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
;
256 size_t pfxlen
, entlen
;
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
= st_add(pfxlen
, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
277 ent
= xmalloc(st_add(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 (!fspathcmp(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);
429 * Compute the exact path an alternate is at and returns it. In case of
430 * error NULL is returned and the human readable error is added to `err`
431 * `path` may be relative and should point to $GITDIR.
432 * `err` must not be null.
434 char *compute_alternate_path(const char *path
, struct strbuf
*err
)
436 char *ref_git
= NULL
;
437 const char *repo
, *ref_git_s
;
440 ref_git_s
= real_path_if_valid(path
);
443 strbuf_addf(err
, _("path '%s' does not exist"), path
);
447 * Beware: read_gitfile(), real_path() and mkpath()
448 * return static buffer
450 ref_git
= xstrdup(ref_git_s
);
452 repo
= read_gitfile(ref_git
);
454 repo
= read_gitfile(mkpath("%s/.git", ref_git
));
457 ref_git
= xstrdup(repo
);
460 if (!repo
&& is_directory(mkpath("%s/.git/objects", ref_git
))) {
461 char *ref_git_git
= mkpathdup("%s/.git", ref_git
);
463 ref_git
= ref_git_git
;
464 } else if (!is_directory(mkpath("%s/objects", ref_git
))) {
465 struct strbuf sb
= STRBUF_INIT
;
467 if (get_common_dir(&sb
, ref_git
)) {
469 _("reference repository '%s' as a linked "
470 "checkout is not supported yet."),
475 strbuf_addf(err
, _("reference repository '%s' is not a "
476 "local repository."), path
);
480 if (!access(mkpath("%s/shallow", ref_git
), F_OK
)) {
481 strbuf_addf(err
, _("reference repository '%s' is shallow"),
487 if (!access(mkpath("%s/info/grafts", ref_git
), F_OK
)) {
489 _("reference repository '%s' is grafted"),
504 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
506 struct alternate_object_database
*ent
;
510 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
518 void prepare_alt_odb(void)
525 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
528 alt_odb_tail
= &alt_odb_list
;
529 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
531 read_info_alternates(get_object_directory(), 0);
534 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
535 static int freshen_file(const char *fn
)
538 t
.actime
= t
.modtime
= time(NULL
);
539 return !utime(fn
, &t
);
543 * All of the check_and_freshen functions return 1 if the file exists and was
544 * freshened (if freshening was requested), 0 otherwise. If they return
545 * 0, you should not assume that it is safe to skip a write of the object (it
546 * either does not exist on disk, or has a stale mtime and may be subject to
549 static int check_and_freshen_file(const char *fn
, int freshen
)
551 if (access(fn
, F_OK
))
553 if (freshen
&& !freshen_file(fn
))
558 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
560 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
563 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
565 struct alternate_object_database
*alt
;
567 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
568 fill_sha1_path(alt
->name
, sha1
);
569 if (check_and_freshen_file(alt
->base
, freshen
))
575 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
577 return check_and_freshen_local(sha1
, freshen
) ||
578 check_and_freshen_nonlocal(sha1
, freshen
);
581 int has_loose_object_nonlocal(const unsigned char *sha1
)
583 return check_and_freshen_nonlocal(sha1
, 0);
586 static int has_loose_object(const unsigned char *sha1
)
588 return check_and_freshen(sha1
, 0);
591 static unsigned int pack_used_ctr
;
592 static unsigned int pack_mmap_calls
;
593 static unsigned int peak_pack_open_windows
;
594 static unsigned int pack_open_windows
;
595 static unsigned int pack_open_fds
;
596 static unsigned int pack_max_fds
;
597 static size_t peak_pack_mapped
;
598 static size_t pack_mapped
;
599 struct packed_git
*packed_git
;
601 void pack_report(void)
604 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
605 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
606 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
607 sz_fmt(getpagesize()),
608 sz_fmt(packed_git_window_size
),
609 sz_fmt(packed_git_limit
));
611 "pack_report: pack_used_ctr = %10u\n"
612 "pack_report: pack_mmap_calls = %10u\n"
613 "pack_report: pack_open_windows = %10u / %10u\n"
614 "pack_report: pack_mapped = "
615 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
618 pack_open_windows
, peak_pack_open_windows
,
619 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
623 * Open and mmap the index file at path, perform a couple of
624 * consistency checks, then record its information to p. Return 0 on
627 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
630 struct pack_idx_header
*hdr
;
632 uint32_t version
, nr
, i
, *index
;
633 int fd
= git_open_noatime(path
);
638 if (fstat(fd
, &st
)) {
642 idx_size
= xsize_t(st
.st_size
);
643 if (idx_size
< 4 * 256 + 20 + 20) {
645 return error("index file %s is too small", path
);
647 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
651 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
652 version
= ntohl(hdr
->idx_version
);
653 if (version
< 2 || version
> 2) {
654 munmap(idx_map
, idx_size
);
655 return error("index file %s is version %"PRIu32
656 " and is not supported by this binary"
657 " (try upgrading GIT to a newer version)",
666 index
+= 2; /* skip index header */
667 for (i
= 0; i
< 256; i
++) {
668 uint32_t n
= ntohl(index
[i
]);
670 munmap(idx_map
, idx_size
);
671 return error("non-monotonic index %s", path
);
679 * - 256 index entries 4 bytes each
680 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
681 * - 20-byte SHA1 of the packfile
682 * - 20-byte SHA1 file checksum
684 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
685 munmap(idx_map
, idx_size
);
686 return error("wrong index v1 file size in %s", path
);
688 } else if (version
== 2) {
691 * - 8 bytes of header
692 * - 256 index entries 4 bytes each
693 * - 20-byte sha1 entry * nr
694 * - 4-byte crc entry * nr
695 * - 4-byte offset entry * nr
696 * - 20-byte SHA1 of the packfile
697 * - 20-byte SHA1 file checksum
698 * And after the 4-byte offset table might be a
699 * variable sized table containing 8-byte entries
700 * for offsets larger than 2^31.
702 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
703 unsigned long max_size
= min_size
;
705 max_size
+= (nr
- 1)*8;
706 if (idx_size
< min_size
|| idx_size
> max_size
) {
707 munmap(idx_map
, idx_size
);
708 return error("wrong index v2 file size in %s", path
);
710 if (idx_size
!= min_size
&&
712 * make sure we can deal with large pack offsets.
713 * 31-bit signed offset won't be enough, neither
714 * 32-bit unsigned one will be.
716 (sizeof(off_t
) <= 4)) {
717 munmap(idx_map
, idx_size
);
718 return error("pack too large for current definition of off_t in %s", path
);
722 p
->index_version
= version
;
723 p
->index_data
= idx_map
;
724 p
->index_size
= idx_size
;
729 int open_pack_index(struct packed_git
*p
)
738 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
739 die("BUG: pack_name does not end in .pack");
740 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
741 ret
= check_packed_git_idx(idx_name
, p
);
746 static void scan_windows(struct packed_git
*p
,
747 struct packed_git
**lru_p
,
748 struct pack_window
**lru_w
,
749 struct pack_window
**lru_l
)
751 struct pack_window
*w
, *w_l
;
753 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
755 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
765 static int unuse_one_window(struct packed_git
*current
)
767 struct packed_git
*p
, *lru_p
= NULL
;
768 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
771 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
772 for (p
= packed_git
; p
; p
= p
->next
)
773 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
775 munmap(lru_w
->base
, lru_w
->len
);
776 pack_mapped
-= lru_w
->len
;
778 lru_l
->next
= lru_w
->next
;
780 lru_p
->windows
= lru_w
->next
;
788 void release_pack_memory(size_t need
)
790 size_t cur
= pack_mapped
;
791 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
795 static void mmap_limit_check(size_t length
)
797 static size_t limit
= 0;
799 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
804 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
805 (uintmax_t)length
, (uintmax_t)limit
);
808 void *xmmap_gently(void *start
, size_t length
,
809 int prot
, int flags
, int fd
, off_t offset
)
813 mmap_limit_check(length
);
814 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
815 if (ret
== MAP_FAILED
) {
818 release_pack_memory(length
);
819 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
824 void *xmmap(void *start
, size_t length
,
825 int prot
, int flags
, int fd
, off_t offset
)
827 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
828 if (ret
== MAP_FAILED
)
829 die_errno("mmap failed");
833 void close_pack_windows(struct packed_git
*p
)
836 struct pack_window
*w
= p
->windows
;
839 die("pack '%s' still has open windows to it",
841 munmap(w
->base
, w
->len
);
842 pack_mapped
-= w
->len
;
844 p
->windows
= w
->next
;
849 static int close_pack_fd(struct packed_git
*p
)
861 static void close_pack(struct packed_git
*p
)
863 close_pack_windows(p
);
868 void close_all_packs(void)
870 struct packed_git
*p
;
872 for (p
= packed_git
; p
; p
= p
->next
)
874 die("BUG! Want to close pack marked 'do-not-close'");
881 * The LRU pack is the one with the oldest MRU window, preferring packs
882 * with no used windows, or the oldest mtime if it has no windows allocated.
