Git 2.45
[git.git] / object-file.c
blob610b1f465c4248e8c0520687049a707a8497195e
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "git-compat-util.h"
10 #include "abspath.h"
11 #include "config.h"
12 #include "convert.h"
13 #include "environment.h"
14 #include "gettext.h"
15 #include "hex.h"
16 #include "string-list.h"
17 #include "lockfile.h"
18 #include "pack.h"
19 #include "commit.h"
20 #include "run-command.h"
21 #include "refs.h"
22 #include "bulk-checkin.h"
23 #include "repository.h"
24 #include "replace-object.h"
25 #include "streaming.h"
26 #include "dir.h"
27 #include "list.h"
28 #include "quote.h"
29 #include "packfile.h"
30 #include "object-file.h"
31 #include "object-store.h"
32 #include "oidtree.h"
33 #include "path.h"
34 #include "promisor-remote.h"
35 #include "setup.h"
36 #include "submodule.h"
37 #include "fsck.h"
38 #include "loose.h"
39 #include "object-file-convert.h"
41 /* The maximum size for an object header. */
42 #define MAX_HEADER_LEN 32
45 #define EMPTY_TREE_SHA1_BIN_LITERAL \
46 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
47 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
48 #define EMPTY_TREE_SHA256_BIN_LITERAL \
49 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
50 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
51 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
52 "\x53\x21"
54 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
55 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
56 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
57 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
58 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
59 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
60 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
61 "\x18\x13"
63 static const struct object_id empty_tree_oid = {
64 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
65 .algo = GIT_HASH_SHA1,
67 static const struct object_id empty_blob_oid = {
68 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
69 .algo = GIT_HASH_SHA1,
71 static const struct object_id null_oid_sha1 = {
72 .hash = {0},
73 .algo = GIT_HASH_SHA1,
75 static const struct object_id empty_tree_oid_sha256 = {
76 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
77 .algo = GIT_HASH_SHA256,
79 static const struct object_id empty_blob_oid_sha256 = {
80 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
81 .algo = GIT_HASH_SHA256,
83 static const struct object_id null_oid_sha256 = {
84 .hash = {0},
85 .algo = GIT_HASH_SHA256,
88 static void git_hash_sha1_init(git_hash_ctx *ctx)
90 git_SHA1_Init(&ctx->sha1);
93 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
95 git_SHA1_Clone(&dst->sha1, &src->sha1);
98 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
100 git_SHA1_Update(&ctx->sha1, data, len);
103 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
105 git_SHA1_Final(hash, &ctx->sha1);
108 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
110 git_SHA1_Final(oid->hash, &ctx->sha1);
111 memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
112 oid->algo = GIT_HASH_SHA1;
116 static void git_hash_sha256_init(git_hash_ctx *ctx)
118 git_SHA256_Init(&ctx->sha256);
121 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
123 git_SHA256_Clone(&dst->sha256, &src->sha256);
126 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
128 git_SHA256_Update(&ctx->sha256, data, len);
131 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
133 git_SHA256_Final(hash, &ctx->sha256);
136 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
138 git_SHA256_Final(oid->hash, &ctx->sha256);
140 * This currently does nothing, so the compiler should optimize it out,
141 * but keep it in case we extend the hash size again.
143 memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
144 oid->algo = GIT_HASH_SHA256;
147 static void git_hash_unknown_init(git_hash_ctx *ctx UNUSED)
149 BUG("trying to init unknown hash");
152 static void git_hash_unknown_clone(git_hash_ctx *dst UNUSED,
153 const git_hash_ctx *src UNUSED)
155 BUG("trying to clone unknown hash");
158 static void git_hash_unknown_update(git_hash_ctx *ctx UNUSED,
159 const void *data UNUSED,
160 size_t len UNUSED)
162 BUG("trying to update unknown hash");
165 static void git_hash_unknown_final(unsigned char *hash UNUSED,
166 git_hash_ctx *ctx UNUSED)
168 BUG("trying to finalize unknown hash");
171 static void git_hash_unknown_final_oid(struct object_id *oid UNUSED,
172 git_hash_ctx *ctx UNUSED)
174 BUG("trying to finalize unknown hash");
177 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
179 .name = NULL,
180 .format_id = 0x00000000,
181 .rawsz = 0,
182 .hexsz = 0,
183 .blksz = 0,
184 .init_fn = git_hash_unknown_init,
185 .clone_fn = git_hash_unknown_clone,
186 .update_fn = git_hash_unknown_update,
187 .final_fn = git_hash_unknown_final,
188 .final_oid_fn = git_hash_unknown_final_oid,
189 .empty_tree = NULL,
190 .empty_blob = NULL,
191 .null_oid = NULL,
194 .name = "sha1",
195 .format_id = GIT_SHA1_FORMAT_ID,
196 .rawsz = GIT_SHA1_RAWSZ,
197 .hexsz = GIT_SHA1_HEXSZ,
198 .blksz = GIT_SHA1_BLKSZ,
199 .init_fn = git_hash_sha1_init,
200 .clone_fn = git_hash_sha1_clone,
201 .update_fn = git_hash_sha1_update,
202 .final_fn = git_hash_sha1_final,
203 .final_oid_fn = git_hash_sha1_final_oid,
204 .empty_tree = &empty_tree_oid,
205 .empty_blob = &empty_blob_oid,
206 .null_oid = &null_oid_sha1,
209 .name = "sha256",
210 .format_id = GIT_SHA256_FORMAT_ID,
211 .rawsz = GIT_SHA256_RAWSZ,
212 .hexsz = GIT_SHA256_HEXSZ,
213 .blksz = GIT_SHA256_BLKSZ,
214 .init_fn = git_hash_sha256_init,
215 .clone_fn = git_hash_sha256_clone,
216 .update_fn = git_hash_sha256_update,
217 .final_fn = git_hash_sha256_final,
218 .final_oid_fn = git_hash_sha256_final_oid,
219 .empty_tree = &empty_tree_oid_sha256,
220 .empty_blob = &empty_blob_oid_sha256,
221 .null_oid = &null_oid_sha256,
225 const struct object_id *null_oid(void)
227 return the_hash_algo->null_oid;
230 const char *empty_tree_oid_hex(void)
232 static char buf[GIT_MAX_HEXSZ + 1];
233 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
236 const char *empty_blob_oid_hex(void)
238 static char buf[GIT_MAX_HEXSZ + 1];
239 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
242 int hash_algo_by_name(const char *name)
244 int i;
245 if (!name)
246 return GIT_HASH_UNKNOWN;
247 for (i = 1; i < GIT_HASH_NALGOS; i++)
248 if (!strcmp(name, hash_algos[i].name))
249 return i;
250 return GIT_HASH_UNKNOWN;
253 int hash_algo_by_id(uint32_t format_id)
255 int i;
256 for (i = 1; i < GIT_HASH_NALGOS; i++)
257 if (format_id == hash_algos[i].format_id)
258 return i;
259 return GIT_HASH_UNKNOWN;
262 int hash_algo_by_length(int len)
264 int i;
265 for (i = 1; i < GIT_HASH_NALGOS; i++)
266 if (len == hash_algos[i].rawsz)
267 return i;
268 return GIT_HASH_UNKNOWN;
272 * This is meant to hold a *small* number of objects that you would
273 * want repo_read_object_file() to be able to return, but yet you do not want
274 * to write them into the object store (e.g. a browse-only
275 * application).
277 static struct cached_object {
278 struct object_id oid;
279 enum object_type type;
280 void *buf;
281 unsigned long size;
282 } *cached_objects;
283 static int cached_object_nr, cached_object_alloc;
285 static struct cached_object empty_tree = {
286 .oid = {
287 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
289 .type = OBJ_TREE,
290 .buf = "",
293 static struct cached_object *find_cached_object(const struct object_id *oid)
295 int i;
296 struct cached_object *co = cached_objects;
298 for (i = 0; i < cached_object_nr; i++, co++) {
299 if (oideq(&co->oid, oid))
300 return co;
302 if (oideq(oid, the_hash_algo->empty_tree))
303 return &empty_tree;
304 return NULL;
308 static int get_conv_flags(unsigned flags)
310 if (flags & HASH_RENORMALIZE)
311 return CONV_EOL_RENORMALIZE;
312 else if (flags & HASH_WRITE_OBJECT)
313 return global_conv_flags_eol | CONV_WRITE_OBJECT;
314 else
315 return 0;
319 int mkdir_in_gitdir(const char *path)
321 if (mkdir(path, 0777)) {
322 int saved_errno = errno;
323 struct stat st;
324 struct strbuf sb = STRBUF_INIT;
326 if (errno != EEXIST)
327 return -1;
329 * Are we looking at a path in a symlinked worktree
330 * whose original repository does not yet have it?
331 * e.g. .git/rr-cache pointing at its original
332 * repository in which the user hasn't performed any
333 * conflict resolution yet?
335 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
336 strbuf_readlink(&sb, path, st.st_size) ||
337 !is_absolute_path(sb.buf) ||
338 mkdir(sb.buf, 0777)) {
339 strbuf_release(&sb);
340 errno = saved_errno;
341 return -1;
343 strbuf_release(&sb);
345 return adjust_shared_perm(path);
348 static enum scld_error safe_create_leading_directories_1(char *path, int share)
350 char *next_component = path + offset_1st_component(path);
351 enum scld_error ret = SCLD_OK;
353 while (ret == SCLD_OK && next_component) {
354 struct stat st;
355 char *slash = next_component, slash_character;
357 while (*slash && !is_dir_sep(*slash))
358 slash++;
360 if (!*slash)
361 break;
363 next_component = slash + 1;
364 while (is_dir_sep(*next_component))
365 next_component++;
366 if (!*next_component)
367 break;
369 slash_character = *slash;
370 *slash = '\0';
371 if (!stat(path, &st)) {
372 /* path exists */
373 if (!S_ISDIR(st.st_mode)) {
374 errno = ENOTDIR;
375 ret = SCLD_EXISTS;
377 } else if (mkdir(path, 0777)) {
378 if (errno == EEXIST &&
379 !stat(path, &st) && S_ISDIR(st.st_mode))
380 ; /* somebody created it since we checked */
381 else if (errno == ENOENT)
383 * Either mkdir() failed because
384 * somebody just pruned the containing
385 * directory, or stat() failed because
386 * the file that was in our way was
387 * just removed. Either way, inform
388 * the caller that it might be worth
389 * trying again:
391 ret = SCLD_VANISHED;
392 else
393 ret = SCLD_FAILED;
394 } else if (share && adjust_shared_perm(path)) {
395 ret = SCLD_PERMS;
397 *slash = slash_character;
399 return ret;
402 enum scld_error safe_create_leading_directories(char *path)
404 return safe_create_leading_directories_1(path, 1);
407 enum scld_error safe_create_leading_directories_no_share(char *path)
409 return safe_create_leading_directories_1(path, 0);
412 enum scld_error safe_create_leading_directories_const(const char *path)
414 int save_errno;
415 /* path points to cache entries, so xstrdup before messing with it */
416 char *buf = xstrdup(path);
417 enum scld_error result = safe_create_leading_directories(buf);
419 save_errno = errno;
420 free(buf);
421 errno = save_errno;
422 return result;
425 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
427 int i;
428 for (i = 0; i < the_hash_algo->rawsz; i++) {
429 static char hex[] = "0123456789abcdef";
430 unsigned int val = oid->hash[i];
431 strbuf_addch(buf, hex[val >> 4]);
432 strbuf_addch(buf, hex[val & 0xf]);
433 if (!i)
434 strbuf_addch(buf, '/');
438 static const char *odb_loose_path(struct object_directory *odb,
439 struct strbuf *buf,
440 const struct object_id *oid)
442 strbuf_reset(buf);
443 strbuf_addstr(buf, odb->path);
444 strbuf_addch(buf, '/');
445 fill_loose_path(buf, oid);
446 return buf->buf;
449 const char *loose_object_path(struct repository *r, struct strbuf *buf,
450 const struct object_id *oid)
452 return odb_loose_path(r->objects->odb, buf, oid);
456 * Return non-zero iff the path is usable as an alternate object database.
