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 "replace-object.h"
27#include "streaming.h"
28#include "dir.h"
29#include "list.h"
30#include "mergesort.h"
31#include "quote.h"
32#include "packfile.h"
33#include "fetch-object.h"
34#include "object-store.h"
35
36/* The maximum size for an object header. */
37#define MAX_HEADER_LEN 32
38
39
40#define EMPTY_TREE_SHA1_BIN_LITERAL \
41 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
42 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
43
44#define EMPTY_BLOB_SHA1_BIN_LITERAL \
45 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
46 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
47
48const unsigned char null_sha1[GIT_MAX_RAWSZ];
49const struct object_id null_oid;
50static const struct object_id empty_tree_oid = {
51 EMPTY_TREE_SHA1_BIN_LITERAL
52};
53static const struct object_id empty_blob_oid = {
54 EMPTY_BLOB_SHA1_BIN_LITERAL
55};
56
57static void git_hash_sha1_init(git_hash_ctx *ctx)
58{
59 git_SHA1_Init(&ctx->sha1);
60}
61
62static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
63{
64 git_SHA1_Update(&ctx->sha1, data, len);
65}
66
67static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
68{
69 git_SHA1_Final(hash, &ctx->sha1);
70}
71
72static void git_hash_unknown_init(git_hash_ctx *ctx)
73{
74 BUG("trying to init unknown hash");
75}
76
77static void git_hash_unknown_update(git_hash_ctx *ctx, const void *data, size_t len)
78{
79 BUG("trying to update unknown hash");
80}
81
82static void git_hash_unknown_final(unsigned char *hash, git_hash_ctx *ctx)
83{
84 BUG("trying to finalize unknown hash");
85}
86
87const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
88 {
89 NULL,
90 0x00000000,
91 0,
92 0,
93 0,
94 git_hash_unknown_init,
95 git_hash_unknown_update,
96 git_hash_unknown_final,
97 NULL,
98 NULL,
99 },
100 {
101 "sha1",
102 /* "sha1", big-endian */
103 0x73686131,
104 GIT_SHA1_RAWSZ,
105 GIT_SHA1_HEXSZ,
106 GIT_SHA1_BLKSZ,
107 git_hash_sha1_init,
108 git_hash_sha1_update,
109 git_hash_sha1_final,
110 &empty_tree_oid,
111 &empty_blob_oid,
112 },
113};
114
115const char *empty_tree_oid_hex(void)
116{
117 static char buf[GIT_MAX_HEXSZ + 1];
118 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
119}
120
121const char *empty_blob_oid_hex(void)
122{
123 static char buf[GIT_MAX_HEXSZ + 1];
124 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
125}
126
127int hash_algo_by_name(const char *name)
128{
129 int i;
130 if (!name)
131 return GIT_HASH_UNKNOWN;
132 for (i = 1; i < GIT_HASH_NALGOS; i++)
133 if (!strcmp(name, hash_algos[i].name))
134 return i;
135 return GIT_HASH_UNKNOWN;
136}
137
138int hash_algo_by_id(uint32_t format_id)
139{
140 int i;
141 for (i = 1; i < GIT_HASH_NALGOS; i++)
142 if (format_id == hash_algos[i].format_id)
143 return i;
144 return GIT_HASH_UNKNOWN;
145}
146
147
148/*
149 * This is meant to hold a *small* number of objects that you would
150 * want read_sha1_file() to be able to return, but yet you do not want
151 * to write them into the object store (e.g. a browse-only
152 * application).
153 */
154static struct cached_object {
155 struct object_id oid;
156 enum object_type type;
157 void *buf;
158 unsigned long size;
159} *cached_objects;
160static int cached_object_nr, cached_object_alloc;
161
162static struct cached_object empty_tree = {
163 { EMPTY_TREE_SHA1_BIN_LITERAL },
164 OBJ_TREE,
165 "",
166 0
167};
168
169static struct cached_object *find_cached_object(const struct object_id *oid)
170{
171 int i;
172 struct cached_object *co = cached_objects;
173
174 for (i = 0; i < cached_object_nr; i++, co++) {
175 if (oideq(&co->oid, oid))
176 return co;
177 }
178 if (oideq(oid, the_hash_algo->empty_tree))
179 return &empty_tree;
180 return NULL;
181}
182
183
184static int get_conv_flags(unsigned flags)
185{
186 if (flags & HASH_RENORMALIZE)
187 return CONV_EOL_RENORMALIZE;
188 else if (flags & HASH_WRITE_OBJECT)
189 return global_conv_flags_eol | CONV_WRITE_OBJECT;
190 else
191 return 0;
192}
193
194
195int mkdir_in_gitdir(const char *path)
196{
197 if (mkdir(path, 0777)) {
198 int saved_errno = errno;
199 struct stat st;
200 struct strbuf sb = STRBUF_INIT;
201
202 if (errno != EEXIST)
203 return -1;
204 /*
205 * Are we looking at a path in a symlinked worktree
206 * whose original repository does not yet have it?
207 * e.g. .git/rr-cache pointing at its original
208 * repository in which the user hasn't performed any
209 * conflict resolution yet?
210 */
211 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
212 strbuf_readlink(&sb, path, st.st_size) ||
213 !is_absolute_path(sb.buf) ||
214 mkdir(sb.buf, 0777)) {
215 strbuf_release(&sb);
216 errno = saved_errno;
217 return -1;
218 }
219 strbuf_release(&sb);
220 }
221 return adjust_shared_perm(path);
222}
223
224enum scld_error safe_create_leading_directories(char *path)
225{
226 char *next_component = path + offset_1st_component(path);
227 enum scld_error ret = SCLD_OK;
228
229 while (ret == SCLD_OK && next_component) {
230 struct stat st;
231 char *slash = next_component, slash_character;
232
233 while (*slash && !is_dir_sep(*slash))
234 slash++;
235
236 if (!*slash)
237 break;
238
239 next_component = slash + 1;
240 while (is_dir_sep(*next_component))
241 next_component++;
242 if (!*next_component)
243 break;
244
245 slash_character = *slash;
246 *slash = '\0';
247 if (!stat(path, &st)) {
248 /* path exists */
249 if (!S_ISDIR(st.st_mode)) {
250 errno = ENOTDIR;
251 ret = SCLD_EXISTS;
252 }
253 } else if (mkdir(path, 0777)) {
254 if (errno == EEXIST &&
255 !stat(path, &st) && S_ISDIR(st.st_mode))
256 ; /* somebody created it since we checked */
257 else if (errno == ENOENT)
258 /*
259 * Either mkdir() failed because
260 * somebody just pruned the containing
261 * directory, or stat() failed because
262 * the file that was in our way was
263 * just removed. Either way, inform
264 * the caller that it might be worth
265 * trying again:
266 */
267 ret = SCLD_VANISHED;
268 else
269 ret = SCLD_FAILED;
270 } else if (adjust_shared_perm(path)) {
271 ret = SCLD_PERMS;
272 }
273 *slash = slash_character;
274 }
275 return ret;
276}
277
278enum scld_error safe_create_leading_directories_const(const char *path)
279{
280 int save_errno;
281 /* path points to cache entries, so xstrdup before messing with it */
282 char *buf = xstrdup(path);
283 enum scld_error result = safe_create_leading_directories(buf);
284
285 save_errno = errno;
286 free(buf);
287 errno = save_errno;
288 return result;
289}
290
291int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
292{
293 /*
294 * The number of times we will try to remove empty directories
295 * in the way of path. This is only 1 because if another
296 * process is racily creating directories that conflict with
297 * us, we don't want to fight against them.
298 */
299 int remove_directories_remaining = 1;
300
301 /*
302 * The number of times that we will try to create the
303 * directories containing path. We are willing to attempt this
304 * more than once, because another process could be trying to
305 * clean up empty directories at the same time as we are
306 * trying to create them.