884 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
886 struct pack_window
*w
, *this_mru_w
;
887 int has_windows_inuse
= 0;
890 * Reject this pack if it has windows and the previously selected
891 * one does not. If this pack does not have windows, reject
892 * it if the pack file is newer than the previously selected one.
894 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
897 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
899 * Reject this pack if any of its windows are in use,
900 * but the previously selected pack did not have any
901 * inuse windows. Otherwise, record that this pack
902 * has windows in use.
905 if (*accept_windows_inuse
)
906 has_windows_inuse
= 1;
911 if (w
->last_used
> this_mru_w
->last_used
)
915 * Reject this pack if it has windows that have been
916 * used more recently than the previously selected pack.
917 * If the previously selected pack had windows inuse and
918 * we have not encountered a window in this pack that is
919 * inuse, skip this check since we prefer a pack with no
920 * inuse windows to one that has inuse windows.
922 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
923 this_mru_w
->last_used
> (*mru_w
)->last_used
)
932 *accept_windows_inuse
= has_windows_inuse
;
935 static int close_one_pack(void)
937 struct packed_git
*p
, *lru_p
= NULL
;
938 struct pack_window
*mru_w
= NULL
;
939 int accept_windows_inuse
= 1;
941 for (p
= packed_git
; p
; p
= p
->next
) {
942 if (p
->pack_fd
== -1)
944 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
948 return close_pack_fd(lru_p
);
953 void unuse_pack(struct pack_window
**w_cursor
)
955 struct pack_window
*w
= *w_cursor
;
962 void close_pack_index(struct packed_git
*p
)
965 munmap((void *)p
->index_data
, p
->index_size
);
966 p
->index_data
= NULL
;
971 * This is used by git-repack in case a newly created pack happens to
972 * contain the same set of objects as an existing one. In that case
973 * the resulting file might be different even if its name would be the
974 * same. It is best to close any reference to the old pack before it is
975 * replaced on disk. Of course no index pointers or windows for given pack
976 * must subsist at this point. If ever objects from this pack are requested
977 * again, the new version of the pack will be reinitialized through
978 * reprepare_packed_git().
980 void free_pack_by_name(const char *pack_name
)
982 struct packed_git
*p
, **pp
= &packed_git
;
986 if (strcmp(pack_name
, p
->pack_name
) == 0) {
987 clear_delta_base_cache();
989 free(p
->bad_object_sha1
);
991 if (last_found_pack
== p
)
992 last_found_pack
= NULL
;
1000 static unsigned int get_max_fd_limit(void)
1002 #ifdef RLIMIT_NOFILE
1006 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
1007 return lim
.rlim_cur
;
1013 long open_max
= sysconf(_SC_OPEN_MAX
);
1017 * Otherwise, we got -1 for one of the two
1020 * (1) sysconf() did not understand _SC_OPEN_MAX
1021 * and signaled an error with -1; or
1022 * (2) sysconf() said there is no limit.
1024 * We _could_ clear errno before calling sysconf() to
1025 * tell these two cases apart and return a huge number
1026 * in the latter case to let the caller cap it to a
1027 * value that is not so selfish, but letting the
1028 * fallback OPEN_MAX codepath take care of these cases
1037 return 1; /* see the caller ;-) */
1042 * Do not call this directly as this leaks p->pack_fd on error return;
1043 * call open_packed_git() instead.
1045 static int open_packed_git_1(struct packed_git
*p
)
1048 struct pack_header hdr
;
1049 unsigned char sha1
[20];
1050 unsigned char *idx_sha1
;
1053 if (!p
->index_data
&& open_pack_index(p
))
1054 return error("packfile %s index unavailable", p
->pack_name
);
1056 if (!pack_max_fds
) {
1057 unsigned int max_fds
= get_max_fd_limit();
1059 /* Save 3 for stdin/stdout/stderr, 22 for work */
1061 pack_max_fds
= max_fds
- 25;
1066 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
1069 p
->pack_fd
= git_open_noatime(p
->pack_name
);
1070 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
1074 /* If we created the struct before we had the pack we lack size. */
1075 if (!p
->pack_size
) {
1076 if (!S_ISREG(st
.st_mode
))
1077 return error("packfile %s not a regular file", p
->pack_name
);
1078 p
->pack_size
= st
.st_size
;
1079 } else if (p
->pack_size
!= st
.st_size
)
1080 return error("packfile %s size changed", p
->pack_name
);
1082 /* We leave these file descriptors open with sliding mmap;
1083 * there is no point keeping them open across exec(), though.
1085 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
1087 return error("cannot determine file descriptor flags");
1088 fd_flag
|= FD_CLOEXEC
;
1089 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
1090 return error("cannot set FD_CLOEXEC");
1092 /* Verify we recognize this pack file format. */
1093 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
1094 return error("file %s is far too short to be a packfile", p
->pack_name
);
1095 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
1096 return error("file %s is not a GIT packfile", p
->pack_name
);
1097 if (!pack_version_ok(hdr
.hdr_version
))
1098 return error("packfile %s is version %"PRIu32
" and not"
1099 " supported (try upgrading GIT to a newer version)",
1100 p
->pack_name
, ntohl(hdr
.hdr_version
));
1102 /* Verify the pack matches its index. */
1103 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
1104 return error("packfile %s claims to have %"PRIu32
" objects"
1105 " while index indicates %"PRIu32
" objects",
1106 p
->pack_name
, ntohl(hdr
.hdr_entries
),
1108 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1109 return error("end of packfile %s is unavailable", p
->pack_name
);
1110 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1111 return error("packfile %s signature is unavailable", p
->pack_name
);
1112 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1113 if (hashcmp(sha1
, idx_sha1
))
1114 return error("packfile %s does not match index", p
->pack_name
);
1118 static int open_packed_git(struct packed_git
*p
)
1120 if (!open_packed_git_1(p
))
1126 static int in_window(struct pack_window
*win
, off_t offset
)
1128 /* We must promise at least 20 bytes (one hash) after the
1129 * offset is available from this window, otherwise the offset
1130 * is not actually in this window and a different window (which
1131 * has that one hash excess) must be used. This is to support
1132 * the object header and delta base parsing routines below.
1134 off_t win_off
= win
->offset
;
1135 return win_off
<= offset
1136 && (offset
+ 20) <= (win_off
+ win
->len
);
1139 unsigned char *use_pack(struct packed_git
*p
,
1140 struct pack_window
**w_cursor
,
1142 unsigned long *left
)
1144 struct pack_window
*win
= *w_cursor
;
1146 /* Since packfiles end in a hash of their content and it's
1147 * pointless to ask for an offset into the middle of that
1148 * hash, and the in_window function above wouldn't match
1149 * don't allow an offset too close to the end of the file.
1151 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1152 die("packfile %s cannot be accessed", p
->pack_name
);
1153 if (offset
> (p
->pack_size
- 20))
1154 die("offset beyond end of packfile (truncated pack?)");
1156 die(_("offset before end of packfile (broken .idx?)"));
1158 if (!win
|| !in_window(win
, offset
)) {
1161 for (win
= p
->windows
; win
; win
= win
->next
) {
1162 if (in_window(win
, offset
))
1166 size_t window_align
= packed_git_window_size
/ 2;
1169 if (p
->pack_fd
== -1 && open_packed_git(p
))
1170 die("packfile %s cannot be accessed", p
->pack_name
);
1172 win
= xcalloc(1, sizeof(*win
));
1173 win
->offset
= (offset
/ window_align
) * window_align
;
1174 len
= p
->pack_size
- win
->offset
;
1175 if (len
> packed_git_window_size
)
1176 len
= packed_git_window_size
;
1177 win
->len
= (size_t)len
;
1178 pack_mapped
+= win
->len
;
1179 while (packed_git_limit
< pack_mapped
1180 && unuse_one_window(p
))
1182 win
->base
= xmmap(NULL
, win
->len
,
1183 PROT_READ
, MAP_PRIVATE
,
1184 p
->pack_fd
, win
->offset
);
1185 if (win
->base
== MAP_FAILED
)
1186 die_errno("packfile %s cannot be mapped",
1188 if (!win
->offset
&& win
->len
== p
->pack_size
1189 && !p
->do_not_close
)
1192 pack_open_windows
++;
1193 if (pack_mapped
> peak_pack_mapped
)
1194 peak_pack_mapped
= pack_mapped
;
1195 if (pack_open_windows
> peak_pack_open_windows
)
1196 peak_pack_open_windows
= pack_open_windows
;
1197 win
->next
= p
->windows
;
1201 if (win
!= *w_cursor
) {
1202 win
->last_used
= pack_used_ctr
++;
1206 offset
-= win
->offset
;
1208 *left
= win
->len
- xsize_t(offset
);
1209 return win
->base
+ offset
;
1212 static struct packed_git
*alloc_packed_git(int extra
)
1214 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
1215 memset(p
, 0, sizeof(*p
));
1220 static void try_to_free_pack_memory(size_t size
)
1222 release_pack_memory(size
);
1225 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1227 static int have_set_try_to_free_routine
;
1230 struct packed_git
*p
;
1232 if (!have_set_try_to_free_routine
) {
1233 have_set_try_to_free_routine
= 1;
1234 set_try_to_free_routine(try_to_free_pack_memory
);
1238 * Make sure a corresponding .pack file exists and that
1239 * the index looks sane.
