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
)
170 /* path points to cache entries, so xstrdup before messing with it */
171 char *buf
= xstrdup(path
);
172 enum scld_error result
= safe_create_leading_directories(buf
);
180 static void fill_sha1_path(struct strbuf
*buf
, const unsigned char *sha1
)
183 for (i
= 0; i
< 20; i
++) {
184 static char hex
[] = "0123456789abcdef";
185 unsigned int val
= sha1
[i
];
186 strbuf_addch(buf
, hex
[val
>> 4]);
187 strbuf_addch(buf
, hex
[val
& 0xf]);
189 strbuf_addch(buf
, '/');
193 const char *sha1_file_name(const unsigned char *sha1
)
195 static struct strbuf buf
= STRBUF_INIT
;
198 strbuf_addf(&buf
, "%s/", get_object_directory());
200 fill_sha1_path(&buf
, sha1
);
204 struct strbuf
*alt_scratch_buf(struct alternate_object_database
*alt
)
206 strbuf_setlen(&alt
->scratch
, alt
->base_len
);
207 return &alt
->scratch
;
210 static const char *alt_sha1_path(struct alternate_object_database
*alt
,
211 const unsigned char *sha1
)
213 struct strbuf
*buf
= alt_scratch_buf(alt
);
214 fill_sha1_path(buf
, sha1
);
219 * Return the name of the pack or index file with the specified sha1
220 * in its filename. *base and *name are scratch space that must be
221 * provided by the caller. which should be "pack" or "idx".
223 static char *sha1_get_pack_name(const unsigned char *sha1
,
228 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
229 sha1_to_hex(sha1
), which
);
233 char *sha1_pack_name(const unsigned char *sha1
)
235 static struct strbuf buf
= STRBUF_INIT
;
236 return sha1_get_pack_name(sha1
, &buf
, "pack");
239 char *sha1_pack_index_name(const unsigned char *sha1
)
241 static struct strbuf buf
= STRBUF_INIT
;
242 return sha1_get_pack_name(sha1
, &buf
, "idx");
245 struct alternate_object_database
*alt_odb_list
;
246 static struct alternate_object_database
**alt_odb_tail
;
249 * Return non-zero iff the path is usable as an alternate object database.
251 static int alt_odb_usable(struct strbuf
*path
, const char *normalized_objdir
)
253 struct alternate_object_database
*alt
;
255 /* Detect cases where alternate disappeared */
256 if (!is_directory(path
->buf
)) {
257 error("object directory %s does not exist; "
258 "check .git/objects/info/alternates.",
264 * Prevent the common mistake of listing the same
265 * thing twice, or object directory itself.
267 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
268 if (!fspathcmp(path
->buf
, alt
->path
))
271 if (!fspathcmp(path
->buf
, normalized_objdir
))
278 * Prepare alternate object database registry.
280 * The variable alt_odb_list points at the list of struct
281 * alternate_object_database. The elements on this list come from
282 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
283 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
284 * whose contents is similar to that environment variable but can be
285 * LF separated. Its base points at a statically allocated buffer that
286 * contains "/the/directory/corresponding/to/.git/objects/...", while
287 * its name points just after the slash at the end of ".git/objects/"
288 * in the example above, and has enough space to hold 40-byte hex
289 * SHA1, an extra slash for the first level indirection, and the
292 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
293 int depth
, const char *normalized_objdir
)
295 struct alternate_object_database
*ent
;
296 struct strbuf pathbuf
= STRBUF_INIT
;
298 if (!is_absolute_path(entry
) && relative_base
) {
299 strbuf_addstr(&pathbuf
, real_path(relative_base
));
300 strbuf_addch(&pathbuf
, '/');
302 strbuf_addstr(&pathbuf
, entry
);
304 if (strbuf_normalize_path(&pathbuf
) < 0 && relative_base
) {
305 error("unable to normalize alternate object path: %s",
307 strbuf_release(&pathbuf
);
312 * The trailing slash after the directory name is given by
313 * this function at the end. Remove duplicates.
315 while (pathbuf
.len
&& pathbuf
.buf
[pathbuf
.len
- 1] == '/')
316 strbuf_setlen(&pathbuf
, pathbuf
.len
- 1);
318 if (!alt_odb_usable(&pathbuf
, normalized_objdir
)) {
319 strbuf_release(&pathbuf
);
323 ent
= alloc_alt_odb(pathbuf
.buf
);
325 /* add the alternate entry */
327 alt_odb_tail
= &(ent
->next
);
330 /* recursively add alternates */
331 read_info_alternates(pathbuf
.buf
, depth
+ 1);
333 strbuf_release(&pathbuf
);
337 static const char *parse_alt_odb_entry(const char *string
,
345 if (*string
== '#') {
346 /* comment; consume up to next separator */
347 end
= strchrnul(string
, sep
);
348 } else if (*string
== '"' && !unquote_c_style(out
, string
, &end
)) {
350 * quoted path; unquote_c_style has copied the
351 * data for us and set "end". Broken quoting (e.g.,
352 * an entry that doesn't end with a quote) falls
353 * back to the unquoted case below.
356 /* normal, unquoted path */
357 end
= strchrnul(string
, sep
);
358 strbuf_add(out
, string
, end
- string
);
366 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
367 const char *relative_base
, int depth
)
369 struct strbuf objdirbuf
= STRBUF_INIT
;
370 struct strbuf entry
= STRBUF_INIT
;
373 error("%s: ignoring alternate object stores, nesting too deep.",
378 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
379 if (strbuf_normalize_path(&objdirbuf
) < 0)
380 die("unable to normalize object directory: %s",
384 alt
= parse_alt_odb_entry(alt
, sep
, &entry
);
387 link_alt_odb_entry(entry
.buf
, relative_base
, depth
, objdirbuf
.buf
);
389 strbuf_release(&entry
);
390 strbuf_release(&objdirbuf
);
393 void read_info_alternates(const char * relative_base
, int depth
)
401 path
= xstrfmt("%s/info/alternates", relative_base
);
406 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
410 mapsz
= xsize_t(st
.st_size
);
411 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
414 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
419 struct alternate_object_database
*alloc_alt_odb(const char *dir
)
421 struct alternate_object_database
*ent
;
423 FLEX_ALLOC_STR(ent
, path
, dir
);
424 strbuf_init(&ent
->scratch
, 0);
425 strbuf_addf(&ent
->scratch
, "%s/", dir
);
426 ent
->base_len
= ent
->scratch
.len
;
431 void add_to_alternates_file(const char *reference
)
433 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
434 char *alts
= git_pathdup("objects/info/alternates");
437 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
438 out
= fdopen_lock_file(lock
, "w");
440 die_errno("unable to fdopen alternates lockfile");
442 in
= fopen(alts
, "r");
444 struct strbuf line
= STRBUF_INIT
;
447 while (strbuf_getline(&line
, in
) != EOF
) {
448 if (!strcmp(reference
, line
.buf
)) {
452 fprintf_or_die(out
, "%s\n", line
.buf
);
455 strbuf_release(&line
);
459 rollback_lock_file(lock
);
463 else if (errno
!= ENOENT
)
464 die_errno("unable to read alternates file");
467 fprintf_or_die(out
, "%s\n", reference
);
468 if (commit_lock_file(lock
))
469 die_errno("unable to move new alternates file into place");
471 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
476 void add_to_alternates_memory(const char *reference
)
479 * Make sure alternates are initialized, or else our entry may be
480 * overwritten when they are.
484 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
488 * Compute the exact path an alternate is at and returns it. In case of
489 * error NULL is returned and the human readable error is added to `err`
490 * `path` may be relative and should point to $GITDIR.
491 * `err` must not be null.
