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];
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
46 static struct cached_object
{
47 unsigned char sha1
[20];
48 enum object_type type
;
52 static int cached_object_nr
, cached_object_alloc
;
54 static struct cached_object empty_tree
= {
55 EMPTY_TREE_SHA1_BIN_LITERAL
,
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
67 static struct packed_git
*last_found_pack
;
69 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
72 struct cached_object
*co
= cached_objects
;
74 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
75 if (!hashcmp(co
->sha1
, sha1
))
78 if (!hashcmp(sha1
, empty_tree
.sha1
))
83 int mkdir_in_gitdir(const char *path
)
85 if (mkdir(path
, 0777)) {
86 int saved_errno
= errno
;
88 struct strbuf sb
= STRBUF_INIT
;
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
99 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
100 strbuf_readlink(&sb
, path
, st
.st_size
) ||
101 !is_absolute_path(sb
.buf
) ||
102 mkdir(sb
.buf
, 0777)) {
109 return adjust_shared_perm(path
);
112 enum scld_error
safe_create_leading_directories(char *path
)
114 char *next_component
= path
+ offset_1st_component(path
);
115 enum scld_error ret
= SCLD_OK
;
117 while (ret
== SCLD_OK
&& next_component
) {
119 char *slash
= next_component
, slash_character
;
121 while (*slash
&& !is_dir_sep(*slash
))
127 next_component
= slash
+ 1;
128 while (is_dir_sep(*next_component
))
130 if (!*next_component
)
133 slash_character
= *slash
;
135 if (!stat(path
, &st
)) {
137 if (!S_ISDIR(st
.st_mode
))
139 } else if (mkdir(path
, 0777)) {
140 if (errno
== EEXIST
&&
141 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
142 ; /* somebody created it since we checked */
143 else if (errno
== ENOENT
)
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
156 } else if (adjust_shared_perm(path
)) {
159 *slash
= slash_character
;
164 enum scld_error
safe_create_leading_directories_const(const char *path
)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf
= xstrdup(path
);
168 enum scld_error result
= safe_create_leading_directories(buf
);
173 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
176 for (i
= 0; i
< 20; i
++) {
177 static char hex
[] = "0123456789abcdef";
178 unsigned int val
= sha1
[i
];
179 char *pos
= pathbuf
+ i
*2 + (i
> 0);
180 *pos
++ = hex
[val
>> 4];
181 *pos
= hex
[val
& 0xf];
185 const char *sha1_file_name(const unsigned char *sha1
)
187 static char buf
[PATH_MAX
];
191 objdir
= get_object_directory();
192 len
= strlen(objdir
);
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len
+ 43 > PATH_MAX
)
196 die("insanely long object directory %s", objdir
);
197 memcpy(buf
, objdir
, len
);
201 fill_sha1_path(buf
+ len
+ 1, sha1
);
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
210 static char *sha1_get_pack_name(const unsigned char *sha1
,
211 char **name
, char **base
, const char *which
)
213 static const char hex
[] = "0123456789abcdef";
218 const char *sha1_file_directory
= get_object_directory();
219 int len
= strlen(sha1_file_directory
);
220 *base
= xmalloc(len
+ 60);
221 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory
, which
);
223 *name
= *base
+ len
+ 11;
228 for (i
= 0; i
< 20; i
++) {
229 unsigned int val
= *sha1
++;
230 *buf
++ = hex
[val
>> 4];
231 *buf
++ = hex
[val
& 0xf];
237 char *sha1_pack_name(const unsigned char *sha1
)
239 static char *name
, *base
;
241 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
244 char *sha1_pack_index_name(const unsigned char *sha1
)
246 static char *name
, *base
;
248 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
251 struct alternate_object_database
*alt_odb_list
;
252 static struct alternate_object_database
**alt_odb_tail
;
255 * Prepare alternate object database registry.
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
269 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
270 int depth
, const char *normalized_objdir
)
272 struct alternate_object_database
*ent
;
273 struct alternate_object_database
*alt
;
275 struct strbuf pathbuf
= STRBUF_INIT
;
277 if (!is_absolute_path(entry
) && relative_base
) {
278 strbuf_addstr(&pathbuf
, real_path(relative_base
));
279 strbuf_addch(&pathbuf
, '/');
281 strbuf_addstr(&pathbuf
, entry
);
283 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
285 pfxlen
= strlen(pathbuf
.buf
);
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
291 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
294 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent
= xmalloc(sizeof(*ent
) + entlen
);
296 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
297 strbuf_release(&pathbuf
);
299 ent
->name
= ent
->base
+ pfxlen
+ 1;
300 ent
->base
[pfxlen
+ 3] = '/';
301 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent
->base
)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
315 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
316 if (pfxlen
== alt
->name
- alt
->base
- 1 &&
317 !memcmp(ent
->base
, alt
->base
, pfxlen
)) {
322 if (!strcmp_icase(ent
->base
, normalized_objdir
)) {
327 /* add the alternate entry */
329 alt_odb_tail
= &(ent
->next
);
332 /* recursively add alternates */
333 read_info_alternates(ent
->base
, depth
+ 1);
335 ent
->base
[pfxlen
] = '/';
340 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
341 const char *relative_base
, int depth
)
343 struct string_list entries
= STRING_LIST_INIT_NODUP
;
346 struct strbuf objdirbuf
= STRBUF_INIT
;
349 error("%s: ignoring alternate object stores, nesting too deep.",
354 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
355 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
357 alt_copy
= xmemdupz(alt
, len
);
358 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
359 for (i
= 0; i
< entries
.nr
; i
++) {
360 const char *entry
= entries
.items
[i
].string
;
361 if (entry
[0] == '\0' || entry
[0] == '#')
363 if (!is_absolute_path(entry
) && depth
) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base
, entry
);
367 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
370 string_list_clear(&entries
, 0);
372 strbuf_release(&objdirbuf
);
375 void read_info_alternates(const char * relative_base
, int depth
)
383 path
= xstrfmt("%s/info/alternates", relative_base
);
384 fd
= git_open_noatime(path
);
388 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
392 mapsz
= xsize_t(st
.st_size
);
393 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
396 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
401 void add_to_alternates_file(const char *reference
)
403 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
404 char *alts
= git_pathdup("objects/info/alternates");
407 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
408 out
= fdopen_lock_file(lock
, "w");
410 die_errno("unable to fdopen alternates lockfile");
412 in
= fopen(alts
, "r");
414 struct strbuf line
= STRBUF_INIT
;
417 while (strbuf_getline(&line
, in
, '\n') != EOF
) {
418 if (!strcmp(reference
, line
.buf
)) {
422 fprintf_or_die(out
, "%s\n", line
.buf
);
425 strbuf_release(&line
);
429 rollback_lock_file(lock
);
433 else if (errno
!= ENOENT
)
434 die_errno("unable to read alternates file");
437 fprintf_or_die(out
, "%s\n", reference
);
438 if (commit_lock_file(lock
))
439 die_errno("unable to move new alternates file into place");
441 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
446 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
448 struct alternate_object_database
*ent
;
452 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
460 void prepare_alt_odb(void)
467 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
470 alt_odb_tail
= &alt_odb_list
;
471 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
473 read_info_alternates(get_object_directory(), 0);
476 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
477 static int freshen_file(const char *fn
)
480 t
.actime
= t
.modtime
= time(NULL
);
481 return !utime(fn
, &t
);
485 * All of the check_and_freshen functions return 1 if the file exists and was
486 * freshened (if freshening was requested), 0 otherwise. If they return
487 * 0, you should not assume that it is safe to skip a write of the object (it
488 * either does not exist on disk, or has a stale mtime and may be subject to
491 static int check_and_freshen_file(const char *fn
, int freshen
)
493 if (access(fn
, F_OK
))
495 if (freshen
&& !freshen_file(fn
))
500 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
502 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
505 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
507 struct alternate_object_database
*alt
;
509 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
510 fill_sha1_path(alt
->name
, sha1
);
511 if (check_and_freshen_file(alt
->base
, freshen
))
517 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
519 return check_and_freshen_local(sha1
, freshen
) ||
520 check_and_freshen_nonlocal(sha1
, freshen
);
523 int has_loose_object_nonlocal(const unsigned char *sha1
)
525 return check_and_freshen_nonlocal(sha1
, 0);
528 static int has_loose_object(const unsigned char *sha1
)
530 return check_and_freshen(sha1
, 0);
533 static unsigned int pack_used_ctr
;
534 static unsigned int pack_mmap_calls
;
535 static unsigned int peak_pack_open_windows
;
536 static unsigned int pack_open_windows
;
537 static unsigned int pack_open_fds
;
538 static unsigned int pack_max_fds
;
539 static size_t peak_pack_mapped
;
540 static size_t pack_mapped
;
541 struct packed_git
*packed_git
;
543 void pack_report(void)
546 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
547 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
548 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
549 sz_fmt(getpagesize()),
550 sz_fmt(packed_git_window_size
),
551 sz_fmt(packed_git_limit
));
553 "pack_report: pack_used_ctr = %10u\n"
554 "pack_report: pack_mmap_calls = %10u\n"
555 "pack_report: pack_open_windows = %10u / %10u\n"
556 "pack_report: pack_mapped = "
557 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
560 pack_open_windows
, peak_pack_open_windows
,
561 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
565 * Open and mmap the index file at path, perform a couple of
566 * consistency checks, then record its information to p. Return 0 on
569 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
572 struct pack_idx_header
*hdr
;
574 uint32_t version
, nr
, i
, *index
;
575 int fd
= git_open_noatime(path
);
580 if (fstat(fd
, &st
)) {
584 idx_size
= xsize_t(st
.st_size
);
585 if (idx_size
< 4 * 256 + 20 + 20) {
587 return error("index file %s is too small", path
);
589 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
593 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
594 version
= ntohl(hdr
->idx_version
);
595 if (version
< 2 || version
> 2) {
596 munmap(idx_map
, idx_size
);
597 return error("index file %s is version %"PRIu32
598 " and is not supported by this binary"
599 " (try upgrading GIT to a newer version)",
608 index
+= 2; /* skip index header */
609 for (i
= 0; i
< 256; i
++) {
610 uint32_t n
= ntohl(index
[i
]);
612 munmap(idx_map
, idx_size
);
613 return error("non-monotonic index %s", path
);
621 * - 256 index entries 4 bytes each
622 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
623 * - 20-byte SHA1 of the packfile
624 * - 20-byte SHA1 file checksum
626 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
627 munmap(idx_map
, idx_size
);
628 return error("wrong index v1 file size in %s", path
);
630 } else if (version
== 2) {
633 * - 8 bytes of header
634 * - 256 index entries 4 bytes each
635 * - 20-byte sha1 entry * nr
636 * - 4-byte crc entry * nr
637 * - 4-byte offset entry * nr
638 * - 20-byte SHA1 of the packfile
639 * - 20-byte SHA1 file checksum
640 * And after the 4-byte offset table might be a
641 * variable sized table containing 8-byte entries
642 * for offsets larger than 2^31.
