test-ctype: test iscntrl, ispunct, isxdigit and isprint
[git/debian.git] / object-file.c
blob26290554bb4c5f1d679194fe7d3f2b5c78a273e2
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 "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 "hash-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 "object-store.h"
34 #include "promisor-remote.h"
35 #include "submodule.h"
37 /* The maximum size for an object header. */
38 #define MAX_HEADER_LEN 32
41 #define EMPTY_TREE_SHA1_BIN_LITERAL \
42 "\x4b\x82\x5d\xc6\x42\xcb\x6e\xb9\xa0\x60" \
43 "\xe5\x4b\xf8\xd6\x92\x88\xfb\xee\x49\x04"
44 #define EMPTY_TREE_SHA256_BIN_LITERAL \
45 "\x6e\xf1\x9b\x41\x22\x5c\x53\x69\xf1\xc1" \
46 "\x04\xd4\x5d\x8d\x85\xef\xa9\xb0\x57\xb5" \
47 "\x3b\x14\xb4\xb9\xb9\x39\xdd\x74\xde\xcc" \
48 "\x53\x21"
50 #define EMPTY_BLOB_SHA1_BIN_LITERAL \
51 "\xe6\x9d\xe2\x9b\xb2\xd1\xd6\x43\x4b\x8b" \
52 "\x29\xae\x77\x5a\xd8\xc2\xe4\x8c\x53\x91"
53 #define EMPTY_BLOB_SHA256_BIN_LITERAL \
54 "\x47\x3a\x0f\x4c\x3b\xe8\xa9\x36\x81\xa2" \
55 "\x67\xe3\xb1\xe9\xa7\xdc\xda\x11\x85\x43" \
56 "\x6f\xe1\x41\xf7\x74\x91\x20\xa3\x03\x72" \
57 "\x18\x13"
59 static const struct object_id empty_tree_oid = {
60 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
61 .algo = GIT_HASH_SHA1,
63 static const struct object_id empty_blob_oid = {
64 .hash = EMPTY_BLOB_SHA1_BIN_LITERAL,
65 .algo = GIT_HASH_SHA1,
67 static const struct object_id null_oid_sha1 = {
68 .hash = {0},
69 .algo = GIT_HASH_SHA1,
71 static const struct object_id empty_tree_oid_sha256 = {
72 .hash = EMPTY_TREE_SHA256_BIN_LITERAL,
73 .algo = GIT_HASH_SHA256,
75 static const struct object_id empty_blob_oid_sha256 = {
76 .hash = EMPTY_BLOB_SHA256_BIN_LITERAL,
77 .algo = GIT_HASH_SHA256,
79 static const struct object_id null_oid_sha256 = {
80 .hash = {0},
81 .algo = GIT_HASH_SHA256,
84 static void git_hash_sha1_init(git_hash_ctx *ctx)
86 git_SHA1_Init(&ctx->sha1);
89 static void git_hash_sha1_clone(git_hash_ctx *dst, const git_hash_ctx *src)
91 git_SHA1_Clone(&dst->sha1, &src->sha1);
94 static void git_hash_sha1_update(git_hash_ctx *ctx, const void *data, size_t len)
96 git_SHA1_Update(&ctx->sha1, data, len);
99 static void git_hash_sha1_final(unsigned char *hash, git_hash_ctx *ctx)
101 git_SHA1_Final(hash, &ctx->sha1);
104 static void git_hash_sha1_final_oid(struct object_id *oid, git_hash_ctx *ctx)
106 git_SHA1_Final(oid->hash, &ctx->sha1);
107 memset(oid->hash + GIT_SHA1_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA1_RAWSZ);
108 oid->algo = GIT_HASH_SHA1;
112 static void git_hash_sha256_init(git_hash_ctx *ctx)
114 git_SHA256_Init(&ctx->sha256);
117 static void git_hash_sha256_clone(git_hash_ctx *dst, const git_hash_ctx *src)
119 git_SHA256_Clone(&dst->sha256, &src->sha256);
122 static void git_hash_sha256_update(git_hash_ctx *ctx, const void *data, size_t len)
124 git_SHA256_Update(&ctx->sha256, data, len);
127 static void git_hash_sha256_final(unsigned char *hash, git_hash_ctx *ctx)
129 git_SHA256_Final(hash, &ctx->sha256);
132 static void git_hash_sha256_final_oid(struct object_id *oid, git_hash_ctx *ctx)
134 git_SHA256_Final(oid->hash, &ctx->sha256);
136 * This currently does nothing, so the compiler should optimize it out,
137 * but keep it in case we extend the hash size again.
139 memset(oid->hash + GIT_SHA256_RAWSZ, 0, GIT_MAX_RAWSZ - GIT_SHA256_RAWSZ);
140 oid->algo = GIT_HASH_SHA256;
143 static void git_hash_unknown_init(git_hash_ctx *ctx UNUSED)
145 BUG("trying to init unknown hash");
148 static void git_hash_unknown_clone(git_hash_ctx *dst UNUSED,
149 const git_hash_ctx *src UNUSED)
151 BUG("trying to clone unknown hash");
154 static void git_hash_unknown_update(git_hash_ctx *ctx UNUSED,
155 const void *data UNUSED,
156 size_t len UNUSED)
158 BUG("trying to update unknown hash");
161 static void git_hash_unknown_final(unsigned char *hash UNUSED,
162 git_hash_ctx *ctx UNUSED)
164 BUG("trying to finalize unknown hash");
167 static void git_hash_unknown_final_oid(struct object_id *oid UNUSED,
168 git_hash_ctx *ctx UNUSED)
170 BUG("trying to finalize unknown hash");
173 const struct git_hash_algo hash_algos[GIT_HASH_NALGOS] = {
175 .name = NULL,
176 .format_id = 0x00000000,
177 .rawsz = 0,
178 .hexsz = 0,
179 .blksz = 0,
180 .init_fn = git_hash_unknown_init,
181 .clone_fn = git_hash_unknown_clone,
182 .update_fn = git_hash_unknown_update,
183 .final_fn = git_hash_unknown_final,
184 .final_oid_fn = git_hash_unknown_final_oid,
185 .empty_tree = NULL,
186 .empty_blob = NULL,
187 .null_oid = NULL,
190 .name = "sha1",
191 .format_id = GIT_SHA1_FORMAT_ID,
192 .rawsz = GIT_SHA1_RAWSZ,
193 .hexsz = GIT_SHA1_HEXSZ,
194 .blksz = GIT_SHA1_BLKSZ,
195 .init_fn = git_hash_sha1_init,
196 .clone_fn = git_hash_sha1_clone,
197 .update_fn = git_hash_sha1_update,
198 .final_fn = git_hash_sha1_final,
199 .final_oid_fn = git_hash_sha1_final_oid,
200 .empty_tree = &empty_tree_oid,
201 .empty_blob = &empty_blob_oid,
202 .null_oid = &null_oid_sha1,
205 .name = "sha256",
206 .format_id = GIT_SHA256_FORMAT_ID,
207 .rawsz = GIT_SHA256_RAWSZ,
208 .hexsz = GIT_SHA256_HEXSZ,
209 .blksz = GIT_SHA256_BLKSZ,
210 .init_fn = git_hash_sha256_init,
211 .clone_fn = git_hash_sha256_clone,
212 .update_fn = git_hash_sha256_update,
213 .final_fn = git_hash_sha256_final,
214 .final_oid_fn = git_hash_sha256_final_oid,
215 .empty_tree = &empty_tree_oid_sha256,
216 .empty_blob = &empty_blob_oid_sha256,
217 .null_oid = &null_oid_sha256,
221 const struct object_id *null_oid(void)
223 return the_hash_algo->null_oid;
226 const char *empty_tree_oid_hex(void)
228 static char buf[GIT_MAX_HEXSZ + 1];
229 return oid_to_hex_r(buf, the_hash_algo->empty_tree);
232 const char *empty_blob_oid_hex(void)
234 static char buf[GIT_MAX_HEXSZ + 1];
235 return oid_to_hex_r(buf, the_hash_algo->empty_blob);
238 int hash_algo_by_name(const char *name)
240 int i;
241 if (!name)
242 return GIT_HASH_UNKNOWN;
243 for (i = 1; i < GIT_HASH_NALGOS; i++)
244 if (!strcmp(name, hash_algos[i].name))
245 return i;
246 return GIT_HASH_UNKNOWN;
249 int hash_algo_by_id(uint32_t format_id)
251 int i;
252 for (i = 1; i < GIT_HASH_NALGOS; i++)
253 if (format_id == hash_algos[i].format_id)
254 return i;
255 return GIT_HASH_UNKNOWN;
258 int hash_algo_by_length(int len)
260 int i;
261 for (i = 1; i < GIT_HASH_NALGOS; i++)
262 if (len == hash_algos[i].rawsz)
263 return i;
264 return GIT_HASH_UNKNOWN;
268 * This is meant to hold a *small* number of objects that you would
269 * want read_object_file() to be able to return, but yet you do not want
270 * to write them into the object store (e.g. a browse-only
271 * application).
273 static struct cached_object {
274 struct object_id oid;
275 enum object_type type;
276 void *buf;
277 unsigned long size;
278 } *cached_objects;
279 static int cached_object_nr, cached_object_alloc;
281 static struct cached_object empty_tree = {
282 .oid = {
283 .hash = EMPTY_TREE_SHA1_BIN_LITERAL,
285 .type = OBJ_TREE,
286 .buf = "",
289 static struct cached_object *find_cached_object(const struct object_id *oid)
291 int i;
292 struct cached_object *co = cached_objects;
294 for (i = 0; i < cached_object_nr; i++, co++) {
295 if (oideq(&co->oid, oid))
296 return co;
298 if (oideq(oid, the_hash_algo->empty_tree))
299 return &empty_tree;
300 return NULL;
304 static int get_conv_flags(unsigned flags)
306 if (flags & HASH_RENORMALIZE)
307 return CONV_EOL_RENORMALIZE;
308 else if (flags & HASH_WRITE_OBJECT)
309 return global_conv_flags_eol | CONV_WRITE_OBJECT;
310 else
311 return 0;
315 int mkdir_in_gitdir(const char *path)
317 if (mkdir(path, 0777)) {
318 int saved_errno = errno;
319 struct stat st;
320 struct strbuf sb = STRBUF_INIT;
322 if (errno != EEXIST)
323 return -1;
325 * Are we looking at a path in a symlinked worktree
326 * whose original repository does not yet have it?
327 * e.g. .git/rr-cache pointing at its original
328 * repository in which the user hasn't performed any
329 * conflict resolution yet?
