pack: move static state variables
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "config.h"
11 #include "string-list.h"
12 #include "lockfile.h"
13 #include "delta.h"
14 #include "pack.h"
15 #include "blob.h"
16 #include "commit.h"
17 #include "run-command.h"
18 #include "tag.h"
19 #include "tree.h"
20 #include "tree-walk.h"
21 #include "refs.h"
22 #include "pack-revindex.h"
23 #include "sha1-lookup.h"
24 #include "bulk-checkin.h"
25 #include "streaming.h"
26 #include "dir.h"
27 #include "mru.h"
28 #include "list.h"
29 #include "mergesort.h"
30 #include "quote.h"
31 #include "packfile.h"
32
33 #define SZ_FMT PRIuMAX
34 static inline uintmax_t sz_fmt(size_t s) { return s; }
35
36 const unsigned char null_sha1[20];
37 const struct object_id null_oid;
38 const struct object_id empty_tree_oid = {
39 EMPTY_TREE_SHA1_BIN_LITERAL
40 };
41 const struct object_id empty_blob_oid = {
42 EMPTY_BLOB_SHA1_BIN_LITERAL
43 };
44
45 /*
46 * This is meant to hold a *small* number of objects that you would
47 * want read_sha1_file() to be able to return, but yet you do not want
48 * to write them into the object store (e.g. a browse-only
49 * application).
50 */
51 static struct cached_object {
52 unsigned char sha1[20];
53 enum object_type type;
54 void *buf;
55 unsigned long size;
56 } *cached_objects;
57 static int cached_object_nr, cached_object_alloc;
58
59 static struct cached_object empty_tree = {
60 EMPTY_TREE_SHA1_BIN_LITERAL,
61 OBJ_TREE,
62 "",
63 0
64 };
65
66 static struct cached_object *find_cached_object(const unsigned char *sha1)
67 {
68 int i;
69 struct cached_object *co = cached_objects;
70
71 for (i = 0; i < cached_object_nr; i++, co++) {
72 if (!hashcmp(co->sha1, sha1))
73 return co;
74 }
75 if (!hashcmp(sha1, empty_tree.sha1))
76 return &empty_tree;
77 return NULL;
78 }
79
80 int mkdir_in_gitdir(const char *path)
81 {
82 if (mkdir(path, 0777)) {
83 int saved_errno = errno;
84 struct stat st;
85 struct strbuf sb = STRBUF_INIT;
86
87 if (errno != EEXIST)
88 return -1;
89 /*
90 * Are we looking at a path in a symlinked worktree
91 * whose original repository does not yet have it?
92 * e.g. .git/rr-cache pointing at its original
93 * repository in which the user hasn't performed any
94 * conflict resolution yet?
95 */
96 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
97 strbuf_readlink(&sb, path, st.st_size) ||
98 !is_absolute_path(sb.buf) ||
99 mkdir(sb.buf, 0777)) {
100 strbuf_release(&sb);
101 errno = saved_errno;
102 return -1;
103 }
104 strbuf_release(&sb);
105 }
106 return adjust_shared_perm(path);
107 }
108
109 enum scld_error safe_create_leading_directories(char *path)
110 {
111 char *next_component = path + offset_1st_component(path);
112 enum scld_error ret = SCLD_OK;
113
114 while (ret == SCLD_OK && next_component) {
115 struct stat st;
116 char *slash = next_component, slash_character;
117
118 while (*slash && !is_dir_sep(*slash))
119 slash++;
120
121 if (!*slash)
122 break;
123
124 next_component = slash + 1;
125 while (is_dir_sep(*next_component))
126 next_component++;
127 if (!*next_component)
128 break;
129
130 slash_character = *slash;
131 *slash = '\0';
132 if (!stat(path, &st)) {
133 /* path exists */
134 if (!S_ISDIR(st.st_mode)) {
135 errno = ENOTDIR;
136 ret = SCLD_EXISTS;
137 }
138 } else if (mkdir(path, 0777)) {
139 if (errno == EEXIST &&
140 !stat(path, &st) && S_ISDIR(st.st_mode))
141 ; /* somebody created it since we checked */
142 else if (errno == ENOENT)
143 /*
144 * Either mkdir() failed because
145 * somebody just pruned the containing
146 * directory, or stat() failed because
147 * the file that was in our way was
148 * just removed. Either way, inform
149 * the caller that it might be worth
150 * trying again:
151 */
152 ret = SCLD_VANISHED;
153 else
154 ret = SCLD_FAILED;
155 } else if (adjust_shared_perm(path)) {
156 ret = SCLD_PERMS;
157 }
158 *slash = slash_character;
159 }
160 return ret;
161 }
162
163 enum scld_error safe_create_leading_directories_const(const char *path)
164 {
165 int save_errno;
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
168 enum scld_error result = safe_create_leading_directories(buf);
169
170 save_errno = errno;
171 free(buf);
172 errno = save_errno;
173 return result;
174 }
175
176 int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
177 {
178 /*
179 * The number of times we will try to remove empty directories
180 * in the way of path. This is only 1 because if another
181 * process is racily creating directories that conflict with
182 * us, we don't want to fight against them.
183 */
184 int remove_directories_remaining = 1;
185
186 /*
187 * The number of times that we will try to create the
188 * directories containing path. We are willing to attempt this
189 * more than once, because another process could be trying to
190 * clean up empty directories at the same time as we are
191 * trying to create them.
192 */
193 int create_directories_remaining = 3;
194
195 /* A scratch copy of path, filled lazily if we need it: */
196 struct strbuf path_copy = STRBUF_INIT;
197
198 int ret, save_errno;
199
200 /* Sanity check: */
201 assert(*path);
202
203 retry_fn:
204 ret = fn(path, cb);
205 save_errno = errno;
206 if (!ret)
207 goto out;
208
209 if (errno == EISDIR && remove_directories_remaining-- > 0) {
210 /*
211 * A directory is in the way. Maybe it is empty; try
212 * to remove it:
213 */
214 if (!path_copy.len)
215 strbuf_addstr(&path_copy, path);
216
217 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
218 goto retry_fn;
219 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
220 /*
221 * Maybe the containing directory didn't exist, or
222 * maybe it was just deleted by a process that is
223 * racing with us to clean up empty directories. Try
224 * to create it:
225 */
226 enum scld_error scld_result;
227
228 if (!path_copy.len)
229 strbuf_addstr(&path_copy, path);
230
231 do {
232 scld_result = safe_create_leading_directories(path_copy.buf);
233 if (scld_result == SCLD_OK)
234 goto retry_fn;
235 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
236 }
237
238 out:
239 strbuf_release(&path_copy);
240 errno = save_errno;
241 return ret;
242 }
243
244 static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
245 {
246 int i;
247 for (i = 0; i < 20; i++) {
248 static char hex[] = "0123456789abcdef";
249 unsigned int val = sha1[i];
250 strbuf_addch(buf, hex[val >> 4]);
251 strbuf_addch(buf, hex[val & 0xf]);
252 if (!i)
253 strbuf_addch(buf, '/');
254 }
255 }
256
257 const char *sha1_file_name(const unsigned char *sha1)
258 {
259 static struct strbuf buf = STRBUF_INIT;
260
261 strbuf_reset(&buf);
262 strbuf_addf(&buf, "%s/", get_object_directory());
263
264 fill_sha1_path(&buf, sha1);
265 return buf.buf;
266 }
267
268 struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
269 {
270 strbuf_setlen(&alt->scratch, alt->base_len);
271 return &alt->scratch;
272 }
273
274 static const char *alt_sha1_path(struct alternate_object_database *alt,
275 const unsigned char *sha1)
276 {
277 struct strbuf *buf = alt_scratch_buf(alt);
278 fill_sha1_path(buf, sha1);
279 return buf->buf;
280 }
281
282 struct alternate_object_database *alt_odb_list;
283 static struct alternate_object_database **alt_odb_tail;
284
285 /*
286 * Return non-zero iff the path is usable as an alternate object database.
287 */
288 static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
289 {
290 struct alternate_object_database *alt;
291
292 /* Detect cases where alternate disappeared */
293 if (!is_directory(path->buf)) {
294 error("object directory %s does not exist; "
295 "check .git/objects/info/alternates.",
296 path->buf);
297 return 0;
298 }
299
300 /*
301 * Prevent the common mistake of listing the same
302 * thing twice, or object directory itself.
303 */
304 for (alt = alt_odb_list; alt; alt = alt->next) {
305 if (!fspathcmp(path->buf, alt->path))
306 return 0;
307 }
308 if (!fspathcmp(path->buf, normalized_objdir))
309 return 0;
310
311 return 1;
312 }
313
314 /*
315 * Prepare alternate object database registry.
316 *
317 * The variable alt_odb_list points at the list of struct
318 * alternate_object_database. The elements on this list come from
319 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
320 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
321 * whose contents is similar to that environment variable but can be
322 * LF separated. Its base points at a statically allocated buffer that
323 * contains "/the/directory/corresponding/to/.git/objects/...", while
324 * its name points just after the slash at the end of ".git/objects/"
325 * in the example above, and has enough space to hold 40-byte hex
326 * SHA1, an extra slash for the first level indirection, and the
327 * terminating NUL.
328 */
329 static void read_info_alternates(const char * relative_base, int depth);
330 static int link_alt_odb_entry(const char *entry, const char *relative_base,
331 int depth, const char *normalized_objdir)
332 {
333 struct alternate_object_database *ent;
334 struct strbuf pathbuf = STRBUF_INIT;
335
336 if (!is_absolute_path(entry) && relative_base) {
337 strbuf_realpath(&pathbuf, relative_base, 1);
338 strbuf_addch(&pathbuf, '/');
339 }
340 strbuf_addstr(&pathbuf, entry);
341
342 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
343 error("unable to normalize alternate object path: %s",
344 pathbuf.buf);
345 strbuf_release(&pathbuf);
346 return -1;
347 }
348
349 /*
350 * The trailing slash after the directory name is given by
351 * this function at the end. Remove duplicates.
352 */
353 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
354 strbuf_setlen(&pathbuf, pathbuf.len - 1);
355
356 if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
357 strbuf_release(&pathbuf);
358 return -1;
359 }
360
361 ent = alloc_alt_odb(pathbuf.buf);
362
363 /* add the alternate entry */
364 *alt_odb_tail = ent;
365 alt_odb_tail = &(ent->next);
366 ent->next = NULL;
367
368 /* recursively add alternates */
369 read_info_alternates(pathbuf.buf, depth + 1);
370
371 strbuf_release(&pathbuf);
372 return 0;
373 }
374
375 static const char *parse_alt_odb_entry(const char *string,
376 int sep,
377 struct strbuf *out)
378 {
379 const char *end;
380
381 strbuf_reset(out);
382
383 if (*string == '#') {
384 /* comment; consume up to next separator */
385 end = strchrnul(string, sep);
386 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
387 /*
388 * quoted path; unquote_c_style has copied the
389 * data for us and set "end". Broken quoting (e.g.,
390 * an entry that doesn't end with a quote) falls
391 * back to the unquoted case below.
392 */
393 } else {
394 /* normal, unquoted path */
395 end = strchrnul(string, sep);
396 strbuf_add(out, string, end - string);
397 }
398
399 if (*end)
400 end++;
401 return end;
402 }
403
404 static void link_alt_odb_entries(const char *alt, int len, int sep,
405 const char *relative_base, int depth)
406 {
407 struct strbuf objdirbuf = STRBUF_INIT;
408 struct strbuf entry = STRBUF_INIT;
409
410 if (depth > 5) {
411 error("%s: ignoring alternate object stores, nesting too deep.",
412 relative_base);
413 return;
414 }
415
416 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
417 if (strbuf_normalize_path(&objdirbuf) < 0)
418 die("unable to normalize object directory: %s",
419 objdirbuf.buf);
420
421 while (*alt) {
422 alt = parse_alt_odb_entry(alt, sep, &entry);
423 if (!entry.len)
424 continue;
425 link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
426 }
427 strbuf_release(&entry);
428 strbuf_release(&objdirbuf);
429 }
430
431 static void read_info_alternates(const char * relative_base, int depth)
432 {
433 char *map;
434 size_t mapsz;
435 struct stat st;
436 char *path;
437 int fd;
438
439 path = xstrfmt("%s/info/alternates", relative_base);
440 fd = git_open(path);
441 free(path);
442 if (fd < 0)
443 return;
444 if (fstat(fd, &st) || (st.st_size == 0)) {
445 close(fd);
446 return;
447 }
448 mapsz = xsize_t(st.st_size);
449 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
450 close(fd);
451
452 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
453
454 munmap(map, mapsz);
455 }
456
457 struct alternate_object_database *alloc_alt_odb(const char *dir)
458 {
459 struct alternate_object_database *ent;
460
461 FLEX_ALLOC_STR(ent, path, dir);
462 strbuf_init(&ent->scratch, 0);
463 strbuf_addf(&ent->scratch, "%s/", dir);
464 ent->base_len = ent->scratch.len;
465
466 return ent;
467 }
468
469 void add_to_alternates_file(const char *reference)
470 {
471 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
472 char *alts = git_pathdup("objects/info/alternates");
473 FILE *in, *out;
474
475 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
476 out = fdopen_lock_file(lock, "w");
477 if (!out)
478 die_errno("unable to fdopen alternates lockfile");
479
480 in = fopen(alts, "r");
481 if (in) {
482 struct strbuf line = STRBUF_INIT;
483 int found = 0;
484
485 while (strbuf_getline(&line, in) != EOF) {
486 if (!strcmp(reference, line.buf)) {
487 found = 1;
488 break;
489 }
490 fprintf_or_die(out, "%s\n", line.buf);
491 }
492
493 strbuf_release(&line);
494 fclose(in);
495
496 if (found) {
497 rollback_lock_file(lock);
498 lock = NULL;
499 }
500 }
501 else if (errno != ENOENT)
502 die_errno("unable to read alternates file");
503
504 if (lock) {
505 fprintf_or_die(out, "%s\n", reference);
506 if (commit_lock_file(lock))
507 die_errno("unable to move new alternates file into place");
508 if (alt_odb_tail)
509 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
510 }
511 free(alts);
512 }
513
514 void add_to_alternates_memory(const char *reference)
515 {
516 /*
517 * Make sure alternates are initialized, or else our entry may be
518 * overwritten when they are.