307 */
308 int create_directories_remaining = 3;
309
310 /* A scratch copy of path, filled lazily if we need it: */
311 struct strbuf path_copy = STRBUF_INIT;
312
313 int ret, save_errno;
314
315 /* Sanity check: */
316 assert(*path);
317
318retry_fn:
319 ret = fn(path, cb);
320 save_errno = errno;
321 if (!ret)
322 goto out;
323
324 if (errno == EISDIR && remove_directories_remaining-- > 0) {
325 /*
326 * A directory is in the way. Maybe it is empty; try
327 * to remove it:
328 */
329 if (!path_copy.len)
330 strbuf_addstr(&path_copy, path);
331
332 if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
333 goto retry_fn;
334 } else if (errno == ENOENT && create_directories_remaining-- > 0) {
335 /*
336 * Maybe the containing directory didn't exist, or
337 * maybe it was just deleted by a process that is
338 * racing with us to clean up empty directories. Try
339 * to create it:
340 */
341 enum scld_error scld_result;
342
343 if (!path_copy.len)
344 strbuf_addstr(&path_copy, path);
345
346 do {
347 scld_result = safe_create_leading_directories(path_copy.buf);
348 if (scld_result == SCLD_OK)
349 goto retry_fn;
350 } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
351 }
352
353out:
354 strbuf_release(&path_copy);
355 errno = save_errno;
356 return ret;
357}
358
359static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
360{
361 int i;
362 for (i = 0; i < the_hash_algo->rawsz; i++) {
363 static char hex[] = "0123456789abcdef";
364 unsigned int val = sha1[i];
365 strbuf_addch(buf, hex[val >> 4]);
366 strbuf_addch(buf, hex[val & 0xf]);
367 if (!i)
368 strbuf_addch(buf, '/');
369 }
370}
371
372void sha1_file_name(struct repository *r, struct strbuf *buf, const unsigned char *sha1)
373{
374 strbuf_addstr(buf, r->objects->objectdir);
375 strbuf_addch(buf, '/');
376 fill_sha1_path(buf, sha1);
377}
378
379struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
380{
381 strbuf_setlen(&alt->scratch, alt->base_len);
382 return &alt->scratch;
383}
384
385static const char *alt_sha1_path(struct alternate_object_database *alt,
386 const unsigned char *sha1)
387{
388 struct strbuf *buf = alt_scratch_buf(alt);
389 fill_sha1_path(buf, sha1);
390 return buf->buf;
391}
392
393/*
394 * Return non-zero iff the path is usable as an alternate object database.
395 */
396static int alt_odb_usable(struct raw_object_store *o,
397 struct strbuf *path,
398 const char *normalized_objdir)
399{
400 struct alternate_object_database *alt;
401
402 /* Detect cases where alternate disappeared */
403 if (!is_directory(path->buf)) {
404 error(_("object directory %s does not exist; "
405 "check .git/objects/info/alternates"),
406 path->buf);
407 return 0;
408 }
409
410 /*
411 * Prevent the common mistake of listing the same
412 * thing twice, or object directory itself.
413 */
414 for (alt = o->alt_odb_list; alt; alt = alt->next) {
415 if (!fspathcmp(path->buf, alt->path))
416 return 0;
417 }
418 if (!fspathcmp(path->buf, normalized_objdir))
419 return 0;
420
421 return 1;
422}
423
424/*
425 * Prepare alternate object database registry.
426 *
427 * The variable alt_odb_list points at the list of struct
428 * alternate_object_database. The elements on this list come from
429 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
430 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
431 * whose contents is similar to that environment variable but can be
432 * LF separated. Its base points at a statically allocated buffer that
433 * contains "/the/directory/corresponding/to/.git/objects/...", while
434 * its name points just after the slash at the end of ".git/objects/"
435 * in the example above, and has enough space to hold 40-byte hex
436 * SHA1, an extra slash for the first level indirection, and the
437 * terminating NUL.
438 */
439static void read_info_alternates(struct repository *r,
440 const char *relative_base,
441 int depth);
442static int link_alt_odb_entry(struct repository *r, const char *entry,
443 const char *relative_base, int depth, const char *normalized_objdir)
444{
445 struct alternate_object_database *ent;
446 struct strbuf pathbuf = STRBUF_INIT;
447
448 if (!is_absolute_path(entry) && relative_base) {
449 strbuf_realpath(&pathbuf, relative_base, 1);
450 strbuf_addch(&pathbuf, '/');
451 }
452 strbuf_addstr(&pathbuf, entry);
453
454 if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
455 error(_("unable to normalize alternate object path: %s"),
456 pathbuf.buf);
457 strbuf_release(&pathbuf);
458 return -1;
459 }
460
461 /*
462 * The trailing slash after the directory name is given by
463 * this function at the end. Remove duplicates.
464 */
465 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
466 strbuf_setlen(&pathbuf, pathbuf.len - 1);
467
468 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir)) {
469 strbuf_release(&pathbuf);
470 return -1;
471 }
472
473 ent = alloc_alt_odb(pathbuf.buf);
474
475 /* add the alternate entry */
476 *r->objects->alt_odb_tail = ent;
477 r->objects->alt_odb_tail = &(ent->next);
478 ent->next = NULL;
479
480 /* recursively add alternates */
481 read_info_alternates(r, pathbuf.buf, depth + 1);
482
483 strbuf_release(&pathbuf);
484 return 0;
485}
486
487static const char *parse_alt_odb_entry(const char *string,
488 int sep,
489 struct strbuf *out)
490{
491 const char *end;
492
493 strbuf_reset(out);
494
495 if (*string == '#') {
496 /* comment; consume up to next separator */
497 end = strchrnul(string, sep);
498 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
499 /*
500 * quoted path; unquote_c_style has copied the
501 * data for us and set "end". Broken quoting (e.g.,
502 * an entry that doesn't end with a quote) falls
503 * back to the unquoted case below.
504 */
505 } else {
506 /* normal, unquoted path */
507 end = strchrnul(string, sep);
508 strbuf_add(out, string, end - string);
509 }
510
511 if (*end)
512 end++;
513 return end;
514}
515
516static void link_alt_odb_entries(struct repository *r, const char *alt,
517 int sep, const char *relative_base, int depth)
518{
519 struct strbuf objdirbuf = STRBUF_INIT;
520 struct strbuf entry = STRBUF_INIT;
521
522 if (!alt || !*alt)
523 return;
524
525 if (depth > 5) {
526 error(_("%s: ignoring alternate object stores, nesting too deep"),
527 relative_base);
528 return;
529 }
530
531 strbuf_add_absolute_path(&objdirbuf, r->objects->objectdir);
532 if (strbuf_normalize_path(&objdirbuf) < 0)
533 die(_("unable to normalize object directory: %s"),
534 objdirbuf.buf);
535
536 while (*alt) {
537 alt = parse_alt_odb_entry(alt, sep, &entry);
538 if (!entry.len)
539 continue;
540 link_alt_odb_entry(r, entry.buf,
541 relative_base, depth, objdirbuf.buf);
542 }
543 strbuf_release(&entry);
544 strbuf_release(&objdirbuf);
545}
546
547static void read_info_alternates(struct repository *r,
548 const char *relative_base,
549 int depth)
550{
551 char *path;
552 struct strbuf buf = STRBUF_INIT;
553
554 path = xstrfmt("%s/info/alternates", relative_base);
555 if (strbuf_read_file(&buf, path, 1024) < 0) {
556 warn_on_fopen_errors(path);
557 free(path);
558 return;
559 }
560
561 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
562 strbuf_release(&buf);
563 free(path);
564}
565
566struct alternate_object_database *alloc_alt_odb(const char *dir)
567{
568 struct alternate_object_database *ent;
569
570 FLEX_ALLOC_STR(ent, path, dir);
571 strbuf_init(&ent->scratch, 0);
572 strbuf_addf(&ent->scratch, "%s/", dir);
573 ent->base_len = ent->scratch.len;
574
575 return ent;
576}
577
578void add_to_alternates_file(const char *reference)
579{
580 struct lock_file lock = LOCK_INIT;
581 char *alts = git_pathdup("objects/info/alternates");
582 FILE *in, *out;
583 int found = 0;
584
585 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
586 out = fdopen_lock_file(&lock, "w");
587 if (!out)
588 die_errno(_("unable to fdopen alternates lockfile"));
589
590 in = fopen(alts, "r");
591 if (in) {
592 struct strbuf line = STRBUF_INIT;
593
594 while (strbuf_getline(&line, in) != EOF) {
595 if (!strcmp(reference, line.buf)) {
596 found = 1;
597 break;
598 }
599 fprintf_or_die(out, "%s\n", line.buf);
600 }
601
602 strbuf_release(&line);
603 fclose(in);
604 }
605 else if (errno != ENOENT)
606 die_errno(_("unable to read alternates file"));
607
608 if (found) {
609 rollback_lock_file(&lock);
610 } else {
611 fprintf_or_die(out, "%s\n", reference);
612 if (commit_lock_file(&lock))
613 die_errno(_("unable to move new alternates file into place"));
614 if (the_repository->objects->alt_odb_tail)
615 link_alt_odb_entries(the_repository, reference,
616 '\n', NULL, 0);
617 }
618 free(alts);
619}
620
621void add_to_alternates_memory(const char *reference)
622{
623 /*
624 * Make sure alternates are initialized, or else our entry may be
625 * overwritten when they are.
626 */
627 prepare_alt_odb(the_repository);
628
629 link_alt_odb_entries(the_repository, reference,
630 '\n', NULL, 0);
631}
632
633/*
634 * Compute the exact path an alternate is at and returns it. In case of
635 * error NULL is returned and the human readable error is added to `err`
636 * `path` may be relative and should point to $GIT_DIR.
637 * `err` must not be null.
