_strict)
258 break;
259 default:
260 has_bad_modes = 1;
263 if (o_name) {
264 switch (verify_ordered(o_mode, o_name, mode, name)) {
265 case TREE_UNORDERED:
266 not_properly_sorted = 1;
267 break;
268 case TREE_HAS_DUPS:
269 has_dup_entries = 1;
270 break;
271 default:
272 break;
276 o_mode = mode;
277 o_name = name;
278 o_sha1 = sha1;
280 free(item->buffer);
281 item->buffer = NULL;
283 retval = 0;
284 if (has_full_path) {
285 objwarning(&item->object, "contains full pathnames");
287 if (has_zero_pad) {
288 objwarning(&item->object, "contains zero-padded file modes");
290 if (has_bad_modes) {
291 objwarning(&item->object, "contains bad file modes");
293 if (has_dup_entries) {
294 retval = objerror(&item->object, "contains duplicate file entries");
296 if (not_properly_sorted) {
297 retval = objerror(&item->object, "not properly sorted");
299 return retval;
302 static int fsck_commit(struct commit *commit)
304 char *buffer = commit->buffer;
305 unsigned char tree_sha1[20], sha1[20];
307 if (memcmp(buffer, "tree ", 5))
308 return objerror(&commit->object, "invalid format - expected 'tree' line");
309 if (get_sha1_hex(buffer+5, tree_sha1) || buffer[45] != '\n')
310 return objerror(&commit->object, "invalid 'tree' line format - bad sha1");
311 buffer += 46;
312 while (!memcmp(buffer, "parent ", 7)) {
313 if (get_sha1_hex(buffer+7, sha1) || buffer[47] != '\n')
314 return objerror(&commit->object, "invalid 'parent' line format - bad sha1");
315 buffer += 48;
317 if (memcmp(buffer, "author ", 7))
318 return objerror(&commit->object, "invalid format - expected 'author' line");
319 free(commit->buffer);
320 commit->buffer = NULL;
321 if (!commit->tree)
322 return objerror(&commit->object, "could not load commit's tree %s", tree_sha1);
323 if (!commit->parents && show_root)
324 printf("root %s\n", sha1_to_hex(commit->object.sha1));
325 if (!commit->date)
326 printf("bad commit date in %s\n",
327 sha1_to_hex(commit->object.sha1));
328 return 0;
331 static int fsck_tag(struct tag *tag)
333 struct object *tagged = tag->tagged;
335 if (!tagged) {
336 return objerror(&tag->object, "could not load tagged object");
338 if (!show_tags)
339 return 0;
341 printf("tagged %s %s", typename(tagged->type), sha1_to_hex(tagged->sha1));
342 printf(" (%s) in %s\n", tag->tag, sha1_to_hex(tag->object.sha1));
343 return 0;
346 static int fsck_sha1(unsigned char *sha1)
348 struct object *obj = parse_object(sha1);
349 if (!obj)
350 return error("%s: object corrupt or missing", sha1_to_hex(sha1));
351 if (obj->flags & SEEN)
352 return 0;
353 obj->flags |= SEEN;
354 if (obj->type == OBJ_BLOB)
355 return 0;
356 if (obj->type == OBJ_TREE)
357 return fsck_tree((struct tree *) obj);
358 if (obj->type == OBJ_COMMIT)
359 return fsck_commit((struct commit *) obj);
360 if (obj->type == OBJ_TAG)
361 return fsck_tag((struct tag *) obj);
362 /* By now, parse_object() would've returned NULL instead. */
363 return objerror(obj, "unknown type '%d' (internal fsck error)", obj->type);
367 * This is the sorting chunk size: make it reasonably
368 * big so that we can sort well..