519 */
520 prepare_alt_odb();
521
522 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
523 }
524
525 /*
526 * Compute the exact path an alternate is at and returns it. In case of
527 * error NULL is returned and the human readable error is added to `err`
528 * `path` may be relative and should point to $GITDIR.
529 * `err` must not be null.
530 */
531 char *compute_alternate_path(const char *path, struct strbuf *err)
532 {
533 char *ref_git = NULL;
534 const char *repo, *ref_git_s;
535 int seen_error = 0;
536
537 ref_git_s = real_path_if_valid(path);
538 if (!ref_git_s) {
539 seen_error = 1;
540 strbuf_addf(err, _("path '%s' does not exist"), path);
541 goto out;
542 } else
543 /*
544 * Beware: read_gitfile(), real_path() and mkpath()
545 * return static buffer
546 */
547 ref_git = xstrdup(ref_git_s);
548
549 repo = read_gitfile(ref_git);
550 if (!repo)
551 repo = read_gitfile(mkpath("%s/.git", ref_git));
552 if (repo) {
553 free(ref_git);
554 ref_git = xstrdup(repo);
555 }
556
557 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
558 char *ref_git_git = mkpathdup("%s/.git", ref_git);
559 free(ref_git);
560 ref_git = ref_git_git;
561 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
562 struct strbuf sb = STRBUF_INIT;
563 seen_error = 1;
564 if (get_common_dir(&sb, ref_git)) {
565 strbuf_addf(err,
566 _("reference repository '%s' as a linked "
567 "checkout is not supported yet."),
568 path);
569 goto out;
570 }
571
572 strbuf_addf(err, _("reference repository '%s' is not a "
573 "local repository."), path);
574 goto out;
575 }
576
577 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
578 strbuf_addf(err, _("reference repository '%s' is shallow"),
579 path);
580 seen_error = 1;
581 goto out;
582 }
583
584 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
585 strbuf_addf(err,
586 _("reference repository '%s' is grafted"),
587 path);
588 seen_error = 1;
589 goto out;
590 }
591
592 out:
593 if (seen_error) {
594 FREE_AND_NULL(ref_git);
595 }
596
597 return ref_git;
598 }
599
600 int foreach_alt_odb(alt_odb_fn fn, void *cb)
601 {
602 struct alternate_object_database *ent;
603 int r = 0;
604
605 prepare_alt_odb();
606 for (ent = alt_odb_list; ent; ent = ent->next) {
607 r = fn(ent, cb);
608 if (r)
609 break;
610 }
611 return r;
612 }
613
614 void prepare_alt_odb(void)
615 {
616 const char *alt;
617
618 if (alt_odb_tail)
619 return;
620
621 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
622 if (!alt) alt = "";
623
624 alt_odb_tail = &alt_odb_list;
625 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
626
627 read_info_alternates(get_object_directory(), 0);
628 }
629
630 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
631 static int freshen_file(const char *fn)
632 {
633 struct utimbuf t;
634 t.actime = t.modtime = time(NULL);
635 return !utime(fn, &t);
636 }
637
638 /*
639 * All of the check_and_freshen functions return 1 if the file exists and was
640 * freshened (if freshening was requested), 0 otherwise. If they return
641 * 0, you should not assume that it is safe to skip a write of the object (it
642 * either does not exist on disk, or has a stale mtime and may be subject to
643 * pruning).
644 */
645 int check_and_freshen_file(const char *fn, int freshen)
646 {
647 if (access(fn, F_OK))
648 return 0;
649 if (freshen && !freshen_file(fn))
650 return 0;
651 return 1;
652 }
653
654 static int check_and_freshen_local(const unsigned char *sha1, int freshen)
655 {
656 return check_and_freshen_file(sha1_file_name(sha1), freshen);
657 }
658
659 static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
660 {
661 struct alternate_object_database *alt;
662 prepare_alt_odb();
663 for (alt = alt_odb_list; alt; alt = alt->next) {
664 const char *path = alt_sha1_path(alt, sha1);
665 if (check_and_freshen_file(path, freshen))
666 return 1;
667 }
668 return 0;
669 }
670
671 static int check_and_freshen(const unsigned char *sha1, int freshen)
672 {
673 return check_and_freshen_local(sha1, freshen) ||
674 check_and_freshen_nonlocal(sha1, freshen);
675 }
676
677 int has_loose_object_nonlocal(const unsigned char *sha1)
678 {
679 return check_and_freshen_nonlocal(sha1, 0);
680 }
681
682 static int has_loose_object(const unsigned char *sha1)
683 {
684 return check_and_freshen(sha1, 0);
685 }
686
687 void pack_report(void)
688 {
689 fprintf(stderr,
690 "pack_report: getpagesize() = %10" SZ_FMT "\n"
691 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
692 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
693 sz_fmt(getpagesize()),
694 sz_fmt(packed_git_window_size),
695 sz_fmt(packed_git_limit));
696 fprintf(stderr,
697 "pack_report: pack_used_ctr = %10u\n"
698 "pack_report: pack_mmap_calls = %10u\n"
699 "pack_report: pack_open_windows = %10u / %10u\n"
700 "pack_report: pack_mapped = "
701 "%10" SZ_FMT " / %10" SZ_FMT "\n",
702 pack_used_ctr,
703 pack_mmap_calls,
704 pack_open_windows, peak_pack_open_windows,
705 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
706 }
707
708 /*
709 * Open and mmap the index file at path, perform a couple of
710 * consistency checks, then record its information to p. Return 0 on
711 * success.
712 */
713 static int check_packed_git_idx(const char *path, struct packed_git *p)
714 {
715 void *idx_map;
716 struct pack_idx_header *hdr;
717 size_t idx_size;
718 uint32_t version, nr, i, *index;
719 int fd = git_open(path);
720 struct stat st;
721
722 if (fd < 0)
723 return -1;
724 if (fstat(fd, &st)) {
725 close(fd);
726 return -1;
727 }
728 idx_size = xsize_t(st.st_size);
729 if (idx_size < 4 * 256 + 20 + 20) {
730 close(fd);
731 return error("index file %s is too small", path);
732 }
733 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
734 close(fd);
735
736 hdr = idx_map;
737 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
738 version = ntohl(hdr->idx_version);
739 if (version < 2 || version > 2) {
740 munmap(idx_map, idx_size);
741 return error("index file %s is version %"PRIu32
742 " and is not supported by this binary"
743 " (try upgrading GIT to a newer version)",
744 path, version);
745 }
746 } else
747 version = 1;
748
749 nr = 0;
750 index = idx_map;
751 if (version > 1)
752 index += 2; /* skip index header */
753 for (i = 0; i < 256; i++) {
754 uint32_t n = ntohl(index[i]);
755 if (n < nr) {
756 munmap(idx_map, idx_size);
757 return error("non-monotonic index %s", path);
758 }
759 nr = n;
760 }
761
762 if (version == 1) {
763 /*
764 * Total size:
765 * - 256 index entries 4 bytes each
766 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
767 * - 20-byte SHA1 of the packfile
768 * - 20-byte SHA1 file checksum
769 */
770 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
771 munmap(idx_map, idx_size);
772 return error("wrong index v1 file size in %s", path);
773 }
774 } else if (version == 2) {
775 /*
776 * Minimum size:
777 * - 8 bytes of header
778 * - 256 index entries 4 bytes each
779 * - 20-byte sha1 entry * nr
780 * - 4-byte crc entry * nr
781 * - 4-byte offset entry * nr
782 * - 20-byte SHA1 of the packfile
783 * - 20-byte SHA1 file checksum
784 * And after the 4-byte offset table might be a
785 * variable sized table containing 8-byte entries
786 * for offsets larger than 2^31.
787 */
788 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
789 unsigned long max_size = min_size;
790 if (nr)
791 max_size += (nr - 1)*8;
792 if (idx_size < min_size || idx_size > max_size) {
793 munmap(idx_map, idx_size);
794 return error("wrong index v2 file size in %s", path);
795 }
796 if (idx_size != min_size &&
797 /*
798 * make sure we can deal with large pack offsets.
799 * 31-bit signed offset won't be enough, neither
800 * 32-bit unsigned one will be.
801 */
802 (sizeof(off_t) <= 4)) {
803 munmap(idx_map, idx_size);
804 return error("pack too large for current definition of off_t in %s", path);
805 }
806 }
807
808 p->index_version = version;
809 p->index_data = idx_map;
810 p->index_size = idx_size;
811 p->num_objects = nr;
812 return 0;
813 }
814
815 int open_pack_index(struct packed_git *p)
816 {
817 char *idx_name;
818 size_t len;
819 int ret;
820
821 if (p->index_data)
822 return 0;
823
824 if (!strip_suffix(p->pack_name, ".pack", &len))
825 die("BUG: pack_name does not end in .pack");
826 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
827 ret = check_packed_git_idx(idx_name, p);
828 free(idx_name);
829 return ret;
830 }
831
832 static void scan_windows(struct packed_git *p,
833 struct packed_git **lru_p,
834 struct pack_window **lru_w,
835 struct pack_window **lru_l)
836 {
837 struct pack_window *w, *w_l;
838
839 for (w_l = NULL, w = p->windows; w; w = w->next) {
840 if (!w->inuse_cnt) {
841 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
842 *lru_p = p;
843 *lru_w = w;
844 *lru_l = w_l;
845 }
846 }
847 w_l = w;
848 }
849 }
850
851 static int unuse_one_window(struct packed_git *current)
852 {
853 struct packed_git *p, *lru_p = NULL;
854 struct pack_window *lru_w = NULL, *lru_l = NULL;
855
856 if (current)
857 scan_windows(current, &lru_p, &lru_w, &lru_l);
858 for (p = packed_git; p; p = p->next)
859 scan_windows(p, &lru_p, &lru_w, &lru_l);
860 if (lru_p) {
861 munmap(lru_w->base, lru_w->len);
862 pack_mapped -= lru_w->len;
863 if (lru_l)
864 lru_l->next = lru_w->next;
865 else
866 lru_p->windows = lru_w->next;
867 free(lru_w);
868 pack_open_windows--;
869 return 1;
870 }
871 return 0;
872 }
873
874 void release_pack_memory(size_t need)
875 {
876 size_t cur = pack_mapped;
877 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
878 ; /* nothing */
879 }
880
881 static void mmap_limit_check(size_t length)
882 {
883 static size_t limit = 0;
884 if (!limit) {
885 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
886 if (!limit)
887 limit = SIZE_MAX;
888 }
889 if (length > limit)
890 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
891 (uintmax_t)length, (uintmax_t)limit);
892 }
893
894 void *xmmap_gently(void *start, size_t length,
895 int prot, int flags, int fd, off_t offset)
896 {
897 void *ret;
898
899 mmap_limit_check(length);
900 ret = mmap(start, length, prot, flags, fd, offset);
901 if (ret == MAP_FAILED) {
902 if (!length)
903 return NULL;
904 release_pack_memory(length);
905 ret = mmap(start, length, prot, flags, fd, offset);
906 }
907 return ret;
908 }
909
910 void *xmmap(void *start, size_t length,
911 int prot, int flags, int fd, off_t offset)
912 {
913 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
914 if (ret == MAP_FAILED)
915 die_errno("mmap failed");
916 return ret;
917 }
918
919 void close_pack_windows(struct packed_git *p)
920 {
921 while (p->windows) {
922 struct pack_window *w = p->windows;
923
924 if (w->inuse_cnt)
925 die("pack '%s' still has open windows to it",
926 p->pack_name);
927 munmap(w->base, w->len);
928 pack_mapped -= w->len;
929 pack_open_windows--;
930 p->windows = w->next;
931 free(w);
932 }
933 }
934
935 static int close_pack_fd(struct packed_git *p)
936 {
937 if (p->pack_fd < 0)
938 return 0;
939
940 close(p->pack_fd);
941 pack_open_fds--;
942 p->pack_fd = -1;
943
944 return 1;
945 }
946
947 static void close_pack(struct packed_git *p)
948 {
949 close_pack_windows(p);
950 close_pack_fd(p);
951 close_pack_index(p);
952 }
953
954 void close_all_packs(void)
955 {
956 struct packed_git *p;
957
958 for (p = packed_git; p; p = p->next)
959 if (p->do_not_close)
960 die("BUG: want to close pack marked 'do-not-close'");
961 else
962 close_pack(p);
963 }
964
965
966 /*
967 * The LRU pack is the one with the oldest MRU window, preferring packs
968 * with no used windows, or the oldest mtime if it has no windows allocated.
969 */
970 static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
971 {
972 struct pack_window *w, *this_mru_w;
973 int has_windows_inuse = 0;
974
975 /*
976 * Reject this pack if it has windows and the previously selected
977 * one does not. If this pack does not have windows, reject
978 * it if the pack file is newer than the previously selected one.
979 */
980 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
981 return;
982
983 for (w = this_mru_w = p->windows; w; w = w->next) {
984 /*
985 * Reject this pack if any of its windows are in use,
986 * but the previously selected pack did not have any
987 * inuse windows. Otherwise, record that this pack
988 * has windows in use.
989 */
990 if (w->inuse_cnt) {
991 if (*accept_windows_inuse)
992 has_windows_inuse = 1;
993 else
994 return;
995 }
996
997 if (w->last_used > this_mru_w->last_used)
998 this_mru_w = w;
999
1000 /*
1001 * Reject this pack if it has windows that have been
1002 * used more recently than the previously selected pack.
1003 * If the previously selected pack had windows inuse and
1004 * we have not encountered a window in this pack that is
1005 * inuse, skip this check since we prefer a pack with no
1006 * inuse windows to one that has inuse windows.
1007 */
1008 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
1009 this_mru_w->last_used > (*mru_w)->last_used)
1010 return;
1011 }
1012
1013 /*
1014 * Select this pack.