644 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
645 unsigned long max_size
= min_size
;
647 max_size
+= (nr
- 1)*8;
648 if (idx_size
< min_size
|| idx_size
> max_size
) {
649 munmap(idx_map
, idx_size
);
650 return error("wrong index v2 file size in %s", path
);
652 if (idx_size
!= min_size
&&
654 * make sure we can deal with large pack offsets.
655 * 31-bit signed offset won't be enough, neither
656 * 32-bit unsigned one will be.
658 (sizeof(off_t
) <= 4)) {
659 munmap(idx_map
, idx_size
);
660 return error("pack too large for current definition of off_t in %s", path
);
664 p
->index_version
= version
;
665 p
->index_data
= idx_map
;
666 p
->index_size
= idx_size
;
671 int open_pack_index(struct packed_git
*p
)
680 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
681 die("BUG: pack_name does not end in .pack");
682 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
683 ret
= check_packed_git_idx(idx_name
, p
);
688 static void scan_windows(struct packed_git
*p
,
689 struct packed_git
**lru_p
,
690 struct pack_window
**lru_w
,
691 struct pack_window
**lru_l
)
693 struct pack_window
*w
, *w_l
;
695 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
697 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
707 static int unuse_one_window(struct packed_git
*current
)
709 struct packed_git
*p
, *lru_p
= NULL
;
710 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
713 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
714 for (p
= packed_git
; p
; p
= p
->next
)
715 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
717 munmap(lru_w
->base
, lru_w
->len
);
718 pack_mapped
-= lru_w
->len
;
720 lru_l
->next
= lru_w
->next
;
722 lru_p
->windows
= lru_w
->next
;
730 void release_pack_memory(size_t need
)
732 size_t cur
= pack_mapped
;
733 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
737 static void mmap_limit_check(size_t length
)
739 static size_t limit
= 0;
741 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
746 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
747 (uintmax_t)length
, (uintmax_t)limit
);
750 void *xmmap_gently(void *start
, size_t length
,
751 int prot
, int flags
, int fd
, off_t offset
)
755 mmap_limit_check(length
);
756 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
757 if (ret
== MAP_FAILED
) {
760 release_pack_memory(length
);
761 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
766 void *xmmap(void *start
, size_t length
,
767 int prot
, int flags
, int fd
, off_t offset
)
769 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
770 if (ret
== MAP_FAILED
)
771 die_errno("mmap failed");
775 void close_pack_windows(struct packed_git
*p
)
778 struct pack_window
*w
= p
->windows
;
781 die("pack '%s' still has open windows to it",
783 munmap(w
->base
, w
->len
);
784 pack_mapped
-= w
->len
;
786 p
->windows
= w
->next
;
792 * The LRU pack is the one with the oldest MRU window, preferring packs
793 * with no used windows, or the oldest mtime if it has no windows allocated.
795 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
797 struct pack_window
*w
, *this_mru_w
;
798 int has_windows_inuse
= 0;
801 * Reject this pack if it has windows and the previously selected
802 * one does not. If this pack does not have windows, reject
803 * it if the pack file is newer than the previously selected one.
805 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
808 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
810 * Reject this pack if any of its windows are in use,
811 * but the previously selected pack did not have any
812 * inuse windows. Otherwise, record that this pack
813 * has windows in use.
816 if (*accept_windows_inuse
)
817 has_windows_inuse
= 1;
822 if (w
->last_used
> this_mru_w
->last_used
)
826 * Reject this pack if it has windows that have been
827 * used more recently than the previously selected pack.
828 * If the previously selected pack had windows inuse and
829 * we have not encountered a window in this pack that is
830 * inuse, skip this check since we prefer a pack with no
831 * inuse windows to one that has inuse windows.
833 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
834 this_mru_w
->last_used
> (*mru_w
)->last_used
)
843 *accept_windows_inuse
= has_windows_inuse
;
846 static int close_one_pack(void)
848 struct packed_git
*p
, *lru_p
= NULL
;
849 struct pack_window
*mru_w
= NULL
;
850 int accept_windows_inuse
= 1;
852 for (p
= packed_git
; p
; p
= p
->next
) {
853 if (p
->pack_fd
== -1)
855 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
859 close(lru_p
->pack_fd
);
868 void unuse_pack(struct pack_window
**w_cursor
)
870 struct pack_window
*w
= *w_cursor
;
877 void close_pack_index(struct packed_git
*p
)
880 munmap((void *)p
->index_data
, p
->index_size
);
881 p
->index_data
= NULL
;
886 * This is used by git-repack in case a newly created pack happens to
887 * contain the same set of objects as an existing one. In that case
888 * the resulting file might be different even if its name would be the
889 * same. It is best to close any reference to the old pack before it is
890 * replaced on disk. Of course no index pointers or windows for given pack
891 * must subsist at this point. If ever objects from this pack are requested
892 * again, the new version of the pack will be reinitialized through
893 * reprepare_packed_git().
895 void free_pack_by_name(const char *pack_name
)
897 struct packed_git
*p
, **pp
= &packed_git
;
901 if (strcmp(pack_name
, p
->pack_name
) == 0) {
902 clear_delta_base_cache();
903 close_pack_windows(p
);
904 if (p
->pack_fd
!= -1) {
909 free(p
->bad_object_sha1
);
911 if (last_found_pack
== p
)
912 last_found_pack
= NULL
;
920 static unsigned int get_max_fd_limit(void)
926 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
933 long open_max
= sysconf(_SC_OPEN_MAX
);
937 * Otherwise, we got -1 for one of the two
940 * (1) sysconf() did not understand _SC_OPEN_MAX
941 * and signaled an error with -1; or
942 * (2) sysconf() said there is no limit.
944 * We _could_ clear errno before calling sysconf() to
945 * tell these two cases apart and return a huge number
946 * in the latter case to let the caller cap it to a
947 * value that is not so selfish, but letting the
948 * fallback OPEN_MAX codepath take care of these cases
957 return 1; /* see the caller ;-) */
962 * Do not call this directly as this leaks p->pack_fd on error return;
963 * call open_packed_git() instead.
965 static int open_packed_git_1(struct packed_git
*p
)
968 struct pack_header hdr
;
969 unsigned char sha1
[20];
970 unsigned char *idx_sha1
;
973 if (!p
->index_data
&& open_pack_index(p
))
974 return error("packfile %s index unavailable", p
->pack_name
);
977 unsigned int max_fds
= get_max_fd_limit();
979 /* Save 3 for stdin/stdout/stderr, 22 for work */
981 pack_max_fds
= max_fds
- 25;
986 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
989 p
->pack_fd
= git_open_noatime(p
->pack_name
);
990 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
994 /* If we created the struct before we had the pack we lack size. */
996 if (!S_ISREG(st
.st_mode
))
997 return error("packfile %s not a regular file", p
->pack_name
);
998 p
->pack_size
= st
.st_size
;
999 } else if (p
->pack_size
!= st
.st_size
)
1000 return error("packfile %s size changed", p
->pack_name
);
1002 /* We leave these file descriptors open with sliding mmap;
1003 * there is no point keeping them open across exec(), though.