331 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
332 strbuf_readlink(&sb, path, st.st_size) ||
333 !is_absolute_path(sb.buf) ||
334 mkdir(sb.buf, 0777)) {
335 strbuf_release(&sb);
336 errno = saved_errno;
337 return -1;
339 strbuf_release(&sb);
341 return adjust_shared_perm(path);
344 static enum scld_error safe_create_leading_directories_1(char *path, int share)
346 char *next_component = path + offset_1st_component(path);
347 enum scld_error ret = SCLD_OK;
349 while (ret == SCLD_OK && next_component) {
350 struct stat st;
351 char *slash = next_component, slash_character;
353 while (*slash && !is_dir_sep(*slash))
354 slash++;
356 if (!*slash)
357 break;
359 next_component = slash + 1;
360 while (is_dir_sep(*next_component))
361 next_component++;
362 if (!*next_component)
363 break;
365 slash_character = *slash;
366 *slash = '\0';
367 if (!stat(path, &st)) {
368 /* path exists */
369 if (!S_ISDIR(st.st_mode)) {
370 errno = ENOTDIR;
371 ret = SCLD_EXISTS;
373 } else if (mkdir(path, 0777)) {
374 if (errno == EEXIST &&
375 !stat(path, &st) && S_ISDIR(st.st_mode))
376 ; /* somebody created it since we checked */
377 else if (errno == ENOENT)
379 * Either mkdir() failed because
380 * somebody just pruned the containing
381 * directory, or stat() failed because
382 * the file that was in our way was
383 * just removed. Either way, inform
384 * the caller that it might be worth
385 * trying again:
387 ret = SCLD_VANISHED;
388 else
389 ret = SCLD_FAILED;
390 } else if (share && adjust_shared_perm(path)) {
391 ret = SCLD_PERMS;
393 *slash = slash_character;
395 return ret;
398 enum scld_error safe_create_leading_directories(char *path)
400 return safe_create_leading_directories_1(path, 1);
403 enum scld_error safe_create_leading_directories_no_share(char *path)
405 return safe_create_leading_directories_1(path, 0);
408 enum scld_error safe_create_leading_directories_const(const char *path)
410 int save_errno;
411 /* path points to cache entries, so xstrdup before messing with it */
412 char *buf = xstrdup(path);
413 enum scld_error result = safe_create_leading_directories(buf);
415 save_errno = errno;
416 free(buf);
417 errno = save_errno;
418 return result;
421 static void fill_loose_path(struct strbuf *buf, const struct object_id *oid)
423 int i;
424 for (i = 0; i < the_hash_algo->rawsz; i++) {
425 static char hex[] = "0123456789abcdef";
426 unsigned int val = oid->hash[i];
427 strbuf_addch(buf, hex[val >> 4]);
428 strbuf_addch(buf, hex[val & 0xf]);
429 if (!i)
430 strbuf_addch(buf, '/');
434 static const char *odb_loose_path(struct object_directory *odb,
435 struct strbuf *buf,
436 const struct object_id *oid)
438 strbuf_reset(buf);
439 strbuf_addstr(buf, odb->path);
440 strbuf_addch(buf, '/');
441 fill_loose_path(buf, oid);
442 return buf->buf;
445 const char *loose_object_path(struct repository *r, struct strbuf *buf,
446 const struct object_id *oid)
448 return odb_loose_path(r->objects->odb, buf, oid);
452 * Return non-zero iff the path is usable as an alternate object database.
454 static int alt_odb_usable(struct raw_object_store *o,
455 struct strbuf *path,
456 const char *normalized_objdir, khiter_t *pos)
458 int r;
460 /* Detect cases where alternate disappeared */
461 if (!is_directory(path->buf)) {
462 error(_("object directory %s does not exist; "
463 "check .git/objects/info/alternates"),
464 path->buf);
465 return 0;
469 * Prevent the common mistake of listing the same
470 * thing twice, or object directory itself.
472 if (!o->odb_by_path) {
473 khiter_t p;
475 o->odb_by_path = kh_init_odb_path_map();
476 assert(!o->odb->next);
477 p = kh_put_odb_path_map(o->odb_by_path, o->odb->path, &r);
478 assert(r == 1); /* never used */
479 kh_value(o->odb_by_path, p) = o->odb;
481 if (fspatheq(path->buf, normalized_objdir))
482 return 0;
483 *pos = kh_put_odb_path_map(o->odb_by_path, path->buf, &r);
484 /* r: 0 = exists, 1 = never used, 2 = deleted */
485 return r == 0 ? 0 : 1;
489 * Prepare alternate object database registry.
491 * The variable alt_odb_list points at the list of struct
492 * object_directory. The elements on this list come from
493 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
494 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
495 * whose contents is similar to that environment variable but can be
496 * LF separated. Its base points at a statically allocated buffer that
497 * contains "/the/directory/corresponding/to/.git/objects/...", while
498 * its name points just after the slash at the end of ".git/objects/"
499 * in the example above, and has enough space to hold all hex characters
500 * of the object ID, an extra slash for the first level indirection, and
501 * the terminating NUL.
503 static void read_info_alternates(struct repository *r,
504 const char *relative_base,
505 int depth);
506 static int link_alt_odb_entry(struct repository *r, const struct strbuf *entry,
507 const char *relative_base, int depth, const char *normalized_objdir)
509 struct object_directory *ent;
510 struct strbuf pathbuf = STRBUF_INIT;
511 struct strbuf tmp = STRBUF_INIT;
512 khiter_t pos;
513 int ret = -1;
515 if (!is_absolute_path(entry->buf) && relative_base) {
516 strbuf_realpath(&pathbuf, relative_base, 1);
517 strbuf_addch(&pathbuf, '/');
519 strbuf_addbuf(&pathbuf, entry);
521 if (!strbuf_realpath(&tmp, pathbuf.buf, 0)) {
522 error(_("unable to normalize alternate object path: %s"),
523 pathbuf.buf);
524 goto error;
526 strbuf_swap(&pathbuf, &tmp);
529 * The trailing slash after the directory name is given by
530 * this function at the end. Remove duplicates.
532 while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
533 strbuf_setlen(&pathbuf, pathbuf.len - 1);
535 if (!alt_odb_usable(r->objects, &pathbuf, normalized_objdir, &pos))
536 goto error;
538 CALLOC_ARRAY(ent, 1);
539 /* pathbuf.buf is already in r->objects->odb_by_path */
540 ent->path = strbuf_detach(&pathbuf, NULL);
542 /* add the alternate entry */
543 *r->objects->odb_tail = ent;
544 r->objects->odb_tail = &(ent->next);
545 ent->next = NULL;
546 assert(r->objects->odb_by_path);
547 kh_value(r->objects->odb_by_path, pos) = ent;
549 /* recursively add alternates */
550 read_info_alternates(r, ent->path, depth + 1);
551 ret = 0;
552 error:
553 strbuf_release(&tmp);
554 strbuf_release(&pathbuf);
555 return ret;
558 static const char *parse_alt_odb_entry(const char *string,
559 int sep,
560 struct strbuf *out)
562 const char *end;
564 strbuf_reset(out);
566 if (*string == '#') {
567 /* comment; consume up to next separator */
568 end = strchrnul(string, sep);
569 } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
571 * quoted path; unquote_c_style has copied the
572 * data for us and set "end". Broken quoting (e.g.,
573 * an entry that doesn't end with a quote) falls
574 * back to the unquoted case below.
576 } else {
577 /* normal, unquoted path */
578 end = strchrnul(string, sep);
579 strbuf_add(out, string, end - string);
582 if (*end)
583 end++;
584 return end;
587 static void link_alt_odb_entries(struct repository *r, const char *alt,
588 int sep, const char *relative_base, int depth)
590 struct strbuf objdirbuf = STRBUF_INIT;
591 struct strbuf entry = STRBUF_INIT;
593 if (!alt || !*alt)
594 return;
596 if (depth > 5) {
597 error(_("%s: ignoring alternate object stores, nesting too deep"),
598 relative_base);
599 return;
602 strbuf_realpath(&objdirbuf, r->objects->odb->path, 1);
604 while (*alt) {
605 alt = parse_alt_odb_entry(alt, sep, &entry);
606 if (!entry.len)
607 continue;
608 link_alt_odb_entry(r, &entry,
609 relative_base, depth, objdirbuf.buf);
611 strbuf_release(&entry);
612 strbuf_release(&objdirbuf);
615 static void read_info_alternates(struct repository *r,
616 const char *relative_base,
617 int depth)
619 char *path;
620 struct strbuf buf = STRBUF_INIT;
622 path = xstrfmt("%s/info/alternates", relative_base);
623 if (strbuf_read_file(&buf, path, 1024) < 0) {
624 warn_on_fopen_errors(path);
625 free(path);
626 return;
629 link_alt_odb_entries(r, buf.buf, '\n', relative_base, depth);
630 strbuf_release(&buf);
631 free(path);
634 void add_to_alternates_file(const char *reference)
636 struct lock_file lock = LOCK_INIT;
637 char *alts = git_pathdup("objects/info/alternates");
638 FILE *in, *out;
639 int found = 0;
641 hold_lock_file_for_update(&lock, alts, LOCK_DIE_ON_ERROR);
642 out = fdopen_lock_file(&lock, "w");
643 if (!out)
644 die_errno(_("unable to fdopen alternates lockfile"));
646 in = fopen(alts, "r");
647 if (in) {
648 struct strbuf line = STRBUF_INIT;
650 while (strbuf_getline(&line, in) != EOF) {
651 if (!strcmp(reference, line.buf)) {
652 found = 1;
653 break;
655 fprintf_or_die(out, "%s\n", line.buf);
658 strbuf_release(&line);
659 fclose(in);
661 else if (errno != ENOENT)
662 die_errno(_("unable to read alternates file"));
664 if (found) {
665 rollback_lock_file(&lock);
666 } else {
667 fprintf_or_die(out, "%s\n", reference);
668 if (commit_lock_file(&lock))
669 die_errno(_("unable to move new alternates file into place"));
670 if (the_repository->objects->loaded_alternates)
671 link_alt_odb_entries(the_repository, reference,
672 '\n', NULL, 0);
674 free(alts);
677 void add_to_alternates_memory(const char *reference)
680 * Make sure alternates are initialized, or else our entry may be
681 * overwritten when they are.
683 prepare_alt_odb(the_repository);
685 link_alt_odb_entries(the_repository, reference,
686 '\n', NULL, 0);
689 struct object_directory *set_temporary_primary_odb(const char *dir, int will_destroy)
691 struct object_directory *new_odb;
694 * Make sure alternates are initialized, or else our entry may be
695 * overwritten when they are.
697 prepare_alt_odb(the_repository);
700 * Make a new primary odb and link the old primary ODB in as an
701 * alternate
703 new_odb = xcalloc(1, sizeof(*new_odb));
704 new_odb->path = xstrdup(dir);
707 * Disable ref updates while a temporary odb is active, since
708 * the objects in the database may roll back.
