2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
16 #include "tree-walk.h"
18 #include "pack-revindex.h"
19 #include "sha1-lookup.h"
22 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
23 #define O_NOATIME 01000000
29 #define SZ_FMT PRIuMAX
30 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
32 const unsigned char null_sha1
[20];
34 static int git_open_noatime(const char *name
, struct packed_git
*p
);
37 * This is meant to hold a *small* number of objects that you would
38 * want read_sha1_file() to be able to return, but yet you do not want
39 * to write them into the object store (e.g. a browse-only
42 static struct cached_object
{
43 unsigned char sha1
[20];
44 enum object_type type
;
48 static int cached_object_nr
, cached_object_alloc
;
50 static struct cached_object empty_tree
= {
51 EMPTY_TREE_SHA1_BIN_LITERAL
,
57 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
60 struct cached_object
*co
= cached_objects
;
62 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
63 if (!hashcmp(co
->sha1
, sha1
))
66 if (!hashcmp(sha1
, empty_tree
.sha1
))
71 int mkdir_in_gitdir(const char *path
)
73 if (mkdir(path
, 0777)) {
74 int saved_errno
= errno
;
76 struct strbuf sb
= STRBUF_INIT
;
81 * Are we looking at a path in a symlinked worktree
82 * whose original repository does not yet have it?
83 * e.g. .git/rr-cache pointing at its original
84 * repository in which the user hasn't performed any
85 * conflict resolution yet?
87 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
88 strbuf_readlink(&sb
, path
, st
.st_size
) ||
89 !is_absolute_path(sb
.buf
) ||
90 mkdir(sb
.buf
, 0777)) {
97 return adjust_shared_perm(path
);
100 int safe_create_leading_directories(char *path
)
102 char *pos
= path
+ offset_1st_component(path
);
106 pos
= strchr(pos
, '/');
109 while (*++pos
== '/')
114 if (!stat(path
, &st
)) {
116 if (!S_ISDIR(st
.st_mode
)) {
121 else if (mkdir(path
, 0777)) {
125 else if (adjust_shared_perm(path
)) {
134 int safe_create_leading_directories_const(const char *path
)
136 /* path points to cache entries, so xstrdup before messing with it */
137 char *buf
= xstrdup(path
);
138 int result
= safe_create_leading_directories(buf
);
143 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
146 for (i
= 0; i
< 20; i
++) {
147 static char hex
[] = "0123456789abcdef";
148 unsigned int val
= sha1
[i
];
149 char *pos
= pathbuf
+ i
*2 + (i
> 0);
150 *pos
++ = hex
[val
>> 4];
151 *pos
= hex
[val
& 0xf];
156 * NOTE! This returns a statically allocated buffer, so you have to be
157 * careful about using it. Do an "xstrdup()" if you need to save the
160 * Also note that this returns the location for creating. Reading
161 * SHA1 file can happen from any alternate directory listed in the
162 * DB_ENVIRONMENT environment variable if it is not found in
163 * the primary object database.
165 char *sha1_file_name(const unsigned char *sha1
)
167 static char buf
[PATH_MAX
];
171 objdir
= get_object_directory();
172 len
= strlen(objdir
);
174 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
175 if (len
+ 43 > PATH_MAX
)
176 die("insanely long object directory %s", objdir
);
177 memcpy(buf
, objdir
, len
);
181 fill_sha1_path(buf
+ len
+ 1, sha1
);
185 static char *sha1_get_pack_name(const unsigned char *sha1
,
186 char **name
, char **base
, const char *which
)
188 static const char hex
[] = "0123456789abcdef";
193 const char *sha1_file_directory
= get_object_directory();
194 int len
= strlen(sha1_file_directory
);
195 *base
= xmalloc(len
+ 60);
196 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
197 sha1_file_directory
, which
);
198 *name
= *base
+ len
+ 11;
203 for (i
= 0; i
< 20; i
++) {
204 unsigned int val
= *sha1
++;
205 *buf
++ = hex
[val
>> 4];
206 *buf
++ = hex
[val
& 0xf];
212 char *sha1_pack_name(const unsigned char *sha1
)
214 static char *name
, *base
;
216 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
219 char *sha1_pack_index_name(const unsigned char *sha1
)
221 static char *name
, *base
;
223 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
226 struct alternate_object_database
*alt_odb_list
;
227 static struct alternate_object_database
**alt_odb_tail
;
229 static void read_info_alternates(const char * alternates
, int depth
);
232 * Prepare alternate object database registry.