1005 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
1007 return error("cannot determine file descriptor flags");
1008 fd_flag
|= FD_CLOEXEC
;
1009 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
1010 return error("cannot set FD_CLOEXEC");
1012 /* Verify we recognize this pack file format. */
1013 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
1014 return error("file %s is far too short to be a packfile", p
->pack_name
);
1015 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
1016 return error("file %s is not a GIT packfile", p
->pack_name
);
1017 if (!pack_version_ok(hdr
.hdr_version
))
1018 return error("packfile %s is version %"PRIu32
" and not"
1019 " supported (try upgrading GIT to a newer version)",
1020 p
->pack_name
, ntohl(hdr
.hdr_version
));
1022 /* Verify the pack matches its index. */
1023 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
1024 return error("packfile %s claims to have %"PRIu32
" objects"
1025 " while index indicates %"PRIu32
" objects",
1026 p
->pack_name
, ntohl(hdr
.hdr_entries
),
1028 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1029 return error("end of packfile %s is unavailable", p
->pack_name
);
1030 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1031 return error("packfile %s signature is unavailable", p
->pack_name
);
1032 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1033 if (hashcmp(sha1
, idx_sha1
))
1034 return error("packfile %s does not match index", p
->pack_name
);
1038 static int open_packed_git(struct packed_git
*p
)
1040 if (!open_packed_git_1(p
))
1042 if (p
->pack_fd
!= -1) {
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
) {
1118 pack_open_windows
++;
1119 if (pack_mapped
> peak_pack_mapped
)
1120 peak_pack_mapped
= pack_mapped
;
1121 if (pack_open_windows
> peak_pack_open_windows
)
1122 peak_pack_open_windows
= pack_open_windows
;
1123 win
->next
= p
->windows
;
1127 if (win
!= *w_cursor
) {
1128 win
->last_used
= pack_used_ctr
++;
1132 offset
-= win
->offset
;
1134 *left
= win
->len
- xsize_t(offset
);
1135 return win
->base
+ offset
;
1138 static struct packed_git
*alloc_packed_git(int extra
)
1140 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1141 memset(p
, 0, sizeof(*p
));
1146 static void try_to_free_pack_memory(size_t size
)
1148 release_pack_memory(size
);
1151 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1153 static int have_set_try_to_free_routine
;
1156 struct packed_git
*p
;
1158 if (!have_set_try_to_free_routine
) {
1159 have_set_try_to_free_routine
= 1;
1160 set_try_to_free_routine(try_to_free_pack_memory
);
1164 * Make sure a corresponding .pack file exists and that
1165 * the index looks sane.
1167 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1171 * ".pack" is long enough to hold any suffix we're adding (and
1172 * the use xsnprintf double-checks that)
1174 alloc
= path_len
+ strlen(".pack") + 1;
1175 p
= alloc_packed_git(alloc
);
1176 memcpy(p
->pack_name
, path
, path_len
);
1178 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1179 if (!access(p
->pack_name
, F_OK
))
1182 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1183 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1188 /* ok, it looks sane as far as we can check without
1189 * actually mapping the pack file.
1191 p
->pack_size
= st
.st_size
;
1192 p
->pack_local
= local
;
1193 p
->mtime
= st
.st_mtime
;
1194 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1199 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1201 const char *path
= sha1_pack_name(sha1
);
1202 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1204 strcpy(p
->pack_name
, path
);
1205 hashcpy(p
->sha1
, sha1
);
1206 if (check_packed_git_idx(idx_path
, p
)) {
1214 void install_packed_git(struct packed_git
*pack
)
1216 if (pack
->pack_fd
!= -1)
1219 pack
->next
= packed_git
;
1223 void (*report_garbage
)(const char *desc
, const char *path
);
1225 static void report_helper(const struct string_list
*list
,
1226 int seen_bits
, int first
, int last
)
1229 switch (seen_bits
) {
1231 msg
= "no corresponding .idx or .pack";
1234 msg
= "no corresponding .idx";
1237 msg
= "no corresponding .pack";
1242 for (; first
< last
; first
++)
1243 report_garbage(msg
, list
->items
[first
].string
);
1246 static void report_pack_garbage(struct string_list
*list
)
1248 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1250 if (!report_garbage
)
1253 string_list_sort(list
);
1255 for (i
= 0; i
< list
->nr
; i
++) {
1256 const char *path
= list
->items
[i
].string
;
1257 if (baselen
!= -1 &&
1258 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1259 report_helper(list
, seen_bits
, first
, i
);
1263 if (baselen
== -1) {
1264 const char *dot
= strrchr(path
, '.');
1266 report_garbage("garbage found", path
);
1269 baselen
= dot
- path
+ 1;
1272 if (!strcmp(path
+ baselen
, "pack"))
1274 else if (!strcmp(path
+ baselen
, "idx"))
1277 report_helper(list
, seen_bits
, first
, list
->nr
);
1280 static void prepare_packed_git_one(char *objdir
, int local
)
1282 struct strbuf path
= STRBUF_INIT
;
1286 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1288 strbuf_addstr(&path
, objdir
);
1289 strbuf_addstr(&path
, "/pack");
1290 dir
= opendir(path
.buf
);
1292 if (errno
!= ENOENT
)
1293 error("unable to open object pack directory: %s: %s",
1294 path
.buf
, strerror(errno
));
1295 strbuf_release(&path
);
1298 strbuf_addch(&path
, '/');
1299 dirnamelen
= path
.len
;
1300 while ((de
= readdir(dir
)) != NULL
) {
1301 struct packed_git
*p
;
1304 if (is_dot_or_dotdot(de
->d_name
))
1307 strbuf_setlen(&path
, dirnamelen
);
1308 strbuf_addstr(&path
, de
->d_name
);
1310 base_len
= path
.len
;
1311 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1312 /* Don't reopen a pack we already have. */
1313 for (p
= packed_git
; p
; p
= p
->next
) {
1315 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1317 !memcmp(p
->pack_name
, path
.buf
, len
))
1322 * See if it really is a valid .idx file with
1323 * corresponding .pack file that we can map.
1325 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1326 install_packed_git(p
);
1329 if (!report_garbage
)
1332 if (ends_with(de
->d_name
, ".idx") ||
1333 ends_with(de
->d_name
, ".pack") ||
1334 ends_with(de
->d_name
, ".bitmap") ||
1335 ends_with(de
->d_name
, ".keep"))
1336 string_list_append(&garbage
, path
.buf
);
1338 report_garbage("garbage found", path
.buf
);
1341 report_pack_garbage(&garbage
);
1342 string_list_clear(&garbage
, 0);
1343 strbuf_release(&path
);
1346 static int sort_pack(const void *a_
, const void *b_
)
1348 struct packed_git
*a
= *((struct packed_git
**)a_
);
1349 struct packed_git
*b
= *((struct packed_git
**)b_
);
1353 * Local packs tend to contain objects specific to our
1354 * variant of the project than remote ones. In addition,
1355 * remote ones could be on a network mounted filesystem.
1356 * Favor local ones for these reasons.
1358 st
= a
->pack_local
- b
->pack_local
;
1363 * Younger packs tend to contain more recent objects,
1364 * and more recent objects tend to get accessed more
1367 if (a
->mtime
< b
->mtime
)
1369 else if (a
->mtime
== b
->mtime
)
1374 static void rearrange_packed_git(void)
1376 struct packed_git
**ary
, *p
;
1379 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1384 /* prepare an array of packed_git for easier sorting */
1385 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1386 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1389 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1391 /* link them back again */
1392 for (i
= 0; i
< n
- 1; i
++)
1393 ary
[i
]->next
= ary
[i
+ 1];
1394 ary
[n
- 1]->next
= NULL
;
1395 packed_git
= ary
[0];
1400 static int prepare_packed_git_run_once
= 0;
1401 void prepare_packed_git(void)
1403 struct alternate_object_database
*alt
;
1405 if (prepare_packed_git_run_once
)
1407 prepare_packed_git_one(get_object_directory(), 1);
1409 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1411 prepare_packed_git_one(alt
->base
, 0);
1412 alt
->name
[-1] = '/';
1414 rearrange_packed_git();
1415 prepare_packed_git_run_once
= 1;
1418 void reprepare_packed_git(void)
1420 prepare_packed_git_run_once
= 0;
1421 prepare_packed_git();
1424 static void mark_bad_packed_object(struct packed_git
*p
,
1425 const unsigned char *sha1
)
1428 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1429 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1431 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1432 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1433 p
->num_bad_objects
++;
1436 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1438 struct packed_git
*p
;
1441 for (p
= packed_git
; p
; p
= p
->next
)
1442 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1443 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1449 * With an in-core object data in "map", rehash it to make sure the
1450 * object name actually matches "sha1" to detect object corruption.