493 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
495 char *ref_git
= NULL
;
496 const char *repo
, *ref_git_s
;
499 ref_git_s
= real_path_if_valid(path
);
502 strbuf_addf(err
, _("path '%s' does not exist"), path
);
506 * Beware: read_gitfile(), real_path() and mkpath()
507 * return static buffer
509 ref_git
= xstrdup(ref_git_s
);
511 repo
= read_gitfile(ref_git
);
513 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
516 ref_git
= xstrdup(repo
);
519 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
520 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
522 ref_git
= ref_git_git
;
523 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
524 struct strbuf sb
= STRBUF_INIT
;
526 if (get_common_dir(&sb
, ref_git
)) {
528 _("reference repository '%s' as a linked "
529 "checkout is not supported yet."),
534 strbuf_addf(err
, _("reference repository '%s' is not a "
535 "local repository."), path
);
539 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
540 strbuf_addf(err
, _("reference repository '%s' is shallow"),
546 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
548 _("reference repository '%s' is grafted"),
563 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
565 struct alternate_object_database
*ent
;
569 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
577 void prepare_alt_odb(void)
584 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
587 alt_odb_tail
= &alt_odb_list
;
588 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
590 read_info_alternates(get_object_directory(), 0);
593 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
594 static int freshen_file(const char *fn
)
597 t
.actime
= t
.modtime
= time(NULL
);
598 return !utime(fn
, &t
);
602 * All of the check_and_freshen functions return 1 if the file exists and was
603 * freshened (if freshening was requested), 0 otherwise. If they return
604 * 0, you should not assume that it is safe to skip a write of the object (it
605 * either does not exist on disk, or has a stale mtime and may be subject to
608 static int check_and_freshen_file(const char *fn
, int freshen
)
610 if (access(fn
, F_OK
))
612 if (freshen
&& !freshen_file(fn
))
617 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
619 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
622 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
624 struct alternate_object_database
*alt
;
626 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
627 const char *path
= alt_sha1_path(alt
, sha1
);
628 if (check_and_freshen_file(path
, freshen
))
634 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
636 return check_and_freshen_local(sha1
, freshen
) ||
637 check_and_freshen_nonlocal(sha1
, freshen
);
640 int has_loose_object_nonlocal(const unsigned char *sha1
)
642 return check_and_freshen_nonlocal(sha1
, 0);
645 static int has_loose_object(const unsigned char *sha1
)
647 return check_and_freshen(sha1
, 0);
650 static unsigned int pack_used_ctr
;
651 static unsigned int pack_mmap_calls
;
652 static unsigned int peak_pack_open_windows
;
653 static unsigned int pack_open_windows
;
654 static unsigned int pack_open_fds
;
655 static unsigned int pack_max_fds
;
656 static size_t peak_pack_mapped
;
657 static size_t pack_mapped
;
658 struct packed_git
*packed_git
;
660 static struct mru packed_git_mru_storage
;
661 struct mru
*packed_git_mru
= &packed_git_mru_storage
;
663 void pack_report(void)
666 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
667 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
668 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
669 sz_fmt(getpagesize()),
670 sz_fmt(packed_git_window_size
),
671 sz_fmt(packed_git_limit
));
673 "pack_report: pack_used_ctr = %10u\n"
674 "pack_report: pack_mmap_calls = %10u\n"
675 "pack_report: pack_open_windows = %10u / %10u\n"
676 "pack_report: pack_mapped = "
677 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
680 pack_open_windows
, peak_pack_open_windows
,
681 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
685 * Open and mmap the index file at path, perform a couple of
686 * consistency checks, then record its information to p. Return 0 on
689 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
692 struct pack_idx_header
*hdr
;
694 uint32_t version
, nr
, i
, *index
;
695 int fd
= git_open(path
);
700 if (fstat(fd
, &st
)) {
704 idx_size
= xsize_t(st
.st_size
);
705 if (idx_size
< 4 * 256 + 20 + 20) {
707 return error("index file %s is too small", path
);
709 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
713 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
714 version
= ntohl(hdr
->idx_version
);
715 if (version
< 2 || version
> 2) {
716 munmap(idx_map
, idx_size
);
717 return error("index file %s is version %"PRIu32
718 " and is not supported by this binary"
719 " (try upgrading GIT to a newer version)",
728 index
+= 2; /* skip index header */
729 for (i
= 0; i
< 256; i
++) {
730 uint32_t n
= ntohl(index
[i
]);
732 munmap(idx_map
, idx_size
);
733 return error("non-monotonic index %s", path
);
741 * - 256 index entries 4 bytes each
742 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
743 * - 20-byte SHA1 of the packfile
744 * - 20-byte SHA1 file checksum
746 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
747 munmap(idx_map
, idx_size
);
748 return error("wrong index v1 file size in %s", path
);
750 } else if (version
== 2) {
753 * - 8 bytes of header
754 * - 256 index entries 4 bytes each
755 * - 20-byte sha1 entry * nr
756 * - 4-byte crc entry * nr
757 * - 4-byte offset entry * nr
758 * - 20-byte SHA1 of the packfile
759 * - 20-byte SHA1 file checksum
760 * And after the 4-byte offset table might be a
761 * variable sized table containing 8-byte entries
762 * for offsets larger than 2^31.
764 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
765 unsigned long max_size
= min_size
;
767 max_size
+= (nr
- 1)*8;
768 if (idx_size
< min_size
|| idx_size
> max_size
) {
769 munmap(idx_map
, idx_size
);
770 return error("wrong index v2 file size in %s", path
);
772 if (idx_size
!= min_size
&&
774 * make sure we can deal with large pack offsets.
775 * 31-bit signed offset won't be enough, neither
776 * 32-bit unsigned one will be.
778 (sizeof(off_t
) <= 4)) {
779 munmap(idx_map
, idx_size
);
780 return error("pack too large for current definition of off_t in %s", path
);
784 p
->index_version
= version
;
785 p
->index_data
= idx_map
;
786 p
->index_size
= idx_size
;
791 int open_pack_index(struct packed_git
*p
)
800 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
801 die("BUG: pack_name does not end in .pack");
802 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
803 ret
= check_packed_git_idx(idx_name
, p
);
808 static void scan_windows(struct packed_git
*p
,
809 struct packed_git
**lru_p
,
810 struct pack_window
**lru_w
,
811 struct pack_window
**lru_l
)
813 struct pack_window
*w
, *w_l
;
815 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
817 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
827 static int unuse_one_window(struct packed_git
*current
)
829 struct packed_git
*p
, *lru_p
= NULL
;
830 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
833 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
834 for (p
= packed_git
; p
; p
= p
->next
)
835 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
837 munmap(lru_w
->base
, lru_w
->len
);
838 pack_mapped
-= lru_w
->len
;
840 lru_l
->next
= lru_w
->next
;
842 lru_p
->windows
= lru_w
->next
;
850 void release_pack_memory(size_t need
)
852 size_t cur
= pack_mapped
;
853 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
857 static void mmap_limit_check(size_t length
)
859 static size_t limit
= 0;
861 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
866 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
867 (uintmax_t)length
, (uintmax_t)limit
);
870 void *xmmap_gently(void *start
, size_t length
,
871 int prot
, int flags
, int fd
, off_t offset
)
875 mmap_limit_check(length
);
876 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
877 if (ret
== MAP_FAILED
) {
880 release_pack_memory(length
);
881 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
886 void *xmmap(void *start
, size_t length
,
887 int prot
, int flags
, int fd
, off_t offset
)
889 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
890 if (ret
== MAP_FAILED
)
891 die_errno("mmap failed");
895 void close_pack_windows(struct packed_git
*p
)
898 struct pack_window
*w
= p
->windows
;
901 die("pack '%s' still has open windows to it",
903 munmap(w
->base
, w
->len
);
904 pack_mapped
-= w
->len
;
906 p
->windows
= w
->next
;
911 static int close_pack_fd(struct packed_git
*p
)
923 static void close_pack(struct packed_git
*p
)
925 close_pack_windows(p
);
930 void close_all_packs(void)
932 struct packed_git
*p
;
934 for (p
= packed_git
; p
; p
= p
->next
)
936 die("BUG: want to close pack marked 'do-not-close'");
943 * The LRU pack is the one with the oldest MRU window, preferring packs
944 * with no used windows, or the oldest mtime if it has no windows allocated.
946 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
948 struct pack_window
*w
, *this_mru_w
;
949 int has_windows_inuse
= 0;
952 * Reject this pack if it has windows and the previously selected
953 * one does not. If this pack does not have windows, reject
954 * it if the pack file is newer than the previously selected one.
956 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
959 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
961 * Reject this pack if any of its windows are in use,
962 * but the previously selected pack did not have any
963 * inuse windows. Otherwise, record that this pack
964 * has windows in use.
967 if (*accept_windows_inuse
)
968 has_windows_inuse
= 1;
973 if (w
->last_used
> this_mru_w
->last_used
)
977 * Reject this pack if it has windows that have been
978 * used more recently than the previously selected pack.
979 * If the previously selected pack had windows inuse and
980 * we have not encountered a window in this pack that is
981 * inuse, skip this check since we prefer a pack with no
982 * inuse windows to one that has inuse windows.
984 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
985 this_mru_w
->last_used
> (*mru_w
)->last_used
)
994 *accept_windows_inuse
= has_windows_inuse
;
997 static int close_one_pack(void)
999 struct packed_git
*p
, *lru_p
= NULL
;
1000 struct pack_window
*mru_w
= NULL
;
1001 int accept_windows_inuse
= 1;
1003 for (p
= packed_git
; p
; p
= p
->next
) {
1004 if (p
->pack_fd
== -1)
1006 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
1010 return close_pack_fd(lru_p
);
1015 void unuse_pack(struct pack_window
**w_cursor
)
1017 struct pack_window
*w
= *w_cursor
;
1024 void close_pack_index(struct packed_git
*p
)
1026 if (p
->index_data
) {
1027 munmap((void *)p
->index_data
, p
->index_size
);
1028 p
->index_data
= NULL
;
1032 static unsigned int get_max_fd_limit(void)
1034 #ifdef RLIMIT_NOFILE
1038 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
1039 return lim
.rlim_cur
;
1045 long open_max
= sysconf(_SC_OPEN_MAX
);
1049 * Otherwise, we got -1 for one of the two
1052 * (1) sysconf() did not understand _SC_OPEN_MAX
1053 * and signaled an error with -1; or
1054 * (2) sysconf() said there is no limit.
1056 * We _could_ clear errno before calling sysconf() to
1057 * tell these two cases apart and return a huge number
1058 * in the latter case to let the caller cap it to a
1059 * value that is not so selfish, but letting the
1060 * fallback OPEN_MAX codepath take care of these cases
1069 return 1; /* see the caller ;-) */
1074 * Do not call this directly as this leaks p->pack_fd on error return;
1075 * call open_packed_git() instead.
1077 static int open_packed_git_1(struct packed_git
*p
)
1080 struct pack_header hdr
;
1081 unsigned char sha1
[20];
1082 unsigned char *idx_sha1
;
1085 if (!p
->index_data
&& open_pack_index(p
))
1086 return error("packfile %s index unavailable", p
->pack_name
);
1088 if (!pack_max_fds
) {
1089 unsigned int max_fds
= get_max_fd_limit();
1091 /* Save 3 for stdin/stdout/stderr, 22 for work */
1093 pack_max_fds
= max_fds
- 25;
1098 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
1101 p
->pack_fd
= git_open(p
->pack_name
);
1102 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
1106 /* If we created the struct before we had the pack we lack size. */
1107 if (!p
->pack_size
) {
1108 if (!S_ISREG(st
.st_mode
))
1109 return error("packfile %s not a regular file", p
->pack_name
);
1110 p
->pack_size
= st
.st_size
;
1111 } else if (p
->pack_size
!= st
.st_size
)
1112 return error("packfile %s size changed", p
->pack_name
);
1114 /* We leave these file descriptors open with sliding mmap;
1115 * there is no point keeping them open across exec(), though.