234 * The variable alt_odb_list points at the list of struct
235 * alternate_object_database. The elements on this list come from
236 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
237 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
238 * whose contents is similar to that environment variable but can be
239 * LF separated. Its base points at a statically allocated buffer that
240 * contains "/the/directory/corresponding/to/.git/objects/...", while
241 * its name points just after the slash at the end of ".git/objects/"
242 * in the example above, and has enough space to hold 40-byte hex
243 * SHA1, an extra slash for the first level indirection, and the
246 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
248 const char *objdir
= get_object_directory();
249 struct alternate_object_database
*ent
;
250 struct alternate_object_database
*alt
;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
253 int entlen
= pfxlen
+ 43;
256 if (!is_absolute_path(entry
) && relative_base
) {
257 /* Relative alt-odb */
259 base_len
= strlen(relative_base
) + 1;
263 ent
= xmalloc(sizeof(*ent
) + entlen
);
265 if (!is_absolute_path(entry
) && relative_base
) {
266 memcpy(ent
->base
, relative_base
, base_len
- 1);
267 ent
->base
[base_len
- 1] = '/';
268 memcpy(ent
->base
+ base_len
, entry
, len
);
271 memcpy(ent
->base
, entry
, pfxlen
);
273 ent
->name
= ent
->base
+ pfxlen
+ 1;
274 ent
->base
[pfxlen
+ 3] = '/';
275 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
277 /* Detect cases where alternate disappeared */
278 if (!is_directory(ent
->base
)) {
279 error("object directory %s does not exist; "
280 "check .git/objects/info/alternates.",
286 /* Prevent the common mistake of listing the same
287 * thing twice, or object directory itself.
289 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
290 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
295 if (!memcmp(ent
->base
, objdir
, pfxlen
)) {
300 /* add the alternate entry */
302 alt_odb_tail
= &(ent
->next
);
305 /* recursively add alternates */
306 read_info_alternates(ent
->base
, depth
+ 1);
308 ent
->base
[pfxlen
] = '/';
313 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
314 const char *relative_base
, int depth
)
316 const char *cp
, *last
;
319 error("%s: ignoring alternate object stores, nesting too deep.",
327 if (cp
< ep
&& *cp
== '#') {
328 while (cp
< ep
&& *cp
!= sep
)
333 while (cp
< ep
&& *cp
!= sep
)
336 if (!is_absolute_path(last
) && depth
) {
337 error("%s: ignoring relative alternate object store %s",
338 relative_base
, last
);
340 link_alt_odb_entry(last
, cp
- last
,
341 relative_base
, depth
);
344 while (cp
< ep
&& *cp
== sep
)
350 static void read_info_alternates(const char * relative_base
, int depth
)
355 const char alt_file_name
[] = "info/alternates";
356 /* Given that relative_base is no longer than PATH_MAX,
357 ensure that "path" has enough space to append "/", the
358 file name, "info/alternates", and a trailing NUL. */
359 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
362 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
363 fd
= git_open_noatime(path
, NULL
);
366 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
370 mapsz
= xsize_t(st
.st_size
);
371 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
374 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
379 void add_to_alternates_file(const char *reference
)
381 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
382 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
383 char *alt
= mkpath("%s/objects\n", reference
);
384 write_or_die(fd
, alt
, strlen(alt
));
385 if (commit_lock_file(lock
))
386 die("could not close alternates file");
388 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
391 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
393 struct alternate_object_database
*ent
;
396 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
401 void prepare_alt_odb(void)
408 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
411 alt_odb_tail
= &alt_odb_list
;
412 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
414 read_info_alternates(get_object_directory(), 0);
417 static int has_loose_object_local(const unsigned char *sha1
)
419 char *name
= sha1_file_name(sha1
);
420 return !access(name
, F_OK
);
423 int has_loose_object_nonlocal(const unsigned char *sha1
)
425 struct alternate_object_database
*alt
;
427 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
428 fill_sha1_path(alt
->name
, sha1
);
429 if (!access(alt
->base
, F_OK
))
435 static int has_loose_object(const unsigned char *sha1
)
437 return has_loose_object_local(sha1
) ||
438 has_loose_object_nonlocal(sha1
);
441 static unsigned int pack_used_ctr
;
442 static unsigned int pack_mmap_calls
;
443 static unsigned int peak_pack_open_windows
;
444 static unsigned int pack_open_windows
;
445 static unsigned int pack_open_fds
;
446 static unsigned int pack_max_fds
;
447 static size_t peak_pack_mapped
;
448 static size_t pack_mapped
;
449 struct packed_git
*packed_git
;
451 void pack_report(void)
454 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
455 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
456 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
457 sz_fmt(getpagesize()),
458 sz_fmt(packed_git_window_size