1451 * With "map" == NULL, try reading the object named with "sha1" using
1452 * the streaming interface and rehash it to do the same.
1454 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1455 unsigned long size
, const char *type
)
1457 unsigned char real_sha1
[20];
1458 enum object_type obj_type
;
1459 struct git_istream
*st
;
1465 hash_sha1_file(map
, size
, type
, real_sha1
);
1466 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1469 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1473 /* Generate the header */
1474 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1478 git_SHA1_Update(&c
, hdr
, hdrlen
);
1480 char buf
[1024 * 16];
1481 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1489 git_SHA1_Update(&c
, buf
, readlen
);
1491 git_SHA1_Final(real_sha1
, &c
);
1493 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1496 int git_open_noatime(const char *name
)
1498 static int sha1_file_open_flag
= O_NOATIME
;
1504 fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1508 /* Might the failure be due to O_NOATIME? */
1509 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1510 sha1_file_open_flag
= 0;
1518 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1520 struct alternate_object_database
*alt
;
1522 if (!lstat(sha1_file_name(sha1
), st
))
1527 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1528 fill_sha1_path(alt
->name
, sha1
);
1529 if (!lstat(alt
->base
, st
))
1536 static int open_sha1_file(const unsigned char *sha1
)
1539 struct alternate_object_database
*alt
;
1540 int most_interesting_errno
;
1542 fd
= git_open_noatime(sha1_file_name(sha1
));
1545 most_interesting_errno
= errno
;
1548 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1549 fill_sha1_path(alt
->name
, sha1
);
1550 fd
= git_open_noatime(alt
->base
);
1553 if (most_interesting_errno
== ENOENT
)
1554 most_interesting_errno
= errno
;
1556 errno
= most_interesting_errno
;
1560 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1565 fd
= open_sha1_file(sha1
);
1570 if (!fstat(fd
, &st
)) {
1571 *size
= xsize_t(st
.st_size
);
1573 /* mmap() is forbidden on empty files */
1574 error("object file %s is empty", sha1_file_name(sha1
));
1577 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1584 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1585 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1588 unsigned long size
, c
;
1589 unsigned long used
= 0;
1592 *type
= (c
>> 4) & 7;
1596 if (len
<= used
|| bitsizeof(long) <= shift
) {
1597 error("bad object header");
1602 size
+= (c
& 0x7f) << shift
;
1609 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1611 /* Get the data stream */
1612 memset(stream
, 0, sizeof(*stream
));
1613 stream
->next_in
= map
;
1614 stream
->avail_in
= mapsize
;
1615 stream
->next_out
= buffer
;
1616 stream
->avail_out
= bufsiz
;
1618 git_inflate_init(stream
);
1619 return git_inflate(stream
, 0);
1622 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1623 unsigned long mapsize
, void *buffer
,
1624 unsigned long bufsiz
, struct strbuf
*header
)
1628 status
= unpack_sha1_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1631 * Check if entire header is unpacked in the first iteration.
1633 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1637 * buffer[0..bufsiz] was not large enough. Copy the partial
1638 * result out to header, and then append the result of further
1639 * reading the stream.
1641 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1642 stream
->next_out
= buffer
;
1643 stream
->avail_out
= bufsiz
;
1646 status
= git_inflate(stream
, 0);
1647 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1648 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1650 stream
->next_out
= buffer
;
1651 stream
->avail_out
= bufsiz
;
1652 } while (status
!= Z_STREAM_END
);
1656 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1658 int bytes
= strlen(buffer
) + 1;
1659 unsigned char *buf
= xmallocz(size
);
1663 n
= stream
->total_out
- bytes
;
1666 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1668 if (bytes
<= size
) {
1670 * The above condition must be (bytes <= size), not
1671 * (bytes < size). In other words, even though we
1672 * expect no more output and set avail_out to zero,
1673 * the input zlib stream may have bytes that express
1674 * "this concludes the stream", and we *do* want to
1677 * Otherwise we would not be able to test that we
1678 * consumed all the input to reach the expected size;
1679 * we also want to check that zlib tells us that all
1680 * went well with status == Z_STREAM_END at the end.
1682 stream
->next_out
= buf
+ bytes
;
1683 stream
->avail_out
= size
- bytes
;
1684 while (status
== Z_OK
)
1685 status
= git_inflate(stream
, Z_FINISH
);
1687 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1688 git_inflate_end(stream
);
1693 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1694 else if (stream
->avail_in
)
1695 error("garbage at end of loose object '%s'",
1702 * We used to just use "sscanf()", but that's actually way
1703 * too permissive for what we want to check. So do an anal
1704 * object header parse by hand.
1706 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1709 const char *type_buf
= hdr
;
1711 int type
, type_len
= 0;
1714 * The type can be of any size but is followed by
1724 type
= type_from_string_gently(type_buf
, type_len
, 1);
1726 strbuf_add(oi
->typename
, type_buf
, type_len
);
1728 * Set type to 0 if its an unknown object and
1729 * we're obtaining the type using '--allow-unkown-type'
1732 if ((flags
& LOOKUP_UNKNOWN_OBJECT
) && (type
< 0))
1735 die("invalid object type");
1740 * The length must follow immediately, and be in canonical
1741 * decimal format (ie "010" is not valid).
1743 size
= *hdr
++ - '0';
1748 unsigned long c
= *hdr
- '0';
1752 size
= size
* 10 + c
;
1760 * The length must be followed by a zero byte
1762 return *hdr
? -1 : type
;
1765 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1767 struct object_info oi
;
1772 return parse_sha1_header_extended(hdr
, &oi
, LOOKUP_REPLACE_OBJECT
);
1775 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1781 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1782 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1785 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1788 unsigned long get_size_from_delta(struct packed_git
*p
,
1789 struct pack_window
**w_curs
,
1792 const unsigned char *data
;
1793 unsigned char delta_head
[20], *in
;
1797 memset(&stream
, 0, sizeof(stream
));
1798 stream
.next_out
= delta_head
;
1799 stream
.avail_out
= sizeof(delta_head
);
1801 git_inflate_init(&stream
);
1803 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1804 stream
.next_in
= in
;
1805 st
= git_inflate(&stream
, Z_FINISH
);
1806 curpos
+= stream
.next_in
- in
;
1807 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1808 stream
.total_out
< sizeof(delta_head
));
1809 git_inflate_end(&stream
);
1810 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1811 error("delta data unpack-initial failed");
1815 /* Examine the initial part of the delta to figure out
1820 /* ignore base size */
1821 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1823 /* Read the result size */
1824 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1827 static off_t
get_delta_base(struct packed_git
*p
,
1828 struct pack_window
**w_curs
,
1830 enum object_type type
,
1831 off_t delta_obj_offset
)
1833 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1836 /* use_pack() assured us we have [base_info, base_info + 20)
1837 * as a range that we can look at without walking off the
1838 * end of the mapped window. Its actually the hash size
1839 * that is assured. An OFS_DELTA longer than the hash size
1840 * is stupid, as then a REF_DELTA would be smaller to store.
1842 if (type
== OBJ_OFS_DELTA
) {
1844 unsigned char c
= base_info
[used
++];
1845 base_offset
= c
& 127;
1848 if (!base_offset
|| MSB(base_offset
, 7))
1849 return 0; /* overflow */
1850 c
= base_info
[used
++];
1851 base_offset
= (base_offset
<< 7) + (c
& 127);
1853 base_offset
= delta_obj_offset
- base_offset
;
1854 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1855 return 0; /* out of bound */
1857 } else if (type
== OBJ_REF_DELTA
) {
1858 /* The base entry _must_ be in the same pack */
1859 base_offset
= find_pack_entry_one(base_info
, p
);
1862 die("I am totally screwed");
1867 * Like get_delta_base above, but we return the sha1 instead of the pack
1868 * offset. This means it is cheaper for REF deltas (we do not have to do
1869 * the final object lookup), but more expensive for OFS deltas (we
1870 * have to load the revidx to convert the offset back into a sha1).