458 static int alt_odb_usable(struct raw_object_store *o,
459 struct strbuf *path,
460 const char *normalized_objdir, khiter_t *pos)
462 int r;
464 /* Detect cases where alternate disappeared */
465 if (!is_directory(path->buf)) {
466 error(_("object directory %s does not exist; "
467 "check .git/objects/info/alternates"),
468 path->buf);
469 return 0;
473 * Prevent the common mistake of listing the same
474 * thing twice, or object directory itself.
476 if (!o->odb_by_path) {
477 khiter_t p;
479 o->odb_by_path = kh_init_odb_path_map();
480 assert(!o->odb->next);
481 p = kh_put_odb_path_map(o->odb_by_path, o->odb->path, &r);
482 assert(r == 1); /* never used */
483 kh_value(o->odb_by_path, p) = o->odb;
485 if (fspatheq(path->buf, normalized_objdir))
486 return 0;
487 *pos = kh_put_odb_path_map(o->odb_by_path, path->buf, &r);
488 /* r: 0 = exists, 1 = never used, 2 = deleted */
489 return r == 0 ? 0 : 1;
493 * Prepare alternate object database registry.
495 * The variable alt_odb_list points at the list of struct
496 * object_directory. The elements on this list come from
497 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
498 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
499 * whose contents is similar to that environment variable but can be
500 * LF separated. Its base points at a statically allocated buffer that
501 * contains "/the/directory/corresponding/to/.git/objects/...", while
502 * its name points just after the slash at the end of ".git/objects/"
503 * in the example above, and has enough space to hold all hex characters
504 * of the object ID, an extra slash for the first level indirection, and
505 * the terminating NUL.
507 static void read_info_alternates(struct repository *r,
508 const char *relative_base,
509 int depth);
510 static int link_alt_odb_entry(struct repository *r, const struct strbuf *entry,
511 const char *relative_base, int depth, const char *normalized_objdir)
513 struct object_directory *ent;
514 struct strbuf pathbuf = STRBUF_INIT;
515 struct strbuf tmp = STRBUF_INIT;
516 khiter_t pos;
517 int ret = -1;
519 if (!is_absolute_path(entry->buf) && relative_base) {
520 strbuf_realpath(&pathbuf, relative_base, 1);
521 strbuf_addch(&pathbuf, '/');
523 strbuf_addbuf(&pathbuf, entry);
525 if (!strbuf_realpath(&tmp, pathbuf.buf, 0)) {
526 error(_("unable to normalize alternate object path: %s"),
527 pathbuf.buf);
528 goto error;
530 strbuf_swap(&pathbuf, &tmp);
533 * The trailing slash after the directory name is given by
534 * this function at the end. Remove duplicates.
536 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
537 strbuf_setlen(&pathbuf, pathbuf.len - 1);
539 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir, &pos))
540 goto error;
542 CALLOC_ARRAY(ent, 1);
543 /* pathbuf.buf is already in r->objects->odb_by_path */
544 ent->path = strbuf_detach(&pathbuf, NULL);
546 /* add the alternate entry */
547 *r->objects->odb_tail = ent;
548 r->objects->odb_tail = &(ent->next);
549 ent->next = NULL;
550 assert(r->objects->odb_by_path);
551 kh_value(r->objects->odb_by_path, pos) = ent;
553 /* recursively add alternates */
554 read_info_alternates(r, ent->path, depth + 1);
555 ret = 0;
556 error:
557 strbuf_release(&tmp);
558 strbuf_release(&pathbuf);
559 return ret;
562 static const char *parse_alt_odb_entry(const char *string,
563 int sep,
564 struct strbuf *out)
566 const char *end;
568 strbuf_reset(out);
570 if (*string == '#') {
571 /* comment; consume up to next separator */
572 end = strchrnul(string, sep);
573 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
575 * quoted path; unquote_c_style has copied the
576 * data for us and set "end". Broken quoting (e.g.,
577 * an entry that doesn't end with a quote) falls
578 * back to the unquoted case below.
580 } else {
581 /* normal, unquoted path */
582 end = strchrnul(string, sep);
583 strbuf_add(out, string, end - string);
586 if (*end)
587 end++;
588 return end;
591 static void link_alt_odb_entries(struct repository *r, const char *alt,
592 int sep, const char *relative_base, int depth)
594 struct strbuf objdirbuf = STRBUF_INIT;
595 struct strbuf entry = STRBUF_INIT;
597 if (!alt || !*alt)
598 return;
600 if (depth > 5) {
601 error(_("%s: ignoring alternate object stores, nesting too deep"),
602 relative_base);
603 return;
606 strbuf_realpath(&objdirbuf, r->objects->odb->path, 1);
608 while (*alt) {
609 alt = parse_alt_odb_entry(alt, sep, &entry);
610 if (!entry.len)
611 continue;
612 link_alt_odb_entry(r, &entry,
613 relative_base, depth, objdirbuf.buf);
615 strbuf_release(&entry);
616 strbuf_release(&objdirbuf);
619 static void read_info_alternates(struct repository *r,
620 const char *relative_base,
621 int depth)
623 char *path;
624 struct strbuf buf = STRBUF_INIT;
626 path = xstrfmt("%s/info/alternates", relative_base);
627 if (strbuf_read_file(&buf, path, 1024) < 0) {
628 warn_on_fopen_errors(path);
629 free(path);
630 return;
633 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
634 strbuf_release(&buf);
635 free(path);
638 void add_to_alternates_file(const char *reference)
640 struct lock_file lock = LOCK_INIT;
641 char *alts = git_pathdup("objects/info/alternates");
642 FILE *in, *out;
643 int found = 0;
645 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
646 out = fdopen_lock_file(&lock, "w");
647 if (!out)
648 die_errno(_("unable to fdopen alternates lockfile"));
650 in = fopen(alts, "r");
651 if (in) {
652 struct strbuf line = STRBUF_INIT;
654 while (strbuf_getline(&line, in) != EOF) {
655 if (!strcmp(reference, line.buf)) {
656 found = 1;
657 break;
659 fprintf_or_die(out, "%s\n", line.buf);
662 strbuf_release(&line);
663 fclose(in);
665 else if (errno != ENOENT)
666 die_errno(_("unable to read alternates file"));
668 if (found) {
669 rollback_lock_file(&lock);
670 } else {
671 fprintf_or_die(out, "%s\n", reference);
672 if (commit_lock_file(&lock))
673 die_errno(_("unable to move new alternates file into place"));
674 if (the_repository->objects->loaded_alternates)
675 link_alt_odb_entries(the_repository, reference,
676 '\n', NULL, 0);
678 free(alts);
681 void add_to_alternates_memory(const char *reference)
684 * Make sure alternates are initialized, or else our entry may be
685 * overwritten when they are.
687 prepare_alt_odb(the_repository);
689 link_alt_odb_entries(the_repository, reference,
690 '\n', NULL, 0);
693 struct object_directory *set_temporary_primary_odb(const char *dir, int will_destroy)
695 struct object_directory *new_odb;
698 * Make sure alternates are initialized, or else our entry may be
699 * overwritten when they are.
701 prepare_alt_odb(the_repository);
704 * Make a new primary odb and link the old primary ODB in as an
705 * alternate
707 new_odb = xcalloc(1, sizeof(*new_odb));
708 new_odb->path = xstrdup(dir);
711 * Disable ref updates while a temporary odb is active, since
712 * the objects in the database may roll back.
714 new_odb->disable_ref_updates = 1;
715 new_odb->will_destroy = will_destroy;
716 new_odb->next = the_repository->objects->odb;
717 the_repository->objects->odb = new_odb;
718 return new_odb->next;
721 void restore_primary_odb(struct object_directory *restore_odb, const char *old_path)
723 struct object_directory *cur_odb = the_repository->objects->odb;
725 if (strcmp(old_path, cur_odb->path))
726 BUG("expected %s as primary object store; found %s",
727 old_path, cur_odb->path);
729 if (cur_odb->next != restore_odb)
730 BUG("we expect the old primary object store to be the first alternate");
732 the_repository->objects->odb = restore_odb;
733 free_object_directory(cur_odb);
737 * Compute the exact path an alternate is at and returns it. In case of
738 * error NULL is returned and the human readable error is added to `err`
739 * `path` may be relative and should point to $GIT_DIR.
740 * `err` must not be null.