370 #define MAX_SHA1_ENTRIES (1024)
372 struct sha1_entry {
373 unsigned long ino;
374 unsigned char sha1[20];
377 static struct {
378 unsigned long nr;
379 struct sha1_entry *entry[MAX_SHA1_ENTRIES];
380 } sha1_list;
382 static int ino_compare(const void *_a, const void *_b)
384 const struct sha1_entry *a = _a, *b = _b;
385 unsigned long ino1 = a->ino, ino2 = b->ino;
386 return ino1 < ino2 ? -1 : ino1 > ino2 ? 1 : 0;
389 static void fsck_sha1_list(void)
391 int i, nr = sha1_list.nr;
393 if (SORT_DIRENT)
394 qsort(sha1_list.entry, nr,
395 sizeof(struct sha1_entry *), ino_compare);
396 for (i = 0; i < nr; i++) {
397 struct sha1_entry *entry = sha1_list.entry[i];
398 unsigned char *sha1 = entry->sha1;
400 sha1_list.entry[i] = NULL;
401 fsck_sha1(sha1);
402 free(entry);
404 sha1_list.nr = 0;
407 static void add_sha1_list(unsigned char *sha1, unsigned long ino)
409 struct sha1_entry *entry = xmalloc(sizeof(*entry));
410 int nr;
412 entry->ino = ino;
413 hashcpy(entry->sha1, sha1);
414 nr = sha1_list.nr;
415 if (nr == MAX_SHA1_ENTRIES) {
416 fsck_sha1_list();
417 nr = 0;
419 sha1_list.entry[nr] = entry;
420 sha1_list.nr = ++nr;
423 static void fsck_dir(int i, char *path)
425 DIR *dir = opendir(path);
426 struct dirent *de;
428 if (!dir)
429 return;
431 while ((de = readdir(dir)) != NULL) {
432 char name[100];
433 unsigned char sha1[20];
434 int len = strlen(de->d_name);
436 switch (len) {
437 case 2:
438 if (de->d_name[1] != '.')
439 break;
440 case 1:
441 if (de->d_name[0] != '.')
442 break;
443 continue;
444 case 38:
445 sprintf(name, "%02x", i);
446 memcpy(name+2, de->d_name, len+1);
447 if (get_sha1_hex(name, sha1) < 0)
448 break;
449 add_sha1_list(sha1, DIRENT_SORT_HINT(de));
450 continue;
452 fprintf(stderr, "bad sha1 file: %s/%s\n", path, de->d_name);
454 closedir(dir);
457 static int default_refs;
459 static int fsck_handle_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
460 const char *email, unsigned long timestamp, int tz,
461 const char *message, void *cb_data)
463 struct object *obj;
465 if (!is_null_sha1(osha1)) {
466 obj = lookup_object(osha1);
467 if (obj) {
468 obj->used = 1;
469 mark_reachable(obj, REACHABLE);
472 obj = lookup_object(nsha1);
473 if (obj) {
474 obj->used = 1;
475 mark_reachable(obj, REACHABLE);
477 return 0;
480 static int fsck_handle_reflog(const char *logname, const unsigned char *sha1, int flag, void *cb_data)
482 for_each_reflog_ent(logname, fsck_handle_reflog_ent, NULL);
483 return 0;
486 static int fsck_handle_ref(const char *refname, const unsigned char *sha1, int flag, void *cb_data)
488 struct object *obj;
490 obj = lookup_object(sha1);
491 if (!obj) {
492 if (has_sha1_file(sha1)) {
493 default_refs++;
494 return 0; /* it is in a pack */
496 error("%s: invalid sha1 pointer %s", refname, sha1_to_hex(sha1));
497 /* We'll continue with the rest despite the error.. */
498 return 0;
500 default_refs++;
501 obj->used = 1;
502 mark_reachable(obj, REACHABLE);
504 return 0;
507 static void get_default_heads(void)
509 for_each_ref(fsck_handle_ref, NULL);
510 for_each_reflog(fsck_handle_reflog, NULL);
513 * Not having any default heads isn't really fatal, but
514 * it does mean that "--unreachable" no longer makes any
515 * sense (since in this case everything will obviously
516 * be unreachable by definition.
518 * Showing dangling objects is valid, though (as those
519 * dangling objects are likely lost heads).
521 * So we just print a warning about it, and clear the
522 * "show_unreachable" flag.