1015 */
1016 *mru_w = this_mru_w;
1017 *lru_p = p;
1018 *accept_windows_inuse = has_windows_inuse;
1019 }
1020
1021 static int close_one_pack(void)
1022 {
1023 struct packed_git *p, *lru_p = NULL;
1024 struct pack_window *mru_w = NULL;
1025 int accept_windows_inuse = 1;
1026
1027 for (p = packed_git; p; p = p->next) {
1028 if (p->pack_fd == -1)
1029 continue;
1030 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
1031 }
1032
1033 if (lru_p)
1034 return close_pack_fd(lru_p);
1035
1036 return 0;
1037 }
1038
1039 void unuse_pack(struct pack_window **w_cursor)
1040 {
1041 struct pack_window *w = *w_cursor;
1042 if (w) {
1043 w->inuse_cnt--;
1044 *w_cursor = NULL;
1045 }
1046 }
1047
1048 void close_pack_index(struct packed_git *p)
1049 {
1050 if (p->index_data) {
1051 munmap((void *)p->index_data, p->index_size);
1052 p->index_data = NULL;
1053 }
1054 }
1055
1056 static unsigned int get_max_fd_limit(void)
1057 {
1058 #ifdef RLIMIT_NOFILE
1059 {
1060 struct rlimit lim;
1061
1062 if (!getrlimit(RLIMIT_NOFILE, &lim))
1063 return lim.rlim_cur;
1064 }
1065 #endif
1066
1067 #ifdef _SC_OPEN_MAX
1068 {
1069 long open_max = sysconf(_SC_OPEN_MAX);
1070 if (0 < open_max)
1071 return open_max;
1072 /*
1073 * Otherwise, we got -1 for one of the two
1074 * reasons:
1075 *
1076 * (1) sysconf() did not understand _SC_OPEN_MAX
1077 * and signaled an error with -1; or
1078 * (2) sysconf() said there is no limit.
1079 *
1080 * We _could_ clear errno before calling sysconf() to
1081 * tell these two cases apart and return a huge number
1082 * in the latter case to let the caller cap it to a
1083 * value that is not so selfish, but letting the
1084 * fallback OPEN_MAX codepath take care of these cases
1085 * is a lot simpler.
1086 */
1087 }
1088 #endif
1089
1090 #ifdef OPEN_MAX
1091 return OPEN_MAX;
1092 #else
1093 return 1; /* see the caller ;-) */
1094 #endif
1095 }
1096
1097 /*
1098 * Do not call this directly as this leaks p->pack_fd on error return;
1099 * call open_packed_git() instead.
1100 */
1101 static int open_packed_git_1(struct packed_git *p)
1102 {
1103 struct stat st;
1104 struct pack_header hdr;
1105 unsigned char sha1[20];
1106 unsigned char *idx_sha1;
1107 long fd_flag;
1108
1109 if (!p->index_data && open_pack_index(p))
1110 return error("packfile %s index unavailable", p->pack_name);
1111
1112 if (!pack_max_fds) {
1113 unsigned int max_fds = get_max_fd_limit();
1114
1115 /* Save 3 for stdin/stdout/stderr, 22 for work */
1116 if (25 < max_fds)
1117 pack_max_fds = max_fds - 25;
1118 else
1119 pack_max_fds = 1;
1120 }
1121
1122 while (pack_max_fds <= pack_open_fds && close_one_pack())
1123 ; /* nothing */
1124
1125 p->pack_fd = git_open(p->pack_name);
1126 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
1127 return -1;
1128 pack_open_fds++;
1129
1130 /* If we created the struct before we had the pack we lack size. */
1131 if (!p->pack_size) {
1132 if (!S_ISREG(st.st_mode))
1133 return error("packfile %s not a regular file", p->pack_name);
1134 p->pack_size = st.st_size;
1135 } else if (p->pack_size != st.st_size)
1136 return error("packfile %s size changed", p->pack_name);
1137
1138 /* We leave these file descriptors open with sliding mmap;
1139 * there is no point keeping them open across exec(), though.
1140 */
1141 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
1142 if (fd_flag < 0)
1143 return error("cannot determine file descriptor flags");
1144 fd_flag |= FD_CLOEXEC;
1145 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
1146 return error("cannot set FD_CLOEXEC");
1147
1148 /* Verify we recognize this pack file format. */
1149 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
1150 return error("file %s is far too short to be a packfile", p->pack_name);
1151 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
1152 return error("file %s is not a GIT packfile", p->pack_name);
1153 if (!pack_version_ok(hdr.hdr_version))
1154 return error("packfile %s is version %"PRIu32" and not"
1155 " supported (try upgrading GIT to a newer version)",
1156 p->pack_name, ntohl(hdr.hdr_version));
1157
1158 /* Verify the pack matches its index. */
1159 if (p->num_objects != ntohl(hdr.hdr_entries))
1160 return error("packfile %s claims to have %"PRIu32" objects"
1161 " while index indicates %"PRIu32" objects",
1162 p->pack_name, ntohl(hdr.hdr_entries),
1163 p->num_objects);
1164 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
1165 return error("end of packfile %s is unavailable", p->pack_name);
1166 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
1167 return error("packfile %s signature is unavailable", p->pack_name);
1168 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
1169 if (hashcmp(sha1, idx_sha1))
1170 return error("packfile %s does not match index", p->pack_name);
1171 return 0;
1172 }
1173
1174 static int open_packed_git(struct packed_git *p)
1175 {
1176 if (!open_packed_git_1(p))
1177 return 0;
1178 close_pack_fd(p);
1179 return -1;
1180 }
1181
1182 static int in_window(struct pack_window *win, off_t offset)
1183 {
1184 /* We must promise at least 20 bytes (one hash) after the
1185 * offset is available from this window, otherwise the offset
1186 * is not actually in this window and a different window (which
1187 * has that one hash excess) must be used. This is to support
1188 * the object header and delta base parsing routines below.
1189 */
1190 off_t win_off = win->offset;
1191 return win_off <= offset
1192 && (offset + 20) <= (win_off + win->len);
1193 }
1194
1195 unsigned char *use_pack(struct packed_git *p,
1196 struct pack_window **w_cursor,
1197 off_t offset,
1198 unsigned long *left)
1199 {
1200 struct pack_window *win = *w_cursor;
1201
1202 /* Since packfiles end in a hash of their content and it's
1203 * pointless to ask for an offset into the middle of that
1204 * hash, and the in_window function above wouldn't match
1205 * don't allow an offset too close to the end of the file.
1206 */
1207 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1208 die("packfile %s cannot be accessed", p->pack_name);
1209 if (offset > (p->pack_size - 20))
1210 die("offset beyond end of packfile (truncated pack?)");
1211 if (offset < 0)
1212 die(_("offset before end of packfile (broken .idx?)"));
1213
1214 if (!win || !in_window(win, offset)) {
1215 if (win)
1216 win->inuse_cnt--;
1217 for (win = p->windows; win; win = win->next) {
1218 if (in_window(win, offset))
1219 break;
1220 }
1221 if (!win) {
1222 size_t window_align = packed_git_window_size / 2;
1223 off_t len;
1224
1225 if (p->pack_fd == -1 && open_packed_git(p))
1226 die("packfile %s cannot be accessed", p->pack_name);
1227
1228 win = xcalloc(1, sizeof(*win));
1229 win->offset = (offset / window_align) * window_align;
1230 len = p->pack_size - win->offset;
1231 if (len > packed_git_window_size)
1232 len = packed_git_window_size;
1233 win->len = (size_t)len;
1234 pack_mapped += win->len;
1235 while (packed_git_limit < pack_mapped
1236 && unuse_one_window(p))
1237 ; /* nothing */
1238 win->base = xmmap(NULL, win->len,
1239 PROT_READ, MAP_PRIVATE,
1240 p->pack_fd, win->offset);
1241 if (win->base == MAP_FAILED)
1242 die_errno("packfile %s cannot be mapped",
1243 p->pack_name);
1244 if (!win->offset && win->len == p->pack_size
1245 && !p->do_not_close)
1246 close_pack_fd(p);
1247 pack_mmap_calls++;
1248 pack_open_windows++;
1249 if (pack_mapped > peak_pack_mapped)
1250 peak_pack_mapped = pack_mapped;
1251 if (pack_open_windows > peak_pack_open_windows)
1252 peak_pack_open_windows = pack_open_windows;
1253 win->next = p->windows;
1254 p->windows = win;
1255 }
1256 }
1257 if (win != *w_cursor) {
1258 win->last_used = pack_used_ctr++;
1259 win->inuse_cnt++;
1260 *w_cursor = win;
1261 }
1262 offset -= win->offset;
1263 if (left)
1264 *left = win->len - xsize_t(offset);
1265 return win->base + offset;
1266 }
1267
1268 static struct packed_git *alloc_packed_git(int extra)
1269 {
1270 struct packed_git *p = xmalloc(st_add(sizeof(*p), extra));
1271 memset(p, 0, sizeof(*p));
1272 p->pack_fd = -1;
1273 return p;
1274 }
1275
1276 static void try_to_free_pack_memory(size_t size)
1277 {
1278 release_pack_memory(size);
1279 }
1280
1281 struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1282 {
1283 static int have_set_try_to_free_routine;
1284 struct stat st;
1285 size_t alloc;
1286 struct packed_git *p;
1287
1288 if (!have_set_try_to_free_routine) {
1289 have_set_try_to_free_routine = 1;
1290 set_try_to_free_routine(try_to_free_pack_memory);
1291 }
1292
1293 /*
1294 * Make sure a corresponding .pack file exists and that
1295 * the index looks sane.
1296 */
1297 if (!strip_suffix_mem(path, &path_len, ".idx"))
1298 return NULL;
1299
1300 /*
1301 * ".pack" is long enough to hold any suffix we're adding (and
1302 * the use xsnprintf double-checks that)
1303 */
1304 alloc = st_add3(path_len, strlen(".pack"), 1);
1305 p = alloc_packed_git(alloc);
1306 memcpy(p->pack_name, path, path_len);
1307
1308 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
1309 if (!access(p->pack_name, F_OK))
1310 p->pack_keep = 1;
1311
1312 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
1313 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
1314 free(p);
1315 return NULL;
1316 }
1317
1318 /* ok, it looks sane as far as we can check without
1319 * actually mapping the pack file.
1320 */
1321 p->pack_size = st.st_size;
1322 p->pack_local = local;
1323 p->mtime = st.st_mtime;
1324 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1325 hashclr(p->sha1);
1326 return p;
1327 }
1328
1329 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
1330 {
1331 const char *path = sha1_pack_name(sha1);
1332 size_t alloc = st_add(strlen(path), 1);
1333 struct packed_git *p = alloc_packed_git(alloc);
1334
1335 memcpy(p->pack_name, path, alloc); /* includes NUL */
1336 hashcpy(p->sha1, sha1);
1337 if (check_packed_git_idx(idx_path, p)) {
1338 free(p);
1339 return NULL;
1340 }
1341
1342 return p;
1343 }
1344
1345 void install_packed_git(struct packed_git *pack)
1346 {
1347 if (pack->pack_fd != -1)
1348 pack_open_fds++;
1349
1350 pack->next = packed_git;
1351 packed_git = pack;
1352 }
1353
1354 void (*report_garbage)(unsigned seen_bits, const char *path);
1355
1356 static void report_helper(const struct string_list *list,
1357 int seen_bits, int first, int last)
1358 {
1359 if (seen_bits == (PACKDIR_FILE_PACK|PACKDIR_FILE_IDX))
1360 return;
1361
1362 for (; first < last; first++)
1363 report_garbage(seen_bits, list->items[first].string);
1364 }
1365
1366 static void report_pack_garbage(struct string_list *list)
1367 {
1368 int i, baselen = -1, first = 0, seen_bits = 0;
1369
1370 if (!report_garbage)
1371 return;
1372
1373 string_list_sort(list);
1374
1375 for (i = 0; i < list->nr; i++) {
1376 const char *path = list->items[i].string;
1377 if (baselen != -1 &&
1378 strncmp(path, list->items[first].string, baselen)) {
1379 report_helper(list, seen_bits, first, i);
1380 baselen = -1;
1381 seen_bits = 0;
1382 }
1383 if (baselen == -1) {
1384 const char *dot = strrchr(path, '.');
1385 if (!dot) {
1386 report_garbage(PACKDIR_FILE_GARBAGE, path);
1387 continue;
1388 }
1389 baselen = dot - path + 1;
1390 first = i;
1391 }
1392 if (!strcmp(path + baselen, "pack"))
1393 seen_bits |= 1;
1394 else if (!strcmp(path + baselen, "idx"))
1395 seen_bits |= 2;
1396 }
1397 report_helper(list, seen_bits, first, list->nr);
1398 }
1399
1400 static void prepare_packed_git_one(char *objdir, int local)
1401 {
1402 struct strbuf path = STRBUF_INIT;
1403 size_t dirnamelen;
1404 DIR *dir;
1405 struct dirent *de;
1406 struct string_list garbage = STRING_LIST_INIT_DUP;
1407
1408 strbuf_addstr(&path, objdir);
1409 strbuf_addstr(&path, "/pack");
1410 dir = opendir(path.buf);
1411 if (!dir) {
1412 if (errno != ENOENT)
1413 error_errno("unable to open object pack directory: %s",
1414 path.buf);
1415 strbuf_release(&path);
1416 return;
1417 }
1418 strbuf_addch(&path, '/');
1419 dirnamelen = path.len;
1420 while ((de = readdir(dir)) != NULL) {
1421 struct packed_git *p;
1422 size_t base_len;
1423
1424 if (is_dot_or_dotdot(de->d_name))
1425 continue;
1426
1427 strbuf_setlen(&path, dirnamelen);
1428 strbuf_addstr(&path, de->d_name);
1429
1430 base_len = path.len;
1431 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
1432 /* Don't reopen a pack we already have. */
1433 for (p = packed_git; p; p = p->next) {
1434 size_t len;
1435 if (strip_suffix(p->pack_name, ".pack", &len) &&
1436 len == base_len &&
1437 !memcmp(p->pack_name, path.buf, len))
1438 break;
1439 }
1440 if (p == NULL &&
1441 /*
1442 * See if it really is a valid .idx file with
1443 * corresponding .pack file that we can map.
1444 */
1445 (p = add_packed_git(path.buf, path.len, local)) != NULL)
1446 install_packed_git(p);
1447 }
1448
1449 if (!report_garbage)
1450 continue;
1451
1452 if (ends_with(de->d_name, ".idx") ||
1453 ends_with(de->d_name, ".pack") ||
1454 ends_with(de->d_name, ".bitmap") ||
1455 ends_with(de->d_name, ".keep"))
1456 string_list_append(&garbage, path.buf);
1457 else
1458 report_garbage(PACKDIR_FILE_GARBAGE, path.buf);
1459 }
1460 closedir(dir);
1461 report_pack_garbage(&garbage);
1462 string_list_clear(&garbage, 0);
1463 strbuf_release(&path);
1464 }
1465
1466 static int approximate_object_count_valid;
1467
1468 /*
1469 * Give a fast, rough count of the number of objects in the repository. This
1470 * ignores loose objects completely. If you have a lot of them, then either
1471 * you should repack because your performance will be awful, or they are
1472 * all unreachable objects about to be pruned, in which case they're not really
1473 * interesting as a measure of repo size in the first place.
