2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
16 #include "run-command.h"
19 #include "tree-walk.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
28 #include "mergesort.h"
32 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
33 #define O_NOATIME 01000000
39 #define SZ_FMT PRIuMAX
40 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
42 const unsigned char null_sha1
[20];
43 const struct object_id null_oid
;
44 const struct object_id empty_tree_oid
= {
45 EMPTY_TREE_SHA1_BIN_LITERAL
47 const struct object_id empty_blob_oid
= {
48 EMPTY_BLOB_SHA1_BIN_LITERAL
52 * This is meant to hold a *small* number of objects that you would
53 * want read_sha1_file() to be able to return, but yet you do not want
54 * to write them into the object store (e.g. a browse-only
57 static struct cached_object
{
58 unsigned char sha1
[20];
59 enum object_type type
;
63 static int cached_object_nr
, cached_object_alloc
;
65 static struct cached_object empty_tree
= {
66 EMPTY_TREE_SHA1_BIN_LITERAL
,
72 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
75 struct cached_object
*co
= cached_objects
;
77 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
78 if (!hashcmp(co
->sha1
, sha1
))
81 if (!hashcmp(sha1
, empty_tree
.sha1
))
86 int mkdir_in_gitdir(const char *path
)
88 if (mkdir(path
, 0777)) {
89 int saved_errno
= errno
;
91 struct strbuf sb
= STRBUF_INIT
;
96 * Are we looking at a path in a symlinked worktree
97 * whose original repository does not yet have it?
98 * e.g. .git/rr-cache pointing at its original
99 * repository in which the user hasn't performed any
100 * conflict resolution yet?
102 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
103 strbuf_readlink(&sb
, path
, st
.st_size
) ||
104 !is_absolute_path(sb
.buf
) ||
105 mkdir(sb
.buf
, 0777)) {
112 return adjust_shared_perm(path
);
115 enum scld_error
safe_create_leading_directories(char *path
)
117 char *next_component
= path
+ offset_1st_component(path
);
118 enum scld_error ret
= SCLD_OK
;
120 while (ret
== SCLD_OK
&& next_component
) {
122 char *slash
= next_component
, slash_character
;
124 while (*slash
&& !is_dir_sep(*slash
))
130 next_component
= slash
+ 1;
131 while (is_dir_sep(*next_component
))
133 if (!*next_component
)
136 slash_character
= *slash
;
138 if (!stat(path
, &st
)) {
140 if (!S_ISDIR(st
.st_mode
))
142 } else if (mkdir(path
, 0777)) {
143 if (errno
== EEXIST
&&
144 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
145 ; /* somebody created it since we checked */
146 else if (errno
== ENOENT
)
148 * Either mkdir() failed because
149 * somebody just pruned the containing
150 * directory, or stat() failed because
151 * the file that was in our way was
152 * just removed. Either way, inform
153 * the caller that it might be worth
159 } else if (adjust_shared_perm(path
)) {
162 *slash
= slash_character
;
167 enum scld_error
safe_create_leading_directories_const(const char *path
)
169 /* path points to cache entries, so xstrdup before messing with it */
170 char *buf
= xstrdup(path
);
171 enum scld_error result
= safe_create_leading_directories(buf
);
176 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
179 for (i
= 0; i
< 20; i
++) {
180 static char hex
[] = "0123456789abcdef";
181 unsigned int val
= sha1
[i
];
182 strbuf_addch(buf
, hex
[val
>> 4]);
183 strbuf_addch(buf
, hex
[val
& 0xf]);
185 strbuf_addch(buf
, '/');
189 const char *sha1_file_name(const unsigned char *sha1
)
191 static struct strbuf buf
= STRBUF_INIT
;
194 strbuf_addf(&buf
, "%s/", get_object_directory());
196 fill_sha1_path(&buf
, sha1
);
200 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
202 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
203 return &alt
->scratch
;
206 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
207 const unsigned char *sha1
)
209 struct strbuf
*buf
= alt_scratch_buf(alt
);
210 fill_sha1_path(buf
, sha1
);
215 * Return the name of the pack or index file with the specified sha1
216 * in its filename. *base and *name are scratch space that must be
217 * provided by the caller. which should be "pack" or "idx".
219 static char *sha1_get_pack_name(const unsigned char *sha1
,
224 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
225 sha1_to_hex(sha1
), which
);
229 char *sha1_pack_name(const unsigned char *sha1
)
231 static struct strbuf buf
= STRBUF_INIT
;
232 return sha1_get_pack_name(sha1
, &buf
, "pack");
235 char *sha1_pack_index_name(const unsigned char *sha1
)
237 static struct strbuf buf
= STRBUF_INIT
;
238 return sha1_get_pack_name(sha1
, &buf
, "idx");
241 struct alternate_object_database
*alt_odb_list
;
242 static struct alternate_object_database
**alt_odb_tail
;
245 * Return non-zero iff the path is usable as an alternate object database.
247 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
249 struct alternate_object_database
*alt
;
251 /* Detect cases where alternate disappeared */
252 if (!is_directory(path
->buf
)) {
253 error("object directory %s does not exist; "
254 "check .git/objects/info/alternates.",
260 * Prevent the common mistake of listing the same
261 * thing twice, or object directory itself.
263 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
264 if (!fspathcmp(path
->buf
, alt
->path
))
267 if (!fspathcmp(path
->buf
, normalized_objdir
))
274 * Prepare alternate object database registry.
276 * The variable alt_odb_list points at the list of struct
277 * alternate_object_database. The elements on this list come from
278 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
279 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
280 * whose contents is similar to that environment variable but can be
281 * LF separated. Its base points at a statically allocated buffer that
282 * contains "/the/directory/corresponding/to/.git/objects/...", while
283 * its name points just after the slash at the end of ".git/objects/"
284 * in the example above, and has enough space to hold 40-byte hex
285 * SHA1, an extra slash for the first level indirection, and the
288 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
289 int depth
, const char *normalized_objdir
)
291 struct alternate_object_database
*ent
;
292 struct strbuf pathbuf
= STRBUF_INIT
;
294 if (!is_absolute_path(entry
) && relative_base
) {
295 strbuf_addstr(&pathbuf
, real_path(relative_base
));
296 strbuf_addch(&pathbuf
, '/');
298 strbuf_addstr(&pathbuf
, entry
);
300 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
301 error("unable to normalize alternate object path: %s",
303 strbuf_release(&pathbuf
);
308 * The trailing slash after the directory name is given by
309 * this function at the end. Remove duplicates.
311 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
312 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
314 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
315 strbuf_release(&pathbuf
);
319 ent
= alloc_alt_odb(pathbuf
.buf
);
321 /* add the alternate entry */
323 alt_odb_tail
= &(ent
->next
);
326 /* recursively add alternates */
327 read_info_alternates(pathbuf
.buf
, depth
+ 1);
329 strbuf_release(&pathbuf
);
333 static const char *parse_alt_odb_entry(const char *string
,
341 if (*string
== '#') {
342 /* comment; consume up to next separator */
343 end
= strchrnul(string
, sep
);
344 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
346 * quoted path; unquote_c_style has copied the
347 * data for us and set "end". Broken quoting (e.g.,
348 * an entry that doesn't end with a quote) falls
349 * back to the unquoted case below.
352 /* normal, unquoted path */
353 end
= strchrnul(string
, sep
);
354 strbuf_add(out
, string
, end
- string
);
362 static void link_alt_odb_entries(const char *alt
, int sep
,
363 const char *relative_base
, int depth
)
365 struct strbuf objdirbuf
= STRBUF_INIT
;
366 struct strbuf entry
= STRBUF_INIT
;
369 error("%s: ignoring alternate object stores, nesting too deep.",
374 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
375 if (strbuf_normalize_path(&objdirbuf
) < 0)
376 die("unable to normalize object directory: %s",
380 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
383 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
385 strbuf_release(&entry
);
386 strbuf_release(&objdirbuf
);
389 void read_info_alternates(const char * relative_base
, int depth
)
392 struct strbuf buf
= STRBUF_INIT
;
394 path
= xstrfmt("%s/info/alternates", relative_base
);
395 if (strbuf_read_file(&buf
, path
, 1024) < 0) {
400 link_alt_odb_entries(buf
.buf
, '\n', relative_base
, depth
);
401 strbuf_release(&buf
);
405 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
407 struct alternate_object_database
*ent
;
409 FLEX_ALLOC_STR(ent
, path
, dir
);
410 strbuf_init(&ent
->scratch
, 0);
411 strbuf_addf(&ent
->scratch
, "%s/", dir
);
412 ent
->base_len
= ent
->scratch
.len
;
417 void add_to_alternates_file(const char *reference
)
419 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
420 char *alts
= git_pathdup("objects/info/alternates");
423 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
424 out
= fdopen_lock_file(lock
, "w");
426 die_errno("unable to fdopen alternates lockfile");
428 in
= fopen(alts
, "r");
430 struct strbuf line
= STRBUF_INIT
;
433 while (strbuf_getline(&line
, in
) != EOF
) {
434 if (!strcmp(reference
, line
.buf
)) {
438 fprintf_or_die(out
, "%s\n", line
.buf
);
441 strbuf_release(&line
);
445 rollback_lock_file(lock
);
449 else if (errno
!= ENOENT
)
450 die_errno("unable to read alternates file");
453 fprintf_or_die(out
, "%s\n", reference
);
454 if (commit_lock_file(lock
))
455 die_errno("unable to move new alternates file into place");
457 link_alt_odb_entries(reference
, '\n', NULL
, 0);
462 void add_to_alternates_memory(const char *reference
)
465 * Make sure alternates are initialized, or else our entry may be
466 * overwritten when they are.
470 link_alt_odb_entries(reference
, '\n', NULL
, 0);
474 * Compute the exact path an alternate is at and returns it. In case of
475 * error NULL is returned and the human readable error is added to `err`
476 * `path` may be relative and should point to $GITDIR.