1117 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
1119 return error("cannot determine file descriptor flags");
1120 fd_flag
|= FD_CLOEXEC
;
1121 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
1122 return error("cannot set FD_CLOEXEC");
1124 /* Verify we recognize this pack file format. */
1125 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
1126 return error("file %s is far too short to be a packfile", p
->pack_name
);
1127 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
1128 return error("file %s is not a GIT packfile", p
->pack_name
);
1129 if (!pack_version_ok(hdr
.hdr_version
))
1130 return error("packfile %s is version %"PRIu32
" and not"
1131 " supported (try upgrading GIT to a newer version)",
1132 p
->pack_name
, ntohl(hdr
.hdr_version
));
1134 /* Verify the pack matches its index. */
1135 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
1136 return error("packfile %s claims to have %"PRIu32
" objects"
1137 " while index indicates %"PRIu32
" objects",
1138 p
->pack_name
, ntohl(hdr
.hdr_entries
),
1140 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1141 return error("end of packfile %s is unavailable", p
->pack_name
);
1142 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1143 return error("packfile %s signature is unavailable", p
->pack_name
);
1144 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1145 if (hashcmp(sha1
, idx_sha1
))
1146 return error("packfile %s does not match index", p
->pack_name
);
1150 static int open_packed_git(struct packed_git
*p
)
1152 if (!open_packed_git_1(p
))
1158 static int in_window(struct pack_window
*win
, off_t offset
)
1160 /* We must promise at least 20 bytes (one hash) after the
1161 * offset is available from this window, otherwise the offset
1162 * is not actually in this window and a different window (which
1163 * has that one hash excess) must be used. This is to support
1164 * the object header and delta base parsing routines below.
1166 off_t win_off
= win
->offset
;
1167 return win_off
<= offset
1168 && (offset
+ 20) <= (win_off
+ win
->len
);
1171 unsigned char *use_pack(struct packed_git
*p
,
1172 struct pack_window
**w_cursor
,
1174 unsigned long *left
)
1176 struct pack_window
*win
= *w_cursor
;
1178 /* Since packfiles end in a hash of their content and it's
1179 * pointless to ask for an offset into the middle of that
1180 * hash, and the in_window function above wouldn't match
1181 * don't allow an offset too close to the end of the file.
1183 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1184 die("packfile %s cannot be accessed", p
->pack_name
);
1185 if (offset
> (p
->pack_size
- 20))
1186 die("offset beyond end of packfile (truncated pack?)");
1188 die(_("offset before end of packfile (broken .idx?)"));
1190 if (!win
|| !in_window(win
, offset
)) {
1193 for (win
= p
->windows
; win
; win
= win
->next
) {
1194 if (in_window(win
, offset
))
1198 size_t window_align
= packed_git_window_size
/ 2;
1201 if (p
->pack_fd
== -1 && open_packed_git(p
))
1202 die("packfile %s cannot be accessed", p
->pack_name
);
1204 win
= xcalloc(1, sizeof(*win
));
1205 win
->offset
= (offset
/ window_align
) * window_align
;
1206 len
= p
->pack_size
- win
->offset
;
1207 if (len
> packed_git_window_size
)
1208 len
= packed_git_window_size
;
1209 win
->len
= (size_t)len
;
1210 pack_mapped
+= win
->len
;
1211 while (packed_git_limit
< pack_mapped
1212 && unuse_one_window(p
))
1214 win
->base
= xmmap(NULL
, win
->len
,
1215 PROT_READ
, MAP_PRIVATE
,
1216 p
->pack_fd
, win
->offset
);
1217 if (win
->base
== MAP_FAILED
)
1218 die_errno("packfile %s cannot be mapped",
1220 if (!win
->offset
&& win
->len
== p
->pack_size
1221 && !p
->do_not_close
)
1224 pack_open_windows
++;
1225 if (pack_mapped
> peak_pack_mapped
)
1226 peak_pack_mapped
= pack_mapped
;
1227 if (pack_open_windows
> peak_pack_open_windows
)
1228 peak_pack_open_windows
= pack_open_windows
;
1229 win
->next
= p
->windows
;
1233 if (win
!= *w_cursor
) {
1234 win
->last_used
= pack_used_ctr
++;
1238 offset
-= win
->offset
;
1240 *left
= win
->len
- xsize_t(offset
);
1241 return win
->base
+ offset
;
1244 static struct packed_git
*alloc_packed_git(int extra
)
1246 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
1247 memset(p
, 0, sizeof(*p
));
1252 static void try_to_free_pack_memory(size_t size
)
1254 release_pack_memory(size
);
1257 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1259 static int have_set_try_to_free_routine
;
1262 struct packed_git
*p
;
1264 if (!have_set_try_to_free_routine
) {
1265 have_set_try_to_free_routine
= 1;
1266 set_try_to_free_routine(try_to_free_pack_memory
);
1270 * Make sure a corresponding .pack file exists and that
1271 * the index looks sane.
1273 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1277 * ".pack" is long enough to hold any suffix we're adding (and
1278 * the use xsnprintf double-checks that)
1280 alloc
= st_add3(path_len
, strlen(".pack"), 1);
1281 p
= alloc_packed_git(alloc
);
1282 memcpy(p
->pack_name
, path
, path_len
);
1284 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1285 if (!access(p
->pack_name
, F_OK
))
1288 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1289 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1294 /* ok, it looks sane as far as we can check without
1295 * actually mapping the pack file.
1297 p
->pack_size
= st
.st_size
;
1298 p
->pack_local
= local
;
1299 p
->mtime
= st
.st_mtime
;
1300 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1305 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1307 const char *path
= sha1_pack_name(sha1
);
1308 size_t alloc
= st_add(strlen(path
), 1);
1309 struct packed_git
*p
= alloc_packed_git(alloc
);
1311 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
1312 hashcpy(p
->sha1
, sha1
);
1313 if (check_packed_git_idx(idx_path
, p
)) {
1321 void install_packed_git(struct packed_git
*pack
)
1323 if (pack
->pack_fd
!= -1)
1326 pack
->next
= packed_git
;
1330 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
1332 static void report_helper(const struct string_list
*list
,
1333 int seen_bits
, int first
, int last
)
1335 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
1338 for (; first
< last
; first
++)
1339 report_garbage(seen_bits
, list
->items
[first
].string
);
1342 static void report_pack_garbage(struct string_list
*list
)
1344 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1346 if (!report_garbage
)
1349 string_list_sort(list
);
1351 for (i
= 0; i
< list
->nr
; i
++) {
1352 const char *path
= list
->items
[i
].string
;
1353 if (baselen
!= -1 &&
1354 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1355 report_helper(list
, seen_bits
, first
, i
);
1359 if (baselen
== -1) {
1360 const char *dot
= strrchr(path
, '.');
1362 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
1365 baselen
= dot
- path
+ 1;
1368 if (!strcmp(path
+ baselen
, "pack"))
1370 else if (!strcmp(path
+ baselen
, "idx"))
1373 report_helper(list
, seen_bits
, first
, list
->nr
);
1376 static void prepare_packed_git_one(char *objdir
, int local
)
1378 struct strbuf path
= STRBUF_INIT
;
1382 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1384 strbuf_addstr(&path
, objdir
);
1385 strbuf_addstr(&path
, "/pack");
1386 dir
= opendir(path
.buf
);
1388 if (errno
!= ENOENT
)
1389 error_errno("unable to open object pack directory: %s",
1391 strbuf_release(&path
);
1394 strbuf_addch(&path
, '/');
1395 dirnamelen
= path
.len
;
1396 while ((de
= readdir(dir
)) != NULL
) {
1397 struct packed_git
*p
;
1400 if (is_dot_or_dotdot(de
->d_name
))
1403 strbuf_setlen(&path
, dirnamelen
);
1404 strbuf_addstr(&path
, de
->d_name
);
1406 base_len
= path
.len
;
1407 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1408 /* Don't reopen a pack we already have. */
1409 for (p
= packed_git
; p
; p
= p
->next
) {
1411 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1413 !memcmp(p
->pack_name
, path
.buf
, len
))
1418 * See if it really is a valid .idx file with
1419 * corresponding .pack file that we can map.
1421 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1422 install_packed_git(p
);
1425 if (!report_garbage
)
1428 if (ends_with(de
->d_name
, ".idx") ||
1429 ends_with(de
->d_name
, ".pack") ||
1430 ends_with(de
->d_name
, ".bitmap") ||
1431 ends_with(de
->d_name
, ".keep"))
1432 string_list_append(&garbage
, path
.buf
);
1434 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
1437 report_pack_garbage(&garbage
);
1438 string_list_clear(&garbage
, 0);
1439 strbuf_release(&path
);
1442 static int approximate_object_count_valid
;
1445 * Give a fast, rough count of the number of objects in the repository. This
1446 * ignores loose objects completely. If you have a lot of them, then either
1447 * you should repack because your performance will be awful, or they are
1448 * all unreachable objects about to be pruned, in which case they're not really
1449 * interesting as a measure of repo size in the first place.