1474 */
1475 unsigned long approximate_object_count(void)
1476 {
1477 static unsigned long count;
1478 if (!approximate_object_count_valid) {
1479 struct packed_git *p;
1480
1481 prepare_packed_git();
1482 count = 0;
1483 for (p = packed_git; p; p = p->next) {
1484 if (open_pack_index(p))
1485 continue;
1486 count += p->num_objects;
1487 }
1488 }
1489 return count;
1490 }
1491
1492 static void *get_next_packed_git(const void *p)
1493 {
1494 return ((const struct packed_git *)p)->next;
1495 }
1496
1497 static void set_next_packed_git(void *p, void *next)
1498 {
1499 ((struct packed_git *)p)->next = next;
1500 }
1501
1502 static int sort_pack(const void *a_, const void *b_)
1503 {
1504 const struct packed_git *a = a_;
1505 const struct packed_git *b = b_;
1506 int st;
1507
1508 /*
1509 * Local packs tend to contain objects specific to our
1510 * variant of the project than remote ones. In addition,
1511 * remote ones could be on a network mounted filesystem.
1512 * Favor local ones for these reasons.
1513 */
1514 st = a->pack_local - b->pack_local;
1515 if (st)
1516 return -st;
1517
1518 /*
1519 * Younger packs tend to contain more recent objects,
1520 * and more recent objects tend to get accessed more
1521 * often.
1522 */
1523 if (a->mtime < b->mtime)
1524 return 1;
1525 else if (a->mtime == b->mtime)
1526 return 0;
1527 return -1;
1528 }
1529
1530 static void rearrange_packed_git(void)
1531 {
1532 packed_git = llist_mergesort(packed_git, get_next_packed_git,
1533 set_next_packed_git, sort_pack);
1534 }
1535
1536 static void prepare_packed_git_mru(void)
1537 {
1538 struct packed_git *p;
1539
1540 mru_clear(packed_git_mru);
1541 for (p = packed_git; p; p = p->next)
1542 mru_append(packed_git_mru, p);
1543 }
1544
1545 static int prepare_packed_git_run_once = 0;
1546 void prepare_packed_git(void)
1547 {
1548 struct alternate_object_database *alt;
1549
1550 if (prepare_packed_git_run_once)
1551 return;
1552 prepare_packed_git_one(get_object_directory(), 1);
1553 prepare_alt_odb();
1554 for (alt = alt_odb_list; alt; alt = alt->next)
1555 prepare_packed_git_one(alt->path, 0);
1556 rearrange_packed_git();
1557 prepare_packed_git_mru();
1558 prepare_packed_git_run_once = 1;
1559 }
1560
1561 void reprepare_packed_git(void)
1562 {
1563 approximate_object_count_valid = 0;
1564 prepare_packed_git_run_once = 0;
1565 prepare_packed_git();
1566 }
1567
1568 static void mark_bad_packed_object(struct packed_git *p,
1569 const unsigned char *sha1)
1570 {
1571 unsigned i;
1572 for (i = 0; i < p->num_bad_objects; i++)
1573 if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
1574 return;
1575 p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
1576 st_mult(GIT_MAX_RAWSZ,
1577 st_add(p->num_bad_objects, 1)));
1578 hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
1579 p->num_bad_objects++;
1580 }
1581
1582 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1583 {
1584 struct packed_git *p;
1585 unsigned i;
1586
1587 for (p = packed_git; p; p = p->next)
1588 for (i = 0; i < p->num_bad_objects; i++)
1589 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1590 return p;
1591 return NULL;
1592 }
1593
1594 /*
1595 * With an in-core object data in "map", rehash it to make sure the
1596 * object name actually matches "sha1" to detect object corruption.
1597 * With "map" == NULL, try reading the object named with "sha1" using
1598 * the streaming interface and rehash it to do the same.
1599 */
1600 int check_sha1_signature(const unsigned char *sha1, void *map,
1601 unsigned long size, const char *type)
1602 {
1603 unsigned char real_sha1[20];
1604 enum object_type obj_type;
1605 struct git_istream *st;
1606 git_SHA_CTX c;
1607 char hdr[32];
1608 int hdrlen;
1609
1610 if (map) {
1611 hash_sha1_file(map, size, type, real_sha1);
1612 return hashcmp(sha1, real_sha1) ? -1 : 0;
1613 }
1614
1615 st = open_istream(sha1, &obj_type, &size, NULL);
1616 if (!st)
1617 return -1;
1618
1619 /* Generate the header */
1620 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
1621
1622 /* Sha1.. */
1623 git_SHA1_Init(&c);
1624 git_SHA1_Update(&c, hdr, hdrlen);
1625 for (;;) {
1626 char buf[1024 * 16];
1627 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1628
1629 if (readlen < 0) {
1630 close_istream(st);
1631 return -1;
1632 }
1633 if (!readlen)
1634 break;
1635 git_SHA1_Update(&c, buf, readlen);
1636 }
1637 git_SHA1_Final(real_sha1, &c);
1638 close_istream(st);
1639 return hashcmp(sha1, real_sha1) ? -1 : 0;
1640 }
1641
1642 int git_open_cloexec(const char *name, int flags)
1643 {
1644 int fd;
1645 static int o_cloexec = O_CLOEXEC;
1646
1647 fd = open(name, flags | o_cloexec);
1648 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1649 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1650 o_cloexec &= ~O_CLOEXEC;
1651 fd = open(name, flags | o_cloexec);
1652 }
1653
1654 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1655 {
1656 static int fd_cloexec = FD_CLOEXEC;
1657
1658 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1659 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1660 int flags = fcntl(fd, F_GETFD);
1661 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1662 fd_cloexec = 0;
1663 }
1664 }
1665 #endif
1666 return fd;
1667 }
1668
1669 /*
1670 * Find "sha1" as a loose object in the local repository or in an alternate.
1671 * Returns 0 on success, negative on failure.
1672 *
1673 * The "path" out-parameter will give the path of the object we found (if any).
1674 * Note that it may point to static storage and is only valid until another
1675 * call to sha1_file_name(), etc.
1676 */
1677 static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
1678 const char **path)
1679 {
1680 struct alternate_object_database *alt;
1681
1682 *path = sha1_file_name(sha1);
1683 if (!lstat(*path, st))
1684 return 0;
1685
1686 prepare_alt_odb();
1687 errno = ENOENT;
1688 for (alt = alt_odb_list; alt; alt = alt->next) {
1689 *path = alt_sha1_path(alt, sha1);
1690 if (!lstat(*path, st))
1691 return 0;
1692 }
1693
1694 return -1;
1695 }
1696
1697 /*
1698 * Like stat_sha1_file(), but actually open the object and return the
1699 * descriptor. See the caveats on the "path" parameter above.
1700 */
1701 static int open_sha1_file(const unsigned char *sha1, const char **path)
1702 {
1703 int fd;
1704 struct alternate_object_database *alt;
1705 int most_interesting_errno;
1706
1707 *path = sha1_file_name(sha1);
1708 fd = git_open(*path);
1709 if (fd >= 0)
1710 return fd;
1711 most_interesting_errno = errno;
1712
1713 prepare_alt_odb();
1714 for (alt = alt_odb_list; alt; alt = alt->next) {
1715 *path = alt_sha1_path(alt, sha1);
1716 fd = git_open(*path);
1717 if (fd >= 0)
1718 return fd;
1719 if (most_interesting_errno == ENOENT)
1720 most_interesting_errno = errno;
1721 }
1722 errno = most_interesting_errno;
1723 return -1;
1724 }
1725
1726 /*
1727 * Map the loose object at "path" if it is not NULL, or the path found by
1728 * searching for a loose object named "sha1".
1729 */
1730 static void *map_sha1_file_1(const char *path,
1731 const unsigned char *sha1,
1732 unsigned long *size)
1733 {
1734 void *map;
1735 int fd;
1736
1737 if (path)
1738 fd = git_open(path);
1739 else
1740 fd = open_sha1_file(sha1, &path);
1741 map = NULL;
1742 if (fd >= 0) {
1743 struct stat st;
1744
1745 if (!fstat(fd, &st)) {
1746 *size = xsize_t(st.st_size);
1747 if (!*size) {
1748 /* mmap() is forbidden on empty files */
1749 error("object file %s is empty", path);
1750 return NULL;
1751 }
1752 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1753 }
1754 close(fd);
1755 }
1756 return map;
1757 }
1758
1759 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1760 {
1761 return map_sha1_file_1(NULL, sha1, size);
1762 }
1763
1764 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1765 unsigned long len, enum object_type *type, unsigned long *sizep)
1766 {
1767 unsigned shift;
1768 unsigned long size, c;
1769 unsigned long used = 0;
1770
1771 c = buf[used++];
1772 *type = (c >> 4) & 7;
1773 size = c & 15;
1774 shift = 4;
1775 while (c & 0x80) {
1776 if (len <= used || bitsizeof(long) <= shift) {
1777 error("bad object header");
1778 size = used = 0;
1779 break;
1780 }
1781 c = buf[used++];
1782 size += (c & 0x7f) << shift;
1783 shift += 7;
1784 }
1785 *sizep = size;
1786 return used;
1787 }
1788
1789 static int unpack_sha1_short_header(git_zstream *stream,
1790 unsigned char *map, unsigned long mapsize,
1791 void *buffer, unsigned long bufsiz)
1792 {
1793 /* Get the data stream */
1794 memset(stream, 0, sizeof(*stream));
1795 stream->next_in = map;
1796 stream->avail_in = mapsize;
1797 stream->next_out = buffer;
1798 stream->avail_out = bufsiz;
1799
1800 git_inflate_init(stream);
1801 return git_inflate(stream, 0);
1802 }
1803
1804 int unpack_sha1_header(git_zstream *stream,
1805 unsigned char *map, unsigned long mapsize,
1806 void *buffer, unsigned long bufsiz)
1807 {
1808 int status = unpack_sha1_short_header(stream, map, mapsize,
1809 buffer, bufsiz);
1810
1811 if (status < Z_OK)
1812 return status;
1813
1814 /* Make sure we have the terminating NUL */
1815 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1816 return -1;
1817 return 0;
1818 }
1819
1820 static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1821 unsigned long mapsize, void *buffer,
1822 unsigned long bufsiz, struct strbuf *header)
1823 {
1824 int status;
1825
1826 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1827 if (status < Z_OK)
1828 return -1;
1829
1830 /*
1831 * Check if entire header is unpacked in the first iteration.
1832 */
1833 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1834 return 0;
1835
1836 /*
1837 * buffer[0..bufsiz] was not large enough. Copy the partial
1838 * result out to header, and then append the result of further
1839 * reading the stream.
1840 */
1841 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1842 stream->next_out = buffer;
1843 stream->avail_out = bufsiz;
1844
1845 do {
1846 status = git_inflate(stream, 0);
1847 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1848 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1849 return 0;
1850 stream->next_out = buffer;
1851 stream->avail_out = bufsiz;
1852 } while (status != Z_STREAM_END);
1853 return -1;
1854 }
1855
1856 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1857 {
1858 int bytes = strlen(buffer) + 1;
1859 unsigned char *buf = xmallocz(size);
1860 unsigned long n;
1861 int status = Z_OK;
1862
1863 n = stream->total_out - bytes;
1864 if (n > size)
1865 n = size;
1866 memcpy(buf, (char *) buffer + bytes, n);
1867 bytes = n;
1868 if (bytes <= size) {
1869 /*
1870 * The above condition must be (bytes <= size), not
1871 * (bytes < size). In other words, even though we
1872 * expect no more output and set avail_out to zero,
1873 * the input zlib stream may have bytes that express
1874 * "this concludes the stream", and we *do* want to
1875 * eat that input.
1876 *
1877 * Otherwise we would not be able to test that we
1878 * consumed all the input to reach the expected size;
1879 * we also want to check that zlib tells us that all
1880 * went well with status == Z_STREAM_END at the end.
1881 */
1882 stream->next_out = buf + bytes;
1883 stream->avail_out = size - bytes;
1884 while (status == Z_OK)
1885 status = git_inflate(stream, Z_FINISH);
1886 }
1887 if (status == Z_STREAM_END && !stream->avail_in) {
1888 git_inflate_end(stream);
1889 return buf;
1890 }
1891
1892 if (status < 0)
1893 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1894 else if (stream->avail_in)
1895 error("garbage at end of loose object '%s'",
1896 sha1_to_hex(sha1));
1897 free(buf);
1898 return NULL;
1899 }
1900
1901 /*
1902 * We used to just use "sscanf()", but that's actually way
1903 * too permissive for what we want to check. So do an anal
1904 * object header parse by hand.
1905 */
1906 static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1907 unsigned int flags)
1908 {
1909 const char *type_buf = hdr;
1910 unsigned long size;
1911 int type, type_len = 0;
1912
1913 /*
1914 * The type can be of any size but is followed by
1915 * a space.
1916 */
1917 for (;;) {
1918 char c = *hdr++;
1919 if (!c)
1920 return -1;
1921 if (c == ' ')
1922 break;
1923 type_len++;
1924 }
1925
1926 type = type_from_string_gently(type_buf, type_len, 1);
1927 if (oi->typename)
1928 strbuf_add(oi->typename, type_buf, type_len);
1929 /*
1930 * Set type to 0 if its an unknown object and
1931 * we're obtaining the type using '--allow-unknown-type'
1932 * option.
1933 */
1934 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1935 type = 0;
1936 else if (type < 0)
1937 die("invalid object type");
1938 if (oi->typep)
1939 *oi->typep = type;
1940
1941 /*
1942 * The length must follow immediately, and be in canonical
1943 * decimal format (ie "010" is not valid).
