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 "repository.h"
26#include "streaming.h"
27#include "dir.h"
28#include "list.h"
29#include "mergesort.h"
30#include "quote.h"
31#include "packfile.h"
32#include "fetch-object.h"
33#include "object-store.h"
34
35const unsigned char null_sha1[GIT_MAX_RAWSZ];
36const struct object_id null_oid;
37const struct object_id empty_tree_oid = {
38 EMPTY_TREE_SHA1_BIN_LITERAL
39};
40const struct object_id empty_blob_oid = {
41 EMPTY_BLOB_SHA1_BIN_LITERAL
42};
43
44static void git_hash_sha1_init(void *ctx)
45{
46 git_SHA1_Init((git_SHA_CTX *)ctx);
47}
48
49static void git_hash_sha1_update(void *ctx, const void *data, size_t len)
50{
51 git_SHA1_Update((git_SHA_CTX *)ctx, data, len);
52}
53
54static void git_hash_sha1_final(unsigned char *hash, void *ctx)
55{
56 git_SHA1_Final(hash, (git_SHA_CTX *)ctx);
57}
58
59static void git_hash_unknown_init(void *ctx)
60{
61 die("trying to init unknown hash");
62}
63
64static void git_hash_unknown_update(void *ctx, const void *data, size_t len)
65{
66 die("trying to update unknown hash");
67}
68
69static void git_hash_unknown_final(unsigned char *hash, void *ctx)
70{
71 die("trying to finalize unknown hash");
72}
73
74const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
75 {
76 NULL,
77 0x00000000,
78 0,
79 0,
80 0,
81 git_hash_unknown_init,
82 git_hash_unknown_update,
83 git_hash_unknown_final,
84 NULL,
85 NULL,
86 },
87 {
88 "sha-1",
89 /* "sha1", big-endian */
90 0x73686131,
91 sizeof(git_SHA_CTX),
92 GIT_SHA1_RAWSZ,
93 GIT_SHA1_HEXSZ,
94 git_hash_sha1_init,
95 git_hash_sha1_update,
96 git_hash_sha1_final,
97 &empty_tree_oid,
98 &empty_blob_oid,
99 },
100};
101
102/*
103 * This is meant to hold a *small* number of objects that you would
104 * want read_sha1_file() to be able to return, but yet you do not want
105 * to write them into the object store (e.g. a browse-only
106 * application).
107 */
108static struct cached_object {
109 unsigned char sha1[20];
110 enum object_type type;
111 void *buf;
112 unsigned long size;
113} *cached_objects;
114static int cached_object_nr, cached_object_alloc;
115
116static struct cached_object empty_tree = {
117 EMPTY_TREE_SHA1_BIN_LITERAL,
118 OBJ_TREE,
119 "",
120 0
121};
122
123static struct cached_object *find_cached_object(const unsigned char *sha1)
124{
125 int i;
126 struct cached_object *co = cached_objects;
127
128 for (i = 0; i < cached_object_nr; i++, co++) {
129 if (!hashcmp(co->sha1, sha1))
130 return co;
131 }
132 if (!hashcmp(sha1, empty_tree.sha1))
133 return &empty_tree;
134 return NULL;
135}
136
137
138static int get_conv_flags(unsigned flags)
139{
140 if (flags & HASH_RENORMALIZE)
141 return CONV_EOL_RENORMALIZE;
142 else if (flags & HASH_WRITE_OBJECT)
143 return global_conv_flags_eol;
144 else
145 return 0;
146}
147
148
149int mkdir_in_gitdir(const char *path)
150{
151 if (mkdir(path, 0777)) {
152 int saved_errno = errno;
153 struct stat st;
154 struct strbuf sb = STRBUF_INIT;
155
156 if (errno != EEXIST)
157 return -1;
158 /*
159 * Are we looking at a path in a symlinked worktree
160 * whose original repository does not yet have it?
161 * e.g. .git/rr-cache pointing at its original
162 * repository in which the user hasn't performed any
163 * conflict resolution yet?
164 */
165 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
166 strbuf_readlink(&sb, path, st.st_size) ||
167 !is_absolute_path(sb.buf) ||
168 mkdir(sb.buf, 0777)) {
169 strbuf_release(&sb);
170 errno = saved_errno;
171 return -1;
172 }
173 strbuf_release(&sb);
174 }
175 return adjust_shared_perm(path);
176}
177
178enum scld_error safe_create_leading_directories(char *path)
179{
180 char *next_component = path + offset_1st_component(path);
181 enum scld_error ret = SCLD_OK;
182
183 while (ret == SCLD_OK && next_component) {
184 struct stat st;
185 char *slash = next_component, slash_character;
186
187 while (*slash && !is_dir_sep(*slash))
188 slash++;
189
190 if (!*slash)
191 break;
192
193 next_component = slash + 1;
194 while (is_dir_sep(*next_component))
195 next_component++;
196 if (!*next_component)
197 break;
198
199 slash_character = *slash;
200 *slash = '\0';
201 if (!stat(path, &st)) {
202 /* path exists */
203 if (!S_ISDIR(st.st_mode)) {
204 errno = ENOTDIR;
205 ret = SCLD_EXISTS;
206 }
207 } else if (mkdir(path, 0777)) {
208 if (errno == EEXIST &&
209 !stat(path, &st) && S_ISDIR(st.st_mode))
210 ; /* somebody created it since we checked */
211 else if (errno == ENOENT)
212 /*
213 * Either mkdir() failed because
214 * somebody just pruned the containing
215 * directory, or stat() failed because
216 * the file that was in our way was
217 * just removed. Either way, inform
218 * the caller that it might be worth
219 * trying again:
220 */
221 ret = SCLD_VANISHED;
222 else
223 ret = SCLD_FAILED;
224 } else if (adjust_shared_perm(path)) {
225 ret = SCLD_PERMS;
226 }
227 *slash = slash_character;
228 }
229 return ret;
230}
231
232enum scld_error safe_create_leading_directories_const(const char *path)
233{
234 int save_errno;
235 /* path points to cache entries, so xstrdup before messing with it */
236 char *buf = xstrdup(path);
237 enum scld_error result = safe_create_leading_directories(buf);
238
239 save_errno = errno;
240 free(buf);
241 errno = save_errno;
242 return result;
243}
244
245int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
246{
247 /*
248 * The number of times we will try to remove empty directories
249 * in the way of path. This is only 1 because if another
250 * process is racily creating directories that conflict with
251 * us, we don't want to fight against them.
252 */
253 int remove_directories_remaining = 1;
254
255 /*
256 * The number of times that we will try to create the
257 * directories containing path. We are willing to attempt this
258 * more than once, because another process could be trying to
259 * clean up empty directories at the same time as we are
260 * trying to create them.
261 */
262 int create_directories_remaining = 3;
263
264 /* A scratch copy of path, filled lazily if we need it: */
265 struct strbuf path_copy = STRBUF_INIT;
266
267 int ret, save_errno;
268
269 /* Sanity check: */
270 assert(*path);
271
272retry_fn:
273 ret = fn(path, cb);
274 save_errno = errno;
275 if (!ret)
276 goto out;
277
278 if (errno == EISDIR && remove_directories_remaining-- > 0) {
279 /*
280 * A directory is in the way. Maybe it is empty; try
281 * to remove it:
282 */
283 if (!path_copy.len)
284 strbuf_addstr(&path_copy, path);
285
286 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
287 goto retry_fn;
288 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
289 /*
290 * Maybe the containing directory didn't exist, or
291 * maybe it was just deleted by a process that is
292 * racing with us to clean up empty directories. Try
293 * to create it:
294 */
295 enum scld_error scld_result;
296
297 if (!path_copy.len)
298 strbuf_addstr(&path_copy, path);
299
300 do {
301 scld_result = safe_create_leading_directories(path_copy.buf);
302 if (scld_result == SCLD_OK)
303 goto retry_fn;
304 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
305 }
306
307out:
308 strbuf_release(&path_copy);
309 errno = save_errno;
310 return ret;
311}
312
313static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
314{
315 int i;
316 for (i = 0; i < 20; i++) {
317 static char hex[] = "0123456789abcdef";
318 unsigned int val = sha1[i];
319 strbuf_addch(buf, hex[val >> 4]);
320 strbuf_addch(buf, hex[val & 0xf]);
321 if (!i)
322 strbuf_addch(buf, '/');
323 }
324}
325
326void sha1_file_name(struct strbuf *buf, const unsigned char *sha1)
327{
328 strbuf_addstr(buf, get_object_directory());
329 strbuf_addch(buf, '/');
330 fill_sha1_path(buf, sha1);
331}
332
333struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
334{
335 strbuf_setlen(&alt->scratch, alt->base_len);
336 return &alt->scratch;
337}
338
339static const char *alt_sha1_path(struct alternate_object_database *alt,
340 const unsigned char *sha1)
341{
342 struct strbuf *buf = alt_scratch_buf(alt);
343 fill_sha1_path(buf, sha1);
344 return buf->buf;
345}
346
347/*
348 * Return non-zero iff the path is usable as an alternate object database.
