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 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
38 const unsigned char null_sha1
[20];
39 const struct object_id null_oid
;
42 * This is meant to hold a *small* number of objects that you would
43 * want read_sha1_file() to be able to return, but yet you do not want
44 * to write them into the object store (e.g. a browse-only
47 static struct cached_object
{
48 unsigned char sha1
[20];
49 enum object_type type
;
53 static int cached_object_nr
, cached_object_alloc
;
55 static struct cached_object empty_tree
= {
56 EMPTY_TREE_SHA1_BIN_LITERAL
,
63 * A pointer to the last packed_git in which an object was found.
64 * When an object is sought, we look in this packfile first, because
65 * objects that are looked up at similar times are often in the same
66 * packfile as one another.
68 static struct packed_git
*last_found_pack
;
70 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
73 struct cached_object
*co
= cached_objects
;
75 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
76 if (!hashcmp(co
->sha1
, sha1
))
79 if (!hashcmp(sha1
, empty_tree
.sha1
))
84 int mkdir_in_gitdir(const char *path
)
86 if (mkdir(path
, 0777)) {
87 int saved_errno
= errno
;
89 struct strbuf sb
= STRBUF_INIT
;
94 * Are we looking at a path in a symlinked worktree
95 * whose original repository does not yet have it?
96 * e.g. .git/rr-cache pointing at its original
97 * repository in which the user hasn't performed any
98 * conflict resolution yet?
100 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
101 strbuf_readlink(&sb
, path
, st
.st_size
) ||
102 !is_absolute_path(sb
.buf
) ||
103 mkdir(sb
.buf
, 0777)) {
110 return adjust_shared_perm(path
);
113 enum scld_error
safe_create_leading_directories(char *path
)
115 char *next_component
= path
+ offset_1st_component(path
);
116 enum scld_error ret
= SCLD_OK
;
118 while (ret
== SCLD_OK
&& next_component
) {
120 char *slash
= next_component
, slash_character
;
122 while (*slash
&& !is_dir_sep(*slash
))
128 next_component
= slash
+ 1;
129 while (is_dir_sep(*next_component
))
131 if (!*next_component
)
134 slash_character
= *slash
;
136 if (!stat(path
, &st
)) {
138 if (!S_ISDIR(st
.st_mode
))
140 } else if (mkdir(path
, 0777)) {
141 if (errno
== EEXIST
&&
142 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
143 ; /* somebody created it since we checked */
144 else if (errno
== ENOENT
)
146 * Either mkdir() failed because
147 * somebody just pruned the containing
148 * directory, or stat() failed because
149 * the file that was in our way was
150 * just removed. Either way, inform
151 * the caller that it might be worth
157 } else if (adjust_shared_perm(path
)) {
160 *slash
= slash_character
;
165 enum scld_error
safe_create_leading_directories_const(const char *path
)
167 /* path points to cache entries, so xstrdup before messing with it */
168 char *buf
= xstrdup(path
);
169 enum scld_error result
= safe_create_leading_directories(buf
);
174 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
177 for (i
= 0; i
< 20; i
++) {
178 static char hex
[] = "0123456789abcdef";
179 unsigned int val
= sha1
[i
];
180 char *pos
= pathbuf
+ i
*2 + (i
> 0);
181 *pos
++ = hex
[val
>> 4];
182 *pos
= hex
[val
& 0xf];
186 const char *sha1_file_name(const unsigned char *sha1
)
188 static char buf
[PATH_MAX
];
192 objdir
= get_object_directory();
193 len
= strlen(objdir
);
195 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
196 if (len
+ 43 > PATH_MAX
)
197 die("insanely long object directory %s", objdir
);
198 memcpy(buf
, objdir
, len
);
202 fill_sha1_path(buf
+ len
+ 1, sha1
);
207 * Return the name of the pack or index file with the specified sha1
208 * in its filename. *base and *name are scratch space that must be
209 * provided by the caller. which should be "pack" or "idx".
211 static char *sha1_get_pack_name(const unsigned char *sha1
,
216 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
217 sha1_to_hex(sha1
), which
);
221 char *sha1_pack_name(const unsigned char *sha1
)
223 static struct strbuf buf
= STRBUF_INIT
;
224 return sha1_get_pack_name(sha1
, &buf
, "pack");
227 char *sha1_pack_index_name(const unsigned char *sha1
)
229 static struct strbuf buf
= STRBUF_INIT
;
230 return sha1_get_pack_name(sha1
, &buf
, "idx");
233 struct alternate_object_database
*alt_odb_list
;
234 static struct alternate_object_database
**alt_odb_tail
;
237 * Prepare alternate object database registry.
239 * The variable alt_odb_list points at the list of struct
240 * alternate_object_database. The elements on this list come from
241 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
242 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
243 * whose contents is similar to that environment variable but can be
244 * LF separated. Its base points at a statically allocated buffer that
245 * contains "/the/directory/corresponding/to/.git/objects/...", while
246 * its name points just after the slash at the end of ".git/objects/"
247 * in the example above, and has enough space to hold 40-byte hex
248 * SHA1, an extra slash for the first level indirection, and the
251 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
252 int depth
, const char *normalized_objdir
)
254 struct alternate_object_database
*ent
;
255 struct alternate_object_database
*alt
;
257 struct strbuf pathbuf
= STRBUF_INIT
;
259 if (!is_absolute_path(entry
) && relative_base
) {
260 strbuf_addstr(&pathbuf
, real_path(relative_base
));
261 strbuf_addch(&pathbuf
, '/');
263 strbuf_addstr(&pathbuf
, entry
);
265 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
267 pfxlen
= strlen(pathbuf
.buf
);
270 * The trailing slash after the directory name is given by
271 * this function at the end. Remove duplicates.
273 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
276 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
277 ent
= xmalloc(sizeof(*ent
) + entlen
);
278 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
279 strbuf_release(&pathbuf
);
281 ent
->name
= ent
->base
+ pfxlen
+ 1;
282 ent
->base
[pfxlen
+ 3] = '/';
283 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
285 /* Detect cases where alternate disappeared */
286 if (!is_directory(ent
->base
)) {
287 error("object directory %s does not exist; "
288 "check .git/objects/info/alternates.",
294 /* Prevent the common mistake of listing the same
295 * thing twice, or object directory itself.
297 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
298 if (pfxlen
== alt
->name
- alt
->base
- 1 &&
299 !memcmp(ent
->base
, alt
->base
, pfxlen
)) {
304 if (!strcmp_icase(ent
->base
, normalized_objdir
)) {
309 /* add the alternate entry */
311 alt_odb_tail
= &(ent
->next
);
314 /* recursively add alternates */
315 read_info_alternates(ent
->base
, depth
+ 1);
317 ent
->base
[pfxlen
] = '/';
322 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
323 const char *relative_base
, int depth
)
325 struct string_list entries
= STRING_LIST_INIT_NODUP
;
328 struct strbuf objdirbuf
= STRBUF_INIT
;
331 error("%s: ignoring alternate object stores, nesting too deep.",
336 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
337 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
339 alt_copy
= xmemdupz(alt
, len
);
340 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
341 for (i
= 0; i
< entries
.nr
; i
++) {
342 const char *entry
= entries
.items
[i
].string
;
343 if (entry
[0] == '\0' || entry
[0] == '#')
345 if (!is_absolute_path(entry
) && depth
) {
346 error("%s: ignoring relative alternate object store %s",
347 relative_base
, entry
);
349 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
352 string_list_clear(&entries
, 0);
354 strbuf_release(&objdirbuf
);
357 void read_info_alternates(const char * relative_base
, int depth
)
365 path
= xstrfmt("%s/info/alternates", relative_base
);
366 fd
= git_open_noatime(path
);
370 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
374 mapsz
= xsize_t(st
.st_size
);
375 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
378 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
383 void add_to_alternates_file(const char *reference
)
385 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
386 char *alts
= git_pathdup("objects/info/alternates");
389 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
390 out
= fdopen_lock_file(lock
, "w");
392 die_errno("unable to fdopen alternates lockfile");
394 in
= fopen(alts
, "r");
396 struct strbuf line
= STRBUF_INIT
;
399 while (strbuf_getline(&line
, in
) != EOF
) {
400 if (!strcmp(reference
, line
.buf
)) {
404 fprintf_or_die(out
, "%s\n", line
.buf
);
407 strbuf_release(&line
);
411 rollback_lock_file(lock
);
415 else if (errno
!= ENOENT
)
416 die_errno("unable to read alternates file");
419 fprintf_or_die(out
, "%s\n", reference
);
420 if (commit_lock_file(lock
))
421 die_errno("unable to move new alternates file into place");
423 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
428 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
430 struct alternate_object_database
*ent
;
434 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
442 void prepare_alt_odb(void)
449 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
452 alt_odb_tail
= &alt_odb_list
;
453 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
455 read_info_alternates(get_object_directory(), 0);
458 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
459 static int freshen_file(const char *fn
)
462 t
.actime
= t
.modtime
= time(NULL
);
463 return !utime(fn
, &t
);
467 * All of the check_and_freshen functions return 1 if the file exists and was
468 * freshened (if freshening was requested), 0 otherwise. If they return
469 * 0, you should not assume that it is safe to skip a write of the object (it
470 * either does not exist on disk, or has a stale mtime and may be subject to