1241 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1245 * ".pack" is long enough to hold any suffix we're adding (and
1246 * the use xsnprintf double-checks that)
1248 alloc
= st_add3(path_len
, strlen(".pack"), 1);
1249 p
= alloc_packed_git(alloc
);
1250 memcpy(p
->pack_name
, path
, path_len
);
1252 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1253 if (!access(p
->pack_name
, F_OK
))
1256 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1257 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1262 /* ok, it looks sane as far as we can check without
1263 * actually mapping the pack file.
1265 p
->pack_size
= st
.st_size
;
1266 p
->pack_local
= local
;
1267 p
->mtime
= st
.st_mtime
;
1268 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1273 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1275 const char *path
= sha1_pack_name(sha1
);
1276 size_t alloc
= st_add(strlen(path
), 1);
1277 struct packed_git
*p
= alloc_packed_git(alloc
);
1279 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
1280 hashcpy(p
->sha1
, sha1
);
1281 if (check_packed_git_idx(idx_path
, p
)) {
1289 void install_packed_git(struct packed_git
*pack
)
1291 if (pack
->pack_fd
!= -1)
1294 pack
->next
= packed_git
;
1298 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
1300 static void report_helper(const struct string_list
*list
,
1301 int seen_bits
, int first
, int last
)
1303 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
1306 for (; first
< last
; first
++)
1307 report_garbage(seen_bits
, list
->items
[first
].string
);
1310 static void report_pack_garbage(struct string_list
*list
)
1312 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1314 if (!report_garbage
)
1317 string_list_sort(list
);
1319 for (i
= 0; i
< list
->nr
; i
++) {
1320 const char *path
= list
->items
[i
].string
;
1321 if (baselen
!= -1 &&
1322 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1323 report_helper(list
, seen_bits
, first
, i
);
1327 if (baselen
== -1) {
1328 const char *dot
= strrchr(path
, '.');
1330 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
1333 baselen
= dot
- path
+ 1;
1336 if (!strcmp(path
+ baselen
, "pack"))
1338 else if (!strcmp(path
+ baselen
, "idx"))
1341 report_helper(list
, seen_bits
, first
, list
->nr
);
1344 static void prepare_packed_git_one(char *objdir
, int local
)
1346 struct strbuf path
= STRBUF_INIT
;
1350 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1352 strbuf_addstr(&path
, objdir
);
1353 strbuf_addstr(&path
, "/pack");
1354 dir
= opendir(path
.buf
);
1356 if (errno
!= ENOENT
)
1357 error_errno("unable to open object pack directory: %s",
1359 strbuf_release(&path
);
1362 strbuf_addch(&path
, '/');
1363 dirnamelen
= path
.len
;
1364 while ((de
= readdir(dir
)) != NULL
) {
1365 struct packed_git
*p
;
1368 if (is_dot_or_dotdot(de
->d_name
))
1371 strbuf_setlen(&path
, dirnamelen
);
1372 strbuf_addstr(&path
, de
->d_name
);
1374 base_len
= path
.len
;
1375 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1376 /* Don't reopen a pack we already have. */
1377 for (p
= packed_git
; p
; p
= p
->next
) {
1379 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1381 !memcmp(p
->pack_name
, path
.buf
, len
))
1386 * See if it really is a valid .idx file with
1387 * corresponding .pack file that we can map.
1389 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1390 install_packed_git(p
);
1393 if (!report_garbage
)
1396 if (ends_with(de
->d_name
, ".idx") ||
1397 ends_with(de
->d_name
, ".pack") ||
1398 ends_with(de
->d_name
, ".bitmap") ||
1399 ends_with(de
->d_name
, ".keep"))
1400 string_list_append(&garbage
, path
.buf
);
1402 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
1405 report_pack_garbage(&garbage
);
1406 string_list_clear(&garbage
, 0);
1407 strbuf_release(&path
);
1410 static int sort_pack(const void *a_
, const void *b_
)
1412 struct packed_git
*a
= *((struct packed_git
**)a_
);
1413 struct packed_git
*b
= *((struct packed_git
**)b_
);
1417 * Local packs tend to contain objects specific to our
1418 * variant of the project than remote ones. In addition,
1419 * remote ones could be on a network mounted filesystem.
1420 * Favor local ones for these reasons.
1422 st
= a
->pack_local
- b
->pack_local
;
1427 * Younger packs tend to contain more recent objects,
1428 * and more recent objects tend to get accessed more
1431 if (a
->mtime
< b
->mtime
)
1433 else if (a
->mtime
== b
->mtime
)
1438 static void rearrange_packed_git(void)
1440 struct packed_git
**ary
, *p
;
1443 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1448 /* prepare an array of packed_git for easier sorting */
1449 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1450 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1453 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1455 /* link them back again */
1456 for (i
= 0; i
< n
- 1; i
++)
1457 ary
[i
]->next
= ary
[i
+ 1];
1458 ary
[n
- 1]->next
= NULL
;
1459 packed_git
= ary
[0];
1464 static int prepare_packed_git_run_once
= 0;
1465 void prepare_packed_git(void)
1467 struct alternate_object_database
*alt
;
1469 if (prepare_packed_git_run_once
)
1471 prepare_packed_git_one(get_object_directory(), 1);
1473 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1475 prepare_packed_git_one(alt
->base
, 0);
1476 alt
->name
[-1] = '/';
1478 rearrange_packed_git();
1479 prepare_packed_git_run_once
= 1;
1482 void reprepare_packed_git(void)
1484 prepare_packed_git_run_once
= 0;
1485 prepare_packed_git();
1488 static void mark_bad_packed_object(struct packed_git
*p
,
1489 const unsigned char *sha1
)
1492 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1493 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
1495 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1496 st_mult(GIT_SHA1_RAWSZ
,
1497 st_add(p
->num_bad_objects
, 1)));
1498 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
1499 p
->num_bad_objects
++;
1502 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1504 struct packed_git
*p
;
1507 for (p
= packed_git
; p
; p
= p
->next
)
1508 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1509 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1515 * With an in-core object data in "map", rehash it to make sure the
1516 * object name actually matches "sha1" to detect object corruption.
1517 * With "map" == NULL, try reading the object named with "sha1" using
1518 * the streaming interface and rehash it to do the same.