742 char *compute_alternate_path(const char *path, struct strbuf *err)
744 char *ref_git = NULL;
745 const char *repo;
746 int seen_error = 0;
748 ref_git = real_pathdup(path, 0);
749 if (!ref_git) {
750 seen_error = 1;
751 strbuf_addf(err, _("path '%s' does not exist"), path);
752 goto out;
755 repo = read_gitfile(ref_git);
756 if (!repo)
757 repo = read_gitfile(mkpath("%s/.git", ref_git));
758 if (repo) {
759 free(ref_git);
760 ref_git = xstrdup(repo);
763 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
764 char *ref_git_git = mkpathdup("%s/.git", ref_git);
765 free(ref_git);
766 ref_git = ref_git_git;
767 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
768 struct strbuf sb = STRBUF_INIT;
769 seen_error = 1;
770 if (get_common_dir(&sb, ref_git)) {
771 strbuf_addf(err,
772 _("reference repository '%s' as a linked "
773 "checkout is not supported yet."),
774 path);
775 goto out;
778 strbuf_addf(err, _("reference repository '%s' is not a "
779 "local repository."), path);
780 goto out;
783 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
784 strbuf_addf(err, _("reference repository '%s' is shallow"),
785 path);
786 seen_error = 1;
787 goto out;
790 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
791 strbuf_addf(err,
792 _("reference repository '%s' is grafted"),
793 path);
794 seen_error = 1;
795 goto out;
798 out:
799 if (seen_error) {
800 FREE_AND_NULL(ref_git);
803 return ref_git;
806 struct object_directory *find_odb(struct repository *r, const char *obj_dir)
808 struct object_directory *odb;
809 char *obj_dir_real = real_pathdup(obj_dir, 1);
810 struct strbuf odb_path_real = STRBUF_INIT;
812 prepare_alt_odb(r);
813 for (odb = r->objects->odb; odb; odb = odb->next) {
814 strbuf_realpath(&odb_path_real, odb->path, 1);
815 if (!strcmp(obj_dir_real, odb_path_real.buf))
816 break;
819 free(obj_dir_real);
820 strbuf_release(&odb_path_real);
822 if (!odb)
823 die(_("could not find object directory matching %s"), obj_dir);
824 return odb;
827 static void fill_alternate_refs_command(struct child_process *cmd,
828 const char *repo_path)
830 const char *value;
832 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
833 cmd->use_shell = 1;
835 strvec_push(&cmd->args, value);
836 strvec_push(&cmd->args, repo_path);
837 } else {
838 cmd->git_cmd = 1;
840 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
841 strvec_push(&cmd->args, "for-each-ref");
842 strvec_push(&cmd->args, "--format=%(objectname)");
844 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
845 strvec_push(&cmd->args, "--");
846 strvec_split(&cmd->args, value);
850 strvec_pushv(&cmd->env, (const char **)local_repo_env);
851 cmd->out = -1;
854 static void read_alternate_refs(const char *path,
855 alternate_ref_fn *cb,
856 void *data)
858 struct child_process cmd = CHILD_PROCESS_INIT;
859 struct strbuf line = STRBUF_INIT;
860 FILE *fh;
862 fill_alternate_refs_command(&cmd, path);
864 if (start_command(&cmd))
865 return;
867 fh = xfdopen(cmd.out, "r");
868 while (strbuf_getline_lf(&line, fh) != EOF) {
869 struct object_id oid;
870 const char *p;
872 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
873 warning(_("invalid line while parsing alternate refs: %s"),
874 line.buf);
875 break;
878 cb(&oid, data);
881 fclose(fh);
882 finish_command(&cmd);
883 strbuf_release(&line);
886 struct alternate_refs_data {
887 alternate_ref_fn *fn;
888 void *data;
891 static int refs_from_alternate_cb(struct object_directory *e,
892 void *data)
894 struct strbuf path = STRBUF_INIT;
895 size_t base_len;
896 struct alternate_refs_data *cb = data;
898 if (!strbuf_realpath(&path, e->path, 0))
899 goto out;
900 if (!strbuf_strip_suffix(&path, "/objects"))
901 goto out;
902 base_len = path.len;
904 /* Is this a git repository with refs? */
905 strbuf_addstr(&path, "/refs");
906 if (!is_directory(path.buf))
907 goto out;
908 strbuf_setlen(&path, base_len);
910 read_alternate_refs(path.buf, cb->fn, cb->data);
912 out:
913 strbuf_release(&path);
914 return 0;
917 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
919 struct alternate_refs_data cb;
920 cb.fn = fn;
921 cb.data = data;
922 foreach_alt_odb(refs_from_alternate_cb, &cb);
925 int foreach_alt_odb(alt_odb_fn fn, void *cb)
927 struct object_directory *ent;
928 int r = 0;
930 prepare_alt_odb(the_repository);
931 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
932 r = fn(ent, cb);
933 if (r)
934 break;
936 return r;
939 void prepare_alt_odb(struct repository *r)
941 if (r->objects->loaded_alternates)
942 return;
944 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
946 read_info_alternates(r, r->objects->odb->path, 0);
947 r->objects->loaded_alternates = 1;
950 int has_alt_odb(struct repository *r)
952 prepare_alt_odb(r);
953 return !!r->objects->odb->next;
956 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
957 static int freshen_file(const char *fn)
959 return !utime(fn, NULL);
963 * All of the check_and_freshen functions return 1 if the file exists and was
964 * freshened (if freshening was requested), 0 otherwise. If they return
965 * 0, you should not assume that it is safe to skip a write of the object (it
966 * either does not exist on disk, or has a stale mtime and may be subject to
967 * pruning).
969 int check_and_freshen_file(const char *fn, int freshen)
971 if (access(fn, F_OK))
972 return 0;
973 if (freshen && !freshen_file(fn))
974 return 0;
975 return 1;
978 static int check_and_freshen_odb(struct object_directory *odb,
979 const struct object_id *oid,
980 int freshen)
982 static struct strbuf path = STRBUF_INIT;
983 odb_loose_path(odb, &path, oid);
984 return check_and_freshen_file(path.buf, freshen);
987 static int check_and_freshen_local(const struct object_id *oid, int freshen)
989 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
992 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
994 struct object_directory *odb;
996 prepare_alt_odb(the_repository);
997 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
998 if (check_and_freshen_odb(odb, oid, freshen))
999 return 1;
1001 return 0;
1004 static int check_and_freshen(const struct object_id *oid, int freshen)
1006 return check_and_freshen_local(oid, freshen) ||
1007 check_and_freshen_nonlocal(oid, freshen);
1010 int has_loose_object_nonlocal(const struct object_id *oid)
1012 return check_and_freshen_nonlocal(oid, 0);
1015 int has_loose_object(const struct object_id *oid)
1017 return check_and_freshen(oid, 0);
1020 static void mmap_limit_check(size_t length)
1022 static size_t limit = 0;
1023 if (!limit) {
1024 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1025 if (!limit)
1026 limit = SIZE_MAX;
1028 if (length > limit)
1029 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1030 (uintmax_t)length, (uintmax_t)limit);
1033 void *xmmap_gently(void *start, size_t length,
1034 int prot, int flags, int fd, off_t offset)
1036 void *ret;
1038 mmap_limit_check(length);
1039 ret = mmap(start, length, prot, flags, fd, offset);
1040 if (ret == MAP_FAILED && !length)
1041 ret = NULL;
1042 return ret;
1045 const char *mmap_os_err(void)
1047 static const char blank[] = "";
1048 #if defined(__linux__)
1049 if (errno == ENOMEM) {
1050 /* this continues an existing error message: */
1051 static const char enomem[] =
1052 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1053 return enomem;
1055 #endif /* OS-specific bits */
1056 return blank;
1059 void *xmmap(void *start, size_t length,
1060 int prot, int flags, int fd, off_t offset)
1062 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1063 if (ret == MAP_FAILED)
1064 die_errno(_("mmap failed%s"), mmap_os_err());
1065 return ret;
1068 static int format_object_header_literally(char *str, size_t size,
1069 const char *type, size_t objsize)
1071 return xsnprintf(str, size, "%s %"PRIuMAX, type, (uintmax_t)objsize) + 1;
1074 int format_object_header(char *str, size_t size, enum object_type type,
1075 size_t objsize)
1077 const char *name = type_name(type);
1079 if (!name)
1080 BUG("could not get a type name for 'enum object_type' value %d", type);
1082 return format_object_header_literally(str, size, name, objsize);
1085 int check_object_signature(struct repository *r, const struct object_id *oid,
1086 void *buf, unsigned long size,
1087 enum object_type type)
1089 const struct git_hash_algo *algo =
1090 oid->algo ? &hash_algos[oid->algo] : r->hash_algo;
1091 struct object_id real_oid;
1093 hash_object_file(algo, buf, size, type, &real_oid);
1095 return !oideq(oid, &real_oid) ? -1 : 0;
1098 int stream_object_signature(struct repository *r, const struct object_id *oid)
1100 struct object_id real_oid;
1101 unsigned long size;
1102 enum object_type obj_type;
1103 struct git_istream *st;
1104 git_hash_ctx c;
1105 char hdr[MAX_HEADER_LEN];
1106 int hdrlen;
1108 st = open_istream(r, oid, &obj_type, &size, NULL);
1109 if (!st)
1110 return -1;
1112 /* Generate the header */
1113 hdrlen = format_object_header(hdr, sizeof(hdr), obj_type, size);
1115 /* Sha1.. */
1116 r->hash_algo->init_fn(&c);
1117 r->hash_algo->update_fn(&c, hdr, hdrlen);
1118 for (;;) {
1119 char buf[1024 * 16];
1120 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1122 if (readlen < 0) {
1123 close_istream(st);
1124 return -1;
1126 if (!readlen)
1127 break;
1128 r->hash_algo->update_fn(&c, buf, readlen);
1130 r->hash_algo->final_oid_fn(&real_oid, &c);
1131 close_istream(st);
1132 return !oideq(oid, &real_oid) ? -1 : 0;
1135 int git_open_cloexec(const char *name, int flags)
1137 int fd;
1138 static int o_cloexec = O_CLOEXEC;
1140 fd = open(name, flags | o_cloexec);
1141 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1142 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1143 o_cloexec &= ~O_CLOEXEC;
1144 fd = open(name, flags | o_cloexec);
1147 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1149 static int fd_cloexec = FD_CLOEXEC;
1151 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1152 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1153 int flags = fcntl(fd, F_GETFD);
1154 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1155 fd_cloexec = 0;
1158 #endif
1159 return fd;
1163 * Find "oid" as a loose object in the local repository or in an alternate.
1164 * Returns 0 on success, negative on failure.
1166 * The "path" out-parameter will give the path of the object we found (if any).
1167 * Note that it may point to static storage and is only valid until another
1168 * call to stat_loose_object().
1170 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1171 struct stat *st, const char **path)
1173 struct object_directory *odb;
1174 static struct strbuf buf = STRBUF_INIT;
1176 prepare_alt_odb(r);
1177 for (odb = r->objects->odb; odb; odb = odb->next) {
1178 *path = odb_loose_path(odb, &buf, oid);
1179 if (!lstat(*path, st))
1180 return 0;
1183 return -1;
1187 * Like stat_loose_object(), but actually open the object and return the
1188 * descriptor. See the caveats on the "path" parameter above.
1190 static int open_loose_object(struct repository *r,
1191 const struct object_id *oid, const char **path)
1193 int fd;
1194 struct object_directory *odb;
1195 int most_interesting_errno = ENOENT;
1196 static struct strbuf buf = STRBUF_INIT;
1198 prepare_alt_odb(r);
1199 for (odb = r->objects->odb; odb; odb = odb->next) {
1200 *path = odb_loose_path(odb, &buf, oid);
1201 fd = git_open(*path);
1202 if (fd >= 0)
1203 return fd;
1205 if (most_interesting_errno == ENOENT)
1206 most_interesting_errno = errno;
1208 errno = most_interesting_errno;
1209 return -1;
1212 static int quick_has_loose(struct repository *r,
1213 const struct object_id *oid)
1215 struct object_directory *odb;
1217 prepare_alt_odb(r);
1218 for (odb = r->objects->odb; odb; odb = odb->next) {
1219 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1220 return 1;
1222 return 0;
1226 * Map and close the given loose object fd. The path argument is used for
1227 * error reporting.