1451 unsigned long approximate_object_count(void)
1453 static unsigned long count
;
1454 if (!approximate_object_count_valid
) {
1455 struct packed_git
*p
;
1457 prepare_packed_git();
1459 for (p
= packed_git
; p
; p
= p
->next
) {
1460 if (open_pack_index(p
))
1462 count
+= p
->num_objects
;
1468 static void *get_next_packed_git(const void *p
)
1470 return ((const struct packed_git
*)p
)->next
;
1473 static void set_next_packed_git(void *p
, void *next
)
1475 ((struct packed_git
*)p
)->next
= next
;
1478 static int sort_pack(const void *a_
, const void *b_
)
1480 const struct packed_git
*a
= a_
;
1481 const struct packed_git
*b
= b_
;
1485 * Local packs tend to contain objects specific to our
1486 * variant of the project than remote ones. In addition,
1487 * remote ones could be on a network mounted filesystem.
1488 * Favor local ones for these reasons.
1490 st
= a
->pack_local
- b
->pack_local
;
1495 * Younger packs tend to contain more recent objects,
1496 * and more recent objects tend to get accessed more
1499 if (a
->mtime
< b
->mtime
)
1501 else if (a
->mtime
== b
->mtime
)
1506 static void rearrange_packed_git(void)
1508 packed_git
= llist_mergesort(packed_git
, get_next_packed_git
,
1509 set_next_packed_git
, sort_pack
);
1512 static void prepare_packed_git_mru(void)
1514 struct packed_git
*p
;
1516 mru_clear(packed_git_mru
);
1517 for (p
= packed_git
; p
; p
= p
->next
)
1518 mru_append(packed_git_mru
, p
);
1521 static int prepare_packed_git_run_once
= 0;
1522 void prepare_packed_git(void)
1524 struct alternate_object_database
*alt
;
1526 if (prepare_packed_git_run_once
)
1528 prepare_packed_git_one(get_object_directory(), 1);
1530 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
)
1531 prepare_packed_git_one(alt
->path
, 0);
1532 rearrange_packed_git();
1533 prepare_packed_git_mru();
1534 prepare_packed_git_run_once
= 1;
1537 void reprepare_packed_git(void)
1539 approximate_object_count_valid
= 0;
1540 prepare_packed_git_run_once
= 0;
1541 prepare_packed_git();
1544 static void mark_bad_packed_object(struct packed_git
*p
,
1545 const unsigned char *sha1
)
1548 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1549 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
1551 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1552 st_mult(GIT_SHA1_RAWSZ
,
1553 st_add(p
->num_bad_objects
, 1)));
1554 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
1555 p
->num_bad_objects
++;
1558 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1560 struct packed_git
*p
;
1563 for (p
= packed_git
; p
; p
= p
->next
)
1564 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1565 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1571 * With an in-core object data in "map", rehash it to make sure the
1572 * object name actually matches "sha1" to detect object corruption.
1573 * With "map" == NULL, try reading the object named with "sha1" using
1574 * the streaming interface and rehash it to do the same.
1576 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1577 unsigned long size
, const char *type
)
1579 unsigned char real_sha1
[20];
1580 enum object_type obj_type
;
1581 struct git_istream
*st
;
1587 hash_sha1_file(map
, size
, type
, real_sha1
);
1588 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1591 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1595 /* Generate the header */
1596 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1600 git_SHA1_Update(&c
, hdr
, hdrlen
);
1602 char buf
[1024 * 16];
1603 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1611 git_SHA1_Update(&c
, buf
, readlen
);
1613 git_SHA1_Final(real_sha1
, &c
);
1615 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1618 int git_open(const char *name
)
1620 static int sha1_file_open_flag
= O_NOATIME
| O_CLOEXEC
;
1626 fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1630 /* Try again w/o O_CLOEXEC: the kernel might not support it */
1631 if ((sha1_file_open_flag
& O_CLOEXEC
) && errno
== EINVAL
) {
1632 sha1_file_open_flag
&= ~O_CLOEXEC
;
1636 /* Might the failure be due to O_NOATIME? */
1637 if (errno
!= ENOENT
&& (sha1_file_open_flag
& O_NOATIME
)) {
1638 sha1_file_open_flag
&= ~O_NOATIME
;
1645 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1647 struct alternate_object_database
*alt
;
1649 if (!lstat(sha1_file_name(sha1
), st
))
1654 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1655 const char *path
= alt_sha1_path(alt
, sha1
);
1656 if (!lstat(path
, st
))
1663 static int open_sha1_file(const unsigned char *sha1
)
1666 struct alternate_object_database
*alt
;
1667 int most_interesting_errno
;
1669 fd
= git_open(sha1_file_name(sha1
));
1672 most_interesting_errno
= errno
;
1675 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1676 const char *path
= alt_sha1_path(alt
, sha1
);
1677 fd
= git_open(path
);
1680 if (most_interesting_errno
== ENOENT
)
1681 most_interesting_errno
= errno
;
1683 errno
= most_interesting_errno
;
1687 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1692 fd
= open_sha1_file(sha1
);
1697 if (!fstat(fd
, &st
)) {
1698 *size
= xsize_t(st
.st_size
);
1700 /* mmap() is forbidden on empty files */
1701 error("object file %s is empty", sha1_file_name(sha1
));
1704 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1711 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1712 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1715 unsigned long size
, c
;
1716 unsigned long used
= 0;
1719 *type
= (c
>> 4) & 7;
1723 if (len
<= used
|| bitsizeof(long) <= shift
) {
1724 error("bad object header");
1729 size
+= (c
& 0x7f) << shift
;
1736 static int unpack_sha1_short_header(git_zstream
*stream
,
1737 unsigned char *map
, unsigned long mapsize
,
1738 void *buffer
, unsigned long bufsiz
)
1740 /* Get the data stream */
1741 memset(stream
, 0, sizeof(*stream
));
1742 stream
->next_in
= map
;
1743 stream
->avail_in
= mapsize
;
1744 stream
->next_out
= buffer
;
1745 stream
->avail_out
= bufsiz
;
1747 git_inflate_init(stream
);
1748 return git_inflate(stream
, 0);
1751 int unpack_sha1_header(git_zstream
*stream
,
1752 unsigned char *map
, unsigned long mapsize
,
1753 void *buffer
, unsigned long bufsiz
)
1755 int status
= unpack_sha1_short_header(stream
, map
, mapsize
,
1761 /* Make sure we have the terminating NUL */
1762 if (!memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1767 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1768 unsigned long mapsize
, void *buffer
,
1769 unsigned long bufsiz
, struct strbuf
*header
)
1773 status
= unpack_sha1_short_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1778 * Check if entire header is unpacked in the first iteration.
1780 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1784 * buffer[0..bufsiz] was not large enough. Copy the partial
1785 * result out to header, and then append the result of further
1786 * reading the stream.
1788 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1789 stream
->next_out
= buffer
;
1790 stream
->avail_out
= bufsiz
;
1793 status
= git_inflate(stream
, 0);
1794 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1795 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1797 stream
->next_out
= buffer
;
1798 stream
->avail_out
= bufsiz
;
1799 } while (status
!= Z_STREAM_END
);
1803 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1805 int bytes
= strlen(buffer
) + 1;
1806 unsigned char *buf
= xmallocz(size
);
1810 n
= stream
->total_out
- bytes
;
1813 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1815 if (bytes
<= size
) {
1817 * The above condition must be (bytes <= size), not
1818 * (bytes < size). In other words, even though we
1819 * expect no more output and set avail_out to zero,
1820 * the input zlib stream may have bytes that express
1821 * "this concludes the stream", and we *do* want to
1824 * Otherwise we would not be able to test that we
1825 * consumed all the input to reach the expected size;
1826 * we also want to check that zlib tells us that all
1827 * went well with status == Z_STREAM_END at the end.
1829 stream
->next_out
= buf
+ bytes
;
1830 stream
->avail_out
= size
- bytes
;
1831 while (status
== Z_OK
)
1832 status
= git_inflate(stream
, Z_FINISH
);
1834 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1835 git_inflate_end(stream
);
1840 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1841 else if (stream
->avail_in
)
1842 error("garbage at end of loose object '%s'",
1849 * We used to just use "sscanf()", but that's actually way
1850 * too permissive for what we want to check. So do an anal
1851 * object header parse by hand.
1853 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1856 const char *type_buf
= hdr
;
1858 int type
, type_len
= 0;
1861 * The type can be of any size but is followed by
1873 type
= type_from_string_gently(type_buf
, type_len
, 1);
1875 strbuf_add(oi
->typename
, type_buf
, type_len
);
1877 * Set type to 0 if its an unknown object and
1878 * we're obtaining the type using '--allow-unknown-type'
1881 if ((flags
& LOOKUP_UNKNOWN_OBJECT
) && (type
< 0))
1884 die("invalid object type");
1889 * The length must follow immediately, and be in canonical
1890 * decimal format (ie "010" is not valid).
1892 size
= *hdr
++ - '0';
1897 unsigned long c
= *hdr
- '0';
1901 size
= size
* 10 + c
;
1909 * The length must be followed by a zero byte
1911 return *hdr
? -1 : type
;
1914 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1916 struct object_info oi
= OBJECT_INFO_INIT
;
1919 return parse_sha1_header_extended(hdr
, &oi
, LOOKUP_REPLACE_OBJECT
);
1922 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1928 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1929 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1932 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1935 unsigned long get_size_from_delta(struct packed_git
*p
,
1936 struct pack_window
**w_curs
,
1939 const unsigned char *data
;
1940 unsigned char delta_head
[20], *in
;
1944 memset(&stream
, 0, sizeof(stream
));
1945 stream
.next_out
= delta_head
;
1946 stream
.avail_out
= sizeof(delta_head
);
1948 git_inflate_init(&stream
);
1950 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1951 stream
.next_in
= in
;
1952 st
= git_inflate(&stream
, Z_FINISH
);
1953 curpos
+= stream
.next_in
- in
;
1954 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1955 stream
.total_out
< sizeof(delta_head
));
1956 git_inflate_end(&stream
);
1957 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1958 error("delta data unpack-initial failed");
1962 /* Examine the initial part of the delta to figure out
1967 /* ignore base size */
1968 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1970 /* Read the result size */
1971 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1974 static off_t
get_delta_base(struct packed_git
*p
,
1975 struct pack_window
**w_curs
,
1977 enum object_type type
,
1978 off_t delta_obj_offset
)
1980 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1983 /* use_pack() assured us we have [base_info, base_info + 20)
1984 * as a range that we can look at without walking off the
1985 * end of the mapped window. Its actually the hash size
1986 * that is assured. An OFS_DELTA longer than the hash size
1987 * is stupid, as then a REF_DELTA would be smaller to store.