710 new_odb->disable_ref_updates = 1;
711 new_odb->will_destroy = will_destroy;
712 new_odb->next = the_repository->objects->odb;
713 the_repository->objects->odb = new_odb;
714 return new_odb->next;
717 void restore_primary_odb(struct object_directory *restore_odb, const char *old_path)
719 struct object_directory *cur_odb = the_repository->objects->odb;
721 if (strcmp(old_path, cur_odb->path))
722 BUG("expected %s as primary object store; found %s",
723 old_path, cur_odb->path);
725 if (cur_odb->next != restore_odb)
726 BUG("we expect the old primary object store to be the first alternate");
728 the_repository->objects->odb = restore_odb;
729 free_object_directory(cur_odb);
733 * Compute the exact path an alternate is at and returns it. In case of
734 * error NULL is returned and the human readable error is added to `err`
735 * `path` may be relative and should point to $GIT_DIR.
736 * `err` must not be null.
738 char *compute_alternate_path(const char *path, struct strbuf *err)
740 char *ref_git = NULL;
741 const char *repo;
742 int seen_error = 0;
744 ref_git = real_pathdup(path, 0);
745 if (!ref_git) {
746 seen_error = 1;
747 strbuf_addf(err, _("path '%s' does not exist"), path);
748 goto out;
751 repo = read_gitfile(ref_git);
752 if (!repo)
753 repo = read_gitfile(mkpath("%s/.git", ref_git));
754 if (repo) {
755 free(ref_git);
756 ref_git = xstrdup(repo);
759 if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
760 char *ref_git_git = mkpathdup("%s/.git", ref_git);
761 free(ref_git);
762 ref_git = ref_git_git;
763 } else if (!is_directory(mkpath("%s/objects", ref_git))) {
764 struct strbuf sb = STRBUF_INIT;
765 seen_error = 1;
766 if (get_common_dir(&sb, ref_git)) {
767 strbuf_addf(err,
768 _("reference repository '%s' as a linked "
769 "checkout is not supported yet."),
770 path);
771 goto out;
774 strbuf_addf(err, _("reference repository '%s' is not a "
775 "local repository."), path);
776 goto out;
779 if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
780 strbuf_addf(err, _("reference repository '%s' is shallow"),
781 path);
782 seen_error = 1;
783 goto out;
786 if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
787 strbuf_addf(err,
788 _("reference repository '%s' is grafted"),
789 path);
790 seen_error = 1;
791 goto out;
794 out:
795 if (seen_error) {
796 FREE_AND_NULL(ref_git);
799 return ref_git;
802 struct object_directory *find_odb(struct repository *r, const char *obj_dir)
804 struct object_directory *odb;
805 char *obj_dir_real = real_pathdup(obj_dir, 1);
806 struct strbuf odb_path_real = STRBUF_INIT;
808 prepare_alt_odb(r);
809 for (odb = r->objects->odb; odb; odb = odb->next) {
810 strbuf_realpath(&odb_path_real, odb->path, 1);
811 if (!strcmp(obj_dir_real, odb_path_real.buf))
812 break;
815 free(obj_dir_real);
816 strbuf_release(&odb_path_real);
818 if (!odb)
819 die(_("could not find object directory matching %s"), obj_dir);
820 return odb;
823 static void fill_alternate_refs_command(struct child_process *cmd,
824 const char *repo_path)
826 const char *value;
828 if (!git_config_get_value("core.alternateRefsCommand", &value)) {
829 cmd->use_shell = 1;
831 strvec_push(&cmd->args, value);
832 strvec_push(&cmd->args, repo_path);
833 } else {
834 cmd->git_cmd = 1;
836 strvec_pushf(&cmd->args, "--git-dir=%s", repo_path);
837 strvec_push(&cmd->args, "for-each-ref");
838 strvec_push(&cmd->args, "--format=%(objectname)");
840 if (!git_config_get_value("core.alternateRefsPrefixes", &value)) {
841 strvec_push(&cmd->args, "--");
842 strvec_split(&cmd->args, value);
846 strvec_pushv(&cmd->env, (const char **)local_repo_env);
847 cmd->out = -1;
850 static void read_alternate_refs(const char *path,
851 alternate_ref_fn *cb,
852 void *data)
854 struct child_process cmd = CHILD_PROCESS_INIT;
855 struct strbuf line = STRBUF_INIT;
856 FILE *fh;
858 fill_alternate_refs_command(&cmd, path);
860 if (start_command(&cmd))
861 return;
863 fh = xfdopen(cmd.out, "r");
864 while (strbuf_getline_lf(&line, fh) != EOF) {
865 struct object_id oid;
866 const char *p;
868 if (parse_oid_hex(line.buf, &oid, &p) || *p) {
869 warning(_("invalid line while parsing alternate refs: %s"),
870 line.buf);
871 break;
874 cb(&oid, data);
877 fclose(fh);
878 finish_command(&cmd);
879 strbuf_release(&line);
882 struct alternate_refs_data {
883 alternate_ref_fn *fn;
884 void *data;
887 static int refs_from_alternate_cb(struct object_directory *e,
888 void *data)
890 struct strbuf path = STRBUF_INIT;
891 size_t base_len;
892 struct alternate_refs_data *cb = data;
894 if (!strbuf_realpath(&path, e->path, 0))
895 goto out;
896 if (!strbuf_strip_suffix(&path, "/objects"))
897 goto out;
898 base_len = path.len;
900 /* Is this a git repository with refs? */
901 strbuf_addstr(&path, "/refs");
902 if (!is_directory(path.buf))
903 goto out;
904 strbuf_setlen(&path, base_len);
906 read_alternate_refs(path.buf, cb->fn, cb->data);
908 out:
909 strbuf_release(&path);
910 return 0;
913 void for_each_alternate_ref(alternate_ref_fn fn, void *data)
915 struct alternate_refs_data cb;
916 cb.fn = fn;
917 cb.data = data;
918 foreach_alt_odb(refs_from_alternate_cb, &cb);
921 int foreach_alt_odb(alt_odb_fn fn, void *cb)
923 struct object_directory *ent;
924 int r = 0;
926 prepare_alt_odb(the_repository);
927 for (ent = the_repository->objects->odb->next; ent; ent = ent->next) {
928 r = fn(ent, cb);
929 if (r)
930 break;
932 return r;
935 void prepare_alt_odb(struct repository *r)
937 if (r->objects->loaded_alternates)
938 return;
940 link_alt_odb_entries(r, r->objects->alternate_db, PATH_SEP, NULL, 0);
942 read_info_alternates(r, r->objects->odb->path, 0);
943 r->objects->loaded_alternates = 1;
946 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
947 static int freshen_file(const char *fn)
949 return !utime(fn, NULL);
953 * All of the check_and_freshen functions return 1 if the file exists and was
954 * freshened (if freshening was requested), 0 otherwise. If they return
955 * 0, you should not assume that it is safe to skip a write of the object (it
956 * either does not exist on disk, or has a stale mtime and may be subject to
957 * pruning).
959 int check_and_freshen_file(const char *fn, int freshen)
961 if (access(fn, F_OK))
962 return 0;
963 if (freshen && !freshen_file(fn))
964 return 0;
965 return 1;
968 static int check_and_freshen_odb(struct object_directory *odb,
969 const struct object_id *oid,
970 int freshen)
972 static struct strbuf path = STRBUF_INIT;
973 odb_loose_path(odb, &path, oid);
974 return check_and_freshen_file(path.buf, freshen);
977 static int check_and_freshen_local(const struct object_id *oid, int freshen)
979 return check_and_freshen_odb(the_repository->objects->odb, oid, freshen);
982 static int check_and_freshen_nonlocal(const struct object_id *oid, int freshen)
984 struct object_directory *odb;
986 prepare_alt_odb(the_repository);
987 for (odb = the_repository->objects->odb->next; odb; odb = odb->next) {
988 if (check_and_freshen_odb(odb, oid, freshen))
989 return 1;
991 return 0;
994 static int check_and_freshen(const struct object_id *oid, int freshen)
996 return check_and_freshen_local(oid, freshen) ||
997 check_and_freshen_nonlocal(oid, freshen);
1000 int has_loose_object_nonlocal(const struct object_id *oid)
1002 return check_and_freshen_nonlocal(oid, 0);
1005 int has_loose_object(const struct object_id *oid)
1007 return check_and_freshen(oid, 0);
1010 static void mmap_limit_check(size_t length)
1012 static size_t limit = 0;
1013 if (!limit) {
1014 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
1015 if (!limit)
1016 limit = SIZE_MAX;
1018 if (length > limit)
1019 die(_("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX),
1020 (uintmax_t)length, (uintmax_t)limit);
1023 void *xmmap_gently(void *start, size_t length,
1024 int prot, int flags, int fd, off_t offset)
1026 void *ret;
1028 mmap_limit_check(length);
1029 ret = mmap(start, length, prot, flags, fd, offset);
1030 if (ret == MAP_FAILED && !length)
1031 ret = NULL;
1032 return ret;
1035 const char *mmap_os_err(void)
1037 static const char blank[] = "";
1038 #if defined(__linux__)
1039 if (errno == ENOMEM) {
1040 /* this continues an existing error message: */
1041 static const char enomem[] =
1042 ", check sys.vm.max_map_count and/or RLIMIT_DATA";
1043 return enomem;
1045 #endif /* OS-specific bits */
1046 return blank;
1049 void *xmmap(void *start, size_t length,
1050 int prot, int flags, int fd, off_t offset)
1052 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
1053 if (ret == MAP_FAILED)
1054 die_errno(_("mmap failed%s"), mmap_os_err());
1055 return ret;
1058 static int format_object_header_literally(char *str, size_t size,
1059 const char *type, size_t objsize)
1061 return xsnprintf(str, size, "%s %"PRIuMAX, type, (uintmax_t)objsize) + 1;
1064 int format_object_header(char *str, size_t size, enum object_type type,
1065 size_t objsize)
1067 const char *name = type_name(type);
1069 if (!name)
1070 BUG("could not get a type name for 'enum object_type' value %d", type);
1072 return format_object_header_literally(str, size, name, objsize);
1075 int check_object_signature(struct repository *r, const struct object_id *oid,
1076 void *buf, unsigned long size,
1077 enum object_type type)
1079 struct object_id real_oid;
1081 hash_object_file(r->hash_algo, buf, size, type, &real_oid);
1083 return !oideq(oid, &real_oid) ? -1 : 0;
1086 int stream_object_signature(struct repository *r, const struct object_id *oid)
1088 struct object_id real_oid;
1089 unsigned long size;
1090 enum object_type obj_type;
1091 struct git_istream *st;
1092 git_hash_ctx c;
1093 char hdr[MAX_HEADER_LEN];
1094 int hdrlen;
1096 st = open_istream(r, oid, &obj_type, &size, NULL);
1097 if (!st)
1098 return -1;
1100 /* Generate the header */
1101 hdrlen = format_object_header(hdr, sizeof(hdr), obj_type, size);
1103 /* Sha1.. */
1104 r->hash_algo->init_fn(&c);
1105 r->hash_algo->update_fn(&c, hdr, hdrlen);
1106 for (;;) {
1107 char buf[1024 * 16];
1108 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1110 if (readlen < 0) {
1111 close_istream(st);
1112 return -1;
1114 if (!readlen)
1115 break;
1116 r->hash_algo->update_fn(&c, buf, readlen);
1118 r->hash_algo->final_oid_fn(&real_oid, &c);
1119 close_istream(st);
1120 return !oideq(oid, &real_oid) ? -1 : 0;
1123 int git_open_cloexec(const char *name, int flags)
1125 int fd;
1126 static int o_cloexec = O_CLOEXEC;
1128 fd = open(name, flags | o_cloexec);
1129 if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
1130 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1131 o_cloexec &= ~O_CLOEXEC;
1132 fd = open(name, flags | o_cloexec);
1135 #if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
1137 static int fd_cloexec = FD_CLOEXEC;
1139 if (!o_cloexec && 0 <= fd && fd_cloexec) {
1140 /* Opened w/o O_CLOEXEC? try with fcntl(2) to add it */
1141 int flags = fcntl(fd, F_GETFD);
1142 if (fcntl(fd, F_SETFD, flags | fd_cloexec))
1143 fd_cloexec = 0;
1146 #endif
1147 return fd;
1151 * Find "oid" as a loose object in the local repository or in an alternate.