1872 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1873 struct pack_window
**w_curs
,
1875 enum object_type type
,
1876 off_t delta_obj_offset
)
1878 if (type
== OBJ_REF_DELTA
) {
1879 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1881 } else if (type
== OBJ_OFS_DELTA
) {
1882 struct revindex_entry
*revidx
;
1883 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1884 type
, delta_obj_offset
);
1889 revidx
= find_pack_revindex(p
, base_offset
);
1893 return nth_packed_object_sha1(p
, revidx
->nr
);
1898 int unpack_object_header(struct packed_git
*p
,
1899 struct pack_window
**w_curs
,
1901 unsigned long *sizep
)
1903 unsigned char *base
;
1906 enum object_type type
;
1908 /* use_pack() assures us we have [base, base + 20) available
1909 * as a range that we can look at. (Its actually the hash
1910 * size that is assured.) With our object header encoding
1911 * the maximum deflated object size is 2^137, which is just
1912 * insane, so we know won't exceed what we have been given.
1914 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1915 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1924 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1927 struct revindex_entry
*revidx
;
1928 const unsigned char *sha1
;
1929 revidx
= find_pack_revindex(p
, obj_offset
);
1932 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1933 mark_bad_packed_object(p
, sha1
);
1934 type
= sha1_object_info(sha1
, NULL
);
1935 if (type
<= OBJ_NONE
)
1940 #define POI_STACK_PREALLOC 64
1942 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1944 enum object_type type
,
1945 struct pack_window
**w_curs
,
1948 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1949 off_t
*poi_stack
= small_poi_stack
;
1950 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1952 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1955 /* Push the object we're going to leave behind */
1956 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1957 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1958 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1959 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1961 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1963 poi_stack
[poi_stack_nr
++] = obj_offset
;
1964 /* If parsing the base offset fails, just unwind */
1965 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1968 curpos
= obj_offset
= base_offset
;
1969 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1970 if (type
<= OBJ_NONE
) {
1971 /* If getting the base itself fails, we first
1972 * retry the base, otherwise unwind */
1973 type
= retry_bad_packed_offset(p
, base_offset
);
1974 if (type
> OBJ_NONE
)
1988 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1989 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1994 if (poi_stack
!= small_poi_stack
)
1999 while (poi_stack_nr
) {
2000 obj_offset
= poi_stack
[--poi_stack_nr
];
2001 type
= retry_bad_packed_offset(p
, obj_offset
);
2002 if (type
> OBJ_NONE
)
2009 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
2010 struct object_info
*oi
)
2012 struct pack_window
*w_curs
= NULL
;
2014 off_t curpos
= obj_offset
;
2015 enum object_type type
;
2018 * We always get the representation type, but only convert it to
2019 * a "real" type later if the caller is interested.
2021 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2024 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2025 off_t tmp_pos
= curpos
;
2026 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
2032 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2033 if (*oi
->sizep
== 0) {
2042 if (oi
->disk_sizep
) {
2043 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2044 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2048 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
2049 if (*oi
->typep
< 0) {
2055 if (oi
->delta_base_sha1
) {
2056 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2057 const unsigned char *base
;
2059 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2066 hashcpy(oi
->delta_base_sha1
, base
);
2068 hashclr(oi
->delta_base_sha1
);
2072 unuse_pack(&w_curs
);
2076 static void *unpack_compressed_entry(struct packed_git
*p
,
2077 struct pack_window
**w_curs
,
2083 unsigned char *buffer
, *in
;
2085 buffer
= xmallocz_gently(size
);
2088 memset(&stream
, 0, sizeof(stream
));
2089 stream
.next_out
= buffer
;
2090 stream
.avail_out
= size
+ 1;
2092 git_inflate_init(&stream
);
2094 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2095 stream
.next_in
= in
;
2096 st
= git_inflate(&stream
, Z_FINISH
);
2097 if (!stream
.avail_out
)
2098 break; /* the payload is larger than it should be */
2099 curpos
+= stream
.next_in
- in
;
2100 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2101 git_inflate_end(&stream
);
2102 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2110 #define MAX_DELTA_CACHE (256)
2112 static size_t delta_base_cached
;
2114 static struct delta_base_cache_lru_list
{
2115 struct delta_base_cache_lru_list
*prev
;
2116 struct delta_base_cache_lru_list
*next
;
2117 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
2119 static struct delta_base_cache_entry
{
2120 struct delta_base_cache_lru_list lru
;
2122 struct packed_git
*p
;
2125 enum object_type type
;
2126 } delta_base_cache
[MAX_DELTA_CACHE
];
2128 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2132 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
2133 hash
+= (hash
>> 8) + (hash
>> 16);
2134 return hash
% MAX_DELTA_CACHE
;
2137 static struct delta_base_cache_entry
*
2138 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2140 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2141 return delta_base_cache
+ hash
;
2144 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
2145 struct packed_git
*p
, off_t base_offset
)
2147 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
2150 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2152 struct delta_base_cache_entry
*ent
;
2153 ent
= get_delta_base_cache_entry(p
, base_offset
);
2154 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
2157 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2160 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2161 ent
->lru
.prev
->next
= ent
->lru
.next
;
2162 delta_base_cached
-= ent
->size
;
2165 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2166 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2168 struct delta_base_cache_entry
*ent
;
2171 ent
= get_delta_base_cache_entry(p
, base_offset
);
2173 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2174 return unpack_entry(p
, base_offset
, type
, base_size
);
2179 clear_delta_base_cache_entry(ent
);
2181 ret
= xmemdupz(ent
->data
, ent
->size
);
2183 *base_size
= ent
->size
;
2187 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2192 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2193 ent
->lru
.prev
->next
= ent
->lru
.next
;
2194 delta_base_cached
-= ent
->size
;
2198 void clear_delta_base_cache(void)
2201 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2202 release_delta_base_cache(&delta_base_cache
[p
]);
2205 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2206 void *base
, unsigned long base_size
, enum object_type type
)
2208 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2209 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2210 struct delta_base_cache_lru_list
*lru
;
2212 release_delta_base_cache(ent
);
2213 delta_base_cached
+= base_size
;
2215 for (lru
= delta_base_cache_lru
.next
;
2216 delta_base_cached
> delta_base_cache_limit
2217 && lru
!= &delta_base_cache_lru
;
2219 struct delta_base_cache_entry
*f
= (void *)lru
;
2220 if (f
->type
== OBJ_BLOB
)
2221 release_delta_base_cache(f
);
2223 for (lru
= delta_base_cache_lru
.next
;
2224 delta_base_cached
> delta_base_cache_limit
2225 && lru
!= &delta_base_cache_lru
;
2227 struct delta_base_cache_entry
*f
= (void *)lru
;
2228 release_delta_base_cache(f
);
2232 ent
->base_offset
= base_offset
;
2235 ent
->size
= base_size
;
2236 ent
->lru
.next
= &delta_base_cache_lru
;
2237 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2238 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2239 delta_base_cache_lru
.prev
= &ent
->lru
;
2242 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2243 unsigned long *size
);
2245 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2247 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2248 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2249 p
->pack_name
, (uintmax_t)obj_offset
);
2252 int do_check_packed_object_crc
;
2254 #define UNPACK_ENTRY_STACK_PREALLOC 64
2255 struct unpack_entry_stack_ent
{
2261 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2262 enum object_type
*final_type
, unsigned long *final_size
)
2264 struct pack_window
*w_curs
= NULL
;
2265 off_t curpos
= obj_offset
;
2268 enum object_type type
;
2269 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2270 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2271 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2272 int base_from_cache
= 0;
2274 write_pack_access_log(p
, obj_offset
);
2276 /* PHASE 1: drill down to the innermost base object */
2280 struct delta_base_cache_entry
*ent
;
2282 ent
= get_delta_base_cache_entry(p
, curpos
);
2283 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2287 clear_delta_base_cache_entry(ent
);
2288 base_from_cache
= 1;
2292 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2293 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2294 unsigned long len
= revidx
[1].offset
- obj_offset
;
2295 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2296 const unsigned char *sha1
=
2297 nth_packed_object_sha1(p
, revidx
->nr
);
2298 error("bad packed object CRC for %s",
2300 mark_bad_packed_object(p
, sha1
);
2301 unuse_pack(&w_curs
);
2306 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2307 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2310 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2312 error("failed to validate delta base reference "
2313 "at offset %"PRIuMAX
" from %s",
2314 (uintmax_t)curpos
, p
->pack_name
);
2315 /* bail to phase 2, in hopes of recovery */
2320 /* push object, proceed to base */
2321 if (delta_stack_nr
>= delta_stack_alloc
2322 && delta_stack
== small_delta_stack
) {
2323 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2324 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2325 memcpy(delta_stack
, small_delta_stack
,
2326 sizeof(*delta_stack
)*delta_stack_nr
);
2328 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2330 i
= delta_stack_nr
++;
2331 delta_stack
[i
].obj_offset
= obj_offset
;
2332 delta_stack
[i
].curpos
= curpos
;
2333 delta_stack
[i
].size
= size
;
2335 curpos
= obj_offset
= base_offset
;
2338 /* PHASE 2: handle the base */
2343 die("BUG in unpack_entry: left loop at a valid delta");
2349 if (!base_from_cache
)
2350 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2354 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2355 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2358 /* PHASE 3: apply deltas in order */
2361 * 'data' holds the base data, or NULL if there was corruption
2363 while (delta_stack_nr
) {
2366 unsigned long delta_size
, base_size
= size
;
2372 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2376 * We're probably in deep shit, but let's try to fetch
2377 * the required base anyway from another pack or loose.