1229 static void *map_fd(int fd, const char *path, unsigned long *size)
1231 void *map = NULL;
1232 struct stat st;
1234 if (!fstat(fd, &st)) {
1235 *size = xsize_t(st.st_size);
1236 if (!*size) {
1237 /* mmap() is forbidden on empty files */
1238 error(_("object file %s is empty"), path);
1239 close(fd);
1240 return NULL;
1242 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1244 close(fd);
1245 return map;
1248 void *map_loose_object(struct repository *r,
1249 const struct object_id *oid,
1250 unsigned long *size)
1252 const char *p;
1253 int fd = open_loose_object(r, oid, &p);
1255 if (fd < 0)
1256 return NULL;
1257 return map_fd(fd, p, size);
1260 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1261 unsigned char *map,
1262 unsigned long mapsize,
1263 void *buffer,
1264 unsigned long bufsiz,
1265 struct strbuf *header)
1267 int status;
1269 /* Get the data stream */
1270 memset(stream, 0, sizeof(*stream));
1271 stream->next_in = map;
1272 stream->avail_in = mapsize;
1273 stream->next_out = buffer;
1274 stream->avail_out = bufsiz;
1276 git_inflate_init(stream);
1277 obj_read_unlock();
1278 status = git_inflate(stream, 0);
1279 obj_read_lock();
1280 if (status < Z_OK)
1281 return ULHR_BAD;
1284 * Check if entire header is unpacked in the first iteration.
1286 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1287 return ULHR_OK;
1290 * We have a header longer than MAX_HEADER_LEN. The "header"
1291 * here is only non-NULL when we run "cat-file
1292 * --allow-unknown-type".
1294 if (!header)
1295 return ULHR_TOO_LONG;
1298 * buffer[0..bufsiz] was not large enough. Copy the partial
1299 * result out to header, and then append the result of further
1300 * reading the stream.
1302 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1303 stream->next_out = buffer;
1304 stream->avail_out = bufsiz;
1306 do {
1307 obj_read_unlock();
1308 status = git_inflate(stream, 0);
1309 obj_read_lock();
1310 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1311 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1312 return 0;
1313 stream->next_out = buffer;
1314 stream->avail_out = bufsiz;
1315 } while (status != Z_STREAM_END);
1316 return ULHR_TOO_LONG;
1319 static void *unpack_loose_rest(git_zstream *stream,
1320 void *buffer, unsigned long size,
1321 const struct object_id *oid)
1323 int bytes = strlen(buffer) + 1;
1324 unsigned char *buf = xmallocz(size);
1325 unsigned long n;
1326 int status = Z_OK;
1328 n = stream->total_out - bytes;
1329 if (n > size)
1330 n = size;
1331 memcpy(buf, (char *) buffer + bytes, n);
1332 bytes = n;
1333 if (bytes <= size) {
1335 * The above condition must be (bytes <= size), not
1336 * (bytes < size). In other words, even though we
1337 * expect no more output and set avail_out to zero,
1338 * the input zlib stream may have bytes that express
1339 * "this concludes the stream", and we *do* want to
1340 * eat that input.
1342 * Otherwise we would not be able to test that we
1343 * consumed all the input to reach the expected size;
1344 * we also want to check that zlib tells us that all
1345 * went well with status == Z_STREAM_END at the end.
1347 stream->next_out = buf + bytes;
1348 stream->avail_out = size - bytes;
1349 while (status == Z_OK) {
1350 obj_read_unlock();
1351 status = git_inflate(stream, Z_FINISH);
1352 obj_read_lock();
1355 if (status == Z_STREAM_END && !stream->avail_in) {
1356 git_inflate_end(stream);
1357 return buf;
1360 if (status < 0)
1361 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1362 else if (stream->avail_in)
1363 error(_("garbage at end of loose object '%s'"),
1364 oid_to_hex(oid));
1365 free(buf);
1366 return NULL;
1370 * We used to just use "sscanf()", but that's actually way
1371 * too permissive for what we want to check. So do an anal
1372 * object header parse by hand.
1374 int parse_loose_header(const char *hdr, struct object_info *oi)
1376 const char *type_buf = hdr;
1377 size_t size;
1378 int type, type_len = 0;
1381 * The type can be of any size but is followed by
1382 * a space.
1384 for (;;) {
1385 char c = *hdr++;
1386 if (!c)
1387 return -1;
1388 if (c == ' ')
1389 break;
1390 type_len++;
1393 type = type_from_string_gently(type_buf, type_len, 1);
1394 if (oi->type_name)
1395 strbuf_add(oi->type_name, type_buf, type_len);
1396 if (oi->typep)
1397 *oi->typep = type;
1400 * The length must follow immediately, and be in canonical
1401 * decimal format (ie "010" is not valid).
1403 size = *hdr++ - '0';
1404 if (size > 9)
1405 return -1;
1406 if (size) {
1407 for (;;) {
1408 unsigned long c = *hdr - '0';
1409 if (c > 9)
1410 break;
1411 hdr++;
1412 size = st_add(st_mult(size, 10), c);
1416 if (oi->sizep)
1417 *oi->sizep = cast_size_t_to_ulong(size);
1420 * The length must be followed by a zero byte
1422 if (*hdr)
1423 return -1;
1426 * The format is valid, but the type may still be bogus. The
1427 * Caller needs to check its oi->typep.
1429 return 0;
1432 static int loose_object_info(struct repository *r,
1433 const struct object_id *oid,
1434 struct object_info *oi, int flags)
1436 int status = 0;
1437 int fd;
1438 unsigned long mapsize;
1439 const char *path;
1440 void *map;
1441 git_zstream stream;
1442 char hdr[MAX_HEADER_LEN];
1443 struct strbuf hdrbuf = STRBUF_INIT;
1444 unsigned long size_scratch;
1445 enum object_type type_scratch;
1446 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1448 if (oi->delta_base_oid)
1449 oidclr(oi->delta_base_oid);
1452 * If we don't care about type or size, then we don't
1453 * need to look inside the object at all. Note that we
1454 * do not optimize out the stat call, even if the
1455 * caller doesn't care about the disk-size, since our
1456 * return value implicitly indicates whether the
1457 * object even exists.
1459 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1460 struct stat st;
1461 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1462 return quick_has_loose(r, oid) ? 0 : -1;
1463 if (stat_loose_object(r, oid, &st, &path) < 0)
1464 return -1;
1465 if (oi->disk_sizep)
1466 *oi->disk_sizep = st.st_size;
1467 return 0;
1470 fd = open_loose_object(r, oid, &path);
1471 if (fd < 0) {
1472 if (errno != ENOENT)
1473 error_errno(_("unable to open loose object %s"), oid_to_hex(oid));
1474 return -1;
1476 map = map_fd(fd, path, &mapsize);
1477 if (!map)
1478 return -1;
1480 if (!oi->sizep)
1481 oi->sizep = &size_scratch;
1482 if (!oi->typep)
1483 oi->typep = &type_scratch;
1485 if (oi->disk_sizep)
1486 *oi->disk_sizep = mapsize;
1488 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1489 allow_unknown ? &hdrbuf : NULL)) {
1490 case ULHR_OK:
1491 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1492 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1493 else if (!allow_unknown && *oi->typep < 0)
1494 die(_("invalid object type"));
1496 if (!oi->contentp)
1497 break;
1498 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1499 if (*oi->contentp)
1500 goto cleanup;
1502 status = -1;
1503 break;
1504 case ULHR_BAD:
1505 status = error(_("unable to unpack %s header"),
1506 oid_to_hex(oid));
1507 break;
1508 case ULHR_TOO_LONG:
1509 status = error(_("header for %s too long, exceeds %d bytes"),
1510 oid_to_hex(oid), MAX_HEADER_LEN);
1511 break;
1514 if (status && (flags & OBJECT_INFO_DIE_IF_CORRUPT))
1515 die(_("loose object %s (stored in %s) is corrupt"),
1516 oid_to_hex(oid), path);
1518 git_inflate_end(&stream);
1519 cleanup:
1520 munmap(map, mapsize);
1521 if (oi->sizep == &size_scratch)
1522 oi->sizep = NULL;
1523 strbuf_release(&hdrbuf);
1524 if (oi->typep == &type_scratch)
1525 oi->typep = NULL;
1526 oi->whence = OI_LOOSE;
1527 return status;
1530 int obj_read_use_lock = 0;
1531 pthread_mutex_t obj_read_mutex;
1533 void enable_obj_read_lock(void)
1535 if (obj_read_use_lock)
1536 return;
1538 obj_read_use_lock = 1;
1539 init_recursive_mutex(&obj_read_mutex);
1542 void disable_obj_read_lock(void)
1544 if (!obj_read_use_lock)
1545 return;
1547 obj_read_use_lock = 0;
1548 pthread_mutex_destroy(&obj_read_mutex);
1551 int fetch_if_missing = 1;
1553 static int do_oid_object_info_extended(struct repository *r,
1554 const struct object_id *oid,
1555 struct object_info *oi, unsigned flags)
1557 static struct object_info blank_oi = OBJECT_INFO_INIT;
1558 struct cached_object *co;
1559 struct pack_entry e;
1560 int rtype;
1561 const struct object_id *real = oid;
1562 int already_retried = 0;
1565 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1566 real = lookup_replace_object(r, oid);
1568 if (is_null_oid(real))
1569 return -1;
1571 if (!oi)
1572 oi = &blank_oi;
1574 co = find_cached_object(real);
1575 if (co) {
1576 if (oi->typep)
1577 *(oi->typep) = co->type;
1578 if (oi->sizep)
1579 *(oi->sizep) = co->size;
1580 if (oi->disk_sizep)
1581 *(oi->disk_sizep) = 0;
1582 if (oi->delta_base_oid)
1583 oidclr(oi->delta_base_oid);
1584 if (oi->type_name)
1585 strbuf_addstr(oi->type_name, type_name(co->type));
1586 if (oi->contentp)
1587 *oi->contentp = xmemdupz(co->buf, co->size);
1588 oi->whence = OI_CACHED;
1589 return 0;
1592 while (1) {
1593 if (find_pack_entry(r, real, &e))
1594 break;
1596 /* Most likely it's a loose object. */
1597 if (!loose_object_info(r, real, oi, flags))
1598 return 0;
1600 /* Not a loose object; someone else may have just packed it. */
1601 if (!(flags & OBJECT_INFO_QUICK)) {
1602 reprepare_packed_git(r);
1603 if (find_pack_entry(r, real, &e))
1604 break;
1608 * If r is the_repository, this might be an attempt at
1609 * accessing a submodule object as if it were in the_repository
1610 * (having called add_submodule_odb() on that submodule's ODB).
1611 * If any such ODBs exist, register them and try again.
