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"
30 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
31 #define O_NOATIME 01000000
37 #define SZ_FMT PRIuMAX
38 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
40 const unsigned char null_sha1
[20];
41 const struct object_id null_oid
;
44 * This is meant to hold a *small* number of objects that you would
45 * want read_sha1_file() to be able to return, but yet you do not want
46 * to write them into the object store (e.g. a browse-only
49 static struct cached_object
{
50 unsigned char sha1
[20];
51 enum object_type type
;
55 static int cached_object_nr
, cached_object_alloc
;
57 static struct cached_object empty_tree
= {
58 EMPTY_TREE_SHA1_BIN_LITERAL
,
64 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
67 struct cached_object
*co
= cached_objects
;
69 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
70 if (!hashcmp(co
->sha1
, sha1
))
73 if (!hashcmp(sha1
, empty_tree
.sha1
))
78 int mkdir_in_gitdir(const char *path
)
80 if (mkdir(path
, 0777)) {
81 int saved_errno
= errno
;
83 struct strbuf sb
= STRBUF_INIT
;
88 * Are we looking at a path in a symlinked worktree
89 * whose original repository does not yet have it?
90 * e.g. .git/rr-cache pointing at its original
91 * repository in which the user hasn't performed any
92 * conflict resolution yet?
94 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
95 strbuf_readlink(&sb
, path
, st
.st_size
) ||
96 !is_absolute_path(sb
.buf
) ||
97 mkdir(sb
.buf
, 0777)) {
104 return adjust_shared_perm(path
);
107 enum scld_error
safe_create_leading_directories(char *path
)
109 char *next_component
= path
+ offset_1st_component(path
);
110 enum scld_error ret
= SCLD_OK
;
112 while (ret
== SCLD_OK
&& next_component
) {
114 char *slash
= next_component
, slash_character
;
116 while (*slash
&& !is_dir_sep(*slash
))
122 next_component
= slash
+ 1;
123 while (is_dir_sep(*next_component
))
125 if (!*next_component
)
128 slash_character
= *slash
;
130 if (!stat(path
, &st
)) {
132 if (!S_ISDIR(st
.st_mode
))
134 } else if (mkdir(path
, 0777)) {
135 if (errno
== EEXIST
&&
136 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
137 ; /* somebody created it since we checked */
138 else if (errno
== ENOENT
)
140 * Either mkdir() failed because
141 * somebody just pruned the containing
142 * directory, or stat() failed because
143 * the file that was in our way was
144 * just removed. Either way, inform
145 * the caller that it might be worth
151 } else if (adjust_shared_perm(path
)) {
154 *slash
= slash_character
;
159 enum scld_error
safe_create_leading_directories_const(const char *path
)
161 /* path points to cache entries, so xstrdup before messing with it */
162 char *buf
= xstrdup(path
);
163 enum scld_error result
= safe_create_leading_directories(buf
);
168 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
171 for (i
= 0; i
< 20; i
++) {
172 static char hex
[] = "0123456789abcdef";
173 unsigned int val
= sha1
[i
];
174 char *pos
= pathbuf
+ i
*2 + (i
> 0);
175 *pos
++ = hex
[val
>> 4];
176 *pos
= hex
[val
& 0xf];
180 const char *sha1_file_name(const unsigned char *sha1
)
182 static char buf
[PATH_MAX
];
186 objdir
= get_object_directory();
187 len
= strlen(objdir
);
189 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
190 if (len
+ 43 > PATH_MAX
)
191 die("insanely long object directory %s", objdir
);
192 memcpy(buf
, objdir
, len
);
196 fill_sha1_path(buf
+ len
+ 1, sha1
);
201 * Return the name of the pack or index file with the specified sha1
202 * in its filename. *base and *name are scratch space that must be
203 * provided by the caller. which should be "pack" or "idx".
205 static char *sha1_get_pack_name(const unsigned char *sha1
,
210 strbuf_addf(buf
, "%s/pack/pack-%s.%s", get_object_directory(),
211 sha1_to_hex(sha1
), which
);
215 char *sha1_pack_name(const unsigned char *sha1
)
217 static struct strbuf buf
= STRBUF_INIT
;
218 return sha1_get_pack_name(sha1
, &buf
, "pack");
221 char *sha1_pack_index_name(const unsigned char *sha1
)
223 static struct strbuf buf
= STRBUF_INIT
;
224 return sha1_get_pack_name(sha1
, &buf
, "idx");
227 struct alternate_object_database
*alt_odb_list
;
228 static struct alternate_object_database
**alt_odb_tail
;
231 * Prepare alternate object database registry.
233 * The variable alt_odb_list points at the list of struct
234 * alternate_object_database. The elements on this list come from
235 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
236 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
237 * whose contents is similar to that environment variable but can be
238 * LF separated. Its base points at a statically allocated buffer that
239 * contains "/the/directory/corresponding/to/.git/objects/...", while
240 * its name points just after the slash at the end of ".git/objects/"
241 * in the example above, and has enough space to hold 40-byte hex
242 * SHA1, an extra slash for the first level indirection, and the
245 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
246 int depth
, const char *normalized_objdir
)
248 struct alternate_object_database
*ent
;
249 struct alternate_object_database
*alt
;
250 size_t pfxlen
, entlen
;
251 struct strbuf pathbuf
= STRBUF_INIT
;
253 if (!is_absolute_path(entry
) && relative_base
) {
254 strbuf_addstr(&pathbuf
, real_path(relative_base
));
255 strbuf_addch(&pathbuf
, '/');
257 strbuf_addstr(&pathbuf
, entry
);
259 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
261 pfxlen
= strlen(pathbuf
.buf
);
264 * The trailing slash after the directory name is given by
265 * this function at the end. Remove duplicates.
267 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
270 entlen
= st_add(pfxlen
, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
271 ent
= xmalloc(st_add(sizeof(*ent
), entlen
));
272 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
273 strbuf_release(&pathbuf
);
275 ent
->name
= ent
->base
+ pfxlen
+ 1;
276 ent
->base
[pfxlen
+ 3] = '/';
277 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
279 /* Detect cases where alternate disappeared */
280 if (!is_directory(ent
->base
)) {
281 error("object directory %s does not exist; "
282 "check .git/objects/info/alternates.",
288 /* Prevent the common mistake of listing the same
289 * thing twice, or object directory itself.
291 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
292 if (pfxlen
== alt
->name
- alt
->base
- 1 &&
293 !memcmp(ent
->base
, alt
->base
, pfxlen
)) {
298 if (!fspathcmp(ent
->base
, normalized_objdir
)) {
303 /* add the alternate entry */
305 alt_odb_tail
= &(ent
->next
);
308 /* recursively add alternates */
309 read_info_alternates(ent
->base
, depth
+ 1);
311 ent
->base
[pfxlen
] = '/';
316 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
317 const char *relative_base
, int depth
)
319 struct string_list entries
= STRING_LIST_INIT_NODUP
;
322 struct strbuf objdirbuf
= STRBUF_INIT
;
325 error("%s: ignoring alternate object stores, nesting too deep.",
330 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
331 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
333 alt_copy
= xmemdupz(alt
, len
);
334 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
335 for (i
= 0; i
< entries
.nr
; i
++) {
336 const char *entry
= entries
.items
[i
].string
;
337 if (entry
[0] == '\0' || entry
[0] == '#')
339 if (!is_absolute_path(entry
) && depth
) {
340 error("%s: ignoring relative alternate object store %s",
341 relative_base
, entry
);
343 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
346 string_list_clear(&entries
, 0);
348 strbuf_release(&objdirbuf
);
351 void read_info_alternates(const char * relative_base
, int depth
)
359 path
= xstrfmt("%s/info/alternates", relative_base
);
360 fd
= git_open_noatime(path
);
364 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
368 mapsz
= xsize_t(st
.st_size
);
369 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
372 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
377 void add_to_alternates_file(const char *reference
)
379 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
380 char *alts
= git_pathdup("objects/info/alternates");
383 hold_lock_file_for_update(lock
, alts
, LOCK_DIE_ON_ERROR
);
384 out
= fdopen_lock_file(lock
, "w");
386 die_errno("unable to fdopen alternates lockfile");
388 in
= fopen(alts
, "r");
390 struct strbuf line
= STRBUF_INIT
;
393 while (strbuf_getline(&line
, in
) != EOF
) {
394 if (!strcmp(reference
, line
.buf
)) {
398 fprintf_or_die(out
, "%s\n", line
.buf
);
401 strbuf_release(&line
);
405 rollback_lock_file(lock
);
409 else if (errno
!= ENOENT
)
410 die_errno("unable to read alternates file");
413 fprintf_or_die(out
, "%s\n", reference
);
414 if (commit_lock_file(lock
))
415 die_errno("unable to move new alternates file into place");
417 link_alt_odb_entries(reference
, strlen(reference
), '\n', NULL
, 0);
422 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
424 struct alternate_object_database
*ent
;
428 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
436 void prepare_alt_odb(void)
443 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
446 alt_odb_tail
= &alt_odb_list
;
447 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
449 read_info_alternates(get_object_directory(), 0);
452 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
453 static int freshen_file(const char *fn
)
456 t
.actime
= t
.modtime
= time(NULL
);
457 return !utime(fn
, &t
);
461 * All of the check_and_freshen functions return 1 if the file exists and was
462 * freshened (if freshening was requested), 0 otherwise. If they return
463 * 0, you should not assume that it is safe to skip a write of the object (it
464 * either does not exist on disk, or has a stale mtime and may be subject to
467 static int check_and_freshen_file(const char *fn
, int freshen
)
469 if (access(fn
, F_OK
))
471 if (freshen
&& !freshen_file(fn
))
476 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
478 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
481 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
483 struct alternate_object_database
*alt
;
485 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