638 */
639char *compute_alternate_path(const char *path, struct strbuf *err)
640{
641 char *ref_git = NULL;
642 const char *repo, *ref_git_s;
643 int seen_error = 0;
644
645 ref_git_s = real_path_if_valid(path);
646 if (!ref_git_s) {
647 seen_error = 1;
648 strbuf_addf(err, _("path '%s' does not exist"), path);
649 goto out;
650 } else
651 /*
652 * Beware: read_gitfile(), real_path() and mkpath()
653 * return static buffer
654 */
655 ref_git = xstrdup(ref_git_s);
656
657 repo = read_gitfile(ref_git);
658 if (!repo)
659 repo = read_gitfile(mkpath("%s/.git", ref_git));
660 if (repo) {
661 free(ref_git);
662 ref_git = xstrdup(repo);
663 }
664
665 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
666 char *ref_git_git = mkpathdup("%s/.git", ref_git);
667 free(ref_git);
668 ref_git = ref_git_git;
669 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
670 struct strbuf sb = STRBUF_INIT;
671 seen_error = 1;
672 if (get_common_dir(&sb, ref_git)) {
673 strbuf_addf(err,
674 _("reference repository '%s' as a linked "
675 "checkout is not supported yet."),
676 path);
677 goto out;
678 }
679
680 strbuf_addf(err, _("reference repository '%s' is not a "
681 "local repository."), path);
682 goto out;
683 }
684
685 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
686 strbuf_addf(err, _("reference repository '%s' is shallow"),
687 path);
688 seen_error = 1;
689 goto out;
690 }
691
692 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
693 strbuf_addf(err,
694 _("reference repository '%s' is grafted"),
695 path);
696 seen_error = 1;
697 goto out;
698 }
699
700out:
701 if (seen_error) {
702 FREE_AND_NULL(ref_git);
703 }
704
705 return ref_git;
706}
707
708int foreach_alt_odb(alt_odb_fn fn, void *cb)
709{
710 struct alternate_object_database *ent;
711 int r = 0;
712
713 prepare_alt_odb(the_repository);
714 for (ent = the_repository->objects->alt_odb_list; ent; ent = ent->next) {
715 r = fn(ent, cb);
716 if (r)
717 break;
718 }
719 return r;
720}
721
722void prepare_alt_odb(struct repository *r)
723{
724 if (r->objects->alt_odb_tail)
725 return;
726
727 r->objects->alt_odb_tail = &r->objects->alt_odb_list;
728 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
729
730 read_info_alternates(r, r->objects->objectdir, 0);
731}
732
733/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
734static int freshen_file(const char *fn)
735{
736 struct utimbuf t;
737 t.actime = t.modtime = time(NULL);
738 return !utime(fn, &t);
739}
740
741/*
742 * All of the check_and_freshen functions return 1 if the file exists and was
743 * freshened (if freshening was requested), 0 otherwise. If they return
744 * 0, you should not assume that it is safe to skip a write of the object (it
745 * either does not exist on disk, or has a stale mtime and may be subject to
746 * pruning).
747 */
748int check_and_freshen_file(const char *fn, int freshen)
749{
750 if (access(fn, F_OK))
751 return 0;
752 if (freshen && !freshen_file(fn))
753 return 0;
754 return 1;
755}
756
757static int check_and_freshen_local(const struct object_id *oid, int freshen)
758{
759 static struct strbuf buf = STRBUF_INIT;
760
761 strbuf_reset(&buf);
762 sha1_file_name(the_repository, &buf, oid->hash);
763
764 return check_and_freshen_file(buf.buf, freshen);
765}
766
767static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
768{
769 struct alternate_object_database *alt;
770 prepare_alt_odb(the_repository);
771 for (alt = the_repository->objects->alt_odb_list; alt; alt = alt->next) {
772 const char *path = alt_sha1_path(alt, oid->hash);
773 if (check_and_freshen_file(path, freshen))
774 return 1;
775 }
776 return 0;
777}
778
779static int check_and_freshen(const struct object_id *oid, int freshen)
780{
781 return check_and_freshen_local(oid, freshen) ||
782 check_and_freshen_nonlocal(oid, freshen);
783}
784
785int has_loose_object_nonlocal(const struct object_id *oid)
786{
787 return check_and_freshen_nonlocal(oid, 0);
788}
789
790static int has_loose_object(const struct object_id *oid)
791{
792 return check_and_freshen(oid, 0);
793}
794
795static void mmap_limit_check(size_t length)
796{
797 static size_t limit = 0;
798 if (!limit) {
799 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
800 if (!limit)
801 limit = SIZE_MAX;
802 }
803 if (length > limit)
804 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
805 (uintmax_t)length, (uintmax_t)limit);
806}
807
808void *xmmap_gently(void *start, size_t length,
809 int prot, int flags, int fd, off_t offset)
810{
811 void *ret;
812
813 mmap_limit_check(length);
814 ret = mmap(start, length, prot, flags, fd, offset);
815 if (ret == MAP_FAILED) {
816 if (!length)
817 return NULL;
818 release_pack_memory(length);
819 ret = mmap(start, length, prot, flags, fd, offset);
820 }
821 return ret;
822}
823
824void *xmmap(void *start, size_t length,
825 int prot, int flags, int fd, off_t offset)
826{
827 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
828 if (ret == MAP_FAILED)
829 die_errno(_("mmap failed"));
830 return ret;
831}
832
833/*
834 * With an in-core object data in "map", rehash it to make sure the
835 * object name actually matches "sha1" to detect object corruption.
836 * With "map" == NULL, try reading the object named with "sha1" using
837 * the streaming interface and rehash it to do the same.
838 */
839int check_object_signature(const struct object_id *oid, void *map,
840 unsigned long size, const char *type)
841{
842 struct object_id real_oid;
843 enum object_type obj_type;
844 struct git_istream *st;
845 git_hash_ctx c;
846 char hdr[MAX_HEADER_LEN];
847 int hdrlen;
848
849 if (map) {
850 hash_object_file(map, size, type, &real_oid);
851 return !oideq(oid, &real_oid) ? -1 : 0;
852 }
853
854 st = open_istream(oid, &obj_type, &size, NULL);
855 if (!st)
856 return -1;
857
858 /* Generate the header */
859 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(obj_type), size) + 1;
860
861 /* Sha1.. */
862 the_hash_algo->init_fn(&c);
863 the_hash_algo->update_fn(&c, hdr, hdrlen);
864 for (;;) {
865 char buf[1024 * 16];
866 ssize_t readlen = read_istream(st, buf, sizeof(buf));
867
868 if (readlen < 0) {
869 close_istream(st);
870 return -1;
871 }
872 if (!readlen)
873 break;
874 the_hash_algo->update_fn(&c, buf, readlen);
875 }
876 the_hash_algo->final_fn(real_oid.hash, &c);
877 close_istream(st);
878 return !oideq(oid, &real_oid) ? -1 : 0;
879}
880
881int git_open_cloexec(const char *name, int flags)
882{
883 int fd;
884 static int o_cloexec = O_CLOEXEC;
885
886 fd = open(name, flags | o_cloexec);
887 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
888 /* Try again w/o O_CLOEXEC: the kernel might not support it */
889 o_cloexec &= ~O_CLOEXEC;
890 fd = open(name, flags | o_cloexec);
891 }
892
893#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
894 {
895 static int fd_cloexec = FD_CLOEXEC;
896
897 if (!o_cloexec && 0 <= fd && fd_cloexec) {
898 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
899 int flags = fcntl(fd, F_GETFD);
900 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
901 fd_cloexec = 0;
902 }
903 }
904#endif
905 return fd;
906}
907
908/*
909 * Find "sha1" as a loose object in the local repository or in an alternate.
910 * Returns 0 on success, negative on failure.
911 *
912 * The "path" out-parameter will give the path of the object we found (if any).
913 * Note that it may point to static storage and is only valid until another
914 * call to sha1_file_name(), etc.
915 */
916static int stat_sha1_file(struct repository *r, const unsigned char *sha1,
917 struct stat *st, const char **path)
918{
919 struct alternate_object_database *alt;
920 static struct strbuf buf = STRBUF_INIT;
921
922 strbuf_reset(&buf);
923 sha1_file_name(r, &buf, sha1);
924 *path = buf.buf;
925
926 if (!lstat(*path, st))
927 return 0;
928
929 prepare_alt_odb(r);
930 errno = ENOENT;
931 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
932 *path = alt_sha1_path(alt, sha1);
933 if (!lstat(*path, st))
934 return 0;
935 }
936
937 return -1;
938}
939
940/*
941 * Like stat_sha1_file(), but actually open the object and return the
942 * descriptor. See the caveats on the "path" parameter above.