1613 if (r == the_repository &&
1614 register_all_submodule_odb_as_alternates())
1615 /* We added some alternates; retry */
1616 continue;
1618 /* Check if it is a missing object */
1619 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1620 !already_retried &&
1621 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1622 promisor_remote_get_direct(r, real, 1);
1623 already_retried = 1;
1624 continue;
1627 if (flags & OBJECT_INFO_DIE_IF_CORRUPT) {
1628 const struct packed_git *p;
1629 if ((flags & OBJECT_INFO_LOOKUP_REPLACE) && !oideq(real, oid))
1630 die(_("replacement %s not found for %s"),
1631 oid_to_hex(real), oid_to_hex(oid));
1632 if ((p = has_packed_and_bad(r, real)))
1633 die(_("packed object %s (stored in %s) is corrupt"),
1634 oid_to_hex(real), p->pack_name);
1636 return -1;
1639 if (oi == &blank_oi)
1641 * We know that the caller doesn't actually need the
1642 * information below, so return early.
1644 return 0;
1645 rtype = packed_object_info(r, e.p, e.offset, oi);
1646 if (rtype < 0) {
1647 mark_bad_packed_object(e.p, real);
1648 return do_oid_object_info_extended(r, real, oi, 0);
1649 } else if (oi->whence == OI_PACKED) {
1650 oi->u.packed.offset = e.offset;
1651 oi->u.packed.pack = e.p;
1652 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1653 rtype == OBJ_OFS_DELTA);
1656 return 0;
1659 static int oid_object_info_convert(struct repository *r,
1660 const struct object_id *input_oid,
1661 struct object_info *input_oi, unsigned flags)
1663 const struct git_hash_algo *input_algo = &hash_algos[input_oid->algo];
1664 int do_die = flags & OBJECT_INFO_DIE_IF_CORRUPT;
1665 struct strbuf type_name = STRBUF_INIT;
1666 struct object_id oid, delta_base_oid;
1667 struct object_info new_oi, *oi;
1668 unsigned long size;
1669 void *content;
1670 int ret;
1672 if (repo_oid_to_algop(r, input_oid, the_hash_algo, &oid)) {
1673 if (do_die)
1674 die(_("missing mapping of %s to %s"),
1675 oid_to_hex(input_oid), the_hash_algo->name);
1676 return -1;
1679 /* Is new_oi needed? */
1680 oi = input_oi;
1681 if (input_oi && (input_oi->delta_base_oid || input_oi->sizep ||
1682 input_oi->contentp)) {
1683 new_oi = *input_oi;
1684 /* Does delta_base_oid need to be converted? */
1685 if (input_oi->delta_base_oid)
1686 new_oi.delta_base_oid = &delta_base_oid;
1687 /* Will the attributes differ when converted? */
1688 if (input_oi->sizep || input_oi->contentp) {
1689 new_oi.contentp = &content;
1690 new_oi.sizep = &size;
1691 new_oi.type_name = &type_name;
1693 oi = &new_oi;
1696 ret = oid_object_info_extended(r, &oid, oi, flags);
1697 if (ret)
1698 return -1;
1699 if (oi == input_oi)
1700 return ret;
1702 if (new_oi.contentp) {
1703 struct strbuf outbuf = STRBUF_INIT;
1704 enum object_type type;
1706 type = type_from_string_gently(type_name.buf, type_name.len,
1707 !do_die);
1708 if (type == -1)
1709 return -1;
1710 if (type != OBJ_BLOB) {
1711 ret = convert_object_file(&outbuf,
1712 the_hash_algo, input_algo,
1713 content, size, type, !do_die);
1714 if (ret == -1)
1715 return -1;
1716 free(content);
1717 size = outbuf.len;
1718 content = strbuf_detach(&outbuf, NULL);
1720 if (input_oi->sizep)
1721 *input_oi->sizep = size;
1722 if (input_oi->contentp)
1723 *input_oi->contentp = content;
1724 else
1725 free(content);
1726 if (input_oi->type_name)
1727 *input_oi->type_name = type_name;
1728 else
1729 strbuf_release(&type_name);
1731 if (new_oi.delta_base_oid == &delta_base_oid) {
1732 if (repo_oid_to_algop(r, &delta_base_oid, input_algo,
1733 input_oi->delta_base_oid)) {
1734 if (do_die)
1735 die(_("missing mapping of %s to %s"),
1736 oid_to_hex(&delta_base_oid),
1737 input_algo->name);
1738 return -1;
1741 input_oi->whence = new_oi.whence;
1742 input_oi->u = new_oi.u;
1743 return ret;
1746 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1747 struct object_info *oi, unsigned flags)
1749 int ret;
1751 if (oid->algo && (hash_algo_by_ptr(r->hash_algo) != oid->algo))
1752 return oid_object_info_convert(r, oid, oi, flags);
1754 obj_read_lock();
1755 ret = do_oid_object_info_extended(r, oid, oi, flags);
1756 obj_read_unlock();
1757 return ret;
1761 /* returns enum object_type or negative */
1762 int oid_object_info(struct repository *r,
1763 const struct object_id *oid,
1764 unsigned long *sizep)
1766 enum object_type type;
1767 struct object_info oi = OBJECT_INFO_INIT;
1769 oi.typep = &type;
1770 oi.sizep = sizep;
1771 if (oid_object_info_extended(r, oid, &oi,
1772 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1773 return -1;
1774 return type;
1777 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1778 struct object_id *oid)
1780 struct cached_object *co;
1782 hash_object_file(the_hash_algo, buf, len, type, oid);
1783 if (repo_has_object_file_with_flags(the_repository, oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1784 find_cached_object(oid))
1785 return 0;
1786 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1787 co = &cached_objects[cached_object_nr++];
1788 co->size = len;
1789 co->type = type;
1790 co->buf = xmalloc(len);
1791 memcpy(co->buf, buf, len);
1792 oidcpy(&co->oid, oid);
1793 return 0;
1797 * This function dies on corrupt objects; the callers who want to
1798 * deal with them should arrange to call oid_object_info_extended() and give
1799 * error messages themselves.
1801 void *repo_read_object_file(struct repository *r,
1802 const struct object_id *oid,
1803 enum object_type *type,
1804 unsigned long *size)
1806 struct object_info oi = OBJECT_INFO_INIT;
1807 unsigned flags = OBJECT_INFO_DIE_IF_CORRUPT | OBJECT_INFO_LOOKUP_REPLACE;
1808 void *data;
1810 oi.typep = type;
1811 oi.sizep = size;
1812 oi.contentp = &data;
1813 if (oid_object_info_extended(r, oid, &oi, flags))
1814 return NULL;
1816 return data;
1819 void *read_object_with_reference(struct repository *r,
1820 const struct object_id *oid,
1821 enum object_type required_type,
1822 unsigned long *size,
1823 struct object_id *actual_oid_return)
1825 enum object_type type;
1826 void *buffer;
1827 unsigned long isize;
1828 struct object_id actual_oid;
1830 oidcpy(&actual_oid, oid);
1831 while (1) {
1832 int ref_length = -1;
1833 const char *ref_type = NULL;
1835 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1836 if (!buffer)
1837 return NULL;
1838 if (type == required_type) {
1839 *size = isize;
1840 if (actual_oid_return)
1841 oidcpy(actual_oid_return, &actual_oid);
1842 return buffer;
1844 /* Handle references */
1845 else if (type == OBJ_COMMIT)
1846 ref_type = "tree ";
1847 else if (type == OBJ_TAG)
1848 ref_type = "object ";
1849 else {
1850 free(buffer);
1851 return NULL;
1853 ref_length = strlen(ref_type);
1855 if (ref_length + the_hash_algo->hexsz > isize ||
1856 memcmp(buffer, ref_type, ref_length) ||
1857 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1858 free(buffer);
1859 return NULL;
1861 free(buffer);
1862 /* Now we have the ID of the referred-to object in
1863 * actual_oid. Check again. */
1867 static void hash_object_body(const struct git_hash_algo *algo, git_hash_ctx *c,
1868 const void *buf, unsigned long len,
1869 struct object_id *oid,
1870 char *hdr, int *hdrlen)
1872 algo->init_fn(c);
1873 algo->update_fn(c, hdr, *hdrlen);
1874 algo->update_fn(c, buf, len);
1875 algo->final_oid_fn(oid, c);
1878 static void write_object_file_prepare(const struct git_hash_algo *algo,
1879 const void *buf, unsigned long len,
1880 enum object_type type, struct object_id *oid,
1881 char *hdr, int *hdrlen)
1883 git_hash_ctx c;
1885 /* Generate the header */
1886 *hdrlen = format_object_header(hdr, *hdrlen, type, len);
1888 /* Sha1.. */
1889 hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen);
1892 static void write_object_file_prepare_literally(const struct git_hash_algo *algo,
1893 const void *buf, unsigned long len,
1894 const char *type, struct object_id *oid,
1895 char *hdr, int *hdrlen)
1897 git_hash_ctx c;
1899 *hdrlen = format_object_header_literally(hdr, *hdrlen, type, len);
1900 hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen);
1904 * Move the just written object into its final resting place.
1906 int finalize_object_file(const char *tmpfile, const char *filename)
1908 int ret = 0;
1910 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1911 goto try_rename;
1912 else if (link(tmpfile, filename))
1913 ret = errno;
1916 * Coda hack - coda doesn't like cross-directory links,
1917 * so we fall back to a rename, which will mean that it
1918 * won't be able to check collisions, but that's not a
1919 * big deal.
1921 * The same holds for FAT formatted media.
1923 * When this succeeds, we just return. We have nothing
1924 * left to unlink.
1926 if (ret && ret != EEXIST) {
1927 try_rename:
1928 if (!rename(tmpfile, filename))
1929 goto out;
1930 ret = errno;
1932 unlink_or_warn(tmpfile);
1933 if (ret) {
1934 if (ret != EEXIST) {
1935 return error_errno(_("unable to write file %s"), filename);
1937 /* FIXME!!! Collision check here ? */
1940 out:
1941 if (adjust_shared_perm(filename))
1942 return error(_("unable to set permission to '%s'"), filename);
1943 return 0;
1946 static void hash_object_file_literally(const struct git_hash_algo *algo,
1947 const void *buf, unsigned long len,
1948 const char *type, struct object_id *oid)
1950 char hdr[MAX_HEADER_LEN];
1951 int hdrlen = sizeof(hdr);
1953 write_object_file_prepare_literally(algo, buf, len, type, oid, hdr, &hdrlen);
1956 void hash_object_file(const struct git_hash_algo *algo, const void *buf,
1957 unsigned long len, enum object_type type,
1958 struct object_id *oid)
1960 hash_object_file_literally(algo, buf, len, type_name(type), oid);
1963 /* Finalize a file on disk, and close it. */
1964 static void close_loose_object(int fd, const char *filename)
1966 if (the_repository->objects->odb->will_destroy)
1967 goto out;
1969 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
1970 fsync_loose_object_bulk_checkin(fd, filename);
1971 else if (fsync_object_files > 0)
1972 fsync_or_die(fd, filename);
1973 else
1974 fsync_component_or_die(FSYNC_COMPONENT_LOOSE_OBJECT, fd,
1975 filename);
1977 out:
1978 if (close(fd) != 0)
1979 die_errno(_("error when closing loose object file"));
1982 /* Size of directory component, including the ending '/' */
1983 static inline int directory_size(const char *filename)
1985 const char *s = strrchr(filename, '/');
1986 if (!s)
1987 return 0;
1988 return s - filename + 1;
1992 * This creates a temporary file in the same directory as the final
1993 * 'filename'
1995 * We want to avoid cross-directory filename renames, because those
1996 * can have problems on various filesystems (FAT, NFS, Coda).