2378 * This is costly but should happen only in the presence
2379 * of a corrupted pack, and is better than failing outright.
2381 struct revindex_entry
*revidx
;
2382 const unsigned char *base_sha1
;
2383 revidx
= find_pack_revindex(p
, obj_offset
);
2385 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2386 error("failed to read delta base object %s"
2387 " at offset %"PRIuMAX
" from %s",
2388 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2390 mark_bad_packed_object(p
, base_sha1
);
2391 base
= read_object(base_sha1
, &type
, &base_size
);
2395 i
= --delta_stack_nr
;
2396 obj_offset
= delta_stack
[i
].obj_offset
;
2397 curpos
= delta_stack
[i
].curpos
;
2398 delta_size
= delta_stack
[i
].size
;
2403 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2406 error("failed to unpack compressed delta "
2407 "at offset %"PRIuMAX
" from %s",
2408 (uintmax_t)curpos
, p
->pack_name
);
2413 data
= patch_delta(base
, base_size
,
2414 delta_data
, delta_size
,
2418 * We could not apply the delta; warn the user, but keep going.
2419 * Our failure will be noticed either in the next iteration of
2420 * the loop, or if this is the final delta, in the caller when
2421 * we return NULL. Those code paths will take care of making
2422 * a more explicit warning and retrying with another copy of
2426 error("failed to apply delta");
2434 unuse_pack(&w_curs
);
2436 if (delta_stack
!= small_delta_stack
)
2442 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2445 const unsigned char *index
= p
->index_data
;
2447 if (open_pack_index(p
))
2449 index
= p
->index_data
;
2451 if (n
>= p
->num_objects
)
2454 if (p
->index_version
== 1) {
2455 return index
+ 24 * n
+ 4;
2458 return index
+ 20 * n
;
2462 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2464 const unsigned char *index
= p
->index_data
;
2466 if (p
->index_version
== 1) {
2467 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2470 index
+= 8 + p
->num_objects
* (20 + 4);
2471 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2472 if (!(off
& 0x80000000))
2474 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2475 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2476 ntohl(*((uint32_t *)(index
+ 4)));
2480 off_t
find_pack_entry_one(const unsigned char *sha1
,
2481 struct packed_git
*p
)
2483 const uint32_t *level1_ofs
= p
->index_data
;
2484 const unsigned char *index
= p
->index_data
;
2485 unsigned hi
, lo
, stride
;
2486 static int use_lookup
= -1;
2487 static int debug_lookup
= -1;
2489 if (debug_lookup
< 0)
2490 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2493 if (open_pack_index(p
))
2495 level1_ofs
= p
->index_data
;
2496 index
= p
->index_data
;
2498 if (p
->index_version
> 1) {
2503 hi
= ntohl(level1_ofs
[*sha1
]);
2504 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2505 if (p
->index_version
> 1) {
2513 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2514 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2517 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2519 int pos
= sha1_entry_pos(index
, stride
, 0,
2520 lo
, hi
, p
->num_objects
, sha1
);
2523 return nth_packed_object_offset(p
, pos
);
2527 unsigned mi
= (lo
+ hi
) / 2;
2528 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2531 printf("lo %u hi %u rg %u mi %u\n",
2532 lo
, hi
, hi
- lo
, mi
);
2534 return nth_packed_object_offset(p
, mi
);
2543 int is_pack_valid(struct packed_git
*p
)
2545 /* An already open pack is known to be valid. */
2546 if (p
->pack_fd
!= -1)
2549 /* If the pack has one window completely covering the
2550 * file size, the pack is known to be valid even if
2551 * the descriptor is not currently open.
2554 struct pack_window
*w
= p
->windows
;
2556 if (!w
->offset
&& w
->len
== p
->pack_size
)
2560 /* Force the pack to open to prove its valid. */
2561 return !open_packed_git(p
);
2564 static int fill_pack_entry(const unsigned char *sha1
,
2565 struct pack_entry
*e
,
2566 struct packed_git
*p
)
2570 if (p
->num_bad_objects
) {
2572 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2573 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2577 offset
= find_pack_entry_one(sha1
, p
);
2582 * We are about to tell the caller where they can locate the
2583 * requested object. We better make sure the packfile is
2584 * still here and can be accessed before supplying that
2585 * answer, as it may have been deleted since the index was
2588 if (!is_pack_valid(p
))
2592 hashcpy(e
->sha1
, sha1
);
2597 * Iff a pack file contains the object named by sha1, return true and
2598 * store its location to e.
2600 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2602 struct packed_git
*p
;
2604 prepare_packed_git();
2608 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2611 for (p
= packed_git
; p
; p
= p
->next
) {
2612 if (p
== last_found_pack
)
2613 continue; /* we already checked this one */
2615 if (fill_pack_entry(sha1
, e
, p
)) {
2616 last_found_pack
= p
;
2623 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2624 struct packed_git
*packs
)
2626 struct packed_git
*p
;
2628 for (p
= packs
; p
; p
= p
->next
) {
2629 if (find_pack_entry_one(sha1
, p
))
2636 static int sha1_loose_object_info(const unsigned char *sha1
,
2637 struct object_info
*oi
,
2641 unsigned long mapsize
;
2645 struct strbuf hdrbuf
= STRBUF_INIT
;
2647 if (oi
->delta_base_sha1
)
2648 hashclr(oi
->delta_base_sha1
);
2651 * If we don't care about type or size, then we don't
2652 * need to look inside the object at all. Note that we
2653 * do not optimize out the stat call, even if the
2654 * caller doesn't care about the disk-size, since our
2655 * return value implicitly indicates whether the
2656 * object even exists.
2658 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
) {
2660 if (stat_sha1_file(sha1
, &st
) < 0)
2663 *oi
->disk_sizep
= st
.st_size
;
2667 map
= map_sha1_file(sha1
, &mapsize
);
2671 *oi
->disk_sizep
= mapsize
;
2672 if ((flags
& LOOKUP_UNKNOWN_OBJECT
)) {
2673 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2674 status
= error("unable to unpack %s header with --allow-unknown-type",
2676 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2677 status
= error("unable to unpack %s header",
2681 else if (hdrbuf
.len
) {
2682 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2683 status
= error("unable to parse %s header with --allow-unknown-type",
2685 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2686 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2687 git_inflate_end(&stream
);
2688 munmap(map
, mapsize
);
2689 if (status
&& oi
->typep
)
2690 *oi
->typep
= status
;
2691 strbuf_release(&hdrbuf
);
2695 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2697 struct cached_object
*co
;
2698 struct pack_entry e
;
2700 enum object_type real_type
;
2701 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2703 co
= find_cached_object(real
);
2706 *(oi
->typep
) = co
->type
;
2708 *(oi
->sizep
) = co
->size
;
2710 *(oi
->disk_sizep
) = 0;
2711 if (oi
->delta_base_sha1
)
2712 hashclr(oi
->delta_base_sha1
);
2714 strbuf_addstr(oi
->typename
, typename(co
->type
));
2715 oi
->whence
= OI_CACHED
;
2719 if (!find_pack_entry(real
, &e
)) {
2720 /* Most likely it's a loose object. */
2721 if (!sha1_loose_object_info(real
, oi
, flags
)) {
2722 oi
->whence
= OI_LOOSE
;
2726 /* Not a loose object; someone else may have just packed it. */
2727 reprepare_packed_git();
2728 if (!find_pack_entry(real
, &e
))
2733 * packed_object_info() does not follow the delta chain to
2734 * find out the real type, unless it is given oi->typep.
2736 if (oi
->typename
&& !oi
->typep
)
2737 oi
->typep
= &real_type
;
2739 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2741 mark_bad_packed_object(e
.p
, real
);
2742 if (oi
->typep
== &real_type
)
2744 return sha1_object_info_extended(real
, oi
, 0);
2745 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2746 oi
->whence
= OI_DBCACHED
;
2748 oi
->whence
= OI_PACKED
;
2749 oi
->u
.packed
.offset
= e
.offset
;
2750 oi
->u
.packed
.pack
= e
.p
;
2751 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2752 rtype
== OBJ_OFS_DELTA
);
2755 strbuf_addstr(oi
->typename
, typename(*oi
->typep
));
2756 if (oi
->typep
== &real_type
)
2762 /* returns enum object_type or negative */
2763 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2765 enum object_type type
;
2766 struct object_info oi
= {NULL
};
2770 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2775 static void *read_packed_sha1(const unsigned char *sha1
,
2776 enum object_type
*type
, unsigned long *size
)
2778 struct pack_entry e
;
2781 if (!find_pack_entry(sha1
, &e
))
2783 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2786 * We're probably in deep shit, but let's try to fetch
2787 * the required object anyway from another pack or loose.