486 fill_sha1_path(alt
->name
, sha1
);
487 if (check_and_freshen_file(alt
->base
, freshen
))
493 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
495 return check_and_freshen_local(sha1
, freshen
) ||
496 check_and_freshen_nonlocal(sha1
, freshen
);
499 int has_loose_object_nonlocal(const unsigned char *sha1
)
501 return check_and_freshen_nonlocal(sha1
, 0);
504 static int has_loose_object(const unsigned char *sha1
)
506 return check_and_freshen(sha1
, 0);
509 static unsigned int pack_used_ctr
;
510 static unsigned int pack_mmap_calls
;
511 static unsigned int peak_pack_open_windows
;
512 static unsigned int pack_open_windows
;
513 static unsigned int pack_open_fds
;
514 static unsigned int pack_max_fds
;
515 static size_t peak_pack_mapped
;
516 static size_t pack_mapped
;
517 struct packed_git
*packed_git
;
519 static struct mru packed_git_mru_storage
;
520 struct mru
*packed_git_mru
= &packed_git_mru_storage
;
522 void pack_report(void)
525 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
526 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
527 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
528 sz_fmt(getpagesize()),
529 sz_fmt(packed_git_window_size
),
530 sz_fmt(packed_git_limit
));
532 "pack_report: pack_used_ctr = %10u\n"
533 "pack_report: pack_mmap_calls = %10u\n"
534 "pack_report: pack_open_windows = %10u / %10u\n"
535 "pack_report: pack_mapped = "
536 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
539 pack_open_windows
, peak_pack_open_windows
,
540 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
544 * Open and mmap the index file at path, perform a couple of
545 * consistency checks, then record its information to p. Return 0 on
548 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
551 struct pack_idx_header
*hdr
;
553 uint32_t version
, nr
, i
, *index
;
554 int fd
= git_open_noatime(path
);
559 if (fstat(fd
, &st
)) {
563 idx_size
= xsize_t(st
.st_size
);
564 if (idx_size
< 4 * 256 + 20 + 20) {
566 return error("index file %s is too small", path
);
568 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
572 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
573 version
= ntohl(hdr
->idx_version
);
574 if (version
< 2 || version
> 2) {
575 munmap(idx_map
, idx_size
);
576 return error("index file %s is version %"PRIu32
577 " and is not supported by this binary"
578 " (try upgrading GIT to a newer version)",
587 index
+= 2; /* skip index header */
588 for (i
= 0; i
< 256; i
++) {
589 uint32_t n
= ntohl(index
[i
]);
591 munmap(idx_map
, idx_size
);
592 return error("non-monotonic index %s", path
);
600 * - 256 index entries 4 bytes each
601 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
602 * - 20-byte SHA1 of the packfile
603 * - 20-byte SHA1 file checksum
605 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
606 munmap(idx_map
, idx_size
);
607 return error("wrong index v1 file size in %s", path
);
609 } else if (version
== 2) {
612 * - 8 bytes of header
613 * - 256 index entries 4 bytes each
614 * - 20-byte sha1 entry * nr
615 * - 4-byte crc entry * nr
616 * - 4-byte offset entry * nr
617 * - 20-byte SHA1 of the packfile
618 * - 20-byte SHA1 file checksum
619 * And after the 4-byte offset table might be a
620 * variable sized table containing 8-byte entries
621 * for offsets larger than 2^31.
623 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
624 unsigned long max_size
= min_size
;
626 max_size
+= (nr
- 1)*8;
627 if (idx_size
< min_size
|| idx_size
> max_size
) {
628 munmap(idx_map
, idx_size
);
629 return error("wrong index v2 file size in %s", path
);
631 if (idx_size
!= min_size
&&
633 * make sure we can deal with large pack offsets.
634 * 31-bit signed offset won't be enough, neither
635 * 32-bit unsigned one will be.
637 (sizeof(off_t
) <= 4)) {
638 munmap(idx_map
, idx_size
);
639 return error("pack too large for current definition of off_t in %s", path
);
643 p
->index_version
= version
;
644 p
->index_data
= idx_map
;
645 p
->index_size
= idx_size
;
650 int open_pack_index(struct packed_git
*p
)
659 if (!strip_suffix(p
->pack_name
, ".pack", &len
))
660 die("BUG: pack_name does not end in .pack");
661 idx_name
= xstrfmt("%.*s.idx", (int)len
, p
->pack_name
);
662 ret
= check_packed_git_idx(idx_name
, p
);
667 static void scan_windows(struct packed_git
*p
,
668 struct packed_git
**lru_p
,
669 struct pack_window
**lru_w
,
670 struct pack_window
**lru_l
)
672 struct pack_window
*w
, *w_l
;
674 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
676 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
686 static int unuse_one_window(struct packed_git
*current
)
688 struct packed_git
*p
, *lru_p
= NULL
;
689 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
692 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
693 for (p
= packed_git
; p
; p
= p
->next
)
694 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
696 munmap(lru_w
->base
, lru_w
->len
);
697 pack_mapped
-= lru_w
->len
;
699 lru_l
->next
= lru_w
->next
;
701 lru_p
->windows
= lru_w
->next
;
709 void release_pack_memory(size_t need
)
711 size_t cur
= pack_mapped
;
712 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
716 static void mmap_limit_check(size_t length
)
718 static size_t limit
= 0;
720 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
725 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
726 (uintmax_t)length
, (uintmax_t)limit
);
729 void *xmmap_gently(void *start
, size_t length
,
730 int prot
, int flags
, int fd
, off_t offset
)
734 mmap_limit_check(length
);
735 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
736 if (ret
== MAP_FAILED
) {
739 release_pack_memory(length
);
740 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
745 void *xmmap(void *start
, size_t length
,
746 int prot
, int flags
, int fd
, off_t offset
)
748 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
749 if (ret
== MAP_FAILED
)
750 die_errno("mmap failed");
754 void close_pack_windows(struct packed_git
*p
)
757 struct pack_window
*w
= p
->windows
;
760 die("pack '%s' still has open windows to it",
762 munmap(w
->base
, w
->len
);
763 pack_mapped
-= w
->len
;
765 p
->windows
= w
->next
;
770 static int close_pack_fd(struct packed_git
*p
)
782 static void close_pack(struct packed_git
*p
)
784 close_pack_windows(p
);
789 void close_all_packs(void)
791 struct packed_git
*p
;
793 for (p
= packed_git
; p
; p
= p
->next
)
795 die("BUG: want to close pack marked 'do-not-close'");
802 * The LRU pack is the one with the oldest MRU window, preferring packs
803 * with no used windows, or the oldest mtime if it has no windows allocated.
805 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
807 struct pack_window
*w
, *this_mru_w
;
808 int has_windows_inuse
= 0;
811 * Reject this pack if it has windows and the previously selected
812 * one does not. If this pack does not have windows, reject
813 * it if the pack file is newer than the previously selected one.
815 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
818 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
820 * Reject this pack if any of its windows are in use,
821 * but the previously selected pack did not have any
822 * inuse windows. Otherwise, record that this pack
823 * has windows in use.
826 if (*accept_windows_inuse
)
827 has_windows_inuse
= 1;
832 if (w
->last_used
> this_mru_w
->last_used
)
836 * Reject this pack if it has windows that have been
837 * used more recently than the previously selected pack.
838 * If the previously selected pack had windows inuse and
839 * we have not encountered a window in this pack that is
840 * inuse, skip this check since we prefer a pack with no
841 * inuse windows to one that has inuse windows.
843 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
844 this_mru_w
->last_used
> (*mru_w
)->last_used
)
853 *accept_windows_inuse
= has_windows_inuse
;
856 static int close_one_pack(void)
858 struct packed_git
*p
, *lru_p
= NULL
;
859 struct pack_window
*mru_w
= NULL
;
860 int accept_windows_inuse
= 1;
862 for (p
= packed_git
; p
; p
= p
->next
) {
863 if (p
->pack_fd
== -1)
865 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
869 return close_pack_fd(lru_p
);
874 void unuse_pack(struct pack_window
**w_cursor
)
876 struct pack_window
*w
= *w_cursor
;
883 void close_pack_index(struct packed_git
*p
)
886 munmap((void *)p
->index_data
, p
->index_size
);
887 p
->index_data
= NULL
;
891 static unsigned int get_max_fd_limit(void)
897 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
904 long open_max
= sysconf(_SC_OPEN_MAX
);
908 * Otherwise, we got -1 for one of the two
911 * (1) sysconf() did not understand _SC_OPEN_MAX
912 * and signaled an error with -1; or
913 * (2) sysconf() said there is no limit.
915 * We _could_ clear errno before calling sysconf() to
916 * tell these two cases apart and return a huge number
917 * in the latter case to let the caller cap it to a
918 * value that is not so selfish, but letting the
919 * fallback OPEN_MAX codepath take care of these cases
928 return 1; /* see the caller ;-) */
933 * Do not call this directly as this leaks p->pack_fd on error return;
934 * call open_packed_git() instead.
936 static int open_packed_git_1(struct packed_git
*p
)
939 struct pack_header hdr
;
940 unsigned char sha1
[20];
941 unsigned char *idx_sha1
;
944 if (!p
->index_data
&& open_pack_index(p
))
945 return error("packfile %s index unavailable", p
->pack_name
);
948 unsigned int max_fds
= get_max_fd_limit();
950 /* Save 3 for stdin/stdout/stderr, 22 for work */
952 pack_max_fds
= max_fds
- 25;
957 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
960 p
->pack_fd
= git_open_noatime(p
->pack_name
);
961 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
965 /* If we created the struct before we had the pack we lack size. */
967 if (!S_ISREG(st
.st_mode
))
968 return error("packfile %s not a regular file", p
->pack_name
);
969 p
->pack_size
= st
.st_size
;
970 } else if (p
->pack_size
!= st
.st_size
)
971 return error("packfile %s size changed", p
->pack_name
);
973 /* We leave these file descriptors open with sliding mmap;
974 * there is no point keeping them open across exec(), though.