477 * `err` must not be null.
479 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
481 char *ref_git
= NULL
;
482 const char *repo
, *ref_git_s
;
485 ref_git_s
= real_path_if_valid(path
);
488 strbuf_addf(err
, _("path '%s' does not exist"), path
);
492 * Beware: read_gitfile(), real_path() and mkpath()
493 * return static buffer
495 ref_git
= xstrdup(ref_git_s
);
497 repo
= read_gitfile(ref_git
);
499 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
502 ref_git
= xstrdup(repo
);
505 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
506 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
508 ref_git
= ref_git_git
;
509 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
510 struct strbuf sb
= STRBUF_INIT
;
512 if (get_common_dir(&sb
, ref_git
)) {
514 _("reference repository '%s' as a linked "
515 "checkout is not supported yet."),
520 strbuf_addf(err
, _("reference repository '%s' is not a "
521 "local repository."), path
);
525 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
526 strbuf_addf(err
, _("reference repository '%s' is shallow"),
532 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
534 _("reference repository '%s' is grafted"),
549 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
551 struct alternate_object_database
*ent
;
555 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
563 void prepare_alt_odb(void)
570 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
573 alt_odb_tail
= &alt_odb_list
;
574 link_alt_odb_entries(alt
, PATH_SEP
, NULL
, 0);
576 read_info_alternates(get_object_directory(), 0);
579 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
580 static int freshen_file(const char *fn
)
583 t
.actime
= t
.modtime
= time(NULL
);
584 return !utime(fn
, &t
);
588 * All of the check_and_freshen functions return 1 if the file exists and was
589 * freshened (if freshening was requested), 0 otherwise. If they return
590 * 0, you should not assume that it is safe to skip a write of the object (it
591 * either does not exist on disk, or has a stale mtime and may be subject to
594 static int check_and_freshen_file(const char *fn
, int freshen
)
596 if (access(fn
, F_OK
))
598 if (freshen
&& !freshen_file(fn
))
603 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
605 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
608 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
610 struct alternate_object_database
*alt
;
612 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
613 const char *path
= alt_sha1_path(alt
, sha1
);
614 if (check_and_freshen_file(path
, freshen
))
620 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
622 return check_and_freshen_local(sha1
, freshen
) ||
623 check_and_freshen_nonlocal(sha1
, freshen
);
626 int has_loose_object_nonlocal(const unsigned char *sha1
)
628 return check_and_freshen_nonlocal(sha1
, 0);
631 static int has_loose_object(const unsigned char *sha1
)
633 return check_and_freshen(sha1
, 0);
636 static unsigned int pack_used_ctr
;
637 static unsigned int pack_mmap_calls
;
638 static unsigned int peak_pack_open_windows
;
639 static unsigned int pack_open_windows
;
640 static unsigned int pack_open_fds
;
641 static unsigned int pack_max_fds
;
642 static size_t peak_pack_mapped
;
643 static size_t pack_mapped
;
644 struct packed_git
*packed_git
;
646 static struct mru packed_git_mru_storage
;
647 struct mru
*packed_git_mru
= &packed_git_mru_storage
;
649 void pack_report(void)
652 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
653 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
654 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
655 sz_fmt(getpagesize()),
656 sz_fmt(packed_git_window_size
),
657 sz_fmt(packed_git_limit
));
659 "pack_report: pack_used_ctr = %10u\n"
660 "pack_report: pack_mmap_calls = %10u\n"
661 "pack_report: pack_open_windows = %10u / %10u\n"
662 "pack_report: pack_mapped = "
663 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
666 pack_open_windows
, peak_pack_open_windows
,
667 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
671 * Open and mmap the index file at path, perform a couple of
672 * consistency checks, then record its information to p. Return 0 on
675 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
678 struct pack_idx_header
*hdr
;
680 uint32_t version
, nr
, i
, *index
;
681 int fd
= git_open(path
);
686 if (fstat(fd
, &st
)) {
690 idx_size
= xsize_t(st
.st_size
);
691 if (idx_size
< 4 * 256 + 20 + 20) {
693 return error("index file %s is too small", path
);
695 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
699 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
700 version
= ntohl(hdr
->idx_version
);
701 if (version
< 2 || version
> 2) {
702 munmap(idx_map
, idx_size
);
703 return error("index file %s is version %"PRIu32
704 " and is not supported by this binary"
705 " (try upgrading GIT to a newer version)",
714 index
+= 2; /* skip index header */
715 for (i
= 0; i
< 256; i
++) {
716 uint32_t n
= ntohl(index
[i
]);
718 munmap(idx_map
, idx_size
);
719 return error("non-monotonic index %s", path
);
727 * - 256 index entries 4 bytes each
728 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
729 * - 20-byte SHA1 of the packfile
730 * - 20-byte SHA1 file checksum
732 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
733 munmap(idx_map
, idx_size
);
734 return error("wrong index v1 file size in %s", path
);
736 } else if (version
== 2) {
739 * - 8 bytes of header
740 * - 256 index entries 4 bytes each
741 * - 20-byte sha1 entry * nr
742 * - 4-byte crc entry * nr
743 * - 4-byte offset entry * nr
744 * - 20-byte SHA1 of the packfile
745 * - 20-byte SHA1 file checksum
746 * And after the 4-byte offset table might be a
747 * variable sized table containing 8-byte entries
748 * for offsets larger than 2^31.
750 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
751 unsigned long max_size
= min_size
;
753 max_size
+= (nr
- 1)*8;
754 if (idx_size
< min_size
|| idx_size
> max_size
) {
755 munmap(idx_map
, idx_size
);
756 return error("wrong index v2 file size in %s", path
);
758 if (idx_size
!= min_size
&&
760 * make sure we can deal with large pack offsets.
761 * 31-bit signed offset won't be enough, neither
762 * 32-bit unsigned one will be.
764 (sizeof(off_t
) <= 4)) {
765 munmap(idx_map
, idx_size
);
766 return error("pack too large for current definition of off_t in %s", path
);
770 p
->index_version
= version
;
771 p
->index_data
= idx_map
;
772 p
->index_size
= idx_size
;
777 int open_pack_index(struct packed_git
*p
)
786 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
787 die("BUG: pack_name does not end in .pack");
788 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
789 ret
= check_packed_git_idx(idx_name
, p
);
794 static void scan_windows(struct packed_git
*p
,
795 struct packed_git
**lru_p
,
796 struct pack_window
**lru_w
,
797 struct pack_window
**lru_l
)
799 struct pack_window
*w
, *w_l
;
801 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
803 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
813 static int unuse_one_window(struct packed_git
*current
)
815 struct packed_git
*p
, *lru_p
= NULL
;
816 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
819 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
820 for (p
= packed_git
; p
; p
= p
->next
)
821 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
823 munmap(lru_w
->base
, lru_w
->len
);
824 pack_mapped
-= lru_w
->len
;
826 lru_l
->next
= lru_w
->next
;
828 lru_p
->windows
= lru_w
->next
;
836 void release_pack_memory(size_t need
)
838 size_t cur
= pack_mapped
;
839 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
843 static void mmap_limit_check(size_t length
)
845 static size_t limit
= 0;
847 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
852 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
853 (uintmax_t)length
, (uintmax_t)limit
);
856 void *xmmap_gently(void *start
, size_t length
,
857 int prot
, int flags
, int fd
, off_t offset
)
861 mmap_limit_check(length
);
862 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
863 if (ret
== MAP_FAILED
) {
866 release_pack_memory(length
);
867 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
872 void *xmmap(void *start
, size_t length
,
873 int prot
, int flags
, int fd
, off_t offset
)
875 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
876 if (ret
== MAP_FAILED
)
877 die_errno("mmap failed");
881 void close_pack_windows(struct packed_git
*p
)
884 struct pack_window
*w
= p
->windows
;
887 die("pack '%s' still has open windows to it",
889 munmap(w
->base
, w
->len
);
890 pack_mapped
-= w
->len
;
892 p
->windows
= w
->next
;
897 static int close_pack_fd(struct packed_git
*p
)
909 static void close_pack(struct packed_git
*p
)
911 close_pack_windows(p
);
916 void close_all_packs(void)
918 struct packed_git
*p
;
920 for (p
= packed_git
; p
; p
= p
->next
)
922 die("BUG: want to close pack marked 'do-not-close'");
929 * The LRU pack is the one with the oldest MRU window, preferring packs
930 * with no used windows, or the oldest mtime if it has no windows allocated.
932 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
934 struct pack_window
*w
, *this_mru_w
;
935 int has_windows_inuse
= 0;
938 * Reject this pack if it has windows and the previously selected
939 * one does not. If this pack does not have windows, reject
940 * it if the pack file is newer than the previously selected one.
942 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
945 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
947 * Reject this pack if any of its windows are in use,
948 * but the previously selected pack did not have any
949 * inuse windows. Otherwise, record that this pack
950 * has windows in use.
953 if (*accept_windows_inuse
)
954 has_windows_inuse
= 1;
959 if (w
->last_used
> this_mru_w
->last_used
)
963 * Reject this pack if it has windows that have been
964 * used more recently than the previously selected pack.
965 * If the previously selected pack had windows inuse and
966 * we have not encountered a window in this pack that is
967 * inuse, skip this check since we prefer a pack with no
968 * inuse windows to one that has inuse windows.