1520 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1521 unsigned long size
, const char *type
)
1523 unsigned char real_sha1
[20];
1524 enum object_type obj_type
;
1525 struct git_istream
*st
;
1531 hash_sha1_file(map
, size
, type
, real_sha1
);
1532 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1535 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1539 /* Generate the header */
1540 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1544 git_SHA1_Update(&c
, hdr
, hdrlen
);
1546 char buf
[1024 * 16];
1547 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1555 git_SHA1_Update(&c
, buf
, readlen
);
1557 git_SHA1_Final(real_sha1
, &c
);
1559 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1562 int git_open_noatime(const char *name
)
1564 static int sha1_file_open_flag
= O_NOATIME
;
1570 fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1574 /* Might the failure be due to O_NOATIME? */
1575 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1576 sha1_file_open_flag
= 0;
1584 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1586 struct alternate_object_database
*alt
;
1588 if (!lstat(sha1_file_name(sha1
), st
))
1593 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1594 fill_sha1_path(alt
->name
, sha1
);
1595 if (!lstat(alt
->base
, st
))
1602 static int open_sha1_file(const unsigned char *sha1
)
1605 struct alternate_object_database
*alt
;
1606 int most_interesting_errno
;
1608 fd
= git_open_noatime(sha1_file_name(sha1
));
1611 most_interesting_errno
= errno
;
1614 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1615 fill_sha1_path(alt
->name
, sha1
);
1616 fd
= git_open_noatime(alt
->base
);
1619 if (most_interesting_errno
== ENOENT
)
1620 most_interesting_errno
= errno
;
1622 errno
= most_interesting_errno
;
1626 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1631 fd
= open_sha1_file(sha1
);
1636 if (!fstat(fd
, &st
)) {
1637 *size
= xsize_t(st
.st_size
);
1639 /* mmap() is forbidden on empty files */
1640 error("object file %s is empty", sha1_file_name(sha1
));
1643 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1650 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1651 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1654 unsigned long size
, c
;
1655 unsigned long used
= 0;
1658 *type
= (c
>> 4) & 7;
1662 if (len
<= used
|| bitsizeof(long) <= shift
) {
1663 error("bad object header");
1668 size
+= (c
& 0x7f) << shift
;
1675 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1677 /* Get the data stream */
1678 memset(stream
, 0, sizeof(*stream
));
1679 stream
->next_in
= map
;
1680 stream
->avail_in
= mapsize
;
1681 stream
->next_out
= buffer
;
1682 stream
->avail_out
= bufsiz
;
1684 git_inflate_init(stream
);
1685 return git_inflate(stream
, 0);
1688 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1689 unsigned long mapsize
, void *buffer
,
1690 unsigned long bufsiz
, struct strbuf
*header
)
1694 status
= unpack_sha1_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1697 * Check if entire header is unpacked in the first iteration.
1699 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1703 * buffer[0..bufsiz] was not large enough. Copy the partial
1704 * result out to header, and then append the result of further
1705 * reading the stream.
1707 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1708 stream
->next_out
= buffer
;
1709 stream
->avail_out
= bufsiz
;
1712 status
= git_inflate(stream
, 0);
1713 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1714 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1716 stream
->next_out
= buffer
;
1717 stream
->avail_out
= bufsiz
;
1718 } while (status
!= Z_STREAM_END
);
1722 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1724 int bytes
= strlen(buffer
) + 1;
1725 unsigned char *buf
= xmallocz(size
);
1729 n
= stream
->total_out
- bytes
;
1732 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1734 if (bytes
<= size
) {
1736 * The above condition must be (bytes <= size), not
1737 * (bytes < size). In other words, even though we
1738 * expect no more output and set avail_out to zero,
1739 * the input zlib stream may have bytes that express
1740 * "this concludes the stream", and we *do* want to
1743 * Otherwise we would not be able to test that we
1744 * consumed all the input to reach the expected size;
1745 * we also want to check that zlib tells us that all
1746 * went well with status == Z_STREAM_END at the end.
1748 stream
->next_out
= buf
+ bytes
;
1749 stream
->avail_out
= size
- bytes
;
1750 while (status
== Z_OK
)
1751 status
= git_inflate(stream
, Z_FINISH
);
1753 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1754 git_inflate_end(stream
);
1759 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1760 else if (stream
->avail_in
)
1761 error("garbage at end of loose object '%s'",
1768 * We used to just use "sscanf()", but that's actually way
1769 * too permissive for what we want to check. So do an anal
1770 * object header parse by hand.
1772 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1775 const char *type_buf
= hdr
;
1777 int type
, type_len
= 0;
1780 * The type can be of any size but is followed by
1790 type
= type_from_string_gently(type_buf
, type_len
, 1);
1792 strbuf_add(oi
->typename
, type_buf
, type_len
);
1794 * Set type to 0 if its an unknown object and
1795 * we're obtaining the type using '--allow-unkown-type'
1798 if ((flags
& LOOKUP_UNKNOWN_OBJECT
) && (type
< 0))
1801 die("invalid object type");
1806 * The length must follow immediately, and be in canonical
1807 * decimal format (ie "010" is not valid).
1809 size
= *hdr
++ - '0';
1814 unsigned long c
= *hdr
- '0';
1818 size
= size
* 10 + c
;
1826 * The length must be followed by a zero byte
1828 return *hdr
? -1 : type
;
1831 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1833 struct object_info oi
;
1838 return parse_sha1_header_extended(hdr
, &oi
, LOOKUP_REPLACE_OBJECT
);
1841 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1847 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1848 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1851 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1854 unsigned long get_size_from_delta(struct packed_git
*p
,
1855 struct pack_window
**w_curs
,
1858 const unsigned char *data
;
1859 unsigned char delta_head
[20], *in
;
1863 memset(&stream
, 0, sizeof(stream
));
1864 stream
.next_out
= delta_head
;
1865 stream
.avail_out
= sizeof(delta_head
);
1867 git_inflate_init(&stream
);
1869 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1870 stream
.next_in
= in
;
1871 st
= git_inflate(&stream
, Z_FINISH
);
1872 curpos
+= stream
.next_in
- in
;
1873 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1874 stream
.total_out
< sizeof(delta_head
));
1875 git_inflate_end(&stream
);
1876 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1877 error("delta data unpack-initial failed");
1881 /* Examine the initial part of the delta to figure out
1886 /* ignore base size */
1887 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1889 /* Read the result size */
1890 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1893 static off_t
get_delta_base(struct packed_git
*p
,
1894 struct pack_window
**w_curs
,
1896 enum object_type type
,
1897 off_t delta_obj_offset
)
1899 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1902 /* use_pack() assured us we have [base_info, base_info + 20)
1903 * as a range that we can look at without walking off the
1904 * end of the mapped window. Its actually the hash size
1905 * that is assured. An OFS_DELTA longer than the hash size
1906 * is stupid, as then a REF_DELTA would be smaller to store.
1908 if (type
== OBJ_OFS_DELTA
) {
1910 unsigned char c
= base_info
[used
++];
1911 base_offset
= c
& 127;
1914 if (!base_offset
|| MSB(base_offset
, 7))
1915 return 0; /* overflow */
1916 c
= base_info
[used
++];
1917 base_offset
= (base_offset
<< 7) + (c
& 127);
1919 base_offset
= delta_obj_offset
- base_offset
;
1920 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1921 return 0; /* out of bound */
1923 } else if (type
== OBJ_REF_DELTA
) {
1924 /* The base entry _must_ be in the same pack */
1925 base_offset
= find_pack_entry_one(base_info
, p
);
1928 die("I am totally screwed");
1933 * Like get_delta_base above, but we return the sha1 instead of the pack
1934 * offset. This means it is cheaper for REF deltas (we do not have to do
1935 * the final object lookup), but more expensive for OFS deltas (we
1936 * have to load the revidx to convert the offset back into a sha1).
1938 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1939 struct pack_window
**w_curs
,
1941 enum object_type type
,
1942 off_t delta_obj_offset
)
1944 if (type
== OBJ_REF_DELTA
) {
1945 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1947 } else if (type
== OBJ_OFS_DELTA
) {
1948 struct revindex_entry
*revidx
;
1949 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1950 type
, delta_obj_offset
);
1955 revidx
= find_pack_revindex(p
, base_offset
);
1959 return nth_packed_object_sha1(p
, revidx
->nr
);
1964 int unpack_object_header(struct packed_git
*p
,
1965 struct pack_window
**w_curs
,
1967 unsigned long *sizep
)
1969 unsigned char *base
;
1972 enum object_type type
;
1974 /* use_pack() assures us we have [base, base + 20) available
1975 * as a range that we can look at. (Its actually the hash
1976 * size that is assured.) With our object header encoding
1977 * the maximum deflated object size is 2^137, which is just
1978 * insane, so we know won't exceed what we have been given.
1980 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1981 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1990 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1993 struct revindex_entry
*revidx
;
1994 const unsigned char *sha1
;
1995 revidx
= find_pack_revindex(p
, obj_offset
);
1998 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1999 mark_bad_packed_object(p
, sha1
);
2000 type
= sha1_object_info(sha1
, NULL
);
2001 if (type
<= OBJ_NONE
)
2006 #define POI_STACK_PREALLOC 64
2008 static enum object_type
packed_to_object_type(struct packed_git
*p
,
2010 enum object_type type
,
2011 struct pack_window
**w_curs
,
2014 off_t small_poi_stack
[POI_STACK_PREALLOC
];
2015 off_t
*poi_stack
= small_poi_stack
;
2016 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
2018 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2021 /* Push the object we're going to leave behind */
2022 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
2023 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
2024 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
2025 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
2027 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
2029 poi_stack
[poi_stack_nr
++] = obj_offset
;
2030 /* If parsing the base offset fails, just unwind */
2031 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
2034 curpos
= obj_offset
= base_offset
;
2035 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
2036 if (type
<= OBJ_NONE
) {
2037 /* If getting the base itself fails, we first
2038 * retry the base, otherwise unwind */
2039 type
= retry_bad_packed_offset(p
, base_offset
);
2040 if (type
> OBJ_NONE
)
2054 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2055 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2060 if (poi_stack
!= small_poi_stack
)
2065 while (poi_stack_nr
) {
2066 obj_offset
= poi_stack
[--poi_stack_nr
];
2067 type
= retry_bad_packed_offset(p
, obj_offset
);
2068 if (type
> OBJ_NONE
)
2075 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
2076 struct object_info
*oi
)
2078 struct pack_window
*w_curs
= NULL
;
2080 off_t curpos
= obj_offset
;
2081 enum object_type type
;
2084 * We always get the representation type, but only convert it to
2085 * a "real" type later if the caller is interested.