943 */
944static int open_sha1_file(struct repository *r,
945 const unsigned char *sha1, const char **path)
946{
947 int fd;
948 struct alternate_object_database *alt;
949 int most_interesting_errno;
950 static struct strbuf buf = STRBUF_INIT;
951
952 strbuf_reset(&buf);
953 sha1_file_name(r, &buf, sha1);
954 *path = buf.buf;
955
956 fd = git_open(*path);
957 if (fd >= 0)
958 return fd;
959 most_interesting_errno = errno;
960
961 prepare_alt_odb(r);
962 for (alt = r->objects->alt_odb_list; alt; alt = alt->next) {
963 *path = alt_sha1_path(alt, sha1);
964 fd = git_open(*path);
965 if (fd >= 0)
966 return fd;
967 if (most_interesting_errno == ENOENT)
968 most_interesting_errno = errno;
969 }
970 errno = most_interesting_errno;
971 return -1;
972}
973
974/*
975 * Map the loose object at "path" if it is not NULL, or the path found by
976 * searching for a loose object named "sha1".
977 */
978static void *map_sha1_file_1(struct repository *r, const char *path,
979 const unsigned char *sha1, unsigned long *size)
980{
981 void *map;
982 int fd;
983
984 if (path)
985 fd = git_open(path);
986 else
987 fd = open_sha1_file(r, sha1, &path);
988 map = NULL;
989 if (fd >= 0) {
990 struct stat st;
991
992 if (!fstat(fd, &st)) {
993 *size = xsize_t(st.st_size);
994 if (!*size) {
995 /* mmap() is forbidden on empty files */
996 error(_("object file %s is empty"), path);
997 return NULL;
998 }
999 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1000 }
1001 close(fd);
1002 }
1003 return map;
1004}
1005
1006void *map_sha1_file(struct repository *r,
1007 const unsigned char *sha1, unsigned long *size)
1008{
1009 return map_sha1_file_1(r, NULL, sha1, size);
1010}
1011
1012static int unpack_sha1_short_header(git_zstream *stream,
1013 unsigned char *map, unsigned long mapsize,
1014 void *buffer, unsigned long bufsiz)
1015{
1016 /* Get the data stream */
1017 memset(stream, 0, sizeof(*stream));
1018 stream->next_in = map;
1019 stream->avail_in = mapsize;
1020 stream->next_out = buffer;
1021 stream->avail_out = bufsiz;
1022
1023 git_inflate_init(stream);
1024 return git_inflate(stream, 0);
1025}
1026
1027int unpack_sha1_header(git_zstream *stream,
1028 unsigned char *map, unsigned long mapsize,
1029 void *buffer, unsigned long bufsiz)
1030{
1031 int status = unpack_sha1_short_header(stream, map, mapsize,
1032 buffer, bufsiz);
1033
1034 if (status < Z_OK)
1035 return status;
1036
1037 /* Make sure we have the terminating NUL */
1038 if (!memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1039 return -1;
1040 return 0;
1041}
1042
1043static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1044 unsigned long mapsize, void *buffer,
1045 unsigned long bufsiz, struct strbuf *header)
1046{
1047 int status;
1048
1049 status = unpack_sha1_short_header(stream, map, mapsize, buffer, bufsiz);
1050 if (status < Z_OK)
1051 return -1;
1052
1053 /*
1054 * Check if entire header is unpacked in the first iteration.
1055 */
1056 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1057 return 0;
1058
1059 /*
1060 * buffer[0..bufsiz] was not large enough. Copy the partial
1061 * result out to header, and then append the result of further
1062 * reading the stream.
1063 */
1064 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1065 stream->next_out = buffer;
1066 stream->avail_out = bufsiz;
1067
1068 do {
1069 status = git_inflate(stream, 0);
1070 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1071 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1072 return 0;
1073 stream->next_out = buffer;
1074 stream->avail_out = bufsiz;
1075 } while (status != Z_STREAM_END);
1076 return -1;
1077}
1078
1079static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1080{
1081 int bytes = strlen(buffer) + 1;
1082 unsigned char *buf = xmallocz(size);
1083 unsigned long n;
1084 int status = Z_OK;
1085
1086 n = stream->total_out - bytes;
1087 if (n > size)
1088 n = size;
1089 memcpy(buf, (char *) buffer + bytes, n);
1090 bytes = n;
1091 if (bytes <= size) {
1092 /*
1093 * The above condition must be (bytes <= size), not
1094 * (bytes < size). In other words, even though we
1095 * expect no more output and set avail_out to zero,
1096 * the input zlib stream may have bytes that express
1097 * "this concludes the stream", and we *do* want to
1098 * eat that input.
1099 *
1100 * Otherwise we would not be able to test that we
1101 * consumed all the input to reach the expected size;
1102 * we also want to check that zlib tells us that all
1103 * went well with status == Z_STREAM_END at the end.
1104 */
1105 stream->next_out = buf + bytes;
1106 stream->avail_out = size - bytes;
1107 while (status == Z_OK)
1108 status = git_inflate(stream, Z_FINISH);
1109 }
1110 if (status == Z_STREAM_END && !stream->avail_in) {
1111 git_inflate_end(stream);
1112 return buf;
1113 }
1114
1115 if (status < 0)
1116 error(_("corrupt loose object '%s'"), sha1_to_hex(sha1));
1117 else if (stream->avail_in)
1118 error(_("garbage at end of loose object '%s'"),
1119 sha1_to_hex(sha1));
1120 free(buf);
1121 return NULL;
1122}
1123
1124/*
1125 * We used to just use "sscanf()", but that's actually way
1126 * too permissive for what we want to check. So do an anal
1127 * object header parse by hand.
1128 */
1129static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1130 unsigned int flags)
1131{
1132 const char *type_buf = hdr;
1133 unsigned long size;
1134 int type, type_len = 0;
1135
1136 /*
1137 * The type can be of any size but is followed by
1138 * a space.
1139 */
1140 for (;;) {
1141 char c = *hdr++;
1142 if (!c)
1143 return -1;
1144 if (c == ' ')
1145 break;
1146 type_len++;
1147 }
1148
1149 type = type_from_string_gently(type_buf, type_len, 1);
1150 if (oi->type_name)
1151 strbuf_add(oi->type_name, type_buf, type_len);
1152 /*
1153 * Set type to 0 if its an unknown object and
1154 * we're obtaining the type using '--allow-unknown-type'
1155 * option.
1156 */
1157 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE) && (type < 0))
1158 type = 0;
1159 else if (type < 0)
1160 die(_("invalid object type"));
1161 if (oi->typep)
1162 *oi->typep = type;
1163
1164 /*
1165 * The length must follow immediately, and be in canonical
1166 * decimal format (ie "010" is not valid).
1167 */
1168 size = *hdr++ - '0';
1169 if (size > 9)
1170 return -1;
1171 if (size) {
1172 for (;;) {
1173 unsigned long c = *hdr - '0';
1174 if (c > 9)
1175 break;
1176 hdr++;
1177 size = size * 10 + c;
1178 }
1179 }
1180
1181 if (oi->sizep)
1182 *oi->sizep = size;
1183
1184 /*
1185 * The length must be followed by a zero byte
1186 */
1187 return *hdr ? -1 : type;
1188}
1189
1190int parse_sha1_header(const char *hdr, unsigned long *sizep)
1191{
1192 struct object_info oi = OBJECT_INFO_INIT;
1193
1194 oi.sizep = sizep;
1195 return parse_sha1_header_extended(hdr, &oi, 0);
1196}
1197
1198static int sha1_loose_object_info(struct repository *r,
1199 const unsigned char *sha1,
1200 struct object_info *oi, int flags)
1201{
1202 int status = 0;
1203 unsigned long mapsize;
1204 void *map;
1205 git_zstream stream;
1206 char hdr[MAX_HEADER_LEN];
1207 struct strbuf hdrbuf = STRBUF_INIT;
1208 unsigned long size_scratch;
1209
1210 if (oi->delta_base_sha1)
1211 hashclr(oi->delta_base_sha1);
1212
1213 /*
1214 * If we don't care about type or size, then we don't
1215 * need to look inside the object at all. Note that we
1216 * do not optimize out the stat call, even if the
1217 * caller doesn't care about the disk-size, since our
1218 * return value implicitly indicates whether the
1219 * object even exists.