970 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
971 this_mru_w
->last_used
> (*mru_w
)->last_used
)
980 *accept_windows_inuse
= has_windows_inuse
;
983 static int close_one_pack(void)
985 struct packed_git
*p
, *lru_p
= NULL
;
986 struct pack_window
*mru_w
= NULL
;
987 int accept_windows_inuse
= 1;
989 for (p
= packed_git
; p
; p
= p
->next
) {
990 if (p
->pack_fd
== -1)
992 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
996 return close_pack_fd(lru_p
);
1001 void unuse_pack(struct pack_window
**w_cursor
)
1003 struct pack_window
*w
= *w_cursor
;
1010 void close_pack_index(struct packed_git
*p
)
1012 if (p
->index_data
) {
1013 munmap((void *)p
->index_data
, p
->index_size
);
1014 p
->index_data
= NULL
;
1018 static unsigned int get_max_fd_limit(void)
1020 #ifdef RLIMIT_NOFILE
1024 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
1025 return lim
.rlim_cur
;
1031 long open_max
= sysconf(_SC_OPEN_MAX
);
1035 * Otherwise, we got -1 for one of the two
1038 * (1) sysconf() did not understand _SC_OPEN_MAX
1039 * and signaled an error with -1; or
1040 * (2) sysconf() said there is no limit.
1042 * We _could_ clear errno before calling sysconf() to
1043 * tell these two cases apart and return a huge number
1044 * in the latter case to let the caller cap it to a
1045 * value that is not so selfish, but letting the
1046 * fallback OPEN_MAX codepath take care of these cases
1055 return 1; /* see the caller ;-) */
1060 * Do not call this directly as this leaks p->pack_fd on error return;
1061 * call open_packed_git() instead.
1063 static int open_packed_git_1(struct packed_git
*p
)
1066 struct pack_header hdr
;
1067 unsigned char sha1
[20];
1068 unsigned char *idx_sha1
;
1071 if (!p
->index_data
&& open_pack_index(p
))
1072 return error("packfile %s index unavailable", p
->pack_name
);
1074 if (!pack_max_fds
) {
1075 unsigned int max_fds
= get_max_fd_limit();
1077 /* Save 3 for stdin/stdout/stderr, 22 for work */
1079 pack_max_fds
= max_fds
- 25;
1084 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
1087 p
->pack_fd
= git_open(p
->pack_name
);
1088 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
1092 /* If we created the struct before we had the pack we lack size. */
1093 if (!p
->pack_size
) {
1094 if (!S_ISREG(st
.st_mode
))
1095 return error("packfile %s not a regular file", p
->pack_name
);
1096 p
->pack_size
= st
.st_size
;
1097 } else if (p
->pack_size
!= st
.st_size
)
1098 return error("packfile %s size changed", p
->pack_name
);
1100 /* We leave these file descriptors open with sliding mmap;
1101 * there is no point keeping them open across exec(), though.
1103 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
1105 return error("cannot determine file descriptor flags");
1106 fd_flag
|= FD_CLOEXEC
;
1107 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
1108 return error("cannot set FD_CLOEXEC");
1110 /* Verify we recognize this pack file format. */
1111 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
1112 return error("file %s is far too short to be a packfile", p
->pack_name
);
1113 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
1114 return error("file %s is not a GIT packfile", p
->pack_name
);
1115 if (!pack_version_ok(hdr
.hdr_version
))
1116 return error("packfile %s is version %"PRIu32
" and not"
1117 " supported (try upgrading GIT to a newer version)",
1118 p
->pack_name
, ntohl(hdr
.hdr_version
));
1120 /* Verify the pack matches its index. */
1121 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
1122 return error("packfile %s claims to have %"PRIu32
" objects"
1123 " while index indicates %"PRIu32
" objects",
1124 p
->pack_name
, ntohl(hdr
.hdr_entries
),
1126 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1127 return error("end of packfile %s is unavailable", p
->pack_name
);
1128 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1129 return error("packfile %s signature is unavailable", p
->pack_name
);
1130 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1131 if (hashcmp(sha1
, idx_sha1
))
1132 return error("packfile %s does not match index", p
->pack_name
);
1136 static int open_packed_git(struct packed_git
*p
)
1138 if (!open_packed_git_1(p
))
1144 static int in_window(struct pack_window
*win
, off_t offset
)
1146 /* We must promise at least 20 bytes (one hash) after the
1147 * offset is available from this window, otherwise the offset
1148 * is not actually in this window and a different window (which
1149 * has that one hash excess) must be used. This is to support
1150 * the object header and delta base parsing routines below.
1152 off_t win_off
= win
->offset
;
1153 return win_off
<= offset
1154 && (offset
+ 20) <= (win_off
+ win
->len
);
1157 unsigned char *use_pack(struct packed_git
*p
,
1158 struct pack_window
**w_cursor
,
1160 unsigned long *left
)
1162 struct pack_window
*win
= *w_cursor
;
1164 /* Since packfiles end in a hash of their content and it's
1165 * pointless to ask for an offset into the middle of that
1166 * hash, and the in_window function above wouldn't match
1167 * don't allow an offset too close to the end of the file.
1169 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1170 die("packfile %s cannot be accessed", p
->pack_name
);
1171 if (offset
> (p
->pack_size
- 20))
1172 die("offset beyond end of packfile (truncated pack?)");
1174 die(_("offset before end of packfile (broken .idx?)"));
1176 if (!win
|| !in_window(win
, offset
)) {
1179 for (win
= p
->windows
; win
; win
= win
->next
) {
1180 if (in_window(win
, offset
))
1184 size_t window_align
= packed_git_window_size
/ 2;
1187 if (p
->pack_fd
== -1 && open_packed_git(p
))
1188 die("packfile %s cannot be accessed", p
->pack_name
);
1190 win
= xcalloc(1, sizeof(*win
));
1191 win
->offset
= (offset
/ window_align
) * window_align
;
1192 len
= p
->pack_size
- win
->offset
;
1193 if (len
> packed_git_window_size
)
1194 len
= packed_git_window_size
;
1195 win
->len
= (size_t)len
;
1196 pack_mapped
+= win
->len
;
1197 while (packed_git_limit
< pack_mapped
1198 && unuse_one_window(p
))
1200 win
->base
= xmmap(NULL
, win
->len
,
1201 PROT_READ
, MAP_PRIVATE
,
1202 p
->pack_fd
, win
->offset
);
1203 if (win
->base
== MAP_FAILED
)
1204 die_errno("packfile %s cannot be mapped",
1206 if (!win
->offset
&& win
->len
== p
->pack_size
1207 && !p
->do_not_close
)
1210 pack_open_windows
++;
1211 if (pack_mapped
> peak_pack_mapped
)
1212 peak_pack_mapped
= pack_mapped
;
1213 if (pack_open_windows
> peak_pack_open_windows
)
1214 peak_pack_open_windows
= pack_open_windows
;
1215 win
->next
= p
->windows
;
1219 if (win
!= *w_cursor
) {
1220 win
->last_used
= pack_used_ctr
++;
1224 offset
-= win
->offset
;
1226 *left
= win
->len
- xsize_t(offset
);
1227 return win
->base
+ offset
;
1230 static struct packed_git
*alloc_packed_git(int extra
)
1232 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
1233 memset(p
, 0, sizeof(*p
));
1238 static void try_to_free_pack_memory(size_t size
)
1240 release_pack_memory(size
);
1243 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1245 static int have_set_try_to_free_routine
;
1248 struct packed_git
*p
;
1250 if (!have_set_try_to_free_routine
) {
1251 have_set_try_to_free_routine
= 1;
1252 set_try_to_free_routine(try_to_free_pack_memory
);
1256 * Make sure a corresponding .pack file exists and that
1257 * the index looks sane.
1259 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1263 * ".pack" is long enough to hold any suffix we're adding (and
1264 * the use xsnprintf double-checks that)
1266 alloc
= st_add3(path_len
, strlen(".pack"), 1);
1267 p
= alloc_packed_git(alloc
);
1268 memcpy(p
->pack_name
, path
, path_len
);
1270 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1271 if (!access(p
->pack_name
, F_OK
))
1274 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1275 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1280 /* ok, it looks sane as far as we can check without
1281 * actually mapping the pack file.
1283 p
->pack_size
= st
.st_size
;
1284 p
->pack_local
= local
;
1285 p
->mtime
= st
.st_mtime
;
1286 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1291 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1293 const char *path
= sha1_pack_name(sha1
);
1294 size_t alloc
= st_add(strlen(path
), 1);
1295 struct packed_git
*p
= alloc_packed_git(alloc
);
1297 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
1298 hashcpy(p
->sha1
, sha1
);
1299 if (check_packed_git_idx(idx_path
, p
)) {
1307 void install_packed_git(struct packed_git
*pack
)
1309 if (pack
->pack_fd
!= -1)
1312 pack
->next
= packed_git
;
1316 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
1318 static void report_helper(const struct string_list
*list
,
1319 int seen_bits
, int first
, int last
)
1321 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
1324 for (; first
< last
; first
++)
1325 report_garbage(seen_bits
, list
->items
[first
].string
);
1328 static void report_pack_garbage(struct string_list
*list
)
1330 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1332 if (!report_garbage
)
1335 string_list_sort(list
);
1337 for (i
= 0; i
< list
->nr
; i
++) {
1338 const char *path
= list
->items
[i
].string
;
1339 if (baselen
!= -1 &&
1340 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1341 report_helper(list
, seen_bits
, first
, i
);
1345 if (baselen
== -1) {
1346 const char *dot
= strrchr(path
, '.');
1348 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
1351 baselen
= dot
- path
+ 1;
1354 if (!strcmp(path
+ baselen
, "pack"))
1356 else if (!strcmp(path
+ baselen
, "idx"))
1359 report_helper(list
, seen_bits
, first
, list
->nr
);
1362 static void prepare_packed_git_one(char *objdir
, int local
)
1364 struct strbuf path
= STRBUF_INIT
;
1368 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1370 strbuf_addstr(&path
, objdir
);
1371 strbuf_addstr(&path
, "/pack");
1372 dir
= opendir(path
.buf
);
1374 if (errno
!= ENOENT
)
1375 error_errno("unable to open object pack directory: %s",
1377 strbuf_release(&path
);
1380 strbuf_addch(&path
, '/');
1381 dirnamelen
= path
.len
;
1382 while ((de
= readdir(dir
)) != NULL
) {
1383 struct packed_git
*p
;
1386 if (is_dot_or_dotdot(de
->d_name
))
1389 strbuf_setlen(&path
, dirnamelen
);
1390 strbuf_addstr(&path
, de
->d_name
);
1392 base_len
= path
.len
;
1393 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1394 /* Don't reopen a pack we already have. */
1395 for (p
= packed_git
; p
; p
= p
->next
) {
1397 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1399 !memcmp(p
->pack_name
, path
.buf
, len
))
1404 * See if it really is a valid .idx file with
1405 * corresponding .pack file that we can map.