349 */
350static int alt_odb_usable(struct raw_object_store *o,
351 struct strbuf *path,
352 const char *normalized_objdir)
353{
354 struct alternate_object_database *alt;
355
356 /* Detect cases where alternate disappeared */
357 if (!is_directory(path->buf)) {
358 error("object directory %s does not exist; "
359 "check .git/objects/info/alternates.",
360 path->buf);
361 return 0;
362 }
363
364 /*
365 * Prevent the common mistake of listing the same
366 * thing twice, or object directory itself.
367 */
368 for (alt = o->alt_odb_list; alt; alt = alt->next) {
369 if (!fspathcmp(path->buf, alt->path))
370 return 0;
371 }
372 if (!fspathcmp(path->buf, normalized_objdir))
373 return 0;
374
375 return 1;
376}
377
378/*
379 * Prepare alternate object database registry.
380 *
381 * The variable alt_odb_list points at the list of struct
382 * alternate_object_database. The elements on this list come from
383 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
384 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
385 * whose contents is similar to that environment variable but can be
386 * LF separated. Its base points at a statically allocated buffer that
387 * contains "/the/directory/corresponding/to/.git/objects/...", while
388 * its name points just after the slash at the end of ".git/objects/"
389 * in the example above, and has enough space to hold 40-byte hex
390 * SHA1, an extra slash for the first level indirection, and the
391 * terminating NUL.
392 */
393#define read_info_alternates(r, rb, d) read_info_alternates_##r(rb, d)
394static void read_info_alternates_the_repository(const char *relative_base,
395 int depth);
396#define link_alt_odb_entry(r, e, rb, d, n) link_alt_odb_entry_##r(e, rb, d, n)
397static int link_alt_odb_entry_the_repository(const char *entry,
398 const char *relative_base, int depth, const char *normalized_objdir)
399{
400 struct alternate_object_database *ent;
401 struct strbuf pathbuf = STRBUF_INIT;
402
403 if (!is_absolute_path(entry) && relative_base) {
404 strbuf_realpath(&pathbuf, relative_base, 1);
405 strbuf_addch(&pathbuf, '/');
406 }
407 strbuf_addstr(&pathbuf, entry);
408
409 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
410 error("unable to normalize alternate object path: %s",
411 pathbuf.buf);
412 strbuf_release(&pathbuf);
413 return -1;
414 }
415
416 /*
417 * The trailing slash after the directory name is given by
418 * this function at the end. Remove duplicates.
419 */
420 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
421 strbuf_setlen(&pathbuf, pathbuf.len - 1);
422
423 if (!alt_odb_usable(the_repository->objects, &pathbuf, normalized_objdir)) {
424 strbuf_release(&pathbuf);
425 return -1;
426 }
427
428 ent = alloc_alt_odb(pathbuf.buf);
429
430 /* add the alternate entry */
431 *the_repository->objects->alt_odb_tail = ent;
432 the_repository->objects->alt_odb_tail = &(ent->next);
433 ent->next = NULL;
434
435 /* recursively add alternates */
436 read_info_alternates(the_repository, pathbuf.buf, depth + 1);
437
438 strbuf_release(&pathbuf);
439 return 0;
440}
441
442static const char *parse_alt_odb_entry(const char *string,
443 int sep,
444 struct strbuf *out)
445{
446 const char *end;
447
448 strbuf_reset(out);
449
450 if (*string == '#') {
451 /* comment; consume up to next separator */
452 end = strchrnul(string, sep);
453 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
454 /*
455 * quoted path; unquote_c_style has copied the
456 * data for us and set "end". Broken quoting (e.g.,
457 * an entry that doesn't end with a quote) falls
458 * back to the unquoted case below.
459 */
460 } else {
461 /* normal, unquoted path */
462 end = strchrnul(string, sep);
463 strbuf_add(out, string, end - string);
464 }
465
466 if (*end)
467 end++;
468 return end;
469}
470
471#define link_alt_odb_entries(r, a, s, rb, d) \
472 link_alt_odb_entries_##r(a, s, rb, d)
473static void link_alt_odb_entries_the_repository(const char *alt,
474 int sep,
475 const char *relative_base,
476 int depth)
477{
478 struct strbuf objdirbuf = STRBUF_INIT;
479 struct strbuf entry = STRBUF_INIT;
480
481 if (!alt || !*alt)
482 return;
483
484 if (depth > 5) {
485 error("%s: ignoring alternate object stores, nesting too deep.",
486 relative_base);
487 return;
488 }
489
490 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
491 if (strbuf_normalize_path(&objdirbuf) < 0)
492 die("unable to normalize object directory: %s",
493 objdirbuf.buf);
494
495 while (*alt) {
496 alt = parse_alt_odb_entry(alt, sep, &entry);
497 if (!entry.len)
498 continue;
499 link_alt_odb_entry(the_repository, entry.buf,
500 relative_base, depth, objdirbuf.buf);
501 }
502 strbuf_release(&entry);
503 strbuf_release(&objdirbuf);
504}
505
506static void read_info_alternates_the_repository(const char *relative_base,
507 int depth)
508{
509 char *path;
510 struct strbuf buf = STRBUF_INIT;
511
512 path = xstrfmt("%s/info/alternates", relative_base);
513 if (strbuf_read_file(&buf, path, 1024) < 0) {
514 warn_on_fopen_errors(path);
515 free(path);
516 return;
517 }
518
519 link_alt_odb_entries(the_repository, buf.buf, '\n', relative_base, depth);
520 strbuf_release(&buf);
521 free(path);
522}
523
524struct alternate_object_database *alloc_alt_odb(const char *dir)
525{
526 struct alternate_object_database *ent;
527
528 FLEX_ALLOC_STR(ent, path, dir);
529 strbuf_init(&ent->scratch, 0);
530 strbuf_addf(&ent->scratch, "%s/", dir);
531 ent->base_len = ent->scratch.len;
532
533 return ent;
534}
535
536void add_to_alternates_file(const char *reference)
537{
538 struct lock_file lock = LOCK_INIT;
539 char *alts = git_pathdup("objects/info/alternates");
540 FILE *in, *out;
541 int found = 0;
542
543 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
544 out = fdopen_lock_file(&lock, "w");
545 if (!out)
546 die_errno("unable to fdopen alternates lockfile");
547
548 in = fopen(alts, "r");
549 if (in) {
550 struct strbuf line = STRBUF_INIT;
551
552 while (strbuf_getline(&line, in) != EOF) {
553 if (!strcmp(reference, line.buf)) {
554 found = 1;
555 break;
556 }
557 fprintf_or_die(out, "%s\n", line.buf);
558 }
559
560 strbuf_release(&line);
561 fclose(in);
562 }
563 else if (errno != ENOENT)
564 die_errno("unable to read alternates file");
565
566 if (found) {
567 rollback_lock_file(&lock);
568 } else {
569 fprintf_or_die(out, "%s\n", reference);
570 if (commit_lock_file(&lock))
571 die_errno("unable to move new alternates file into place");
572 if (the_repository->objects->alt_odb_tail)
573 link_alt_odb_entries(the_repository, reference,
574 '\n', NULL, 0);
575 }
576 free(alts);
577}
578
579void add_to_alternates_memory(const char *reference)
580{
581 /*
582 * Make sure alternates are initialized, or else our entry may be
583 * overwritten when they are.
584 */
585 prepare_alt_odb();
586
587 link_alt_odb_entries(the_repository, reference,
588 '\n', NULL, 0);
589}
590
591/*
592 * Compute the exact path an alternate is at and returns it. In case of
593 * error NULL is returned and the human readable error is added to `err`
594 * `path` may be relative and should point to $GITDIR.