473 static int check_and_freshen_file(const char *fn
, int freshen
)
475 if (access(fn
, F_OK
))
477 if (freshen
&& !freshen_file(fn
))
482 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
484 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
487 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
489 struct alternate_object_database
*alt
;
491 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
492 fill_sha1_path(alt
->name
, sha1
);
493 if (check_and_freshen_file(alt
->base
, freshen
))
499 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
501 return check_and_freshen_local(sha1
, freshen
) ||
502 check_and_freshen_nonlocal(sha1
, freshen
);
505 int has_loose_object_nonlocal(const unsigned char *sha1
)
507 return check_and_freshen_nonlocal(sha1
, 0);
510 static int has_loose_object(const unsigned char *sha1
)
512 return check_and_freshen(sha1
, 0);
515 static unsigned int pack_used_ctr
;
516 static unsigned int pack_mmap_calls
;
517 static unsigned int peak_pack_open_windows
;
518 static unsigned int pack_open_windows
;
519 static unsigned int pack_open_fds
;
520 static unsigned int pack_max_fds
;
521 static size_t peak_pack_mapped
;
522 static size_t pack_mapped
;
523 struct packed_git
*packed_git
;
525 void pack_report(void)
528 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
529 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
530 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
531 sz_fmt(getpagesize()),
532 sz_fmt(packed_git_window_size
),
533 sz_fmt(packed_git_limit
));
535 "pack_report: pack_used_ctr = %10u\n"
536 "pack_report: pack_mmap_calls = %10u\n"
537 "pack_report: pack_open_windows = %10u / %10u\n"
538 "pack_report: pack_mapped = "
539 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
542 pack_open_windows
, peak_pack_open_windows
,
543 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
547 * Open and mmap the index file at path, perform a couple of
548 * consistency checks, then record its information to p. Return 0 on
551 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
554 struct pack_idx_header
*hdr
;
556 uint32_t version
, nr
, i
, *index
;
557 int fd
= git_open_noatime(path
);
562 if (fstat(fd
, &st
)) {
566 idx_size
= xsize_t(st
.st_size
);
567 if (idx_size
< 4 * 256 + 20 + 20) {
569 return error("index file %s is too small", path
);
571 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
575 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
576 version
= ntohl(hdr
->idx_version
);
577 if (version
< 2 || version
> 2) {
578 munmap(idx_map
, idx_size
);
579 return error("index file %s is version %"PRIu32
580 " and is not supported by this binary"
581 " (try upgrading GIT to a newer version)",
590 index
+= 2; /* skip index header */
591 for (i
= 0; i
< 256; i
++) {
592 uint32_t n
= ntohl(index
[i
]);
594 munmap(idx_map
, idx_size
);
595 return error("non-monotonic index %s", path
);
603 * - 256 index entries 4 bytes each
604 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
608 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
609 munmap(idx_map
, idx_size
);
610 return error("wrong index v1 file size in %s", path
);
612 } else if (version
== 2) {
615 * - 8 bytes of header
616 * - 256 index entries 4 bytes each
617 * - 20-byte sha1 entry * nr
618 * - 4-byte crc entry * nr
619 * - 4-byte offset entry * nr
620 * - 20-byte SHA1 of the packfile
621 * - 20-byte SHA1 file checksum
622 * And after the 4-byte offset table might be a
623 * variable sized table containing 8-byte entries
624 * for offsets larger than 2^31.
626 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
627 unsigned long max_size
= min_size
;
629 max_size
+= (nr
- 1)*8;
630 if (idx_size
< min_size
|| idx_size
> max_size
) {
631 munmap(idx_map
, idx_size
);
632 return error("wrong index v2 file size in %s", path
);
634 if (idx_size
!= min_size
&&
636 * make sure we can deal with large pack offsets.
637 * 31-bit signed offset won't be enough, neither
638 * 32-bit unsigned one will be.
640 (sizeof(off_t
) <= 4)) {
641 munmap(idx_map
, idx_size
);
642 return error("pack too large for current definition of off_t in %s", path
);
646 p
->index_version
= version
;
647 p
->index_data
= idx_map
;
648 p
->index_size
= idx_size
;
653 int open_pack_index(struct packed_git
*p
)
662 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
663 die("BUG: pack_name does not end in .pack");
664 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
665 ret
= check_packed_git_idx(idx_name
, p
);
670 static void scan_windows(struct packed_git
*p
,
671 struct packed_git
**lru_p
,
672 struct pack_window
**lru_w
,
673 struct pack_window
**lru_l
)
675 struct pack_window
*w
, *w_l
;
677 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
679 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
689 static int unuse_one_window(struct packed_git
*current
)
691 struct packed_git
*p
, *lru_p
= NULL
;
692 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
695 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
696 for (p
= packed_git
; p
; p
= p
->next
)
697 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
699 munmap(lru_w
->base
, lru_w
->len
);
700 pack_mapped
-= lru_w
->len
;
702 lru_l
->next
= lru_w
->next
;
704 lru_p
->windows
= lru_w
->next
;
712 void release_pack_memory(size_t need
)
714 size_t cur
= pack_mapped
;
715 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
719 static void mmap_limit_check(size_t length
)
721 static size_t limit
= 0;
723 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
728 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
729 (uintmax_t)length
, (uintmax_t)limit
);
732 void *xmmap_gently(void *start
, size_t length
,
733 int prot
, int flags
, int fd
, off_t offset
)
737 mmap_limit_check(length
);
738 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
739 if (ret
== MAP_FAILED
) {
742 release_pack_memory(length
);
743 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
748 void *xmmap(void *start
, size_t length
,
749 int prot
, int flags
, int fd
, off_t offset
)
751 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
752 if (ret
== MAP_FAILED
)
753 die_errno("mmap failed");
757 void close_pack_windows(struct packed_git
*p
)
760 struct pack_window
*w
= p
->windows
;
763 die("pack '%s' still has open windows to it",
765 munmap(w
->base
, w
->len
);
766 pack_mapped
-= w
->len
;
768 p
->windows
= w
->next
;
773 static int close_pack_fd(struct packed_git
*p
)
785 static void close_pack(struct packed_git
*p
)
787 close_pack_windows(p
);
792 void close_all_packs(void)
794 struct packed_git
*p
;
796 for (p
= packed_git
; p
; p
= p
->next
)
798 die("BUG! Want to close pack marked 'do-not-close'");
805 * The LRU pack is the one with the oldest MRU window, preferring packs
806 * with no used windows, or the oldest mtime if it has no windows allocated.
808 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
810 struct pack_window
*w
, *this_mru_w
;
811 int has_windows_inuse
= 0;
814 * Reject this pack if it has windows and the previously selected
815 * one does not. If this pack does not have windows, reject
816 * it if the pack file is newer than the previously selected one.
818 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
821 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
823 * Reject this pack if any of its windows are in use,
824 * but the previously selected pack did not have any
825 * inuse windows. Otherwise, record that this pack
826 * has windows in use.
829 if (*accept_windows_inuse
)
830 has_windows_inuse
= 1;
835 if (w
->last_used
> this_mru_w
->last_used
)
839 * Reject this pack if it has windows that have been
840 * used more recently than the previously selected pack.
841 * If the previously selected pack had windows inuse and
842 * we have not encountered a window in this pack that is
843 * inuse, skip this check since we prefer a pack with no
844 * inuse windows to one that has inuse windows.
846 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
847 this_mru_w
->last_used
> (*mru_w
)->last_used
)
856 *accept_windows_inuse
= has_windows_inuse
;
859 static int close_one_pack(void)
861 struct packed_git
*p
, *lru_p
= NULL
;
862 struct pack_window
*mru_w
= NULL
;
863 int accept_windows_inuse
= 1;
865 for (p
= packed_git
; p
; p
= p
->next
) {
866 if (p
->pack_fd
== -1)
868 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
872 return close_pack_fd(lru_p
);
877 void unuse_pack(struct pack_window
**w_cursor
)
879 struct pack_window
*w
= *w_cursor
;
886 void close_pack_index(struct packed_git
*p
)
889 munmap((void *)p
->index_data
, p
->index_size
);
890 p
->index_data
= NULL
;
895 * This is used by git-repack in case a newly created pack happens to
896 * contain the same set of objects as an existing one. In that case
897 * the resulting file might be different even if its name would be the
898 * same. It is best to close any reference to the old pack before it is
899 * replaced on disk. Of course no index pointers or windows for given pack
900 * must subsist at this point. If ever objects from this pack are requested
901 * again, the new version of the pack will be reinitialized through
902 * reprepare_packed_git().
904 void free_pack_by_name(const char *pack_name
)
906 struct packed_git
*p
, **pp
= &packed_git
;
910 if (strcmp(pack_name
, p
->pack_name
) == 0) {
911 clear_delta_base_cache();
913 free(p
->bad_object_sha1
);
915 if (last_found_pack
== p
)
916 last_found_pack
= NULL
;
924 static unsigned int get_max_fd_limit(void)
930 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
937 long open_max
= sysconf(_SC_OPEN_MAX
);
941 * Otherwise, we got -1 for one of the two
944 * (1) sysconf() did not understand _SC_OPEN_MAX
945 * and signaled an error with -1; or
946 * (2) sysconf() said there is no limit.