1944 */
1945 size = *hdr++ - '0';
1946 if (size > 9)
1947 return -1;
1948 if (size) {
1949 for (;;) {
1950 unsigned long c = *hdr - '0';
1951 if (c > 9)
1952 break;
1953 hdr++;
1954 size = size * 10 + c;
1955 }
1956 }
1957
1958 if (oi->sizep)
1959 *oi->sizep = size;
1960
1961 /*
1962 * The length must be followed by a zero byte
1963 */
1964 return *hdr ? -1 : type;
1965 }
1966
1967 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1968 {
1969 struct object_info oi = OBJECT_INFO_INIT;
1970
1971 oi.sizep = sizep;
1972 return parse_sha1_header_extended(hdr, &oi, 0);
1973 }
1974
1975 unsigned long get_size_from_delta(struct packed_git *p,
1976 struct pack_window **w_curs,
1977 off_t curpos)
1978 {
1979 const unsigned char *data;
1980 unsigned char delta_head[20], *in;
1981 git_zstream stream;
1982 int st;
1983
1984 memset(&stream, 0, sizeof(stream));
1985 stream.next_out = delta_head;
1986 stream.avail_out = sizeof(delta_head);
1987
1988 git_inflate_init(&stream);
1989 do {
1990 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1991 stream.next_in = in;
1992 st = git_inflate(&stream, Z_FINISH);
1993 curpos += stream.next_in - in;
1994 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1995 stream.total_out < sizeof(delta_head));
1996 git_inflate_end(&stream);
1997 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1998 error("delta data unpack-initial failed");
1999 return 0;
2000 }
2001
2002 /* Examine the initial part of the delta to figure out
2003 * the result size.
2004 */
2005 data = delta_head;
2006
2007 /* ignore base size */
2008 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2009
2010 /* Read the result size */
2011 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
2012 }
2013
2014 static off_t get_delta_base(struct packed_git *p,
2015 struct pack_window **w_curs,
2016 off_t *curpos,
2017 enum object_type type,
2018 off_t delta_obj_offset)
2019 {
2020 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
2021 off_t base_offset;
2022
2023 /* use_pack() assured us we have [base_info, base_info + 20)
2024 * as a range that we can look at without walking off the
2025 * end of the mapped window. Its actually the hash size
2026 * that is assured. An OFS_DELTA longer than the hash size
2027 * is stupid, as then a REF_DELTA would be smaller to store.
2028 */
2029 if (type == OBJ_OFS_DELTA) {
2030 unsigned used = 0;
2031 unsigned char c = base_info[used++];
2032 base_offset = c & 127;
2033 while (c & 128) {
2034 base_offset += 1;
2035 if (!base_offset || MSB(base_offset, 7))
2036 return 0; /* overflow */
2037 c = base_info[used++];
2038 base_offset = (base_offset << 7) + (c & 127);
2039 }
2040 base_offset = delta_obj_offset - base_offset;
2041 if (base_offset <= 0 || base_offset >= delta_obj_offset)
2042 return 0; /* out of bound */
2043 *curpos += used;
2044 } else if (type == OBJ_REF_DELTA) {
2045 /* The base entry _must_ be in the same pack */
2046 base_offset = find_pack_entry_one(base_info, p);
2047 *curpos += 20;
2048 } else
2049 die("I am totally screwed");
2050 return base_offset;
2051 }
2052
2053 /*
2054 * Like get_delta_base above, but we return the sha1 instead of the pack
2055 * offset. This means it is cheaper for REF deltas (we do not have to do
2056 * the final object lookup), but more expensive for OFS deltas (we
2057 * have to load the revidx to convert the offset back into a sha1).
2058 */
2059 static const unsigned char *get_delta_base_sha1(struct packed_git *p,
2060 struct pack_window **w_curs,
2061 off_t curpos,
2062 enum object_type type,
2063 off_t delta_obj_offset)
2064 {
2065 if (type == OBJ_REF_DELTA) {
2066 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
2067 return base;
2068 } else if (type == OBJ_OFS_DELTA) {
2069 struct revindex_entry *revidx;
2070 off_t base_offset = get_delta_base(p, w_curs, &curpos,
2071 type, delta_obj_offset);
2072
2073 if (!base_offset)
2074 return NULL;
2075
2076 revidx = find_pack_revindex(p, base_offset);
2077 if (!revidx)
2078 return NULL;
2079
2080 return nth_packed_object_sha1(p, revidx->nr);
2081 } else
2082 return NULL;
2083 }
2084
2085 int unpack_object_header(struct packed_git *p,
2086 struct pack_window **w_curs,
2087 off_t *curpos,
2088 unsigned long *sizep)
2089 {
2090 unsigned char *base;
2091 unsigned long left;
2092 unsigned long used;
2093 enum object_type type;
2094
2095 /* use_pack() assures us we have [base, base + 20) available
2096 * as a range that we can look at. (Its actually the hash
2097 * size that is assured.) With our object header encoding
2098 * the maximum deflated object size is 2^137, which is just
2099 * insane, so we know won't exceed what we have been given.
2100 */
2101 base = use_pack(p, w_curs, *curpos, &left);
2102 used = unpack_object_header_buffer(base, left, &type, sizep);
2103 if (!used) {
2104 type = OBJ_BAD;
2105 } else
2106 *curpos += used;
2107
2108 return type;
2109 }
2110
2111 static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
2112 {
2113 int type;
2114 struct revindex_entry *revidx;
2115 const unsigned char *sha1;
2116 revidx = find_pack_revindex(p, obj_offset);
2117 if (!revidx)
2118 return OBJ_BAD;
2119 sha1 = nth_packed_object_sha1(p, revidx->nr);
2120 mark_bad_packed_object(p, sha1);
2121 type = sha1_object_info(sha1, NULL);
2122 if (type <= OBJ_NONE)
2123 return OBJ_BAD;
2124 return type;
2125 }
2126
2127 #define POI_STACK_PREALLOC 64
2128
2129 static enum object_type packed_to_object_type(struct packed_git *p,
2130 off_t obj_offset,
2131 enum object_type type,
2132 struct pack_window **w_curs,
2133 off_t curpos)
2134 {
2135 off_t small_poi_stack[POI_STACK_PREALLOC];
2136 off_t *poi_stack = small_poi_stack;
2137 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
2138
2139 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2140 off_t base_offset;
2141 unsigned long size;
2142 /* Push the object we're going to leave behind */
2143 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
2144 poi_stack_alloc = alloc_nr(poi_stack_nr);
2145 ALLOC_ARRAY(poi_stack, poi_stack_alloc);
2146 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
2147 } else {
2148 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
2149 }
2150 poi_stack[poi_stack_nr++] = obj_offset;
2151 /* If parsing the base offset fails, just unwind */
2152 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
2153 if (!base_offset)
2154 goto unwind;
2155 curpos = obj_offset = base_offset;
2156 type = unpack_object_header(p, w_curs, &curpos, &size);
2157 if (type <= OBJ_NONE) {
2158 /* If getting the base itself fails, we first
2159 * retry the base, otherwise unwind */
2160 type = retry_bad_packed_offset(p, base_offset);
2161 if (type > OBJ_NONE)
2162 goto out;
2163 goto unwind;
2164 }
2165 }
2166
2167 switch (type) {
2168 case OBJ_BAD:
2169 case OBJ_COMMIT:
2170 case OBJ_TREE:
2171 case OBJ_BLOB:
2172 case OBJ_TAG:
2173 break;
2174 default:
2175 error("unknown object type %i at offset %"PRIuMAX" in %s",
2176 type, (uintmax_t)obj_offset, p->pack_name);
2177 type = OBJ_BAD;
2178 }
2179
2180 out:
2181 if (poi_stack != small_poi_stack)
2182 free(poi_stack);
2183 return type;
2184
2185 unwind:
2186 while (poi_stack_nr) {
2187 obj_offset = poi_stack[--poi_stack_nr];
2188 type = retry_bad_packed_offset(p, obj_offset);
2189 if (type > OBJ_NONE)
2190 goto out;
2191 }
2192 type = OBJ_BAD;
2193 goto out;
2194 }
2195
2196 static struct hashmap delta_base_cache;
2197 static size_t delta_base_cached;
2198
2199 static LIST_HEAD(delta_base_cache_lru);
2200
2201 struct delta_base_cache_key {
2202 struct packed_git *p;
2203 off_t base_offset;
2204 };
2205
2206 struct delta_base_cache_entry {
2207 struct hashmap hash;
2208 struct delta_base_cache_key key;
2209 struct list_head lru;
2210 void *data;
2211 unsigned long size;
2212 enum object_type type;
2213 };
2214
2215 static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
2216 {
2217 unsigned int hash;
2218
2219 hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
2220 hash += (hash >> 8) + (hash >> 16);
2221 return hash;
2222 }
2223
2224 static struct delta_base_cache_entry *
2225 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
2226 {
2227 struct hashmap_entry entry;
2228 struct delta_base_cache_key key;
2229
2230 if (!delta_base_cache.cmpfn)
2231 return NULL;
2232
2233 hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
2234 key.p = p;
2235 key.base_offset = base_offset;
2236 return hashmap_get(&delta_base_cache, &entry, &key);
2237 }
2238
2239 static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
2240 const struct delta_base_cache_key *b)
2241 {
2242 return a->p == b->p && a->base_offset == b->base_offset;
2243 }
2244
2245 static int delta_base_cache_hash_cmp(const void *unused_cmp_data,
2246 const void *va, const void *vb,
2247 const void *vkey)
2248 {
2249 const struct delta_base_cache_entry *a = va, *b = vb;
2250 const struct delta_base_cache_key *key = vkey;
2251 if (key)
2252 return !delta_base_cache_key_eq(&a->key, key);
2253 else
2254 return !delta_base_cache_key_eq(&a->key, &b->key);
2255 }
2256
2257 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2258 {
2259 return !!get_delta_base_cache_entry(p, base_offset);
2260 }
2261
2262 /*
2263 * Remove the entry from the cache, but do _not_ free the associated
2264 * entry data. The caller takes ownership of the "data" buffer, and
2265 * should copy out any fields it wants before detaching.
2266 */
2267 static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2268 {
2269 hashmap_remove(&delta_base_cache, ent, &ent->key);
2270 list_del(&ent->lru);
2271 delta_base_cached -= ent->size;
2272 free(ent);
2273 }
2274
2275 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
2276 unsigned long *base_size, enum object_type *type)
2277 {
2278 struct delta_base_cache_entry *ent;
2279
2280 ent = get_delta_base_cache_entry(p, base_offset);
2281 if (!ent)
2282 return unpack_entry(p, base_offset, type, base_size);
2283
2284 if (type)
2285 *type = ent->type;
2286 if (base_size)
2287 *base_size = ent->size;
2288 return xmemdupz(ent->data, ent->size);
2289 }
2290
2291 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2292 {
2293 free(ent->data);
2294 detach_delta_base_cache_entry(ent);
2295 }
2296
2297 void clear_delta_base_cache(void)
2298 {
2299 struct list_head *lru, *tmp;
2300 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2301 struct delta_base_cache_entry *entry =
2302 list_entry(lru, struct delta_base_cache_entry, lru);
2303 release_delta_base_cache(entry);
2304 }
2305 }
2306
2307 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2308 void *base, unsigned long base_size, enum object_type type)
2309 {
2310 struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
2311 struct list_head *lru, *tmp;
2312
2313 delta_base_cached += base_size;
2314
2315 list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
2316 struct delta_base_cache_entry *f =
2317 list_entry(lru, struct delta_base_cache_entry, lru);
2318 if (delta_base_cached <= delta_base_cache_limit)
2319 break;
2320 release_delta_base_cache(f);
2321 }
2322
2323 ent->key.p = p;
2324 ent->key.base_offset = base_offset;
2325 ent->type = type;
2326 ent->data = base;
2327 ent->size = base_size;
2328 list_add_tail(&ent->lru, &delta_base_cache_lru);
2329
2330 if (!delta_base_cache.cmpfn)
2331 hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, NULL, 0);
2332 hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
2333 hashmap_add(&delta_base_cache, ent);
2334 }
2335
2336 int packed_object_info(struct packed_git *p, off_t obj_offset,
2337 struct object_info *oi)
2338 {
2339 struct pack_window *w_curs = NULL;
2340 unsigned long size;
2341 off_t curpos = obj_offset;
2342 enum object_type type;
2343
2344 /*
2345 * We always get the representation type, but only convert it to
2346 * a "real" type later if the caller is interested.