1152 * Returns 0 on success, negative on failure.
1154 * The "path" out-parameter will give the path of the object we found (if any).
1155 * Note that it may point to static storage and is only valid until another
1156 * call to stat_loose_object().
1158 static int stat_loose_object(struct repository *r, const struct object_id *oid,
1159 struct stat *st, const char **path)
1161 struct object_directory *odb;
1162 static struct strbuf buf = STRBUF_INIT;
1164 prepare_alt_odb(r);
1165 for (odb = r->objects->odb; odb; odb = odb->next) {
1166 *path = odb_loose_path(odb, &buf, oid);
1167 if (!lstat(*path, st))
1168 return 0;
1171 return -1;
1175 * Like stat_loose_object(), but actually open the object and return the
1176 * descriptor. See the caveats on the "path" parameter above.
1178 static int open_loose_object(struct repository *r,
1179 const struct object_id *oid, const char **path)
1181 int fd;
1182 struct object_directory *odb;
1183 int most_interesting_errno = ENOENT;
1184 static struct strbuf buf = STRBUF_INIT;
1186 prepare_alt_odb(r);
1187 for (odb = r->objects->odb; odb; odb = odb->next) {
1188 *path = odb_loose_path(odb, &buf, oid);
1189 fd = git_open(*path);
1190 if (fd >= 0)
1191 return fd;
1193 if (most_interesting_errno == ENOENT)
1194 most_interesting_errno = errno;
1196 errno = most_interesting_errno;
1197 return -1;
1200 static int quick_has_loose(struct repository *r,
1201 const struct object_id *oid)
1203 struct object_directory *odb;
1205 prepare_alt_odb(r);
1206 for (odb = r->objects->odb; odb; odb = odb->next) {
1207 if (oidtree_contains(odb_loose_cache(odb, oid), oid))
1208 return 1;
1210 return 0;
1214 * Map the loose object at "path" if it is not NULL, or the path found by
1215 * searching for a loose object named "oid".
1217 static void *map_loose_object_1(struct repository *r, const char *path,
1218 const struct object_id *oid, unsigned long *size)
1220 void *map;
1221 int fd;
1223 if (path)
1224 fd = git_open(path);
1225 else
1226 fd = open_loose_object(r, oid, &path);
1227 map = NULL;
1228 if (fd >= 0) {
1229 struct stat st;
1231 if (!fstat(fd, &st)) {
1232 *size = xsize_t(st.st_size);
1233 if (!*size) {
1234 /* mmap() is forbidden on empty files */
1235 error(_("object file %s is empty"), path);
1236 close(fd);
1237 return NULL;
1239 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1241 close(fd);
1243 return map;
1246 void *map_loose_object(struct repository *r,
1247 const struct object_id *oid,
1248 unsigned long *size)
1250 return map_loose_object_1(r, NULL, oid, size);
1253 enum unpack_loose_header_result unpack_loose_header(git_zstream *stream,
1254 unsigned char *map,
1255 unsigned long mapsize,
1256 void *buffer,
1257 unsigned long bufsiz,
1258 struct strbuf *header)
1260 int status;
1262 /* Get the data stream */
1263 memset(stream, 0, sizeof(*stream));
1264 stream->next_in = map;
1265 stream->avail_in = mapsize;
1266 stream->next_out = buffer;
1267 stream->avail_out = bufsiz;
1269 git_inflate_init(stream);
1270 obj_read_unlock();
1271 status = git_inflate(stream, 0);
1272 obj_read_lock();
1273 if (status < Z_OK)
1274 return ULHR_BAD;
1277 * Check if entire header is unpacked in the first iteration.
1279 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1280 return ULHR_OK;
1283 * We have a header longer than MAX_HEADER_LEN. The "header"
1284 * here is only non-NULL when we run "cat-file
1285 * --allow-unknown-type".
1287 if (!header)
1288 return ULHR_TOO_LONG;
1291 * buffer[0..bufsiz] was not large enough. Copy the partial
1292 * result out to header, and then append the result of further
1293 * reading the stream.
1295 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1296 stream->next_out = buffer;
1297 stream->avail_out = bufsiz;
1299 do {
1300 obj_read_unlock();
1301 status = git_inflate(stream, 0);
1302 obj_read_lock();
1303 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1304 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1305 return 0;
1306 stream->next_out = buffer;
1307 stream->avail_out = bufsiz;
1308 } while (status != Z_STREAM_END);
1309 return ULHR_TOO_LONG;
1312 static void *unpack_loose_rest(git_zstream *stream,
1313 void *buffer, unsigned long size,
1314 const struct object_id *oid)
1316 int bytes = strlen(buffer) + 1;
1317 unsigned char *buf = xmallocz(size);
1318 unsigned long n;
1319 int status = Z_OK;
1321 n = stream->total_out - bytes;
1322 if (n > size)
1323 n = size;
1324 memcpy(buf, (char *) buffer + bytes, n);
1325 bytes = n;
1326 if (bytes <= size) {
1328 * The above condition must be (bytes <= size), not
1329 * (bytes < size). In other words, even though we
1330 * expect no more output and set avail_out to zero,
1331 * the input zlib stream may have bytes that express
1332 * "this concludes the stream", and we *do* want to
1333 * eat that input.
1335 * Otherwise we would not be able to test that we
1336 * consumed all the input to reach the expected size;
1337 * we also want to check that zlib tells us that all
1338 * went well with status == Z_STREAM_END at the end.
1340 stream->next_out = buf + bytes;
1341 stream->avail_out = size - bytes;
1342 while (status == Z_OK) {
1343 obj_read_unlock();
1344 status = git_inflate(stream, Z_FINISH);
1345 obj_read_lock();
1348 if (status == Z_STREAM_END && !stream->avail_in) {
1349 git_inflate_end(stream);
1350 return buf;
1353 if (status < 0)
1354 error(_("corrupt loose object '%s'"), oid_to_hex(oid));
1355 else if (stream->avail_in)
1356 error(_("garbage at end of loose object '%s'"),
1357 oid_to_hex(oid));
1358 free(buf);
1359 return NULL;
1363 * We used to just use "sscanf()", but that's actually way
1364 * too permissive for what we want to check. So do an anal
1365 * object header parse by hand.
1367 int parse_loose_header(const char *hdr, struct object_info *oi)
1369 const char *type_buf = hdr;
1370 size_t size;
1371 int type, type_len = 0;
1374 * The type can be of any size but is followed by
1375 * a space.
1377 for (;;) {
1378 char c = *hdr++;
1379 if (!c)
1380 return -1;
1381 if (c == ' ')
1382 break;
1383 type_len++;
1386 type = type_from_string_gently(type_buf, type_len, 1);
1387 if (oi->type_name)
1388 strbuf_add(oi->type_name, type_buf, type_len);
1389 if (oi->typep)
1390 *oi->typep = type;
1393 * The length must follow immediately, and be in canonical
1394 * decimal format (ie "010" is not valid).
1396 size = *hdr++ - '0';
1397 if (size > 9)
1398 return -1;
1399 if (size) {
1400 for (;;) {
1401 unsigned long c = *hdr - '0';
1402 if (c > 9)
1403 break;
1404 hdr++;
1405 size = st_add(st_mult(size, 10), c);
1409 if (oi->sizep)
1410 *oi->sizep = cast_size_t_to_ulong(size);
1413 * The length must be followed by a zero byte
1415 if (*hdr)
1416 return -1;
1419 * The format is valid, but the type may still be bogus. The
1420 * Caller needs to check its oi->typep.
1422 return 0;
1425 static int loose_object_info(struct repository *r,
1426 const struct object_id *oid,
1427 struct object_info *oi, int flags)
1429 int status = 0;
1430 unsigned long mapsize;
1431 void *map;
1432 git_zstream stream;
1433 char hdr[MAX_HEADER_LEN];
1434 struct strbuf hdrbuf = STRBUF_INIT;
1435 unsigned long size_scratch;
1436 enum object_type type_scratch;
1437 int allow_unknown = flags & OBJECT_INFO_ALLOW_UNKNOWN_TYPE;
1439 if (oi->delta_base_oid)
1440 oidclr(oi->delta_base_oid);
1443 * If we don't care about type or size, then we don't
1444 * need to look inside the object at all. Note that we
1445 * do not optimize out the stat call, even if the
1446 * caller doesn't care about the disk-size, since our
1447 * return value implicitly indicates whether the
1448 * object even exists.