),
459 sz_fmt(packed_git_limit
));
461 "pack_report: pack_used_ctr = %10u\n"
462 "pack_report: pack_mmap_calls = %10u\n"
463 "pack_report: pack_open_windows = %10u / %10u\n"
464 "pack_report: pack_mapped = "
465 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
468 pack_open_windows
, peak_pack_open_windows
,
469 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
472 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
475 struct pack_idx_header
*hdr
;
477 uint32_t version
, nr
, i
, *index
;
478 int fd
= git_open_noatime(path
, p
);
483 if (fstat(fd
, &st
)) {
487 idx_size
= xsize_t(st
.st_size
);
488 if (idx_size
< 4 * 256 + 20 + 20) {
490 return error("index file %s is too small", path
);
492 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
496 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
497 version
= ntohl(hdr
->idx_version
);
498 if (version
< 2 || version
> 2) {
499 munmap(idx_map
, idx_size
);
500 return error("index file %s is version %"PRIu32
501 " and is not supported by this binary"
502 " (try upgrading GIT to a newer version)",
511 index
+= 2; /* skip index header */
512 for (i
= 0; i
< 256; i
++) {
513 uint32_t n
= ntohl(index
[i
]);
515 munmap(idx_map
, idx_size
);
516 return error("non-monotonic index %s", path
);
524 * - 256 index entries 4 bytes each
525 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526 * - 20-byte SHA1 of the packfile
527 * - 20-byte SHA1 file checksum
529 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
530 munmap(idx_map
, idx_size
);
531 return error("wrong index v1 file size in %s", path
);
533 } else if (version
== 2) {
536 * - 8 bytes of header
537 * - 256 index entries 4 bytes each
538 * - 20-byte sha1 entry * nr
539 * - 4-byte crc entry * nr
540 * - 4-byte offset entry * nr
541 * - 20-byte SHA1 of the packfile
542 * - 20-byte SHA1 file checksum
543 * And after the 4-byte offset table might be a
544 * variable sized table containing 8-byte entries
545 * for offsets larger than 2^31.
547 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
548 unsigned long max_size
= min_size
;
550 max_size
+= (nr
- 1)*8;
551 if (idx_size
< min_size
|| idx_size
> max_size
) {
552 munmap(idx_map
, idx_size
);
553 return error("wrong index v2 file size in %s", path
);
555 if (idx_size
!= min_size
&&
557 * make sure we can deal with large pack offsets.
558 * 31-bit signed offset won't be enough, neither
559 * 32-bit unsigned one will be.
561 (sizeof(off_t
) <= 4)) {
562 munmap(idx_map
, idx_size
);
563 return error("pack too large for current definition of off_t in %s", path
);
567 p
->index_version
= version
;
568 p
->index_data
= idx_map
;
569 p
->index_size
= idx_size
;
574 int open_pack_index(struct packed_git
*p
)
582 idx_name
= xstrdup(p
->pack_name
);
583 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
584 ret
= check_packed_git_idx(idx_name
, p
);
589 static void scan_windows(struct packed_git
*p
,
590 struct packed_git
**lru_p
,
591 struct pack_window
**lru_w
,
592 struct pack_window
**lru_l
)
594 struct pack_window
*w
, *w_l
;
596 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
598 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
608 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
610 struct packed_git
*p
, *lru_p
= NULL
;
611 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
614 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
615 for (p
= packed_git
; p
; p
= p
->next
)
616 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
618 munmap(lru_w
->base
, lru_w
->len
);
619 pack_mapped
-= lru_w
->len
;
621 lru_l
->next
= lru_w
->next
;
623 lru_p
->windows
= lru_w
->next
;
624 if (!lru_p
->windows
&& lru_p
->pack_fd
!= -1
625 && lru_p
->pack_fd
!= keep_fd
) {
626 close(lru_p
->pack_fd
);
638 void release_pack_memory(size_t need
, int fd
)
640 size_t cur
= pack_mapped
;
641 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
645 void *xmmap(void *start
, size_t length
,
646 int prot
, int flags
, int fd
, off_t offset
)
648 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
649 if (ret
== MAP_FAILED
) {
652 release_pack_memory(length
, fd
);
653 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
654 if (ret
== MAP_FAILED
)
655 die_errno("Out of memory? mmap failed");
660 void close_pack_windows(struct packed_git
*p
)
663 struct pack_window
*w
= p
->windows
;
666 die("pack '%s' still has open windows to it",
668 munmap(w
->base
, w
->len
);
669 pack_mapped
-= w
->len
;
671 p
->windows
= w
->next
;
676 void unuse_pack(struct pack_window
**w_cursor
)
678 struct pack_window
*w
= *w_cursor
;
685 void close_pack_index(struct packed_git
*p
)
688 munmap((void *)p
->index_data
, p
->index_size
);
689 p
->index_data
= NULL
;
694 * This is used by git-repack in case a newly created pack happens to
695 * contain the same set of objects as an existing one. In that case
696 * the resulting file might be different even if its name would be the
697 * same. It is best to close any reference to the old pack before it is
698 * replaced on disk. Of course no index pointers nor windows for given pack
699 * must subsist at this point. If ever objects from this pack are requested
700 * again, the new version of the pack will be reinitialized through
701 * reprepare_packed_git().