1989 if (type
== OBJ_OFS_DELTA
) {
1991 unsigned char c
= base_info
[used
++];
1992 base_offset
= c
& 127;
1995 if (!base_offset
|| MSB(base_offset
, 7))
1996 return 0; /* overflow */
1997 c
= base_info
[used
++];
1998 base_offset
= (base_offset
<< 7) + (c
& 127);
2000 base_offset
= delta_obj_offset
- base_offset
;
2001 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
2002 return 0; /* out of bound */
2004 } else if (type
== OBJ_REF_DELTA
) {
2005 /* The base entry _must_ be in the same pack */
2006 base_offset
= find_pack_entry_one(base_info
, p
);
2009 die("I am totally screwed");
2014 * Like get_delta_base above, but we return the sha1 instead of the pack
2015 * offset. This means it is cheaper for REF deltas (we do not have to do
2016 * the final object lookup), but more expensive for OFS deltas (we
2017 * have to load the revidx to convert the offset back into a sha1).
2019 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
2020 struct pack_window
**w_curs
,
2022 enum object_type type
,
2023 off_t delta_obj_offset
)
2025 if (type
== OBJ_REF_DELTA
) {
2026 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
2028 } else if (type
== OBJ_OFS_DELTA
) {
2029 struct revindex_entry
*revidx
;
2030 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
2031 type
, delta_obj_offset
);
2036 revidx
= find_pack_revindex(p
, base_offset
);
2040 return nth_packed_object_sha1(p
, revidx
->nr
);
2045 int unpack_object_header(struct packed_git
*p
,
2046 struct pack_window
**w_curs
,
2048 unsigned long *sizep
)
2050 unsigned char *base
;
2053 enum object_type type
;
2055 /* use_pack() assures us we have [base, base + 20) available
2056 * as a range that we can look at. (Its actually the hash
2057 * size that is assured.) With our object header encoding
2058 * the maximum deflated object size is 2^137, which is just
2059 * insane, so we know won't exceed what we have been given.
2061 base
= use_pack(p
, w_curs
, *curpos
, &left
);
2062 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
2071 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
2074 struct revindex_entry
*revidx
;
2075 const unsigned char *sha1
;
2076 revidx
= find_pack_revindex(p
, obj_offset
);
2079 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2080 mark_bad_packed_object(p
, sha1
);
2081 type
= sha1_object_info(sha1
, NULL
);
2082 if (type
<= OBJ_NONE
)
2087 #define POI_STACK_PREALLOC 64
2089 static enum object_type
packed_to_object_type(struct packed_git
*p
,
2091 enum object_type type
,
2092 struct pack_window
**w_curs
,
2095 off_t small_poi_stack
[POI_STACK_PREALLOC
];
2096 off_t
*poi_stack
= small_poi_stack
;
2097 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
2099 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2102 /* Push the object we're going to leave behind */
2103 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
2104 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
2105 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
2106 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
2108 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
2110 poi_stack
[poi_stack_nr
++] = obj_offset
;
2111 /* If parsing the base offset fails, just unwind */
2112 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
2115 curpos
= obj_offset
= base_offset
;
2116 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
2117 if (type
<= OBJ_NONE
) {
2118 /* If getting the base itself fails, we first
2119 * retry the base, otherwise unwind */
2120 type
= retry_bad_packed_offset(p
, base_offset
);
2121 if (type
> OBJ_NONE
)
2135 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2136 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2141 if (poi_stack
!= small_poi_stack
)
2146 while (poi_stack_nr
) {
2147 obj_offset
= poi_stack
[--poi_stack_nr
];
2148 type
= retry_bad_packed_offset(p
, obj_offset
);
2149 if (type
> OBJ_NONE
)
2156 int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
2157 struct object_info
*oi
)
2159 struct pack_window
*w_curs
= NULL
;
2161 off_t curpos
= obj_offset
;
2162 enum object_type type
;
2165 * We always get the representation type, but only convert it to
2166 * a "real" type later if the caller is interested.
2168 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2171 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2172 off_t tmp_pos
= curpos
;
2173 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
2179 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2180 if (*oi
->sizep
== 0) {
2189 if (oi
->disk_sizep
) {
2190 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2191 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2195 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
2196 if (*oi
->typep
< 0) {
2202 if (oi
->delta_base_sha1
) {
2203 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2204 const unsigned char *base
;
2206 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2213 hashcpy(oi
->delta_base_sha1
, base
);
2215 hashclr(oi
->delta_base_sha1
);
2219 unuse_pack(&w_curs
);
2223 static void *unpack_compressed_entry(struct packed_git
*p
,
2224 struct pack_window
**w_curs
,
2230 unsigned char *buffer
, *in
;
2232 buffer
= xmallocz_gently(size
);
2235 memset(&stream
, 0, sizeof(stream
));
2236 stream
.next_out
= buffer
;
2237 stream
.avail_out
= size
+ 1;
2239 git_inflate_init(&stream
);
2241 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2242 stream
.next_in
= in
;
2243 st
= git_inflate(&stream
, Z_FINISH
);
2244 if (!stream
.avail_out
)
2245 break; /* the payload is larger than it should be */
2246 curpos
+= stream
.next_in
- in
;
2247 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2248 git_inflate_end(&stream
);
2249 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2257 static struct hashmap delta_base_cache
;
2258 static size_t delta_base_cached
;
2260 static LIST_HEAD(delta_base_cache_lru
);
2262 struct delta_base_cache_key
{
2263 struct packed_git
*p
;
2267 struct delta_base_cache_entry
{
2268 struct hashmap hash
;
2269 struct delta_base_cache_key key
;
2270 struct list_head lru
;
2273 enum object_type type
;
2276 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2280 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
2281 hash
+= (hash
>> 8) + (hash
>> 16);
2285 static struct delta_base_cache_entry
*
2286 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2288 struct hashmap_entry entry
;
2289 struct delta_base_cache_key key
;
2291 if (!delta_base_cache
.cmpfn
)
2294 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
2296 key
.base_offset
= base_offset
;
2297 return hashmap_get(&delta_base_cache
, &entry
, &key
);
2300 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
2301 const struct delta_base_cache_key
*b
)
2303 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
2306 static int delta_base_cache_hash_cmp(const void *va
, const void *vb
,
2309 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
2310 const struct delta_base_cache_key
*key
= vkey
;
2312 return !delta_base_cache_key_eq(&a
->key
, key
);
2314 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
2317 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2319 return !!get_delta_base_cache_entry(p
, base_offset
);
2323 * Remove the entry from the cache, but do _not_ free the associated
2324 * entry data. The caller takes ownership of the "data" buffer, and
2325 * should copy out any fields it wants before detaching.
2327 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2329 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
2330 list_del(&ent
->lru
);
2331 delta_base_cached
-= ent
->size
;
2335 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2336 unsigned long *base_size
, enum object_type
*type
)
2338 struct delta_base_cache_entry
*ent
;
2340 ent
= get_delta_base_cache_entry(p
, base_offset
);
2342 return unpack_entry(p
, base_offset
, type
, base_size
);
2345 *base_size
= ent
->size
;
2346 return xmemdupz(ent
->data
, ent
->size
);
2349 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2352 detach_delta_base_cache_entry(ent
);
2355 void clear_delta_base_cache(void)
2357 struct hashmap_iter iter
;
2358 struct delta_base_cache_entry
*entry
;
2359 for (entry
= hashmap_iter_first(&delta_base_cache
, &iter
);
2361 entry
= hashmap_iter_next(&iter
)) {
2362 release_delta_base_cache(entry
);
2366 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2367 void *base
, unsigned long base_size
, enum object_type type
)
2369 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
2370 struct list_head
*lru
, *tmp
;
2372 delta_base_cached
+= base_size
;
2374 list_for_each_safe(lru
, tmp
, &delta_base_cache_lru
) {
2375 struct delta_base_cache_entry
*f
=
2376 list_entry(lru
, struct delta_base_cache_entry
, lru
);
2377 if (delta_base_cached
<= delta_base_cache_limit
)
2379 release_delta_base_cache(f
);
2383 ent
->key
.base_offset
= base_offset
;
2386 ent
->size
= base_size
;
2387 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
2389 if (!delta_base_cache
.cmpfn
)
2390 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, 0);
2391 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
2392 hashmap_add(&delta_base_cache
, ent
);
2395 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2396 unsigned long *size
);
2398 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2400 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2401 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2402 p
->pack_name
, (uintmax_t)obj_offset
);
2405 int do_check_packed_object_crc
;
2407 #define UNPACK_ENTRY_STACK_PREALLOC 64
2408 struct unpack_entry_stack_ent
{
2414 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2415 enum object_type
*final_type
, unsigned long *final_size
)
2417 struct pack_window
*w_curs
= NULL
;
2418 off_t curpos
= obj_offset
;
2421 enum object_type type
;
2422 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2423 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2424 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2425 int base_from_cache
= 0;
2427 write_pack_access_log(p
, obj_offset
);
2429 /* PHASE 1: drill down to the innermost base object */
2433 struct delta_base_cache_entry
*ent
;
2435 ent
= get_delta_base_cache_entry(p
, curpos
);
2440 detach_delta_base_cache_entry(ent
);
2441 base_from_cache
= 1;
2445 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2446 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2447 off_t len
= revidx
[1].offset
- obj_offset
;
2448 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2449 const unsigned char *sha1
=
2450 nth_packed_object_sha1(p
, revidx
->nr
);
2451 error("bad packed object CRC for %s",
2453 mark_bad_packed_object(p
, sha1
);
2454 unuse_pack(&w_curs
);
2459 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2460 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2463 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2465 error("failed to validate delta base reference "
2466 "at offset %"PRIuMAX
" from %s",
2467 (uintmax_t)curpos
, p
->pack_name
);
2468 /* bail to phase 2, in hopes of recovery */
2473 /* push object, proceed to base */
2474 if (delta_stack_nr
>= delta_stack_alloc
2475 && delta_stack
== small_delta_stack
) {
2476 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2477 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
2478 memcpy(delta_stack
, small_delta_stack
,
2479 sizeof(*delta_stack
)*delta_stack_nr
);
2481 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2483 i
= delta_stack_nr
++;
2484 delta_stack
[i
].obj_offset
= obj_offset
;
2485 delta_stack
[i
].curpos
= curpos
;
2486 delta_stack
[i
].size
= size
;
2488 curpos
= obj_offset
= base_offset
;
2491 /* PHASE 2: handle the base */
2496 die("BUG: unpack_entry: left loop at a valid delta");
2502 if (!base_from_cache
)
2503 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2507 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2508 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2511 /* PHASE 3: apply deltas in order */
2514 * 'data' holds the base data, or NULL if there was corruption
2516 while (delta_stack_nr
) {
2519 unsigned long delta_size
, base_size
= size
;
2525 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2529 * We're probably in deep shit, but let's try to fetch
2530 * the required base anyway from another pack or loose.