976 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
978 return error("cannot determine file descriptor flags");
979 fd_flag
|= FD_CLOEXEC
;
980 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
981 return error("cannot set FD_CLOEXEC");
983 /* Verify we recognize this pack file format. */
984 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
985 return error("file %s is far too short to be a packfile", p
->pack_name
);
986 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
987 return error("file %s is not a GIT packfile", p
->pack_name
);
988 if (!pack_version_ok(hdr
.hdr_version
))
989 return error("packfile %s is version %"PRIu32
" and not"
990 " supported (try upgrading GIT to a newer version)",
991 p
->pack_name
, ntohl(hdr
.hdr_version
));
993 /* Verify the pack matches its index. */
994 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
995 return error("packfile %s claims to have %"PRIu32
" objects"
996 " while index indicates %"PRIu32
" objects",
997 p
->pack_name
, ntohl(hdr
.hdr_entries
),
999 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
1000 return error("end of packfile %s is unavailable", p
->pack_name
);
1001 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
1002 return error("packfile %s signature is unavailable", p
->pack_name
);
1003 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
1004 if (hashcmp(sha1
, idx_sha1
))
1005 return error("packfile %s does not match index", p
->pack_name
);
1009 static int open_packed_git(struct packed_git
*p
)
1011 if (!open_packed_git_1(p
))
1017 static int in_window(struct pack_window
*win
, off_t offset
)
1019 /* We must promise at least 20 bytes (one hash) after the
1020 * offset is available from this window, otherwise the offset
1021 * is not actually in this window and a different window (which
1022 * has that one hash excess) must be used. This is to support
1023 * the object header and delta base parsing routines below.
1025 off_t win_off
= win
->offset
;
1026 return win_off
<= offset
1027 && (offset
+ 20) <= (win_off
+ win
->len
);
1030 unsigned char *use_pack(struct packed_git
*p
,
1031 struct pack_window
**w_cursor
,
1033 unsigned long *left
)
1035 struct pack_window
*win
= *w_cursor
;
1037 /* Since packfiles end in a hash of their content and it's
1038 * pointless to ask for an offset into the middle of that
1039 * hash, and the in_window function above wouldn't match
1040 * don't allow an offset too close to the end of the file.
1042 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1043 die("packfile %s cannot be accessed", p
->pack_name
);
1044 if (offset
> (p
->pack_size
- 20))
1045 die("offset beyond end of packfile (truncated pack?)");
1047 die(_("offset before end of packfile (broken .idx?)"));
1049 if (!win
|| !in_window(win
, offset
)) {
1052 for (win
= p
->windows
; win
; win
= win
->next
) {
1053 if (in_window(win
, offset
))
1057 size_t window_align
= packed_git_window_size
/ 2;
1060 if (p
->pack_fd
== -1 && open_packed_git(p
))
1061 die("packfile %s cannot be accessed", p
->pack_name
);
1063 win
= xcalloc(1, sizeof(*win
));
1064 win
->offset
= (offset
/ window_align
) * window_align
;
1065 len
= p
->pack_size
- win
->offset
;
1066 if (len
> packed_git_window_size
)
1067 len
= packed_git_window_size
;
1068 win
->len
= (size_t)len
;
1069 pack_mapped
+= win
->len
;
1070 while (packed_git_limit
< pack_mapped
1071 && unuse_one_window(p
))
1073 win
->base
= xmmap(NULL
, win
->len
,
1074 PROT_READ
, MAP_PRIVATE
,
1075 p
->pack_fd
, win
->offset
);
1076 if (win
->base
== MAP_FAILED
)
1077 die_errno("packfile %s cannot be mapped",
1079 if (!win
->offset
&& win
->len
== p
->pack_size
1080 && !p
->do_not_close
)
1083 pack_open_windows
++;
1084 if (pack_mapped
> peak_pack_mapped
)
1085 peak_pack_mapped
= pack_mapped
;
1086 if (pack_open_windows
> peak_pack_open_windows
)
1087 peak_pack_open_windows
= pack_open_windows
;
1088 win
->next
= p
->windows
;
1092 if (win
!= *w_cursor
) {
1093 win
->last_used
= pack_used_ctr
++;
1097 offset
-= win
->offset
;
1099 *left
= win
->len
- xsize_t(offset
);
1100 return win
->base
+ offset
;
1103 static struct packed_git
*alloc_packed_git(int extra
)
1105 struct packed_git
*p
= xmalloc(st_add(sizeof(*p
), extra
));
1106 memset(p
, 0, sizeof(*p
));
1111 static void try_to_free_pack_memory(size_t size
)
1113 release_pack_memory(size
);
1116 struct packed_git
*add_packed_git(const char *path
, size_t path_len
, int local
)
1118 static int have_set_try_to_free_routine
;
1121 struct packed_git
*p
;
1123 if (!have_set_try_to_free_routine
) {
1124 have_set_try_to_free_routine
= 1;
1125 set_try_to_free_routine(try_to_free_pack_memory
);
1129 * Make sure a corresponding .pack file exists and that
1130 * the index looks sane.
1132 if (!strip_suffix_mem(path
, &path_len
, ".idx"))
1136 * ".pack" is long enough to hold any suffix we're adding (and
1137 * the use xsnprintf double-checks that)
1139 alloc
= st_add3(path_len
, strlen(".pack"), 1);
1140 p
= alloc_packed_git(alloc
);
1141 memcpy(p
->pack_name
, path
, path_len
);
1143 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".keep");
1144 if (!access(p
->pack_name
, F_OK
))
1147 xsnprintf(p
->pack_name
+ path_len
, alloc
- path_len
, ".pack");
1148 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1153 /* ok, it looks sane as far as we can check without
1154 * actually mapping the pack file.
1156 p
->pack_size
= st
.st_size
;
1157 p
->pack_local
= local
;
1158 p
->mtime
= st
.st_mtime
;
1159 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1164 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1166 const char *path
= sha1_pack_name(sha1
);
1167 size_t alloc
= st_add(strlen(path
), 1);
1168 struct packed_git
*p
= alloc_packed_git(alloc
);
1170 memcpy(p
->pack_name
, path
, alloc
); /* includes NUL */
1171 hashcpy(p
->sha1
, sha1
);
1172 if (check_packed_git_idx(idx_path
, p
)) {
1180 void install_packed_git(struct packed_git
*pack
)
1182 if (pack
->pack_fd
!= -1)
1185 pack
->next
= packed_git
;
1189 void (*report_garbage
)(unsigned seen_bits
, const char *path
);
1191 static void report_helper(const struct string_list
*list
,
1192 int seen_bits
, int first
, int last
)
1194 if (seen_bits
== (PACKDIR_FILE_PACK
|PACKDIR_FILE_IDX
))
1197 for (; first
< last
; first
++)
1198 report_garbage(seen_bits
, list
->items
[first
].string
);
1201 static void report_pack_garbage(struct string_list
*list
)
1203 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1205 if (!report_garbage
)
1208 string_list_sort(list
);
1210 for (i
= 0; i
< list
->nr
; i
++) {
1211 const char *path
= list
->items
[i
].string
;
1212 if (baselen
!= -1 &&
1213 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1214 report_helper(list
, seen_bits
, first
, i
);
1218 if (baselen
== -1) {
1219 const char *dot
= strrchr(path
, '.');
1221 report_garbage(PACKDIR_FILE_GARBAGE
, path
);
1224 baselen
= dot
- path
+ 1;
1227 if (!strcmp(path
+ baselen
, "pack"))
1229 else if (!strcmp(path
+ baselen
, "idx"))
1232 report_helper(list
, seen_bits
, first
, list
->nr
);
1235 static void prepare_packed_git_one(char *objdir
, int local
)
1237 struct strbuf path
= STRBUF_INIT
;
1241 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1243 strbuf_addstr(&path
, objdir
);
1244 strbuf_addstr(&path
, "/pack");
1245 dir
= opendir(path
.buf
);
1247 if (errno
!= ENOENT
)
1248 error_errno("unable to open object pack directory: %s",
1250 strbuf_release(&path
);
1253 strbuf_addch(&path
, '/');
1254 dirnamelen
= path
.len
;
1255 while ((de
= readdir(dir
)) != NULL
) {
1256 struct packed_git
*p
;
1259 if (is_dot_or_dotdot(de
->d_name
))
1262 strbuf_setlen(&path
, dirnamelen
);
1263 strbuf_addstr(&path
, de
->d_name
);
1265 base_len
= path
.len
;
1266 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1267 /* Don't reopen a pack we already have. */
1268 for (p
= packed_git
; p
; p
= p
->next
) {
1270 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1272 !memcmp(p
->pack_name
, path
.buf
, len
))
1277 * See if it really is a valid .idx file with
1278 * corresponding .pack file that we can map.
1280 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1281 install_packed_git(p
);
1284 if (!report_garbage
)
1287 if (ends_with(de
->d_name
, ".idx") ||
1288 ends_with(de
->d_name
, ".pack") ||
1289 ends_with(de
->d_name
, ".bitmap") ||
1290 ends_with(de
->d_name
, ".keep"))
1291 string_list_append(&garbage
, path
.buf
);
1293 report_garbage(PACKDIR_FILE_GARBAGE
, path
.buf
);
1296 report_pack_garbage(&garbage
);
1297 string_list_clear(&garbage
, 0);
1298 strbuf_release(&path
);
1301 static int sort_pack(const void *a_
, const void *b_
)
1303 struct packed_git
*a
= *((struct packed_git
**)a_
);
1304 struct packed_git
*b
= *((struct packed_git
**)b_
);
1308 * Local packs tend to contain objects specific to our
1309 * variant of the project than remote ones. In addition,
1310 * remote ones could be on a network mounted filesystem.