2347 */
2348 if (oi->contentp) {
2349 *oi->contentp = cache_or_unpack_entry(p, obj_offset, oi->sizep,
2350 &type);
2351 if (!*oi->contentp)
2352 type = OBJ_BAD;
2353 } else {
2354 type = unpack_object_header(p, &w_curs, &curpos, &size);
2355 }
2356
2357 if (!oi->contentp && oi->sizep) {
2358 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2359 off_t tmp_pos = curpos;
2360 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2361 type, obj_offset);
2362 if (!base_offset) {
2363 type = OBJ_BAD;
2364 goto out;
2365 }
2366 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2367 if (*oi->sizep == 0) {
2368 type = OBJ_BAD;
2369 goto out;
2370 }
2371 } else {
2372 *oi->sizep = size;
2373 }
2374 }
2375
2376 if (oi->disk_sizep) {
2377 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2378 *oi->disk_sizep = revidx[1].offset - obj_offset;
2379 }
2380
2381 if (oi->typep || oi->typename) {
2382 enum object_type ptot;
2383 ptot = packed_to_object_type(p, obj_offset, type, &w_curs,
2384 curpos);
2385 if (oi->typep)
2386 *oi->typep = ptot;
2387 if (oi->typename) {
2388 const char *tn = typename(ptot);
2389 if (tn)
2390 strbuf_addstr(oi->typename, tn);
2391 }
2392 if (ptot < 0) {
2393 type = OBJ_BAD;
2394 goto out;
2395 }
2396 }
2397
2398 if (oi->delta_base_sha1) {
2399 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2400 const unsigned char *base;
2401
2402 base = get_delta_base_sha1(p, &w_curs, curpos,
2403 type, obj_offset);
2404 if (!base) {
2405 type = OBJ_BAD;
2406 goto out;
2407 }
2408
2409 hashcpy(oi->delta_base_sha1, base);
2410 } else
2411 hashclr(oi->delta_base_sha1);
2412 }
2413
2414 oi->whence = in_delta_base_cache(p, obj_offset) ? OI_DBCACHED :
2415 OI_PACKED;
2416
2417 out:
2418 unuse_pack(&w_curs);
2419 return type;
2420 }
2421
2422 static void *unpack_compressed_entry(struct packed_git *p,
2423 struct pack_window **w_curs,
2424 off_t curpos,
2425 unsigned long size)
2426 {
2427 int st;
2428 git_zstream stream;
2429 unsigned char *buffer, *in;
2430
2431 buffer = xmallocz_gently(size);
2432 if (!buffer)
2433 return NULL;
2434 memset(&stream, 0, sizeof(stream));
2435 stream.next_out = buffer;
2436 stream.avail_out = size + 1;
2437
2438 git_inflate_init(&stream);
2439 do {
2440 in = use_pack(p, w_curs, curpos, &stream.avail_in);
2441 stream.next_in = in;
2442 st = git_inflate(&stream, Z_FINISH);
2443 if (!stream.avail_out)
2444 break; /* the payload is larger than it should be */
2445 curpos += stream.next_in - in;
2446 } while (st == Z_OK || st == Z_BUF_ERROR);
2447 git_inflate_end(&stream);
2448 if ((st != Z_STREAM_END) || stream.total_out != size) {
2449 free(buffer);
2450 return NULL;
2451 }
2452
2453 return buffer;
2454 }
2455
2456 static void *read_object(const unsigned char *sha1, enum object_type *type,
2457 unsigned long *size);
2458
2459 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2460 {
2461 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2462 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2463 p->pack_name, (uintmax_t)obj_offset);
2464 }
2465
2466 int do_check_packed_object_crc;
2467
2468 #define UNPACK_ENTRY_STACK_PREALLOC 64
2469 struct unpack_entry_stack_ent {
2470 off_t obj_offset;
2471 off_t curpos;
2472 unsigned long size;
2473 };
2474
2475 void *unpack_entry(struct packed_git *p, off_t obj_offset,
2476 enum object_type *final_type, unsigned long *final_size)
2477 {
2478 struct pack_window *w_curs = NULL;
2479 off_t curpos = obj_offset;
2480 void *data = NULL;
2481 unsigned long size;
2482 enum object_type type;
2483 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2484 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2485 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2486 int base_from_cache = 0;
2487
2488 write_pack_access_log(p, obj_offset);
2489
2490 /* PHASE 1: drill down to the innermost base object */
2491 for (;;) {
2492 off_t base_offset;
2493 int i;
2494 struct delta_base_cache_entry *ent;
2495
2496 ent = get_delta_base_cache_entry(p, curpos);
2497 if (ent) {
2498 type = ent->type;
2499 data = ent->data;
2500 size = ent->size;
2501 detach_delta_base_cache_entry(ent);
2502 base_from_cache = 1;
2503 break;
2504 }
2505
2506 if (do_check_packed_object_crc && p->index_version > 1) {
2507 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2508 off_t len = revidx[1].offset - obj_offset;
2509 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2510 const unsigned char *sha1 =
2511 nth_packed_object_sha1(p, revidx->nr);
2512 error("bad packed object CRC for %s",
2513 sha1_to_hex(sha1));
2514 mark_bad_packed_object(p, sha1);
2515 data = NULL;
2516 goto out;
2517 }
2518 }
2519
2520 type = unpack_object_header(p, &w_curs, &curpos, &size);
2521 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2522 break;
2523
2524 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2525 if (!base_offset) {
2526 error("failed to validate delta base reference "
2527 "at offset %"PRIuMAX" from %s",
2528 (uintmax_t)curpos, p->pack_name);
2529 /* bail to phase 2, in hopes of recovery */
2530 data = NULL;
2531 break;
2532 }
2533
2534 /* push object, proceed to base */
2535 if (delta_stack_nr >= delta_stack_alloc
2536 && delta_stack == small_delta_stack) {
2537 delta_stack_alloc = alloc_nr(delta_stack_nr);
2538 ALLOC_ARRAY(delta_stack, delta_stack_alloc);
2539 memcpy(delta_stack, small_delta_stack,
2540 sizeof(*delta_stack)*delta_stack_nr);
2541 } else {
2542 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2543 }
2544 i = delta_stack_nr++;
2545 delta_stack[i].obj_offset = obj_offset;
2546 delta_stack[i].curpos = curpos;
2547 delta_stack[i].size = size;
2548
2549 curpos = obj_offset = base_offset;
2550 }
2551
2552 /* PHASE 2: handle the base */
2553 switch (type) {
2554 case OBJ_OFS_DELTA:
2555 case OBJ_REF_DELTA:
2556 if (data)
2557 die("BUG: unpack_entry: left loop at a valid delta");
2558 break;
2559 case OBJ_COMMIT:
2560 case OBJ_TREE:
2561 case OBJ_BLOB:
2562 case OBJ_TAG:
2563 if (!base_from_cache)
2564 data = unpack_compressed_entry(p, &w_curs, curpos, size);
2565 break;
2566 default:
2567 data = NULL;
2568 error("unknown object type %i at offset %"PRIuMAX" in %s",
2569 type, (uintmax_t)obj_offset, p->pack_name);
2570 }
2571
2572 /* PHASE 3: apply deltas in order */
2573
2574 /* invariants:
2575 * 'data' holds the base data, or NULL if there was corruption
2576 */
2577 while (delta_stack_nr) {
2578 void *delta_data;
2579 void *base = data;
2580 void *external_base = NULL;
2581 unsigned long delta_size, base_size = size;
2582 int i;
2583
2584 data = NULL;
2585
2586 if (base)
2587 add_delta_base_cache(p, obj_offset, base, base_size, type);
2588
2589 if (!base) {
2590 /*
2591 * We're probably in deep shit, but let's try to fetch
2592 * the required base anyway from another pack or loose.
2593 * This is costly but should happen only in the presence
2594 * of a corrupted pack, and is better than failing outright.
2595 */
2596 struct revindex_entry *revidx;
2597 const unsigned char *base_sha1;
2598 revidx = find_pack_revindex(p, obj_offset);
2599 if (revidx) {
2600 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2601 error("failed to read delta base object %s"
2602 " at offset %"PRIuMAX" from %s",
2603 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2604 p->pack_name);
2605 mark_bad_packed_object(p, base_sha1);
2606 base = read_object(base_sha1, &type, &base_size);
2607 external_base = base;
2608 }
2609 }
2610
2611 i = --delta_stack_nr;
2612 obj_offset = delta_stack[i].obj_offset;
2613 curpos = delta_stack[i].curpos;
2614 delta_size = delta_stack[i].size;
2615
2616 if (!base)
2617 continue;
2618
2619 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2620
2621 if (!delta_data) {
2622 error("failed to unpack compressed delta "
2623 "at offset %"PRIuMAX" from %s",
2624 (uintmax_t)curpos, p->pack_name);
2625 data = NULL;
2626 free(external_base);
2627 continue;
2628 }
2629
2630 data = patch_delta(base, base_size,
2631 delta_data, delta_size,
2632 &size);
2633
2634 /*
2635 * We could not apply the delta; warn the user, but keep going.
2636 * Our failure will be noticed either in the next iteration of
2637 * the loop, or if this is the final delta, in the caller when
2638 * we return NULL. Those code paths will take care of making
2639 * a more explicit warning and retrying with another copy of
2640 * the object.
2641 */
2642 if (!data)
2643 error("failed to apply delta");
2644
2645 free(delta_data);
2646 free(external_base);
2647 }
2648
2649 if (final_type)
2650 *final_type = type;
2651 if (final_size)
2652 *final_size = size;
2653
2654 out:
2655 unuse_pack(&w_curs);
2656
2657 if (delta_stack != small_delta_stack)
2658 free(delta_stack);
2659
2660 return data;
2661 }
2662
2663 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
2664 uint32_t n)
2665 {
2666 const unsigned char *index = p->index_data;
2667 if (!index) {
2668 if (open_pack_index(p))
2669 return NULL;
2670 index = p->index_data;
2671 }
2672 if (n >= p->num_objects)
2673 return NULL;
2674 index += 4 * 256;
2675 if (p->index_version == 1) {
2676 return index + 24 * n + 4;
2677 } else {
2678 index += 8;
2679 return index + 20 * n;
2680 }
2681 }
2682
2683 const struct object_id *nth_packed_object_oid(struct object_id *oid,
2684 struct packed_git *p,
2685 uint32_t n)
2686 {
2687 const unsigned char *hash = nth_packed_object_sha1(p, n);
2688 if (!hash)
2689 return NULL;
2690 hashcpy(oid->hash, hash);
2691 return oid;
2692 }
2693
2694 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
2695 {
2696 const unsigned char *ptr = vptr;
2697 const unsigned char *start = p->index_data;
2698 const unsigned char *end = start + p->index_size;
2699 if (ptr < start)
2700 die(_("offset before start of pack index for %s (corrupt index?)"),
2701 p->pack_name);
2702 /* No need to check for underflow; .idx files must be at least 8 bytes */
2703 if (ptr >= end - 8)
2704 die(_("offset beyond end of pack index for %s (truncated index?)"),
2705 p->pack_name);
2706 }
2707
2708 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
2709 {
2710 const unsigned char *index = p->index_data;
2711 index += 4 * 256;
2712 if (p->index_version == 1) {
2713 return ntohl(*((uint32_t *)(index + 24 * n)));
2714 } else {
2715 uint32_t off;
2716 index += 8 + p->num_objects * (20 + 4);
2717 off = ntohl(*((uint32_t *)(index + 4 * n)));
2718 if (!(off & 0x80000000))
2719 return off;
2720 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2721 check_pack_index_ptr(p, index);
2722 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2723 ntohl(*((uint32_t *)(index + 4)));
2724 }
2725 }
2726
2727 off_t find_pack_entry_one(const unsigned char *sha1,
2728 struct packed_git *p)
2729 {
2730 const uint32_t *level1_ofs = p->index_data;
2731 const unsigned char *index = p->index_data;
2732 unsigned hi, lo, stride;
2733 static int debug_lookup = -1;
2734
2735 if (debug_lookup < 0)
2736 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
2737
2738 if (!index) {
2739 if (open_pack_index(p))
2740 return 0;
2741 level1_ofs = p->index_data;
2742 index = p->index_data;
2743 }
2744 if (p->index_version > 1) {
2745 level1_ofs += 2;
2746 index += 8;
2747 }
2748 index += 4 * 256;
2749 hi = ntohl(level1_ofs[*sha1]);
2750 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2751 if (p->index_version > 1) {
2752 stride = 20;
2753 } else {
2754 stride = 24;
2755 index += 4;
2756 }
2757
2758 if (debug_lookup)
2759 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2760 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2761
2762 while (lo < hi) {
2763 unsigned mi = (lo + hi) / 2;
2764 int cmp = hashcmp(index + mi * stride, sha1);
2765
2766 if (debug_lookup)
2767 printf("lo %u hi %u rg %u mi %u\n",
2768 lo, hi, hi - lo, mi);
2769 if (!cmp)
2770 return nth_packed_object_offset(p, mi);
2771 if (cmp > 0)
2772 hi = mi;
2773 else
2774 lo = mi+1;
2775 }
2776 return 0;
2777 }
2778
2779 int is_pack_valid(struct packed_git *p)
2780 {
2781 /* An already open pack is known to be valid. */
2782 if (p->pack_fd != -1)
2783 return 1;
2784
2785 /* If the pack has one window completely covering the
2786 * file size, the pack is known to be valid even if
2787 * the descriptor is not currently open.
2788 */
2789 if (p->windows) {
2790 struct pack_window *w = p->windows;
2791
2792 if (!w->offset && w->len == p->pack_size)
2793 return 1;
2794 }
2795
2796 /* Force the pack to open to prove its valid. */
2797 return !open_packed_git(p);
2798 }
2799
2800 static int fill_pack_entry(const unsigned char *sha1,
2801 struct pack_entry *e,
2802 struct packed_git *p)
2803 {
2804 off_t offset;
2805
2806 if (p->num_bad_objects) {
2807 unsigned i;
2808 for (i = 0; i < p->num_bad_objects; i++)
2809 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2810 return 0;
2811 }
2812
2813 offset = find_pack_entry_one(sha1, p);
2814 if (!offset)
2815 return 0;
2816
2817 /*
2818 * We are about to tell the caller where they can locate the
2819 * requested object. We better make sure the packfile is
2820 * still here and can be accessed before supplying that
2821 * answer, as it may have been deleted since the index was
2822 * loaded!
2823 */
2824 if (!is_pack_valid(p))
2825 return 0;
2826 e->offset = offset;
2827 e->p = p;
2828 hashcpy(e->sha1, sha1);
2829 return 1;
2830 }
2831
2832 /*
2833 * Iff a pack file contains the object named by sha1, return true and
2834 * store its location to e.
2835 */
2836 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2837 {
2838 struct mru_entry *p;
2839
2840 prepare_packed_git();
2841 if (!packed_git)
2842 return 0;
2843
2844 for (p = packed_git_mru->head; p; p = p->next) {
2845 if (fill_pack_entry(sha1, e, p->item)) {
2846 mru_mark(packed_git_mru, p);
2847 return 1;
2848 }
2849 }
2850 return 0;
2851 }
2852
2853 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2854 struct packed_git *packs)
2855 {
2856 struct packed_git *p;
2857
2858 for (p = packs; p; p = p->next) {
2859 if (find_pack_entry_one(sha1, p))
2860 return p;
2861 }
2862 return NULL;
2863
2864 }
2865
2866 static int sha1_loose_object_info(const unsigned char *sha1,
2867 struct object_info *oi,
2868 int flags)
2869 {
2870 int status = 0;
2871 unsigned long mapsize;
2872 void *map;
2873 git_zstream stream;
2874 char hdr[32];
2875 struct strbuf hdrbuf = STRBUF_INIT;
2876 unsigned long size_scratch;
2877
2878 if (oi->delta_base_sha1)
2879 hashclr(oi->delta_base_sha1);
2880
2881 /*
2882 * If we don't care about type or size, then we don't
2883 * need to look inside the object at all. Note that we
2884 * do not optimize out the stat call, even if the
2885 * caller doesn't care about the disk-size, since our
2886 * return value implicitly indicates whether the
2887 * object even exists.