1998 static int create_tmpfile(struct strbuf *tmp, const char *filename)
2000 int fd, dirlen = directory_size(filename);
2002 strbuf_reset(tmp);
2003 strbuf_add(tmp, filename, dirlen);
2004 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
2005 fd = git_mkstemp_mode(tmp->buf, 0444);
2006 if (fd < 0 && dirlen && errno == ENOENT) {
2008 * Make sure the directory exists; note that the contents
2009 * of the buffer are undefined after mkstemp returns an
2010 * error, so we have to rewrite the whole buffer from
2011 * scratch.
2013 strbuf_reset(tmp);
2014 strbuf_add(tmp, filename, dirlen - 1);
2015 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
2016 return -1;
2017 if (adjust_shared_perm(tmp->buf))
2018 return -1;
2020 /* Try again */
2021 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
2022 fd = git_mkstemp_mode(tmp->buf, 0444);
2024 return fd;
2028 * Common steps for loose object writers to start writing loose
2029 * objects:
2031 * - Create tmpfile for the loose object.
2032 * - Setup zlib stream for compression.
2033 * - Start to feed header to zlib stream.
2035 * Returns a "fd", which should later be provided to
2036 * end_loose_object_common().
2038 static int start_loose_object_common(struct strbuf *tmp_file,
2039 const char *filename, unsigned flags,
2040 git_zstream *stream,
2041 unsigned char *buf, size_t buflen,
2042 git_hash_ctx *c, git_hash_ctx *compat_c,
2043 char *hdr, int hdrlen)
2045 struct repository *repo = the_repository;
2046 const struct git_hash_algo *algo = repo->hash_algo;
2047 const struct git_hash_algo *compat = repo->compat_hash_algo;
2048 int fd;
2050 fd = create_tmpfile(tmp_file, filename);
2051 if (fd < 0) {
2052 if (flags & HASH_SILENT)
2053 return -1;
2054 else if (errno == EACCES)
2055 return error(_("insufficient permission for adding "
2056 "an object to repository database %s"),
2057 get_object_directory());
2058 else
2059 return error_errno(
2060 _("unable to create temporary file"));
2063 /* Setup zlib stream for compression */
2064 git_deflate_init(stream, zlib_compression_level);
2065 stream->next_out = buf;
2066 stream->avail_out = buflen;
2067 algo->init_fn(c);
2068 if (compat && compat_c)
2069 compat->init_fn(compat_c);
2071 /* Start to feed header to zlib stream */
2072 stream->next_in = (unsigned char *)hdr;
2073 stream->avail_in = hdrlen;
2074 while (git_deflate(stream, 0) == Z_OK)
2075 ; /* nothing */
2076 algo->update_fn(c, hdr, hdrlen);
2077 if (compat && compat_c)
2078 compat->update_fn(compat_c, hdr, hdrlen);
2080 return fd;
2084 * Common steps for the inner git_deflate() loop for writing loose
2085 * objects. Returns what git_deflate() returns.
2087 static int write_loose_object_common(git_hash_ctx *c, git_hash_ctx *compat_c,
2088 git_zstream *stream, const int flush,
2089 unsigned char *in0, const int fd,
2090 unsigned char *compressed,
2091 const size_t compressed_len)
2093 struct repository *repo = the_repository;
2094 const struct git_hash_algo *algo = repo->hash_algo;
2095 const struct git_hash_algo *compat = repo->compat_hash_algo;
2096 int ret;
2098 ret = git_deflate(stream, flush ? Z_FINISH : 0);
2099 algo->update_fn(c, in0, stream->next_in - in0);
2100 if (compat && compat_c)
2101 compat->update_fn(compat_c, in0, stream->next_in - in0);
2102 if (write_in_full(fd, compressed, stream->next_out - compressed) < 0)
2103 die_errno(_("unable to write loose object file"));
2104 stream->next_out = compressed;
2105 stream->avail_out = compressed_len;
2107 return ret;
2111 * Common steps for loose object writers to end writing loose objects:
2113 * - End the compression of zlib stream.
2114 * - Get the calculated oid to "oid".
2116 static int end_loose_object_common(git_hash_ctx *c, git_hash_ctx *compat_c,
2117 git_zstream *stream, struct object_id *oid,
2118 struct object_id *compat_oid)
2120 struct repository *repo = the_repository;
2121 const struct git_hash_algo *algo = repo->hash_algo;
2122 const struct git_hash_algo *compat = repo->compat_hash_algo;
2123 int ret;
2125 ret = git_deflate_end_gently(stream);
2126 if (ret != Z_OK)
2127 return ret;
2128 algo->final_oid_fn(oid, c);
2129 if (compat && compat_c)
2130 compat->final_oid_fn(compat_oid, compat_c);
2132 return Z_OK;
2135 static int write_loose_object(const struct object_id *oid, char *hdr,
2136 int hdrlen, const void *buf, unsigned long len,
2137 time_t mtime, unsigned flags)
2139 int fd, ret;
2140 unsigned char compressed[4096];
2141 git_zstream stream;
2142 git_hash_ctx c;
2143 struct object_id parano_oid;
2144 static struct strbuf tmp_file = STRBUF_INIT;
2145 static struct strbuf filename = STRBUF_INIT;
2147 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
2148 prepare_loose_object_bulk_checkin();
2150 loose_object_path(the_repository, &filename, oid);
2152 fd = start_loose_object_common(&tmp_file, filename.buf, flags,
2153 &stream, compressed, sizeof(compressed),
2154 &c, NULL, hdr, hdrlen);
2155 if (fd < 0)
2156 return -1;
2158 /* Then the data itself.. */
2159 stream.next_in = (void *)buf;
2160 stream.avail_in = len;
2161 do {
2162 unsigned char *in0 = stream.next_in;
2164 ret = write_loose_object_common(&c, NULL, &stream, 1, in0, fd,
2165 compressed, sizeof(compressed));
2166 } while (ret == Z_OK);
2168 if (ret != Z_STREAM_END)
2169 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
2170 ret);
2171 ret = end_loose_object_common(&c, NULL, &stream, &parano_oid, NULL);
2172 if (ret != Z_OK)
2173 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
2174 ret);
2175 if (!oideq(oid, &parano_oid))
2176 die(_("confused by unstable object source data for %s"),
2177 oid_to_hex(oid));
2179 close_loose_object(fd, tmp_file.buf);
2181 if (mtime) {
2182 struct utimbuf utb;
2183 utb.actime = mtime;
2184 utb.modtime = mtime;
2185 if (utime(tmp_file.buf, &utb) < 0 &&
2186 !(flags & HASH_SILENT))
2187 warning_errno(_("failed utime() on %s"), tmp_file.buf);
2190 return finalize_object_file(tmp_file.buf, filename.buf);
2193 static int freshen_loose_object(const struct object_id *oid)
2195 return check_and_freshen(oid, 1);
2198 static int freshen_packed_object(const struct object_id *oid)
2200 struct pack_entry e;
2201 if (!find_pack_entry(the_repository, oid, &e))
2202 return 0;
2203 if (e.p->is_cruft)
2204 return 0;
2205 if (e.p->freshened)
2206 return 1;
2207 if (!freshen_file(e.p->pack_name))
2208 return 0;
2209 e.p->freshened = 1;
2210 return 1;
2213 int stream_loose_object(struct input_stream *in_stream, size_t len,
2214 struct object_id *oid)
2216 const struct git_hash_algo *compat = the_repository->compat_hash_algo;
2217 struct object_id compat_oid;
2218 int fd, ret, err = 0, flush = 0;
2219 unsigned char compressed[4096];
2220 git_zstream stream;
2221 git_hash_ctx c, compat_c;
2222 struct strbuf tmp_file = STRBUF_INIT;
2223 struct strbuf filename = STRBUF_INIT;
2224 int dirlen;
2225 char hdr[MAX_HEADER_LEN];
2226 int hdrlen;
2228 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
2229 prepare_loose_object_bulk_checkin();
2231 /* Since oid is not determined, save tmp file to odb path. */
2232 strbuf_addf(&filename, "%s/", get_object_directory());
2233 hdrlen = format_object_header(hdr, sizeof(hdr), OBJ_BLOB, len);
2236 * Common steps for write_loose_object and stream_loose_object to
2237 * start writing loose objects:
2239 * - Create tmpfile for the loose object.
2240 * - Setup zlib stream for compression.
2241 * - Start to feed header to zlib stream.
2243 fd = start_loose_object_common(&tmp_file, filename.buf, 0,
2244 &stream, compressed, sizeof(compressed),
2245 &c, &compat_c, hdr, hdrlen);
2246 if (fd < 0) {
2247 err = -1;
2248 goto cleanup;
2251 /* Then the data itself.. */
2252 do {
2253 unsigned char *in0 = stream.next_in;
2255 if (!stream.avail_in && !in_stream->is_finished) {
2256 const void *in = in_stream->read(in_stream, &stream.avail_in);
2257 stream.next_in = (void *)in;
2258 in0 = (unsigned char *)in;
2259 /* All data has been read. */
2260 if (in_stream->is_finished)
2261 flush = 1;
2263 ret = write_loose_object_common(&c, &compat_c, &stream, flush, in0, fd,
2264 compressed, sizeof(compressed));
2266 * Unlike write_loose_object(), we do not have the entire
2267 * buffer. If we get Z_BUF_ERROR due to too few input bytes,
2268 * then we'll replenish them in the next input_stream->read()
2269 * call when we loop.
2271 } while (ret == Z_OK || ret == Z_BUF_ERROR);
2273 if (stream.total_in != len + hdrlen)
2274 die(_("write stream object %ld != %"PRIuMAX), stream.total_in,
2275 (uintmax_t)len + hdrlen);
2278 * Common steps for write_loose_object and stream_loose_object to
2279 * end writing loose oject:
2281 * - End the compression of zlib stream.
2282 * - Get the calculated oid.