948 * We _could_ clear errno before calling sysconf() to
949 * tell these two cases apart and return a huge number
950 * in the latter case to let the caller cap it to a
951 * value that is not so selfish, but letting the
952 * fallback OPEN_MAX codepath take care of these cases
961 return 1; /* see the caller ;-) */
966 * Do not call this directly as this leaks p->pack_fd on error return;
967 * call open_packed_git() instead.
969 static int open_packed_git_1(struct packed_git
*p
)
972 struct pack_header hdr
;
973 unsigned char sha1
[20];
974 unsigned char *idx_sha1
;
977 if (!p
->index_data
&& open_pack_index(p
))
978 return error("packfile %s index unavailable", p
->pack_name
);
981 unsigned int max_fds
= get_max_fd_limit();
983 /* Save 3 for stdin/stdout/stderr, 22 for work */
985 pack_max_fds
= max_fds
- 25;
990 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
993 p
->pack_fd
= git_open_noatime(p
->pack_name
);
994 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
998 /* If we created the struct before we had the pack we lack size. */
1000 if (!S_ISREG(st
.st_mode
))
1001 return error("packfile %s not a regular file", p
->pack_name
);
1002 p
->pack_size
= st
.st_size
;
1003 } else if (p
->pack_size
!= st
.st_size
)
1004 return error("packfile %s size changed", p
->pack_name
);
1006 /* We leave these file descriptors open with sliding mmap;
1007 * there is no point keeping them open across exec(), though.
1009 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
1011 return error("cannot determine file descriptor flags");
1012 fd_flag
|= FD_CLOEXEC
;
1013 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
1014 return error("cannot set FD_CLOEXEC");
1016 /* Verify we recognize this pack file format. */
1017 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
1018 return error("file %s is far too short to be a packfile", p
->pack_name
);
1019 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
1020 return error("file %s is not a GIT packfile", p
->pack_name
);
1021 if (!pack_version_ok(hdr
.hdr_version
))
1022 return error("packfile %s is version %"PRIu32
" and not"
1023 " supported (try upgrading GIT to a newer version)",
1024 p
->pack_name
, ntohl(hdr
.hdr_version
));
1026 /* Verify the pack matches its index. */
1027 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
1028 return error("packfile %s claims to have %"PRIu32
" objects"
1029 " while index indicates %"PRIu32
" objects",
1030 p
->pack_name
, ntohl(hdr
.hdr_entries
),
1032 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1033 return error("end of packfile %s is unavailable", p
->pack_name
);
1034 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1035 return error("packfile %s signature is unavailable", p
->pack_name
);
1036 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1037 if (hashcmp(sha1
, idx_sha1
))
1038 return error("packfile %s does not match index", p
->pack_name
);
1042 static int open_packed_git(struct packed_git
*p
)
1044 if (!open_packed_git_1(p
))
1050 static int in_window(struct pack_window
*win
, off_t offset
)
1052 /* We must promise at least 20 bytes (one hash) after the
1053 * offset is available from this window, otherwise the offset
1054 * is not actually in this window and a different window (which
1055 * has that one hash excess) must be used. This is to support
1056 * the object header and delta base parsing routines below.
1058 off_t win_off
= win
->offset
;
1059 return win_off
<= offset
1060 && (offset
+ 20) <= (win_off
+ win
->len
);
1063 unsigned char *use_pack(struct packed_git
*p
,
1064 struct pack_window
**w_cursor
,
1066 unsigned long *left
)
1068 struct pack_window
*win
= *w_cursor
;
1070 /* Since packfiles end in a hash of their content and it's
1071 * pointless to ask for an offset into the middle of that
1072 * hash, and the in_window function above wouldn't match
1073 * don't allow an offset too close to the end of the file.
1075 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1076 die("packfile %s cannot be accessed", p
->pack_name
);
1077 if (offset
> (p
->pack_size
- 20))
1078 die("offset beyond end of packfile (truncated pack?)");
1080 if (!win
|| !in_window(win
, offset
)) {
1083 for (win
= p
->windows
; win
; win
= win
->next
) {
1084 if (in_window(win
, offset
))
1088 size_t window_align
= packed_git_window_size
/ 2;
1091 if (p
->pack_fd
== -1 && open_packed_git(p
))
1092 die("packfile %s cannot be accessed", p
->pack_name
);
1094 win
= xcalloc(1, sizeof(*win
));
1095 win
->offset
= (offset
/ window_align
) * window_align
;
1096 len
= p
->pack_size
- win
->offset
;
1097 if (len
> packed_git_window_size
)
1098 len
= packed_git_window_size
;
1099 win
->len
= (size_t)len
;
1100 pack_mapped
+= win
->len
;
1101 while (packed_git_limit
< pack_mapped
1102 && unuse_one_window(p
))
1104 win
->base
= xmmap(NULL
, win
->len
,
1105 PROT_READ
, MAP_PRIVATE
,
1106 p
->pack_fd
, win
->offset
);
1107 if (win
->base
== MAP_FAILED
)
1108 die("packfile %s cannot be mapped: %s",
1111 if (!win
->offset
&& win
->len
== p
->pack_size
1112 && !p
->do_not_close
)
1115 pack_open_windows
++;
1116 if (pack_mapped
> peak_pack_mapped
)
1117 peak_pack_mapped
= pack_mapped
;
1118 if (pack_open_windows
> peak_pack_open_windows
)
1119 peak_pack_open_windows
= pack_open_windows
;
1120 win
->next
= p
->windows
;
1124 if (win
!= *w_cursor
) {
1125 win
->last_used
= pack_used_ctr
++;
1129 offset
-= win
->offset
;
1131 *left
= win
->len
- xsize_t(offset
);
1132 return win
->base
+ offset
;
1135 static struct packed_git
*alloc_packed_git(int extra
)
1137 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1138 memset(p
, 0, sizeof(*p
));
1143 static void try_to_free_pack_memory(size_t size
)
1145 release_pack_memory(size
);
1148 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1150 static int have_set_try_to_free_routine
;
1153 struct packed_git
*p
;
1155 if (!have_set_try_to_free_routine
) {
1156 have_set_try_to_free_routine
= 1;
1157 set_try_to_free_routine(try_to_free_pack_memory
);
1161 * Make sure a corresponding .pack file exists and that
1162 * the index looks sane.
1164 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1168 * ".pack" is long enough to hold any suffix we're adding (and
1169 * the use xsnprintf double-checks that)
1171 alloc
= path_len
+ strlen(".pack") + 1;
1172 p
= alloc_packed_git(alloc
);
1173 memcpy(p
->pack_name
, path
, path_len
);
1175 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1176 if (!access(p
->pack_name
, F_OK
))
1179 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1180 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1185 /* ok, it looks sane as far as we can check without
1186 * actually mapping the pack file.
1188 p
->pack_size
= st
.st_size
;
1189 p
->pack_local
= local
;
1190 p
->mtime
= st
.st_mtime
;
1191 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1196 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1198 const char *path
= sha1_pack_name(sha1
);
1199 int alloc
= strlen(path
) + 1;
1200 struct packed_git
*p
= alloc_packed_git(alloc
);
1202 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
1203 hashcpy(p
->sha1
, sha1
);
1204 if (check_packed_git_idx(idx_path
, p
)) {
1212 void install_packed_git(struct packed_git
*pack
)
1214 if (pack
->pack_fd
!= -1)
1217 pack
->next
= packed_git
;
1221 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
1223 static void report_helper(const struct string_list
*list
,
1224 int seen_bits
, int first
, int last
)
1226 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
1229 for (; first
< last
; first
++)
1230 report_garbage(seen_bits
, list
->items
[first
].string
);
1233 static void report_pack_garbage(struct string_list
*list
)
1235 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1237 if (!report_garbage
)
1240 string_list_sort(list
);
1242 for (i
= 0; i
< list
->nr
; i
++) {
1243 const char *path
= list
->items
[i
].string
;
1244 if (baselen
!= -1 &&
1245 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1246 report_helper(list
, seen_bits
, first
, i
);
1250 if (baselen
== -1) {
1251 const char *dot
= strrchr(path
, '.');
1253 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
1256 baselen
= dot
- path
+ 1;
1259 if (!strcmp(path
+ baselen
, "pack"))
1261 else if (!strcmp(path
+ baselen
, "idx"))
1264 report_helper(list
, seen_bits
, first
, list
->nr
);
1267 static void prepare_packed_git_one(char *objdir
, int local
)
1269 struct strbuf path
= STRBUF_INIT
;
1273 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1275 strbuf_addstr(&path
, objdir
);
1276 strbuf_addstr(&path
, "/pack");
1277 dir
= opendir(path
.buf
);
1279 if (errno
!= ENOENT
)
1280 error("unable to open object pack directory: %s: %s",
1281 path
.buf
, strerror(errno
));
1282 strbuf_release(&path
);
1285 strbuf_addch(&path
, '/');
1286 dirnamelen
= path
.len
;
1287 while ((de
= readdir(dir
)) != NULL
) {
1288 struct packed_git
*p
;
1291 if (is_dot_or_dotdot(de
->d_name
))
1294 strbuf_setlen(&path
, dirnamelen
);
1295 strbuf_addstr(&path
, de
->d_name
);
1297 base_len
= path
.len
;
1298 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1299 /* Don't reopen a pack we already have. */
1300 for (p
= packed_git
; p
; p
= p
->next
) {
1302 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1304 !memcmp(p
->pack_name
, path
.buf
, len
))
1309 * See if it really is a valid .idx file with
1310 * corresponding .pack file that we can map.