1407 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1408 install_packed_git(p
);
1411 if (!report_garbage
)
1414 if (ends_with(de
->d_name
, ".idx") ||
1415 ends_with(de
->d_name
, ".pack") ||
1416 ends_with(de
->d_name
, ".bitmap") ||
1417 ends_with(de
->d_name
, ".keep"))
1418 string_list_append(&garbage
, path
.buf
);
1420 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
1423 report_pack_garbage(&garbage
);
1424 string_list_clear(&garbage
, 0);
1425 strbuf_release(&path
);
1428 static int approximate_object_count_valid
;
1431 * Give a fast, rough count of the number of objects in the repository. This
1432 * ignores loose objects completely. If you have a lot of them, then either
1433 * you should repack because your performance will be awful, or they are
1434 * all unreachable objects about to be pruned, in which case they're not really
1435 * interesting as a measure of repo size in the first place.
1437 unsigned long approximate_object_count(void)
1439 static unsigned long count
;
1440 if (!approximate_object_count_valid
) {
1441 struct packed_git
*p
;
1443 prepare_packed_git();
1445 for (p
= packed_git
; p
; p
= p
->next
) {
1446 if (open_pack_index(p
))
1448 count
+= p
->num_objects
;
1454 static void *get_next_packed_git(const void *p
)
1456 return ((const struct packed_git
*)p
)->next
;
1459 static void set_next_packed_git(void *p
, void *next
)
1461 ((struct packed_git
*)p
)->next
= next
;
1464 static int sort_pack(const void *a_
, const void *b_
)
1466 const struct packed_git
*a
= a_
;
1467 const struct packed_git
*b
= b_
;
1471 * Local packs tend to contain objects specific to our
1472 * variant of the project than remote ones. In addition,
1473 * remote ones could be on a network mounted filesystem.
1474 * Favor local ones for these reasons.
1476 st
= a
->pack_local
- b
->pack_local
;
1481 * Younger packs tend to contain more recent objects,
1482 * and more recent objects tend to get accessed more
1485 if (a
->mtime
< b
->mtime
)
1487 else if (a
->mtime
== b
->mtime
)
1492 static void rearrange_packed_git(void)
1494 packed_git
= llist_mergesort(packed_git
, get_next_packed_git
,
1495 set_next_packed_git
, sort_pack
);
1498 static void prepare_packed_git_mru(void)
1500 struct packed_git
*p
;
1502 mru_clear(packed_git_mru
);
1503 for (p
= packed_git
; p
; p
= p
->next
)
1504 mru_append(packed_git_mru
, p
);
1507 static int prepare_packed_git_run_once
= 0;
1508 void prepare_packed_git(void)
1510 struct alternate_object_database
*alt
;
1512 if (prepare_packed_git_run_once
)
1514 prepare_packed_git_one(get_object_directory(), 1);
1516 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
)
1517 prepare_packed_git_one(alt
->path
, 0);
1518 rearrange_packed_git();
1519 prepare_packed_git_mru();
1520 prepare_packed_git_run_once
= 1;
1523 void reprepare_packed_git(void)
1525 approximate_object_count_valid
= 0;
1526 prepare_packed_git_run_once
= 0;
1527 prepare_packed_git();
1530 static void mark_bad_packed_object(struct packed_git
*p
,
1531 const unsigned char *sha1
)
1534 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1535 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
1537 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1538 st_mult(GIT_SHA1_RAWSZ
,
1539 st_add(p
->num_bad_objects
, 1)));
1540 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
1541 p
->num_bad_objects
++;
1544 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1546 struct packed_git
*p
;
1549 for (p
= packed_git
; p
; p
= p
->next
)
1550 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1551 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1557 * With an in-core object data in "map", rehash it to make sure the
1558 * object name actually matches "sha1" to detect object corruption.
1559 * With "map" == NULL, try reading the object named with "sha1" using
1560 * the streaming interface and rehash it to do the same.
1562 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1563 unsigned long size
, const char *type
)
1565 unsigned char real_sha1
[20];
1566 enum object_type obj_type
;
1567 struct git_istream
*st
;
1573 hash_sha1_file(map
, size
, type
, real_sha1
);
1574 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1577 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1581 /* Generate the header */
1582 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1586 git_SHA1_Update(&c
, hdr
, hdrlen
);
1588 char buf
[1024 * 16];
1589 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1597 git_SHA1_Update(&c
, buf
, readlen
);
1599 git_SHA1_Final(real_sha1
, &c
);
1601 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1604 int git_open(const char *name
)
1606 static int sha1_file_open_flag
= O_NOATIME
| O_CLOEXEC
;
1612 fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1616 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1617 if ((sha1_file_open_flag
& O_CLOEXEC
) && errno
== EINVAL
) {
1618 sha1_file_open_flag
&= ~O_CLOEXEC
;
1622 /* Might the failure be due to O_NOATIME? */
1623 if (errno
!= ENOENT
&& (sha1_file_open_flag
& O_NOATIME
)) {
1624 sha1_file_open_flag
&= ~O_NOATIME
;
1631 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1633 struct alternate_object_database
*alt
;
1635 if (!lstat(sha1_file_name(sha1
), st
))
1640 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1641 const char *path
= alt_sha1_path(alt
, sha1
);
1642 if (!lstat(path
, st
))
1649 static int open_sha1_file(const unsigned char *sha1
)
1652 struct alternate_object_database
*alt
;
1653 int most_interesting_errno
;
1655 fd
= git_open(sha1_file_name(sha1
));
1658 most_interesting_errno
= errno
;
1661 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1662 const char *path
= alt_sha1_path(alt
, sha1
);
1663 fd
= git_open(path
);
1666 if (most_interesting_errno
== ENOENT
)
1667 most_interesting_errno
= errno
;
1669 errno
= most_interesting_errno
;
1673 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1678 fd
= open_sha1_file(sha1
);
1683 if (!fstat(fd
, &st
)) {
1684 *size
= xsize_t(st
.st_size
);
1686 /* mmap() is forbidden on empty files */
1687 error("object file %s is empty", sha1_file_name(sha1
));
1690 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1697 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1698 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1701 unsigned long size
, c
;
1702 unsigned long used
= 0;
1705 *type
= (c
>> 4) & 7;
1709 if (len
<= used
|| bitsizeof(long) <= shift
) {
1710 error("bad object header");
1715 size
+= (c
& 0x7f) << shift
;
1722 static int unpack_sha1_short_header(git_zstream
*stream
,
1723 unsigned char *map
, unsigned long mapsize
,
1724 void *buffer
, unsigned long bufsiz
)
1726 /* Get the data stream */
1727 memset(stream
, 0, sizeof(*stream
));
1728 stream
->next_in
= map
;
1729 stream
->avail_in
= mapsize
;
1730 stream
->next_out
= buffer
;
1731 stream
->avail_out
= bufsiz
;
1733 git_inflate_init(stream
);
1734 return git_inflate(stream
, 0);
1737 int unpack_sha1_header(git_zstream
*stream
,
1738 unsigned char *map
, unsigned long mapsize
,
1739 void *buffer
, unsigned long bufsiz
)
1741 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
1747 /* Make sure we have the terminating NUL */
1748 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1753 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1754 unsigned long mapsize
, void *buffer
,
1755 unsigned long bufsiz
, struct strbuf
*header
)
1759 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1764 * Check if entire header is unpacked in the first iteration.
1766 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1770 * buffer[0..bufsiz] was not large enough. Copy the partial
1771 * result out to header, and then append the result of further
1772 * reading the stream.
1774 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1775 stream
->next_out
= buffer
;
1776 stream
->avail_out
= bufsiz
;
1779 status
= git_inflate(stream
, 0);
1780 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1781 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1783 stream
->next_out
= buffer
;
1784 stream
->avail_out
= bufsiz
;
1785 } while (status
!= Z_STREAM_END
);
1789 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1791 int bytes
= strlen(buffer
) + 1;
1792 unsigned char *buf
= xmallocz(size
);
1796 n
= stream
->total_out
- bytes
;
1799 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1801 if (bytes
<= size
) {
1803 * The above condition must be (bytes <= size), not
1804 * (bytes < size). In other words, even though we
1805 * expect no more output and set avail_out to zero,
1806 * the input zlib stream may have bytes that express
1807 * "this concludes the stream", and we *do* want to
1810 * Otherwise we would not be able to test that we
1811 * consumed all the input to reach the expected size;
1812 * we also want to check that zlib tells us that all
1813 * went well with status == Z_STREAM_END at the end.
1815 stream
->next_out
= buf
+ bytes
;
1816 stream
->avail_out
= size
- bytes
;
1817 while (status
== Z_OK
)
1818 status
= git_inflate(stream
, Z_FINISH
);
1820 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1821 git_inflate_end(stream
);
1826 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1827 else if (stream
->avail_in
)
1828 error("garbage at end of loose object '%s'",
1835 * We used to just use "sscanf()", but that's actually way
1836 * too permissive for what we want to check. So do an anal
1837 * object header parse by hand.
1839 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1842 const char *type_buf
= hdr
;
1844 int type
, type_len
= 0;
1847 * The type can be of any size but is followed by
1859 type
= type_from_string_gently(type_buf
, type_len
, 1);
1861 strbuf_add(oi
->typename
, type_buf
, type_len
);
1863 * Set type to 0 if its an unknown object and
1864 * we're obtaining the type using '--allow-unknown-type'
1867 if ((flags
& LOOKUP_UNKNOWN_OBJECT
) && (type
< 0))
1870 die("invalid object type");
1875 * The length must follow immediately, and be in canonical
1876 * decimal format (ie "010" is not valid).