595 * `err` must not be null.
596 */
597char *compute_alternate_path(const char *path, struct strbuf *err)
598{
599 char *ref_git = NULL;
600 const char *repo, *ref_git_s;
601 int seen_error = 0;
602
603 ref_git_s = real_path_if_valid(path);
604 if (!ref_git_s) {
605 seen_error = 1;
606 strbuf_addf(err, _("path '%s' does not exist"), path);
607 goto out;
608 } else
609 /*
610 * Beware: read_gitfile(), real_path() and mkpath()
611 * return static buffer
612 */
613 ref_git = xstrdup(ref_git_s);
614
615 repo = read_gitfile(ref_git);
616 if (!repo)
617 repo = read_gitfile(mkpath("%s/.git", ref_git));
618 if (repo) {
619 free(ref_git);
620 ref_git = xstrdup(repo);
621 }
622
623 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
624 char *ref_git_git = mkpathdup("%s/.git", ref_git);
625 free(ref_git);
626 ref_git = ref_git_git;
627 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
628 struct strbuf sb = STRBUF_INIT;
629 seen_error = 1;
630 if (get_common_dir(&sb, ref_git)) {
631 strbuf_addf(err,
632 _("reference repository '%s' as a linked "
633 "checkout is not supported yet."),
634 path);
635 goto out;
636 }
637
638 strbuf_addf(err, _("reference repository '%s' is not a "
639 "local repository."), path);
640 goto out;
641 }
642
643 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
644 strbuf_addf(err, _("reference repository '%s' is shallow"),
645 path);
646 seen_error = 1;
647 goto out;
648 }
649
650 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
651 strbuf_addf(err,
652 _("reference repository '%s' is grafted"),
653 path);
654 seen_error = 1;
655 goto out;
656 }
657
658out:
659 if (seen_error) {
660 FREE_AND_NULL(ref_git);
661 }
662
663 return ref_git;
664}
665
666int foreach_alt_odb(alt_odb_fn fn, void *cb)
667{
668 struct alternate_object_database *ent;
669 int r = 0;
670
671 prepare_alt_odb();
672 for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
673 r = fn(ent, cb);
674 if (r)
675 break;
676 }
677 return r;
678}
679
680void prepare_alt_odb(void)
681{
682 if (the_repository->objects->alt_odb_tail)
683 return;
684
685 the_repository->objects->alt_odb_tail =
686 &the_repository->objects->alt_odb_list;
687 link_alt_odb_entries(the_repository,
688 the_repository->objects->alternate_db,
689 PATH_SEP, NULL, 0);
690
691 read_info_alternates(the_repository, get_object_directory(), 0);
692}
693
694/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
695static int freshen_file(const char *fn)
696{
697 struct utimbuf t;
698 t.actime = t.modtime = time(NULL);
699 return !utime(fn, &t);
700}
701
702/*
703 * All of the check_and_freshen functions return 1 if the file exists and was
704 * freshened (if freshening was requested), 0 otherwise. If they return
705 * 0, you should not assume that it is safe to skip a write of the object (it
706 * either does not exist on disk, or has a stale mtime and may be subject to
707 * pruning).
708 */
709int check_and_freshen_file(const char *fn, int freshen)
710{
711 if (access(fn, F_OK))
712 return 0;
713 if (freshen && !freshen_file(fn))
714 return 0;
715 return 1;
716}
717
718static int check_and_freshen_local(const unsigned char *sha1, int freshen)
719{
720 static struct strbuf buf = STRBUF_INIT;
721
722 strbuf_reset(&buf);
723 sha1_file_name(&buf, sha1);
724
725 return check_and_freshen_file(buf.buf, freshen);
726}
727
728static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
729{
730 struct alternate_object_database *alt;
731 prepare_alt_odb();
732 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
733 const char *path = alt_sha1_path(alt, sha1);
734 if (check_and_freshen_file(path, freshen))
735 return 1;
736 }
737 return 0;
738}
739
740static int check_and_freshen(const unsigned char *sha1, int freshen)
741{
742 return check_and_freshen_local(sha1, freshen) ||
743 check_and_freshen_nonlocal(sha1, freshen);
744}
745
746int has_loose_object_nonlocal(const unsigned char *sha1)
747{
748 return check_and_freshen_nonlocal(sha1, 0);
749}
750
751static int has_loose_object(const unsigned char *sha1)
752{
753 return check_and_freshen(sha1, 0);
754}
755
756static void mmap_limit_check(size_t length)
757{
758 static size_t limit = 0;
759 if (!limit) {
760 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
761 if (!limit)
762 limit = SIZE_MAX;
763 }
764 if (length > limit)
765 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
766 (uintmax_t)length, (uintmax_t)limit);
767}
768
769void *xmmap_gently(void *start, size_t length,
770 int prot, int flags, int fd, off_t offset)
771{
772 void *ret;
773
774 mmap_limit_check(length);
775 ret = mmap(start, length, prot, flags, fd, offset);
776 if (ret == MAP_FAILED) {
777 if (!length)
778 return NULL;
779 release_pack_memory(length);
780 ret = mmap(start, length, prot, flags, fd, offset);
781 }
782 return ret;
783}
784
785void *xmmap(void *start, size_t length,
786 int prot, int flags, int fd, off_t offset)
787{
788 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
789 if (ret == MAP_FAILED)
790 die_errno("mmap failed");
791 return ret;
792}
793
794/*
795 * With an in-core object data in "map", rehash it to make sure the
796 * object name actually matches "sha1" to detect object corruption.
797 * With "map" == NULL, try reading the object named with "sha1" using
798 * the streaming interface and rehash it to do the same.
799 */
800int check_sha1_signature(const unsigned char *sha1, void *map,
801 unsigned long size, const char *type)
802{
803 unsigned char real_sha1[20];
804 enum object_type obj_type;
805 struct git_istream *st;
806 git_SHA_CTX c;
807 char hdr[32];
808 int hdrlen;
809
810 if (map) {
811 hash_sha1_file(map, size, type, real_sha1);
812 return hashcmp(sha1, real_sha1) ? -1 : 0;
813 }
814
815 st = open_istream(sha1, &obj_type, &size, NULL);
816 if (!st)
817 return -1;
818
819 /* Generate the header */
820 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
821
822 /* Sha1.. */
823 git_SHA1_Init(&c);
824 git_SHA1_Update(&c, hdr, hdrlen);
825 for (;;) {
826 char buf[1024 * 16];
827 ssize_t readlen = read_istream(st, buf, sizeof(buf));
828
829 if (readlen < 0) {
830 close_istream(st);
831 return -1;
832 }
833 if (!readlen)
834 break;
835 git_SHA1_Update(&c, buf, readlen);
836 }
837 git_SHA1_Final(real_sha1, &c);
838 close_istream(st);
839 return hashcmp(sha1, real_sha1) ? -1 : 0;
840}
841
842int git_open_cloexec(const char *name, int flags)
843{
844 int fd;
845 static int o_cloexec = O_CLOEXEC;
846
847 fd = open(name, flags | o_cloexec);
848 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
849 /* Try again w/o O_CLOEXEC: the kernel might not support it */
850 o_cloexec &= ~O_CLOEXEC;
851 fd = open(name, flags | o_cloexec);
852 }
853
854#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
855 {
856 static int fd_cloexec = FD_CLOEXEC;
857
858 if (!o_cloexec && 0 <= fd && fd_cloexec) {
859 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
860 int flags = fcntl(fd, F_GETFD);
861 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
862 fd_cloexec = 0;
863 }
864 }
865#endif
866 return fd;
867}
868
869/*
870 * Find "sha1" as a loose object in the local repository or in an alternate.
871 * Returns 0 on success, negative on failure.
872 *
873 * The "path" out-parameter will give the path of the object we found (if any).
874 * Note that it may point to static storage and is only valid until another
875 * call to sha1_file_name(), etc.