1450 if (!oi->typep && !oi->type_name && !oi->sizep && !oi->contentp) {
1451 const char *path;
1452 struct stat st;
1453 if (!oi->disk_sizep && (flags & OBJECT_INFO_QUICK))
1454 return quick_has_loose(r, oid) ? 0 : -1;
1455 if (stat_loose_object(r, oid, &st, &path) < 0)
1456 return -1;
1457 if (oi->disk_sizep)
1458 *oi->disk_sizep = st.st_size;
1459 return 0;
1462 map = map_loose_object(r, oid, &mapsize);
1463 if (!map)
1464 return -1;
1466 if (!oi->sizep)
1467 oi->sizep = &size_scratch;
1468 if (!oi->typep)
1469 oi->typep = &type_scratch;
1471 if (oi->disk_sizep)
1472 *oi->disk_sizep = mapsize;
1474 switch (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
1475 allow_unknown ? &hdrbuf : NULL)) {
1476 case ULHR_OK:
1477 if (parse_loose_header(hdrbuf.len ? hdrbuf.buf : hdr, oi) < 0)
1478 status = error(_("unable to parse %s header"), oid_to_hex(oid));
1479 else if (!allow_unknown && *oi->typep < 0)
1480 die(_("invalid object type"));
1482 if (!oi->contentp)
1483 break;
1484 *oi->contentp = unpack_loose_rest(&stream, hdr, *oi->sizep, oid);
1485 if (*oi->contentp)
1486 goto cleanup;
1488 status = -1;
1489 break;
1490 case ULHR_BAD:
1491 status = error(_("unable to unpack %s header"),
1492 oid_to_hex(oid));
1493 break;
1494 case ULHR_TOO_LONG:
1495 status = error(_("header for %s too long, exceeds %d bytes"),
1496 oid_to_hex(oid), MAX_HEADER_LEN);
1497 break;
1500 git_inflate_end(&stream);
1501 cleanup:
1502 munmap(map, mapsize);
1503 if (oi->sizep == &size_scratch)
1504 oi->sizep = NULL;
1505 strbuf_release(&hdrbuf);
1506 if (oi->typep == &type_scratch)
1507 oi->typep = NULL;
1508 oi->whence = OI_LOOSE;
1509 return status;
1512 int obj_read_use_lock = 0;
1513 pthread_mutex_t obj_read_mutex;
1515 void enable_obj_read_lock(void)
1517 if (obj_read_use_lock)
1518 return;
1520 obj_read_use_lock = 1;
1521 init_recursive_mutex(&obj_read_mutex);
1524 void disable_obj_read_lock(void)
1526 if (!obj_read_use_lock)
1527 return;
1529 obj_read_use_lock = 0;
1530 pthread_mutex_destroy(&obj_read_mutex);
1533 int fetch_if_missing = 1;
1535 static int do_oid_object_info_extended(struct repository *r,
1536 const struct object_id *oid,
1537 struct object_info *oi, unsigned flags)
1539 static struct object_info blank_oi = OBJECT_INFO_INIT;
1540 struct cached_object *co;
1541 struct pack_entry e;
1542 int rtype;
1543 const struct object_id *real = oid;
1544 int already_retried = 0;
1547 if (flags & OBJECT_INFO_LOOKUP_REPLACE)
1548 real = lookup_replace_object(r, oid);
1550 if (is_null_oid(real))
1551 return -1;
1553 if (!oi)
1554 oi = &blank_oi;
1556 co = find_cached_object(real);
1557 if (co) {
1558 if (oi->typep)
1559 *(oi->typep) = co->type;
1560 if (oi->sizep)
1561 *(oi->sizep) = co->size;
1562 if (oi->disk_sizep)
1563 *(oi->disk_sizep) = 0;
1564 if (oi->delta_base_oid)
1565 oidclr(oi->delta_base_oid);
1566 if (oi->type_name)
1567 strbuf_addstr(oi->type_name, type_name(co->type));
1568 if (oi->contentp)
1569 *oi->contentp = xmemdupz(co->buf, co->size);
1570 oi->whence = OI_CACHED;
1571 return 0;
1574 while (1) {
1575 if (find_pack_entry(r, real, &e))
1576 break;
1578 if (flags & OBJECT_INFO_IGNORE_LOOSE)
1579 return -1;
1581 /* Most likely it's a loose object. */
1582 if (!loose_object_info(r, real, oi, flags))
1583 return 0;
1585 /* Not a loose object; someone else may have just packed it. */
1586 if (!(flags & OBJECT_INFO_QUICK)) {
1587 reprepare_packed_git(r);
1588 if (find_pack_entry(r, real, &e))
1589 break;
1593 * If r is the_repository, this might be an attempt at
1594 * accessing a submodule object as if it were in the_repository
1595 * (having called add_submodule_odb() on that submodule's ODB).
1596 * If any such ODBs exist, register them and try again.
1598 if (r == the_repository &&
1599 register_all_submodule_odb_as_alternates())
1600 /* We added some alternates; retry */
1601 continue;
1603 /* Check if it is a missing object */
1604 if (fetch_if_missing && repo_has_promisor_remote(r) &&
1605 !already_retried &&
1606 !(flags & OBJECT_INFO_SKIP_FETCH_OBJECT)) {
1607 promisor_remote_get_direct(r, real, 1);
1608 already_retried = 1;
1609 continue;
1612 return -1;
1615 if (oi == &blank_oi)
1617 * We know that the caller doesn't actually need the
1618 * information below, so return early.
1620 return 0;
1621 rtype = packed_object_info(r, e.p, e.offset, oi);
1622 if (rtype < 0) {
1623 mark_bad_packed_object(e.p, real);
1624 return do_oid_object_info_extended(r, real, oi, 0);
1625 } else if (oi->whence == OI_PACKED) {
1626 oi->u.packed.offset = e.offset;
1627 oi->u.packed.pack = e.p;
1628 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
1629 rtype == OBJ_OFS_DELTA);
1632 return 0;
1635 int oid_object_info_extended(struct repository *r, const struct object_id *oid,
1636 struct object_info *oi, unsigned flags)
1638 int ret;
1639 obj_read_lock();
1640 ret = do_oid_object_info_extended(r, oid, oi, flags);
1641 obj_read_unlock();
1642 return ret;
1646 /* returns enum object_type or negative */
1647 int oid_object_info(struct repository *r,
1648 const struct object_id *oid,
1649 unsigned long *sizep)
1651 enum object_type type;
1652 struct object_info oi = OBJECT_INFO_INIT;
1654 oi.typep = &type;
1655 oi.sizep = sizep;
1656 if (oid_object_info_extended(r, oid, &oi,
1657 OBJECT_INFO_LOOKUP_REPLACE) < 0)
1658 return -1;
1659 return type;
1662 static void *read_object(struct repository *r,
1663 const struct object_id *oid, enum object_type *type,
1664 unsigned long *size)
1666 struct object_info oi = OBJECT_INFO_INIT;
1667 void *content;
1668 oi.typep = type;
1669 oi.sizep = size;
1670 oi.contentp = &content;
1672 if (oid_object_info_extended(r, oid, &oi, 0) < 0)
1673 return NULL;
1674 return content;
1677 int pretend_object_file(void *buf, unsigned long len, enum object_type type,
1678 struct object_id *oid)
1680 struct cached_object *co;
1682 hash_object_file(the_hash_algo, buf, len, type, oid);
1683 if (has_object_file_with_flags(oid, OBJECT_INFO_QUICK | OBJECT_INFO_SKIP_FETCH_OBJECT) ||
1684 find_cached_object(oid))
1685 return 0;
1686 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
1687 co = &cached_objects[cached_object_nr++];
1688 co->size = len;
1689 co->type = type;
1690 co->buf = xmalloc(len);
1691 memcpy(co->buf, buf, len);
1692 oidcpy(&co->oid, oid);
1693 return 0;
1697 * This function dies on corrupt objects; the callers who want to
1698 * deal with them should arrange to call read_object() and give error
1699 * messages themselves.
1701 void *read_object_file_extended(struct repository *r,
1702 const struct object_id *oid,
1703 enum object_type *type,
1704 unsigned long *size,
1705 int lookup_replace)
1707 void *data;
1708 const struct packed_git *p;
1709 const char *path;
1710 struct stat st;
1711 const struct object_id *repl = lookup_replace ?
1712 lookup_replace_object(r, oid) : oid;
1714 errno = 0;
1715 data = read_object(r, repl, type, size);
1716 if (data)
1717 return data;
1719 obj_read_lock();
1720 if (errno && errno != ENOENT)
1721 die_errno(_("failed to read object %s"), oid_to_hex(oid));
1723 /* die if we replaced an object with one that does not exist */
1724 if (repl != oid)
1725 die(_("replacement %s not found for %s"),
1726 oid_to_hex(repl), oid_to_hex(oid));
1728 if (!stat_loose_object(r, repl, &st, &path))
1729 die(_("loose object %s (stored in %s) is corrupt"),
1730 oid_to_hex(repl), path);
1732 if ((p = has_packed_and_bad(r, repl)))
1733 die(_("packed object %s (stored in %s) is corrupt"),
1734 oid_to_hex(repl), p->pack_name);
1735 obj_read_unlock();
1737 return NULL;
1740 void *read_object_with_reference(struct repository *r,
1741 const struct object_id *oid,
1742 enum object_type required_type,
1743 unsigned long *size,
1744 struct object_id *actual_oid_return)
1746 enum object_type type;
1747 void *buffer;
1748 unsigned long isize;
1749 struct object_id actual_oid;
1751 oidcpy(&actual_oid, oid);
1752 while (1) {
1753 int ref_length = -1;
1754 const char *ref_type = NULL;
1756 buffer = repo_read_object_file(r, &actual_oid, &type, &isize);
1757 if (!buffer)
1758 return NULL;
1759 if (type == required_type) {
1760 *size = isize;
1761 if (actual_oid_return)
1762 oidcpy(actual_oid_return, &actual_oid);
1763 return buffer;
1765 /* Handle references */
1766 else if (type == OBJ_COMMIT)
1767 ref_type = "tree ";
1768 else if (type == OBJ_TAG)
1769 ref_type = "object ";
1770 else {
1771 free(buffer);
1772 return NULL;
1774 ref_length = strlen(ref_type);
1776 if (ref_length + the_hash_algo->hexsz > isize ||
1777 memcmp(buffer, ref_type, ref_length) ||
1778 get_oid_hex((char *) buffer + ref_length, &actual_oid)) {
1779 free(buffer);
1780 return NULL;
1782 free(buffer);
1783 /* Now we have the ID of the referred-to object in
1784 * actual_oid. Check again. */
1788 static void hash_object_body(const struct git_hash_algo *algo, git_hash_ctx *c,
1789 const void *buf, unsigned long len,
1790 struct object_id *oid,
1791 char *hdr, int *hdrlen)
1793 algo->init_fn(c);
1794 algo->update_fn(c, hdr, *hdrlen);
1795 algo->update_fn(c, buf, len);
1796 algo->final_oid_fn(oid, c);
1799 static void write_object_file_prepare(const struct git_hash_algo *algo,
1800 const void *buf, unsigned long len,
1801 enum object_type type, struct object_id *oid,
1802 char *hdr, int *hdrlen)
1804 git_hash_ctx c;
1806 /* Generate the header */
1807 *hdrlen = format_object_header(hdr, *hdrlen, type, len);
1809 /* Sha1.. */
1810 hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen);
1813 static void write_object_file_prepare_literally(const struct git_hash_algo *algo,
1814 const void *buf, unsigned long len,
1815 const char *type, struct object_id *oid,
1816 char *hdr, int *hdrlen)
1818 git_hash_ctx c;
1820 *hdrlen = format_object_header_literally(hdr, *hdrlen, type, len);
1821 hash_object_body(algo, &c, buf, len, oid, hdr, hdrlen);
1825 * Move the just written object into its final resting place.
1827 int finalize_object_file(const char *tmpfile, const char *filename)
1829 int ret = 0;
1831 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
1832 goto try_rename;
1833 else if (link(tmpfile, filename))
1834 ret = errno;
1837 * Coda hack - coda doesn't like cross-directory links,
1838 * so we fall back to a rename, which will mean that it
1839 * won't be able to check collisions, but that's not a
1840 * big deal.
1842 * The same holds for FAT formatted media.
1844 * When this succeeds, we just return. We have nothing
1845 * left to unlink.