1311 * Favor local ones for these reasons.
1313 st
= a
->pack_local
- b
->pack_local
;
1318 * Younger packs tend to contain more recent objects,
1319 * and more recent objects tend to get accessed more
1322 if (a
->mtime
< b
->mtime
)
1324 else if (a
->mtime
== b
->mtime
)
1329 static void rearrange_packed_git(void)
1331 struct packed_git
**ary
, *p
;
1334 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1339 /* prepare an array of packed_git for easier sorting */
1340 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1341 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1344 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1346 /* link them back again */
1347 for (i
= 0; i
< n
- 1; i
++)
1348 ary
[i
]->next
= ary
[i
+ 1];
1349 ary
[n
- 1]->next
= NULL
;
1350 packed_git
= ary
[0];
1355 static void prepare_packed_git_mru(void)
1357 struct packed_git
*p
;
1359 mru_clear(packed_git_mru
);
1360 for (p
= packed_git
; p
; p
= p
->next
)
1361 mru_append(packed_git_mru
, p
);
1364 static int prepare_packed_git_run_once
= 0;
1365 void prepare_packed_git(void)
1367 struct alternate_object_database
*alt
;
1369 if (prepare_packed_git_run_once
)
1371 prepare_packed_git_one(get_object_directory(), 1);
1373 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1375 prepare_packed_git_one(alt
->base
, 0);
1376 alt
->name
[-1] = '/';
1378 rearrange_packed_git();
1379 prepare_packed_git_mru();
1380 prepare_packed_git_run_once
= 1;
1383 void reprepare_packed_git(void)
1385 prepare_packed_git_run_once
= 0;
1386 prepare_packed_git();
1389 static void mark_bad_packed_object(struct packed_git
*p
,
1390 const unsigned char *sha1
)
1393 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1394 if (!hashcmp(sha1
, p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* i
))
1396 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
,
1397 st_mult(GIT_SHA1_RAWSZ
,
1398 st_add(p
->num_bad_objects
, 1)));
1399 hashcpy(p
->bad_object_sha1
+ GIT_SHA1_RAWSZ
* p
->num_bad_objects
, sha1
);
1400 p
->num_bad_objects
++;
1403 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1405 struct packed_git
*p
;
1408 for (p
= packed_git
; p
; p
= p
->next
)
1409 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1410 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1416 * With an in-core object data in "map", rehash it to make sure the
1417 * object name actually matches "sha1" to detect object corruption.
1418 * With "map" == NULL, try reading the object named with "sha1" using
1419 * the streaming interface and rehash it to do the same.
1421 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1422 unsigned long size
, const char *type
)
1424 unsigned char real_sha1
[20];
1425 enum object_type obj_type
;
1426 struct git_istream
*st
;
1432 hash_sha1_file(map
, size
, type
, real_sha1
);
1433 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1436 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1440 /* Generate the header */
1441 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(obj_type
), size
) + 1;
1445 git_SHA1_Update(&c
, hdr
, hdrlen
);
1447 char buf
[1024 * 16];
1448 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1456 git_SHA1_Update(&c
, buf
, readlen
);
1458 git_SHA1_Final(real_sha1
, &c
);
1460 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1463 int git_open_noatime(const char *name
)
1465 static int sha1_file_open_flag
= O_NOATIME
;
1471 fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1475 /* Might the failure be due to O_NOATIME? */
1476 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1477 sha1_file_open_flag
= 0;
1485 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1487 struct alternate_object_database
*alt
;
1489 if (!lstat(sha1_file_name(sha1
), st
))
1494 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1495 fill_sha1_path(alt
->name
, sha1
);
1496 if (!lstat(alt
->base
, st
))
1503 static int open_sha1_file(const unsigned char *sha1
)
1506 struct alternate_object_database
*alt
;
1507 int most_interesting_errno
;
1509 fd
= git_open_noatime(sha1_file_name(sha1
));
1512 most_interesting_errno
= errno
;
1515 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1516 fill_sha1_path(alt
->name
, sha1
);
1517 fd
= git_open_noatime(alt
->base
);
1520 if (most_interesting_errno
== ENOENT
)
1521 most_interesting_errno
= errno
;
1523 errno
= most_interesting_errno
;
1527 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1532 fd
= open_sha1_file(sha1
);
1537 if (!fstat(fd
, &st
)) {
1538 *size
= xsize_t(st
.st_size
);
1540 /* mmap() is forbidden on empty files */
1541 error("object file %s is empty", sha1_file_name(sha1
));
1544 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1551 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1552 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1555 unsigned long size
, c
;
1556 unsigned long used
= 0;
1559 *type
= (c
>> 4) & 7;
1563 if (len
<= used
|| bitsizeof(long) <= shift
) {
1564 error("bad object header");
1569 size
+= (c
& 0x7f) << shift
;
1576 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1578 /* Get the data stream */
1579 memset(stream
, 0, sizeof(*stream
));
1580 stream
->next_in
= map
;
1581 stream
->avail_in
= mapsize
;
1582 stream
->next_out
= buffer
;
1583 stream
->avail_out
= bufsiz
;
1585 git_inflate_init(stream
);
1586 return git_inflate(stream
, 0);
1589 static int unpack_sha1_header_to_strbuf(git_zstream
*stream
, unsigned char *map
,
1590 unsigned long mapsize
, void *buffer
,
1591 unsigned long bufsiz
, struct strbuf
*header
)
1595 status
= unpack_sha1_header(stream
, map
, mapsize
, buffer
, bufsiz
);
1598 * Check if entire header is unpacked in the first iteration.
1600 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1604 * buffer[0..bufsiz] was not large enough. Copy the partial
1605 * result out to header, and then append the result of further
1606 * reading the stream.
1608 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1609 stream
->next_out
= buffer
;
1610 stream
->avail_out
= bufsiz
;
1613 status
= git_inflate(stream
, 0);
1614 strbuf_add(header
, buffer
, stream
->next_out
- (unsigned char *)buffer
);
1615 if (memchr(buffer
, '\0', stream
->next_out
- (unsigned char *)buffer
))
1617 stream
->next_out
= buffer
;
1618 stream
->avail_out
= bufsiz
;
1619 } while (status
!= Z_STREAM_END
);
1623 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1625 int bytes
= strlen(buffer
) + 1;
1626 unsigned char *buf
= xmallocz(size
);
1630 n
= stream
->total_out
- bytes
;
1633 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1635 if (bytes
<= size
) {
1637 * The above condition must be (bytes <= size), not
1638 * (bytes < size). In other words, even though we
1639 * expect no more output and set avail_out to zero,
1640 * the input zlib stream may have bytes that express
1641 * "this concludes the stream", and we *do* want to
1644 * Otherwise we would not be able to test that we
1645 * consumed all the input to reach the expected size;
1646 * we also want to check that zlib tells us that all
1647 * went well with status == Z_STREAM_END at the end.
1649 stream
->next_out
= buf
+ bytes
;
1650 stream
->avail_out
= size
- bytes
;
1651 while (status
== Z_OK
)
1652 status
= git_inflate(stream
, Z_FINISH
);
1654 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1655 git_inflate_end(stream
);
1660 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1661 else if (stream
->avail_in
)
1662 error("garbage at end of loose object '%s'",
1669 * We used to just use "sscanf()", but that's actually way
1670 * too permissive for what we want to check. So do an anal
1671 * object header parse by hand.
1673 static int parse_sha1_header_extended(const char *hdr
, struct object_info
*oi
,
1676 const char *type_buf
= hdr
;
1678 int type
, type_len
= 0;
1681 * The type can be of any size but is followed by
1691 type
= type_from_string_gently(type_buf
, type_len
, 1);
1693 strbuf_add(oi
->typename
, type_buf
, type_len
);
1695 * Set type to 0 if its an unknown object and
1696 * we're obtaining the type using '--allow-unknown-type'
1699 if ((flags
& LOOKUP_UNKNOWN_OBJECT
) && (type
< 0))
1702 die("invalid object type");
1707 * The length must follow immediately, and be in canonical
1708 * decimal format (ie "010" is not valid).
1710 size
= *hdr
++ - '0';
1715 unsigned long c
= *hdr
- '0';
1719 size
= size
* 10 + c
;
1727 * The length must be followed by a zero byte
1729 return *hdr
? -1 : type
;
1732 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1734 struct object_info oi
;
1739 return parse_sha1_header_extended(hdr
, &oi
, LOOKUP_REPLACE_OBJECT
);
1742 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1748 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1749 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1752 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1755 unsigned long get_size_from_delta(struct packed_git
*p
,
1756 struct pack_window
**w_curs
,
1759 const unsigned char *data
;
1760 unsigned char delta_head
[20], *in
;
1764 memset(&stream
, 0, sizeof(stream
));
1765 stream
.next_out
= delta_head
;
1766 stream
.avail_out
= sizeof(delta_head
);
1768 git_inflate_init(&stream
);
1770 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1771 stream
.next_in
= in
;
1772 st
= git_inflate(&stream
, Z_FINISH
);
1773 curpos
+= stream
.next_in
- in
;
1774 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1775 stream
.total_out
< sizeof(delta_head
));
1776 git_inflate_end(&stream
);
1777 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1778 error("delta data unpack-initial failed");
1782 /* Examine the initial part of the delta to figure out
1787 /* ignore base size */
1788 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1790 /* Read the result size */
1791 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1794 static off_t
get_delta_base(struct packed_git
*p
,
1795 struct pack_window
**w_curs
,
1797 enum object_type type
,
1798 off_t delta_obj_offset
)
1800 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1803 /* use_pack() assured us we have [base_info, base_info + 20)
1804 * as a range that we can look at without walking off the
1805 * end of the mapped window. Its actually the hash size
1806 * that is assured. An OFS_DELTA longer than the hash size
1807 * is stupid, as then a REF_DELTA would be smaller to store.