876 */
877static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
878 const char **path)
879{
880 struct alternate_object_database *alt;
881 static struct strbuf buf = STRBUF_INIT;
882
883 strbuf_reset(&buf);
884 sha1_file_name(&buf, sha1);
885 *path = buf.buf;
886
887 if (!lstat(*path, st))
888 return 0;
889
890 prepare_alt_odb();
891 errno = ENOENT;
892 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
893 *path = alt_sha1_path(alt, sha1);
894 if (!lstat(*path, st))
895 return 0;
896 }
897
898 return -1;
899}
900
901/*
902 * Like stat_sha1_file(), but actually open the object and return the
903 * descriptor. See the caveats on the "path" parameter above.
904 */
905static int open_sha1_file(const unsigned char *sha1, const char **path)
906{
907 int fd;
908 struct alternate_object_database *alt;
909 int most_interesting_errno;
910 static struct strbuf buf = STRBUF_INIT;
911
912 strbuf_reset(&buf);
913 sha1_file_name(&buf, sha1);
914 *path = buf.buf;
915
916 fd = git_open(*path);
917 if (fd >= 0)
918 return fd;
919 most_interesting_errno = errno;
920
921 prepare_alt_odb();
922 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
923 *path = alt_sha1_path(alt, sha1);
924 fd = git_open(*path);
925 if (fd >= 0)
926 return fd;
927 if (most_interesting_errno == ENOENT)
928 most_interesting_errno = errno;
929 }
930 errno = most_interesting_errno;
931 return -1;
932}
933
934/*
935 * Map the loose object at "path" if it is not NULL, or the path found by
936 * searching for a loose object named "sha1".
937 */
938static void *map_sha1_file_1(const char *path,
939 const unsigned char *sha1,
940 unsigned long *size)
941{
942 void *map;
943 int fd;
944
945 if (path)
946 fd = git_open(path);
947 else
948 fd = open_sha1_file(sha1, &path);
949 map = NULL;
950 if (fd >= 0) {
951 struct stat st;
952
953 if (!fstat(fd, &st)) {
954 *size = xsize_t(st.st_size);
955 if (!*size) {
956 /* mmap() is forbidden on empty files */
957 error("object file %s is empty", path);
958 return NULL;
959 }
960 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
961 }
962 close(fd);
963 }
964 return map;
965}
966
967void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
968{
969 return map_sha1_file_1(NULL, sha1, size);
970}
971
972static int unpack_sha1_short_header(git_zstream *stream,
973 unsigned char *map, unsigned long mapsize,
974 void *buffer, unsigned long bufsiz)
975{
976 /* Get the data stream */
977 memset(stream, 0, sizeof(*stream));
978 stream->next_in = map;
979 stream->avail_in = mapsize;
980 stream->next_out = buffer;
981 stream->avail_out = bufsiz;
982
983 git_inflate_init(stream);
984 return git_inflate(stream, 0);
985}
986
987int unpack_sha1_header(git_zstream *stream,
988 unsigned char *map, unsigned long mapsize,
989 void *buffer, unsigned long bufsiz)
990{
991 int status = unpack_sha1_short_header(stream, map, mapsize,
992 buffer, bufsiz);
993
994 if (status < Z_OK)
995 return status;
996
997 /* Make sure we have the terminating NUL */
998 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
999 return -1;
1000 return 0;
1001}
1002
1003static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1004 unsigned long mapsize, void *buffer,
1005 unsigned long bufsiz, struct strbuf *header)
1006{
1007 int status;
1008
1009 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1010 if (status < Z_OK)
1011 return -1;
1012
1013 /*
1014 * Check if entire header is unpacked in the first iteration.
1015 */
1016 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1017 return 0;
1018
1019 /*
1020 * buffer[0..bufsiz] was not large enough. Copy the partial
1021 * result out to header, and then append the result of further
1022 * reading the stream.
1023 */
1024 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1025 stream->next_out = buffer;
1026 stream->avail_out = bufsiz;
1027
1028 do {
1029 status = git_inflate(stream, 0);
1030 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1031 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1032 return 0;
1033 stream->next_out = buffer;
1034 stream->avail_out = bufsiz;
1035 } while (status != Z_STREAM_END);
1036 return -1;
1037}
1038
1039static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1040{
1041 int bytes = strlen(buffer) + 1;
1042 unsigned char *buf = xmallocz(size);
1043 unsigned long n;
1044 int status = Z_OK;
1045
1046 n = stream->total_out - bytes;
1047 if (n > size)
1048 n = size;
1049 memcpy(buf, (char *) buffer + bytes, n);
1050 bytes = n;
1051 if (bytes <= size) {
1052 /*
1053 * The above condition must be (bytes <= size), not
1054 * (bytes < size). In other words, even though we
1055 * expect no more output and set avail_out to zero,
1056 * the input zlib stream may have bytes that express
1057 * "this concludes the stream", and we *do* want to
1058 * eat that input.
1059 *
1060 * Otherwise we would not be able to test that we
1061 * consumed all the input to reach the expected size;
1062 * we also want to check that zlib tells us that all
1063 * went well with status == Z_STREAM_END at the end.
1064 */
1065 stream->next_out = buf + bytes;
1066 stream->avail_out = size - bytes;
1067 while (status == Z_OK)
1068 status = git_inflate(stream, Z_FINISH);
1069 }
1070 if (status == Z_STREAM_END && !stream->avail_in) {
1071 git_inflate_end(stream);
1072 return buf;
1073 }
1074
1075 if (status < 0)
1076 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1077 else if (stream->avail_in)
1078 error("garbage at end of loose object '%s'",
1079 sha1_to_hex(sha1));
1080 free(buf);
1081 return NULL;
1082}
1083
1084/*
1085 * We used to just use "sscanf()", but that's actually way
1086 * too permissive for what we want to check. So do an anal
1087 * object header parse by hand.
1088 */
1089static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1090 unsigned int flags)
1091{
1092 const char *type_buf = hdr;
1093 unsigned long size;
1094 int type, type_len = 0;
1095
1096 /*
1097 * The type can be of any size but is followed by
1098 * a space.
1099 */
1100 for (;;) {
1101 char c = *hdr++;
1102 if (!c)
1103 return -1;
1104 if (c == ' ')
1105 break;
1106 type_len++;
1107 }
1108
1109 type = type_from_string_gently(type_buf, type_len, 1);
1110 if (oi->typename)
1111 strbuf_add(oi->typename, type_buf, type_len);
1112 /*
1113 * Set type to 0 if its an unknown object and
1114 * we're obtaining the type using '--allow-unknown-type'
1115 * option.
1116 */
1117 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1118 type = 0;
1119 else if (type < 0)
1120 die("invalid object type");
1121 if (oi->typep)
1122 *oi->typep = type;
1123
1124 /*
1125 * The length must follow immediately, and be in canonical
1126 * decimal format (ie "010" is not valid).
1127 */
1128 size = *hdr++ - '0';
1129 if (size > 9)
1130 return -1;
1131 if (size) {
1132 for (;;) {
1133 unsigned long c = *hdr - '0';
1134 if (c > 9)
1135 break;
1136 hdr++;
1137 size = size * 10 + c;
1138 }
1139 }
1140
1141 if (oi->sizep)
1142 *oi->sizep = size;
1143
1144 /*
1145 * The length must be followed by a zero byte
1146 */
1147 return *hdr ? -1 : type;
1148}
1149
1150int parse_sha1_header(const char *hdr, unsigned long *sizep)
1151{
1152 struct object_info oi = OBJECT_INFO_INIT;
1153
1154 oi.sizep = sizep;
1155 return parse_sha1_header_extended(hdr, &oi, 0);
1156}
1157
1158static int sha1_loose_object_info(const unsigned char *sha1,
1159 struct object_info *oi,
1160 int flags)
1161{
1162 int status = 0;
1163 unsigned long mapsize;
1164 void *map;
1165 git_zstream stream;
1166 char hdr[32];
1167 struct strbuf hdrbuf = STRBUF_INIT;
1168 unsigned long size_scratch;
1169
1170 if (oi->delta_base_sha1)
1171 hashclr(oi->delta_base_sha1);
1172
1173 /*
1174 * If we don't care about type or size, then we don't
1175 * need to look inside the object at all. Note that we
1176 * do not optimize out the stat call, even if the
1177 * caller doesn't care about the disk-size, since our
1178 * return value implicitly indicates whether the
1179 * object even exists.