703 void free_pack_by_name(const char *pack_name
)
705 struct packed_git
*p
, **pp
= &packed_git
;
709 if (strcmp(pack_name
, p
->pack_name
) == 0) {
710 clear_delta_base_cache();
711 close_pack_windows(p
);
712 if (p
->pack_fd
!= -1) {
717 free(p
->bad_object_sha1
);
727 * Do not call this directly as this leaks p->pack_fd on error return;
728 * call open_packed_git() instead.
730 static int open_packed_git_1(struct packed_git
*p
)
733 struct pack_header hdr
;
734 unsigned char sha1
[20];
735 unsigned char *idx_sha1
;
738 if (!p
->index_data
&& open_pack_index(p
))
739 return error("packfile %s index unavailable", p
->pack_name
);
743 unsigned int max_fds
;
745 if (getrlimit(RLIMIT_NOFILE
, &lim
))
746 die_errno("cannot get RLIMIT_NOFILE");
748 max_fds
= lim
.rlim_cur
;
750 /* Save 3 for stdin/stdout/stderr, 22 for work */
752 pack_max_fds
= max_fds
- 25;
757 while (pack_max_fds
<= pack_open_fds
&& unuse_one_window(NULL
, -1))
760 p
->pack_fd
= git_open_noatime(p
->pack_name
, p
);
761 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
765 /* If we created the struct before we had the pack we lack size. */
767 if (!S_ISREG(st
.st_mode
))
768 return error("packfile %s not a regular file", p
->pack_name
);
769 p
->pack_size
= st
.st_size
;
770 } else if (p
->pack_size
!= st
.st_size
)
771 return error("packfile %s size changed", p
->pack_name
);
773 /* We leave these file descriptors open with sliding mmap;
774 * there is no point keeping them open across exec(), though.
776 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
778 return error("cannot determine file descriptor flags");
779 fd_flag
|= FD_CLOEXEC
;
780 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
781 return error("cannot set FD_CLOEXEC");
783 /* Verify we recognize this pack file format. */
784 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
785 return error("file %s is far too short to be a packfile", p
->pack_name
);
786 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
787 return error("file %s is not a GIT packfile", p
->pack_name
);
788 if (!pack_version_ok(hdr
.hdr_version
))
789 return error("packfile %s is version %"PRIu32
" and not"
790 " supported (try upgrading GIT to a newer version)",
791 p
->pack_name
, ntohl(hdr
.hdr_version
));
793 /* Verify the pack matches its index. */
794 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
795 return error("packfile %s claims to have %"PRIu32
" objects"
796 " while index indicates %"PRIu32
" objects",
797 p
->pack_name
, ntohl(hdr
.hdr_entries
),
799 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
800 return error("end of packfile %s is unavailable", p
->pack_name
);
801 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
802 return error("packfile %s signature is unavailable", p
->pack_name
);
803 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
804 if (hashcmp(sha1
, idx_sha1
))
805 return error("packfile %s does not match index", p
->pack_name
);
809 static int open_packed_git(struct packed_git
*p
)
811 if (!open_packed_git_1(p
))
813 if (p
->pack_fd
!= -1) {
821 static int in_window(struct pack_window
*win
, off_t offset
)
823 /* We must promise at least 20 bytes (one hash) after the
824 * offset is available from this window, otherwise the offset
825 * is not actually in this window and a different window (which
826 * has that one hash excess) must be used. This is to support
827 * the object header and delta base parsing routines below.