1878 size
= *hdr
++ - '0';
1883 unsigned long c
= *hdr
- '0';
1887 size
= size
* 10 + c
;
1895 * The length must be followed by a zero byte
1897 return *hdr
? -1 : type
;
1900 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1902 struct object_info oi
= OBJECT_INFO_INIT
;
1905 return parse_sha1_header_extended(hdr
, &oi
, LOOKUP_REPLACE_OBJECT
);
1908 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1914 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1915 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1918 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1921 unsigned long get_size_from_delta(struct packed_git
*p
,
1922 struct pack_window
**w_curs
,
1925 const unsigned char *data
;
1926 unsigned char delta_head
[20], *in
;
1930 memset(&stream
, 0, sizeof(stream
));
1931 stream
.next_out
= delta_head
;
1932 stream
.avail_out
= sizeof(delta_head
);
1934 git_inflate_init(&stream
);
1936 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1937 stream
.next_in
= in
;
1938 st
= git_inflate(&stream
, Z_FINISH
);
1939 curpos
+= stream
.next_in
- in
;
1940 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1941 stream
.total_out
< sizeof(delta_head
));
1942 git_inflate_end(&stream
);
1943 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1944 error("delta data unpack-initial failed");
1948 /* Examine the initial part of the delta to figure out
1953 /* ignore base size */
1954 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1956 /* Read the result size */
1957 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1960 static off_t
get_delta_base(struct packed_git
*p
,
1961 struct pack_window
**w_curs
,
1963 enum object_type type
,
1964 off_t delta_obj_offset
)
1966 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1969 /* use_pack() assured us we have [base_info, base_info + 20)
1970 * as a range that we can look at without walking off the
1971 * end of the mapped window. Its actually the hash size
1972 * that is assured. An OFS_DELTA longer than the hash size
1973 * is stupid, as then a REF_DELTA would be smaller to store.
1975 if (type
== OBJ_OFS_DELTA
) {
1977 unsigned char c
= base_info
[used
++];
1978 base_offset
= c
& 127;
1981 if (!base_offset
|| MSB(base_offset
, 7))
1982 return 0; /* overflow */
1983 c
= base_info
[used
++];
1984 base_offset
= (base_offset
<< 7) + (c
& 127);
1986 base_offset
= delta_obj_offset
- base_offset
;
1987 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1988 return 0; /* out of bound */
1990 } else if (type
== OBJ_REF_DELTA
) {
1991 /* The base entry _must_ be in the same pack */
1992 base_offset
= find_pack_entry_one(base_info
, p
);
1995 die("I am totally screwed");
2000 * Like get_delta_base above, but we return the sha1 instead of the pack
2001 * offset. This means it is cheaper for REF deltas (we do not have to do
2002 * the final object lookup), but more expensive for OFS deltas (we
2003 * have to load the revidx to convert the offset back into a sha1).
2005 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
2006 struct pack_window
**w_curs
,
2008 enum object_type type
,
2009 off_t delta_obj_offset
)
2011 if (type
== OBJ_REF_DELTA
) {
2012 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
2014 } else if (type
== OBJ_OFS_DELTA
) {
2015 struct revindex_entry
*revidx
;
2016 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
2017 type
, delta_obj_offset
);
2022 revidx
= find_pack_revindex(p
, base_offset
);
2026 return nth_packed_object_sha1(p
, revidx
->nr
);
2031 int unpack_object_header(struct packed_git
*p
,
2032 struct pack_window
**w_curs
,
2034 unsigned long *sizep
)
2036 unsigned char *base
;
2039 enum object_type type
;
2041 /* use_pack() assures us we have [base, base + 20) available
2042 * as a range that we can look at. (Its actually the hash
2043 * size that is assured.) With our object header encoding
2044 * the maximum deflated object size is 2^137, which is just
2045 * insane, so we know won't exceed what we have been given.
2047 base
= use_pack(p
, w_curs
, *curpos
, &left
);
2048 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
2057 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
2060 struct revindex_entry
*revidx
;
2061 const unsigned char *sha1
;
2062 revidx
= find_pack_revindex(p
, obj_offset
);
2065 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2066 mark_bad_packed_object(p
, sha1
);
2067 type
= sha1_object_info(sha1
, NULL
);
2068 if (type
<= OBJ_NONE
)
2073 #define POI_STACK_PREALLOC 64
2075 static enum object_type
packed_to_object_type(struct packed_git
*p
,
2077 enum object_type type
,
2078 struct pack_window
**w_curs
,
2081 off_t small_poi_stack
[POI_STACK_PREALLOC
];
2082 off_t
*poi_stack
= small_poi_stack
;
2083 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
2085 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2088 /* Push the object we're going to leave behind */
2089 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
2090 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
2091 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
2092 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
2094 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
2096 poi_stack
[poi_stack_nr
++] = obj_offset
;
2097 /* If parsing the base offset fails, just unwind */
2098 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
2101 curpos
= obj_offset
= base_offset
;
2102 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
2103 if (type
<= OBJ_NONE
) {
2104 /* If getting the base itself fails, we first
2105 * retry the base, otherwise unwind */
2106 type
= retry_bad_packed_offset(p
, base_offset
);
2107 if (type
> OBJ_NONE
)
2121 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2122 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2127 if (poi_stack
!= small_poi_stack
)
2132 while (poi_stack_nr
) {
2133 obj_offset
= poi_stack
[--poi_stack_nr
];
2134 type
= retry_bad_packed_offset(p
, obj_offset
);
2135 if (type
> OBJ_NONE
)
2142 int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
2143 struct object_info
*oi
)
2145 struct pack_window
*w_curs
= NULL
;
2147 off_t curpos
= obj_offset
;
2148 enum object_type type
;
2151 * We always get the representation type, but only convert it to
2152 * a "real" type later if the caller is interested.
2154 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2157 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2158 off_t tmp_pos
= curpos
;
2159 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
2165 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2166 if (*oi
->sizep
== 0) {
2175 if (oi
->disk_sizep
) {
2176 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2177 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2181 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
2182 if (*oi
->typep
< 0) {
2188 if (oi
->delta_base_sha1
) {
2189 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2190 const unsigned char *base
;
2192 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2199 hashcpy(oi
->delta_base_sha1
, base
);
2201 hashclr(oi
->delta_base_sha1
);
2205 unuse_pack(&w_curs
);
2209 static void *unpack_compressed_entry(struct packed_git
*p
,
2210 struct pack_window
**w_curs
,
2216 unsigned char *buffer
, *in
;
2218 buffer
= xmallocz_gently(size
);
2221 memset(&stream
, 0, sizeof(stream
));
2222 stream
.next_out
= buffer
;
2223 stream
.avail_out
= size
+ 1;
2225 git_inflate_init(&stream
);
2227 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2228 stream
.next_in
= in
;
2229 st
= git_inflate(&stream
, Z_FINISH
);
2230 if (!stream
.avail_out
)
2231 break; /* the payload is larger than it should be */
2232 curpos
+= stream
.next_in
- in
;
2233 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2234 git_inflate_end(&stream
);
2235 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2243 static struct hashmap delta_base_cache
;
2244 static size_t delta_base_cached
;
2246 static LIST_HEAD(delta_base_cache_lru
);
2248 struct delta_base_cache_key
{
2249 struct packed_git
*p
;
2253 struct delta_base_cache_entry
{
2254 struct hashmap hash
;
2255 struct delta_base_cache_key key
;
2256 struct list_head lru
;
2259 enum object_type type
;
2262 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2266 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
2267 hash
+= (hash
>> 8) + (hash
>> 16);
2271 static struct delta_base_cache_entry
*
2272 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2274 struct hashmap_entry entry
;
2275 struct delta_base_cache_key key
;
2277 if (!delta_base_cache
.cmpfn
)
2280 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
2282 key
.base_offset
= base_offset
;
2283 return hashmap_get(&delta_base_cache
, &entry
, &key
);
2286 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
2287 const struct delta_base_cache_key
*b
)
2289 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
2292 static int delta_base_cache_hash_cmp(const void *va
, const void *vb
,
2295 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
2296 const struct delta_base_cache_key
*key
= vkey
;
2298 return !delta_base_cache_key_eq(&a
->key
, key
);
2300 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
2303 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2305 return !!get_delta_base_cache_entry(p
, base_offset
);
2309 * Remove the entry from the cache, but do _not_ free the associated
2310 * entry data. The caller takes ownership of the "data" buffer, and
2311 * should copy out any fields it wants before detaching.