2888 */
2889 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
2890 const char *path;
2891 struct stat st;
2892 if (stat_sha1_file(sha1, &st, &path) < 0)
2893 return -1;
2894 if (oi->disk_sizep)
2895 *oi->disk_sizep = st.st_size;
2896 return 0;
2897 }
2898
2899 map = map_sha1_file(sha1, &mapsize);
2900 if (!map)
2901 return -1;
2902
2903 if (!oi->sizep)
2904 oi->sizep = &size_scratch;
2905
2906 if (oi->disk_sizep)
2907 *oi->disk_sizep = mapsize;
2908 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
2909 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2910 status = error("unable to unpack %s header with --allow-unknown-type",
2911 sha1_to_hex(sha1));
2912 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2913 status = error("unable to unpack %s header",
2914 sha1_to_hex(sha1));
2915 if (status < 0)
2916 ; /* Do nothing */
2917 else if (hdrbuf.len) {
2918 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2919 status = error("unable to parse %s header with --allow-unknown-type",
2920 sha1_to_hex(sha1));
2921 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
2922 status = error("unable to parse %s header", sha1_to_hex(sha1));
2923
2924 if (status >= 0 && oi->contentp)
2925 *oi->contentp = unpack_sha1_rest(&stream, hdr,
2926 *oi->sizep, sha1);
2927 else
2928 git_inflate_end(&stream);
2929
2930 munmap(map, mapsize);
2931 if (status && oi->typep)
2932 *oi->typep = status;
2933 if (oi->sizep == &size_scratch)
2934 oi->sizep = NULL;
2935 strbuf_release(&hdrbuf);
2936 oi->whence = OI_LOOSE;
2937 return (status < 0) ? status : 0;
2938 }
2939
2940 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
2941 {
2942 static struct object_info blank_oi = OBJECT_INFO_INIT;
2943 struct pack_entry e;
2944 int rtype;
2945 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
2946 lookup_replace_object(sha1) :
2947 sha1;
2948
2949 if (!oi)
2950 oi = &blank_oi;
2951
2952 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
2953 struct cached_object *co = find_cached_object(real);
2954 if (co) {
2955 if (oi->typep)
2956 *(oi->typep) = co->type;
2957 if (oi->sizep)
2958 *(oi->sizep) = co->size;
2959 if (oi->disk_sizep)
2960 *(oi->disk_sizep) = 0;
2961 if (oi->delta_base_sha1)
2962 hashclr(oi->delta_base_sha1);
2963 if (oi->typename)
2964 strbuf_addstr(oi->typename, typename(co->type));
2965 if (oi->contentp)
2966 *oi->contentp = xmemdupz(co->buf, co->size);
2967 oi->whence = OI_CACHED;
2968 return 0;
2969 }
2970 }
2971
2972 if (!find_pack_entry(real, &e)) {
2973 /* Most likely it's a loose object. */
2974 if (!sha1_loose_object_info(real, oi, flags))
2975 return 0;
2976
2977 /* Not a loose object; someone else may have just packed it. */
2978 if (flags & OBJECT_INFO_QUICK) {
2979 return -1;
2980 } else {
2981 reprepare_packed_git();
2982 if (!find_pack_entry(real, &e))
2983 return -1;
2984 }
2985 }
2986
2987 if (oi == &blank_oi)
2988 /*
2989 * We know that the caller doesn't actually need the
2990 * information below, so return early.
2991 */
2992 return 0;
2993
2994 rtype = packed_object_info(e.p, e.offset, oi);
2995 if (rtype < 0) {
2996 mark_bad_packed_object(e.p, real);
2997 return sha1_object_info_extended(real, oi, 0);
2998 } else if (oi->whence == OI_PACKED) {
2999 oi->u.packed.offset = e.offset;
3000 oi->u.packed.pack = e.p;
3001 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
3002 rtype == OBJ_OFS_DELTA);
3003 }
3004
3005 return 0;
3006 }
3007
3008 /* returns enum object_type or negative */
3009 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
3010 {
3011 enum object_type type;
3012 struct object_info oi = OBJECT_INFO_INIT;
3013
3014 oi.typep = &type;
3015 oi.sizep = sizep;
3016 if (sha1_object_info_extended(sha1, &oi,
3017 OBJECT_INFO_LOOKUP_REPLACE) < 0)
3018 return -1;
3019 return type;
3020 }
3021
3022 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
3023 unsigned char *sha1)
3024 {
3025 struct cached_object *co;
3026
3027 hash_sha1_file(buf, len, typename(type), sha1);
3028 if (has_sha1_file(sha1) || find_cached_object(sha1))
3029 return 0;
3030 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
3031 co = &cached_objects[cached_object_nr++];
3032 co->size = len;
3033 co->type = type;
3034 co->buf = xmalloc(len);
3035 memcpy(co->buf, buf, len);
3036 hashcpy(co->sha1, sha1);
3037 return 0;
3038 }
3039
3040 static void *read_object(const unsigned char *sha1, enum object_type *type,
3041 unsigned long *size)
3042 {
3043 struct object_info oi = OBJECT_INFO_INIT;
3044 void *content;
3045 oi.typep = type;
3046 oi.sizep = size;
3047 oi.contentp = &content;
3048
3049 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
3050 return NULL;
3051 return content;
3052 }
3053
3054 /*
3055 * This function dies on corrupt objects; the callers who want to
3056 * deal with them should arrange to call read_object() and give error
3057 * messages themselves.
3058 */
3059 void *read_sha1_file_extended(const unsigned char *sha1,
3060 enum object_type *type,
3061 unsigned long *size,
3062 int lookup_replace)
3063 {
3064 void *data;
3065 const struct packed_git *p;
3066 const char *path;
3067 struct stat st;
3068 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
3069 : sha1;
3070
3071 errno = 0;
3072 data = read_object(repl, type, size);
3073 if (data)
3074 return data;
3075
3076 if (errno && errno != ENOENT)
3077 die_errno("failed to read object %s", sha1_to_hex(sha1));
3078
3079 /* die if we replaced an object with one that does not exist */
3080 if (repl != sha1)
3081 die("replacement %s not found for %s",
3082 sha1_to_hex(repl), sha1_to_hex(sha1));
3083
3084 if (!stat_sha1_file(repl, &st, &path))
3085 die("loose object %s (stored in %s) is corrupt",
3086 sha1_to_hex(repl), path);
3087
3088 if ((p = has_packed_and_bad(repl)) != NULL)
3089 die("packed object %s (stored in %s) is corrupt",
3090 sha1_to_hex(repl), p->pack_name);
3091
3092 return NULL;
3093 }
3094
3095 void *read_object_with_reference(const unsigned char *sha1,
3096 const char *required_type_name,
3097 unsigned long *size,
3098 unsigned char *actual_sha1_return)
3099 {
3100 enum object_type type, required_type;
3101 void *buffer;
3102 unsigned long isize;
3103 unsigned char actual_sha1[20];
3104
3105 required_type = type_from_string(required_type_name);
3106 hashcpy(actual_sha1, sha1);
3107 while (1) {
3108 int ref_length = -1;
3109 const char *ref_type = NULL;
3110
3111 buffer = read_sha1_file(actual_sha1, &type, &isize);
3112 if (!buffer)
3113 return NULL;
3114 if (type == required_type) {
3115 *size = isize;
3116 if (actual_sha1_return)
3117 hashcpy(actual_sha1_return, actual_sha1);
3118 return buffer;
3119 }
3120 /* Handle references */
3121 else if (type == OBJ_COMMIT)
3122 ref_type = "tree ";
3123 else if (type == OBJ_TAG)
3124 ref_type = "object ";
3125 else {
3126 free(buffer);
3127 return NULL;
3128 }
3129 ref_length = strlen(ref_type);
3130
3131 if (ref_length + 40 > isize ||
3132 memcmp(buffer, ref_type, ref_length) ||
3133 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
3134 free(buffer);
3135 return NULL;
3136 }
3137 free(buffer);
3138 /* Now we have the ID of the referred-to object in
3139 * actual_sha1. Check again. */
3140 }
3141 }
3142
3143 static void write_sha1_file_prepare(const void *buf, unsigned long len,
3144 const char *type, unsigned char *sha1,
3145 char *hdr, int *hdrlen)
3146 {
3147 git_SHA_CTX c;
3148
3149 /* Generate the header */
3150 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
3151
3152 /* Sha1.. */
3153 git_SHA1_Init(&c);
3154 git_SHA1_Update(&c, hdr, *hdrlen);
3155 git_SHA1_Update(&c, buf, len);
3156 git_SHA1_Final(sha1, &c);
3157 }
3158
3159 /*
3160 * Move the just written object into its final resting place.
3161 */
3162 int finalize_object_file(const char *tmpfile, const char *filename)
3163 {
3164 int ret = 0;
3165
3166 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
3167 goto try_rename;
3168 else if (link(tmpfile, filename))
3169 ret = errno;
3170
3171 /*
3172 * Coda hack - coda doesn't like cross-directory links,
3173 * so we fall back to a rename, which will mean that it
3174 * won't be able to check collisions, but that's not a
3175 * big deal.
3176 *
3177 * The same holds for FAT formatted media.
3178 *
3179 * When this succeeds, we just return. We have nothing
3180 * left to unlink.
3181 */
3182 if (ret && ret != EEXIST) {
3183 try_rename:
3184 if (!rename(tmpfile, filename))
3185 goto out;
3186 ret = errno;
3187 }
3188 unlink_or_warn(tmpfile);
3189 if (ret) {
3190 if (ret != EEXIST) {
3191 return error_errno("unable to write sha1 filename %s", filename);
3192 }
3193 /* FIXME!!! Collision check here ? */
3194 }
3195
3196 out:
3197 if (adjust_shared_perm(filename))
3198 return error("unable to set permission to '%s'", filename);
3199 return 0;
3200 }
3201
3202 static int write_buffer(int fd, const void *buf, size_t len)
3203 {
3204 if (write_in_full(fd, buf, len) < 0)
3205 return error_errno("file write error");
3206 return 0;
3207 }
3208
3209 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
3210 unsigned char *sha1)
3211 {
3212 char hdr[32];
3213 int hdrlen = sizeof(hdr);
3214 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3215 return 0;
3216 }
3217
3218 /* Finalize a file on disk, and close it. */
3219 static void close_sha1_file(int fd)
3220 {
3221 if (fsync_object_files)
3222 fsync_or_die(fd, "sha1 file");
3223 if (close(fd) != 0)
3224 die_errno("error when closing sha1 file");
3225 }
3226
3227 /* Size of directory component, including the ending '/' */
3228 static inline int directory_size(const char *filename)
3229 {
3230 const char *s = strrchr(filename, '/');
3231 if (!s)
3232 return 0;
3233 return s - filename + 1;
3234 }
3235
3236 /*
3237 * This creates a temporary file in the same directory as the final
3238 * 'filename'
3239 *
3240 * We want to avoid cross-directory filename renames, because those
3241 * can have problems on various filesystems (FAT, NFS, Coda).
3242 */
3243 static int create_tmpfile(struct strbuf *tmp, const char *filename)
3244 {
3245 int fd, dirlen = directory_size(filename);
3246
3247 strbuf_reset(tmp);
3248 strbuf_add(tmp, filename, dirlen);
3249 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3250 fd = git_mkstemp_mode(tmp->buf, 0444);
3251 if (fd < 0 && dirlen && errno == ENOENT) {
3252 /*
3253 * Make sure the directory exists; note that the contents
3254 * of the buffer are undefined after mkstemp returns an
3255 * error, so we have to rewrite the whole buffer from
3256 * scratch.
3257 */
3258 strbuf_reset(tmp);
3259 strbuf_add(tmp, filename, dirlen - 1);
3260 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
3261 return -1;
3262 if (adjust_shared_perm(tmp->buf))
3263 return -1;
3264
3265 /* Try again */
3266 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3267 fd = git_mkstemp_mode(tmp->buf, 0444);
3268 }
3269 return fd;
3270 }
3271
3272 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
3273 const void *buf, unsigned long len, time_t mtime)
3274 {
3275 int fd, ret;
3276 unsigned char compressed[4096];
3277 git_zstream stream;
3278 git_SHA_CTX c;
3279 unsigned char parano_sha1[20];
3280 static struct strbuf tmp_file = STRBUF_INIT;
3281 const char *filename = sha1_file_name(sha1);
3282
3283 fd = create_tmpfile(&tmp_file, filename);
3284 if (fd < 0) {
3285 if (errno == EACCES)
3286 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3287 else
3288 return error_errno("unable to create temporary file");
3289 }
3290
3291 /* Set it up */
3292 git_deflate_init(&stream, zlib_compression_level);
3293 stream.next_out = compressed;
3294 stream.avail_out = sizeof(compressed);
3295 git_SHA1_Init(&c);
3296
3297 /* First header.. */
3298 stream.next_in = (unsigned char *)hdr;
3299 stream.avail_in = hdrlen;
3300 while (git_deflate(&stream, 0) == Z_OK)
3301 ; /* nothing */
3302 git_SHA1_Update(&c, hdr, hdrlen);
3303
3304 /* Then the data itself.. */
3305 stream.next_in = (void *)buf;
3306 stream.avail_in = len;
3307 do {
3308 unsigned char *in0 = stream.next_in;
3309 ret = git_deflate(&stream, Z_FINISH);
3310 git_SHA1_Update(&c, in0, stream.next_in - in0);
3311 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3312 die("unable to write sha1 file");
3313 stream.next_out = compressed;
3314 stream.avail_out = sizeof(compressed);
3315 } while (ret == Z_OK);
3316
3317 if (ret != Z_STREAM_END)
3318 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
3319 ret = git_deflate_end_gently(&stream);
3320 if (ret != Z_OK)
3321 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
3322 git_SHA1_Final(parano_sha1, &c);
3323 if (hashcmp(sha1, parano_sha1) != 0)
3324 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
3325
3326 close_sha1_file(fd);
3327
3328 if (mtime) {
3329 struct utimbuf utb;
3330 utb.actime = mtime;
3331 utb.modtime = mtime;
3332 if (utime(tmp_file.buf, &utb) < 0)
3333 warning_errno("failed utime() on %s", tmp_file.buf);
3334 }
3335
3336 return finalize_object_file(tmp_file.buf, filename);
3337 }
3338
3339 static int freshen_loose_object(const unsigned char *sha1)
3340 {
3341 return check_and_freshen(sha1, 1);
3342 }
3343
3344 static int freshen_packed_object(const unsigned char *sha1)
3345 {
3346 struct pack_entry e;
3347 if (!find_pack_entry(sha1, &e))
3348 return 0;
3349 if (e.p->freshened)
3350 return 1;
3351 if (!freshen_file(e.p->pack_name))
3352 return 0;
3353 e.p->freshened = 1;
3354 return 1;
3355 }
3356
3357 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
3358 {
3359 char hdr[32];
3360 int hdrlen = sizeof(hdr);
3361
3362 /* Normally if we have it in the pack then we do not bother writing
3363 * it out into .git/objects/??/?{38} file.