1847 if (ret && ret != EEXIST) {
1848 try_rename:
1849 if (!rename(tmpfile, filename))
1850 goto out;
1851 ret = errno;
1853 unlink_or_warn(tmpfile);
1854 if (ret) {
1855 if (ret != EEXIST) {
1856 return error_errno(_("unable to write file %s"), filename);
1858 /* FIXME!!! Collision check here ? */
1861 out:
1862 if (adjust_shared_perm(filename))
1863 return error(_("unable to set permission to '%s'"), filename);
1864 return 0;
1867 static int write_buffer(int fd, const void *buf, size_t len)
1869 if (write_in_full(fd, buf, len) < 0)
1870 return error_errno(_("file write error"));
1871 return 0;
1874 static void hash_object_file_literally(const struct git_hash_algo *algo,
1875 const void *buf, unsigned long len,
1876 const char *type, struct object_id *oid)
1878 char hdr[MAX_HEADER_LEN];
1879 int hdrlen = sizeof(hdr);
1881 write_object_file_prepare_literally(algo, buf, len, type, oid, hdr, &hdrlen);
1884 void hash_object_file(const struct git_hash_algo *algo, const void *buf,
1885 unsigned long len, enum object_type type,
1886 struct object_id *oid)
1888 hash_object_file_literally(algo, buf, len, type_name(type), oid);
1891 /* Finalize a file on disk, and close it. */
1892 static void close_loose_object(int fd, const char *filename)
1894 if (the_repository->objects->odb->will_destroy)
1895 goto out;
1897 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
1898 fsync_loose_object_bulk_checkin(fd, filename);
1899 else if (fsync_object_files > 0)
1900 fsync_or_die(fd, filename);
1901 else
1902 fsync_component_or_die(FSYNC_COMPONENT_LOOSE_OBJECT, fd,
1903 filename);
1905 out:
1906 if (close(fd) != 0)
1907 die_errno(_("error when closing loose object file"));
1910 /* Size of directory component, including the ending '/' */
1911 static inline int directory_size(const char *filename)
1913 const char *s = strrchr(filename, '/');
1914 if (!s)
1915 return 0;
1916 return s - filename + 1;
1920 * This creates a temporary file in the same directory as the final
1921 * 'filename'
1923 * We want to avoid cross-directory filename renames, because those
1924 * can have problems on various filesystems (FAT, NFS, Coda).
1926 static int create_tmpfile(struct strbuf *tmp, const char *filename)
1928 int fd, dirlen = directory_size(filename);
1930 strbuf_reset(tmp);
1931 strbuf_add(tmp, filename, dirlen);
1932 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
1933 fd = git_mkstemp_mode(tmp->buf, 0444);
1934 if (fd < 0 && dirlen && errno == ENOENT) {
1936 * Make sure the directory exists; note that the contents
1937 * of the buffer are undefined after mkstemp returns an
1938 * error, so we have to rewrite the whole buffer from
1939 * scratch.
1941 strbuf_reset(tmp);
1942 strbuf_add(tmp, filename, dirlen - 1);
1943 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
1944 return -1;
1945 if (adjust_shared_perm(tmp->buf))
1946 return -1;
1948 /* Try again */
1949 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
1950 fd = git_mkstemp_mode(tmp->buf, 0444);
1952 return fd;
1956 * Common steps for loose object writers to start writing loose
1957 * objects:
1959 * - Create tmpfile for the loose object.
1960 * - Setup zlib stream for compression.
1961 * - Start to feed header to zlib stream.
1963 * Returns a "fd", which should later be provided to
1964 * end_loose_object_common().
1966 static int start_loose_object_common(struct strbuf *tmp_file,
1967 const char *filename, unsigned flags,
1968 git_zstream *stream,
1969 unsigned char *buf, size_t buflen,
1970 git_hash_ctx *c,
1971 char *hdr, int hdrlen)
1973 int fd;
1975 fd = create_tmpfile(tmp_file, filename);
1976 if (fd < 0) {
1977 if (flags & HASH_SILENT)
1978 return -1;
1979 else if (errno == EACCES)
1980 return error(_("insufficient permission for adding "
1981 "an object to repository database %s"),
1982 get_object_directory());
1983 else
1984 return error_errno(
1985 _("unable to create temporary file"));
1988 /* Setup zlib stream for compression */
1989 git_deflate_init(stream, zlib_compression_level);
1990 stream->next_out = buf;
1991 stream->avail_out = buflen;
1992 the_hash_algo->init_fn(c);
1994 /* Start to feed header to zlib stream */
1995 stream->next_in = (unsigned char *)hdr;
1996 stream->avail_in = hdrlen;
1997 while (git_deflate(stream, 0) == Z_OK)
1998 ; /* nothing */
1999 the_hash_algo->update_fn(c, hdr, hdrlen);
2001 return fd;
2005 * Common steps for the inner git_deflate() loop for writing loose
2006 * objects. Returns what git_deflate() returns.
2008 static int write_loose_object_common(git_hash_ctx *c,
2009 git_zstream *stream, const int flush,
2010 unsigned char *in0, const int fd,
2011 unsigned char *compressed,
2012 const size_t compressed_len)
2014 int ret;
2016 ret = git_deflate(stream, flush ? Z_FINISH : 0);
2017 the_hash_algo->update_fn(c, in0, stream->next_in - in0);
2018 if (write_buffer(fd, compressed, stream->next_out - compressed) < 0)
2019 die(_("unable to write loose object file"));
2020 stream->next_out = compressed;
2021 stream->avail_out = compressed_len;
2023 return ret;
2027 * Common steps for loose object writers to end writing loose objects:
2029 * - End the compression of zlib stream.
2030 * - Get the calculated oid to "oid".
2032 static int end_loose_object_common(git_hash_ctx *c, git_zstream *stream,
2033 struct object_id *oid)
2035 int ret;
2037 ret = git_deflate_end_gently(stream);
2038 if (ret != Z_OK)
2039 return ret;
2040 the_hash_algo->final_oid_fn(oid, c);
2042 return Z_OK;
2045 static int write_loose_object(const struct object_id *oid, char *hdr,
2046 int hdrlen, const void *buf, unsigned long len,
2047 time_t mtime, unsigned flags)
2049 int fd, ret;
2050 unsigned char compressed[4096];
2051 git_zstream stream;
2052 git_hash_ctx c;
2053 struct object_id parano_oid;
2054 static struct strbuf tmp_file = STRBUF_INIT;
2055 static struct strbuf filename = STRBUF_INIT;
2057 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
2058 prepare_loose_object_bulk_checkin();
2060 loose_object_path(the_repository, &filename, oid);
2062 fd = start_loose_object_common(&tmp_file, filename.buf, flags,
2063 &stream, compressed, sizeof(compressed),
2064 &c, hdr, hdrlen);
2065 if (fd < 0)
2066 return -1;
2068 /* Then the data itself.. */
2069 stream.next_in = (void *)buf;
2070 stream.avail_in = len;
2071 do {
2072 unsigned char *in0 = stream.next_in;
2074 ret = write_loose_object_common(&c, &stream, 1, in0, fd,
2075 compressed, sizeof(compressed));
2076 } while (ret == Z_OK);
2078 if (ret != Z_STREAM_END)
2079 die(_("unable to deflate new object %s (%d)"), oid_to_hex(oid),
2080 ret);
2081 ret = end_loose_object_common(&c, &stream, &parano_oid);
2082 if (ret != Z_OK)
2083 die(_("deflateEnd on object %s failed (%d)"), oid_to_hex(oid),
2084 ret);
2085 if (!oideq(oid, &parano_oid))
2086 die(_("confused by unstable object source data for %s"),
2087 oid_to_hex(oid));
2089 close_loose_object(fd, tmp_file.buf);
2091 if (mtime) {
2092 struct utimbuf utb;
2093 utb.actime = mtime;
2094 utb.modtime = mtime;
2095 if (utime(tmp_file.buf, &utb) < 0 &&
2096 !(flags & HASH_SILENT))
2097 warning_errno(_("failed utime() on %s"), tmp_file.buf);
2100 return finalize_object_file(tmp_file.buf, filename.buf);
2103 static int freshen_loose_object(const struct object_id *oid)
2105 return check_and_freshen(oid, 1);
2108 static int freshen_packed_object(const struct object_id *oid)
2110 struct pack_entry e;
2111 if (!find_pack_entry(the_repository, oid, &e))
2112 return 0;
2113 if (e.p->is_cruft)
2114 return 0;
2115 if (e.p->freshened)
2116 return 1;
2117 if (!freshen_file(e.p->pack_name))
2118 return 0;
2119 e.p->freshened = 1;
2120 return 1;
2123 int stream_loose_object(struct input_stream *in_stream, size_t len,
2124 struct object_id *oid)
2126 int fd, ret, err = 0, flush = 0;
2127 unsigned char compressed[4096];
2128 git_zstream stream;
2129 git_hash_ctx c;
2130 struct strbuf tmp_file = STRBUF_INIT;
2131 struct strbuf filename = STRBUF_INIT;
2132 int dirlen;
2133 char hdr[MAX_HEADER_LEN];
2134 int hdrlen;
2136 if (batch_fsync_enabled(FSYNC_COMPONENT_LOOSE_OBJECT))
2137 prepare_loose_object_bulk_checkin();
2139 /* Since oid is not determined, save tmp file to odb path. */
2140 strbuf_addf(&filename, "%s/", get_object_directory());
2141 hdrlen = format_object_header(hdr, sizeof(hdr), OBJ_BLOB, len);
2144 * Common steps for write_loose_object and stream_loose_object to
2145 * start writing loose objects:
2147 * - Create tmpfile for the loose object.
2148 * - Setup zlib stream for compression.
2149 * - Start to feed header to zlib stream.
2151 fd = start_loose_object_common(&tmp_file, filename.buf, 0,
2152 &stream, compressed, sizeof(compressed),
2153 &c, hdr, hdrlen);
2154 if (fd < 0) {
2155 err = -1;
2156 goto cleanup;
2159 /* Then the data itself.. */
2160 do {
2161 unsigned char *in0 = stream.next_in;
2163 if (!stream.avail_in && !in_stream->is_finished) {
2164 const void *in = in_stream->read(in_stream, &stream.avail_in);
2165 stream.next_in = (void *)in;
2166 in0 = (unsigned char *)in;
2167 /* All data has been read. */
2168 if (in_stream->is_finished)
2169 flush = 1;
2171 ret = write_loose_object_common(&c, &stream, flush, in0, fd,
2172 compressed, sizeof(compressed));
2174 * Unlike write_loose_object(), we do not have the entire
2175 * buffer. If we get Z_BUF_ERROR due to too few input bytes,
2176 * then we'll replenish them in the next input_stream->read()
2177 * call when we loop.