1180 */
1181 if (!oi->typep && !oi->typename && !oi->sizep && !oi->contentp) {
1182 const char *path;
1183 struct stat st;
1184 if (stat_sha1_file(sha1, &st, &path) < 0)
1185 return -1;
1186 if (oi->disk_sizep)
1187 *oi->disk_sizep = st.st_size;
1188 return 0;
1189 }
1190
1191 map = map_sha1_file(sha1, &mapsize);
1192 if (!map)
1193 return -1;
1194
1195 if (!oi->sizep)
1196 oi->sizep = &size_scratch;
1197
1198 if (oi->disk_sizep)
1199 *oi->disk_sizep = mapsize;
1200 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1201 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1202 status = error("unable to unpack %s header with --allow-unknown-type",
1203 sha1_to_hex(sha1));
1204 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1205 status = error("unable to unpack %s header",
1206 sha1_to_hex(sha1));
1207 if (status < 0)
1208 ; /* Do nothing */
1209 else if (hdrbuf.len) {
1210 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1211 status = error("unable to parse %s header with --allow-unknown-type",
1212 sha1_to_hex(sha1));
1213 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1214 status = error("unable to parse %s header", sha1_to_hex(sha1));
1215
1216 if (status >= 0 && oi->contentp) {
1217 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1218 *oi->sizep, sha1);
1219 if (!*oi->contentp) {
1220 git_inflate_end(&stream);
1221 status = -1;
1222 }
1223 } else
1224 git_inflate_end(&stream);
1225
1226 munmap(map, mapsize);
1227 if (status && oi->typep)
1228 *oi->typep = status;
1229 if (oi->sizep == &size_scratch)
1230 oi->sizep = NULL;
1231 strbuf_release(&hdrbuf);
1232 oi->whence = OI_LOOSE;
1233 return (status < 0) ? status : 0;
1234}
1235
1236int fetch_if_missing = 1;
1237
1238int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
1239{
1240 static struct object_info blank_oi = OBJECT_INFO_INIT;
1241 struct pack_entry e;
1242 int rtype;
1243 const unsigned char *real = (flags & OBJECT_INFO_LOOKUP_REPLACE) ?
1244 lookup_replace_object(sha1) :
1245 sha1;
1246 int already_retried = 0;
1247
1248 if (is_null_sha1(real))
1249 return -1;
1250
1251 if (!oi)
1252 oi = &blank_oi;
1253
1254 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1255 struct cached_object *co = find_cached_object(real);
1256 if (co) {
1257 if (oi->typep)
1258 *(oi->typep) = co->type;
1259 if (oi->sizep)
1260 *(oi->sizep) = co->size;
1261 if (oi->disk_sizep)
1262 *(oi->disk_sizep) = 0;
1263 if (oi->delta_base_sha1)
1264 hashclr(oi->delta_base_sha1);
1265 if (oi->typename)
1266 strbuf_addstr(oi->typename, typename(co->type));
1267 if (oi->contentp)
1268 *oi->contentp = xmemdupz(co->buf, co->size);
1269 oi->whence = OI_CACHED;
1270 return 0;
1271 }
1272 }
1273
1274 while (1) {
1275 if (find_pack_entry(real, &e))
1276 break;
1277
1278 /* Most likely it's a loose object. */
1279 if (!sha1_loose_object_info(real, oi, flags))
1280 return 0;
1281
1282 /* Not a loose object; someone else may have just packed it. */
1283 reprepare_packed_git();
1284 if (find_pack_entry(real, &e))
1285 break;
1286
1287 /* Check if it is a missing object */
1288 if (fetch_if_missing && repository_format_partial_clone &&
1289 !already_retried) {
1290 /*
1291 * TODO Investigate haveing fetch_object() return
1292 * TODO error/success and stopping the music here.
1293 */
1294 fetch_object(repository_format_partial_clone, real);
1295 already_retried = 1;
1296 continue;
1297 }
1298
1299 return -1;
1300 }
1301
1302 if (oi == &blank_oi)
1303 /*
1304 * We know that the caller doesn't actually need the
1305 * information below, so return early.
1306 */
1307 return 0;
1308 rtype = packed_object_info(e.p, e.offset, oi);
1309 if (rtype < 0) {
1310 mark_bad_packed_object(e.p, real);
1311 return sha1_object_info_extended(real, oi, 0);
1312 } else if (oi->whence == OI_PACKED) {
1313 oi->u.packed.offset = e.offset;
1314 oi->u.packed.pack = e.p;
1315 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1316 rtype == OBJ_OFS_DELTA);
1317 }
1318
1319 return 0;
1320}
1321
1322/* returns enum object_type or negative */
1323int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
1324{
1325 enum object_type type;
1326 struct object_info oi = OBJECT_INFO_INIT;
1327
1328 oi.typep = &type;
1329 oi.sizep = sizep;
1330 if (sha1_object_info_extended(sha1, &oi,
1331 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1332 return -1;
1333 return type;
1334}
1335
1336static void *read_object(const unsigned char *sha1, enum object_type *type,
1337 unsigned long *size)
1338{
1339 struct object_info oi = OBJECT_INFO_INIT;
1340 void *content;
1341 oi.typep = type;
1342 oi.sizep = size;
1343 oi.contentp = &content;
1344
1345 if (sha1_object_info_extended(sha1, &oi, 0) < 0)
1346 return NULL;
1347 return content;
1348}
1349
1350int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
1351 unsigned char *sha1)
1352{
1353 struct cached_object *co;
1354
1355 hash_sha1_file(buf, len, typename(type), sha1);
1356 if (has_sha1_file(sha1) || find_cached_object(sha1))
1357 return 0;
1358 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1359 co = &cached_objects[cached_object_nr++];
1360 co->size = len;
1361 co->type = type;
1362 co->buf = xmalloc(len);
1363 memcpy(co->buf, buf, len);
1364 hashcpy(co->sha1, sha1);
1365 return 0;
1366}
1367
1368/*
1369 * This function dies on corrupt objects; the callers who want to
1370 * deal with them should arrange to call read_object() and give error
1371 * messages themselves.
1372 */
1373void *read_sha1_file_extended(const unsigned char *sha1,
1374 enum object_type *type,
1375 unsigned long *size,
1376 int lookup_replace)
1377{
1378 void *data;
1379 const struct packed_git *p;
1380 const char *path;
1381 struct stat st;
1382 const unsigned char *repl = lookup_replace ? lookup_replace_object(sha1)
1383 : sha1;
1384
1385 errno = 0;
1386 data = read_object(repl, type, size);
1387 if (data)
1388 return data;
1389
1390 if (errno && errno != ENOENT)
1391 die_errno("failed to read object %s", sha1_to_hex(sha1));
1392
1393 /* die if we replaced an object with one that does not exist */
1394 if (repl != sha1)
1395 die("replacement %s not found for %s",
1396 sha1_to_hex(repl), sha1_to_hex(sha1));
1397
1398 if (!stat_sha1_file(repl, &st, &path))
1399 die("loose object %s (stored in %s) is corrupt",
1400 sha1_to_hex(repl), path);
1401
1402 if ((p = has_packed_and_bad(repl)) != NULL)
1403 die("packed object %s (stored in %s) is corrupt",
1404 sha1_to_hex(repl), p->pack_name);
1405
1406 return NULL;
1407}
1408
1409void *read_object_with_reference(const unsigned char *sha1,
1410 const char *required_type_name,
1411 unsigned long *size,
1412 unsigned char *actual_sha1_return)
1413{
1414 enum object_type type, required_type;
1415 void *buffer;
1416 unsigned long isize;
1417 unsigned char actual_sha1[20];
1418
1419 required_type = type_from_string(required_type_name);
1420 hashcpy(actual_sha1, sha1);
1421 while (1) {
1422 int ref_length = -1;
1423 const char *ref_type = NULL;
1424
1425 buffer = read_sha1_file(actual_sha1, &type, &isize);
1426 if (!buffer)
1427 return NULL;
1428 if (type == required_type) {
1429 *size = isize;
1430 if (actual_sha1_return)
1431 hashcpy(actual_sha1_return, actual_sha1);
1432 return buffer;
1433 }
1434 /* Handle references */
1435 else if (type == OBJ_COMMIT)
1436 ref_type = "tree ";
1437 else if (type == OBJ_TAG)
1438 ref_type = "object ";
1439 else {
1440 free(buffer);
1441 return NULL;
1442 }
1443 ref_length = strlen(ref_type);
1444
1445 if (ref_length + 40 > isize ||
1446 memcmp(buffer, ref_type, ref_length) ||
1447 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1448 free(buffer);
1449 return NULL;
1450 }
1451 free(buffer);
1452 /* Now we have the ID of the referred-to object in
1453 * actual_sha1. Check again. */
1454 }
1455}
1456
1457static void write_sha1_file_prepare(const void *buf, unsigned long len,
1458 const char *type, unsigned char *sha1,
1459 char *hdr, int *hdrlen)
1460{
1461 git_SHA_CTX c;
1462
1463 /* Generate the header */
1464 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1465
1466 /* Sha1.. */
1467 git_SHA1_Init(&c);
1468 git_SHA1_Update(&c, hdr, *hdrlen);
1469 git_SHA1_Update(&c, buf, len);
1470 git_SHA1_Final(sha1, &c);
1471}
1472
1473/*
1474 * Move the just written object into its final resting place.