1312 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1313 install_packed_git(p
);
1316 if (!report_garbage
)
1319 if (ends_with(de
->d_name
, ".idx") ||
1320 ends_with(de
->d_name
, ".pack") ||
1321 ends_with(de
->d_name
, ".bitmap") ||
1322 ends_with(de
->d_name
, ".keep"))
1323 string_list_append(&garbage
, path
.buf
);
1325 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
1328 report_pack_garbage(&garbage
);
1329 string_list_clear(&garbage
, 0);
1330 strbuf_release(&path
);
1333 static int sort_pack(const void *a_
, const void *b_
)
1335 struct packed_git
*a
= *((struct packed_git
**)a_
);
1336 struct packed_git
*b
= *((struct packed_git
**)b_
);
1340 * Local packs tend to contain objects specific to our
1341 * variant of the project than remote ones. In addition,
1342 * remote ones could be on a network mounted filesystem.
1343 * Favor local ones for these reasons.
1345 st
= a
->pack_local
- b
->pack_local
;
1350 * Younger packs tend to contain more recent objects,
1351 * and more recent objects tend to get accessed more
1354 if (a
->mtime
< b
->mtime
)
1356 else if (a
->mtime
== b
->mtime
)
1361 static void rearrange_packed_git(void)
1363 struct packed_git
**ary
, *p
;
1366 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1371 /* prepare an array of packed_git for easier sorting */
1372 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1373 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1376 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1378 /* link them back again */
1379 for (i
= 0; i
< n
- 1; i
++)
1380 ary
[i
]->next
= ary
[i
+ 1];
1381 ary
[n
- 1]->next
= NULL
;
1382 packed_git
= ary
[0];
1387 static int prepare_packed_git_run_once
= 0;
1388 void prepare_packed_git(void)
1390 struct alternate_object_database
*alt
;
1392 if (prepare_packed_git_run_once
)
1394 prepare_packed_git_one(get_object_directory(), 1);
1396 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1398 prepare_packed_git_one(alt
->base
, 0);
1399 alt
->name
[-1] = '/';
1401 rearrange_packed_git();
1402 prepare_packed_git_run_once
= 1;
1405 void reprepare_packed_git(void)
1407 prepare_packed_git_run_once
= 0;
1408 prepare_packed_git();
1411 static void mark_bad_packed_object(struct packed_git
*p
,
1412 const unsigned char *sha1
)
1415 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1416 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1418 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1419 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1420 p
->num_bad_objects
++;
1423 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1425 struct packed_git
*p
;
1428 for (p
= packed_git
; p
; p
= p
->next
)
1429 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1430 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1436 * With an in-core object data in "map", rehash it to make sure the
1437 * object name actually matches "sha1" to detect object corruption.
1438 * With "map" == NULL, try reading the object named with "sha1" using
1439 * the streaming interface and rehash it to do the same.
1441 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1442 unsigned long size
, const char *type
)
1444 unsigned char real_sha1
[20];
1445 enum object_type obj_type
;
1446 struct git_istream
*st
;
1452 hash_sha1_file(map
, size
, type
, real_sha1
);
1453 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1456 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1460 /* Generate the header */
1461 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1465 git_SHA1_Update(&c
, hdr
, hdrlen
);
1467 char buf
[1024 * 16];
1468 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1476 git_SHA1_Update(&c
, buf
, readlen
);
1478 git_SHA1_Final(real_sha1
, &c
);
1480 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1483 int git_open_noatime(const char *name
)
1485 static int sha1_file_open_flag
= O_NOATIME
;
1491 fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1495 /* Might the failure be due to O_NOATIME? */
1496 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1497 sha1_file_open_flag
= 0;
1505 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1507 struct alternate_object_database
*alt
;
1509 if (!lstat(sha1_file_name(sha1
), st
))
1514 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1515 fill_sha1_path(alt
->name
, sha1
);
1516 if (!lstat(alt
->base
, st
))
1523 static int open_sha1_file(const unsigned char *sha1
)
1526 struct alternate_object_database
*alt
;
1527 int most_interesting_errno
;
1529 fd
= git_open_noatime(sha1_file_name(sha1
));
1532 most_interesting_errno
= errno
;
1535 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1536 fill_sha1_path(alt
->name
, sha1
);
1537 fd
= git_open_noatime(alt
->base
);
1540 if (most_interesting_errno
== ENOENT
)
1541 most_interesting_errno
= errno
;
1543 errno
= most_interesting_errno
;
1547 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1552 fd
= open_sha1_file(sha1
);
1557 if (!fstat(fd
, &st
)) {
1558 *size
= xsize_t(st
.st_size
);
1560 /* mmap() is forbidden on empty files */
1561 error("object file %s is empty", sha1_file_name(sha1
));
1564 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1571 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1572 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1575 unsigned long size
, c
;
1576 unsigned long used
= 0;
1579 *type
= (c
>> 4) & 7;
1583 if (len
<= used
|| bitsizeof(long) <= shift
) {
1584 error("bad object header");
1589 size
+= (c
& 0x7f) << shift
;
1596 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1598 /* Get the data stream */
1599 memset(stream
, 0, sizeof(*stream
));
1600 stream
->next_in
= map
;
1601 stream
->avail_in
= mapsize
;
1602 stream
->next_out
= buffer
;
1603 stream
->avail_out
= bufsiz
;
1605 git_inflate_init(stream
);
1606 return git_inflate(stream
, 0);
1609 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1610 unsigned long mapsize
, void *buffer
,
1611 unsigned long bufsiz
, struct strbuf
*header
)
1615 status
= unpack_sha1_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1618 * Check if entire header is unpacked in the first iteration.
1620 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1624 * buffer[0..bufsiz] was not large enough. Copy the partial
1625 * result out to header, and then append the result of further
1626 * reading the stream.
1628 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1629 stream
->next_out
= buffer
;
1630 stream
->avail_out
= bufsiz
;
1633 status
= git_inflate(stream
, 0);
1634 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1635 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1637 stream
->next_out
= buffer
;
1638 stream
->avail_out
= bufsiz
;
1639 } while (status
!= Z_STREAM_END
);
1643 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1645 int bytes
= strlen(buffer
) + 1;
1646 unsigned char *buf
= xmallocz(size
);
1650 n
= stream
->total_out
- bytes
;
1653 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1655 if (bytes
<= size
) {
1657 * The above condition must be (bytes <= size), not
1658 * (bytes < size). In other words, even though we
1659 * expect no more output and set avail_out to zero,
1660 * the input zlib stream may have bytes that express
1661 * "this concludes the stream", and we *do* want to
1664 * Otherwise we would not be able to test that we
1665 * consumed all the input to reach the expected size;
1666 * we also want to check that zlib tells us that all
1667 * went well with status == Z_STREAM_END at the end.
1669 stream
->next_out
= buf
+ bytes
;
1670 stream
->avail_out
= size
- bytes
;
1671 while (status
== Z_OK
)
1672 status
= git_inflate(stream
, Z_FINISH
);
1674 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1675 git_inflate_end(stream
);
1680 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1681 else if (stream
->avail_in
)
1682 error("garbage at end of loose object '%s'",
1689 * We used to just use "sscanf()", but that's actually way
1690 * too permissive for what we want to check. So do an anal
1691 * object header parse by hand.
1693 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1696 const char *type_buf
= hdr
;
1698 int type
, type_len
= 0;
1701 * The type can be of any size but is followed by
1711 type
= type_from_string_gently(type_buf
, type_len
, 1);
1713 strbuf_add(oi
->typename
, type_buf
, type_len
);
1715 * Set type to 0 if its an unknown object and
1716 * we're obtaining the type using '--allow-unkown-type'
1719 if ((flags
& LOOKUP_UNKNOWN_OBJECT
) && (type
< 0))
1722 die("invalid object type");
1727 * The length must follow immediately, and be in canonical
1728 * decimal format (ie "010" is not valid).
1730 size
= *hdr
++ - '0';
1735 unsigned long c
= *hdr
- '0';
1739 size
= size
* 10 + c
;
1747 * The length must be followed by a zero byte
1749 return *hdr
? -1 : type
;
1752 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1754 struct object_info oi
;
1759 return parse_sha1_header_extended(hdr
, &oi
, LOOKUP_REPLACE_OBJECT
);
1762 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1768 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1769 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1772 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1775 unsigned long get_size_from_delta(struct packed_git
*p
,
1776 struct pack_window
**w_curs
,
1779 const unsigned char *data
;
1780 unsigned char delta_head
[20], *in
;
1784 memset(&stream
, 0, sizeof(stream
));
1785 stream
.next_out
= delta_head
;
1786 stream
.avail_out
= sizeof(delta_head
);
1788 git_inflate_init(&stream
);
1790 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1791 stream
.next_in
= in
;
1792 st
= git_inflate(&stream
, Z_FINISH
);
1793 curpos
+= stream
.next_in
- in
;
1794 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1795 stream
.total_out
< sizeof(delta_head
));
1796 git_inflate_end(&stream
);
1797 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1798 error("delta data unpack-initial failed");
1802 /* Examine the initial part of the delta to figure out
1807 /* ignore base size */
1808 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1810 /* Read the result size */
1811 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1814 static off_t
get_delta_base(struct packed_git
*p
,
1815 struct pack_window
**w_curs
,
1817 enum object_type type
,
1818 off_t delta_obj_offset
)
1820 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1823 /* use_pack() assured us we have [base_info, base_info + 20)
1824 * as a range that we can look at without walking off the
1825 * end of the mapped window. Its actually the hash size
1826 * that is assured. An OFS_DELTA longer than the hash size
1827 * is stupid, as then a REF_DELTA would be smaller to store.