2087 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2090 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2091 off_t tmp_pos
= curpos
;
2092 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
2098 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2099 if (*oi
->sizep
== 0) {
2108 if (oi
->disk_sizep
) {
2109 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2110 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2114 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
2115 if (*oi
->typep
< 0) {
2121 if (oi
->delta_base_sha1
) {
2122 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2123 const unsigned char *base
;
2125 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2132 hashcpy(oi
->delta_base_sha1
, base
);
2134 hashclr(oi
->delta_base_sha1
);
2138 unuse_pack(&w_curs
);
2142 static void *unpack_compressed_entry(struct packed_git
*p
,
2143 struct pack_window
**w_curs
,
2149 unsigned char *buffer
, *in
;
2151 buffer
= xmallocz_gently(size
);
2154 memset(&stream
, 0, sizeof(stream
));
2155 stream
.next_out
= buffer
;
2156 stream
.avail_out
= size
+ 1;
2158 git_inflate_init(&stream
);
2160 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2161 stream
.next_in
= in
;
2162 st
= git_inflate(&stream
, Z_FINISH
);
2163 if (!stream
.avail_out
)
2164 break; /* the payload is larger than it should be */
2165 curpos
+= stream
.next_in
- in
;
2166 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2167 git_inflate_end(&stream
);
2168 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2176 #define MAX_DELTA_CACHE (256)
2178 static size_t delta_base_cached
;
2180 static struct delta_base_cache_lru_list
{
2181 struct delta_base_cache_lru_list
*prev
;
2182 struct delta_base_cache_lru_list
*next
;
2183 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
2185 static struct delta_base_cache_entry
{
2186 struct delta_base_cache_lru_list lru
;
2188 struct packed_git
*p
;
2191 enum object_type type
;
2192 } delta_base_cache
[MAX_DELTA_CACHE
];
2194 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2198 hash
= (unsigned long)(intptr_t)p
+ (unsigned long)base_offset
;
2199 hash
+= (hash
>> 8) + (hash
>> 16);
2200 return hash
% MAX_DELTA_CACHE
;
2203 static struct delta_base_cache_entry
*
2204 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2206 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2207 return delta_base_cache
+ hash
;
2210 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
2211 struct packed_git
*p
, off_t base_offset
)
2213 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
2216 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2218 struct delta_base_cache_entry
*ent
;
2219 ent
= get_delta_base_cache_entry(p
, base_offset
);
2220 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
2223 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2226 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2227 ent
->lru
.prev
->next
= ent
->lru
.next
;
2228 delta_base_cached
-= ent
->size
;
2231 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2232 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2234 struct delta_base_cache_entry
*ent
;
2237 ent
= get_delta_base_cache_entry(p
, base_offset
);
2239 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2240 return unpack_entry(p
, base_offset
, type
, base_size
);
2245 clear_delta_base_cache_entry(ent
);
2247 ret
= xmemdupz(ent
->data
, ent
->size
);
2249 *base_size
= ent
->size
;
2253 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2258 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2259 ent
->lru
.prev
->next
= ent
->lru
.next
;
2260 delta_base_cached
-= ent
->size
;
2264 void clear_delta_base_cache(void)
2267 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2268 release_delta_base_cache(&delta_base_cache
[p
]);
2271 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2272 void *base
, unsigned long base_size
, enum object_type type
)
2274 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2275 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2276 struct delta_base_cache_lru_list
*lru
;
2278 release_delta_base_cache(ent
);
2279 delta_base_cached
+= base_size
;
2281 for (lru
= delta_base_cache_lru
.next
;
2282 delta_base_cached
> delta_base_cache_limit
2283 && lru
!= &delta_base_cache_lru
;
2285 struct delta_base_cache_entry
*f
= (void *)lru
;
2286 if (f
->type
== OBJ_BLOB
)
2287 release_delta_base_cache(f
);
2289 for (lru
= delta_base_cache_lru
.next
;
2290 delta_base_cached
> delta_base_cache_limit
2291 && lru
!= &delta_base_cache_lru
;
2293 struct delta_base_cache_entry
*f
= (void *)lru
;
2294 release_delta_base_cache(f
);
2298 ent
->base_offset
= base_offset
;
2301 ent
->size
= base_size
;
2302 ent
->lru
.next
= &delta_base_cache_lru
;
2303 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2304 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2305 delta_base_cache_lru
.prev
= &ent
->lru
;
2308 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2309 unsigned long *size
);
2311 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2313 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2314 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2315 p
->pack_name
, (uintmax_t)obj_offset
);
2318 int do_check_packed_object_crc
;
2320 #define UNPACK_ENTRY_STACK_PREALLOC 64
2321 struct unpack_entry_stack_ent
{
2327 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2328 enum object_type
*final_type
, unsigned long *final_size
)
2330 struct pack_window
*w_curs
= NULL
;
2331 off_t curpos
= obj_offset
;
2334 enum object_type type
;
2335 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2336 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2337 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2338 int base_from_cache
= 0;
2340 write_pack_access_log(p
, obj_offset
);
2342 /* PHASE 1: drill down to the innermost base object */
2346 struct delta_base_cache_entry
*ent
;
2348 ent
= get_delta_base_cache_entry(p
, curpos
);
2349 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2353 clear_delta_base_cache_entry(ent
);
2354 base_from_cache
= 1;
2358 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2359 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2360 off_t len
= revidx
[1].offset
- obj_offset
;
2361 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2362 const unsigned char *sha1
=
2363 nth_packed_object_sha1(p
, revidx
->nr
);
2364 error("bad packed object CRC for %s",
2366 mark_bad_packed_object(p
, sha1
);
2367 unuse_pack(&w_curs
);
2372 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2373 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2376 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2378 error("failed to validate delta base reference "
2379 "at offset %"PRIuMAX
" from %s",
2380 (uintmax_t)curpos
, p
->pack_name
);
2381 /* bail to phase 2, in hopes of recovery */
2386 /* push object, proceed to base */
2387 if (delta_stack_nr
>= delta_stack_alloc
2388 && delta_stack
== small_delta_stack
) {
2389 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2390 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
2391 memcpy(delta_stack
, small_delta_stack
,
2392 sizeof(*delta_stack
)*delta_stack_nr
);
2394 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2396 i
= delta_stack_nr
++;
2397 delta_stack
[i
].obj_offset
= obj_offset
;
2398 delta_stack
[i
].curpos
= curpos
;
2399 delta_stack
[i
].size
= size
;
2401 curpos
= obj_offset
= base_offset
;
2404 /* PHASE 2: handle the base */
2409 die("BUG in unpack_entry: left loop at a valid delta");
2415 if (!base_from_cache
)
2416 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2420 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2421 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2424 /* PHASE 3: apply deltas in order */
2427 * 'data' holds the base data, or NULL if there was corruption
2429 while (delta_stack_nr
) {
2432 unsigned long delta_size
, base_size
= size
;
2438 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2442 * We're probably in deep shit, but let's try to fetch
2443 * the required base anyway from another pack or loose.
2444 * This is costly but should happen only in the presence
2445 * of a corrupted pack, and is better than failing outright.
2447 struct revindex_entry
*revidx
;
2448 const unsigned char *base_sha1
;
2449 revidx
= find_pack_revindex(p
, obj_offset
);
2451 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2452 error("failed to read delta base object %s"
2453 " at offset %"PRIuMAX
" from %s",
2454 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2456 mark_bad_packed_object(p
, base_sha1
);
2457 base
= read_object(base_sha1
, &type
, &base_size
);
2461 i
= --delta_stack_nr
;
2462 obj_offset
= delta_stack
[i
].obj_offset
;
2463 curpos
= delta_stack
[i
].curpos
;
2464 delta_size
= delta_stack
[i
].size
;
2469 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2472 error("failed to unpack compressed delta "
2473 "at offset %"PRIuMAX
" from %s",
2474 (uintmax_t)curpos
, p
->pack_name
);
2479 data
= patch_delta(base
, base_size
,
2480 delta_data
, delta_size
,
2484 * We could not apply the delta; warn the user, but keep going.