1809 if (type
== OBJ_OFS_DELTA
) {
1811 unsigned char c
= base_info
[used
++];
1812 base_offset
= c
& 127;
1815 if (!base_offset
|| MSB(base_offset
, 7))
1816 return 0; /* overflow */
1817 c
= base_info
[used
++];
1818 base_offset
= (base_offset
<< 7) + (c
& 127);
1820 base_offset
= delta_obj_offset
- base_offset
;
1821 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1822 return 0; /* out of bound */
1824 } else if (type
== OBJ_REF_DELTA
) {
1825 /* The base entry _must_ be in the same pack */
1826 base_offset
= find_pack_entry_one(base_info
, p
);
1829 die("I am totally screwed");
1834 * Like get_delta_base above, but we return the sha1 instead of the pack
1835 * offset. This means it is cheaper for REF deltas (we do not have to do
1836 * the final object lookup), but more expensive for OFS deltas (we
1837 * have to load the revidx to convert the offset back into a sha1).
1839 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1840 struct pack_window
**w_curs
,
1842 enum object_type type
,
1843 off_t delta_obj_offset
)
1845 if (type
== OBJ_REF_DELTA
) {
1846 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1848 } else if (type
== OBJ_OFS_DELTA
) {
1849 struct revindex_entry
*revidx
;
1850 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1851 type
, delta_obj_offset
);
1856 revidx
= find_pack_revindex(p
, base_offset
);
1860 return nth_packed_object_sha1(p
, revidx
->nr
);
1865 int unpack_object_header(struct packed_git
*p
,
1866 struct pack_window
**w_curs
,
1868 unsigned long *sizep
)
1870 unsigned char *base
;
1873 enum object_type type
;
1875 /* use_pack() assures us we have [base, base + 20) available
1876 * as a range that we can look at. (Its actually the hash
1877 * size that is assured.) With our object header encoding
1878 * the maximum deflated object size is 2^137, which is just
1879 * insane, so we know won't exceed what we have been given.
1881 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1882 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1891 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1894 struct revindex_entry
*revidx
;
1895 const unsigned char *sha1
;
1896 revidx
= find_pack_revindex(p
, obj_offset
);
1899 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1900 mark_bad_packed_object(p
, sha1
);
1901 type
= sha1_object_info(sha1
, NULL
);
1902 if (type
<= OBJ_NONE
)
1907 #define POI_STACK_PREALLOC 64
1909 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1911 enum object_type type
,
1912 struct pack_window
**w_curs
,
1915 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1916 off_t
*poi_stack
= small_poi_stack
;
1917 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1919 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1922 /* Push the object we're going to leave behind */
1923 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1924 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1925 ALLOC_ARRAY(poi_stack
, poi_stack_alloc
);
1926 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1928 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1930 poi_stack
[poi_stack_nr
++] = obj_offset
;
1931 /* If parsing the base offset fails, just unwind */
1932 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1935 curpos
= obj_offset
= base_offset
;
1936 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1937 if (type
<= OBJ_NONE
) {
1938 /* If getting the base itself fails, we first
1939 * retry the base, otherwise unwind */
1940 type
= retry_bad_packed_offset(p
, base_offset
);
1941 if (type
> OBJ_NONE
)
1955 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1956 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1961 if (poi_stack
!= small_poi_stack
)
1966 while (poi_stack_nr
) {
1967 obj_offset
= poi_stack
[--poi_stack_nr
];
1968 type
= retry_bad_packed_offset(p
, obj_offset
);
1969 if (type
> OBJ_NONE
)
1976 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1977 struct object_info
*oi
)
1979 struct pack_window
*w_curs
= NULL
;
1981 off_t curpos
= obj_offset
;
1982 enum object_type type
;
1985 * We always get the representation type, but only convert it to
1986 * a "real" type later if the caller is interested.
1988 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1991 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1992 off_t tmp_pos
= curpos
;
1993 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1999 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
2000 if (*oi
->sizep
== 0) {
2009 if (oi
->disk_sizep
) {
2010 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2011 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
2015 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
2016 if (*oi
->typep
< 0) {
2022 if (oi
->delta_base_sha1
) {
2023 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
2024 const unsigned char *base
;
2026 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
2033 hashcpy(oi
->delta_base_sha1
, base
);
2035 hashclr(oi
->delta_base_sha1
);
2039 unuse_pack(&w_curs
);
2043 static void *unpack_compressed_entry(struct packed_git
*p
,
2044 struct pack_window
**w_curs
,
2050 unsigned char *buffer
, *in
;
2052 buffer
= xmallocz_gently(size
);
2055 memset(&stream
, 0, sizeof(stream
));
2056 stream
.next_out
= buffer
;
2057 stream
.avail_out
= size
+ 1;
2059 git_inflate_init(&stream
);
2061 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
2062 stream
.next_in
= in
;
2063 st
= git_inflate(&stream
, Z_FINISH
);
2064 if (!stream
.avail_out
)
2065 break; /* the payload is larger than it should be */
2066 curpos
+= stream
.next_in
- in
;
2067 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2068 git_inflate_end(&stream
);
2069 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2077 static struct hashmap delta_base_cache
;
2078 static size_t delta_base_cached
;
2080 static LIST_HEAD(delta_base_cache_lru
);
2082 struct delta_base_cache_key
{
2083 struct packed_git
*p
;
2087 struct delta_base_cache_entry
{
2088 struct hashmap hash
;
2089 struct delta_base_cache_key key
;
2090 struct list_head lru
;
2093 enum object_type type
;
2096 static unsigned int pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2100 hash
= (unsigned int)(intptr_t)p
+ (unsigned int)base_offset
;
2101 hash
+= (hash
>> 8) + (hash
>> 16);
2105 static struct delta_base_cache_entry
*
2106 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2108 struct hashmap_entry entry
;
2109 struct delta_base_cache_key key
;
2111 if (!delta_base_cache
.cmpfn
)
2114 hashmap_entry_init(&entry
, pack_entry_hash(p
, base_offset
));
2116 key
.base_offset
= base_offset
;
2117 return hashmap_get(&delta_base_cache
, &entry
, &key
);
2120 static int delta_base_cache_key_eq(const struct delta_base_cache_key
*a
,
2121 const struct delta_base_cache_key
*b
)
2123 return a
->p
== b
->p
&& a
->base_offset
== b
->base_offset
;
2126 static int delta_base_cache_hash_cmp(const void *va
, const void *vb
,
2129 const struct delta_base_cache_entry
*a
= va
, *b
= vb
;
2130 const struct delta_base_cache_key
*key
= vkey
;
2132 return !delta_base_cache_key_eq(&a
->key
, key
);
2134 return !delta_base_cache_key_eq(&a
->key
, &b
->key
);
2137 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2139 return !!get_delta_base_cache_entry(p
, base_offset
);
2143 * Remove the entry from the cache, but do _not_ free the associated
2144 * entry data. The caller takes ownership of the "data" buffer, and
2145 * should copy out any fields it wants before detaching.
2147 static void detach_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2149 hashmap_remove(&delta_base_cache
, ent
, &ent
->key
);
2150 list_del(&ent
->lru
);
2151 delta_base_cached
-= ent
->size
;
2155 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2156 unsigned long *base_size
, enum object_type
*type
)
2158 struct delta_base_cache_entry
*ent
;
2160 ent
= get_delta_base_cache_entry(p
, base_offset
);
2162 return unpack_entry(p
, base_offset
, type
, base_size
);
2165 *base_size
= ent
->size
;
2166 return xmemdupz(ent
->data
, ent
->size
);
2169 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2172 detach_delta_base_cache_entry(ent
);
2175 void clear_delta_base_cache(void)
2177 struct hashmap_iter iter
;
2178 struct delta_base_cache_entry
*entry
;
2179 for (entry
= hashmap_iter_first(&delta_base_cache
, &iter
);
2181 entry
= hashmap_iter_next(&iter
)) {
2182 release_delta_base_cache(entry
);
2186 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2187 void *base
, unsigned long base_size
, enum object_type type
)
2189 struct delta_base_cache_entry
*ent
= xmalloc(sizeof(*ent
));
2190 struct list_head
*lru
;
2192 delta_base_cached
+= base_size
;
2194 list_for_each(lru
, &delta_base_cache_lru
) {
2195 struct delta_base_cache_entry
*f
=
2196 list_entry(lru
, struct delta_base_cache_entry
, lru
);
2197 if (delta_base_cached
<= delta_base_cache_limit
)
2199 release_delta_base_cache(f
);
2203 ent
->key
.base_offset
= base_offset
;
2206 ent
->size
= base_size
;
2207 list_add_tail(&ent
->lru
, &delta_base_cache_lru
);
2209 if (!delta_base_cache
.cmpfn
)
2210 hashmap_init(&delta_base_cache
, delta_base_cache_hash_cmp
, 0);
2211 hashmap_entry_init(ent
, pack_entry_hash(p
, base_offset
));
2212 hashmap_add(&delta_base_cache
, ent
);
2215 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2216 unsigned long *size
);
2218 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2220 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2221 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2222 p
->pack_name
, (uintmax_t)obj_offset
);
2225 int do_check_packed_object_crc
;
2227 #define UNPACK_ENTRY_STACK_PREALLOC 64
2228 struct unpack_entry_stack_ent
{
2234 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2235 enum object_type
*final_type
, unsigned long *final_size
)
2237 struct pack_window
*w_curs
= NULL
;
2238 off_t curpos
= obj_offset
;
2241 enum object_type type
;
2242 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2243 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2244 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2245 int base_from_cache
= 0;
2247 write_pack_access_log(p
, obj_offset
);
2249 /* PHASE 1: drill down to the innermost base object */
2253 struct delta_base_cache_entry
*ent
;
2255 ent
= get_delta_base_cache_entry(p
, curpos
);
2260 detach_delta_base_cache_entry(ent
);
2261 base_from_cache
= 1;
2265 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2266 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2267 off_t len
= revidx
[1].offset
- obj_offset
;
2268 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2269 const unsigned char *sha1
=
2270 nth_packed_object_sha1(p
, revidx
->nr
);
2271 error("bad packed object CRC for %s",
2273 mark_bad_packed_object(p
, sha1
);
2274 unuse_pack(&w_curs
);
2279 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2280 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2283 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2285 error("failed to validate delta base reference "
2286 "at offset %"PRIuMAX
" from %s",
2287 (uintmax_t)curpos
, p
->pack_name
);
2288 /* bail to phase 2, in hopes of recovery */
2293 /* push object, proceed to base */
2294 if (delta_stack_nr
>= delta_stack_alloc
2295 && delta_stack
== small_delta_stack
) {
2296 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2297 ALLOC_ARRAY(delta_stack
, delta_stack_alloc
);
2298 memcpy(delta_stack
, small_delta_stack
,
2299 sizeof(*delta_stack
)*delta_stack_nr
);
2301 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2303 i
= delta_stack_nr
++;
2304 delta_stack
[i
].obj_offset
= obj_offset
;
2305 delta_stack
[i
].curpos
= curpos
;
2306 delta_stack
[i
].size
= size
;
2308 curpos
= obj_offset
= base_offset
;
2311 /* PHASE 2: handle the base */
2316 die("BUG: unpack_entry: left loop at a valid delta");
2322 if (!base_from_cache
)
2323 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2327 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2328 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2331 /* PHASE 3: apply deltas in order */
2334 * 'data' holds the base data, or NULL if there was corruption
2336 while (delta_stack_nr
) {
2339 unsigned long delta_size
, base_size
= size
;
2345 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2349 * We're probably in deep shit, but let's try to fetch
2350 * the required base anyway from another pack or loose.