3364 */
3365 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
3366 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3367 return 0;
3368 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3369 }
3370
3371 int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3372 unsigned char *sha1, unsigned flags)
3373 {
3374 char *header;
3375 int hdrlen, status = 0;
3376
3377 /* type string, SP, %lu of the length plus NUL must fit this */
3378 hdrlen = strlen(type) + 32;
3379 header = xmalloc(hdrlen);
3380 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3381
3382 if (!(flags & HASH_WRITE_OBJECT))
3383 goto cleanup;
3384 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
3385 goto cleanup;
3386 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3387
3388 cleanup:
3389 free(header);
3390 return status;
3391 }
3392
3393 int force_object_loose(const unsigned char *sha1, time_t mtime)
3394 {
3395 void *buf;
3396 unsigned long len;
3397 enum object_type type;
3398 char hdr[32];
3399 int hdrlen;
3400 int ret;
3401
3402 if (has_loose_object(sha1))
3403 return 0;
3404 buf = read_object(sha1, &type, &len);
3405 if (!buf)
3406 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
3407 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
3408 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3409 free(buf);
3410
3411 return ret;
3412 }
3413
3414 int has_pack_index(const unsigned char *sha1)
3415 {
3416 struct stat st;
3417 if (stat(sha1_pack_index_name(sha1), &st))
3418 return 0;
3419 return 1;
3420 }
3421
3422 int has_sha1_pack(const unsigned char *sha1)
3423 {
3424 struct pack_entry e;
3425 return find_pack_entry(sha1, &e);
3426 }
3427
3428 int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
3429 {
3430 if (!startup_info->have_repository)
3431 return 0;
3432 return sha1_object_info_extended(sha1, NULL,
3433 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
3434 }
3435
3436 int has_object_file(const struct object_id *oid)
3437 {
3438 return has_sha1_file(oid->hash);
3439 }
3440
3441 int has_object_file_with_flags(const struct object_id *oid, int flags)
3442 {
3443 return has_sha1_file_with_flags(oid->hash, flags);
3444 }
3445
3446 static void check_tree(const void *buf, size_t size)
3447 {
3448 struct tree_desc desc;
3449 struct name_entry entry;
3450
3451 init_tree_desc(&desc, buf, size);
3452 while (tree_entry(&desc, &entry))
3453 /* do nothing
3454 * tree_entry() will die() on malformed entries */
3455 ;
3456 }
3457
3458 static void check_commit(const void *buf, size_t size)
3459 {
3460 struct commit c;
3461 memset(&c, 0, sizeof(c));
3462 if (parse_commit_buffer(&c, buf, size))
3463 die("corrupt commit");
3464 }
3465
3466 static void check_tag(const void *buf, size_t size)
3467 {
3468 struct tag t;
3469 memset(&t, 0, sizeof(t));
3470 if (parse_tag_buffer(&t, buf, size))
3471 die("corrupt tag");
3472 }
3473
3474 static int index_mem(unsigned char *sha1, void *buf, size_t size,
3475 enum object_type type,
3476 const char *path, unsigned flags)
3477 {
3478 int ret, re_allocated = 0;
3479 int write_object = flags & HASH_WRITE_OBJECT;
3480
3481 if (!type)
3482 type = OBJ_BLOB;
3483
3484 /*
3485 * Convert blobs to git internal format
3486 */
3487 if ((type == OBJ_BLOB) && path) {
3488 struct strbuf nbuf = STRBUF_INIT;
3489 if (convert_to_git(&the_index, path, buf, size, &nbuf,
3490 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
3491 buf = strbuf_detach(&nbuf, &size);
3492 re_allocated = 1;
3493 }
3494 }
3495 if (flags & HASH_FORMAT_CHECK) {
3496 if (type == OBJ_TREE)
3497 check_tree(buf, size);
3498 if (type == OBJ_COMMIT)
3499 check_commit(buf, size);
3500 if (type == OBJ_TAG)
3501 check_tag(buf, size);
3502 }
3503
3504 if (write_object)
3505 ret = write_sha1_file(buf, size, typename(type), sha1);
3506 else
3507 ret = hash_sha1_file(buf, size, typename(type), sha1);
3508 if (re_allocated)
3509 free(buf);
3510 return ret;
3511 }
3512
3513 static int index_stream_convert_blob(unsigned char *sha1, int fd,
3514 const char *path, unsigned flags)
3515 {
3516 int ret;
3517 const int write_object = flags & HASH_WRITE_OBJECT;
3518 struct strbuf sbuf = STRBUF_INIT;
3519
3520 assert(path);
3521 assert(would_convert_to_git_filter_fd(path));
3522
3523 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
3524 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3525
3526 if (write_object)
3527 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3528 sha1);
3529 else
3530 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3531 sha1);
3532 strbuf_release(&sbuf);
3533 return ret;
3534 }
3535
3536 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3537 const char *path, unsigned flags)
3538 {
3539 struct strbuf sbuf = STRBUF_INIT;
3540 int ret;
3541
3542 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3543 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3544 else
3545 ret = -1;
3546 strbuf_release(&sbuf);
3547 return ret;
3548 }
3549
3550 #define SMALL_FILE_SIZE (32*1024)
3551
3552 static int index_core(unsigned char *sha1, int fd, size_t size,
3553 enum object_type type, const char *path,
3554 unsigned flags)
3555 {
3556 int ret;
3557
3558 if (!size) {
3559 ret = index_mem(sha1, "", size, type, path, flags);
3560 } else if (size <= SMALL_FILE_SIZE) {
3561 char *buf = xmalloc(size);
3562 if (size == read_in_full(fd, buf, size))
3563 ret = index_mem(sha1, buf, size, type, path, flags);
3564 else
3565 ret = error_errno("short read");
3566 free(buf);
3567 } else {
3568 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
3569 ret = index_mem(sha1, buf, size, type, path, flags);
3570 munmap(buf, size);
3571 }
3572 return ret;
3573 }
3574
3575 /*
3576 * This creates one packfile per large blob unless bulk-checkin
3577 * machinery is "plugged".
3578 *
3579 * This also bypasses the usual "convert-to-git" dance, and that is on
3580 * purpose. We could write a streaming version of the converting
3581 * functions and insert that before feeding the data to fast-import
3582 * (or equivalent in-core API described above). However, that is
3583 * somewhat complicated, as we do not know the size of the filter
3584 * result, which we need to know beforehand when writing a git object.
3585 * Since the primary motivation for trying to stream from the working
3586 * tree file and to avoid mmaping it in core is to deal with large
3587 * binary blobs, they generally do not want to get any conversion, and
3588 * callers should avoid this code path when filters are requested.
3589 */
3590 static int index_stream(unsigned char *sha1, int fd, size_t size,
3591 enum object_type type, const char *path,
3592 unsigned flags)
3593 {
3594 return index_bulk_checkin(sha1, fd, size, type, path, flags);
3595 }
3596
3597 int index_fd(unsigned char *sha1, int fd, struct stat *st,
3598 enum object_type type, const char *path, unsigned flags)
3599 {
3600 int ret;
3601
3602 /*
3603 * Call xsize_t() only when needed to avoid potentially unnecessary
3604 * die() for large files.
3605 */
3606 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3607 ret = index_stream_convert_blob(sha1, fd, path, flags);
3608 else if (!S_ISREG(st->st_mode))
3609 ret = index_pipe(sha1, fd, type, path, flags);
3610 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
3611 (path && would_convert_to_git(&the_index, path)))
3612 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3613 flags);
3614 else
3615 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3616 flags);
3617 close(fd);
3618 return ret;
3619 }
3620
3621 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
3622 {
3623 int fd;
3624 struct strbuf sb = STRBUF_INIT;
3625
3626 switch (st->st_mode & S_IFMT) {
3627 case S_IFREG:
3628 fd = open(path, O_RDONLY);
3629 if (fd < 0)
3630 return error_errno("open(\"%s\")", path);
3631 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
3632 return error("%s: failed to insert into database",
3633 path);
3634 break;
3635 case S_IFLNK:
3636 if (strbuf_readlink(&sb, path, st->st_size))
3637 return error_errno("readlink(\"%s\")", path);
3638 if (!(flags & HASH_WRITE_OBJECT))
3639 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3640 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
3641 return error("%s: failed to insert into database",
3642 path);
3643 strbuf_release(&sb);
3644 break;
3645 case S_IFDIR:
3646 return resolve_gitlink_ref(path, "HEAD", sha1);
3647 default:
3648 return error("%s: unsupported file type", path);
3649 }
3650 return 0;
3651 }
3652
3653 int read_pack_header(int fd, struct pack_header *header)
3654 {
3655 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3656 /* "eof before pack header was fully read" */
3657 return PH_ERROR_EOF;
3658
3659 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3660 /* "protocol error (pack signature mismatch detected)" */
3661 return PH_ERROR_PACK_SIGNATURE;
3662 if (!pack_version_ok(header->hdr_version))
3663 /* "protocol error (pack version unsupported)" */
3664 return PH_ERROR_PROTOCOL;
3665 return 0;
3666 }
3667
3668 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3669 {
3670 enum object_type type = sha1_object_info(sha1, NULL);
3671 if (type < 0)
3672 die("%s is not a valid object", sha1_to_hex(sha1));
3673 if (type != expect)
3674 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3675 typename(expect));
3676 }
3677
3678 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
3679 struct strbuf *path,
3680 each_loose_object_fn obj_cb,
3681 each_loose_cruft_fn cruft_cb,
3682 each_loose_subdir_fn subdir_cb,
3683 void *data)
3684 {
3685 size_t origlen, baselen;
3686 DIR *dir;
3687 struct dirent *de;
3688 int r = 0;
3689
3690 if (subdir_nr > 0xff)
3691 BUG("invalid loose object subdirectory: %x", subdir_nr);
3692
3693 origlen = path->len;
3694 strbuf_complete(path, '/');
3695 strbuf_addf(path, "%02x", subdir_nr);
3696 baselen = path->len;
3697
3698 dir = opendir(path->buf);
3699 if (!dir) {
3700 if (errno != ENOENT)
3701 r = error_errno("unable to open %s", path->buf);
3702 strbuf_setlen(path, origlen);
3703 return r;
3704 }
3705
3706 while ((de = readdir(dir))) {
3707 if (is_dot_or_dotdot(de->d_name))
3708 continue;
3709
3710 strbuf_setlen(path, baselen);
3711 strbuf_addf(path, "/%s", de->d_name);
3712
3713 if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2) {
3714 char hex[GIT_MAX_HEXSZ+1];
3715 struct object_id oid;
3716
3717 xsnprintf(hex, sizeof(hex), "%02x%s",
3718 subdir_nr, de->d_name);
3719 if (!get_oid_hex(hex, &oid)) {
3720 if (obj_cb) {
3721 r = obj_cb(&oid, path->buf, data);
3722 if (r)
3723 break;
3724 }
3725 continue;
3726 }
3727 }
3728
3729 if (cruft_cb) {
3730 r = cruft_cb(de->d_name, path->buf, data);
3731 if (r)
3732 break;
3733 }
3734 }
3735 closedir(dir);
3736
3737 strbuf_setlen(path, baselen);
3738 if (!r && subdir_cb)
3739 r = subdir_cb(subdir_nr, path->buf, data);
3740
3741 strbuf_setlen(path, origlen);
3742
3743 return r;
3744 }
3745
3746 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
3747 each_loose_object_fn obj_cb,
3748 each_loose_cruft_fn cruft_cb,
3749 each_loose_subdir_fn subdir_cb,
3750 void *data)
3751 {
3752 int r = 0;
3753 int i;
3754
3755 for (i = 0; i < 256; i++) {
3756 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
3757 subdir_cb, data);
3758 if (r)
3759 break;
3760 }
3761
3762 return r;
3763 }
3764
3765 int for_each_loose_file_in_objdir(const char *path,
3766 each_loose_object_fn obj_cb,
3767 each_loose_cruft_fn cruft_cb,
3768 each_loose_subdir_fn subdir_cb,
3769 void *data)
3770 {
3771 struct strbuf buf = STRBUF_INIT;
3772 int r;
3773
3774 strbuf_addstr(&buf, path);
3775 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3776 subdir_cb, data);
3777 strbuf_release(&buf);
3778
3779 return r;
3780 }
3781
3782 struct loose_alt_odb_data {
3783 each_loose_object_fn *cb;
3784 void *data;
3785 };
3786
3787 static int loose_from_alt_odb(struct alternate_object_database *alt,
3788 void *vdata)
3789 {
3790 struct loose_alt_odb_data *data = vdata;
3791 struct strbuf buf = STRBUF_INIT;
3792 int r;
3793
3794 strbuf_addstr(&buf, alt->path);
3795 r = for_each_loose_file_in_objdir_buf(&buf,
3796 data->cb, NULL, NULL,
3797 data->data);
3798 strbuf_release(&buf);
3799 return r;
3800 }
3801
3802 int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
3803 {
3804 struct loose_alt_odb_data alt;
3805 int r;
3806
3807 r = for_each_loose_file_in_objdir(get_object_directory(),
3808 cb, NULL, NULL, data);
3809 if (r)
3810 return r;
3811
3812 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
3813 return 0;
3814
3815 alt.cb = cb;
3816 alt.data = data;
3817 return foreach_alt_odb(loose_from_alt_odb, &alt);
3818 }
3819
3820 static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn cb, void *data)
3821 {
3822 uint32_t i;
3823 int r = 0;
3824
3825 for (i = 0; i < p->num_objects; i++) {
3826 struct object_id oid;
3827
3828 if (!nth_packed_object_oid(&oid, p, i))
3829 return error("unable to get sha1 of object %u in %s",
3830 i, p->pack_name);
3831
3832 r = cb(&oid, p, i, data);
3833 if (r)
3834 break;
3835 }
3836 return r;
3837 }
3838
3839 int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
3840 {
3841 struct packed_git *p;
3842 int r = 0;
3843 int pack_errors = 0;
3844
3845 prepare_packed_git();
3846 for (p = packed_git; p; p = p->next) {
3847 if ((flags & FOR_EACH_OBJECT_LOCAL_ONLY) && !p->pack_local)
3848 continue;
3849 if (open_pack_index(p)) {
3850 pack_errors = 1;
3851 continue;
3852 }
3853 r = for_each_object_in_pack(p, cb, data);
3854 if (r)
3855 break;
3856 }
3857 return r ? r : pack_errors;
3858 }
3859
3860 static int check_stream_sha1(git_zstream *stream,
3861 const char *hdr,
3862 unsigned long size,
3863 const char *path,
3864 const unsigned char *expected_sha1)
3865