2485 * Our failure will be noticed either in the next iteration of
2486 * the loop, or if this is the final delta, in the caller when
2487 * we return NULL. Those code paths will take care of making
2488 * a more explicit warning and retrying with another copy of
2492 error("failed to apply delta");
2500 unuse_pack(&w_curs
);
2502 if (delta_stack
!= small_delta_stack
)
2508 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2511 const unsigned char *index
= p
->index_data
;
2513 if (open_pack_index(p
))
2515 index
= p
->index_data
;
2517 if (n
>= p
->num_objects
)
2520 if (p
->index_version
== 1) {
2521 return index
+ 24 * n
+ 4;
2524 return index
+ 20 * n
;
2528 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2530 const unsigned char *ptr
= vptr
;
2531 const unsigned char *start
= p
->index_data
;
2532 const unsigned char *end
= start
+ p
->index_size
;
2534 die(_("offset before start of pack index for %s (corrupt index?)"),
2536 /* No need to check for underflow; .idx files must be at least 8 bytes */
2538 die(_("offset beyond end of pack index for %s (truncated index?)"),
2542 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2544 const unsigned char *index
= p
->index_data
;
2546 if (p
->index_version
== 1) {
2547 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2550 index
+= 8 + p
->num_objects
* (20 + 4);
2551 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2552 if (!(off
& 0x80000000))
2554 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2555 check_pack_index_ptr(p
, index
);
2556 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2557 ntohl(*((uint32_t *)(index
+ 4)));
2561 off_t
find_pack_entry_one(const unsigned char *sha1
,
2562 struct packed_git
*p
)
2564 const uint32_t *level1_ofs
= p
->index_data
;
2565 const unsigned char *index
= p
->index_data
;
2566 unsigned hi
, lo
, stride
;
2567 static int use_lookup
= -1;
2568 static int debug_lookup
= -1;
2570 if (debug_lookup
< 0)
2571 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2574 if (open_pack_index(p
))
2576 level1_ofs
= p
->index_data
;
2577 index
= p
->index_data
;
2579 if (p
->index_version
> 1) {
2584 hi
= ntohl(level1_ofs
[*sha1
]);
2585 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2586 if (p
->index_version
> 1) {
2594 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2595 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2598 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2600 int pos
= sha1_entry_pos(index
, stride
, 0,
2601 lo
, hi
, p
->num_objects
, sha1
);
2604 return nth_packed_object_offset(p
, pos
);
2608 unsigned mi
= (lo
+ hi
) / 2;
2609 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2612 printf("lo %u hi %u rg %u mi %u\n",
2613 lo
, hi
, hi
- lo
, mi
);
2615 return nth_packed_object_offset(p
, mi
);
2624 int is_pack_valid(struct packed_git
*p
)
2626 /* An already open pack is known to be valid. */
2627 if (p
->pack_fd
!= -1)
2630 /* If the pack has one window completely covering the
2631 * file size, the pack is known to be valid even if
2632 * the descriptor is not currently open.
2635 struct pack_window
*w
= p
->windows
;
2637 if (!w
->offset
&& w
->len
== p
->pack_size
)
2641 /* Force the pack to open to prove its valid. */
2642 return !open_packed_git(p
);
2645 static int fill_pack_entry(const unsigned char *sha1
,
2646 struct pack_entry
*e
,
2647 struct packed_git
*p
)
2651 if (p
->num_bad_objects
) {
2653 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2654 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2658 offset
= find_pack_entry_one(sha1
, p
);
2663 * We are about to tell the caller where they can locate the
2664 * requested object. We better make sure the packfile is
2665 * still here and can be accessed before supplying that
2666 * answer, as it may have been deleted since the index was
2669 if (!is_pack_valid(p
))
2673 hashcpy(e
->sha1
, sha1
);
2678 * Iff a pack file contains the object named by sha1, return true and
2679 * store its location to e.
2681 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2683 struct packed_git
*p
;
2685 prepare_packed_git();
2689 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2692 for (p
= packed_git
; p
; p
= p
->next
) {
2693 if (p
== last_found_pack
)
2694 continue; /* we already checked this one */
2696 if (fill_pack_entry(sha1
, e
, p
)) {
2697 last_found_pack
= p
;
2704 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2705 struct packed_git
*packs
)
2707 struct packed_git
*p
;
2709 for (p
= packs
; p
; p
= p
->next
) {
2710 if (find_pack_entry_one(sha1
, p
))
2717 static int sha1_loose_object_info(const unsigned char *sha1
,
2718 struct object_info
*oi
,
2722 unsigned long mapsize
;
2726 struct strbuf hdrbuf
= STRBUF_INIT
;
2728 if (oi
->delta_base_sha1
)
2729 hashclr(oi
->delta_base_sha1
);
2732 * If we don't care about type or size, then we don't
2733 * need to look inside the object at all. Note that we
2734 * do not optimize out the stat call, even if the
2735 * caller doesn't care about the disk-size, since our
2736 * return value implicitly indicates whether the
2737 * object even exists.
2739 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
) {
2741 if (stat_sha1_file(sha1
, &st
) < 0)
2744 *oi
->disk_sizep
= st
.st_size
;
2748 map
= map_sha1_file(sha1
, &mapsize
);
2752 *oi
->disk_sizep
= mapsize
;
2753 if ((flags
& LOOKUP_UNKNOWN_OBJECT
)) {
2754 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2755 status
= error("unable to unpack %s header with --allow-unknown-type",
2757 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2758 status
= error("unable to unpack %s header",
2762 else if (hdrbuf
.len
) {
2763 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2764 status
= error("unable to parse %s header with --allow-unknown-type",
2766 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2767 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2768 git_inflate_end(&stream
);
2769 munmap(map
, mapsize
);
2770 if (status
&& oi
->typep
)
2771 *oi
->typep
= status
;
2772 strbuf_release(&hdrbuf
);
2776 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2778 struct cached_object
*co
;
2779 struct pack_entry e
;
2781 enum object_type real_type
;
2782 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2784 co
= find_cached_object(real
);
2787 *(oi
->typep
) = co
->type
;
2789 *(oi
->sizep
) = co
->size
;
2791 *(oi
->disk_sizep
) = 0;
2792 if (oi
->delta_base_sha1
)
2793 hashclr(oi
->delta_base_sha1
);
2795 strbuf_addstr(oi
->typename
, typename(co
->type
));
2796 oi
->whence
= OI_CACHED
;
2800 if (!find_pack_entry(real
, &e
)) {
2801 /* Most likely it's a loose object. */
2802 if (!sha1_loose_object_info(real
, oi
, flags
)) {
2803 oi
->whence
= OI_LOOSE
;
2807 /* Not a loose object; someone else may have just packed it. */
2808 reprepare_packed_git();
2809 if (!find_pack_entry(real
, &e
))
2814 * packed_object_info() does not follow the delta chain to
2815 * find out the real type, unless it is given oi->typep.
2817 if (oi
->typename
&& !oi
->typep
)
2818 oi
->typep
= &real_type
;
2820 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2822 mark_bad_packed_object(e
.p
, real
);
2823 if (oi
->typep
== &real_type
)
2825 return sha1_object_info_extended(real
, oi
, 0);
2826 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2827 oi
->whence
= OI_DBCACHED
;
2829 oi
->whence
= OI_PACKED
;
2830 oi
->u
.packed
.offset
= e
.offset
;
2831 oi
->u
.packed
.pack
= e
.p
;
2832 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2833 rtype
== OBJ_OFS_DELTA
);
2836 strbuf_addstr(oi
->typename
, typename(*oi
->typep
));
2837 if (oi
->typep
== &real_type
)
2843 /* returns enum object_type or negative */
2844 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2846 enum object_type type
;
2847 struct object_info oi
= {NULL
};
2851 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2856 static void *read_packed_sha1(const unsigned char *sha1
,
2857 enum object_type
*type
, unsigned long *size
)
2859 struct pack_entry e
;
2862 if (!find_pack_entry(sha1
, &e
))
2864 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2867 * We're probably in deep shit, but let's try to fetch
2868 * the required object anyway from another pack or loose.
2869 * This should happen only in the presence of a corrupted
2870 * pack, and is better than failing outright.