829 off_t win_off
= win
->offset
;
830 return win_off
<= offset
831 && (offset
+ 20) <= (win_off
+ win
->len
);
834 unsigned char *use_pack(struct packed_git
*p
,
835 struct pack_window
**w_cursor
,
839 struct pack_window
*win
= *w_cursor
;
841 /* Since packfiles end in a hash of their content and it's
842 * pointless to ask for an offset into the middle of that
843 * hash, and the in_window function above wouldn't match
844 * don't allow an offset too close to the end of the file.
846 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
847 die("packfile %s cannot be accessed", p
->pack_name
);
848 if (offset
> (p
->pack_size
- 20))
849 die("offset beyond end of packfile (truncated pack?)");
851 if (!win
|| !in_window(win
, offset
)) {
854 for (win
= p
->windows
; win
; win
= win
->next
) {
855 if (in_window(win
, offset
))
859 size_t window_align
= packed_git_window_size
/ 2;
862 if (p
->pack_fd
== -1 && open_packed_git(p
))
863 die("packfile %s cannot be accessed", p
->pack_name
);
865 win
= xcalloc(1, sizeof(*win
));
866 win
->offset
= (offset
/ window_align
) * window_align
;
867 len
= p
->pack_size
- win
->offset
;
868 if (len
> packed_git_window_size
)
869 len
= packed_git_window_size
;
870 win
->len
= (size_t)len
;
871 pack_mapped
+= win
->len
;
872 while (packed_git_limit
< pack_mapped
873 && unuse_one_window(p
, p
->pack_fd
))
875 win
->base
= xmmap(NULL
, win
->len
,
876 PROT_READ
, MAP_PRIVATE
,
877 p
->pack_fd
, win
->offset
);
878 if (win
->base
== MAP_FAILED
)
879 die("packfile %s cannot be mapped: %s",
882 if (!win
->offset
&& win
->len
== p
->pack_size
883 && !p
->do_not_close
) {
890 if (pack_mapped
> peak_pack_mapped
)
891 peak_pack_mapped
= pack_mapped
;
892 if (pack_open_windows
> peak_pack_open_windows
)
893 peak_pack_open_windows
= pack_open_windows
;
894 win
->next
= p
->windows
;
898 if (win
!= *w_cursor
) {
899 win
->last_used
= pack_used_ctr
++;
903 offset
-= win
->offset
;
905 *left
= win
->len
- xsize_t(offset
);
906 return win
->base
+ offset
;
909 static struct packed_git
*alloc_packed_git(int extra
)
911 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
912 memset(p
, 0, sizeof(*p
));
917 static void try_to_free_pack_memory(size_t size
)
919 release_pack_memory(size
, -1);
922 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
924 static int have_set_try_to_free_routine
;
926 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
928 if (!have_set_try_to_free_routine
) {
929 have_set_try_to_free_routine
= 1;
930 set_try_to_free_routine(try_to_free_pack_memory
);
934 * Make sure a corresponding .pack file exists and that
935 * the index looks sane.
937 path_len
-= strlen(".idx");
942 memcpy(p
->pack_name
, path
, path_len
);
944 strcpy(p
->pack_name
+ path_len
, ".keep");
945 if (!access(p
->pack_name
, F_OK
))
948 strcpy(p
->pack_name
+ path_len
, ".pack");
949 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
954 /* ok, it looks sane as far as we can check without
955 * actually mapping the pack file.
957 p
->pack_size
= st
.st_size
;
958 p
->pack_local
= local
;
959 p
->mtime
= st
.st_mtime
;
960 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
965 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
967 const char *path
= sha1_pack_name(sha1
);
968 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
970 strcpy(p
->pack_name
, path
);
971 hashcpy(p
->sha1
, sha1
);
972 if (check_packed_git_idx(idx_path
, p
)) {
980 void install_packed_git(struct packed_git
*pack
)
982 if (pack
->pack_fd
!= -1)
985 pack
->next
= packed_git
;
989 static void prepare_packed_git_one(char *objdir
, int local
)
991 /* Ensure that this buffer is large enough so that we can
992 append "/pack/" without clobbering the stack even if
993 strlen(objdir) were PATH_MAX. */
994 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
999 sprintf(path
, "%s/pack", objdir
);
1001 dir
= opendir(path
);
1003 if (errno
!= ENOENT
)
1004 error("unable to open object pack directory: %s: %s",
1005 path
, strerror(errno
));
1009 while ((de
= readdir(dir
)) != NULL
) {
1010 int namelen
= strlen(de
->d_name
);
1011 struct packed_git
*p
;
1013 if (!has_extension(de
->d_name
, ".idx"))
1016 if (len
+ namelen
+ 1 > sizeof(path
))
1019 /* Don't reopen a pack we already have. */
1020 strcpy(path
+ len
, de
->d_name
);
1021 for (p
= packed_git
; p
; p
= p
->next
) {
1022 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1027 /* See if it really is a valid .idx file with corresponding
1028 * .pack file that we can map.