2179 } while (ret == Z_OK || ret == Z_BUF_ERROR);
2181 if (stream.total_in != len + hdrlen)
2182 die(_("write stream object %ld != %"PRIuMAX), stream.total_in,
2183 (uintmax_t)len + hdrlen);
2186 * Common steps for write_loose_object and stream_loose_object to
2187 * end writing loose oject:
2189 * - End the compression of zlib stream.
2190 * - Get the calculated oid.
2192 if (ret != Z_STREAM_END)
2193 die(_("unable to stream deflate new object (%d)"), ret);
2194 ret = end_loose_object_common(&c, &stream, oid);
2195 if (ret != Z_OK)
2196 die(_("deflateEnd on stream object failed (%d)"), ret);
2197 close_loose_object(fd, tmp_file.buf);
2199 if (freshen_packed_object(oid) || freshen_loose_object(oid)) {
2200 unlink_or_warn(tmp_file.buf);
2201 goto cleanup;
2204 loose_object_path(the_repository, &filename, oid);
2206 /* We finally know the object path, and create the missing dir. */
2207 dirlen = directory_size(filename.buf);
2208 if (dirlen) {
2209 struct strbuf dir = STRBUF_INIT;
2210 strbuf_add(&dir, filename.buf, dirlen);
2212 if (mkdir_in_gitdir(dir.buf) && errno != EEXIST) {
2213 err = error_errno(_("unable to create directory %s"), dir.buf);
2214 strbuf_release(&dir);
2215 goto cleanup;
2217 strbuf_release(&dir);
2220 err = finalize_object_file(tmp_file.buf, filename.buf);
2221 cleanup:
2222 strbuf_release(&tmp_file);
2223 strbuf_release(&filename);
2224 return err;
2227 int write_object_file_flags(const void *buf, unsigned long len,
2228 enum object_type type, struct object_id *oid,
2229 unsigned flags)
2231 char hdr[MAX_HEADER_LEN];
2232 int hdrlen = sizeof(hdr);
2234 /* Normally if we have it in the pack then we do not bother writing
2235 * it out into .git/objects/??/?{38} file.
2237 write_object_file_prepare(the_hash_algo, buf, len, type, oid, hdr,
2238 &hdrlen);
2239 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2240 return 0;
2241 return write_loose_object(oid, hdr, hdrlen, buf, len, 0, flags);
2244 int write_object_file_literally(const void *buf, unsigned long len,
2245 const char *type, struct object_id *oid,
2246 unsigned flags)
2248 char *header;
2249 int hdrlen, status = 0;
2251 /* type string, SP, %lu of the length plus NUL must fit this */
2252 hdrlen = strlen(type) + MAX_HEADER_LEN;
2253 header = xmalloc(hdrlen);
2254 write_object_file_prepare_literally(the_hash_algo, buf, len, type,
2255 oid, header, &hdrlen);
2257 if (!(flags & HASH_WRITE_OBJECT))
2258 goto cleanup;
2259 if (freshen_packed_object(oid) || freshen_loose_object(oid))
2260 goto cleanup;
2261 status = write_loose_object(oid, header, hdrlen, buf, len, 0, 0);
2263 cleanup:
2264 free(header);
2265 return status;
2268 int force_object_loose(const struct object_id *oid, time_t mtime)
2270 void *buf;
2271 unsigned long len;
2272 enum object_type type;
2273 char hdr[MAX_HEADER_LEN];
2274 int hdrlen;
2275 int ret;
2277 if (has_loose_object(oid))
2278 return 0;
2279 buf = read_object(the_repository, oid, &type, &len);
2280 if (!buf)
2281 return error(_("cannot read object for %s"), oid_to_hex(oid));
2282 hdrlen = format_object_header(hdr, sizeof(hdr), type, len);
2283 ret = write_loose_object(oid, hdr, hdrlen, buf, len, mtime, 0);
2284 free(buf);
2286 return ret;
2289 int has_object(struct repository *r, const struct object_id *oid,
2290 unsigned flags)
2292 int quick = !(flags & HAS_OBJECT_RECHECK_PACKED);
2293 unsigned object_info_flags = OBJECT_INFO_SKIP_FETCH_OBJECT |
2294 (quick ? OBJECT_INFO_QUICK : 0);
2296 if (!startup_info->have_repository)
2297 return 0;
2298 return oid_object_info_extended(r, oid, NULL, object_info_flags) >= 0;
2301 int repo_has_object_file_with_flags(struct repository *r,
2302 const struct object_id *oid, int flags)
2304 if (!startup_info->have_repository)
2305 return 0;
2306 return oid_object_info_extended(r, oid, NULL, flags) >= 0;
2309 int repo_has_object_file(struct repository *r,
2310 const struct object_id *oid)
2312 return repo_has_object_file_with_flags(r, oid, 0);
2315 static void check_tree(const void *buf, size_t size)
2317 struct tree_desc desc;
2318 struct name_entry entry;
2320 init_tree_desc(&desc, buf, size);
2321 while (tree_entry(&desc, &entry))
2322 /* do nothing
2323 * tree_entry() will die() on malformed entries */
2327 static void check_commit(const void *buf, size_t size)
2329 struct commit c;
2330 memset(&c, 0, sizeof(c));
2331 if (parse_commit_buffer(the_repository, &c, buf, size, 0))
2332 die(_("corrupt commit"));
2335 static void check_tag(const void *buf, size_t size)
2337 struct tag t;
2338 memset(&t, 0, sizeof(t));
2339 if (parse_tag_buffer(the_repository, &t, buf, size))
2340 die(_("corrupt tag"));
2343 static int index_mem(struct index_state *istate,
2344 struct object_id *oid, void *buf, size_t size,
2345 enum object_type type,
2346 const char *path, unsigned flags)
2348 int ret = 0;
2349 int re_allocated = 0;
2350 int write_object = flags & HASH_WRITE_OBJECT;
2352 if (!type)
2353 type = OBJ_BLOB;
2356 * Convert blobs to git internal format
2358 if ((type == OBJ_BLOB) && path) {
2359 struct strbuf nbuf = STRBUF_INIT;
2360 if (convert_to_git(istate, path, buf, size, &nbuf,
2361 get_conv_flags(flags))) {
2362 buf = strbuf_detach(&nbuf, &size);
2363 re_allocated = 1;
2366 if (flags & HASH_FORMAT_CHECK) {
2367 if (type == OBJ_TREE)
2368 check_tree(buf, size);
2369 if (type == OBJ_COMMIT)
2370 check_commit(buf, size);
2371 if (type == OBJ_TAG)
2372 check_tag(buf, size);
2375 if (write_object)
2376 ret = write_object_file(buf, size, type, oid);
2377 else
2378 hash_object_file(the_hash_algo, buf, size, type, oid);
2379 if (re_allocated)
2380 free(buf);
2381 return ret;
2384 static int index_stream_convert_blob(struct index_state *istate,
2385 struct object_id *oid,
2386 int fd,
2387 const char *path,
2388 unsigned flags)
2390 int ret = 0;
2391 const int write_object = flags & HASH_WRITE_OBJECT;
2392 struct strbuf sbuf = STRBUF_INIT;
2394 assert(path);
2395 assert(would_convert_to_git_filter_fd(istate, path));
2397 convert_to_git_filter_fd(istate, path, fd, &sbuf,
2398 get_conv_flags(flags));
2400 if (write_object)
2401 ret = write_object_file(sbuf.buf, sbuf.len, OBJ_BLOB,
2402 oid);
2403 else
2404 hash_object_file(the_hash_algo, sbuf.buf, sbuf.len, OBJ_BLOB,
2405 oid);
2406 strbuf_release(&sbuf);
2407 return ret;
2410 static int index_pipe(struct index_state *istate, struct object_id *oid,
2411 int fd, enum object_type type,
2412 const char *path, unsigned flags)
2414 struct strbuf sbuf = STRBUF_INIT;
2415 int ret;
2417 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2418 ret = index_mem(istate, oid, sbuf.buf, sbuf.len, type, path, flags);
2419 else
2420 ret = -1;
2421 strbuf_release(&sbuf);
2422 return ret;
2425 #define SMALL_FILE_SIZE (32*1024)
2427 static int index_core(struct index_state *istate,
2428 struct object_id *oid, int fd, size_t size,
2429 enum object_type type, const char *path,
2430 unsigned flags)
2432 int ret;
2434 if (!size) {
2435 ret = index_mem(istate, oid, "", size, type, path, flags);
2436 } else if (size <= SMALL_FILE_SIZE) {
2437 char *buf = xmalloc(size);
2438 ssize_t read_result = read_in_full(fd, buf, size);
2439 if (read_result < 0)
2440 ret = error_errno(_("read error while indexing %s"),
2441 path ? path : "<unknown>");
2442 else if (read_result != size)
2443 ret = error(_("short read while indexing %s"),
2444 path ? path : "<unknown>");
2445 else
2446 ret = index_mem(istate, oid, buf, size, type, path, flags);
2447 free(buf);
2448 } else {
2449 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2450 ret = index_mem(istate, oid, buf, size, type, path, flags);
2451 munmap(buf, size);
2453 return ret;
2457 * This creates one packfile per large blob unless bulk-checkin
2458 * machinery is "plugged".
2460 * This also bypasses the usual "convert-to-git" dance, and that is on
2461 * purpose. We could write a streaming version of the converting
2462 * functions and insert that before feeding the data to fast-import
2463 * (or equivalent in-core API described above). However, that is
2464 * somewhat complicated, as we do not know the size of the filter
2465 * result, which we need to know beforehand when writing a git object.
2466 * Since the primary motivation for trying to stream from the working
2467 * tree file and to avoid mmaping it in core is to deal with large
2468 * binary blobs, they generally do not want to get any conversion, and
2469 * callers should avoid this code path when filters are requested.
2471 static int index_stream(struct object_id *oid, int fd, size_t size,
2472 enum object_type type, const char *path,
2473 unsigned flags)
2475 return index_bulk_checkin(oid, fd, size, type, path, flags);
2478 int index_fd(struct index_state *istate, struct object_id *oid,
2479 int fd, struct stat *st,
2480 enum object_type type, const char *path, unsigned flags)
2482 int ret;
2485 * Call xsize_t() only when needed to avoid potentially unnecessary
2486 * die() for large files.