2872 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2873 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2874 mark_bad_packed_object(e
.p
, sha1
);
2875 data
= read_object(sha1
, type
, size
);
2880 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2881 unsigned char *sha1
)
2883 struct cached_object
*co
;
2885 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2886 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2888 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2889 co
= &cached_objects
[cached_object_nr
++];
2892 co
->buf
= xmalloc(len
);
2893 memcpy(co
->buf
, buf
, len
);
2894 hashcpy(co
->sha1
, sha1
);
2898 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2899 unsigned long *size
)
2901 unsigned long mapsize
;
2903 struct cached_object
*co
;
2905 co
= find_cached_object(sha1
);
2909 return xmemdupz(co
->buf
, co
->size
);
2912 buf
= read_packed_sha1(sha1
, type
, size
);
2915 map
= map_sha1_file(sha1
, &mapsize
);
2917 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2918 munmap(map
, mapsize
);
2921 reprepare_packed_git();
2922 return read_packed_sha1(sha1
, type
, size
);
2926 * This function dies on corrupt objects; the callers who want to
2927 * deal with them should arrange to call read_object() and give error
2928 * messages themselves.
2930 void *read_sha1_file_extended(const unsigned char *sha1
,
2931 enum object_type
*type
,
2932 unsigned long *size
,
2936 const struct packed_git
*p
;
2937 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2940 data
= read_object(repl
, type
, size
);
2944 if (errno
&& errno
!= ENOENT
)
2945 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2947 /* die if we replaced an object with one that does not exist */
2949 die("replacement %s not found for %s",
2950 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2952 if (has_loose_object(repl
)) {
2953 const char *path
= sha1_file_name(sha1
);
2955 die("loose object %s (stored in %s) is corrupt",
2956 sha1_to_hex(repl
), path
);
2959 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2960 die("packed object %s (stored in %s) is corrupt",
2961 sha1_to_hex(repl
), p
->pack_name
);
2966 void *read_object_with_reference(const unsigned char *sha1
,
2967 const char *required_type_name
,
2968 unsigned long *size
,
2969 unsigned char *actual_sha1_return
)
2971 enum object_type type
, required_type
;
2973 unsigned long isize
;
2974 unsigned char actual_sha1
[20];
2976 required_type
= type_from_string(required_type_name
);
2977 hashcpy(actual_sha1
, sha1
);
2979 int ref_length
= -1;
2980 const char *ref_type
= NULL
;
2982 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2985 if (type
== required_type
) {
2987 if (actual_sha1_return
)
2988 hashcpy(actual_sha1_return
, actual_sha1
);
2991 /* Handle references */
2992 else if (type
== OBJ_COMMIT
)
2994 else if (type
== OBJ_TAG
)
2995 ref_type
= "object ";
3000 ref_length
= strlen(ref_type
);
3002 if (ref_length
+ 40 > isize
||
3003 memcmp(buffer
, ref_type
, ref_length
) ||
3004 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
3009 /* Now we have the ID of the referred-to object in
3010 * actual_sha1. Check again. */
3014 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
3015 const char *type
, unsigned char *sha1
,
3016 char *hdr
, int *hdrlen
)
3020 /* Generate the header */
3021 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
3025 git_SHA1_Update(&c
, hdr
, *hdrlen
);
3026 git_SHA1_Update(&c
, buf
, len
);
3027 git_SHA1_Final(sha1
, &c
);
3031 * Move the just written object into its final resting place.
3033 int finalize_object_file(const char *tmpfile
, const char *filename
)
3037 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
3039 else if (link(tmpfile
, filename
))
3043 * Coda hack - coda doesn't like cross-directory links,
3044 * so we fall back to a rename, which will mean that it
3045 * won't be able to check collisions, but that's not a
3048 * The same holds for FAT formatted media.
3050 * When this succeeds, we just return. We have nothing
3053 if (ret
&& ret
!= EEXIST
) {
3055 if (!rename(tmpfile
, filename
))
3059 unlink_or_warn(tmpfile
);
3061 if (ret
!= EEXIST
) {
3062 return error_errno("unable to write sha1 filename %s", filename
);
3064 /* FIXME!!! Collision check here ? */
3068 if (adjust_shared_perm(filename
))
3069 return error("unable to set permission to '%s'", filename
);
3073 static int write_buffer(int fd
, const void *buf
, size_t len
)
3075 if (write_in_full(fd
, buf
, len
) < 0)
3076 return error_errno("file write error");
3080 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
3081 unsigned char *sha1
)
3084 int hdrlen
= sizeof(hdr
);
3085 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3089 /* Finalize a file on disk, and close it. */
3090 static void close_sha1_file(int fd
)
3092 if (fsync_object_files
)
3093 fsync_or_die(fd
, "sha1 file");
3095 die_errno("error when closing sha1 file");
3098 /* Size of directory component, including the ending '/' */
3099 static inline int directory_size(const char *filename
)
3101 const char *s
= strrchr(filename
, '/');
3104 return s
- filename
+ 1;
3108 * This creates a temporary file in the same directory as the final
3111 * We want to avoid cross-directory filename renames, because those
3112 * can have problems on various filesystems (FAT, NFS, Coda).
3114 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
3116 int fd
, dirlen
= directory_size(filename
);
3119 strbuf_add(tmp
, filename
, dirlen
);
3120 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
3121 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3122 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3124 * Make sure the directory exists; note that the contents
3125 * of the buffer are undefined after mkstemp returns an
3126 * error, so we have to rewrite the whole buffer from
3130 strbuf_add(tmp
, filename
, dirlen
- 1);
3131 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
3133 if (adjust_shared_perm(tmp
->buf
))
3137 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
3138 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3143 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3144 const void *buf
, unsigned long len
, time_t mtime
)
3147 unsigned char compressed
[4096];
3150 unsigned char parano_sha1
[20];
3151 static struct strbuf tmp_file
= STRBUF_INIT
;
3152 const char *filename
= sha1_file_name(sha1
);
3154 fd
= create_tmpfile(&tmp_file
, filename
);
3156 if (errno
== EACCES
)
3157 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3159 return error_errno("unable to create temporary file");
3163 git_deflate_init(&stream
, zlib_compression_level
);
3164 stream
.next_out
= compressed
;
3165 stream
.avail_out
= sizeof(compressed
);
3168 /* First header.. */
3169 stream
.next_in
= (unsigned char *)hdr
;
3170 stream
.avail_in
= hdrlen
;
3171 while (git_deflate(&stream
, 0) == Z_OK
)
3173 git_SHA1_Update(&c
, hdr
, hdrlen
);
3175 /* Then the data itself.. */
3176 stream
.next_in
= (void *)buf
;
3177 stream
.avail_in
= len
;
3179 unsigned char *in0
= stream
.next_in
;
3180 ret
= git_deflate(&stream
, Z_FINISH
);
3181 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3182 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3183 die("unable to write sha1 file");
3184 stream
.next_out
= compressed
;
3185 stream
.avail_out
= sizeof(compressed
);
3186 } while (ret
== Z_OK
);
3188 if (ret
!= Z_STREAM_END
)
3189 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3190 ret
= git_deflate_end_gently(&stream
);
3192 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3193 git_SHA1_Final(parano_sha1
, &c
);
3194 if (hashcmp(sha1
, parano_sha1
) != 0)
3195 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3197 close_sha1_file(fd
);
3202 utb
.modtime
= mtime
;
3203 if (utime(tmp_file
.buf
, &utb
) < 0)
3204 warning_errno("failed utime() on %s", tmp_file
.buf
);
3207 return finalize_object_file(tmp_file
.buf
, filename
);
3210 static int freshen_loose_object(const unsigned char *sha1
)
3212 return check_and_freshen(sha1
, 1);
3215 static int freshen_packed_object(const unsigned char *sha1
)
3217 struct pack_entry e
;
3218 if (!find_pack_entry(sha1
, &e
))
3222 if (!freshen_file(e
.p
->pack_name
))
3228 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3231 int hdrlen
= sizeof(hdr
);
3233 /* Normally if we have it in the pack then we do not bother writing
3234 * it out into .git/objects/??/?{38} file.