2531 * This is costly but should happen only in the presence
2532 * of a corrupted pack, and is better than failing outright.
2534 struct revindex_entry
*revidx
;
2535 const unsigned char *base_sha1
;
2536 revidx
= find_pack_revindex(p
, obj_offset
);
2538 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2539 error("failed to read delta base object %s"
2540 " at offset %"PRIuMAX
" from %s",
2541 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2543 mark_bad_packed_object(p
, base_sha1
);
2544 base
= read_object(base_sha1
, &type
, &base_size
);
2548 i
= --delta_stack_nr
;
2549 obj_offset
= delta_stack
[i
].obj_offset
;
2550 curpos
= delta_stack
[i
].curpos
;
2551 delta_size
= delta_stack
[i
].size
;
2556 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2559 error("failed to unpack compressed delta "
2560 "at offset %"PRIuMAX
" from %s",
2561 (uintmax_t)curpos
, p
->pack_name
);
2566 data
= patch_delta(base
, base_size
,
2567 delta_data
, delta_size
,
2571 * We could not apply the delta; warn the user, but keep going.
2572 * Our failure will be noticed either in the next iteration of
2573 * the loop, or if this is the final delta, in the caller when
2574 * we return NULL. Those code paths will take care of making
2575 * a more explicit warning and retrying with another copy of
2579 error("failed to apply delta");
2587 unuse_pack(&w_curs
);
2589 if (delta_stack
!= small_delta_stack
)
2595 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2598 const unsigned char *index
= p
->index_data
;
2600 if (open_pack_index(p
))
2602 index
= p
->index_data
;
2604 if (n
>= p
->num_objects
)
2607 if (p
->index_version
== 1) {
2608 return index
+ 24 * n
+ 4;
2611 return index
+ 20 * n
;
2615 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2617 const unsigned char *ptr
= vptr
;
2618 const unsigned char *start
= p
->index_data
;
2619 const unsigned char *end
= start
+ p
->index_size
;
2621 die(_("offset before start of pack index for %s (corrupt index?)"),
2623 /* No need to check for underflow; .idx files must be at least 8 bytes */
2625 die(_("offset beyond end of pack index for %s (truncated index?)"),
2629 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2631 const unsigned char *index
= p
->index_data
;
2633 if (p
->index_version
== 1) {
2634 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2637 index
+= 8 + p
->num_objects
* (20 + 4);
2638 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2639 if (!(off
& 0x80000000))
2641 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2642 check_pack_index_ptr(p
, index
);
2643 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2644 ntohl(*((uint32_t *)(index
+ 4)));
2648 off_t
find_pack_entry_one(const unsigned char *sha1
,
2649 struct packed_git
*p
)
2651 const uint32_t *level1_ofs
= p
->index_data
;
2652 const unsigned char *index
= p
->index_data
;
2653 unsigned hi
, lo
, stride
;
2654 static int use_lookup
= -1;
2655 static int debug_lookup
= -1;
2657 if (debug_lookup
< 0)
2658 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2661 if (open_pack_index(p
))
2663 level1_ofs
= p
->index_data
;
2664 index
= p
->index_data
;
2666 if (p
->index_version
> 1) {
2671 hi
= ntohl(level1_ofs
[*sha1
]);
2672 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2673 if (p
->index_version
> 1) {
2681 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2682 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2685 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2687 int pos
= sha1_entry_pos(index
, stride
, 0,
2688 lo
, hi
, p
->num_objects
, sha1
);
2691 return nth_packed_object_offset(p
, pos
);
2695 unsigned mi
= (lo
+ hi
) / 2;
2696 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2699 printf("lo %u hi %u rg %u mi %u\n",
2700 lo
, hi
, hi
- lo
, mi
);
2702 return nth_packed_object_offset(p
, mi
);
2711 int is_pack_valid(struct packed_git
*p
)
2713 /* An already open pack is known to be valid. */
2714 if (p
->pack_fd
!= -1)
2717 /* If the pack has one window completely covering the
2718 * file size, the pack is known to be valid even if
2719 * the descriptor is not currently open.
2722 struct pack_window
*w
= p
->windows
;
2724 if (!w
->offset
&& w
->len
== p
->pack_size
)
2728 /* Force the pack to open to prove its valid. */
2729 return !open_packed_git(p
);
2732 static int fill_pack_entry(const unsigned char *sha1
,
2733 struct pack_entry
*e
,
2734 struct packed_git
*p
)
2738 if (p
->num_bad_objects
) {
2740 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2741 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2745 offset
= find_pack_entry_one(sha1
, p
);
2750 * We are about to tell the caller where they can locate the
2751 * requested object. We better make sure the packfile is
2752 * still here and can be accessed before supplying that
2753 * answer, as it may have been deleted since the index was
2756 if (!is_pack_valid(p
))
2760 hashcpy(e
->sha1
, sha1
);
2765 * Iff a pack file contains the object named by sha1, return true and
2766 * store its location to e.
2768 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2770 struct mru_entry
*p
;
2772 prepare_packed_git();
2776 for (p
= packed_git_mru
->head
; p
; p
= p
->next
) {
2777 if (fill_pack_entry(sha1
, e
, p
->item
)) {
2778 mru_mark(packed_git_mru
, p
);
2785 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2786 struct packed_git
*packs
)
2788 struct packed_git
*p
;
2790 for (p
= packs
; p
; p
= p
->next
) {
2791 if (find_pack_entry_one(sha1
, p
))
2798 static int sha1_loose_object_info(const unsigned char *sha1
,
2799 struct object_info
*oi
,
2803 unsigned long mapsize
;
2807 struct strbuf hdrbuf
= STRBUF_INIT
;
2809 if (oi
->delta_base_sha1
)
2810 hashclr(oi
->delta_base_sha1
);
2813 * If we don't care about type or size, then we don't
2814 * need to look inside the object at all. Note that we
2815 * do not optimize out the stat call, even if the
2816 * caller doesn't care about the disk-size, since our
2817 * return value implicitly indicates whether the
2818 * object even exists.
2820 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
) {
2822 if (stat_sha1_file(sha1
, &st
) < 0)
2825 *oi
->disk_sizep
= st
.st_size
;
2829 map
= map_sha1_file(sha1
, &mapsize
);
2833 *oi
->disk_sizep
= mapsize
;
2834 if ((flags
& LOOKUP_UNKNOWN_OBJECT
)) {
2835 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2836 status
= error("unable to unpack %s header with --allow-unknown-type",
2838 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2839 status
= error("unable to unpack %s header",
2843 else if (hdrbuf
.len
) {
2844 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2845 status
= error("unable to parse %s header with --allow-unknown-type",
2847 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2848 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2849 git_inflate_end(&stream
);
2850 munmap(map
, mapsize
);
2851 if (status
&& oi
->typep
)
2852 *oi
->typep
= status
;
2853 strbuf_release(&hdrbuf
);
2857 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2859 struct cached_object
*co
;
2860 struct pack_entry e
;
2862 enum object_type real_type
;
2863 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2865 co
= find_cached_object(real
);
2868 *(oi
->typep
) = co
->type
;
2870 *(oi
->sizep
) = co
->size
;
2872 *(oi
->disk_sizep
) = 0;
2873 if (oi
->delta_base_sha1
)
2874 hashclr(oi
->delta_base_sha1
);
2876 strbuf_addstr(oi
->typename
, typename(co
->type
));
2877 oi
->whence
= OI_CACHED
;
2881 if (!find_pack_entry(real
, &e
)) {
2882 /* Most likely it's a loose object. */
2883 if (!sha1_loose_object_info(real
, oi
, flags
)) {
2884 oi
->whence
= OI_LOOSE
;
2888 /* Not a loose object; someone else may have just packed it. */
2889 reprepare_packed_git();
2890 if (!find_pack_entry(real
, &e
))
2895 * packed_object_info() does not follow the delta chain to
2896 * find out the real type, unless it is given oi->typep.