2313 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2315 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
2316 list_del(&ent
->lru
);
2317 delta_base_cached
-= ent
->size
;
2321 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2322 unsigned long *base_size
, enum object_type
*type
)
2324 struct delta_base_cache_entry
*ent
;
2326 ent
= get_delta_base_cache_entry(p
, base_offset
);
2328 return unpack_entry(p
, base_offset
, type
, base_size
);
2331 *base_size
= ent
->size
;
2332 return xmemdupz(ent
->data
, ent
->size
);
2335 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2338 detach_delta_base_cache_entry(ent
);
2341 void clear_delta_base_cache(void)
2343 struct hashmap_iter iter
;
2344 struct delta_base_cache_entry
*entry
;
2345 for (entry
= hashmap_iter_first(&delta_base_cache
, &iter
);
2347 entry
= hashmap_iter_next(&iter
)) {
2348 release_delta_base_cache(entry
);
2352 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2353 void *base
, unsigned long base_size
, enum object_type type
)
2355 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
2356 struct list_head
*lru
, *tmp
;
2358 delta_base_cached
+= base_size
;
2360 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
2361 struct delta_base_cache_entry
*f
=
2362 list_entry(lru
, struct delta_base_cache_entry
, lru
);
2363 if (delta_base_cached
<= delta_base_cache_limit
)
2365 release_delta_base_cache(f
);
2369 ent
->key
.base_offset
= base_offset
;
2372 ent
->size
= base_size
;
2373 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
2375 if (!delta_base_cache
.cmpfn
)
2376 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, 0);
2377 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
2378 hashmap_add(&delta_base_cache
, ent
);
2381 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2382 unsigned long *size
);
2384 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2386 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2387 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2388 p
->pack_name
, (uintmax_t)obj_offset
);
2391 int do_check_packed_object_crc
;
2393 #define UNPACK_ENTRY_STACK_PREALLOC 64
2394 struct unpack_entry_stack_ent
{
2400 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2401 enum object_type
*final_type
, unsigned long *final_size
)
2403 struct pack_window
*w_curs
= NULL
;
2404 off_t curpos
= obj_offset
;
2407 enum object_type type
;
2408 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2409 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2410 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2411 int base_from_cache
= 0;
2413 write_pack_access_log(p
, obj_offset
);
2415 /* PHASE 1: drill down to the innermost base object */
2419 struct delta_base_cache_entry
*ent
;
2421 ent
= get_delta_base_cache_entry(p
, curpos
);
2426 detach_delta_base_cache_entry(ent
);
2427 base_from_cache
= 1;
2431 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2432 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2433 off_t len
= revidx
[1].offset
- obj_offset
;
2434 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2435 const unsigned char *sha1
=
2436 nth_packed_object_sha1(p
, revidx
->nr
);
2437 error("bad packed object CRC for %s",
2439 mark_bad_packed_object(p
, sha1
);
2440 unuse_pack(&w_curs
);
2445 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2446 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2449 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2451 error("failed to validate delta base reference "
2452 "at offset %"PRIuMAX
" from %s",
2453 (uintmax_t)curpos
, p
->pack_name
);
2454 /* bail to phase 2, in hopes of recovery */
2459 /* push object, proceed to base */
2460 if (delta_stack_nr
>= delta_stack_alloc
2461 && delta_stack
== small_delta_stack
) {
2462 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2463 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
2464 memcpy(delta_stack
, small_delta_stack
,
2465 sizeof(*delta_stack
)*delta_stack_nr
);
2467 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2469 i
= delta_stack_nr
++;
2470 delta_stack
[i
].obj_offset
= obj_offset
;
2471 delta_stack
[i
].curpos
= curpos
;
2472 delta_stack
[i
].size
= size
;
2474 curpos
= obj_offset
= base_offset
;
2477 /* PHASE 2: handle the base */
2482 die("BUG: unpack_entry: left loop at a valid delta");
2488 if (!base_from_cache
)
2489 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2493 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2494 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2497 /* PHASE 3: apply deltas in order */
2500 * 'data' holds the base data, or NULL if there was corruption
2502 while (delta_stack_nr
) {
2505 unsigned long delta_size
, base_size
= size
;
2511 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2515 * We're probably in deep shit, but let's try to fetch
2516 * the required base anyway from another pack or loose.
2517 * This is costly but should happen only in the presence
2518 * of a corrupted pack, and is better than failing outright.
2520 struct revindex_entry
*revidx
;
2521 const unsigned char *base_sha1
;
2522 revidx
= find_pack_revindex(p
, obj_offset
);
2524 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2525 error("failed to read delta base object %s"
2526 " at offset %"PRIuMAX
" from %s",
2527 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2529 mark_bad_packed_object(p
, base_sha1
);
2530 base
= read_object(base_sha1
, &type
, &base_size
);
2534 i
= --delta_stack_nr
;
2535 obj_offset
= delta_stack
[i
].obj_offset
;
2536 curpos
= delta_stack
[i
].curpos
;
2537 delta_size
= delta_stack
[i
].size
;
2542 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2545 error("failed to unpack compressed delta "
2546 "at offset %"PRIuMAX
" from %s",
2547 (uintmax_t)curpos
, p
->pack_name
);
2552 data
= patch_delta(base
, base_size
,
2553 delta_data
, delta_size
,
2557 * We could not apply the delta; warn the user, but keep going.
2558 * Our failure will be noticed either in the next iteration of
2559 * the loop, or if this is the final delta, in the caller when
2560 * we return NULL. Those code paths will take care of making
2561 * a more explicit warning and retrying with another copy of
2565 error("failed to apply delta");
2573 unuse_pack(&w_curs
);
2575 if (delta_stack
!= small_delta_stack
)
2581 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2584 const unsigned char *index
= p
->index_data
;
2586 if (open_pack_index(p
))
2588 index
= p
->index_data
;
2590 if (n
>= p
->num_objects
)
2593 if (p
->index_version
== 1) {
2594 return index
+ 24 * n
+ 4;
2597 return index
+ 20 * n
;
2601 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2603 const unsigned char *ptr
= vptr
;
2604 const unsigned char *start
= p
->index_data
;
2605 const unsigned char *end
= start
+ p
->index_size
;
2607 die(_("offset before start of pack index for %s (corrupt index?)"),
2609 /* No need to check for underflow; .idx files must be at least 8 bytes */
2611 die(_("offset beyond end of pack index for %s (truncated index?)"),
2615 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2617 const unsigned char *index
= p
->index_data
;
2619 if (p
->index_version
== 1) {
2620 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2623 index
+= 8 + p
->num_objects
* (20 + 4);
2624 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2625 if (!(off
& 0x80000000))
2627 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2628 check_pack_index_ptr(p
, index
);
2629 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2630 ntohl(*((uint32_t *)(index
+ 4)));
2634 off_t
find_pack_entry_one(const unsigned char *sha1
,
2635 struct packed_git
*p
)
2637 const uint32_t *level1_ofs
= p
->index_data
;
2638 const unsigned char *index
= p
->index_data
;
2639 unsigned hi
, lo
, stride
;
2640 static int use_lookup
= -1;
2641 static int debug_lookup
= -1;
2643 if (debug_lookup
< 0)
2644 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2647 if (open_pack_index(p
))
2649 level1_ofs
= p
->index_data
;
2650 index
= p
->index_data
;
2652 if (p
->index_version
> 1) {
2657 hi
= ntohl(level1_ofs
[*sha1
]);
2658 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2659 if (p
->index_version
> 1) {
2667 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2668 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2671 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2673 int pos
= sha1_entry_pos(index
, stride
, 0,
2674 lo
, hi
, p
->num_objects
, sha1
);
2677 return nth_packed_object_offset(p
, pos
);
2681 unsigned mi
= (lo
+ hi
) / 2;
2682 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2685 printf("lo %u hi %u rg %u mi %u\n",
2686 lo
, hi
, hi
- lo
, mi
);
2688 return nth_packed_object_offset(p
, mi
);
2697 int is_pack_valid(struct packed_git
*p
)
2699 /* An already open pack is known to be valid. */
2700 if (p
->pack_fd
!= -1)
2703 /* If the pack has one window completely covering the
2704 * file size, the pack is known to be valid even if
2705 * the descriptor is not currently open.
2708 struct pack_window
*w
= p
->windows
;
2710 if (!w
->offset
&& w
->len
== p
->pack_size
)
2714 /* Force the pack to open to prove its valid. */
2715 return !open_packed_git(p
);
2718 static int fill_pack_entry(const unsigned char *sha1
,
2719 struct pack_entry
*e
,
2720 struct packed_git
*p
)
2724 if (p
->num_bad_objects
) {
2726 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2727 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2731 offset
= find_pack_entry_one(sha1
, p
);
2736 * We are about to tell the caller where they can locate the
2737 * requested object. We better make sure the packfile is
2738 * still here and can be accessed before supplying that
2739 * answer, as it may have been deleted since the index was
2742 if (!is_pack_valid(p
))
2746 hashcpy(e
->sha1
, sha1
);
2751 * Iff a pack file contains the object named by sha1, return true and
2752 * store its location to e.
2754 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2756 struct mru_entry
*p
;
2758 prepare_packed_git();
2762 for (p
= packed_git_mru
->head
; p
; p
= p
->next
) {
2763 if (fill_pack_entry(sha1
, e
, p
->item
)) {
2764 mru_mark(packed_git_mru
, p
);
2771 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2772 struct packed_git
*packs
)
2774 struct packed_git
*p
;
2776 for (p
= packs
; p
; p
= p
->next
) {
2777 if (find_pack_entry_one(sha1
, p
))
2784 static int sha1_loose_object_info(const unsigned char *sha1
,
2785 struct object_info
*oi
,
2789 unsigned long mapsize
;
2793 struct strbuf hdrbuf
= STRBUF_INIT
;
2795 if (oi
->delta_base_sha1
)
2796 hashclr(oi
->delta_base_sha1
);
2799 * If we don't care about type or size, then we don't
2800 * need to look inside the object at all. Note that we
2801 * do not optimize out the stat call, even if the
2802 * caller doesn't care about the disk-size, since our
2803 * return value implicitly indicates whether the
2804 * object even exists.
2806 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
) {
2808 if (stat_sha1_file(sha1
, &st
) < 0)
2811 *oi
->disk_sizep
= st
.st_size
;
2815 map
= map_sha1_file(sha1
, &mapsize
);
2819 *oi
->disk_sizep
= mapsize
;
2820 if ((flags
& LOOKUP_UNKNOWN_OBJECT
)) {
2821 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2822 status
= error("unable to unpack %s header with --allow-unknown-type",
2824 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2825 status
= error("unable to unpack %s header",
2829 else if (hdrbuf
.len
) {
2830 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2831 status
= error("unable to parse %s header with --allow-unknown-type",
2833 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2834 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2835 git_inflate_end(&stream
);
2836 munmap(map
, mapsize
);
2837 if (status
&& oi
->typep
)
2838 *oi
->typep
= status
;
2839 strbuf_release(&hdrbuf
);
2843 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2845 struct cached_object
*co
;
2846 struct pack_entry e
;
2848 enum object_type real_type
;
2849 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2851 co
= find_cached_object(real
);
2854 *(oi
->typep
) = co
->type
;
2856 *(oi
->sizep
) = co
->size
;
2858 *(oi
->disk_sizep
) = 0;
2859 if (oi
->delta_base_sha1
)
2860 hashclr(oi
->delta_base_sha1
);
2862 strbuf_addstr(oi
->typename
, typename(co
->type
));
2863 oi
->whence
= OI_CACHED
;
2867 if (!find_pack_entry(real
, &e
)) {
2868 /* Most likely it's a loose object. */
2869 if (!sha1_loose_object_info(real
, oi
, flags
)) {
2870 oi
->whence
= OI_LOOSE
;
2874 /* Not a loose object; someone else may have just packed it. */
2875 reprepare_packed_git();
2876 if (!find_pack_entry(real
, &e
))
2881 * packed_object_info() does not follow the delta chain to
2882 * find out the real type, unless it is given oi->typep.