2351 * This is costly but should happen only in the presence
2352 * of a corrupted pack, and is better than failing outright.
2354 struct revindex_entry
*revidx
;
2355 const unsigned char *base_sha1
;
2356 revidx
= find_pack_revindex(p
, obj_offset
);
2358 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2359 error("failed to read delta base object %s"
2360 " at offset %"PRIuMAX
" from %s",
2361 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2363 mark_bad_packed_object(p
, base_sha1
);
2364 base
= read_object(base_sha1
, &type
, &base_size
);
2368 i
= --delta_stack_nr
;
2369 obj_offset
= delta_stack
[i
].obj_offset
;
2370 curpos
= delta_stack
[i
].curpos
;
2371 delta_size
= delta_stack
[i
].size
;
2376 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2379 error("failed to unpack compressed delta "
2380 "at offset %"PRIuMAX
" from %s",
2381 (uintmax_t)curpos
, p
->pack_name
);
2386 data
= patch_delta(base
, base_size
,
2387 delta_data
, delta_size
,
2391 * We could not apply the delta; warn the user, but keep going.
2392 * Our failure will be noticed either in the next iteration of
2393 * the loop, or if this is the final delta, in the caller when
2394 * we return NULL. Those code paths will take care of making
2395 * a more explicit warning and retrying with another copy of
2399 error("failed to apply delta");
2407 unuse_pack(&w_curs
);
2409 if (delta_stack
!= small_delta_stack
)
2415 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2418 const unsigned char *index
= p
->index_data
;
2420 if (open_pack_index(p
))
2422 index
= p
->index_data
;
2424 if (n
>= p
->num_objects
)
2427 if (p
->index_version
== 1) {
2428 return index
+ 24 * n
+ 4;
2431 return index
+ 20 * n
;
2435 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2437 const unsigned char *ptr
= vptr
;
2438 const unsigned char *start
= p
->index_data
;
2439 const unsigned char *end
= start
+ p
->index_size
;
2441 die(_("offset before start of pack index for %s (corrupt index?)"),
2443 /* No need to check for underflow; .idx files must be at least 8 bytes */
2445 die(_("offset beyond end of pack index for %s (truncated index?)"),
2449 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2451 const unsigned char *index
= p
->index_data
;
2453 if (p
->index_version
== 1) {
2454 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2457 index
+= 8 + p
->num_objects
* (20 + 4);
2458 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2459 if (!(off
& 0x80000000))
2461 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2462 check_pack_index_ptr(p
, index
);
2463 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2464 ntohl(*((uint32_t *)(index
+ 4)));
2468 off_t
find_pack_entry_one(const unsigned char *sha1
,
2469 struct packed_git
*p
)
2471 const uint32_t *level1_ofs
= p
->index_data
;
2472 const unsigned char *index
= p
->index_data
;
2473 unsigned hi
, lo
, stride
;
2474 static int use_lookup
= -1;
2475 static int debug_lookup
= -1;
2477 if (debug_lookup
< 0)
2478 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2481 if (open_pack_index(p
))
2483 level1_ofs
= p
->index_data
;
2484 index
= p
->index_data
;
2486 if (p
->index_version
> 1) {
2491 hi
= ntohl(level1_ofs
[*sha1
]);
2492 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2493 if (p
->index_version
> 1) {
2501 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2502 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2505 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2507 int pos
= sha1_entry_pos(index
, stride
, 0,
2508 lo
, hi
, p
->num_objects
, sha1
);
2511 return nth_packed_object_offset(p
, pos
);
2515 unsigned mi
= (lo
+ hi
) / 2;
2516 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2519 printf("lo %u hi %u rg %u mi %u\n",
2520 lo
, hi
, hi
- lo
, mi
);
2522 return nth_packed_object_offset(p
, mi
);
2531 int is_pack_valid(struct packed_git
*p
)
2533 /* An already open pack is known to be valid. */
2534 if (p
->pack_fd
!= -1)
2537 /* If the pack has one window completely covering the
2538 * file size, the pack is known to be valid even if
2539 * the descriptor is not currently open.
2542 struct pack_window
*w
= p
->windows
;
2544 if (!w
->offset
&& w
->len
== p
->pack_size
)
2548 /* Force the pack to open to prove its valid. */
2549 return !open_packed_git(p
);
2552 static int fill_pack_entry(const unsigned char *sha1
,
2553 struct pack_entry
*e
,
2554 struct packed_git
*p
)
2558 if (p
->num_bad_objects
) {
2560 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2561 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2565 offset
= find_pack_entry_one(sha1
, p
);
2570 * We are about to tell the caller where they can locate the
2571 * requested object. We better make sure the packfile is
2572 * still here and can be accessed before supplying that
2573 * answer, as it may have been deleted since the index was
2576 if (!is_pack_valid(p
))
2580 hashcpy(e
->sha1
, sha1
);
2585 * Iff a pack file contains the object named by sha1, return true and
2586 * store its location to e.
2588 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2590 struct mru_entry
*p
;
2592 prepare_packed_git();
2596 for (p
= packed_git_mru
->head
; p
; p
= p
->next
) {
2597 if (fill_pack_entry(sha1
, e
, p
->item
)) {
2598 mru_mark(packed_git_mru
, p
);
2605 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2606 struct packed_git
*packs
)
2608 struct packed_git
*p
;
2610 for (p
= packs
; p
; p
= p
->next
) {
2611 if (find_pack_entry_one(sha1
, p
))
2618 static int sha1_loose_object_info(const unsigned char *sha1
,
2619 struct object_info
*oi
,
2623 unsigned long mapsize
;
2627 struct strbuf hdrbuf
= STRBUF_INIT
;
2629 if (oi
->delta_base_sha1
)
2630 hashclr(oi
->delta_base_sha1
);
2633 * If we don't care about type or size, then we don't
2634 * need to look inside the object at all. Note that we
2635 * do not optimize out the stat call, even if the
2636 * caller doesn't care about the disk-size, since our
2637 * return value implicitly indicates whether the
2638 * object even exists.
2640 if (!oi
->typep
&& !oi
->typename
&& !oi
->sizep
) {
2642 if (stat_sha1_file(sha1
, &st
) < 0)
2645 *oi
->disk_sizep
= st
.st_size
;
2649 map
= map_sha1_file(sha1
, &mapsize
);
2653 *oi
->disk_sizep
= mapsize
;
2654 if ((flags
& LOOKUP_UNKNOWN_OBJECT
)) {
2655 if (unpack_sha1_header_to_strbuf(&stream
, map
, mapsize
, hdr
, sizeof(hdr
), &hdrbuf
) < 0)
2656 status
= error("unable to unpack %s header with --allow-unknown-type",
2658 } else if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2659 status
= error("unable to unpack %s header",
2663 else if (hdrbuf
.len
) {
2664 if ((status
= parse_sha1_header_extended(hdrbuf
.buf
, oi
, flags
)) < 0)
2665 status
= error("unable to parse %s header with --allow-unknown-type",
2667 } else if ((status
= parse_sha1_header_extended(hdr
, oi
, flags
)) < 0)
2668 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2669 git_inflate_end(&stream
);
2670 munmap(map
, mapsize
);
2671 if (status
&& oi
->typep
)
2672 *oi
->typep
= status
;
2673 strbuf_release(&hdrbuf
);
2677 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2679 struct cached_object
*co
;
2680 struct pack_entry e
;
2682 enum object_type real_type
;
2683 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2685 co
= find_cached_object(real
);
2688 *(oi
->typep
) = co
->type
;
2690 *(oi
->sizep
) = co
->size
;
2692 *(oi
->disk_sizep
) = 0;
2693 if (oi
->delta_base_sha1
)
2694 hashclr(oi
->delta_base_sha1
);
2696 strbuf_addstr(oi
->typename
, typename(co
->type
));
2697 oi
->whence
= OI_CACHED
;
2701 if (!find_pack_entry(real
, &e
)) {
2702 /* Most likely it's a loose object. */
2703 if (!sha1_loose_object_info(real
, oi
, flags
)) {
2704 oi
->whence
= OI_LOOSE
;
2708 /* Not a loose object; someone else may have just packed it. */
2709 reprepare_packed_git();
2710 if (!find_pack_entry(real
, &e
))
2715 * packed_object_info() does not follow the delta chain to
2716 * find out the real type, unless it is given oi->typep.