2898 if (oi
->typename
&& !oi
->typep
)
2899 oi
->typep
= &real_type
;
2901 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2903 mark_bad_packed_object(e
.p
, real
);
2904 if (oi
->typep
== &real_type
)
2906 return sha1_object_info_extended(real
, oi
, 0);
2907 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2908 oi
->whence
= OI_DBCACHED
;
2910 oi
->whence
= OI_PACKED
;
2911 oi
->u
.packed
.offset
= e
.offset
;
2912 oi
->u
.packed
.pack
= e
.p
;
2913 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2914 rtype
== OBJ_OFS_DELTA
);
2917 strbuf_addstr(oi
->typename
, typename(*oi
->typep
));
2918 if (oi
->typep
== &real_type
)
2924 /* returns enum object_type or negative */
2925 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2927 enum object_type type
;
2928 struct object_info oi
= OBJECT_INFO_INIT
;
2932 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2937 static void *read_packed_sha1(const unsigned char *sha1
,
2938 enum object_type
*type
, unsigned long *size
)
2940 struct pack_entry e
;
2943 if (!find_pack_entry(sha1
, &e
))
2945 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
);
2948 * We're probably in deep shit, but let's try to fetch
2949 * the required object anyway from another pack or loose.
2950 * This should happen only in the presence of a corrupted
2951 * pack, and is better than failing outright.
2953 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2954 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2955 mark_bad_packed_object(e
.p
, sha1
);
2956 data
= read_object(sha1
, type
, size
);
2961 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2962 unsigned char *sha1
)
2964 struct cached_object
*co
;
2966 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2967 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2969 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2970 co
= &cached_objects
[cached_object_nr
++];
2973 co
->buf
= xmalloc(len
);
2974 memcpy(co
->buf
, buf
, len
);
2975 hashcpy(co
->sha1
, sha1
);
2979 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2980 unsigned long *size
)
2982 unsigned long mapsize
;
2984 struct cached_object
*co
;
2986 co
= find_cached_object(sha1
);
2990 return xmemdupz(co
->buf
, co
->size
);
2993 buf
= read_packed_sha1(sha1
, type
, size
);
2996 map
= map_sha1_file(sha1
, &mapsize
);
2998 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2999 munmap(map
, mapsize
);
3002 reprepare_packed_git();
3003 return read_packed_sha1(sha1
, type
, size
);
3007 * This function dies on corrupt objects; the callers who want to
3008 * deal with them should arrange to call read_object() and give error
3009 * messages themselves.
3011 void *read_sha1_file_extended(const unsigned char *sha1
,
3012 enum object_type
*type
,
3013 unsigned long *size
,
3017 const struct packed_git
*p
;
3018 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
3021 data
= read_object(repl
, type
, size
);
3025 if (errno
&& errno
!= ENOENT
)
3026 die_errno("failed to read object %s", sha1_to_hex(sha1
));
3028 /* die if we replaced an object with one that does not exist */
3030 die("replacement %s not found for %s",
3031 sha1_to_hex(repl
), sha1_to_hex(sha1
));
3033 if (has_loose_object(repl
)) {
3034 const char *path
= sha1_file_name(sha1
);
3036 die("loose object %s (stored in %s) is corrupt",
3037 sha1_to_hex(repl
), path
);
3040 if ((p
= has_packed_and_bad(repl
)) != NULL
)
3041 die("packed object %s (stored in %s) is corrupt",
3042 sha1_to_hex(repl
), p
->pack_name
);
3047 void *read_object_with_reference(const unsigned char *sha1
,
3048 const char *required_type_name
,
3049 unsigned long *size
,
3050 unsigned char *actual_sha1_return
)
3052 enum object_type type
, required_type
;
3054 unsigned long isize
;
3055 unsigned char actual_sha1
[20];
3057 required_type
= type_from_string(required_type_name
);
3058 hashcpy(actual_sha1
, sha1
);
3060 int ref_length
= -1;
3061 const char *ref_type
= NULL
;
3063 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
3066 if (type
== required_type
) {
3068 if (actual_sha1_return
)
3069 hashcpy(actual_sha1_return
, actual_sha1
);
3072 /* Handle references */
3073 else if (type
== OBJ_COMMIT
)
3075 else if (type
== OBJ_TAG
)
3076 ref_type
= "object ";
3081 ref_length
= strlen(ref_type
);
3083 if (ref_length
+ 40 > isize
||
3084 memcmp(buffer
, ref_type
, ref_length
) ||
3085 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
3090 /* Now we have the ID of the referred-to object in
3091 * actual_sha1. Check again. */
3095 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
3096 const char *type
, unsigned char *sha1
,
3097 char *hdr
, int *hdrlen
)
3101 /* Generate the header */
3102 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
3106 git_SHA1_Update(&c
, hdr
, *hdrlen
);
3107 git_SHA1_Update(&c
, buf
, len
);
3108 git_SHA1_Final(sha1
, &c
);
3112 * Move the just written object into its final resting place.
3114 int finalize_object_file(const char *tmpfile
, const char *filename
)
3118 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
3120 else if (link(tmpfile
, filename
))
3124 * Coda hack - coda doesn't like cross-directory links,
3125 * so we fall back to a rename, which will mean that it
3126 * won't be able to check collisions, but that's not a
3129 * The same holds for FAT formatted media.
3131 * When this succeeds, we just return. We have nothing
3134 if (ret
&& ret
!= EEXIST
) {
3136 if (!rename(tmpfile
, filename
))
3140 unlink_or_warn(tmpfile
);
3142 if (ret
!= EEXIST
) {
3143 return error_errno("unable to write sha1 filename %s", filename
);
3145 /* FIXME!!! Collision check here ? */
3149 if (adjust_shared_perm(filename
))
3150 return error("unable to set permission to '%s'", filename
);
3154 static int write_buffer(int fd
, const void *buf
, size_t len
)
3156 if (write_in_full(fd
, buf
, len
) < 0)
3157 return error_errno("file write error");
3161 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
3162 unsigned char *sha1
)
3165 int hdrlen
= sizeof(hdr
);
3166 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3170 /* Finalize a file on disk, and close it. */
3171 static void close_sha1_file(int fd
)
3173 if (fsync_object_files
)
3174 fsync_or_die(fd
, "sha1 file");
3176 die_errno("error when closing sha1 file");
3179 /* Size of directory component, including the ending '/' */
3180 static inline int directory_size(const char *filename
)
3182 const char *s
= strrchr(filename
, '/');
3185 return s
- filename
+ 1;
3189 * This creates a temporary file in the same directory as the final
3192 * We want to avoid cross-directory filename renames, because those
3193 * can have problems on various filesystems (FAT, NFS, Coda).
3195 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
3197 int fd
, dirlen
= directory_size(filename
);
3200 strbuf_add(tmp
, filename
, dirlen
);
3201 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
3202 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3203 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3205 * Make sure the directory exists; note that the contents
3206 * of the buffer are undefined after mkstemp returns an
3207 * error, so we have to rewrite the whole buffer from
3211 strbuf_add(tmp
, filename
, dirlen
- 1);
3212 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
3214 if (adjust_shared_perm(tmp
->buf
))
3218 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
3219 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3224 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3225 const void *buf
, unsigned long len
, time_t mtime
)
3228 unsigned char compressed
[4096];
3231 unsigned char parano_sha1
[20];
3232 static struct strbuf tmp_file
= STRBUF_INIT
;
3233 const char *filename
= sha1_file_name(sha1
);
3235 fd
= create_tmpfile(&tmp_file
, filename
);
3237 if (errno
== EACCES
)
3238 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3240 return error_errno("unable to create temporary file");
3244 git_deflate_init(&stream
, zlib_compression_level
);
3245 stream
.next_out
= compressed
;
3246 stream
.avail_out
= sizeof(compressed
);
3249 /* First header.. */
3250 stream
.next_in
= (unsigned char *)hdr
;
3251 stream
.avail_in
= hdrlen
;
3252 while (git_deflate(&stream
, 0) == Z_OK
)
3254 git_SHA1_Update(&c
, hdr
, hdrlen
);
3256 /* Then the data itself.. */
3257 stream
.next_in
= (void *)buf
;
3258 stream
.avail_in
= len
;
3260 unsigned char *in0
= stream
.next_in
;
3261 ret
= git_deflate(&stream
, Z_FINISH
);
3262 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3263 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3264 die("unable to write sha1 file");
3265 stream
.next_out
= compressed
;
3266 stream
.avail_out
= sizeof(compressed
);
3267 } while (ret
== Z_OK
);
3269 if (ret
!= Z_STREAM_END
)
3270 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3271 ret
= git_deflate_end_gently(&stream
);
3273 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3274 git_SHA1_Final(parano_sha1
, &c
);
3275 if (hashcmp(sha1
, parano_sha1
) != 0)
3276 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3278 close_sha1_file(fd
);
3283 utb
.modtime
= mtime
;
3284 if (utime(tmp_file
.buf
, &utb
) < 0)
3285 warning_errno("failed utime() on %s", tmp_file
.buf
);
3288 return finalize_object_file(tmp_file
.buf
, filename
);
3291 static int freshen_loose_object(const unsigned char *sha1
)
3293 return check_and_freshen(sha1
, 1);
3296 static int freshen_packed_object(const unsigned char *sha1
)
3298 struct pack_entry e
;
3299 if (!find_pack_entry(sha1
, &e
))
3303 if (!freshen_file(e
.p
->pack_name
))
3309 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3312 int hdrlen
= sizeof(hdr
);
3314 /* Normally if we have it in the pack then we do not bother writing
3315 * it out into .git/objects/??/?{38} file.