2788 * This should happen only in the presence of a corrupted
2789 * pack, and is better than failing outright.
2791 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2792 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2793 mark_bad_packed_object(e
.p
, sha1
);
2794 data
= read_object(sha1
, type
, size
);
2799 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2800 unsigned char *sha1
)
2802 struct cached_object
*co
;
2804 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2805 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2807 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2808 co
= &cached_objects
[cached_object_nr
++];
2811 co
->buf
= xmalloc(len
);
2812 memcpy(co
->buf
, buf
, len
);
2813 hashcpy(co
->sha1
, sha1
);
2817 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2818 unsigned long *size
)
2820 unsigned long mapsize
;
2822 struct cached_object
*co
;
2824 co
= find_cached_object(sha1
);
2828 return xmemdupz(co
->buf
, co
->size
);
2831 buf
= read_packed_sha1(sha1
, type
, size
);
2834 map
= map_sha1_file(sha1
, &mapsize
);
2836 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2837 munmap(map
, mapsize
);
2840 reprepare_packed_git();
2841 return read_packed_sha1(sha1
, type
, size
);
2845 * This function dies on corrupt objects; the callers who want to
2846 * deal with them should arrange to call read_object() and give error
2847 * messages themselves.
2849 void *read_sha1_file_extended(const unsigned char *sha1
,
2850 enum object_type
*type
,
2851 unsigned long *size
,
2855 const struct packed_git
*p
;
2856 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2859 data
= read_object(repl
, type
, size
);
2863 if (errno
&& errno
!= ENOENT
)
2864 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2866 /* die if we replaced an object with one that does not exist */
2868 die("replacement %s not found for %s",
2869 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2871 if (has_loose_object(repl
)) {
2872 const char *path
= sha1_file_name(sha1
);
2874 die("loose object %s (stored in %s) is corrupt",
2875 sha1_to_hex(repl
), path
);
2878 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2879 die("packed object %s (stored in %s) is corrupt",
2880 sha1_to_hex(repl
), p
->pack_name
);
2885 void *read_object_with_reference(const unsigned char *sha1
,
2886 const char *required_type_name
,
2887 unsigned long *size
,
2888 unsigned char *actual_sha1_return
)
2890 enum object_type type
, required_type
;
2892 unsigned long isize
;
2893 unsigned char actual_sha1
[20];
2895 required_type
= type_from_string(required_type_name
);
2896 hashcpy(actual_sha1
, sha1
);
2898 int ref_length
= -1;
2899 const char *ref_type
= NULL
;
2901 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2904 if (type
== required_type
) {
2906 if (actual_sha1_return
)
2907 hashcpy(actual_sha1_return
, actual_sha1
);
2910 /* Handle references */
2911 else if (type
== OBJ_COMMIT
)
2913 else if (type
== OBJ_TAG
)
2914 ref_type
= "object ";
2919 ref_length
= strlen(ref_type
);
2921 if (ref_length
+ 40 > isize
||
2922 memcmp(buffer
, ref_type
, ref_length
) ||
2923 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2928 /* Now we have the ID of the referred-to object in
2929 * actual_sha1. Check again. */
2933 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2934 const char *type
, unsigned char *sha1
,
2935 char *hdr
, int *hdrlen
)
2939 /* Generate the header */
2940 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
2944 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2945 git_SHA1_Update(&c
, buf
, len
);
2946 git_SHA1_Final(sha1
, &c
);
2950 * Move the just written object into its final resting place.
2952 int finalize_object_file(const char *tmpfile
, const char *filename
)
2956 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2958 else if (link(tmpfile
, filename
))
2962 * Coda hack - coda doesn't like cross-directory links,
2963 * so we fall back to a rename, which will mean that it
2964 * won't be able to check collisions, but that's not a
2967 * The same holds for FAT formatted media.
2969 * When this succeeds, we just return. We have nothing
2972 if (ret
&& ret
!= EEXIST
) {
2974 if (!rename(tmpfile
, filename
))
2978 unlink_or_warn(tmpfile
);
2980 if (ret
!= EEXIST
) {
2981 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2983 /* FIXME!!! Collision check here ? */
2987 if (adjust_shared_perm(filename
))
2988 return error("unable to set permission to '%s'", filename
);
2992 static int write_buffer(int fd
, const void *buf
, size_t len
)
2994 if (write_in_full(fd
, buf
, len
) < 0)
2995 return error("file write error (%s)", strerror(errno
));
2999 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
3000 unsigned char *sha1
)
3003 int hdrlen
= sizeof(hdr
);
3004 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3008 /* Finalize a file on disk, and close it. */
3009 static void close_sha1_file(int fd
)
3011 if (fsync_object_files
)
3012 fsync_or_die(fd
, "sha1 file");
3014 die_errno("error when closing sha1 file");
3017 /* Size of directory component, including the ending '/' */
3018 static inline int directory_size(const char *filename
)
3020 const char *s
= strrchr(filename
, '/');
3023 return s
- filename
+ 1;
3027 * This creates a temporary file in the same directory as the final
3030 * We want to avoid cross-directory filename renames, because those
3031 * can have problems on various filesystems (FAT, NFS, Coda).
3033 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
3035 int fd
, dirlen
= directory_size(filename
);
3037 if (dirlen
+ 20 > bufsiz
) {
3038 errno
= ENAMETOOLONG
;
3041 memcpy(buffer
, filename
, dirlen
);
3042 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
3043 fd
= git_mkstemp_mode(buffer
, 0444);
3044 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3045 /* Make sure the directory exists */
3046 memcpy(buffer
, filename
, dirlen
);
3047 buffer
[dirlen
-1] = 0;
3048 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
3050 if (adjust_shared_perm(buffer
))
3054 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
3055 fd
= git_mkstemp_mode(buffer
, 0444);
3060 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3061 const void *buf
, unsigned long len
, time_t mtime
)
3064 unsigned char compressed
[4096];
3067 unsigned char parano_sha1
[20];
3068 static char tmp_file
[PATH_MAX
];
3069 const char *filename
= sha1_file_name(sha1
);
3071 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
3073 if (errno
== EACCES
)
3074 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3076 return error("unable to create temporary file: %s", strerror(errno
));
3080 git_deflate_init(&stream
, zlib_compression_level
);
3081 stream
.next_out
= compressed
;
3082 stream
.avail_out
= sizeof(compressed
);
3085 /* First header.. */
3086 stream
.next_in
= (unsigned char *)hdr
;
3087 stream
.avail_in
= hdrlen
;
3088 while (git_deflate(&stream
, 0) == Z_OK
)
3090 git_SHA1_Update(&c
, hdr
, hdrlen
);
3092 /* Then the data itself.. */
3093 stream
.next_in
= (void *)buf
;
3094 stream
.avail_in
= len
;
3096 unsigned char *in0
= stream
.next_in
;
3097 ret
= git_deflate(&stream
, Z_FINISH
);
3098 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3099 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3100 die("unable to write sha1 file");
3101 stream
.next_out
= compressed
;
3102 stream
.avail_out
= sizeof(compressed
);
3103 } while (ret
== Z_OK
);
3105 if (ret
!= Z_STREAM_END
)
3106 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3107 ret
= git_deflate_end_gently(&stream
);
3109 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3110 git_SHA1_Final(parano_sha1
, &c
);
3111 if (hashcmp(sha1
, parano_sha1
) != 0)
3112 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3114 close_sha1_file(fd
);
3119 utb
.modtime
= mtime
;
3120 if (utime(tmp_file
, &utb
) < 0)
3121 warning("failed utime() on %s: %s",
3122 tmp_file
, strerror(errno
));
3125 return finalize_object_file(tmp_file
, filename
);
3128 static int freshen_loose_object(const unsigned char *sha1
)
3130 return check_and_freshen(sha1
, 1);
3133 static int freshen_packed_object(const unsigned char *sha1
)
3135 struct pack_entry e
;
3136 if (!find_pack_entry(sha1
, &e
))
3140 if (!freshen_file(e
.p
->pack_name
))
3146 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3149 int hdrlen
= sizeof(hdr
);
3151 /* Normally if we have it in the pack then we do not bother writing
3152 * it out into .git/objects/??/?{38} file.