2884 if (oi
->typename
&& !oi
->typep
)
2885 oi
->typep
= &real_type
;
2887 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2889 mark_bad_packed_object(e
.p
, real
);
2890 if (oi
->typep
== &real_type
)
2892 return sha1_object_info_extended(real
, oi
, 0);
2893 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2894 oi
->whence
= OI_DBCACHED
;
2896 oi
->whence
= OI_PACKED
;
2897 oi
->u
.packed
.offset
= e
.offset
;
2898 oi
->u
.packed
.pack
= e
.p
;
2899 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2900 rtype
== OBJ_OFS_DELTA
);
2903 strbuf_addstr(oi
->typename
, typename(*oi
->typep
));
2904 if (oi
->typep
== &real_type
)
2910 /* returns enum object_type or negative */
2911 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2913 enum object_type type
;
2914 struct object_info oi
= OBJECT_INFO_INIT
;
2918 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2923 static void *read_packed_sha1(const unsigned char *sha1
,
2924 enum object_type
*type
, unsigned long *size
)
2926 struct pack_entry e
;
2929 if (!find_pack_entry(sha1
, &e
))
2931 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
);
2934 * We're probably in deep shit, but let's try to fetch
2935 * the required object anyway from another pack or loose.
2936 * This should happen only in the presence of a corrupted
2937 * pack, and is better than failing outright.
2939 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2940 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2941 mark_bad_packed_object(e
.p
, sha1
);
2942 data
= read_object(sha1
, type
, size
);
2947 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2948 unsigned char *sha1
)
2950 struct cached_object
*co
;
2952 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2953 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2955 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2956 co
= &cached_objects
[cached_object_nr
++];
2959 co
->buf
= xmalloc(len
);
2960 memcpy(co
->buf
, buf
, len
);
2961 hashcpy(co
->sha1
, sha1
);
2965 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2966 unsigned long *size
)
2968 unsigned long mapsize
;
2970 struct cached_object
*co
;
2972 co
= find_cached_object(sha1
);
2976 return xmemdupz(co
->buf
, co
->size
);
2979 buf
= read_packed_sha1(sha1
, type
, size
);
2982 map
= map_sha1_file(sha1
, &mapsize
);
2984 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2985 munmap(map
, mapsize
);
2988 reprepare_packed_git();
2989 return read_packed_sha1(sha1
, type
, size
);
2993 * This function dies on corrupt objects; the callers who want to
2994 * deal with them should arrange to call read_object() and give error
2995 * messages themselves.
2997 void *read_sha1_file_extended(const unsigned char *sha1
,
2998 enum object_type
*type
,
2999 unsigned long *size
,
3003 const struct packed_git
*p
;
3004 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
3007 data
= read_object(repl
, type
, size
);
3011 if (errno
&& errno
!= ENOENT
)
3012 die_errno("failed to read object %s", sha1_to_hex(sha1
));
3014 /* die if we replaced an object with one that does not exist */
3016 die("replacement %s not found for %s",
3017 sha1_to_hex(repl
), sha1_to_hex(sha1
));
3019 if (has_loose_object(repl
)) {
3020 const char *path
= sha1_file_name(sha1
);
3022 die("loose object %s (stored in %s) is corrupt",
3023 sha1_to_hex(repl
), path
);
3026 if ((p
= has_packed_and_bad(repl
)) != NULL
)
3027 die("packed object %s (stored in %s) is corrupt",
3028 sha1_to_hex(repl
), p
->pack_name
);
3033 void *read_object_with_reference(const unsigned char *sha1
,
3034 const char *required_type_name
,
3035 unsigned long *size
,
3036 unsigned char *actual_sha1_return
)
3038 enum object_type type
, required_type
;
3040 unsigned long isize
;
3041 unsigned char actual_sha1
[20];
3043 required_type
= type_from_string(required_type_name
);
3044 hashcpy(actual_sha1
, sha1
);
3046 int ref_length
= -1;
3047 const char *ref_type
= NULL
;
3049 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
3052 if (type
== required_type
) {
3054 if (actual_sha1_return
)
3055 hashcpy(actual_sha1_return
, actual_sha1
);
3058 /* Handle references */
3059 else if (type
== OBJ_COMMIT
)
3061 else if (type
== OBJ_TAG
)
3062 ref_type
= "object ";
3067 ref_length
= strlen(ref_type
);
3069 if (ref_length
+ 40 > isize
||
3070 memcmp(buffer
, ref_type
, ref_length
) ||
3071 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
3076 /* Now we have the ID of the referred-to object in
3077 * actual_sha1. Check again. */
3081 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
3082 const char *type
, unsigned char *sha1
,
3083 char *hdr
, int *hdrlen
)
3087 /* Generate the header */
3088 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
3092 git_SHA1_Update(&c
, hdr
, *hdrlen
);
3093 git_SHA1_Update(&c
, buf
, len
);
3094 git_SHA1_Final(sha1
, &c
);
3098 * Move the just written object into its final resting place.
3100 int finalize_object_file(const char *tmpfile
, const char *filename
)
3104 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
3106 else if (link(tmpfile
, filename
))
3110 * Coda hack - coda doesn't like cross-directory links,
3111 * so we fall back to a rename, which will mean that it
3112 * won't be able to check collisions, but that's not a
3115 * The same holds for FAT formatted media.
3117 * When this succeeds, we just return. We have nothing
3120 if (ret
&& ret
!= EEXIST
) {
3122 if (!rename(tmpfile
, filename
))
3126 unlink_or_warn(tmpfile
);
3128 if (ret
!= EEXIST
) {
3129 return error_errno("unable to write sha1 filename %s", filename
);
3131 /* FIXME!!! Collision check here ? */
3135 if (adjust_shared_perm(filename
))
3136 return error("unable to set permission to '%s'", filename
);
3140 static int write_buffer(int fd
, const void *buf
, size_t len
)
3142 if (write_in_full(fd
, buf
, len
) < 0)
3143 return error_errno("file write error");
3147 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
3148 unsigned char *sha1
)
3151 int hdrlen
= sizeof(hdr
);
3152 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3156 /* Finalize a file on disk, and close it. */
3157 static void close_sha1_file(int fd
)
3159 if (fsync_object_files
)
3160 fsync_or_die(fd
, "sha1 file");
3162 die_errno("error when closing sha1 file");
3165 /* Size of directory component, including the ending '/' */
3166 static inline int directory_size(const char *filename
)
3168 const char *s
= strrchr(filename
, '/');
3171 return s
- filename
+ 1;
3175 * This creates a temporary file in the same directory as the final
3178 * We want to avoid cross-directory filename renames, because those
3179 * can have problems on various filesystems (FAT, NFS, Coda).
3181 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
3183 int fd
, dirlen
= directory_size(filename
);
3186 strbuf_add(tmp
, filename
, dirlen
);
3187 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
3188 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3189 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3191 * Make sure the directory exists; note that the contents
3192 * of the buffer are undefined after mkstemp returns an
3193 * error, so we have to rewrite the whole buffer from
3197 strbuf_add(tmp
, filename
, dirlen
- 1);
3198 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
3200 if (adjust_shared_perm(tmp
->buf
))
3204 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
3205 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3210 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3211 const void *buf
, unsigned long len
, time_t mtime
)
3214 unsigned char compressed
[4096];
3217 unsigned char parano_sha1
[20];
3218 static struct strbuf tmp_file
= STRBUF_INIT
;
3219 const char *filename
= sha1_file_name(sha1
);
3221 fd
= create_tmpfile(&tmp_file
, filename
);
3223 if (errno
== EACCES
)
3224 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3226 return error_errno("unable to create temporary file");
3230 git_deflate_init(&stream
, zlib_compression_level
);
3231 stream
.next_out
= compressed
;
3232 stream
.avail_out
= sizeof(compressed
);
3235 /* First header.. */
3236 stream
.next_in
= (unsigned char *)hdr
;
3237 stream
.avail_in
= hdrlen
;
3238 while (git_deflate(&stream
, 0) == Z_OK
)
3240 git_SHA1_Update(&c
, hdr
, hdrlen
);
3242 /* Then the data itself.. */
3243 stream
.next_in
= (void *)buf
;
3244 stream
.avail_in
= len
;
3246 unsigned char *in0
= stream
.next_in
;
3247 ret
= git_deflate(&stream
, Z_FINISH
);
3248 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3249 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3250 die("unable to write sha1 file");
3251 stream
.next_out
= compressed
;
3252 stream
.avail_out
= sizeof(compressed
);
3253 } while (ret
== Z_OK
);
3255 if (ret
!= Z_STREAM_END
)
3256 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3257 ret
= git_deflate_end_gently(&stream
);
3259 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3260 git_SHA1_Final(parano_sha1
, &c
);
3261 if (hashcmp(sha1
, parano_sha1
) != 0)
3262 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3264 close_sha1_file(fd
);
3269 utb
.modtime
= mtime
;
3270 if (utime(tmp_file
.buf
, &utb
) < 0)
3271 warning_errno("failed utime() on %s", tmp_file
.buf
);
3274 return finalize_object_file(tmp_file
.buf
, filename
);
3277 static int freshen_loose_object(const unsigned char *sha1
)
3279 return check_and_freshen(sha1
, 1);
3282 static int freshen_packed_object(const unsigned char *sha1
)
3284 struct pack_entry e
;
3285 if (!find_pack_entry(sha1
, &e
))
3289 if (!freshen_file(e
.p
->pack_name
))
3295 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3298 int hdrlen
= sizeof(hdr
);
3300 /* Normally if we have it in the pack then we do not bother writing
3301 * it out into .git/objects/??/?{38} file.