3317 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3318 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3320 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3323 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3324 unsigned char *sha1
, unsigned flags
)
3327 int hdrlen
, status
= 0;
3329 /* type string, SP, %lu of the length plus NUL must fit this */
3330 hdrlen
= strlen(type
) + 32;
3331 header
= xmalloc(hdrlen
);
3332 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3334 if (!(flags
& HASH_WRITE_OBJECT
))
3336 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3338 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3345 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3349 enum object_type type
;
3354 if (has_loose_object(sha1
))
3356 buf
= read_packed_sha1(sha1
, &type
, &len
);
3358 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3359 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3360 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3366 int has_pack_index(const unsigned char *sha1
)
3369 if (stat(sha1_pack_index_name(sha1
), &st
))
3374 int has_sha1_pack(const unsigned char *sha1
)
3376 struct pack_entry e
;
3377 return find_pack_entry(sha1
, &e
);
3380 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3382 struct pack_entry e
;
3384 if (find_pack_entry(sha1
, &e
))
3386 if (has_loose_object(sha1
))
3388 if (flags
& HAS_SHA1_QUICK
)
3390 reprepare_packed_git();
3391 return find_pack_entry(sha1
, &e
);
3394 int has_object_file(const struct object_id
*oid
)
3396 return has_sha1_file(oid
->hash
);
3399 int has_object_file_with_flags(const struct object_id
*oid
, int flags
)
3401 return has_sha1_file_with_flags(oid
->hash
, flags
);
3404 static void check_tree(const void *buf
, size_t size
)
3406 struct tree_desc desc
;
3407 struct name_entry entry
;
3409 init_tree_desc(&desc
, buf
, size
);
3410 while (tree_entry(&desc
, &entry
))
3412 * tree_entry() will die() on malformed entries */
3416 static void check_commit(const void *buf
, size_t size
)
3419 memset(&c
, 0, sizeof(c
));
3420 if (parse_commit_buffer(&c
, buf
, size
))
3421 die("corrupt commit");
3424 static void check_tag(const void *buf
, size_t size
)
3427 memset(&t
, 0, sizeof(t
));
3428 if (parse_tag_buffer(&t
, buf
, size
))
3432 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3433 enum object_type type
,
3434 const char *path
, unsigned flags
)
3436 int ret
, re_allocated
= 0;
3437 int write_object
= flags
& HASH_WRITE_OBJECT
;
3443 * Convert blobs to git internal format
3445 if ((type
== OBJ_BLOB
) && path
) {
3446 struct strbuf nbuf
= STRBUF_INIT
;
3447 if (convert_to_git(path
, buf
, size
, &nbuf
,
3448 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3449 buf
= strbuf_detach(&nbuf
, &size
);
3453 if (flags
& HASH_FORMAT_CHECK
) {
3454 if (type
== OBJ_TREE
)
3455 check_tree(buf
, size
);
3456 if (type
== OBJ_COMMIT
)
3457 check_commit(buf
, size
);
3458 if (type
== OBJ_TAG
)
3459 check_tag(buf
, size
);
3463 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3465 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3471 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3472 const char *path
, unsigned flags
)
3475 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3476 struct strbuf sbuf
= STRBUF_INIT
;
3479 assert(would_convert_to_git_filter_fd(path
));
3481 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3482 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3485 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3488 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3490 strbuf_release(&sbuf
);
3494 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3495 const char *path
, unsigned flags
)
3497 struct strbuf sbuf
= STRBUF_INIT
;
3500 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3501 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3504 strbuf_release(&sbuf
);
3508 #define SMALL_FILE_SIZE (32*1024)
3510 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3511 enum object_type type
, const char *path
,
3517 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3518 } else if (size
<= SMALL_FILE_SIZE
) {
3519 char *buf
= xmalloc(size
);
3520 if (size
== read_in_full(fd
, buf
, size
))
3521 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3523 ret
= error_errno("short read");
3526 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3527 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3534 * This creates one packfile per large blob unless bulk-checkin
3535 * machinery is "plugged".
3537 * This also bypasses the usual "convert-to-git" dance, and that is on
3538 * purpose. We could write a streaming version of the converting
3539 * functions and insert that before feeding the data to fast-import
3540 * (or equivalent in-core API described above). However, that is
3541 * somewhat complicated, as we do not know the size of the filter
3542 * result, which we need to know beforehand when writing a git object.
3543 * Since the primary motivation for trying to stream from the working
3544 * tree file and to avoid mmaping it in core is to deal with large
3545 * binary blobs, they generally do not want to get any conversion, and
3546 * callers should avoid this code path when filters are requested.
3548 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3549 enum object_type type
, const char *path
,
3552 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3555 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3556 enum object_type type
, const char *path
, unsigned flags
)
3561 * Call xsize_t() only when needed to avoid potentially unnecessary
3562 * die() for large files.
3564 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3565 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3566 else if (!S_ISREG(st
->st_mode
))
3567 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3568 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3569 (path
&& would_convert_to_git(path
)))
3570 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3573 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3579 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3582 struct strbuf sb
= STRBUF_INIT
;
3584 switch (st
->st_mode
& S_IFMT
) {
3586 fd
= open(path
, O_RDONLY
);
3588 return error_errno("open(\"%s\")", path
);
3589 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3590 return error("%s: failed to insert into database",
3594 if (strbuf_readlink(&sb
, path
, st
->st_size
))
3595 return error_errno("readlink(\"%s\")", path
);
3596 if (!(flags
& HASH_WRITE_OBJECT
))
3597 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3598 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3599 return error("%s: failed to insert into database",
3601 strbuf_release(&sb
);
3604 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3606 return error("%s: unsupported file type", path
);
3611 int read_pack_header(int fd
, struct pack_header
*header
)
3613 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3614 /* "eof before pack header was fully read" */
3615 return PH_ERROR_EOF
;
3617 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3618 /* "protocol error (pack signature mismatch detected)" */
3619 return PH_ERROR_PACK_SIGNATURE
;
3620 if (!pack_version_ok(header
->hdr_version
))
3621 /* "protocol error (pack version unsupported)" */
3622 return PH_ERROR_PROTOCOL
;
3626 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3628 enum object_type type
= sha1_object_info(sha1
, NULL
);
3630 die("%s is not a valid object", sha1_to_hex(sha1
));
3632 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3636 static int for_each_file_in_obj_subdir(int subdir_nr
,
3637 struct strbuf
*path
,
3638 each_loose_object_fn obj_cb
,
3639 each_loose_cruft_fn cruft_cb
,
3640 each_loose_subdir_fn subdir_cb
,
3643 size_t baselen
= path
->len
;
3644 DIR *dir
= opendir(path
->buf
);
3649 if (errno
== ENOENT
)
3651 return error_errno("unable to open %s", path
->buf
);
3654 while ((de
= readdir(dir
))) {
3655 if (is_dot_or_dotdot(de
->d_name
))
3658 strbuf_setlen(path
, baselen
);
3659 strbuf_addf(path
, "/%s", de
->d_name
);
3661 if (strlen(de
->d_name
) == 38) {
3663 unsigned char sha1
[20];
3665 snprintf(hex
, sizeof(hex
), "%02x%s",
3666 subdir_nr
, de
->d_name
);
3667 if (!get_sha1_hex(hex
, sha1
)) {
3669 r
= obj_cb(sha1
, path
->buf
, data
);
3678 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3685 strbuf_setlen(path
, baselen
);
3686 if (!r
&& subdir_cb
)
3687 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3692 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3693 each_loose_object_fn obj_cb
,
3694 each_loose_cruft_fn cruft_cb
,
3695 each_loose_subdir_fn subdir_cb
,
3698 size_t baselen
= path
->len
;
3702 for (i
= 0; i
< 256; i
++) {
3703 strbuf_addf(path
, "/%02x", i
);
3704 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3706 strbuf_setlen(path
, baselen
);
3714 int for_each_loose_file_in_objdir(const char *path
,
3715 each_loose_object_fn obj_cb
,
3716 each_loose_cruft_fn cruft_cb
,
3717 each_loose_subdir_fn subdir_cb
,
3720 struct strbuf buf
= STRBUF_INIT
;
3723 strbuf_addstr(&buf
, path
);
3724 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3726 strbuf_release(&buf
);
3731 struct loose_alt_odb_data
{
3732 each_loose_object_fn
*cb
;
3736 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3739 struct loose_alt_odb_data
*data
= vdata
;
3740 struct strbuf buf
= STRBUF_INIT
;
3743 strbuf_addstr(&buf
, alt
->path
);
3744 r
= for_each_loose_file_in_objdir_buf(&buf
,
3745 data
->cb
, NULL
, NULL
,
3747 strbuf_release(&buf
);
3751 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3753 struct loose_alt_odb_data alt
;
3756 r
= for_each_loose_file_in_objdir(get_object_directory(),
3757 cb
, NULL
, NULL
, data
);
3761 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3766 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3769 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3774 for (i
= 0; i
< p
->num_objects
; i
++) {
3775 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3778 return error("unable to get sha1 of object %u in %s",
3781 r
= cb(sha1
, p
, i
, data
);
3788 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3790 struct packed_git
*p
;
3792 int pack_errors
= 0;
3794 prepare_packed_git();
3795 for (p
= packed_git
; p
; p
= p
->next
) {
3796 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3798 if (open_pack_index(p
)) {
3802 r
= for_each_object_in_pack(p
, cb
, data
);
3806 return r
? r
: pack_errors
;