3154 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3155 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3157 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3160 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3161 unsigned char *sha1
, unsigned flags
)
3164 int hdrlen
, status
= 0;
3166 /* type string, SP, %lu of the length plus NUL must fit this */
3167 hdrlen
= strlen(type
) + 32;
3168 header
= xmalloc(hdrlen
);
3169 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3171 if (!(flags
& HASH_WRITE_OBJECT
))
3173 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3175 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3182 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3186 enum object_type type
;
3191 if (has_loose_object(sha1
))
3193 buf
= read_packed_sha1(sha1
, &type
, &len
);
3195 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3196 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3197 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3203 int has_pack_index(const unsigned char *sha1
)
3206 if (stat(sha1_pack_index_name(sha1
), &st
))
3211 int has_sha1_pack(const unsigned char *sha1
)
3213 struct pack_entry e
;
3214 return find_pack_entry(sha1
, &e
);
3217 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3219 struct pack_entry e
;
3221 if (find_pack_entry(sha1
, &e
))
3223 if (has_loose_object(sha1
))
3225 if (flags
& HAS_SHA1_QUICK
)
3227 reprepare_packed_git();
3228 return find_pack_entry(sha1
, &e
);
3231 static void check_tree(const void *buf
, size_t size
)
3233 struct tree_desc desc
;
3234 struct name_entry entry
;
3236 init_tree_desc(&desc
, buf
, size
);
3237 while (tree_entry(&desc
, &entry
))
3239 * tree_entry() will die() on malformed entries */
3243 static void check_commit(const void *buf
, size_t size
)
3246 memset(&c
, 0, sizeof(c
));
3247 if (parse_commit_buffer(&c
, buf
, size
))
3248 die("corrupt commit");
3251 static void check_tag(const void *buf
, size_t size
)
3254 memset(&t
, 0, sizeof(t
));
3255 if (parse_tag_buffer(&t
, buf
, size
))
3259 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3260 enum object_type type
,
3261 const char *path
, unsigned flags
)
3263 int ret
, re_allocated
= 0;
3264 int write_object
= flags
& HASH_WRITE_OBJECT
;
3270 * Convert blobs to git internal format
3272 if ((type
== OBJ_BLOB
) && path
) {
3273 struct strbuf nbuf
= STRBUF_INIT
;
3274 if (convert_to_git(path
, buf
, size
, &nbuf
,
3275 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3276 buf
= strbuf_detach(&nbuf
, &size
);
3280 if (flags
& HASH_FORMAT_CHECK
) {
3281 if (type
== OBJ_TREE
)
3282 check_tree(buf
, size
);
3283 if (type
== OBJ_COMMIT
)
3284 check_commit(buf
, size
);
3285 if (type
== OBJ_TAG
)
3286 check_tag(buf
, size
);
3290 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3292 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3298 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3299 const char *path
, unsigned flags
)
3302 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3303 struct strbuf sbuf
= STRBUF_INIT
;
3306 assert(would_convert_to_git_filter_fd(path
));
3308 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3309 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3312 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3315 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3317 strbuf_release(&sbuf
);
3321 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3322 const char *path
, unsigned flags
)
3324 struct strbuf sbuf
= STRBUF_INIT
;
3327 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3328 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3331 strbuf_release(&sbuf
);
3335 #define SMALL_FILE_SIZE (32*1024)
3337 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3338 enum object_type type
, const char *path
,
3344 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3345 } else if (size
<= SMALL_FILE_SIZE
) {
3346 char *buf
= xmalloc(size
);
3347 if (size
== read_in_full(fd
, buf
, size
))
3348 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3350 ret
= error("short read %s", strerror(errno
));
3353 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3354 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3361 * This creates one packfile per large blob unless bulk-checkin
3362 * machinery is "plugged".
3364 * This also bypasses the usual "convert-to-git" dance, and that is on
3365 * purpose. We could write a streaming version of the converting
3366 * functions and insert that before feeding the data to fast-import
3367 * (or equivalent in-core API described above). However, that is
3368 * somewhat complicated, as we do not know the size of the filter
3369 * result, which we need to know beforehand when writing a git object.
3370 * Since the primary motivation for trying to stream from the working
3371 * tree file and to avoid mmaping it in core is to deal with large
3372 * binary blobs, they generally do not want to get any conversion, and
3373 * callers should avoid this code path when filters are requested.
3375 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3376 enum object_type type
, const char *path
,
3379 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3382 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3383 enum object_type type
, const char *path
, unsigned flags
)
3388 * Call xsize_t() only when needed to avoid potentially unnecessary
3389 * die() for large files.
3391 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3392 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3393 else if (!S_ISREG(st
->st_mode
))
3394 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3395 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3396 (path
&& would_convert_to_git(path
)))
3397 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3400 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3406 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3409 struct strbuf sb
= STRBUF_INIT
;
3411 switch (st
->st_mode
& S_IFMT
) {
3413 fd
= open(path
, O_RDONLY
);
3415 return error("open(\"%s\"): %s", path
,
3417 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3418 return error("%s: failed to insert into database",
3422 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3423 char *errstr
= strerror(errno
);
3424 return error("readlink(\"%s\"): %s", path
,
3427 if (!(flags
& HASH_WRITE_OBJECT
))
3428 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3429 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3430 return error("%s: failed to insert into database",
3432 strbuf_release(&sb
);
3435 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3437 return error("%s: unsupported file type", path
);
3442 int read_pack_header(int fd
, struct pack_header
*header
)
3444 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3445 /* "eof before pack header was fully read" */
3446 return PH_ERROR_EOF
;
3448 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3449 /* "protocol error (pack signature mismatch detected)" */
3450 return PH_ERROR_PACK_SIGNATURE
;
3451 if (!pack_version_ok(header
->hdr_version
))
3452 /* "protocol error (pack version unsupported)" */
3453 return PH_ERROR_PROTOCOL
;
3457 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3459 enum object_type type
= sha1_object_info(sha1
, NULL
);
3461 die("%s is not a valid object", sha1_to_hex(sha1
));
3463 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3467 static int for_each_file_in_obj_subdir(int subdir_nr
,
3468 struct strbuf
*path
,
3469 each_loose_object_fn obj_cb
,
3470 each_loose_cruft_fn cruft_cb
,
3471 each_loose_subdir_fn subdir_cb
,
3474 size_t baselen
= path
->len
;
3475 DIR *dir
= opendir(path
->buf
);
3480 if (errno
== ENOENT
)
3482 return error("unable to open %s: %s", path
->buf
, strerror(errno
));
3485 while ((de
= readdir(dir
))) {
3486 if (is_dot_or_dotdot(de
->d_name
))
3489 strbuf_setlen(path
, baselen
);
3490 strbuf_addf(path
, "/%s", de
->d_name
);
3492 if (strlen(de
->d_name
) == 38) {
3494 unsigned char sha1
[20];
3496 snprintf(hex
, sizeof(hex
), "%02x%s",
3497 subdir_nr
, de
->d_name
);
3498 if (!get_sha1_hex(hex
, sha1
)) {
3500 r
= obj_cb(sha1
, path
->buf
, data
);
3509 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3514 strbuf_setlen(path
, baselen
);
3516 if (!r
&& subdir_cb
)
3517 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3523 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3524 each_loose_object_fn obj_cb
,
3525 each_loose_cruft_fn cruft_cb
,
3526 each_loose_subdir_fn subdir_cb
,
3529 size_t baselen
= path
->len
;
3533 for (i
= 0; i
< 256; i
++) {
3534 strbuf_addf(path
, "/%02x", i
);
3535 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3537 strbuf_setlen(path
, baselen
);
3545 int for_each_loose_file_in_objdir(const char *path
,
3546 each_loose_object_fn obj_cb
,
3547 each_loose_cruft_fn cruft_cb
,
3548 each_loose_subdir_fn subdir_cb
,
3551 struct strbuf buf
= STRBUF_INIT
;
3554 strbuf_addstr(&buf
, path
);
3555 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3557 strbuf_release(&buf
);
3562 struct loose_alt_odb_data
{
3563 each_loose_object_fn
*cb
;
3567 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3570 struct loose_alt_odb_data
*data
= vdata
;
3571 struct strbuf buf
= STRBUF_INIT
;
3574 /* copy base not including trailing '/' */
3575 strbuf_add(&buf
, alt
->base
, alt
->name
- alt
->base
- 1);
3576 r
= for_each_loose_file_in_objdir_buf(&buf
,
3577 data
->cb
, NULL
, NULL
,
3579 strbuf_release(&buf
);
3583 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3585 struct loose_alt_odb_data alt
;
3588 r
= for_each_loose_file_in_objdir(get_object_directory(),
3589 cb
, NULL
, NULL
, data
);
3593 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3598 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3601 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3606 for (i
= 0; i
< p
->num_objects
; i
++) {
3607 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3610 return error("unable to get sha1 of object %u in %s",
3613 r
= cb(sha1
, p
, i
, data
);
3620 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3622 struct packed_git
*p
;
3624 int pack_errors
= 0;
3626 prepare_packed_git();
3627 for (p
= packed_git
; p
; p
= p
->next
) {
3628 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3630 if (open_pack_index(p
)) {
3634 r
= for_each_object_in_pack(p
, cb
, data
);
3638 return r
? r
: pack_errors
;