1220 */
1221 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1222 const char *path;
1223 struct stat st;
1224 if (stat_sha1_file(r, sha1, &st, &path) < 0)
1225 return -1;
1226 if (oi->disk_sizep)
1227 *oi->disk_sizep = st.st_size;
1228 return 0;
1229 }
1230
1231 map = map_sha1_file(r, sha1, &mapsize);
1232 if (!map)
1233 return -1;
1234
1235 if (!oi->sizep)
1236 oi->sizep = &size_scratch;
1237
1238 if (oi->disk_sizep)
1239 *oi->disk_sizep = mapsize;
1240 if ((flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE)) {
1241 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
1242 status = error(_("unable to unpack %s header with --allow-unknown-type"),
1243 sha1_to_hex(sha1));
1244 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1245 status = error(_("unable to unpack %s header"),
1246 sha1_to_hex(sha1));
1247 if (status < 0)
1248 ; /* Do nothing */
1249 else if (hdrbuf.len) {
1250 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
1251 status = error(_("unable to parse %s header with --allow-unknown-type"),
1252 sha1_to_hex(sha1));
1253 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
1254 status = error(_("unable to parse %s header"), sha1_to_hex(sha1));
1255
1256 if (status >= 0 && oi->contentp) {
1257 *oi->contentp = unpack_sha1_rest(&stream, hdr,
1258 *oi->sizep, sha1);
1259 if (!*oi->contentp) {
1260 git_inflate_end(&stream);
1261 status = -1;
1262 }
1263 } else
1264 git_inflate_end(&stream);
1265
1266 munmap(map, mapsize);
1267 if (status && oi->typep)
1268 *oi->typep = status;
1269 if (oi->sizep == &size_scratch)
1270 oi->sizep = NULL;
1271 strbuf_release(&hdrbuf);
1272 oi->whence = OI_LOOSE;
1273 return (status < 0) ? status : 0;
1274}
1275
1276int fetch_if_missing = 1;
1277
1278int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1279 struct object_info *oi, unsigned flags)
1280{
1281 static struct object_info blank_oi = OBJECT_INFO_INIT;
1282 struct pack_entry e;
1283 int rtype;
1284 const struct object_id *real = oid;
1285 int already_retried = 0;
1286
1287 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1288 real = lookup_replace_object(r, oid);
1289
1290 if (is_null_oid(real))
1291 return -1;
1292
1293 if (!oi)
1294 oi = &blank_oi;
1295
1296 if (!(flags & OBJECT_INFO_SKIP_CACHED)) {
1297 struct cached_object *co = find_cached_object(real);
1298 if (co) {
1299 if (oi->typep)
1300 *(oi->typep) = co->type;
1301 if (oi->sizep)
1302 *(oi->sizep) = co->size;
1303 if (oi->disk_sizep)
1304 *(oi->disk_sizep) = 0;
1305 if (oi->delta_base_sha1)
1306 hashclr(oi->delta_base_sha1);
1307 if (oi->type_name)
1308 strbuf_addstr(oi->type_name, type_name(co->type));
1309 if (oi->contentp)
1310 *oi->contentp = xmemdupz(co->buf, co->size);
1311 oi->whence = OI_CACHED;
1312 return 0;
1313 }
1314 }
1315
1316 while (1) {
1317 if (find_pack_entry(r, real, &e))
1318 break;
1319
1320 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1321 return -1;
1322
1323 /* Most likely it's a loose object. */
1324 if (!sha1_loose_object_info(r, real->hash, oi, flags))
1325 return 0;
1326
1327 /* Not a loose object; someone else may have just packed it. */
1328 if (!(flags & OBJECT_INFO_QUICK)) {
1329 reprepare_packed_git(r);
1330 if (find_pack_entry(r, real, &e))
1331 break;
1332 }
1333
1334 /* Check if it is a missing object */
1335 if (fetch_if_missing && repository_format_partial_clone &&
1336 !already_retried && r == the_repository) {
1337 /*
1338 * TODO Investigate having fetch_object() return
1339 * TODO error/success and stopping the music here.
1340 * TODO Pass a repository struct through fetch_object,
1341 * such that arbitrary repositories work.
1342 */
1343 fetch_objects(repository_format_partial_clone, real, 1);
1344 already_retried = 1;
1345 continue;
1346 }
1347
1348 return -1;
1349 }
1350
1351 if (oi == &blank_oi)
1352 /*
1353 * We know that the caller doesn't actually need the
1354 * information below, so return early.
1355 */
1356 return 0;
1357 rtype = packed_object_info(r, e.p, e.offset, oi);
1358 if (rtype < 0) {
1359 mark_bad_packed_object(e.p, real->hash);
1360 return oid_object_info_extended(r, real, oi, 0);
1361 } else if (oi->whence == OI_PACKED) {
1362 oi->u.packed.offset = e.offset;
1363 oi->u.packed.pack = e.p;
1364 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1365 rtype == OBJ_OFS_DELTA);
1366 }
1367
1368 return 0;
1369}
1370
1371/* returns enum object_type or negative */
1372int oid_object_info(struct repository *r,
1373 const struct object_id *oid,
1374 unsigned long *sizep)
1375{
1376 enum object_type type;
1377 struct object_info oi = OBJECT_INFO_INIT;
1378
1379 oi.typep = &type;
1380 oi.sizep = sizep;
1381 if (oid_object_info_extended(r, oid, &oi,
1382 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1383 return -1;
1384 return type;
1385}
1386
1387static void *read_object(const unsigned char *sha1, enum object_type *type,
1388 unsigned long *size)
1389{
1390 struct object_id oid;
1391 struct object_info oi = OBJECT_INFO_INIT;
1392 void *content;
1393 oi.typep = type;
1394 oi.sizep = size;
1395 oi.contentp = &content;
1396
1397 hashcpy(oid.hash, sha1);
1398
1399 if (oid_object_info_extended(the_repository, &oid, &oi, 0) < 0)
1400 return NULL;
1401 return content;
1402}
1403
1404int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1405 struct object_id *oid)
1406{
1407 struct cached_object *co;
1408
1409 hash_object_file(buf, len, type_name(type), oid);
1410 if (has_sha1_file(oid->hash) || find_cached_object(oid))
1411 return 0;
1412 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1413 co = &cached_objects[cached_object_nr++];
1414 co->size = len;
1415 co->type = type;
1416 co->buf = xmalloc(len);
1417 memcpy(co->buf, buf, len);
1418 oidcpy(&co->oid, oid);
1419 return 0;
1420}
1421
1422/*
1423 * This function dies on corrupt objects; the callers who want to
1424 * deal with them should arrange to call read_object() and give error
1425 * messages themselves.
1426 */
1427void *read_object_file_extended(const struct object_id *oid,
1428 enum object_type *type,
1429 unsigned long *size,
1430 int lookup_replace)
1431{
1432 void *data;
1433 const struct packed_git *p;
1434 const char *path;
1435 struct stat st;
1436 const struct object_id *repl = lookup_replace ?
1437 lookup_replace_object(the_repository, oid) : oid;
1438
1439 errno = 0;
1440 data = read_object(repl->hash, type, size);
1441 if (data)
1442 return data;
1443
1444 if (errno && errno != ENOENT)
1445 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1446
1447 /* die if we replaced an object with one that does not exist */
1448 if (repl != oid)
1449 die(_("replacement %s not found for %s"),
1450 oid_to_hex(repl), oid_to_hex(oid));
1451
1452 if (!stat_sha1_file(the_repository, repl->hash, &st, &path))
1453 die(_("loose object %s (stored in %s) is corrupt"),
1454 oid_to_hex(repl), path);
1455
1456 if ((p = has_packed_and_bad(repl->hash)) != NULL)
1457 die(_("packed object %s (stored in %s) is corrupt"),
1458 oid_to_hex(repl), p->pack_name);
1459
1460 return NULL;
1461}
1462
1463void *read_object_with_reference(const struct object_id *oid,
1464 const char *required_type_name,
1465 unsigned long *size,
1466 struct object_id *actual_oid_return)
1467{
1468 enum object_type type, required_type;
1469 void *buffer;
1470 unsigned long isize;
1471 struct object_id actual_oid;
1472
1473 required_type = type_from_string(required_type_name);
1474 oidcpy(&actual_oid, oid);
1475 while (1) {
1476 int ref_length = -1;
1477 const char *ref_type = NULL;
1478
1479 buffer = read_object_file(&actual_oid, &type, &isize);
1480 if (!buffer)
1481 return NULL;
1482 if (type == required_type) {
1483 *size = isize;
1484 if (actual_oid_return)
1485 oidcpy(actual_oid_return, &actual_oid);
1486 return buffer;
1487 }
1488 /* Handle references */
1489 else if (type == OBJ_COMMIT)
1490 ref_type = "tree ";
1491 else if (type == OBJ_TAG)
1492 ref_type = "object ";
1493 else {
1494 free(buffer);
1495 return NULL;
1496 }
1497 ref_length = strlen(ref_type);
1498
1499 if (ref_length + the_hash_algo->hexsz > isize ||
1500 memcmp(buffer, ref_type, ref_length) ||
1501 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1502 free(buffer);
1503 return NULL;
1504 }
1505 free(buffer);
1506 /* Now we have the ID of the referred-to object in
1507 * actual_oid. Check again. */
1508 }
1509}
1510
1511static void write_object_file_prepare(const void *buf, unsigned long len,
1512 const char *type, struct object_id *oid,
1513 char *hdr, int *hdrlen)
1514{
1515 git_hash_ctx c;
1516
1517 /* Generate the header */
1518 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
1519
1520 /* Sha1.. */
1521 the_hash_algo->init_fn(&c);
1522 the_hash_algo->update_fn(&c, hdr, *hdrlen);
1523 the_hash_algo->update_fn(&c, buf, len);
1524 the_hash_algo->final_fn(oid->hash, &c);
1525}
1526
1527/*
1528 * Move the just written object into its final resting place.