1475 */
1476int finalize_object_file(const char *tmpfile, const char *filename)
1477{
1478 int ret = 0;
1479
1480 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1481 goto try_rename;
1482 else if (link(tmpfile, filename))
1483 ret = errno;
1484
1485 /*
1486 * Coda hack - coda doesn't like cross-directory links,
1487 * so we fall back to a rename, which will mean that it
1488 * won't be able to check collisions, but that's not a
1489 * big deal.
1490 *
1491 * The same holds for FAT formatted media.
1492 *
1493 * When this succeeds, we just return. We have nothing
1494 * left to unlink.
1495 */
1496 if (ret && ret != EEXIST) {
1497 try_rename:
1498 if (!rename(tmpfile, filename))
1499 goto out;
1500 ret = errno;
1501 }
1502 unlink_or_warn(tmpfile);
1503 if (ret) {
1504 if (ret != EEXIST) {
1505 return error_errno("unable to write sha1 filename %s", filename);
1506 }
1507 /* FIXME!!! Collision check here ? */
1508 }
1509
1510out:
1511 if (adjust_shared_perm(filename))
1512 return error("unable to set permission to '%s'", filename);
1513 return 0;
1514}
1515
1516static int write_buffer(int fd, const void *buf, size_t len)
1517{
1518 if (write_in_full(fd, buf, len) < 0)
1519 return error_errno("file write error");
1520 return 0;
1521}
1522
1523int hash_sha1_file(const void *buf, unsigned long len, const char *type,
1524 unsigned char *sha1)
1525{
1526 char hdr[32];
1527 int hdrlen = sizeof(hdr);
1528 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1529 return 0;
1530}
1531
1532/* Finalize a file on disk, and close it. */
1533static void close_sha1_file(int fd)
1534{
1535 if (fsync_object_files)
1536 fsync_or_die(fd, "sha1 file");
1537 if (close(fd) != 0)
1538 die_errno("error when closing sha1 file");
1539}
1540
1541/* Size of directory component, including the ending '/' */
1542static inline int directory_size(const char *filename)
1543{
1544 const char *s = strrchr(filename, '/');
1545 if (!s)
1546 return 0;
1547 return s - filename + 1;
1548}
1549
1550/*
1551 * This creates a temporary file in the same directory as the final
1552 * 'filename'
1553 *
1554 * We want to avoid cross-directory filename renames, because those
1555 * can have problems on various filesystems (FAT, NFS, Coda).
1556 */
1557static int create_tmpfile(struct strbuf *tmp, const char *filename)
1558{
1559 int fd, dirlen = directory_size(filename);
1560
1561 strbuf_reset(tmp);
1562 strbuf_add(tmp, filename, dirlen);
1563 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1564 fd = git_mkstemp_mode(tmp->buf, 0444);
1565 if (fd < 0 && dirlen && errno == ENOENT) {
1566 /*
1567 * Make sure the directory exists; note that the contents
1568 * of the buffer are undefined after mkstemp returns an
1569 * error, so we have to rewrite the whole buffer from
1570 * scratch.
1571 */
1572 strbuf_reset(tmp);
1573 strbuf_add(tmp, filename, dirlen - 1);
1574 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1575 return -1;
1576 if (adjust_shared_perm(tmp->buf))
1577 return -1;
1578
1579 /* Try again */
1580 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1581 fd = git_mkstemp_mode(tmp->buf, 0444);
1582 }
1583 return fd;
1584}
1585
1586static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
1587 const void *buf, unsigned long len, time_t mtime)
1588{
1589 int fd, ret;
1590 unsigned char compressed[4096];
1591 git_zstream stream;
1592 git_SHA_CTX c;
1593 unsigned char parano_sha1[20];
1594 static struct strbuf tmp_file = STRBUF_INIT;
1595 static struct strbuf filename = STRBUF_INIT;
1596
1597 strbuf_reset(&filename);
1598 sha1_file_name(&filename, sha1);
1599
1600 fd = create_tmpfile(&tmp_file, filename.buf);
1601 if (fd < 0) {
1602 if (errno == EACCES)
1603 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
1604 else
1605 return error_errno("unable to create temporary file");
1606 }
1607
1608 /* Set it up */
1609 git_deflate_init(&stream, zlib_compression_level);
1610 stream.next_out = compressed;
1611 stream.avail_out = sizeof(compressed);
1612 git_SHA1_Init(&c);
1613
1614 /* First header.. */
1615 stream.next_in = (unsigned char *)hdr;
1616 stream.avail_in = hdrlen;
1617 while (git_deflate(&stream, 0) == Z_OK)
1618 ; /* nothing */
1619 git_SHA1_Update(&c, hdr, hdrlen);
1620
1621 /* Then the data itself.. */
1622 stream.next_in = (void *)buf;
1623 stream.avail_in = len;
1624 do {
1625 unsigned char *in0 = stream.next_in;
1626 ret = git_deflate(&stream, Z_FINISH);
1627 git_SHA1_Update(&c, in0, stream.next_in - in0);
1628 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1629 die("unable to write sha1 file");
1630 stream.next_out = compressed;
1631 stream.avail_out = sizeof(compressed);
1632 } while (ret == Z_OK);
1633
1634 if (ret != Z_STREAM_END)
1635 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
1636 ret = git_deflate_end_gently(&stream);
1637 if (ret != Z_OK)
1638 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
1639 git_SHA1_Final(parano_sha1, &c);
1640 if (hashcmp(sha1, parano_sha1) != 0)
1641 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
1642
1643 close_sha1_file(fd);
1644
1645 if (mtime) {
1646 struct utimbuf utb;
1647 utb.actime = mtime;
1648 utb.modtime = mtime;
1649 if (utime(tmp_file.buf, &utb) < 0)
1650 warning_errno("failed utime() on %s", tmp_file.buf);
1651 }
1652
1653 return finalize_object_file(tmp_file.buf, filename.buf);
1654}
1655
1656static int freshen_loose_object(const unsigned char *sha1)
1657{
1658 return check_and_freshen(sha1, 1);
1659}
1660
1661static int freshen_packed_object(const unsigned char *sha1)
1662{
1663 struct pack_entry e;
1664 if (!find_pack_entry(sha1, &e))
1665 return 0;
1666 if (e.p->freshened)
1667 return 1;
1668 if (!freshen_file(e.p->pack_name))
1669 return 0;
1670 e.p->freshened = 1;
1671 return 1;
1672}
1673
1674int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
1675{
1676 char hdr[32];
1677 int hdrlen = sizeof(hdr);
1678
1679 /* Normally if we have it in the pack then we do not bother writing
1680 * it out into .git/objects/??/?{38} file.