3303 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3304 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3306 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3309 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3310 unsigned char *sha1
, unsigned flags
)
3313 int hdrlen
, status
= 0;
3315 /* type string, SP, %lu of the length plus NUL must fit this */
3316 hdrlen
= strlen(type
) + 32;
3317 header
= xmalloc(hdrlen
);
3318 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3320 if (!(flags
& HASH_WRITE_OBJECT
))
3322 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3324 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3331 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3335 enum object_type type
;
3340 if (has_loose_object(sha1
))
3342 buf
= read_packed_sha1(sha1
, &type
, &len
);
3344 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3345 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3346 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3352 int has_pack_index(const unsigned char *sha1
)
3355 if (stat(sha1_pack_index_name(sha1
), &st
))
3360 int has_sha1_pack(const unsigned char *sha1
)
3362 struct pack_entry e
;
3363 return find_pack_entry(sha1
, &e
);
3366 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3368 struct pack_entry e
;
3370 if (find_pack_entry(sha1
, &e
))
3372 if (has_loose_object(sha1
))
3374 if (flags
& HAS_SHA1_QUICK
)
3376 reprepare_packed_git();
3377 return find_pack_entry(sha1
, &e
);
3380 int has_object_file(const struct object_id
*oid
)
3382 return has_sha1_file(oid
->hash
);
3385 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
3387 return has_sha1_file_with_flags(oid
->hash
, flags
);
3390 static void check_tree(const void *buf
, size_t size
)
3392 struct tree_desc desc
;
3393 struct name_entry entry
;
3395 init_tree_desc(&desc
, buf
, size
);
3396 while (tree_entry(&desc
, &entry
))
3398 * tree_entry() will die() on malformed entries */
3402 static void check_commit(const void *buf
, size_t size
)
3405 memset(&c
, 0, sizeof(c
));
3406 if (parse_commit_buffer(&c
, buf
, size
))
3407 die("corrupt commit");
3410 static void check_tag(const void *buf
, size_t size
)
3413 memset(&t
, 0, sizeof(t
));
3414 if (parse_tag_buffer(&t
, buf
, size
))
3418 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3419 enum object_type type
,
3420 const char *path
, unsigned flags
)
3422 int ret
, re_allocated
= 0;
3423 int write_object
= flags
& HASH_WRITE_OBJECT
;
3429 * Convert blobs to git internal format
3431 if ((type
== OBJ_BLOB
) && path
) {
3432 struct strbuf nbuf
= STRBUF_INIT
;
3433 if (convert_to_git(path
, buf
, size
, &nbuf
,
3434 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3435 buf
= strbuf_detach(&nbuf
, &size
);
3439 if (flags
& HASH_FORMAT_CHECK
) {
3440 if (type
== OBJ_TREE
)
3441 check_tree(buf
, size
);
3442 if (type
== OBJ_COMMIT
)
3443 check_commit(buf
, size
);
3444 if (type
== OBJ_TAG
)
3445 check_tag(buf
, size
);
3449 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3451 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3457 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3458 const char *path
, unsigned flags
)
3461 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3462 struct strbuf sbuf
= STRBUF_INIT
;
3465 assert(would_convert_to_git_filter_fd(path
));
3467 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3468 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3471 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3474 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3476 strbuf_release(&sbuf
);
3480 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3481 const char *path
, unsigned flags
)
3483 struct strbuf sbuf
= STRBUF_INIT
;
3486 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3487 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3490 strbuf_release(&sbuf
);
3494 #define SMALL_FILE_SIZE (32*1024)
3496 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3497 enum object_type type
, const char *path
,
3503 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3504 } else if (size
<= SMALL_FILE_SIZE
) {
3505 char *buf
= xmalloc(size
);
3506 if (size
== read_in_full(fd
, buf
, size
))
3507 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3509 ret
= error_errno("short read");
3512 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3513 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3520 * This creates one packfile per large blob unless bulk-checkin
3521 * machinery is "plugged".
3523 * This also bypasses the usual "convert-to-git" dance, and that is on
3524 * purpose. We could write a streaming version of the converting
3525 * functions and insert that before feeding the data to fast-import
3526 * (or equivalent in-core API described above). However, that is
3527 * somewhat complicated, as we do not know the size of the filter
3528 * result, which we need to know beforehand when writing a git object.
3529 * Since the primary motivation for trying to stream from the working
3530 * tree file and to avoid mmaping it in core is to deal with large
3531 * binary blobs, they generally do not want to get any conversion, and
3532 * callers should avoid this code path when filters are requested.
3534 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3535 enum object_type type
, const char *path
,
3538 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3541 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3542 enum object_type type
, const char *path
, unsigned flags
)
3547 * Call xsize_t() only when needed to avoid potentially unnecessary
3548 * die() for large files.
3550 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3551 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3552 else if (!S_ISREG(st
->st_mode
))
3553 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3554 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3555 (path
&& would_convert_to_git(path
)))
3556 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3559 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3565 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3568 struct strbuf sb
= STRBUF_INIT
;
3570 switch (st
->st_mode
& S_IFMT
) {
3572 fd
= open(path
, O_RDONLY
);
3574 return error_errno("open(\"%s\")", path
);
3575 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3576 return error("%s: failed to insert into database",
3580 if (strbuf_readlink(&sb
, path
, st
->st_size
))
3581 return error_errno("readlink(\"%s\")", path
);
3582 if (!(flags
& HASH_WRITE_OBJECT
))
3583 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3584 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3585 return error("%s: failed to insert into database",
3587 strbuf_release(&sb
);
3590 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3592 return error("%s: unsupported file type", path
);
3597 int read_pack_header(int fd
, struct pack_header
*header
)
3599 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3600 /* "eof before pack header was fully read" */
3601 return PH_ERROR_EOF
;
3603 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3604 /* "protocol error (pack signature mismatch detected)" */
3605 return PH_ERROR_PACK_SIGNATURE
;
3606 if (!pack_version_ok(header
->hdr_version
))
3607 /* "protocol error (pack version unsupported)" */
3608 return PH_ERROR_PROTOCOL
;
3612 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3614 enum object_type type
= sha1_object_info(sha1
, NULL
);
3616 die("%s is not a valid object", sha1_to_hex(sha1
));
3618 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3622 static int for_each_file_in_obj_subdir(int subdir_nr
,
3623 struct strbuf
*path
,
3624 each_loose_object_fn obj_cb
,
3625 each_loose_cruft_fn cruft_cb
,
3626 each_loose_subdir_fn subdir_cb
,
3629 size_t baselen
= path
->len
;
3630 DIR *dir
= opendir(path
->buf
);
3635 if (errno
== ENOENT
)
3637 return error_errno("unable to open %s", path
->buf
);
3640 while ((de
= readdir(dir
))) {
3641 if (is_dot_or_dotdot(de
->d_name
))
3644 strbuf_setlen(path
, baselen
);
3645 strbuf_addf(path
, "/%s", de
->d_name
);
3647 if (strlen(de
->d_name
) == 38) {
3649 unsigned char sha1
[20];
3651 snprintf(hex
, sizeof(hex
), "%02x%s",
3652 subdir_nr
, de
->d_name
);
3653 if (!get_sha1_hex(hex
, sha1
)) {
3655 r
= obj_cb(sha1
, path
->buf
, data
);
3664 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3671 strbuf_setlen(path
, baselen
);
3672 if (!r
&& subdir_cb
)
3673 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3678 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3679 each_loose_object_fn obj_cb
,
3680 each_loose_cruft_fn cruft_cb
,
3681 each_loose_subdir_fn subdir_cb
,
3684 size_t baselen
= path
->len
;
3688 for (i
= 0; i
< 256; i
++) {
3689 strbuf_addf(path
, "/%02x", i
);
3690 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3692 strbuf_setlen(path
, baselen
);
3700 int for_each_loose_file_in_objdir(const char *path
,
3701 each_loose_object_fn obj_cb
,
3702 each_loose_cruft_fn cruft_cb
,
3703 each_loose_subdir_fn subdir_cb
,
3706 struct strbuf buf
= STRBUF_INIT
;
3709 strbuf_addstr(&buf
, path
);
3710 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3712 strbuf_release(&buf
);
3717 struct loose_alt_odb_data
{
3718 each_loose_object_fn
*cb
;
3722 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3725 struct loose_alt_odb_data
*data
= vdata
;
3726 struct strbuf buf
= STRBUF_INIT
;
3729 strbuf_addstr(&buf
, alt
->path
);
3730 r
= for_each_loose_file_in_objdir_buf(&buf
,
3731 data
->cb
, NULL
, NULL
,
3733 strbuf_release(&buf
);
3737 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3739 struct loose_alt_odb_data alt
;
3742 r
= for_each_loose_file_in_objdir(get_object_directory(),
3743 cb
, NULL
, NULL
, data
);
3747 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3752 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3755 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3760 for (i
= 0; i
< p
->num_objects
; i
++) {
3761 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3764 return error("unable to get sha1 of object %u in %s",
3767 r
= cb(sha1
, p
, i
, data
);
3774 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3776 struct packed_git
*p
;
3778 int pack_errors
= 0;
3780 prepare_packed_git();
3781 for (p
= packed_git
; p
; p
= p
->next
) {
3782 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3784 if (open_pack_index(p
)) {
3788 r
= for_each_object_in_pack(p
, cb
, data
);
3792 return r
? r
: pack_errors
;