3236 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3237 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3239 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3242 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3243 unsigned char *sha1
, unsigned flags
)
3246 int hdrlen
, status
= 0;
3248 /* type string, SP, %lu of the length plus NUL must fit this */
3249 hdrlen
= strlen(type
) + 32;
3250 header
= xmalloc(hdrlen
);
3251 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3253 if (!(flags
& HASH_WRITE_OBJECT
))
3255 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3257 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3264 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3268 enum object_type type
;
3273 if (has_loose_object(sha1
))
3275 buf
= read_packed_sha1(sha1
, &type
, &len
);
3277 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3278 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3279 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3285 int has_pack_index(const unsigned char *sha1
)
3288 if (stat(sha1_pack_index_name(sha1
), &st
))
3293 int has_sha1_pack(const unsigned char *sha1
)
3295 struct pack_entry e
;
3296 return find_pack_entry(sha1
, &e
);
3299 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3301 struct pack_entry e
;
3303 if (find_pack_entry(sha1
, &e
))
3305 if (has_loose_object(sha1
))
3307 if (flags
& HAS_SHA1_QUICK
)
3309 reprepare_packed_git();
3310 return find_pack_entry(sha1
, &e
);
3313 int has_object_file(const struct object_id
*oid
)
3315 return has_sha1_file(oid
->hash
);
3318 static void check_tree(const void *buf
, size_t size
)
3320 struct tree_desc desc
;
3321 struct name_entry entry
;
3323 init_tree_desc(&desc
, buf
, size
);
3324 while (tree_entry(&desc
, &entry
))
3326 * tree_entry() will die() on malformed entries */
3330 static void check_commit(const void *buf
, size_t size
)
3333 memset(&c
, 0, sizeof(c
));
3334 if (parse_commit_buffer(&c
, buf
, size
))
3335 die("corrupt commit");
3338 static void check_tag(const void *buf
, size_t size
)
3341 memset(&t
, 0, sizeof(t
));
3342 if (parse_tag_buffer(&t
, buf
, size
))
3346 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3347 enum object_type type
,
3348 const char *path
, unsigned flags
)
3350 int ret
, re_allocated
= 0;
3351 int write_object
= flags
& HASH_WRITE_OBJECT
;
3357 * Convert blobs to git internal format
3359 if ((type
== OBJ_BLOB
) && path
) {
3360 struct strbuf nbuf
= STRBUF_INIT
;
3361 if (convert_to_git(path
, buf
, size
, &nbuf
,
3362 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3363 buf
= strbuf_detach(&nbuf
, &size
);
3367 if (flags
& HASH_FORMAT_CHECK
) {
3368 if (type
== OBJ_TREE
)
3369 check_tree(buf
, size
);
3370 if (type
== OBJ_COMMIT
)
3371 check_commit(buf
, size
);
3372 if (type
== OBJ_TAG
)
3373 check_tag(buf
, size
);
3377 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3379 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3385 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3386 const char *path
, unsigned flags
)
3389 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3390 struct strbuf sbuf
= STRBUF_INIT
;
3393 assert(would_convert_to_git_filter_fd(path
));
3395 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3396 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3399 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3402 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3404 strbuf_release(&sbuf
);
3408 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3409 const char *path
, unsigned flags
)
3411 struct strbuf sbuf
= STRBUF_INIT
;
3414 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3415 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3418 strbuf_release(&sbuf
);
3422 #define SMALL_FILE_SIZE (32*1024)
3424 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3425 enum object_type type
, const char *path
,
3431 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3432 } else if (size
<= SMALL_FILE_SIZE
) {
3433 char *buf
= xmalloc(size
);
3434 if (size
== read_in_full(fd
, buf
, size
))
3435 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3437 ret
= error_errno("short read");
3440 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3441 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3448 * This creates one packfile per large blob unless bulk-checkin
3449 * machinery is "plugged".
3451 * This also bypasses the usual "convert-to-git" dance, and that is on
3452 * purpose. We could write a streaming version of the converting
3453 * functions and insert that before feeding the data to fast-import
3454 * (or equivalent in-core API described above). However, that is
3455 * somewhat complicated, as we do not know the size of the filter
3456 * result, which we need to know beforehand when writing a git object.
3457 * Since the primary motivation for trying to stream from the working
3458 * tree file and to avoid mmaping it in core is to deal with large
3459 * binary blobs, they generally do not want to get any conversion, and
3460 * callers should avoid this code path when filters are requested.
3462 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3463 enum object_type type
, const char *path
,
3466 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3469 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3470 enum object_type type
, const char *path
, unsigned flags
)
3475 * Call xsize_t() only when needed to avoid potentially unnecessary
3476 * die() for large files.
3478 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3479 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3480 else if (!S_ISREG(st
->st_mode
))
3481 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3482 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3483 (path
&& would_convert_to_git(path
)))
3484 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3487 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3493 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3496 struct strbuf sb
= STRBUF_INIT
;
3498 switch (st
->st_mode
& S_IFMT
) {
3500 fd
= open(path
, O_RDONLY
);
3502 return error_errno("open(\"%s\")", path
);
3503 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3504 return error("%s: failed to insert into database",
3508 if (strbuf_readlink(&sb
, path
, st
->st_size
))
3509 return error_errno("readlink(\"%s\")", path
);
3510 if (!(flags
& HASH_WRITE_OBJECT
))
3511 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3512 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3513 return error("%s: failed to insert into database",
3515 strbuf_release(&sb
);
3518 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3520 return error("%s: unsupported file type", path
);
3525 int read_pack_header(int fd
, struct pack_header
*header
)
3527 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3528 /* "eof before pack header was fully read" */
3529 return PH_ERROR_EOF
;
3531 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3532 /* "protocol error (pack signature mismatch detected)" */
3533 return PH_ERROR_PACK_SIGNATURE
;
3534 if (!pack_version_ok(header
->hdr_version
))
3535 /* "protocol error (pack version unsupported)" */
3536 return PH_ERROR_PROTOCOL
;
3540 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3542 enum object_type type
= sha1_object_info(sha1
, NULL
);
3544 die("%s is not a valid object", sha1_to_hex(sha1
));
3546 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3550 static int for_each_file_in_obj_subdir(int subdir_nr
,
3551 struct strbuf
*path
,
3552 each_loose_object_fn obj_cb
,
3553 each_loose_cruft_fn cruft_cb
,
3554 each_loose_subdir_fn subdir_cb
,
3557 size_t baselen
= path
->len
;
3558 DIR *dir
= opendir(path
->buf
);
3563 if (errno
== ENOENT
)
3565 return error_errno("unable to open %s", path
->buf
);
3568 while ((de
= readdir(dir
))) {
3569 if (is_dot_or_dotdot(de
->d_name
))
3572 strbuf_setlen(path
, baselen
);
3573 strbuf_addf(path
, "/%s", de
->d_name
);
3575 if (strlen(de
->d_name
) == 38) {
3577 unsigned char sha1
[20];
3579 snprintf(hex
, sizeof(hex
), "%02x%s",
3580 subdir_nr
, de
->d_name
);
3581 if (!get_sha1_hex(hex
, sha1
)) {
3583 r
= obj_cb(sha1
, path
->buf
, data
);
3592 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3599 strbuf_setlen(path
, baselen
);
3600 if (!r
&& subdir_cb
)
3601 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3606 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3607 each_loose_object_fn obj_cb
,
3608 each_loose_cruft_fn cruft_cb
,
3609 each_loose_subdir_fn subdir_cb
,
3612 size_t baselen
= path
->len
;
3616 for (i
= 0; i
< 256; i
++) {
3617 strbuf_addf(path
, "/%02x", i
);
3618 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3620 strbuf_setlen(path
, baselen
);
3628 int for_each_loose_file_in_objdir(const char *path
,
3629 each_loose_object_fn obj_cb
,
3630 each_loose_cruft_fn cruft_cb
,
3631 each_loose_subdir_fn subdir_cb
,
3634 struct strbuf buf
= STRBUF_INIT
;
3637 strbuf_addstr(&buf
, path
);
3638 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3640 strbuf_release(&buf
);
3645 struct loose_alt_odb_data
{
3646 each_loose_object_fn
*cb
;
3650 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3653 struct loose_alt_odb_data
*data
= vdata
;
3654 struct strbuf buf
= STRBUF_INIT
;
3657 /* copy base not including trailing '/' */
3658 strbuf_add(&buf
, alt
->base
, alt
->name
- alt
->base
- 1);
3659 r
= for_each_loose_file_in_objdir_buf(&buf
,
3660 data
->cb
, NULL
, NULL
,
3662 strbuf_release(&buf
);
3666 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3668 struct loose_alt_odb_data alt
;
3671 r
= for_each_loose_file_in_objdir(get_object_directory(),
3672 cb
, NULL
, NULL
, data
);
3676 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3681 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3684 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3689 for (i
= 0; i
< p
->num_objects
; i
++) {
3690 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3693 return error("unable to get sha1 of object %u in %s",
3696 r
= cb(sha1
, p
, i
, data
);
3703 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3705 struct packed_git
*p
;
3707 int pack_errors
= 0;
3709 prepare_packed_git();
3710 for (p
= packed_git
; p
; p
= p
->next
) {
3711 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3713 if (open_pack_index(p
)) {
3717 r
= for_each_object_in_pack(p
, cb
, data
);
3721 return r
? r
: pack_errors
;