1529 */
1530int finalize_object_file(const char *tmpfile, const char *filename)
1531{
1532 int ret = 0;
1533
1534 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1535 goto try_rename;
1536 else if (link(tmpfile, filename))
1537 ret = errno;
1538
1539 /*
1540 * Coda hack - coda doesn't like cross-directory links,
1541 * so we fall back to a rename, which will mean that it
1542 * won't be able to check collisions, but that's not a
1543 * big deal.
1544 *
1545 * The same holds for FAT formatted media.
1546 *
1547 * When this succeeds, we just return. We have nothing
1548 * left to unlink.
1549 */
1550 if (ret && ret != EEXIST) {
1551 try_rename:
1552 if (!rename(tmpfile, filename))
1553 goto out;
1554 ret = errno;
1555 }
1556 unlink_or_warn(tmpfile);
1557 if (ret) {
1558 if (ret != EEXIST) {
1559 return error_errno(_("unable to write sha1 filename %s"), filename);
1560 }
1561 /* FIXME!!! Collision check here ? */
1562 }
1563
1564out:
1565 if (adjust_shared_perm(filename))
1566 return error(_("unable to set permission to '%s'"), filename);
1567 return 0;
1568}
1569
1570static int write_buffer(int fd, const void *buf, size_t len)
1571{
1572 if (write_in_full(fd, buf, len) < 0)
1573 return error_errno(_("file write error"));
1574 return 0;
1575}
1576
1577int hash_object_file(const void *buf, unsigned long len, const char *type,
1578 struct object_id *oid)
1579{
1580 char hdr[MAX_HEADER_LEN];
1581 int hdrlen = sizeof(hdr);
1582 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1583 return 0;
1584}
1585
1586/* Finalize a file on disk, and close it. */
1587static void close_sha1_file(int fd)
1588{
1589 if (fsync_object_files)
1590 fsync_or_die(fd, "sha1 file");
1591 if (close(fd) != 0)
1592 die_errno(_("error when closing sha1 file"));
1593}
1594
1595/* Size of directory component, including the ending '/' */
1596static inline int directory_size(const char *filename)
1597{
1598 const char *s = strrchr(filename, '/');
1599 if (!s)
1600 return 0;
1601 return s - filename + 1;
1602}
1603
1604/*
1605 * This creates a temporary file in the same directory as the final
1606 * 'filename'
1607 *
1608 * We want to avoid cross-directory filename renames, because those
1609 * can have problems on various filesystems (FAT, NFS, Coda).
1610 */
1611static int create_tmpfile(struct strbuf *tmp, const char *filename)
1612{
1613 int fd, dirlen = directory_size(filename);
1614
1615 strbuf_reset(tmp);
1616 strbuf_add(tmp, filename, dirlen);
1617 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1618 fd = git_mkstemp_mode(tmp->buf, 0444);
1619 if (fd < 0 && dirlen && errno == ENOENT) {
1620 /*
1621 * Make sure the directory exists; note that the contents
1622 * of the buffer are undefined after mkstemp returns an
1623 * error, so we have to rewrite the whole buffer from
1624 * scratch.
1625 */
1626 strbuf_reset(tmp);
1627 strbuf_add(tmp, filename, dirlen - 1);
1628 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1629 return -1;
1630 if (adjust_shared_perm(tmp->buf))
1631 return -1;
1632
1633 /* Try again */
1634 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1635 fd = git_mkstemp_mode(tmp->buf, 0444);
1636 }
1637 return fd;
1638}
1639
1640static int write_loose_object(const struct object_id *oid, char *hdr,
1641 int hdrlen, const void *buf, unsigned long len,
1642 time_t mtime)
1643{
1644 int fd, ret;
1645 unsigned char compressed[4096];
1646 git_zstream stream;
1647 git_hash_ctx c;
1648 struct object_id parano_oid;
1649 static struct strbuf tmp_file = STRBUF_INIT;
1650 static struct strbuf filename = STRBUF_INIT;
1651
1652 strbuf_reset(&filename);
1653 sha1_file_name(the_repository, &filename, oid->hash);
1654
1655 fd = create_tmpfile(&tmp_file, filename.buf);
1656 if (fd < 0) {
1657 if (errno == EACCES)
1658 return error(_("insufficient permission for adding an object to repository database %s"), get_object_directory());
1659 else
1660 return error_errno(_("unable to create temporary file"));
1661 }
1662
1663 /* Set it up */
1664 git_deflate_init(&stream, zlib_compression_level);
1665 stream.next_out = compressed;
1666 stream.avail_out = sizeof(compressed);
1667 the_hash_algo->init_fn(&c);
1668
1669 /* First header.. */
1670 stream.next_in = (unsigned char *)hdr;
1671 stream.avail_in = hdrlen;
1672 while (git_deflate(&stream, 0) == Z_OK)
1673 ; /* nothing */
1674 the_hash_algo->update_fn(&c, hdr, hdrlen);
1675
1676 /* Then the data itself.. */
1677 stream.next_in = (void *)buf;
1678 stream.avail_in = len;
1679 do {
1680 unsigned char *in0 = stream.next_in;
1681 ret = git_deflate(&stream, Z_FINISH);
1682 the_hash_algo->update_fn(&c, in0, stream.next_in - in0);
1683 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
1684 die(_("unable to write sha1 file"));
1685 stream.next_out = compressed;
1686 stream.avail_out = sizeof(compressed);
1687 } while (ret == Z_OK);
1688
1689 if (ret != Z_STREAM_END)
1690 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
1691 ret);
1692 ret = git_deflate_end_gently(&stream);
1693 if (ret != Z_OK)
1694 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
1695 ret);
1696 the_hash_algo->final_fn(parano_oid.hash, &c);
1697 if (!oideq(oid, ¶no_oid))
1698 die(_("confused by unstable object source data for %s"),
1699 oid_to_hex(oid));
1700
1701 close_sha1_file(fd);
1702
1703 if (mtime) {
1704 struct utimbuf utb;
1705 utb.actime = mtime;
1706 utb.modtime = mtime;
1707 if (utime(tmp_file.buf, &utb) < 0)
1708 warning_errno(_("failed utime() on %s"), tmp_file.buf);
1709 }
1710
1711 return finalize_object_file(tmp_file.buf, filename.buf);
1712}
1713
1714static int freshen_loose_object(const struct object_id *oid)
1715{
1716 return check_and_freshen(oid, 1);
1717}
1718
1719static int freshen_packed_object(const struct object_id *oid)
1720{
1721 struct pack_entry e;
1722 if (!find_pack_entry(the_repository, oid, &e))
1723 return 0;
1724 if (e.p->freshened)
1725 return 1;
1726 if (!freshen_file(e.p->pack_name))
1727 return 0;
1728 e.p->freshened = 1;
1729 return 1;
1730}
1731
1732int write_object_file(const void *buf, unsigned long len, const char *type,
1733 struct object_id *oid)
1734{
1735 char hdr[MAX_HEADER_LEN];
1736 int hdrlen = sizeof(hdr);
1737
1738 /* Normally if we have it in the pack then we do not bother writing
1739 * it out into .git/objects/??/?{38} file.