1681 */
1682 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1683 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
1684 return 0;
1685 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
1686}
1687
1688int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
1689 struct object_id *oid, unsigned flags)
1690{
1691 char *header;
1692 int hdrlen, status = 0;
1693
1694 /* type string, SP, %lu of the length plus NUL must fit this */
1695 hdrlen = strlen(type) + 32;
1696 header = xmalloc(hdrlen);
1697 write_sha1_file_prepare(buf, len, type, oid->hash, header, &hdrlen);
1698
1699 if (!(flags & HASH_WRITE_OBJECT))
1700 goto cleanup;
1701 if (freshen_packed_object(oid->hash) || freshen_loose_object(oid->hash))
1702 goto cleanup;
1703 status = write_loose_object(oid->hash, header, hdrlen, buf, len, 0);
1704
1705cleanup:
1706 free(header);
1707 return status;
1708}
1709
1710int force_object_loose(const unsigned char *sha1, time_t mtime)
1711{
1712 void *buf;
1713 unsigned long len;
1714 enum object_type type;
1715 char hdr[32];
1716 int hdrlen;
1717 int ret;
1718
1719 if (has_loose_object(sha1))
1720 return 0;
1721 buf = read_object(sha1, &type, &len);
1722 if (!buf)
1723 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1724 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1725 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
1726 free(buf);
1727
1728 return ret;
1729}
1730
1731int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1732{
1733 if (!startup_info->have_repository)
1734 return 0;
1735 return sha1_object_info_extended(sha1, NULL,
1736 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1737}
1738
1739int has_object_file(const struct object_id *oid)
1740{
1741 return has_sha1_file(oid->hash);
1742}
1743
1744int has_object_file_with_flags(const struct object_id *oid, int flags)
1745{
1746 return has_sha1_file_with_flags(oid->hash, flags);
1747}
1748
1749static void check_tree(const void *buf, size_t size)
1750{
1751 struct tree_desc desc;
1752 struct name_entry entry;
1753
1754 init_tree_desc(&desc, buf, size);
1755 while (tree_entry(&desc, &entry))
1756 /* do nothing
1757 * tree_entry() will die() on malformed entries */
1758 ;
1759}
1760
1761static void check_commit(const void *buf, size_t size)
1762{
1763 struct commit c;
1764 memset(&c, 0, sizeof(c));
1765 if (parse_commit_buffer(&c, buf, size))
1766 die("corrupt commit");
1767}
1768
1769static void check_tag(const void *buf, size_t size)
1770{
1771 struct tag t;
1772 memset(&t, 0, sizeof(t));
1773 if (parse_tag_buffer(&t, buf, size))
1774 die("corrupt tag");
1775}
1776
1777static int index_mem(struct object_id *oid, void *buf, size_t size,
1778 enum object_type type,
1779 const char *path, unsigned flags)
1780{
1781 int ret, re_allocated = 0;
1782 int write_object = flags & HASH_WRITE_OBJECT;
1783
1784 if (!type)
1785 type = OBJ_BLOB;
1786
1787 /*
1788 * Convert blobs to git internal format
1789 */
1790 if ((type == OBJ_BLOB) && path) {
1791 struct strbuf nbuf = STRBUF_INIT;
1792 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1793 get_conv_flags(flags))) {
1794 buf = strbuf_detach(&nbuf, &size);
1795 re_allocated = 1;
1796 }
1797 }
1798 if (flags & HASH_FORMAT_CHECK) {
1799 if (type == OBJ_TREE)
1800 check_tree(buf, size);
1801 if (type == OBJ_COMMIT)
1802 check_commit(buf, size);
1803 if (type == OBJ_TAG)
1804 check_tag(buf, size);
1805 }
1806
1807 if (write_object)
1808 ret = write_sha1_file(buf, size, typename(type), oid->hash);
1809 else
1810 ret = hash_sha1_file(buf, size, typename(type), oid->hash);
1811 if (re_allocated)
1812 free(buf);
1813 return ret;
1814}
1815
1816static int index_stream_convert_blob(struct object_id *oid, int fd,
1817 const char *path, unsigned flags)
1818{
1819 int ret;
1820 const int write_object = flags & HASH_WRITE_OBJECT;
1821 struct strbuf sbuf = STRBUF_INIT;
1822
1823 assert(path);
1824 assert(would_convert_to_git_filter_fd(path));
1825
1826 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1827 get_conv_flags(flags));
1828
1829 if (write_object)
1830 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1831 oid->hash);
1832 else
1833 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
1834 oid->hash);
1835 strbuf_release(&sbuf);
1836 return ret;
1837}
1838
1839static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1840 const char *path, unsigned flags)
1841{
1842 struct strbuf sbuf = STRBUF_INIT;
1843 int ret;
1844
1845 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1846 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1847 else
1848 ret = -1;
1849 strbuf_release(&sbuf);
1850 return ret;
1851}
1852
1853#define SMALL_FILE_SIZE (32*1024)
1854
1855static int index_core(struct object_id *oid, int fd, size_t size,
1856 enum object_type type, const char *path,
1857 unsigned flags)
1858{
1859 int ret;
1860
1861 if (!size) {
1862 ret = index_mem(oid, "", size, type, path, flags);
1863 } else if (size <= SMALL_FILE_SIZE) {
1864 char *buf = xmalloc(size);
1865 ssize_t read_result = read_in_full(fd, buf, size);
1866 if (read_result < 0)
1867 ret = error_errno("read error while indexing %s",
1868 path ? path : "<unknown>");
1869 else if (read_result != size)
1870 ret = error("short read while indexing %s",
1871 path ? path : "<unknown>");
1872 else
1873 ret = index_mem(oid, buf, size, type, path, flags);
1874 free(buf);
1875 } else {
1876 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1877 ret = index_mem(oid, buf, size, type, path, flags);
1878 munmap(buf, size);
1879 }
1880 return ret;
1881}
1882
1883/*
1884 * This creates one packfile per large blob unless bulk-checkin
1885 * machinery is "plugged".
1886 *
1887 * This also bypasses the usual "convert-to-git" dance, and that is on
1888 * purpose. We could write a streaming version of the converting
1889 * functions and insert that before feeding the data to fast-import
1890 * (or equivalent in-core API described above). However, that is
1891 * somewhat complicated, as we do not know the size of the filter
1892 * result, which we need to know beforehand when writing a git object.
1893 * Since the primary motivation for trying to stream from the working
1894 * tree file and to avoid mmaping it in core is to deal with large
1895 * binary blobs, they generally do not want to get any conversion, and
1896 * callers should avoid this code path when filters are requested.
1897 */
1898static int index_stream(struct object_id *oid, int fd, size_t size,
1899 enum object_type type, const char *path,
1900 unsigned flags)
1901{
1902 return index_bulk_checkin(oid->hash, fd, size, type, path, flags);
1903}
1904
1905int index_fd(struct object_id *oid, int fd, struct stat *st,
1906 enum object_type type, const char *path, unsigned flags)
1907{
1908 int ret;
1909
1910 /*
1911 * Call xsize_t() only when needed to avoid potentially unnecessary
1912 * die() for large files.