1829 if (type
== OBJ_OFS_DELTA
) {
1831 unsigned char c
= base_info
[used
++];
1832 base_offset
= c
& 127;
1835 if (!base_offset
|| MSB(base_offset
, 7))
1836 return 0; /* overflow */
1837 c
= base_info
[used
++];
1838 base_offset
= (base_offset
<< 7) + (c
& 127);
1840 base_offset
= delta_obj_offset
- base_offset
;
1841 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1842 return 0; /* out of bound */
1844 } else if (type
== OBJ_REF_DELTA
) {
1845 /* The base entry _must_ be in the same pack */
1846 base_offset
= find_pack_entry_one(base_info
, p
);
1849 die("I am totally screwed");
1854 * Like get_delta_base above, but we return the sha1 instead of the pack
1855 * offset. This means it is cheaper for REF deltas (we do not have to do
1856 * the final object lookup), but more expensive for OFS deltas (we
1857 * have to load the revidx to convert the offset back into a sha1).
1859 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1860 struct pack_window
**w_curs
,
1862 enum object_type type
,
1863 off_t delta_obj_offset
)
1865 if (type
== OBJ_REF_DELTA
) {
1866 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1868 } else if (type
== OBJ_OFS_DELTA
) {
1869 struct revindex_entry
*revidx
;
1870 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1871 type
, delta_obj_offset
);
1876 revidx
= find_pack_revindex(p
, base_offset
);
1880 return nth_packed_object_sha1(p
, revidx
->nr
);
1885 int unpack_object_header(struct packed_git
*p
,
1886 struct pack_window
**w_curs
,
1888 unsigned long *sizep
)
1890 unsigned char *base
;
1893 enum object_type type
;
1895 /* use_pack() assures us we have [base, base + 20) available
1896 * as a range that we can look at. (Its actually the hash
1897 * size that is assured.) With our object header encoding
1898 * the maximum deflated object size is 2^137, which is just
1899 * insane, so we know won't exceed what we have been given.
1901 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1902 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1911 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1914 struct revindex_entry
*revidx
;
1915 const unsigned char *sha1
;
1916 revidx
= find_pack_revindex(p
, obj_offset
);
1919 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1920 mark_bad_packed_object(p
, sha1
);
1921 type
= sha1_object_info(sha1
, NULL
);
1922 if (type
<= OBJ_NONE
)
1927 #define POI_STACK_PREALLOC 64
1929 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1931 enum object_type type
,
1932 struct pack_window
**w_curs
,
1935 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1936 off_t
*poi_stack
= small_poi_stack
;
1937 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1939 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1942 /* Push the object we're going to leave behind */
1943 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1944 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1945 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1946 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1948 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1950 poi_stack
[poi_stack_nr
++] = obj_offset
;
1951 /* If parsing the base offset fails, just unwind */
1952 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1955 curpos
= obj_offset
= base_offset
;
1956 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1957 if (type
<= OBJ_NONE
) {
1958 /* If getting the base itself fails, we first
1959 * retry the base, otherwise unwind */
1960 type
= retry_bad_packed_offset(p
, base_offset
);
1961 if (type
> OBJ_NONE
)
1975 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1976 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1981 if (poi_stack
!= small_poi_stack
)
1986 while (poi_stack_nr
) {
1987 obj_offset
= poi_stack
[--poi_stack_nr
];
1988 type
= retry_bad_packed_offset(p
, obj_offset
);
1989 if (type
> OBJ_NONE
)
1996 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1997 struct object_info
*oi
)
1999 struct pack_window
*w_curs
= NULL
;
2001 off_t curpos
= obj_offset
;
2002 enum object_type type
;
2005 * We always get the representation type, but only convert it to
2006 * a "real" type later if the caller is interested.
2008 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2011 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2012 off_t tmp_pos
= curpos
;
2013 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
2019 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2020 if (*oi
->sizep
== 0) {
2029 if (oi
->disk_sizep
) {
2030 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2031 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2035 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
2036 if (*oi
->typep
< 0) {
2042 if (oi
->delta_base_sha1
) {
2043 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2044 const unsigned char *base
;
2046 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2053 hashcpy(oi
->delta_base_sha1
, base
);
2055 hashclr(oi
->delta_base_sha1
);
2059 unuse_pack(&w_curs
);
2063 static void *unpack_compressed_entry(struct packed_git
*p
,
2064 struct pack_window
**w_curs
,
2070 unsigned char *buffer
, *in
;
2072 buffer
= xmallocz_gently(size
);
2075 memset(&stream
, 0, sizeof(stream
));
2076 stream
.next_out
= buffer
;
2077 stream
.avail_out
= size
+ 1;
2079 git_inflate_init(&stream
);
2081 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2082 stream
.next_in
= in
;
2083 st
= git_inflate(&stream
, Z_FINISH
);
2084 if (!stream
.avail_out
)
2085 break; /* the payload is larger than it should be */
2086 curpos
+= stream
.next_in
- in
;
2087 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2088 git_inflate_end(&stream
);
2089 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2097 #define MAX_DELTA_CACHE (256)
2099 static size_t delta_base_cached
;
2101 static struct delta_base_cache_lru_list
{
2102 struct delta_base_cache_lru_list
*prev
;
2103 struct delta_base_cache_lru_list
*next
;
2104 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
2106 static struct delta_base_cache_entry
{
2107 struct delta_base_cache_lru_list lru
;
2109 struct packed_git
*p
;
2112 enum object_type type
;
2113 } delta_base_cache
[MAX_DELTA_CACHE
];
2115 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2119 hash
= (unsigned long)(intptr_t)p
+ (unsigned long)base_offset
;
2120 hash
+= (hash
>> 8) + (hash
>> 16);
2121 return hash
% MAX_DELTA_CACHE
;
2124 static struct delta_base_cache_entry
*
2125 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2127 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2128 return delta_base_cache
+ hash
;
2131 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
2132 struct packed_git
*p
, off_t base_offset
)
2134 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
2137 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2139 struct delta_base_cache_entry
*ent
;
2140 ent
= get_delta_base_cache_entry(p
, base_offset
);
2141 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
2144 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2147 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2148 ent
->lru
.prev
->next
= ent
->lru
.next
;
2149 delta_base_cached
-= ent
->size
;
2152 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2153 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2155 struct delta_base_cache_entry
*ent
;
2158 ent
= get_delta_base_cache_entry(p
, base_offset
);
2160 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2161 return unpack_entry(p
, base_offset
, type
, base_size
);
2166 clear_delta_base_cache_entry(ent
);
2168 ret
= xmemdupz(ent
->data
, ent
->size
);
2170 *base_size
= ent
->size
;
2174 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2179 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2180 ent
->lru
.prev
->next
= ent
->lru
.next
;
2181 delta_base_cached
-= ent
->size
;
2185 void clear_delta_base_cache(void)
2188 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2189 release_delta_base_cache(&delta_base_cache
[p
]);
2192 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2193 void *base
, unsigned long base_size
, enum object_type type
)
2195 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2196 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2197 struct delta_base_cache_lru_list
*lru
;
2199 release_delta_base_cache(ent
);
2200 delta_base_cached
+= base_size
;
2202 for (lru
= delta_base_cache_lru
.next
;
2203 delta_base_cached
> delta_base_cache_limit
2204 && lru
!= &delta_base_cache_lru
;
2206 struct delta_base_cache_entry
*f
= (void *)lru
;
2207 if (f
->type
== OBJ_BLOB
)
2208 release_delta_base_cache(f
);
2210 for (lru
= delta_base_cache_lru
.next
;
2211 delta_base_cached
> delta_base_cache_limit
2212 && lru
!= &delta_base_cache_lru
;
2214 struct delta_base_cache_entry
*f
= (void *)lru
;
2215 release_delta_base_cache(f
);
2219 ent
->base_offset
= base_offset
;
2222 ent
->size
= base_size
;
2223 ent
->lru
.next
= &delta_base_cache_lru
;
2224 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2225 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2226 delta_base_cache_lru
.prev
= &ent
->lru
;
2229 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2230 unsigned long *size
);
2232 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2234 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2235 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2236 p
->pack_name
, (uintmax_t)obj_offset
);
2239 int do_check_packed_object_crc
;
2241 #define UNPACK_ENTRY_STACK_PREALLOC 64
2242 struct unpack_entry_stack_ent
{
2248 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2249 enum object_type
*final_type
, unsigned long *final_size
)
2251 struct pack_window
*w_curs
= NULL
;
2252 off_t curpos
= obj_offset
;
2255 enum object_type type
;
2256 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2257 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2258 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2259 int base_from_cache
= 0;
2261 write_pack_access_log(p
, obj_offset
);
2263 /* PHASE 1: drill down to the innermost base object */
2267 struct delta_base_cache_entry
*ent
;
2269 ent
= get_delta_base_cache_entry(p
, curpos
);
2270 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2274 clear_delta_base_cache_entry(ent
);
2275 base_from_cache
= 1;
2279 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2280 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2281 unsigned long len
= revidx
[1].offset
- obj_offset
;
2282 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2283 const unsigned char *sha1
=
2284 nth_packed_object_sha1(p
, revidx
->nr
);
2285 error("bad packed object CRC for %s",
2287 mark_bad_packed_object(p
, sha1
);
2288 unuse_pack(&w_curs
);
2293 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2294 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2297 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2299 error("failed to validate delta base reference "
2300 "at offset %"PRIuMAX
" from %s",
2301 (uintmax_t)curpos
, p
->pack_name
);
2302 /* bail to phase 2, in hopes of recovery */
2307 /* push object, proceed to base */
2308 if (delta_stack_nr
>= delta_stack_alloc
2309 && delta_stack
== small_delta_stack
) {
2310 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2311 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2312 memcpy(delta_stack
, small_delta_stack
,
2313 sizeof(*delta_stack
)*delta_stack_nr
);
2315 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2317 i
= delta_stack_nr
++;
2318 delta_stack
[i
].obj_offset
= obj_offset
;
2319 delta_stack
[i
].curpos
= curpos
;
2320 delta_stack
[i
].size
= size
;
2322 curpos
= obj_offset
= base_offset
;
2325 /* PHASE 2: handle the base */
2330 die("BUG in unpack_entry: left loop at a valid delta");
2336 if (!base_from_cache
)
2337 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2341 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2342 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2345 /* PHASE 3: apply deltas in order */
2348 * 'data' holds the base data, or NULL if there was corruption
2350 while (delta_stack_nr
) {
2353 unsigned long delta_size
, base_size
= size
;
2359 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2363 * We're probably in deep shit, but let's try to fetch
2364 * the required base anyway from another pack or loose.