2488 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, path))
2489 ret = index_stream_convert_blob(istate, oid, fd, path, flags);
2490 else if (!S_ISREG(st->st_mode))
2491 ret = index_pipe(istate, oid, fd, type, path, flags);
2492 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
2493 (path && would_convert_to_git(istate, path)))
2494 ret = index_core(istate, oid, fd, xsize_t(st->st_size),
2495 type, path, flags);
2496 else
2497 ret = index_stream(oid, fd, xsize_t(st->st_size), type, path,
2498 flags);
2499 close(fd);
2500 return ret;
2503 int index_path(struct index_state *istate, struct object_id *oid,
2504 const char *path, struct stat *st, unsigned flags)
2506 int fd;
2507 struct strbuf sb = STRBUF_INIT;
2508 int rc = 0;
2510 switch (st->st_mode & S_IFMT) {
2511 case S_IFREG:
2512 fd = open(path, O_RDONLY);
2513 if (fd < 0)
2514 return error_errno("open(\"%s\")", path);
2515 if (index_fd(istate, oid, fd, st, OBJ_BLOB, path, flags) < 0)
2516 return error(_("%s: failed to insert into database"),
2517 path);
2518 break;
2519 case S_IFLNK:
2520 if (strbuf_readlink(&sb, path, st->st_size))
2521 return error_errno("readlink(\"%s\")", path);
2522 if (!(flags & HASH_WRITE_OBJECT))
2523 hash_object_file(the_hash_algo, sb.buf, sb.len,
2524 OBJ_BLOB, oid);
2525 else if (write_object_file(sb.buf, sb.len, OBJ_BLOB, oid))
2526 rc = error(_("%s: failed to insert into database"), path);
2527 strbuf_release(&sb);
2528 break;
2529 case S_IFDIR:
2530 return resolve_gitlink_ref(path, "HEAD", oid);
2531 default:
2532 return error(_("%s: unsupported file type"), path);
2534 return rc;
2537 int read_pack_header(int fd, struct pack_header *header)
2539 if (read_in_full(fd, header, sizeof(*header)) != sizeof(*header))
2540 /* "eof before pack header was fully read" */
2541 return PH_ERROR_EOF;
2543 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2544 /* "protocol error (pack signature mismatch detected)" */
2545 return PH_ERROR_PACK_SIGNATURE;
2546 if (!pack_version_ok(header->hdr_version))
2547 /* "protocol error (pack version unsupported)" */
2548 return PH_ERROR_PROTOCOL;
2549 return 0;
2552 void assert_oid_type(const struct object_id *oid, enum object_type expect)
2554 enum object_type type = oid_object_info(the_repository, oid, NULL);
2555 if (type < 0)
2556 die(_("%s is not a valid object"), oid_to_hex(oid));
2557 if (type != expect)
2558 die(_("%s is not a valid '%s' object"), oid_to_hex(oid),
2559 type_name(expect));
2562 int for_each_file_in_obj_subdir(unsigned int subdir_nr,
2563 struct strbuf *path,
2564 each_loose_object_fn obj_cb,
2565 each_loose_cruft_fn cruft_cb,
2566 each_loose_subdir_fn subdir_cb,
2567 void *data)
2569 size_t origlen, baselen;
2570 DIR *dir;
2571 struct dirent *de;
2572 int r = 0;
2573 struct object_id oid;
2575 if (subdir_nr > 0xff)
2576 BUG("invalid loose object subdirectory: %x", subdir_nr);
2578 origlen = path->len;
2579 strbuf_complete(path, '/');
2580 strbuf_addf(path, "%02x", subdir_nr);
2582 dir = opendir(path->buf);
2583 if (!dir) {
2584 if (errno != ENOENT)
2585 r = error_errno(_("unable to open %s"), path->buf);
2586 strbuf_setlen(path, origlen);
2587 return r;
2590 oid.hash[0] = subdir_nr;
2591 strbuf_addch(path, '/');
2592 baselen = path->len;
2594 while ((de = readdir_skip_dot_and_dotdot(dir))) {
2595 size_t namelen;
2597 namelen = strlen(de->d_name);
2598 strbuf_setlen(path, baselen);
2599 strbuf_add(path, de->d_name, namelen);
2600 if (namelen == the_hash_algo->hexsz - 2 &&
2601 !hex_to_bytes(oid.hash + 1, de->d_name,
2602 the_hash_algo->rawsz - 1)) {
2603 oid_set_algo(&oid, the_hash_algo);
2604 if (obj_cb) {
2605 r = obj_cb(&oid, path->buf, data);
2606 if (r)
2607 break;
2609 continue;
2612 if (cruft_cb) {
2613 r = cruft_cb(de->d_name, path->buf, data);
2614 if (r)
2615 break;
2618 closedir(dir);
2620 strbuf_setlen(path, baselen - 1);
2621 if (!r && subdir_cb)
2622 r = subdir_cb(subdir_nr, path->buf, data);
2624 strbuf_setlen(path, origlen);
2626 return r;
2629 int for_each_loose_file_in_objdir_buf(struct strbuf *path,
2630 each_loose_object_fn obj_cb,
2631 each_loose_cruft_fn cruft_cb,
2632 each_loose_subdir_fn subdir_cb,
2633 void *data)
2635 int r = 0;
2636 int i;
2638 for (i = 0; i < 256; i++) {
2639 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
2640 subdir_cb, data);
2641 if (r)
2642 break;
2645 return r;
2648 int for_each_loose_file_in_objdir(const char *path,
2649 each_loose_object_fn obj_cb,
2650 each_loose_cruft_fn cruft_cb,
2651 each_loose_subdir_fn subdir_cb,
2652 void *data)
2654 struct strbuf buf = STRBUF_INIT;
2655 int r;
2657 strbuf_addstr(&buf, path);
2658 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
2659 subdir_cb, data);
2660 strbuf_release(&buf);
2662 return r;
2665 int for_each_loose_object(each_loose_object_fn cb, void *data,
2666 enum for_each_object_flags flags)
2668 struct object_directory *odb;
2670 prepare_alt_odb(the_repository);
2671 for (odb = the_repository->objects->odb; odb; odb = odb->next) {
2672 int r = for_each_loose_file_in_objdir(odb->path, cb, NULL,
2673 NULL, data);
2674 if (r)
2675 return r;
2677 if (flags & FOR_EACH_OBJECT_LOCAL_ONLY)
2678 break;
2681 return 0;
2684 static int append_loose_object(const struct object_id *oid, const char *path,
2685 void *data)
2687 oidtree_insert(data, oid);
2688 return 0;
2691 struct oidtree *odb_loose_cache(struct object_directory *odb,
2692 const struct object_id *oid)
2694 int subdir_nr = oid->hash[0];
2695 struct strbuf buf = STRBUF_INIT;
2696 size_t word_bits = bitsizeof(odb->loose_objects_subdir_seen[0]);
2697 size_t word_index = subdir_nr / word_bits;
2698 size_t mask = (size_t)1u << (subdir_nr % word_bits);
2699 uint32_t *bitmap;
2701 if (subdir_nr < 0 ||
2702 subdir_nr >= bitsizeof(odb->loose_objects_subdir_seen))
2703 BUG("subdir_nr out of range");
2705 bitmap = &odb->loose_objects_subdir_seen[word_index];
2706 if (*bitmap & mask)
2707 return odb->loose_objects_cache;
2708 if (!odb->loose_objects_cache) {
2709 ALLOC_ARRAY(odb->loose_objects_cache, 1);
2710 oidtree_init(odb->loose_objects_cache);
2712 strbuf_addstr(&buf, odb->path);
2713 for_each_file_in_obj_subdir(subdir_nr, &buf,
2714 append_loose_object,
2715 NULL, NULL,
2716 odb->loose_objects_cache);
2717 *bitmap |= mask;
2718 strbuf_release(&buf);
2719 return odb->loose_objects_cache;
2722 void odb_clear_loose_cache(struct object_directory *odb)
2724 oidtree_clear(odb->loose_objects_cache);
2725 FREE_AND_NULL(odb->loose_objects_cache);
2726 memset(&odb->loose_objects_subdir_seen, 0,
2727 sizeof(odb->loose_objects_subdir_seen));
2730 static int check_stream_oid(git_zstream *stream,
2731 const char *hdr,
2732 unsigned long size,
2733 const char *path,
2734 const struct object_id *expected_oid)
2736 git_hash_ctx c;
2737 struct object_id real_oid;
2738 unsigned char buf[4096];
2739 unsigned long total_read;
2740 int status = Z_OK;
2742 the_hash_algo->init_fn(&c);
2743 the_hash_algo->update_fn(&c, hdr, stream->total_out);
2746 * We already read some bytes into hdr, but the ones up to the NUL
2747 * do not count against the object's content size.
2749 total_read = stream->total_out - strlen(hdr) - 1;
2752 * This size comparison must be "<=" to read the final zlib packets;
2753 * see the comment in unpack_loose_rest for details.
2755 while (total_read <= size &&
2756 (status == Z_OK ||
2757 (status == Z_BUF_ERROR && !stream->avail_out))) {
2758 stream->next_out = buf;
2759 stream->avail_out = sizeof(buf);
2760 if (size - total_read < stream->avail_out)
2761 stream->avail_out = size - total_read;
2762 status = git_inflate(stream, Z_FINISH);
2763 the_hash_algo->update_fn(&c, buf, stream->next_out - buf);
2764 total_read += stream->next_out - buf;
2766 git_inflate_end(stream);
2768 if (status != Z_STREAM_END) {
2769 error(_("corrupt loose object '%s'"), oid_to_hex(expected_oid));
2770 return -1;
2772 if (stream->avail_in) {
2773 error(_("garbage at end of loose object '%s'"),
2774 oid_to_hex(expected_oid));
2775 return -1;
2778 the_hash_algo->final_oid_fn(&real_oid, &c);
2779 if (!oideq(expected_oid, &real_oid)) {
2780 error(_("hash mismatch for %s (expected %s)"), path,
2781 oid_to_hex(expected_oid));
2782 return -1;
2785 return 0;
2788 int read_loose_object(const char *path,
2789 const struct object_id *expected_oid,
2790 struct object_id *real_oid,
2791 void **contents,
2792 struct object_info *oi)
2794 int ret = -1;
2795 void *map = NULL;
2796 unsigned long mapsize;
2797 git_zstream stream;
2798 char hdr[MAX_HEADER_LEN];
2799 unsigned long *size = oi->sizep;
2801 map = map_loose_object_1(the_repository, path, NULL, &mapsize);
2802 if (!map) {
2803 error_errno(_("unable to mmap %s"), path);
2804 goto out;
2807 if (unpack_loose_header(&stream, map, mapsize, hdr, sizeof(hdr),
2808 NULL) != ULHR_OK) {
2809 error(_("unable to unpack header of %s"), path);
2810 goto out;
2813 if (parse_loose_header(hdr, oi) < 0) {
2814 error(_("unable to parse header of %s"), path);
2815 git_inflate_end(&stream);
2816 goto out;
2819 if (*oi->typep == OBJ_BLOB && *size > big_file_threshold) {
2820 if (check_stream_oid(&stream, hdr, *size, path, expected_oid) < 0)
2821 goto out;
2822 } else {
2823 *contents = unpack_loose_rest(&stream, hdr, *size, expected_oid);
2824 if (!*contents) {
2825 error(_("unable to unpack contents of %s"), path);
2826 git_inflate_end(&stream);
2827 goto out;
2829 hash_object_file_literally(the_repository->hash_algo,
2830 *contents, *size,
2831 oi->type_name->buf, real_oid);
2832 if (!oideq(expected_oid, real_oid))
2833 goto out;
2836 ret = 0; /* everything checks out */
2838 out:
2839 if (map)
2840 munmap(map, mapsize);
2841 return ret;