1913 */
1914 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
1915 ret = index_stream_convert_blob(oid, fd, path, flags);
1916 else if (!S_ISREG(st->st_mode))
1917 ret = index_pipe(oid, fd, type, path, flags);
1918 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1919 (path && would_convert_to_git(&the_index, path)))
1920 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1921 flags);
1922 else
1923 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1924 flags);
1925 close(fd);
1926 return ret;
1927}
1928
1929int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1930{
1931 int fd;
1932 struct strbuf sb = STRBUF_INIT;
1933 int rc = 0;
1934
1935 switch (st->st_mode & S_IFMT) {
1936 case S_IFREG:
1937 fd = open(path, O_RDONLY);
1938 if (fd < 0)
1939 return error_errno("open(\"%s\")", path);
1940 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
1941 return error("%s: failed to insert into database",
1942 path);
1943 break;
1944 case S_IFLNK:
1945 if (strbuf_readlink(&sb, path, st->st_size))
1946 return error_errno("readlink(\"%s\")", path);
1947 if (!(flags & HASH_WRITE_OBJECT))
1948 hash_sha1_file(sb.buf, sb.len, blob_type, oid->hash);
1949 else if (write_sha1_file(sb.buf, sb.len, blob_type, oid->hash))
1950 rc = error("%s: failed to insert into database", path);
1951 strbuf_release(&sb);
1952 break;
1953 case S_IFDIR:
1954 return resolve_gitlink_ref(path, "HEAD", oid);
1955 default:
1956 return error("%s: unsupported file type", path);
1957 }
1958 return rc;
1959}
1960
1961int read_pack_header(int fd, struct pack_header *header)
1962{
1963 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
1964 /* "eof before pack header was fully read" */
1965 return PH_ERROR_EOF;
1966
1967 if (header->hdr_signature != htonl(PACK_SIGNATURE))
1968 /* "protocol error (pack signature mismatch detected)" */
1969 return PH_ERROR_PACK_SIGNATURE;
1970 if (!pack_version_ok(header->hdr_version))
1971 /* "protocol error (pack version unsupported)" */
1972 return PH_ERROR_PROTOCOL;
1973 return 0;
1974}
1975
1976void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
1977{
1978 enum object_type type = sha1_object_info(sha1, NULL);
1979 if (type < 0)
1980 die("%s is not a valid object", sha1_to_hex(sha1));
1981 if (type != expect)
1982 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
1983 typename(expect));
1984}
1985
1986int for_each_file_in_obj_subdir(unsigned int subdir_nr,
1987 struct strbuf *path,
1988 each_loose_object_fn obj_cb,
1989 each_loose_cruft_fn cruft_cb,
1990 each_loose_subdir_fn subdir_cb,
1991 void *data)
1992{
1993 size_t origlen, baselen;
1994 DIR *dir;
1995 struct dirent *de;
1996 int r = 0;
1997 struct object_id oid;
1998
1999 if (subdir_nr > 0xff)
2000 BUG("invalid loose object subdirectory: %x", subdir_nr);
2001
2002 origlen = path->len;
2003 strbuf_complete(path, '/');
2004 strbuf_addf(path, "%02x", subdir_nr);
2005
2006 dir = opendir(path->buf);
2007 if (!dir) {
2008 if (errno != ENOENT)
2009 r = error_errno("unable to open %s", path->buf);
2010 strbuf_setlen(path, origlen);
2011 return r;
2012 }
2013
2014 oid.hash[0] = subdir_nr;
2015 strbuf_addch(path, '/');
2016 baselen = path->len;
2017
2018 while ((de = readdir(dir))) {
2019 size_t namelen;
2020 if (is_dot_or_dotdot(de->d_name))
2021 continue;
2022
2023 namelen = strlen(de->d_name);
2024 strbuf_setlen(path, baselen);
2025 strbuf_add(path, de->d_name, namelen);
2026 if (namelen == GIT_SHA1_HEXSZ - 2 &&
2027 !hex_to_bytes(oid.hash + 1, de->d_name,
2028 GIT_SHA1_RAWSZ - 1)) {
2029 if (obj_cb) {
2030 r = obj_cb(&oid, path->buf, data);
2031 if (r)
2032 break;
2033 }
2034 continue;
2035 }
2036
2037 if (cruft_cb) {
2038 r = cruft_cb(de->d_name, path->buf, data);
2039 if (r)
2040 break;
2041 }
2042 }
2043 closedir(dir);
2044
2045 strbuf_setlen(path, baselen - 1);
2046 if (!r && subdir_cb)
2047 r = subdir_cb(subdir_nr, path->buf, data);
2048
2049 strbuf_setlen(path, origlen);
2050
2051 return r;
2052}
2053
2054int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2055 each_loose_object_fn obj_cb,
2056 each_loose_cruft_fn cruft_cb,
2057 each_loose_subdir_fn subdir_cb,
2058 void *data)
2059{
2060 int r = 0;
2061 int i;
2062
2063 for (i = 0; i < 256; i++) {
2064 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2065 subdir_cb, data);
2066 if (r)
2067 break;
2068 }
2069
2070 return r;
2071}
2072
2073int for_each_loose_file_in_objdir(const char *path,
2074 each_loose_object_fn obj_cb,
2075 each_loose_cruft_fn cruft_cb,
2076 each_loose_subdir_fn subdir_cb,
2077 void *data)
2078{
2079 struct strbuf buf = STRBUF_INIT;
2080 int r;
2081
2082 strbuf_addstr(&buf, path);
2083 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2084 subdir_cb, data);
2085 strbuf_release(&buf);
2086
2087 return r;
2088}
2089
2090struct loose_alt_odb_data {
2091 each_loose_object_fn *cb;
2092 void *data;
2093};
2094
2095static int loose_from_alt_odb(struct alternate_object_database *alt,
2096 void *vdata)
2097{
2098 struct loose_alt_odb_data *data = vdata;
2099 struct strbuf buf = STRBUF_INIT;
2100 int r;
2101
2102 strbuf_addstr(&buf, alt->path);
2103 r = for_each_loose_file_in_objdir_buf(&buf,
2104 data->cb, NULL, NULL,
2105 data->data);
2106 strbuf_release(&buf);
2107 return r;
2108}
2109
2110int for_each_loose_object(each_loose_object_fn cb, void *data, unsigned flags)
2111{
2112 struct loose_alt_odb_data alt;
2113 int r;
2114
2115 r = for_each_loose_file_in_objdir(get_object_directory(),
2116 cb, NULL, NULL, data);
2117 if (r)
2118 return r;
2119
2120 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2121 return 0;
2122
2123 alt.cb = cb;
2124 alt.data = data;
2125 return foreach_alt_odb(loose_from_alt_odb, &alt);
2126}
2127
2128static int check_stream_sha1(git_zstream *stream,
2129 const char *hdr,
2130 unsigned long size,
2131 const char *path,
2132 const unsigned char *expected_sha1)
2133{
2134 git_SHA_CTX c;
2135 unsigned char real_sha1[GIT_MAX_RAWSZ];
2136 unsigned char buf[4096];
2137 unsigned long total_read;
2138 int status = Z_OK;
2139
2140 git_SHA1_Init(&c);
2141 git_SHA1_Update(&c, hdr, stream->total_out);
2142
2143 /*
2144 * We already read some bytes into hdr, but the ones up to the NUL
2145 * do not count against the object's content size.
2146 */
2147 total_read = stream->total_out - strlen(hdr) - 1;
2148
2149 /*
2150 * This size comparison must be "<=" to read the final zlib packets;
2151 * see the comment in unpack_sha1_rest for details.
2152 */
2153 while (total_read <= size &&
2154 (status == Z_OK || status == Z_BUF_ERROR)) {
2155 stream->next_out = buf;
2156 stream->avail_out = sizeof(buf);
2157 if (size - total_read < stream->avail_out)
2158 stream->avail_out = size - total_read;
2159 status = git_inflate(stream, Z_FINISH);
2160 git_SHA1_Update(&c, buf, stream->next_out - buf);
2161 total_read += stream->next_out - buf;
2162 }
2163 git_inflate_end(stream);
2164
2165 if (status != Z_STREAM_END) {
2166 error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
2167 return -1;
2168 }
2169 if (stream->avail_in) {
2170 error("garbage at end of loose object '%s'",
2171 sha1_to_hex(expected_sha1));
2172 return -1;
2173 }
2174
2175 git_SHA1_Final(real_sha1, &c);
2176 if (hashcmp(expected_sha1, real_sha1)) {
2177 error("sha1 mismatch for %s (expected %s)", path,
2178 sha1_to_hex(expected_sha1));
2179 return -1;
2180 }
2181
2182 return 0;
2183}
2184
2185int read_loose_object(const char *path,
2186 const unsigned char *expected_sha1,
2187 enum object_type *type,
2188 unsigned long *size,
2189 void **contents)
2190{
2191 int ret = -1;
2192 void *map = NULL;
2193 unsigned long mapsize;
2194 git_zstream stream;
2195 char hdr[32];
2196
2197 *contents = NULL;
2198
2199 map = map_sha1_file_1(path, NULL, &mapsize);
2200 if (!map) {
2201 error_errno("unable to mmap %s", path);
2202 goto out;
2203 }
2204
2205 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2206 error("unable to unpack header of %s", path);
2207 goto out;
2208 }
2209
2210 *type = parse_sha1_header(hdr, size);
2211 if (*type < 0) {
2212 error("unable to parse header of %s", path);
2213 git_inflate_end(&stream);
2214 goto out;
2215 }
2216
2217 if (*type == OBJ_BLOB) {
2218 if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
2219 goto out;
2220 } else {
2221 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
2222 if (!*contents) {
2223 error("unable to unpack contents of %s", path);
2224 git_inflate_end(&stream);
2225 goto out;
2226 }
2227 if (check_sha1_signature(expected_sha1, *contents,
2228 *size, typename(*type))) {
2229 error("sha1 mismatch for %s (expected %s)", path,
2230 sha1_to_hex(expected_sha1));
2231 free(*contents);
2232 goto out;
2233 }
2234 }
2235
2236 ret = 0; /* everything checks out */
2237
2238out:
2239 if (map)
2240 munmap(map, mapsize);
2241 return ret;
2242}