2284 if (ret != Z_STREAM_END)
2285 die(_("unable to stream deflate new object (%d)"), ret);
2286 ret = end_loose_object_common(&c, &compat_c, &stream, oid, &compat_oid);
2287 if (ret != Z_OK)
2288 die(_("deflateEnd on stream object failed (%d)"), ret);
2289 close_loose_object(fd, tmp_file.buf);
2291 if (freshen_packed_object(oid) || freshen_loose_object(oid)) {
2292 unlink_or_warn(tmp_file.buf);
2293 goto cleanup;
2296 loose_object_path(the_repository, &filename, oid);
2298 /* We finally know the object path, and create the missing dir. */
2299 dirlen = directory_size(filename.buf);
2300 if (dirlen) {
2301 struct strbuf dir = STRBUF_INIT;
2302 strbuf_add(&dir, filename.buf, dirlen);
2304 if (mkdir_in_gitdir(dir.buf) && errno != EEXIST) {
2305 err = error_errno(_("unable to create directory %s"), dir.buf);
2306 strbuf_release(&dir);
2307 goto cleanup;
2309 strbuf_release(&dir);
2312 err = finalize_object_file(tmp_file.buf, filename.buf);
2313 if (!err && compat)
2314 err = repo_add_loose_object_map(the_repository, oid, &compat_oid);
2315 cleanup:
2316 strbuf_release(&tmp_file);
2317 strbuf_release(&filename);
2318 return err;
2321 int write_object_file_flags(const void *buf, unsigned long len,
2322 enum object_type type, struct object_id *oid,
2323 struct object_id *compat_oid_in, unsigned flags)
2325 struct repository *repo = the_repository;
2326 const struct git_hash_algo *algo = repo->hash_algo;
2327 const struct git_hash_algo *compat = repo->compat_hash_algo;
2328 struct object_id compat_oid;
2329 char hdr[MAX_HEADER_LEN];
2330 int hdrlen = sizeof(hdr);
2332 /* Generate compat_oid */
2333 if (compat) {
2334 if (compat_oid_in)
2335 oidcpy(&compat_oid, compat_oid_in);
2336 else if (type == OBJ_BLOB)
2337 hash_object_file(compat, buf, len, type, &compat_oid);
2338 else {
2339 struct strbuf converted = STRBUF_INIT;
2340 convert_object_file(&converted, algo, compat,
2341 buf, len, type, 0);
2342 hash_object_file(compat, converted.buf, converted.len,
2343 type, &compat_oid);
2344 strbuf_release(&converted);
2348 /* Normally if we have it in the pack then we do not bother writing
2349 * it out into .git/objects/??/?{38} file.
2351 write_object_file_prepare(algo, buf, len, type, oid, hdr, &hdrlen);
2352 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2353 return 0;
2354 if (write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags))
2355 return -1;
2356 if (compat)
2357 return repo_add_loose_object_map(repo, oid, &compat_oid);
2358 return 0;
2361 int write_object_file_literally(const void *buf, unsigned long len,
2362 const char *type, struct object_id *oid,
2363 unsigned flags)
2365 char *header;
2366 struct repository *repo = the_repository;
2367 const struct git_hash_algo *algo = repo->hash_algo;
2368 const struct git_hash_algo *compat = repo->compat_hash_algo;
2369 struct object_id compat_oid;
2370 int hdrlen, status = 0;
2371 int compat_type = -1;
2373 if (compat) {
2374 compat_type = type_from_string_gently(type, -1, 1);
2375 if (compat_type == OBJ_BLOB)
2376 hash_object_file(compat, buf, len, compat_type,
2377 &compat_oid);
2378 else if (compat_type != -1) {
2379 struct strbuf converted = STRBUF_INIT;
2380 convert_object_file(&converted, algo, compat,
2381 buf, len, compat_type, 0);
2382 hash_object_file(compat, converted.buf, converted.len,
2383 compat_type, &compat_oid);
2384 strbuf_release(&converted);
2388 /* type string, SP, %lu of the length plus NUL must fit this */
2389 hdrlen = strlen(type) + MAX_HEADER_LEN;
2390 header = xmalloc(hdrlen);
2391 write_object_file_prepare_literally(the_hash_algo, buf, len, type,
2392 oid, header, &hdrlen);
2394 if (!(flags & HASH_WRITE_OBJECT))
2395 goto cleanup;
2396 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2397 goto cleanup;
2398 status = write_loose_object(oid, header, hdrlen, buf, len, 0, 0);
2399 if (compat_type != -1)
2400 return repo_add_loose_object_map(repo, oid, &compat_oid);
2402 cleanup:
2403 free(header);
2404 return status;
2407 int force_object_loose(const struct object_id *oid, time_t mtime)
2409 struct repository *repo = the_repository;
2410 const struct git_hash_algo *compat = repo->compat_hash_algo;
2411 void *buf;
2412 unsigned long len;
2413 struct object_info oi = OBJECT_INFO_INIT;
2414 struct object_id compat_oid;
2415 enum object_type type;
2416 char hdr[MAX_HEADER_LEN];
2417 int hdrlen;
2418 int ret;
2420 if (has_loose_object(oid))
2421 return 0;
2422 oi.typep = &type;
2423 oi.sizep = &len;
2424 oi.contentp = &buf;
2425 if (oid_object_info_extended(the_repository, oid, &oi, 0))
2426 return error(_("cannot read object for %s"), oid_to_hex(oid));
2427 if (compat) {
2428 if (repo_oid_to_algop(repo, oid, compat, &compat_oid))
2429 return error(_("cannot map object %s to %s"),
2430 oid_to_hex(oid), compat->name);
2432 hdrlen = format_object_header(hdr, sizeof(hdr), type, len);
2433 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime, 0);
2434 if (!ret && compat)
2435 ret = repo_add_loose_object_map(the_repository, oid, &compat_oid);
2436 free(buf);
2438 return ret;
2441 int has_object(struct repository *r, const struct object_id *oid,
2442 unsigned flags)
2444 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2445 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2446 (quick ? OBJECT_INFO_QUICK : 0);
2448 if (!startup_info->have_repository)
2449 return 0;
2450 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2453 int repo_has_object_file_with_flags(struct repository *r,
2454 const struct object_id *oid, int flags)
2456 if (!startup_info->have_repository)
2457 return 0;
2458 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2461 int repo_has_object_file(struct repository *r,
2462 const struct object_id *oid)
2464 return repo_has_object_file_with_flags(r, oid, 0);
2468 * We can't use the normal fsck_error_function() for index_mem(),
2469 * because we don't yet have a valid oid for it to report. Instead,
2470 * report the minimal fsck error here, and rely on the caller to
2471 * give more context.
2473 static int hash_format_check_report(struct fsck_options *opts UNUSED,
2474 const struct object_id *oid UNUSED,
2475 enum object_type object_type UNUSED,
2476 enum fsck_msg_type msg_type UNUSED,
2477 enum fsck_msg_id msg_id UNUSED,
2478 const char *message)
2480 error(_("object fails fsck: %s"), message);
2481 return 1;
2484 static int index_mem(struct index_state *istate,
2485 struct object_id *oid, void *buf, size_t size,
2486 enum object_type type,
2487 const char *path, unsigned flags)
2489 int ret = 0;
2490 int re_allocated = 0;
2491 int write_object = flags & HASH_WRITE_OBJECT;
2493 if (!type)
2494 type = OBJ_BLOB;
2497 * Convert blobs to git internal format
2499 if ((type == OBJ_BLOB) && path) {
2500 struct strbuf nbuf = STRBUF_INIT;
2501 if (convert_to_git(istate, path, buf, size, &nbuf,
2502 get_conv_flags(flags))) {
2503 buf = strbuf_detach(&nbuf, &size);
2504 re_allocated = 1;
2507 if (flags & HASH_FORMAT_CHECK) {
2508 struct fsck_options opts = FSCK_OPTIONS_DEFAULT;
2510 opts.strict = 1;
2511 opts.error_func = hash_format_check_report;
2512 if (fsck_buffer(null_oid(), type, buf, size, &opts))
2513 die(_("refusing to create malformed object"));
2514 fsck_finish(&opts);
2517 if (write_object)
2518 ret = write_object_file(buf, size, type, oid);
2519 else
2520 hash_object_file(the_hash_algo, buf, size, type, oid);
2521 if (re_allocated)
2522 free(buf);
2523 return ret;
2526 static int index_stream_convert_blob(struct index_state *istate,
2527 struct object_id *oid,
2528 int fd,
2529 const char *path,
2530 unsigned flags)
2532 int ret = 0;
2533 const int write_object = flags & HASH_WRITE_OBJECT;
2534 struct strbuf sbuf = STRBUF_INIT;
2536 assert(path);
2537 assert(would_convert_to_git_filter_fd(istate, path));
2539 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2540 get_conv_flags(flags));
2542 if (write_object)
2543 ret = write_object_file(sbuf.buf, sbuf.len, OBJ_BLOB,
2544 oid);
2545 else
2546 hash_object_file(the_hash_algo, sbuf.buf, sbuf.len, OBJ_BLOB,
2547 oid);
2548 strbuf_release(&sbuf);
2549 return ret;
2552 static int index_pipe(struct index_state *istate, struct object_id *oid,
2553 int fd, enum object_type type,
2554 const char *path, unsigned flags)
2556 struct strbuf sbuf = STRBUF_INIT;
2557 int ret;
2559 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2560 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2561 else
2562 ret = -1;
2563 strbuf_release(&sbuf);
2564 return ret;
2567 #define SMALL_FILE_SIZE (32*1024)
2569 static int index_core(struct index_state *istate,
2570 struct object_id *oid, int fd, size_t size,
2571 enum object_type type, const char *path,
2572 unsigned flags)
2574 int ret;
2576 if (!size) {
2577 ret = index_mem(istate, oid, "", size, type, path, flags);
2578 } else if (size <= SMALL_FILE_SIZE) {
2579 char *buf = xmalloc(size);
2580 ssize_t read_result = read_in_full(fd, buf, size);
2581 if (read_result < 0)
2582 ret = error_errno(_("read error while indexing %s"),
2583 path ? path : "<unknown>");
2584 else if (read_result != size)
2585 ret = error(_("short read while indexing %s"),
2586 path ? path : "<unknown>");
2587 else
2588 ret = index_mem(istate, oid, buf, size, type, path, flags);
2589 free(buf);
2590 } else {
2591 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2592 ret = index_mem(istate, oid, buf, size, type, path, flags);
2593 munmap(buf, size);
2595 return ret;
2599 * This creates one packfile per large blob unless bulk-checkin
2600 * machinery is "plugged".
2602 * This also bypasses the usual "convert-to-git" dance, and that is on
2603 * purpose. We could write a streaming version of the converting
2604 * functions and insert that before feeding the data to fast-import
2605 * (or equivalent in-core API described above). However, that is
2606 * somewhat complicated, as we do not know the size of the filter
2607 * result, which we need to know beforehand when writing a git object.