2718 if (oi
->typename
&& !oi
->typep
)
2719 oi
->typep
= &real_type
;
2721 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2723 mark_bad_packed_object(e
.p
, real
);
2724 if (oi
->typep
== &real_type
)
2726 return sha1_object_info_extended(real
, oi
, 0);
2727 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2728 oi
->whence
= OI_DBCACHED
;
2730 oi
->whence
= OI_PACKED
;
2731 oi
->u
.packed
.offset
= e
.offset
;
2732 oi
->u
.packed
.pack
= e
.p
;
2733 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2734 rtype
== OBJ_OFS_DELTA
);
2737 strbuf_addstr(oi
->typename
, typename(*oi
->typep
));
2738 if (oi
->typep
== &real_type
)
2744 /* returns enum object_type or negative */
2745 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2747 enum object_type type
;
2748 struct object_info oi
= {NULL
};
2752 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2757 static void *read_packed_sha1(const unsigned char *sha1
,
2758 enum object_type
*type
, unsigned long *size
)
2760 struct pack_entry e
;
2763 if (!find_pack_entry(sha1
, &e
))
2765 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
);
2768 * We're probably in deep shit, but let's try to fetch
2769 * the required object anyway from another pack or loose.
2770 * This should happen only in the presence of a corrupted
2771 * pack, and is better than failing outright.
2773 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2774 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2775 mark_bad_packed_object(e
.p
, sha1
);
2776 data
= read_object(sha1
, type
, size
);
2781 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2782 unsigned char *sha1
)
2784 struct cached_object
*co
;
2786 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2787 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2789 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2790 co
= &cached_objects
[cached_object_nr
++];
2793 co
->buf
= xmalloc(len
);
2794 memcpy(co
->buf
, buf
, len
);
2795 hashcpy(co
->sha1
, sha1
);
2799 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2800 unsigned long *size
)
2802 unsigned long mapsize
;
2804 struct cached_object
*co
;
2806 co
= find_cached_object(sha1
);
2810 return xmemdupz(co
->buf
, co
->size
);
2813 buf
= read_packed_sha1(sha1
, type
, size
);
2816 map
= map_sha1_file(sha1
, &mapsize
);
2818 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2819 munmap(map
, mapsize
);
2822 reprepare_packed_git();
2823 return read_packed_sha1(sha1
, type
, size
);
2827 * This function dies on corrupt objects; the callers who want to
2828 * deal with them should arrange to call read_object() and give error
2829 * messages themselves.
2831 void *read_sha1_file_extended(const unsigned char *sha1
,
2832 enum object_type
*type
,
2833 unsigned long *size
,
2837 const struct packed_git
*p
;
2838 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2841 data
= read_object(repl
, type
, size
);
2845 if (errno
&& errno
!= ENOENT
)
2846 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2848 /* die if we replaced an object with one that does not exist */
2850 die("replacement %s not found for %s",
2851 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2853 if (has_loose_object(repl
)) {
2854 const char *path
= sha1_file_name(sha1
);
2856 die("loose object %s (stored in %s) is corrupt",
2857 sha1_to_hex(repl
), path
);
2860 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2861 die("packed object %s (stored in %s) is corrupt",
2862 sha1_to_hex(repl
), p
->pack_name
);
2867 void *read_object_with_reference(const unsigned char *sha1
,
2868 const char *required_type_name
,
2869 unsigned long *size
,
2870 unsigned char *actual_sha1_return
)
2872 enum object_type type
, required_type
;
2874 unsigned long isize
;
2875 unsigned char actual_sha1
[20];
2877 required_type
= type_from_string(required_type_name
);
2878 hashcpy(actual_sha1
, sha1
);
2880 int ref_length
= -1;
2881 const char *ref_type
= NULL
;
2883 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2886 if (type
== required_type
) {
2888 if (actual_sha1_return
)
2889 hashcpy(actual_sha1_return
, actual_sha1
);
2892 /* Handle references */
2893 else if (type
== OBJ_COMMIT
)
2895 else if (type
== OBJ_TAG
)
2896 ref_type
= "object ";
2901 ref_length
= strlen(ref_type
);
2903 if (ref_length
+ 40 > isize
||
2904 memcmp(buffer
, ref_type
, ref_length
) ||
2905 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2910 /* Now we have the ID of the referred-to object in
2911 * actual_sha1. Check again. */
2915 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2916 const char *type
, unsigned char *sha1
,
2917 char *hdr
, int *hdrlen
)
2921 /* Generate the header */
2922 *hdrlen
= xsnprintf(hdr
, *hdrlen
, "%s %lu", type
, len
)+1;
2926 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2927 git_SHA1_Update(&c
, buf
, len
);
2928 git_SHA1_Final(sha1
, &c
);
2932 * Move the just written object into its final resting place.
2934 int finalize_object_file(const char *tmpfile
, const char *filename
)
2938 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2940 else if (link(tmpfile
, filename
))
2944 * Coda hack - coda doesn't like cross-directory links,
2945 * so we fall back to a rename, which will mean that it
2946 * won't be able to check collisions, but that's not a
2949 * The same holds for FAT formatted media.
2951 * When this succeeds, we just return. We have nothing
2954 if (ret
&& ret
!= EEXIST
) {
2956 if (!rename(tmpfile
, filename
))
2960 unlink_or_warn(tmpfile
);
2962 if (ret
!= EEXIST
) {
2963 return error_errno("unable to write sha1 filename %s", filename
);
2965 /* FIXME!!! Collision check here ? */
2969 if (adjust_shared_perm(filename
))
2970 return error("unable to set permission to '%s'", filename
);
2974 static int write_buffer(int fd
, const void *buf
, size_t len
)
2976 if (write_in_full(fd
, buf
, len
) < 0)
2977 return error_errno("file write error");
2981 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2982 unsigned char *sha1
)
2985 int hdrlen
= sizeof(hdr
);
2986 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2990 /* Finalize a file on disk, and close it. */
2991 static void close_sha1_file(int fd
)
2993 if (fsync_object_files
)
2994 fsync_or_die(fd
, "sha1 file");
2996 die_errno("error when closing sha1 file");
2999 /* Size of directory component, including the ending '/' */
3000 static inline int directory_size(const char *filename
)
3002 const char *s
= strrchr(filename
, '/');
3005 return s
- filename
+ 1;
3009 * This creates a temporary file in the same directory as the final
3012 * We want to avoid cross-directory filename renames, because those
3013 * can have problems on various filesystems (FAT, NFS, Coda).
3015 static int create_tmpfile(struct strbuf
*tmp
, const char *filename
)
3017 int fd
, dirlen
= directory_size(filename
);
3020 strbuf_add(tmp
, filename
, dirlen
);
3021 strbuf_addstr(tmp
, "tmp_obj_XXXXXX");
3022 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3023 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
3025 * Make sure the directory exists; note that the contents
3026 * of the buffer are undefined after mkstemp returns an
3027 * error, so we have to rewrite the whole buffer from
3031 strbuf_add(tmp
, filename
, dirlen
- 1);
3032 if (mkdir(tmp
->buf
, 0777) && errno
!= EEXIST
)
3034 if (adjust_shared_perm(tmp
->buf
))
3038 strbuf_addstr(tmp
, "/tmp_obj_XXXXXX");
3039 fd
= git_mkstemp_mode(tmp
->buf
, 0444);
3044 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
3045 const void *buf
, unsigned long len
, time_t mtime
)
3048 unsigned char compressed
[4096];
3051 unsigned char parano_sha1
[20];
3052 static struct strbuf tmp_file
= STRBUF_INIT
;
3053 const char *filename
= sha1_file_name(sha1
);
3055 fd
= create_tmpfile(&tmp_file
, filename
);
3057 if (errno
== EACCES
)
3058 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
3060 return error_errno("unable to create temporary file");
3064 git_deflate_init(&stream
, zlib_compression_level
);
3065 stream
.next_out
= compressed
;
3066 stream
.avail_out
= sizeof(compressed
);
3069 /* First header.. */
3070 stream
.next_in
= (unsigned char *)hdr
;
3071 stream
.avail_in
= hdrlen
;
3072 while (git_deflate(&stream
, 0) == Z_OK
)
3074 git_SHA1_Update(&c
, hdr
, hdrlen
);
3076 /* Then the data itself.. */
3077 stream
.next_in
= (void *)buf
;
3078 stream
.avail_in
= len
;
3080 unsigned char *in0
= stream
.next_in
;
3081 ret
= git_deflate(&stream
, Z_FINISH
);
3082 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
3083 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
3084 die("unable to write sha1 file");
3085 stream
.next_out
= compressed
;
3086 stream
.avail_out
= sizeof(compressed
);
3087 } while (ret
== Z_OK
);
3089 if (ret
!= Z_STREAM_END
)
3090 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3091 ret
= git_deflate_end_gently(&stream
);
3093 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3094 git_SHA1_Final(parano_sha1
, &c
);
3095 if (hashcmp(sha1
, parano_sha1
) != 0)
3096 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3098 close_sha1_file(fd
);
3103 utb
.modtime
= mtime
;
3104 if (utime(tmp_file
.buf
, &utb
) < 0)
3105 warning_errno("failed utime() on %s", tmp_file
.buf
);
3108 return finalize_object_file(tmp_file
.buf
, filename
);
3111 static int freshen_loose_object(const unsigned char *sha1
)
3113 return check_and_freshen(sha1
, 1);
3116 static int freshen_packed_object(const unsigned char *sha1
)
3118 struct pack_entry e
;
3119 if (!find_pack_entry(sha1
, &e
))
3123 if (!freshen_file(e
.p
->pack_name
))
3129 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3132 int hdrlen
= sizeof(hdr
);
3134 /* Normally if we have it in the pack then we do not bother writing
3135 * it out into .git/objects/??/?{38} file.