2365 * This is costly but should happen only in the presence
2366 * of a corrupted pack, and is better than failing outright.
2368 struct revindex_entry
*revidx
;
2369 const unsigned char *base_sha1
;
2370 revidx
= find_pack_revindex(p
, obj_offset
);
2372 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2373 error("failed to read delta base object %s"
2374 " at offset %"PRIuMAX
" from %s",
2375 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2377 mark_bad_packed_object(p
, base_sha1
);
2378 base
= read_object(base_sha1
, &type
, &base_size
);
2382 i
= --delta_stack_nr
;
2383 obj_offset
= delta_stack
[i
].obj_offset
;
2384 curpos
= delta_stack
[i
].curpos
;
2385 delta_size
= delta_stack
[i
].size
;
2390 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2393 error("failed to unpack compressed delta "
2394 "at offset %"PRIuMAX
" from %s",
2395 (uintmax_t)curpos
, p
->pack_name
);
2400 data
= patch_delta(base
, base_size
,
2401 delta_data
, delta_size
,
2405 * We could not apply the delta; warn the user, but keep going.
2406 * Our failure will be noticed either in the next iteration of
2407 * the loop, or if this is the final delta, in the caller when
2408 * we return NULL. Those code paths will take care of making
2409 * a more explicit warning and retrying with another copy of
2413 error("failed to apply delta");
2421 unuse_pack(&w_curs
);
2423 if (delta_stack
!= small_delta_stack
)
2429 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2432 const unsigned char *index
= p
->index_data
;
2434 if (open_pack_index(p
))
2436 index
= p
->index_data
;
2438 if (n
>= p
->num_objects
)
2441 if (p
->index_version
== 1) {
2442 return index
+ 24 * n
+ 4;
2445 return index
+ 20 * n
;
2449 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2451 const unsigned char *index
= p
->index_data
;
2453 if (p
->index_version
== 1) {
2454 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2457 index
+= 8 + p
->num_objects
* (20 + 4);
2458 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2459 if (!(off
& 0x80000000))
2461 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2462 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2463 ntohl(*((uint32_t *)(index
+ 4)));
2467 off_t
find_pack_entry_one(const unsigned char *sha1
,
2468 struct packed_git
*p
)
2470 const uint32_t *level1_ofs
= p
->index_data
;
2471 const unsigned char *index
= p
->index_data
;
2472 unsigned hi
, lo
, stride
;
2473 static int use_lookup
= -1;
2474 static int debug_lookup
= -1;
2476 if (debug_lookup
< 0)
2477 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2480 if (open_pack_index(p
))
2482 level1_ofs
= p
->index_data
;
2483 index
= p
->index_data
;
2485 if (p
->index_version
> 1) {
2490 hi
= ntohl(level1_ofs
[*sha1
]);
2491 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2492 if (p
->index_version
> 1) {
2500 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2501 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2504 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2506 int pos
= sha1_entry_pos(index
, stride
, 0,
2507 lo
, hi
, p
->num_objects
, sha1
);
2510 return nth_packed_object_offset(p
, pos
);
2514 unsigned mi
= (lo
+ hi
) / 2;
2515 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2518 printf("lo %u hi %u rg %u mi %u\n",
2519 lo
, hi
, hi
- lo
, mi
);
2521 return nth_packed_object_offset(p
, mi
);
2530 int is_pack_valid(struct packed_git
*p
)
2532 /* An already open pack is known to be valid. */
2533 if (p
->pack_fd
!= -1)
2536 /* If the pack has one window completely covering the
2537 * file size, the pack is known to be valid even if
2538 * the descriptor is not currently open.
2541 struct pack_window
*w
= p
->windows
;
2543 if (!w
->offset
&& w
->len
== p
->pack_size
)
2547 /* Force the pack to open to prove its valid. */
2548 return !open_packed_git(p
);
2551 static int fill_pack_entry(const unsigned char *sha1
,
2552 struct pack_entry
*e
,
2553 struct packed_git
*p
)
2557 if (p
->num_bad_objects
) {
2559 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2560 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2564 offset
= find_pack_entry_one(sha1
, p
);
2569 * We are about to tell the caller where they can locate the
2570 * requested object. We better make sure the packfile is
2571 * still here and can be accessed before supplying that
2572 * answer, as it may have been deleted since the index was
2575 if (!is_pack_valid(p
))
2579 hashcpy(e
->sha1
, sha1
);
2584 * Iff a pack file contains the object named by sha1, return true and
2585 * store its location to e.
2587 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2589 struct packed_git
*p
;
2591 prepare_packed_git();
2595 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2598 for (p
= packed_git
; p
; p
= p
->next
) {
2599 if (p
== last_found_pack
)
2600 continue; /* we already checked this one */
2602 if (fill_pack_entry(sha1
, e
, p
)) {
2603 last_found_pack
= p
;
2610 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2611 struct packed_git
*packs
)
2613 struct packed_git
*p
;
2615 for (p
= packs
; p
; p
= p
->next
) {
2616 if (find_pack_entry_one(sha1
, p
))
2623 static int sha1_loose_object_info(const unsigned char *sha1
,
2624 struct object_info
*oi
,
2628 unsigned long mapsize
;
2632 struct strbuf hdrbuf
= STRBUF_INIT
;
2634 if (oi
->delta_base_sha1
)
2635 hashclr(oi
->delta_base_sha1
);
2638 * If we don't care about type or size, then we don't
2639 * need to look inside the object at all. Note that we
2640 * do not optimize out the stat call, even if the
2641 * caller doesn't care about the disk-size, since our
2642 * return value implicitly indicates whether the
2643 * object even exists.
2645 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
) {
2647 if (stat_sha1_file(sha1
, &st
) < 0)
2650 *oi
->disk_sizep
= st
.st_size
;
2654 map
= map_sha1_file(sha1
, &mapsize
);
2658 *oi
->disk_sizep
= mapsize
;
2659 if ((flags
& LOOKUP_UNKNOWN_OBJECT
)) {
2660 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2661 status
= error("unable to unpack %s header with --allow-unknown-type",
2663 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2664 status
= error("unable to unpack %s header",
2668 else if (hdrbuf
.len
) {
2669 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2670 status
= error("unable to parse %s header with --allow-unknown-type",
2672 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2673 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2674 git_inflate_end(&stream
);
2675 munmap(map
, mapsize
);
2676 if (status
&& oi
->typep
)
2677 *oi
->typep
= status
;
2678 strbuf_release(&hdrbuf
);
2682 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2684 struct cached_object
*co
;
2685 struct pack_entry e
;
2687 enum object_type real_type
;
2688 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2690 co
= find_cached_object(real
);
2693 *(oi
->typep
) = co
->type
;
2695 *(oi
->sizep
) = co
->size
;
2697 *(oi
->disk_sizep
) = 0;
2698 if (oi
->delta_base_sha1
)
2699 hashclr(oi
->delta_base_sha1
);
2701 strbuf_addstr(oi
->typename
, typename(co
->type
));
2702 oi
->whence
= OI_CACHED
;
2706 if (!find_pack_entry(real
, &e
)) {
2707 /* Most likely it's a loose object. */
2708 if (!sha1_loose_object_info(real
, oi
, flags
)) {
2709 oi
->whence
= OI_LOOSE
;
2713 /* Not a loose object; someone else may have just packed it. */
2714 reprepare_packed_git();
2715 if (!find_pack_entry(real
, &e
))
2720 * packed_object_info() does not follow the delta chain to
2721 * find out the real type, unless it is given oi->typep.