1740 */
1741 write_object_file_prepare(buf, len, type, oid, hdr, &hdrlen);
1742 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1743 return 0;
1744 return write_loose_object(oid, hdr, hdrlen, buf, len, 0);
1745}
1746
1747int hash_object_file_literally(const void *buf, unsigned long len,
1748 const char *type, struct object_id *oid,
1749 unsigned flags)
1750{
1751 char *header;
1752 int hdrlen, status = 0;
1753
1754 /* type string, SP, %lu of the length plus NUL must fit this */
1755 hdrlen = strlen(type) + MAX_HEADER_LEN;
1756 header = xmalloc(hdrlen);
1757 write_object_file_prepare(buf, len, type, oid, header, &hdrlen);
1758
1759 if (!(flags & HASH_WRITE_OBJECT))
1760 goto cleanup;
1761 if (freshen_packed_object(oid) || freshen_loose_object(oid))
1762 goto cleanup;
1763 status = write_loose_object(oid, header, hdrlen, buf, len, 0);
1764
1765cleanup:
1766 free(header);
1767 return status;
1768}
1769
1770int force_object_loose(const struct object_id *oid, time_t mtime)
1771{
1772 void *buf;
1773 unsigned long len;
1774 enum object_type type;
1775 char hdr[MAX_HEADER_LEN];
1776 int hdrlen;
1777 int ret;
1778
1779 if (has_loose_object(oid))
1780 return 0;
1781 buf = read_object(oid->hash, &type, &len);
1782 if (!buf)
1783 return error(_("cannot read sha1_file for %s"), oid_to_hex(oid));
1784 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", type_name(type), len) + 1;
1785 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime);
1786 free(buf);
1787
1788 return ret;
1789}
1790
1791int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
1792{
1793 struct object_id oid;
1794 if (!startup_info->have_repository)
1795 return 0;
1796 hashcpy(oid.hash, sha1);
1797 return oid_object_info_extended(the_repository, &oid, NULL,
1798 flags | OBJECT_INFO_SKIP_CACHED) >= 0;
1799}
1800
1801int has_object_file(const struct object_id *oid)
1802{
1803 return has_sha1_file(oid->hash);
1804}
1805
1806int has_object_file_with_flags(const struct object_id *oid, int flags)
1807{
1808 return has_sha1_file_with_flags(oid->hash, flags);
1809}
1810
1811static void check_tree(const void *buf, size_t size)
1812{
1813 struct tree_desc desc;
1814 struct name_entry entry;
1815
1816 init_tree_desc(&desc, buf, size);
1817 while (tree_entry(&desc, &entry))
1818 /* do nothing
1819 * tree_entry() will die() on malformed entries */
1820 ;
1821}
1822
1823static void check_commit(const void *buf, size_t size)
1824{
1825 struct commit c;
1826 memset(&c, 0, sizeof(c));
1827 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
1828 die(_("corrupt commit"));
1829}
1830
1831static void check_tag(const void *buf, size_t size)
1832{
1833 struct tag t;
1834 memset(&t, 0, sizeof(t));
1835 if (parse_tag_buffer(the_repository, &t, buf, size))
1836 die(_("corrupt tag"));
1837}
1838
1839static int index_mem(struct object_id *oid, void *buf, size_t size,
1840 enum object_type type,
1841 const char *path, unsigned flags)
1842{
1843 int ret, re_allocated = 0;
1844 int write_object = flags & HASH_WRITE_OBJECT;
1845
1846 if (!type)
1847 type = OBJ_BLOB;
1848
1849 /*
1850 * Convert blobs to git internal format
1851 */
1852 if ((type == OBJ_BLOB) && path) {
1853 struct strbuf nbuf = STRBUF_INIT;
1854 if (convert_to_git(&the_index, path, buf, size, &nbuf,
1855 get_conv_flags(flags))) {
1856 buf = strbuf_detach(&nbuf, &size);
1857 re_allocated = 1;
1858 }
1859 }
1860 if (flags & HASH_FORMAT_CHECK) {
1861 if (type == OBJ_TREE)
1862 check_tree(buf, size);
1863 if (type == OBJ_COMMIT)
1864 check_commit(buf, size);
1865 if (type == OBJ_TAG)
1866 check_tag(buf, size);
1867 }
1868
1869 if (write_object)
1870 ret = write_object_file(buf, size, type_name(type), oid);
1871 else
1872 ret = hash_object_file(buf, size, type_name(type), oid);
1873 if (re_allocated)
1874 free(buf);
1875 return ret;
1876}
1877
1878static int index_stream_convert_blob(struct object_id *oid, int fd,
1879 const char *path, unsigned flags)
1880{
1881 int ret;
1882 const int write_object = flags & HASH_WRITE_OBJECT;
1883 struct strbuf sbuf = STRBUF_INIT;
1884
1885 assert(path);
1886 assert(would_convert_to_git_filter_fd(&the_index, path));
1887
1888 convert_to_git_filter_fd(&the_index, path, fd, &sbuf,
1889 get_conv_flags(flags));
1890
1891 if (write_object)
1892 ret = write_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1893 oid);
1894 else
1895 ret = hash_object_file(sbuf.buf, sbuf.len, type_name(OBJ_BLOB),
1896 oid);
1897 strbuf_release(&sbuf);
1898 return ret;
1899}
1900
1901static int index_pipe(struct object_id *oid, int fd, enum object_type type,
1902 const char *path, unsigned flags)
1903{
1904 struct strbuf sbuf = STRBUF_INIT;
1905 int ret;
1906
1907 if (strbuf_read(&sbuf, fd, 4096) >= 0)
1908 ret = index_mem(oid, sbuf.buf, sbuf.len, type, path, flags);
1909 else
1910 ret = -1;
1911 strbuf_release(&sbuf);
1912 return ret;
1913}
1914
1915#define SMALL_FILE_SIZE (32*1024)
1916
1917static int index_core(struct object_id *oid, int fd, size_t size,
1918 enum object_type type, const char *path,
1919 unsigned flags)
1920{
1921 int ret;
1922
1923 if (!size) {
1924 ret = index_mem(oid, "", size, type, path, flags);
1925 } else if (size <= SMALL_FILE_SIZE) {
1926 char *buf = xmalloc(size);
1927 ssize_t read_result = read_in_full(fd, buf, size);
1928 if (read_result < 0)
1929 ret = error_errno(_("read error while indexing %s"),
1930 path ? path : "<unknown>");
1931 else if (read_result != size)
1932 ret = error(_("short read while indexing %s"),
1933 path ? path : "<unknown>");
1934 else
1935 ret = index_mem(oid, buf, size, type, path, flags);
1936 free(buf);
1937 } else {
1938 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1939 ret = index_mem(oid, buf, size, type, path, flags);
1940 munmap(buf, size);
1941 }
1942 return ret;
1943}
1944
1945/*
1946 * This creates one packfile per large blob unless bulk-checkin
1947 * machinery is "plugged".
1948 *
1949 * This also bypasses the usual "convert-to-git" dance, and that is on
1950 * purpose. We could write a streaming version of the converting
1951 * functions and insert that before feeding the data to fast-import
1952 * (or equivalent in-core API described above). However, that is
1953 * somewhat complicated, as we do not know the size of the filter
1954 * result, which we need to know beforehand when writing a git object.
1955 * Since the primary motivation for trying to stream from the working
1956 * tree file and to avoid mmaping it in core is to deal with large
1957 * binary blobs, they generally do not want to get any conversion, and
1958 * callers should avoid this code path when filters are requested.
1959 */
1960static int index_stream(struct object_id *oid, int fd, size_t size,
1961 enum object_type type, const char *path,
1962 unsigned flags)
1963{
1964 return index_bulk_checkin(oid, fd, size, type, path, flags);
1965}
1966
1967int index_fd(struct object_id *oid, int fd, struct stat *st,
1968 enum object_type type, const char *path, unsigned flags)
1969{
1970 int ret;
1971
1972 /*
1973 * Call xsize_t() only when needed to avoid potentially unnecessary
1974 * die() for large files.