524 if (!default_refs) {
525 error("No default references");
526 show_unreachable = 0;
530 static void fsck_object_dir(const char *path)
532 int i;
533 for (i = 0; i < 256; i++) {
534 static char dir[4096];
535 sprintf(dir, "%s/%02x", path, i);
536 fsck_dir(i, dir);
538 fsck_sha1_list();
541 static int fsck_head_link(void)
543 unsigned char sha1[20];
544 int flag;
545 const char *head_points_at = resolve_ref("HEAD", sha1, 1, &flag);
547 if (!head_points_at || !(flag & REF_ISSYMREF))
548 return error("HEAD is not a symbolic ref");
549 if (prefixcmp(head_points_at, "refs/heads/"))
550 return error("HEAD points to something strange (%s)",
551 head_points_at);
552 if (is_null_sha1(sha1))
553 return error("HEAD: not a valid git pointer");
554 return 0;
557 static int fsck_cache_tree(struct cache_tree *it)
559 int i;
560 int err = 0;
562 if (0 <= it->entry_count) {
563 struct object *obj = parse_object(it->sha1);
564 if (!obj) {
565 error("%s: invalid sha1 pointer in cache-tree",
566 sha1_to_hex(it->sha1));
567 return 1;
569 mark_reachable(obj, REACHABLE);
570 obj->used = 1;
571 if (obj->type != OBJ_TREE)
572 err |= objerror(obj, "non-tree in cache-tree");
574 for (i = 0; i < it->subtree_nr; i++)
575 err |= fsck_cache_tree(it->down[i]->cache_tree);
576 return err;
579 int cmd_fsck(int argc, char **argv, const char *prefix)
581 int i, heads;
583 track_object_refs = 1;
585 for (i = 1; i < argc; i++) {
586 const char *arg = argv[i];
588 if (!strcmp(arg, "--unreachable")) {
589 show_unreachable = 1;
590 continue;
592 if (!strcmp(arg, "--tags")) {
593 show_tags = 1;
594 continue;
596 if (!strcmp(arg, "--root")) {
597 show_root = 1;
598 continue;
600 if (!strcmp(arg, "--cache")) {
601 keep_cache_objects = 1;
602 continue;
604 if (!strcmp(arg, "--full")) {
605 check_full = 1;
606 continue;
608 if (!strcmp(arg, "--strict")) {
609 check_strict = 1;
610 continue;
612 if (*arg == '-')
613 usage("git-fsck [--tags] [--root] [[--unreachable] [--cache] [--full] [--strict] <head-sha1>*]");
616 fsck_head_link();
617 fsck_object_dir(get_object_directory());
618 if (check_full) {
619 struct alternate_object_database *alt;
620 struct packed_git *p;
621 prepare_alt_odb();
622 for (alt = alt_odb_list; alt; alt = alt->next) {
623 char namebuf[PATH_MAX];
624 int namelen = alt->name - alt->base;
625 memcpy(namebuf, alt->base, namelen);
626 namebuf[namelen - 1] = 0;
627 fsck_object_dir(namebuf);
629 prepare_packed_git();
630 for (p = packed_git; p; p = p->next)
631 /* verify gives error messages itself */
632 verify_pack(p, 0);
634 for (p = packed_git; p; p = p->next) {
635 int num = num_packed_objects(p);
636 for (i = 0; i < num; i++) {
637 unsigned char sha1[20];
638 nth_packed_object_sha1(p, i, sha1);
639 fsck_sha1(sha1);
644 heads = 0;
645 for (i = 1; i < argc; i++) {
646 const char *arg = argv[i];
648 if (*arg == '-')
649 continue;
651 if (!get_sha1(arg, head_sha1)) {
652 struct object *obj = lookup_object(head_sha1);
654 /* Error is printed by lookup_object(). */
655 if (!obj)
656 continue;
658 obj->used = 1;
659 mark_reachable(obj, REACHABLE);
660 heads++;
661 continue;
663 error("invalid parameter: expected sha1, got '%s'", arg);
667 * If we've not been given any explicit head information, do the
668 * default ones from .git/refs. We also consider the index file
669 * in this case (ie this implies --cache).
671 if (!heads) {
672 get_default_heads();
673 keep_cache_objects = 1;
676 if (keep_cache_objects) {
677 int i;
678 read_cache();
679 for (i = 0; i < active_nr; i++) {
680 struct blob *blob = lookup_blob(active_cache[i]->sha1);
681 struct object *obj;
682 if (!blob)
683 continue;
684 obj = &blob->object;
685 obj->used = 1;
686 mark_reachable(obj, REACHABLE);
688 if (active_cache_tree)
689 fsck_cache_tree(active_cache_tree);
692 check_connectivity();
693 return 0;