2723 if (oi
->typename
&& !oi
->typep
)
2724 oi
->typep
= &real_type
;
2726 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2728 mark_bad_packed_object(e
.p
, real
);
2729 if (oi
->typep
== &real_type
)
2731 return sha1_object_info_extended(real
, oi
, 0);
2732 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2733 oi
->whence
= OI_DBCACHED
;
2735 oi
->whence
= OI_PACKED
;
2736 oi
->u
.packed
.offset
= e
.offset
;
2737 oi
->u
.packed
.pack
= e
.p
;
2738 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2739 rtype
== OBJ_OFS_DELTA
);
2742 strbuf_addstr(oi
->typename
, typename(*oi
->typep
));
2743 if (oi
->typep
== &real_type
)
2749 /* returns enum object_type or negative */
2750 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2752 enum object_type type
;
2753 struct object_info oi
= {NULL
};
2757 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2762 static void *read_packed_sha1(const unsigned char *sha1
,
2763 enum object_type
*type
, unsigned long *size
)
2765 struct pack_entry e
;
2768 if (!find_pack_entry(sha1
, &e
))
2770 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2773 * We're probably in deep shit, but let's try to fetch
2774 * the required object anyway from another pack or loose.
2775 * This should happen only in the presence of a corrupted
2776 * pack, and is better than failing outright.
2778 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2779 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2780 mark_bad_packed_object(e
.p
, sha1
);
2781 data
= read_object(sha1
, type
, size
);
2786 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2787 unsigned char *sha1
)
2789 struct cached_object
*co
;
2791 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2792 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2794 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2795 co
= &cached_objects
[cached_object_nr
++];
2798 co
->buf
= xmalloc(len
);
2799 memcpy(co
->buf
, buf
, len
);
2800 hashcpy(co
->sha1
, sha1
);
2804 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2805 unsigned long *size
)
2807 unsigned long mapsize
;
2809 struct cached_object
*co
;
2811 co
= find_cached_object(sha1
);
2815 return xmemdupz(co
->buf
, co
->size
);
2818 buf
= read_packed_sha1(sha1
, type
, size
);
2821 map
= map_sha1_file(sha1
, &mapsize
);
2823 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2824 munmap(map
, mapsize
);
2827 reprepare_packed_git();
2828 return read_packed_sha1(sha1
, type
, size
);
2832 * This function dies on corrupt objects; the callers who want to
2833 * deal with them should arrange to call read_object() and give error
2834 * messages themselves.
2836 void *read_sha1_file_extended(const unsigned char *sha1
,
2837 enum object_type
*type
,
2838 unsigned long *size
,
2842 const struct packed_git
*p
;
2843 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2846 data
= read_object(repl
, type
, size
);
2850 if (errno
&& errno
!= ENOENT
)
2851 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2853 /* die if we replaced an object with one that does not exist */
2855 die("replacement %s not found for %s",
2856 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2858 if (has_loose_object(repl
)) {
2859 const char *path
= sha1_file_name(sha1
);
2861 die("loose object %s (stored in %s) is corrupt",
2862 sha1_to_hex(repl
), path
);
2865 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2866 die("packed object %s (stored in %s) is corrupt",
2867 sha1_to_hex(repl
), p
->pack_name
);
2872 void *read_object_with_reference(const unsigned char *sha1
,
2873 const char *required_type_name
,
2874 unsigned long *size
,
2875 unsigned char *actual_sha1_return
)
2877 enum object_type type
, required_type
;
2879 unsigned long isize
;
2880 unsigned char actual_sha1
[20];
2882 required_type
= type_from_string(required_type_name
);
2883 hashcpy(actual_sha1
, sha1
);
2885 int ref_length
= -1;
2886 const char *ref_type
= NULL
;
2888 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2891 if (type
== required_type
) {
2893 if (actual_sha1_return
)
2894 hashcpy(actual_sha1_return
, actual_sha1
);
2897 /* Handle references */
2898 else if (type
== OBJ_COMMIT
)
2900 else if (type
== OBJ_TAG
)
2901 ref_type
= "object ";
2906 ref_length
= strlen(ref_type
);
2908 if (ref_length
+ 40 > isize
||
2909 memcmp(buffer
, ref_type
, ref_length
) ||
2910 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2915 /* Now we have the ID of the referred-to object in
2916 * actual_sha1. Check again. */
2920 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2921 const char *type
, unsigned char *sha1
,
2922 char *hdr
, int *hdrlen
)
2926 /* Generate the header */
2927 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
2931 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2932 git_SHA1_Update(&c
, buf
, len
);
2933 git_SHA1_Final(sha1
, &c
);
2937 * Move the just written object into its final resting place.
2939 int finalize_object_file(const char *tmpfile
, const char *filename
)
2943 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2945 else if (link(tmpfile
, filename
))
2949 * Coda hack - coda doesn't like cross-directory links,
2950 * so we fall back to a rename, which will mean that it
2951 * won't be able to check collisions, but that's not a
2954 * The same holds for FAT formatted media.
2956 * When this succeeds, we just return. We have nothing
2959 if (ret
&& ret
!= EEXIST
) {
2961 if (!rename(tmpfile
, filename
))
2965 unlink_or_warn(tmpfile
);
2967 if (ret
!= EEXIST
) {
2968 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2970 /* FIXME!!! Collision check here ? */
2974 if (adjust_shared_perm(filename
))
2975 return error("unable to set permission to '%s'", filename
);
2979 static int write_buffer(int fd
, const void *buf
, size_t len
)
2981 if (write_in_full(fd
, buf
, len
) < 0)
2982 return error("file write error (%s)", strerror(errno
));
2986 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2987 unsigned char *sha1
)
2990 int hdrlen
= sizeof(hdr
);
2991 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2995 /* Finalize a file on disk, and close it. */
2996 static void close_sha1_file(int fd
)
2998 if (fsync_object_files
)
2999 fsync_or_die(fd
, "sha1 file");
3001 die_errno("error when closing sha1 file");
3004 /* Size of directory component, including the ending '/' */
3005 static inline int directory_size(const char *filename
)
3007 const char *s
= strrchr(filename
, '/');
3010 return s
- filename
+ 1;
3014 * This creates a temporary file in the same directory as the final
3017 * We want to avoid cross-directory filename renames, because those
3018 * can have problems on various filesystems (FAT, NFS, Coda).
3020 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
3022 int fd
, dirlen
= directory_size(filename
);
3025 strbuf_add(tmp
, filename
, dirlen
);
3026 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
3027 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3028 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3030 * Make sure the directory exists; note that the contents
3031 * of the buffer are undefined after mkstemp returns an
3032 * error, so we have to rewrite the whole buffer from
3036 strbuf_add(tmp
, filename
, dirlen
- 1);
3037 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
3039 if (adjust_shared_perm(tmp
->buf
))
3043 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
3044 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3049 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3050 const void *buf
, unsigned long len
, time_t mtime
)
3053 unsigned char compressed
[4096];
3056 unsigned char parano_sha1
[20];
3057 static struct strbuf tmp_file
= STRBUF_INIT
;
3058 const char *filename
= sha1_file_name(sha1
);
3060 fd
= create_tmpfile(&tmp_file
, filename
);
3062 if (errno
== EACCES
)
3063 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3065 return error("unable to create temporary file: %s", strerror(errno
));
3069 git_deflate_init(&stream
, zlib_compression_level
);
3070 stream
.next_out
= compressed
;
3071 stream
.avail_out
= sizeof(compressed
);
3074 /* First header.. */
3075 stream
.next_in
= (unsigned char *)hdr
;
3076 stream
.avail_in
= hdrlen
;
3077 while (git_deflate(&stream
, 0) == Z_OK
)
3079 git_SHA1_Update(&c
, hdr
, hdrlen
);
3081 /* Then the data itself.. */
3082 stream
.next_in
= (void *)buf
;
3083 stream
.avail_in
= len
;
3085 unsigned char *in0
= stream
.next_in
;
3086 ret
= git_deflate(&stream
, Z_FINISH
);
3087 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3088 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3089 die("unable to write sha1 file");
3090 stream
.next_out
= compressed
;
3091 stream
.avail_out
= sizeof(compressed
);
3092 } while (ret
== Z_OK
);
3094 if (ret
!= Z_STREAM_END
)
3095 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3096 ret
= git_deflate_end_gently(&stream
);
3098 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3099 git_SHA1_Final(parano_sha1
, &c
);
3100 if (hashcmp(sha1
, parano_sha1
) != 0)
3101 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3103 close_sha1_file(fd
);
3108 utb
.modtime
= mtime
;
3109 if (utime(tmp_file
.buf
, &utb
) < 0)
3110 warning("failed utime() on %s: %s",
3111 tmp_file
.buf
, strerror(errno
));
3114 return finalize_object_file(tmp_file
.buf
, filename
);
3117 static int freshen_loose_object(const unsigned char *sha1
)
3119 return check_and_freshen(sha1
, 1);
3122 static int freshen_packed_object(const unsigned char *sha1
)
3124 struct pack_entry e
;
3125 if (!find_pack_entry(sha1
, &e
))
3129 if (!freshen_file(e
.p
->pack_name
))
3135 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3138 int hdrlen
= sizeof(hdr
);
3140 /* Normally if we have it in the pack then we do not bother writing
3141 * it out into .git/objects/??/?{38} file.