3137 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3138 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3140 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3143 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3144 unsigned char *sha1
, unsigned flags
)
3147 int hdrlen
, status
= 0;
3149 /* type string, SP, %lu of the length plus NUL must fit this */
3150 hdrlen
= strlen(type
) + 32;
3151 header
= xmalloc(hdrlen
);
3152 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3154 if (!(flags
& HASH_WRITE_OBJECT
))
3156 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3158 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3165 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3169 enum object_type type
;
3174 if (has_loose_object(sha1
))
3176 buf
= read_packed_sha1(sha1
, &type
, &len
);
3178 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3179 hdrlen
= xsnprintf(hdr
, sizeof(hdr
), "%s %lu", typename(type
), len
) + 1;
3180 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3186 int has_pack_index(const unsigned char *sha1
)
3189 if (stat(sha1_pack_index_name(sha1
), &st
))
3194 int has_sha1_pack(const unsigned char *sha1
)
3196 struct pack_entry e
;
3197 return find_pack_entry(sha1
, &e
);
3200 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3202 struct pack_entry e
;
3204 if (find_pack_entry(sha1
, &e
))
3206 if (has_loose_object(sha1
))
3208 if (flags
& HAS_SHA1_QUICK
)
3210 reprepare_packed_git();
3211 return find_pack_entry(sha1
, &e
);
3214 int has_object_file(const struct object_id
*oid
)
3216 return has_sha1_file(oid
->hash
);
3219 static void check_tree(const void *buf
, size_t size
)
3221 struct tree_desc desc
;
3222 struct name_entry entry
;
3224 init_tree_desc(&desc
, buf
, size
);
3225 while (tree_entry(&desc
, &entry
))
3227 * tree_entry() will die() on malformed entries */
3231 static void check_commit(const void *buf
, size_t size
)
3234 memset(&c
, 0, sizeof(c
));
3235 if (parse_commit_buffer(&c
, buf
, size
))
3236 die("corrupt commit");
3239 static void check_tag(const void *buf
, size_t size
)
3242 memset(&t
, 0, sizeof(t
));
3243 if (parse_tag_buffer(&t
, buf
, size
))
3247 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3248 enum object_type type
,
3249 const char *path
, unsigned flags
)
3251 int ret
, re_allocated
= 0;
3252 int write_object
= flags
& HASH_WRITE_OBJECT
;
3258 * Convert blobs to git internal format
3260 if ((type
== OBJ_BLOB
) && path
) {
3261 struct strbuf nbuf
= STRBUF_INIT
;
3262 if (convert_to_git(path
, buf
, size
, &nbuf
,
3263 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3264 buf
= strbuf_detach(&nbuf
, &size
);
3268 if (flags
& HASH_FORMAT_CHECK
) {
3269 if (type
== OBJ_TREE
)
3270 check_tree(buf
, size
);
3271 if (type
== OBJ_COMMIT
)
3272 check_commit(buf
, size
);
3273 if (type
== OBJ_TAG
)
3274 check_tag(buf
, size
);
3278 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3280 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3286 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3287 const char *path
, unsigned flags
)
3290 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3291 struct strbuf sbuf
= STRBUF_INIT
;
3294 assert(would_convert_to_git_filter_fd(path
));
3296 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3297 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3300 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3303 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3305 strbuf_release(&sbuf
);
3309 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3310 const char *path
, unsigned flags
)
3312 struct strbuf sbuf
= STRBUF_INIT
;
3315 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3316 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3319 strbuf_release(&sbuf
);
3323 #define SMALL_FILE_SIZE (32*1024)
3325 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3326 enum object_type type
, const char *path
,
3332 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3333 } else if (size
<= SMALL_FILE_SIZE
) {
3334 char *buf
= xmalloc(size
);
3335 if (size
== read_in_full(fd
, buf
, size
))
3336 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3338 ret
= error_errno("short read");
3341 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3342 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3349 * This creates one packfile per large blob unless bulk-checkin
3350 * machinery is "plugged".
3352 * This also bypasses the usual "convert-to-git" dance, and that is on
3353 * purpose. We could write a streaming version of the converting
3354 * functions and insert that before feeding the data to fast-import
3355 * (or equivalent in-core API described above). However, that is
3356 * somewhat complicated, as we do not know the size of the filter
3357 * result, which we need to know beforehand when writing a git object.
3358 * Since the primary motivation for trying to stream from the working
3359 * tree file and to avoid mmaping it in core is to deal with large
3360 * binary blobs, they generally do not want to get any conversion, and
3361 * callers should avoid this code path when filters are requested.
3363 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3364 enum object_type type
, const char *path
,
3367 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3370 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3371 enum object_type type
, const char *path
, unsigned flags
)
3376 * Call xsize_t() only when needed to avoid potentially unnecessary
3377 * die() for large files.
3379 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3380 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3381 else if (!S_ISREG(st
->st_mode
))
3382 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3383 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3384 (path
&& would_convert_to_git(path
)))
3385 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3388 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3394 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3397 struct strbuf sb
= STRBUF_INIT
;
3399 switch (st
->st_mode
& S_IFMT
) {
3401 fd
= open(path
, O_RDONLY
);
3403 return error_errno("open(\"%s\")", path
);
3404 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3405 return error("%s: failed to insert into database",
3409 if (strbuf_readlink(&sb
, path
, st
->st_size
))
3410 return error_errno("readlink(\"%s\")", path
);
3411 if (!(flags
& HASH_WRITE_OBJECT
))
3412 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3413 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3414 return error("%s: failed to insert into database",
3416 strbuf_release(&sb
);
3419 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3421 return error("%s: unsupported file type", path
);
3426 int read_pack_header(int fd
, struct pack_header
*header
)
3428 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3429 /* "eof before pack header was fully read" */
3430 return PH_ERROR_EOF
;
3432 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3433 /* "protocol error (pack signature mismatch detected)" */
3434 return PH_ERROR_PACK_SIGNATURE
;
3435 if (!pack_version_ok(header
->hdr_version
))
3436 /* "protocol error (pack version unsupported)" */
3437 return PH_ERROR_PROTOCOL
;
3441 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3443 enum object_type type
= sha1_object_info(sha1
, NULL
);
3445 die("%s is not a valid object", sha1_to_hex(sha1
));
3447 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3451 static int for_each_file_in_obj_subdir(int subdir_nr
,
3452 struct strbuf
*path
,
3453 each_loose_object_fn obj_cb
,
3454 each_loose_cruft_fn cruft_cb
,
3455 each_loose_subdir_fn subdir_cb
,
3458 size_t baselen
= path
->len
;
3459 DIR *dir
= opendir(path
->buf
);
3464 if (errno
== ENOENT
)
3466 return error_errno("unable to open %s", path
->buf
);
3469 while ((de
= readdir(dir
))) {
3470 if (is_dot_or_dotdot(de
->d_name
))
3473 strbuf_setlen(path
, baselen
);
3474 strbuf_addf(path
, "/%s", de
->d_name
);
3476 if (strlen(de
->d_name
) == 38) {
3478 unsigned char sha1
[20];
3480 snprintf(hex
, sizeof(hex
), "%02x%s",
3481 subdir_nr
, de
->d_name
);
3482 if (!get_sha1_hex(hex
, sha1
)) {
3484 r
= obj_cb(sha1
, path
->buf
, data
);
3493 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3500 strbuf_setlen(path
, baselen
);
3501 if (!r
&& subdir_cb
)
3502 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3507 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3508 each_loose_object_fn obj_cb
,
3509 each_loose_cruft_fn cruft_cb
,
3510 each_loose_subdir_fn subdir_cb
,
3513 size_t baselen
= path
->len
;
3517 for (i
= 0; i
< 256; i
++) {
3518 strbuf_addf(path
, "/%02x", i
);
3519 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3521 strbuf_setlen(path
, baselen
);
3529 int for_each_loose_file_in_objdir(const char *path
,
3530 each_loose_object_fn obj_cb
,
3531 each_loose_cruft_fn cruft_cb
,
3532 each_loose_subdir_fn subdir_cb
,
3535 struct strbuf buf
= STRBUF_INIT
;
3538 strbuf_addstr(&buf
, path
);
3539 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3541 strbuf_release(&buf
);
3546 struct loose_alt_odb_data
{
3547 each_loose_object_fn
*cb
;
3551 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3554 struct loose_alt_odb_data
*data
= vdata
;
3555 struct strbuf buf
= STRBUF_INIT
;
3558 /* copy base not including trailing '/' */
3559 strbuf_add(&buf
, alt
->base
, alt
->name
- alt
->base
- 1);
3560 r
= for_each_loose_file_in_objdir_buf(&buf
,
3561 data
->cb
, NULL
, NULL
,
3563 strbuf_release(&buf
);
3567 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3569 struct loose_alt_odb_data alt
;
3572 r
= for_each_loose_file_in_objdir(get_object_directory(),
3573 cb
, NULL
, NULL
, data
);
3577 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3582 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3585 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3590 for (i
= 0; i
< p
->num_objects
; i
++) {
3591 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3594 return error("unable to get sha1 of object %u in %s",
3597 r
= cb(sha1
, p
, i
, data
);
3604 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3606 struct packed_git
*p
;
3608 int pack_errors
= 0;
3610 prepare_packed_git();
3611 for (p
= packed_git
; p
; p
= p
->next
) {
3612 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3614 if (open_pack_index(p
)) {
3618 r
= for_each_object_in_pack(p
, cb
, data
);
3622 return r
? r
: pack_errors
;