1975 */
1976 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(&the_index, path))
1977 ret = index_stream_convert_blob(oid, fd, path, flags);
1978 else if (!S_ISREG(st->st_mode))
1979 ret = index_pipe(oid, fd, type, path, flags);
1980 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
1981 (path && would_convert_to_git(&the_index, path)))
1982 ret = index_core(oid, fd, xsize_t(st->st_size), type, path,
1983 flags);
1984 else
1985 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
1986 flags);
1987 close(fd);
1988 return ret;
1989}
1990
1991int index_path(struct object_id *oid, const char *path, struct stat *st, unsigned flags)
1992{
1993 int fd;
1994 struct strbuf sb = STRBUF_INIT;
1995 int rc = 0;
1996
1997 switch (st->st_mode & S_IFMT) {
1998 case S_IFREG:
1999 fd = open(path, O_RDONLY);
2000 if (fd < 0)
2001 return error_errno("open(\"%s\")", path);
2002 if (index_fd(oid, fd, st, OBJ_BLOB, path, flags) < 0)
2003 return error(_("%s: failed to insert into database"),
2004 path);
2005 break;
2006 case S_IFLNK:
2007 if (strbuf_readlink(&sb, path, st->st_size))
2008 return error_errno("readlink(\"%s\")", path);
2009 if (!(flags & HASH_WRITE_OBJECT))
2010 hash_object_file(sb.buf, sb.len, blob_type, oid);
2011 else if (write_object_file(sb.buf, sb.len, blob_type, oid))
2012 rc = error(_("%s: failed to insert into database"), path);
2013 strbuf_release(&sb);
2014 break;
2015 case S_IFDIR:
2016 return resolve_gitlink_ref(path, "HEAD", oid);
2017 default:
2018 return error(_("%s: unsupported file type"), path);
2019 }
2020 return rc;
2021}
2022
2023int read_pack_header(int fd, struct pack_header *header)
2024{
2025 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2026 /* "eof before pack header was fully read" */
2027 return PH_ERROR_EOF;
2028
2029 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2030 /* "protocol error (pack signature mismatch detected)" */
2031 return PH_ERROR_PACK_SIGNATURE;
2032 if (!pack_version_ok(header->hdr_version))
2033 /* "protocol error (pack version unsupported)" */
2034 return PH_ERROR_PROTOCOL;
2035 return 0;
2036}
2037
2038void assert_oid_type(const struct object_id *oid, enum object_type expect)
2039{
2040 enum object_type type = oid_object_info(the_repository, oid, NULL);
2041 if (type < 0)
2042 die(_("%s is not a valid object"), oid_to_hex(oid));
2043 if (type != expect)
2044 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2045 type_name(expect));
2046}
2047
2048int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2049 struct strbuf *path,
2050 each_loose_object_fn obj_cb,
2051 each_loose_cruft_fn cruft_cb,
2052 each_loose_subdir_fn subdir_cb,
2053 void *data)
2054{
2055 size_t origlen, baselen;
2056 DIR *dir;
2057 struct dirent *de;
2058 int r = 0;
2059 struct object_id oid;
2060
2061 if (subdir_nr > 0xff)
2062 BUG("invalid loose object subdirectory: %x", subdir_nr);
2063
2064 origlen = path->len;
2065 strbuf_complete(path, '/');
2066 strbuf_addf(path, "%02x", subdir_nr);
2067
2068 dir = opendir(path->buf);
2069 if (!dir) {
2070 if (errno != ENOENT)
2071 r = error_errno(_("unable to open %s"), path->buf);
2072 strbuf_setlen(path, origlen);
2073 return r;
2074 }
2075
2076 oid.hash[0] = subdir_nr;
2077 strbuf_addch(path, '/');
2078 baselen = path->len;
2079
2080 while ((de = readdir(dir))) {
2081 size_t namelen;
2082 if (is_dot_or_dotdot(de->d_name))
2083 continue;
2084
2085 namelen = strlen(de->d_name);
2086 strbuf_setlen(path, baselen);
2087 strbuf_add(path, de->d_name, namelen);
2088 if (namelen == the_hash_algo->hexsz - 2 &&
2089 !hex_to_bytes(oid.hash + 1, de->d_name,
2090 the_hash_algo->rawsz - 1)) {
2091 if (obj_cb) {
2092 r = obj_cb(&oid, path->buf, data);
2093 if (r)
2094 break;
2095 }
2096 continue;
2097 }
2098
2099 if (cruft_cb) {
2100 r = cruft_cb(de->d_name, path->buf, data);
2101 if (r)
2102 break;
2103 }
2104 }
2105 closedir(dir);
2106
2107 strbuf_setlen(path, baselen - 1);
2108 if (!r && subdir_cb)
2109 r = subdir_cb(subdir_nr, path->buf, data);
2110
2111 strbuf_setlen(path, origlen);
2112
2113 return r;
2114}
2115
2116int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2117 each_loose_object_fn obj_cb,
2118 each_loose_cruft_fn cruft_cb,
2119 each_loose_subdir_fn subdir_cb,
2120 void *data)
2121{
2122 int r = 0;
2123 int i;
2124
2125 for (i = 0; i < 256; i++) {
2126 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2127 subdir_cb, data);
2128 if (r)
2129 break;
2130 }
2131
2132 return r;
2133}
2134
2135int for_each_loose_file_in_objdir(const char *path,
2136 each_loose_object_fn obj_cb,
2137 each_loose_cruft_fn cruft_cb,
2138 each_loose_subdir_fn subdir_cb,
2139 void *data)
2140{
2141 struct strbuf buf = STRBUF_INIT;
2142 int r;
2143
2144 strbuf_addstr(&buf, path);
2145 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2146 subdir_cb, data);
2147 strbuf_release(&buf);
2148
2149 return r;
2150}
2151
2152struct loose_alt_odb_data {
2153 each_loose_object_fn *cb;
2154 void *data;
2155};
2156
2157static int loose_from_alt_odb(struct alternate_object_database *alt,
2158 void *vdata)
2159{
2160 struct loose_alt_odb_data *data = vdata;
2161 struct strbuf buf = STRBUF_INIT;
2162 int r;
2163
2164 strbuf_addstr(&buf, alt->path);
2165 r = for_each_loose_file_in_objdir_buf(&buf,
2166 data->cb, NULL, NULL,
2167 data->data);
2168 strbuf_release(&buf);
2169 return r;
2170}
2171
2172int for_each_loose_object(each_loose_object_fn cb, void *data,
2173 enum for_each_object_flags flags)
2174{
2175 struct loose_alt_odb_data alt;
2176 int r;
2177
2178 r = for_each_loose_file_in_objdir(get_object_directory(),
2179 cb, NULL, NULL, data);
2180 if (r)
2181 return r;
2182
2183 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2184 return 0;
2185
2186 alt.cb = cb;
2187 alt.data = data;
2188 return foreach_alt_odb(loose_from_alt_odb, &alt);
2189}
2190
2191static int check_stream_sha1(git_zstream *stream,
2192 const char *hdr,
2193 unsigned long size,
2194 const char *path,
2195 const unsigned char *expected_sha1)
2196{
2197 git_hash_ctx c;
2198 unsigned char real_sha1[GIT_MAX_RAWSZ];
2199 unsigned char buf[4096];
2200 unsigned long total_read;
2201 int status = Z_OK;
2202
2203 the_hash_algo->init_fn(&c);
2204 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2205
2206 /*
2207 * We already read some bytes into hdr, but the ones up to the NUL
2208 * do not count against the object's content size.
2209 */
2210 total_read = stream->total_out - strlen(hdr) - 1;
2211
2212 /*
2213 * This size comparison must be "<=" to read the final zlib packets;
2214 * see the comment in unpack_sha1_rest for details.
2215 */
2216 while (total_read <= size &&
2217 (status == Z_OK || status == Z_BUF_ERROR)) {
2218 stream->next_out = buf;
2219 stream->avail_out = sizeof(buf);
2220 if (size - total_read < stream->avail_out)
2221 stream->avail_out = size - total_read;
2222 status = git_inflate(stream, Z_FINISH);
2223 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2224 total_read += stream->next_out - buf;
2225 }
2226 git_inflate_end(stream);
2227
2228 if (status != Z_STREAM_END) {
2229 error(_("corrupt loose object '%s'"), sha1_to_hex(expected_sha1));
2230 return -1;
2231 }
2232 if (stream->avail_in) {
2233 error(_("garbage at end of loose object '%s'"),
2234 sha1_to_hex(expected_sha1));
2235 return -1;
2236 }
2237
2238 the_hash_algo->final_fn(real_sha1, &c);
2239 if (!hasheq(expected_sha1, real_sha1)) {
2240 error(_("sha1 mismatch for %s (expected %s)"), path,
2241 sha1_to_hex(expected_sha1));
2242 return -1;
2243 }
2244
2245 return 0;
2246}
2247
2248int read_loose_object(const char *path,
2249 const struct object_id *expected_oid,
2250 enum object_type *type,
2251 unsigned long *size,
2252 void **contents)
2253{
2254 int ret = -1;
2255 void *map = NULL;
2256 unsigned long mapsize;
2257 git_zstream stream;
2258 char hdr[MAX_HEADER_LEN];
2259
2260 *contents = NULL;
2261
2262 map = map_sha1_file_1(the_repository, path, NULL, &mapsize);
2263 if (!map) {
2264 error_errno(_("unable to mmap %s"), path);
2265 goto out;
2266 }
2267
2268 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
2269 error(_("unable to unpack header of %s"), path);
2270 goto out;
2271 }
2272
2273 *type = parse_sha1_header(hdr, size);
2274 if (*type < 0) {
2275 error(_("unable to parse header of %s"), path);
2276 git_inflate_end(&stream);
2277 goto out;
2278 }
2279
2280 if (*type == OBJ_BLOB && *size > big_file_threshold) {
2281 if (check_stream_sha1(&stream, hdr, *size, path, expected_oid->hash) < 0)
2282 goto out;
2283 } else {
2284 *contents = unpack_sha1_rest(&stream, hdr, *size, expected_oid->hash);
2285 if (!*contents) {
2286 error(_("unable to unpack contents of %s"), path);
2287 git_inflate_end(&stream);
2288 goto out;
2289 }
2290 if (check_object_signature(expected_oid, *contents,
2291 *size, type_name(*type))) {
2292 error(_("sha1 mismatch for %s (expected %s)"), path,
2293 oid_to_hex(expected_oid));
2294 free(*contents);
2295 goto out;
2296 }
2297 }
2298
2299 ret = 0; /* everything checks out */
2300
2301out:
2302 if (map)
2303 munmap(map, mapsize);
2304 return ret;
2305}