2608 * Since the primary motivation for trying to stream from the working
2609 * tree file and to avoid mmaping it in core is to deal with large
2610 * binary blobs, they generally do not want to get any conversion, and
2611 * callers should avoid this code path when filters are requested.
2613 static int index_blob_stream(struct object_id *oid, int fd, size_t size,
2614 const char *path,
2615 unsigned flags)
2617 return index_blob_bulk_checkin(oid, fd, size, path, flags);
2620 int index_fd(struct index_state *istate, struct object_id *oid,
2621 int fd, struct stat *st,
2622 enum object_type type, const char *path, unsigned flags)
2624 int ret;
2627 * Call xsize_t() only when needed to avoid potentially unnecessary
2628 * die() for large files.
2630 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2631 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2632 else if (!S_ISREG(st->st_mode))
2633 ret = index_pipe(istate, oid, fd, type, path, flags);
2634 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2635 (path && would_convert_to_git(istate, path)))
2636 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2637 type, path, flags);
2638 else
2639 ret = index_blob_stream(oid, fd, xsize_t(st->st_size), path,
2640 flags);
2641 close(fd);
2642 return ret;
2645 int index_path(struct index_state *istate, struct object_id *oid,
2646 const char *path, struct stat *st, unsigned flags)
2648 int fd;
2649 struct strbuf sb = STRBUF_INIT;
2650 int rc = 0;
2652 switch (st->st_mode & S_IFMT) {
2653 case S_IFREG:
2654 fd = open(path, O_RDONLY);
2655 if (fd < 0)
2656 return error_errno("open(\"%s\")", path);
2657 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2658 return error(_("%s: failed to insert into database"),
2659 path);
2660 break;
2661 case S_IFLNK:
2662 if (strbuf_readlink(&sb, path, st->st_size))
2663 return error_errno("readlink(\"%s\")", path);
2664 if (!(flags & HASH_WRITE_OBJECT))
2665 hash_object_file(the_hash_algo, sb.buf, sb.len,
2666 OBJ_BLOB, oid);
2667 else if (write_object_file(sb.buf, sb.len, OBJ_BLOB, oid))
2668 rc = error(_("%s: failed to insert into database"), path);
2669 strbuf_release(&sb);
2670 break;
2671 case S_IFDIR:
2672 return resolve_gitlink_ref(path, "HEAD", oid);
2673 default:
2674 return error(_("%s: unsupported file type"), path);
2676 return rc;
2679 int read_pack_header(int fd, struct pack_header *header)
2681 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2682 /* "eof before pack header was fully read" */
2683 return PH_ERROR_EOF;
2685 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2686 /* "protocol error (pack signature mismatch detected)" */
2687 return PH_ERROR_PACK_SIGNATURE;
2688 if (!pack_version_ok(header->hdr_version))
2689 /* "protocol error (pack version unsupported)" */
2690 return PH_ERROR_PROTOCOL;
2691 return 0;
2694 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2696 enum object_type type = oid_object_info(the_repository, oid, NULL);
2697 if (type < 0)
2698 die(_("%s is not a valid object"), oid_to_hex(oid));
2699 if (type != expect)
2700 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2701 type_name(expect));
2704 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2705 struct strbuf *path,
2706 each_loose_object_fn obj_cb,
2707 each_loose_cruft_fn cruft_cb,
2708 each_loose_subdir_fn subdir_cb,
2709 void *data)
2711 size_t origlen, baselen;
2712 DIR *dir;
2713 struct dirent *de;
2714 int r = 0;
2715 struct object_id oid;
2717 if (subdir_nr > 0xff)
2718 BUG("invalid loose object subdirectory: %x", subdir_nr);
2720 origlen = path->len;
2721 strbuf_complete(path, '/');
2722 strbuf_addf(path, "%02x", subdir_nr);
2724 dir = opendir(path->buf);
2725 if (!dir) {
2726 if (errno != ENOENT)
2727 r = error_errno(_("unable to open %s"), path->buf);
2728 strbuf_setlen(path, origlen);
2729 return r;
2732 oid.hash[0] = subdir_nr;
2733 strbuf_addch(path, '/');
2734 baselen = path->len;
2736 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2737 size_t namelen;
2739 namelen = strlen(de->d_name);
2740 strbuf_setlen(path, baselen);
2741 strbuf_add(path, de->d_name, namelen);
2742 if (namelen == the_hash_algo->hexsz - 2 &&
2743 !hex_to_bytes(oid.hash + 1, de->d_name,
2744 the_hash_algo->rawsz - 1)) {
2745 oid_set_algo(&oid, the_hash_algo);
2746 if (obj_cb) {
2747 r = obj_cb(&oid, path->buf, data);
2748 if (r)
2749 break;
2751 continue;
2754 if (cruft_cb) {
2755 r = cruft_cb(de->d_name, path->buf, data);
2756 if (r)
2757 break;
2760 closedir(dir);
2762 strbuf_setlen(path, baselen - 1);
2763 if (!r && subdir_cb)
2764 r = subdir_cb(subdir_nr, path->buf, data);
2766 strbuf_setlen(path, origlen);
2768 return r;
2771 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2772 each_loose_object_fn obj_cb,
2773 each_loose_cruft_fn cruft_cb,
2774 each_loose_subdir_fn subdir_cb,
2775 void *data)
2777 int r = 0;
2778 int i;
2780 for (i = 0; i < 256; i++) {
2781 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2782 subdir_cb, data);
2783 if (r)
2784 break;
2787 return r;
2790 int for_each_loose_file_in_objdir(const char *path,
2791 each_loose_object_fn obj_cb,
2792 each_loose_cruft_fn cruft_cb,
2793 each_loose_subdir_fn subdir_cb,
2794 void *data)
2796 struct strbuf buf = STRBUF_INIT;
2797 int r;
2799 strbuf_addstr(&buf, path);
2800 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2801 subdir_cb, data);
2802 strbuf_release(&buf);
2804 return r;
2807 int for_each_loose_object(each_loose_object_fn cb, void *data,
2808 enum for_each_object_flags flags)
2810 struct object_directory *odb;
2812 prepare_alt_odb(the_repository);
2813 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2814 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2815 NULL, data);
2816 if (r)
2817 return r;
2819 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2820 break;
2823 return 0;
2826 static int append_loose_object(const struct object_id *oid,
2827 const char *path UNUSED,
2828 void *data)
2830 oidtree_insert(data, oid);
2831 return 0;
2834 struct oidtree *odb_loose_cache(struct object_directory *odb,
2835 const struct object_id *oid)
2837 int subdir_nr = oid->hash[0];
2838 struct strbuf buf = STRBUF_INIT;
2839 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2840 size_t word_index = subdir_nr / word_bits;
2841 size_t mask = (size_t)1u << (subdir_nr % word_bits);
2842 uint32_t *bitmap;
2844 if (subdir_nr < 0 ||
2845 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2846 BUG("subdir_nr out of range");
2848 bitmap = &odb->loose_objects_subdir_seen[word_index];
2849 if (*bitmap & mask)
2850 return odb->loose_objects_cache;
2851 if (!odb->loose_objects_cache) {
2852 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2853 oidtree_init(odb->loose_objects_cache);
2855 strbuf_addstr(&buf, odb->path);
2856 for_each_file_in_obj_subdir(subdir_nr, &buf,
2857 append_loose_object,
2858 NULL, NULL,
2859 odb->loose_objects_cache);
2860 *bitmap |= mask;
2861 strbuf_release(&buf);
2862 return odb->loose_objects_cache;
2865 void odb_clear_loose_cache(struct object_directory *odb)
2867 oidtree_clear(odb->loose_objects_cache);
2868 FREE_AND_NULL(odb->loose_objects_cache);
2869 memset(&odb->loose_objects_subdir_seen, 0,
2870 sizeof(odb->loose_objects_subdir_seen));
2873 static int check_stream_oid(git_zstream *stream,
2874 const char *hdr,
2875 unsigned long size,
2876 const char *path,
2877 const struct object_id *expected_oid)
2879 git_hash_ctx c;
2880 struct object_id real_oid;
2881 unsigned char buf[4096];
2882 unsigned long total_read;
2883 int status = Z_OK;
2885 the_hash_algo->init_fn(&c);
2886 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2889 * We already read some bytes into hdr, but the ones up to the NUL
2890 * do not count against the object's content size.
2892 total_read = stream->total_out - strlen(hdr) - 1;
2895 * This size comparison must be "<=" to read the final zlib packets;
2896 * see the comment in unpack_loose_rest for details.
2898 while (total_read <= size &&
2899 (status == Z_OK ||
2900 (status == Z_BUF_ERROR && !stream->avail_out))) {
2901 stream->next_out = buf;
2902 stream->avail_out = sizeof(buf);
2903 if (size - total_read < stream->avail_out)
2904 stream->avail_out = size - total_read;
2905 status = git_inflate(stream, Z_FINISH);
2906 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2907 total_read += stream->next_out - buf;
2909 git_inflate_end(stream);
2911 if (status != Z_STREAM_END) {
2912 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2913 return -1;
2915 if (stream->avail_in) {
2916 error(_("garbage at end of loose object '%s'"),
2917 oid_to_hex(expected_oid));
2918 return -1;
2921 the_hash_algo->final_oid_fn(&real_oid, &c);
2922 if (!oideq(expected_oid, &real_oid)) {
2923 error(_("hash mismatch for %s (expected %s)"), path,
2924 oid_to_hex(expected_oid));
2925 return -1;
2928 return 0;
2931 int read_loose_object(const char *path,
2932 const struct object_id *expected_oid,
2933 struct object_id *real_oid,
2934 void **contents,
2935 struct object_info *oi)
2937 int ret = -1;
2938 int fd;
2939 void *map = NULL;
2940 unsigned long mapsize;
2941 git_zstream stream;
2942 char hdr[MAX_HEADER_LEN];
2943 unsigned long *size = oi->sizep;
2945 fd = git_open(path);
2946 if (fd >= 0)
2947 map = map_fd(fd, path, &mapsize);
2948 if (!map) {
2949 error_errno(_("unable to mmap %s"), path);
2950 goto out;
2953 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2954 NULL) != ULHR_OK) {
2955 error(_("unable to unpack header of %s"), path);
2956 goto out;
2959 if (parse_loose_header(hdr, oi) < 0) {
2960 error(_("unable to parse header of %s"), path);
2961 git_inflate_end(&stream);
2962 goto out;
2965 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2966 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2967 goto out;
2968 } else {
2969 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2970 if (!*contents) {
2971 error(_("unable to unpack contents of %s"), path);
2972 git_inflate_end(&stream);
2973 goto out;
2975 hash_object_file_literally(the_repository->hash_algo,
2976 *contents, *size,
2977 oi->type_name->buf, real_oid);
2978 if (!oideq(expected_oid, real_oid))
2979 goto out;
2982 ret = 0; /* everything checks out */
2984 out:
2985 if (map)
2986 munmap(map, mapsize);
2987 return ret;