3143 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3144 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3146 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3149 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3150 unsigned char *sha1
, unsigned flags
)
3153 int hdrlen
, status
= 0;
3155 /* type string, SP, %lu of the length plus NUL must fit this */
3156 hdrlen
= strlen(type
) + 32;
3157 header
= xmalloc(hdrlen
);
3158 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3160 if (!(flags
& HASH_WRITE_OBJECT
))
3162 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3164 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3171 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3175 enum object_type type
;
3180 if (has_loose_object(sha1
))
3182 buf
= read_packed_sha1(sha1
, &type
, &len
);
3184 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3185 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3186 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3192 int has_pack_index(const unsigned char *sha1
)
3195 if (stat(sha1_pack_index_name(sha1
), &st
))
3200 int has_sha1_pack(const unsigned char *sha1
)
3202 struct pack_entry e
;
3203 return find_pack_entry(sha1
, &e
);
3206 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3208 struct pack_entry e
;
3210 if (find_pack_entry(sha1
, &e
))
3212 if (has_loose_object(sha1
))
3214 if (flags
& HAS_SHA1_QUICK
)
3216 reprepare_packed_git();
3217 return find_pack_entry(sha1
, &e
);
3220 int has_object_file(const struct object_id
*oid
)
3222 return has_sha1_file(oid
->hash
);
3225 static void check_tree(const void *buf
, size_t size
)
3227 struct tree_desc desc
;
3228 struct name_entry entry
;
3230 init_tree_desc(&desc
, buf
, size
);
3231 while (tree_entry(&desc
, &entry
))
3233 * tree_entry() will die() on malformed entries */
3237 static void check_commit(const void *buf
, size_t size
)
3240 memset(&c
, 0, sizeof(c
));
3241 if (parse_commit_buffer(&c
, buf
, size
))
3242 die("corrupt commit");
3245 static void check_tag(const void *buf
, size_t size
)
3248 memset(&t
, 0, sizeof(t
));
3249 if (parse_tag_buffer(&t
, buf
, size
))
3253 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3254 enum object_type type
,
3255 const char *path
, unsigned flags
)
3257 int ret
, re_allocated
= 0;
3258 int write_object
= flags
& HASH_WRITE_OBJECT
;
3264 * Convert blobs to git internal format
3266 if ((type
== OBJ_BLOB
) && path
) {
3267 struct strbuf nbuf
= STRBUF_INIT
;
3268 if (convert_to_git(path
, buf
, size
, &nbuf
,
3269 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3270 buf
= strbuf_detach(&nbuf
, &size
);
3274 if (flags
& HASH_FORMAT_CHECK
) {
3275 if (type
== OBJ_TREE
)
3276 check_tree(buf
, size
);
3277 if (type
== OBJ_COMMIT
)
3278 check_commit(buf
, size
);
3279 if (type
== OBJ_TAG
)
3280 check_tag(buf
, size
);
3284 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3286 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3292 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3293 const char *path
, unsigned flags
)
3296 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3297 struct strbuf sbuf
= STRBUF_INIT
;
3300 assert(would_convert_to_git_filter_fd(path
));
3302 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3303 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3306 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3309 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3311 strbuf_release(&sbuf
);
3315 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3316 const char *path
, unsigned flags
)
3318 struct strbuf sbuf
= STRBUF_INIT
;
3321 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3322 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3325 strbuf_release(&sbuf
);
3329 #define SMALL_FILE_SIZE (32*1024)
3331 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3332 enum object_type type
, const char *path
,
3338 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3339 } else if (size
<= SMALL_FILE_SIZE
) {
3340 char *buf
= xmalloc(size
);
3341 if (size
== read_in_full(fd
, buf
, size
))
3342 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3344 ret
= error("short read %s", strerror(errno
));
3347 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3348 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3355 * This creates one packfile per large blob unless bulk-checkin
3356 * machinery is "plugged".
3358 * This also bypasses the usual "convert-to-git" dance, and that is on
3359 * purpose. We could write a streaming version of the converting
3360 * functions and insert that before feeding the data to fast-import
3361 * (or equivalent in-core API described above). However, that is
3362 * somewhat complicated, as we do not know the size of the filter
3363 * result, which we need to know beforehand when writing a git object.
3364 * Since the primary motivation for trying to stream from the working
3365 * tree file and to avoid mmaping it in core is to deal with large
3366 * binary blobs, they generally do not want to get any conversion, and
3367 * callers should avoid this code path when filters are requested.
3369 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3370 enum object_type type
, const char *path
,
3373 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3376 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3377 enum object_type type
, const char *path
, unsigned flags
)
3382 * Call xsize_t() only when needed to avoid potentially unnecessary
3383 * die() for large files.
3385 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3386 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3387 else if (!S_ISREG(st
->st_mode
))
3388 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3389 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3390 (path
&& would_convert_to_git(path
)))
3391 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3394 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3400 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3403 struct strbuf sb
= STRBUF_INIT
;
3405 switch (st
->st_mode
& S_IFMT
) {
3407 fd
= open(path
, O_RDONLY
);
3409 return error("open(\"%s\"): %s", path
,
3411 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3412 return error("%s: failed to insert into database",
3416 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3417 char *errstr
= strerror(errno
);
3418 return error("readlink(\"%s\"): %s", path
,
3421 if (!(flags
& HASH_WRITE_OBJECT
))
3422 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3423 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3424 return error("%s: failed to insert into database",
3426 strbuf_release(&sb
);
3429 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3431 return error("%s: unsupported file type", path
);
3436 int read_pack_header(int fd
, struct pack_header
*header
)
3438 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3439 /* "eof before pack header was fully read" */
3440 return PH_ERROR_EOF
;
3442 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3443 /* "protocol error (pack signature mismatch detected)" */
3444 return PH_ERROR_PACK_SIGNATURE
;
3445 if (!pack_version_ok(header
->hdr_version
))
3446 /* "protocol error (pack version unsupported)" */
3447 return PH_ERROR_PROTOCOL
;
3451 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3453 enum object_type type
= sha1_object_info(sha1
, NULL
);
3455 die("%s is not a valid object", sha1_to_hex(sha1
));
3457 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3461 static int for_each_file_in_obj_subdir(int subdir_nr
,
3462 struct strbuf
*path
,
3463 each_loose_object_fn obj_cb
,
3464 each_loose_cruft_fn cruft_cb
,
3465 each_loose_subdir_fn subdir_cb
,
3468 size_t baselen
= path
->len
;
3469 DIR *dir
= opendir(path
->buf
);
3474 if (errno
== ENOENT
)
3476 return error("unable to open %s: %s", path
->buf
, strerror(errno
));
3479 while ((de
= readdir(dir
))) {
3480 if (is_dot_or_dotdot(de
->d_name
))
3483 strbuf_setlen(path
, baselen
);
3484 strbuf_addf(path
, "/%s", de
->d_name
);
3486 if (strlen(de
->d_name
) == 38) {
3488 unsigned char sha1
[20];
3490 snprintf(hex
, sizeof(hex
), "%02x%s",
3491 subdir_nr
, de
->d_name
);
3492 if (!get_sha1_hex(hex
, sha1
)) {
3494 r
= obj_cb(sha1
, path
->buf
, data
);
3503 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3510 strbuf_setlen(path
, baselen
);
3511 if (!r
&& subdir_cb
)
3512 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3517 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3518 each_loose_object_fn obj_cb
,
3519 each_loose_cruft_fn cruft_cb
,
3520 each_loose_subdir_fn subdir_cb
,
3523 size_t baselen
= path
->len
;
3527 for (i
= 0; i
< 256; i
++) {
3528 strbuf_addf(path
, "/%02x", i
);
3529 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3531 strbuf_setlen(path
, baselen
);
3539 int for_each_loose_file_in_objdir(const char *path
,
3540 each_loose_object_fn obj_cb
,
3541 each_loose_cruft_fn cruft_cb
,
3542 each_loose_subdir_fn subdir_cb
,
3545 struct strbuf buf
= STRBUF_INIT
;
3548 strbuf_addstr(&buf
, path
);
3549 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3551 strbuf_release(&buf
);
3556 struct loose_alt_odb_data
{
3557 each_loose_object_fn
*cb
;
3561 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3564 struct loose_alt_odb_data
*data
= vdata
;
3565 struct strbuf buf
= STRBUF_INIT
;
3568 /* copy base not including trailing '/' */
3569 strbuf_add(&buf
, alt
->base
, alt
->name
- alt
->base
- 1);
3570 r
= for_each_loose_file_in_objdir_buf(&buf
,
3571 data
->cb
, NULL
, NULL
,
3573 strbuf_release(&buf
);
3577 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3579 struct loose_alt_odb_data alt
;
3582 r
= for_each_loose_file_in_objdir(get_object_directory(),
3583 cb
, NULL
, NULL
, data
);
3587 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3592 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3595 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3600 for (i
= 0; i
< p
->num_objects
; i
++) {
3601 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3604 return error("unable to get sha1 of object %u in %s",
3607 r
= cb(sha1
, p
, i
, data
);
3614 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3616 struct packed_git
*p
;
3618 int pack_errors
= 0;
3620 prepare_packed_git();
3621 for (p
= packed_git
; p
; p
= p
->next
) {
3622 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3624 if (open_pack_index(p
)) {
3628 r
= for_each_object_in_pack(p
, cb
, data
);
3632 return r
? r
: pack_errors
;