1030 p
= add_packed_git(path
, len
+ namelen
, local
);
1033 install_packed_git(p
);
1038 static int sort_pack(const void *a_
, const void *b_
)
1040 struct packed_git
*a
= *((struct packed_git
**)a_
);
1041 struct packed_git
*b
= *((struct packed_git
**)b_
);
1045 * Local packs tend to contain objects specific to our
1046 * variant of the project than remote ones. In addition,
1047 * remote ones could be on a network mounted filesystem.
1048 * Favor local ones for these reasons.
1050 st
= a
->pack_local
- b
->pack_local
;
1055 * Younger packs tend to contain more recent objects,
1056 * and more recent objects tend to get accessed more
1059 if (a
->mtime
< b
->mtime
)
1061 else if (a
->mtime
== b
->mtime
)
1066 static void rearrange_packed_git(void)
1068 struct packed_git
**ary
, *p
;
1071 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1076 /* prepare an array of packed_git for easier sorting */
1077 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1078 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1081 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1083 /* link them back again */
1084 for (i
= 0; i
< n
- 1; i
++)
1085 ary
[i
]->next
= ary
[i
+ 1];
1086 ary
[n
- 1]->next
= NULL
;
1087 packed_git
= ary
[0];
1092 static int prepare_packed_git_run_once
= 0;
1093 void prepare_packed_git(void)
1095 struct alternate_object_database
*alt
;
1097 if (prepare_packed_git_run_once
)
1099 prepare_packed_git_one(get_object_directory(), 1);
1101 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1103 prepare_packed_git_one(alt
->base
, 0);
1104 alt
->name
[-1] = '/';
1106 rearrange_packed_git();
1107 prepare_packed_git_run_once
= 1;
1110 void reprepare_packed_git(void)
1113 prepare_packed_git_run_once
= 0;
1114 prepare_packed_git();
1117 static void mark_bad_packed_object(struct packed_git
*p
,
1118 const unsigned char *sha1
)
1121 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1122 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1124 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1125 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1126 p
->num_bad_objects
++;
1129 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1131 struct packed_git
*p
;
1134 for (p
= packed_git
; p
; p
= p
->next
)
1135 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1136 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1141 int check_sha1_signature(const unsigned char *sha1
, void *map
, unsigned long size
, const char *type
)
1143 unsigned char real_sha1
[20];
1144 hash_sha1_file(map
, size
, type
, real_sha1
);
1145 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1148 static int git_open_noatime(const char *name
, struct packed_git
*p
)
1150 static int sha1_file_open_flag
= O_NOATIME
;
1153 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1157 /* Might the failure be due to O_NOATIME? */
1158 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1159 sha1_file_open_flag
= 0;
1167 static int open_sha1_file(const unsigned char *sha1
)
1170 char *name
= sha1_file_name(sha1
);
1171 struct alternate_object_database
*alt
;
1173 fd
= git_open_noatime(name
, NULL
);
1179 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1181 fill_sha1_path(name
, sha1
);
1182 fd
= git_open_noatime(alt
->base
, NULL
);
1189 static void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1194 fd
= open_sha1_file(sha1
);
1199 if (!fstat(fd
, &st
)) {
1200 *size
= xsize_t(st
.st_size
);
1201 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1208 static int legacy_loose_object(unsigned char *map
)
1213 * Is it a zlib-compressed buffer? If so, the first byte
1214 * must be 0x78 (15-bit window size, deflated), and the
1215 * first 16-bit word is evenly divisible by 31
1217 word
= (map
[0] << 8) + map
[1];
1218 if (map
[0] == 0x78 && !(word
% 31))
1224 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1225 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1228 unsigned long size
, c
;
1229 unsigned long used
= 0;
1232 *type
= (c
>> 4) & 7;
1236 if (len
<= used
|| bitsizeof(long) <= shift
) {
1237 error("bad object header");
1241 size
+= (c
& 0x7f) << shift
;
1248 static int unpack_sha1_header(z_stream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1250 unsigned long size
, used
;
1251 static const char valid_loose_object_type
[8] = {
1253 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1254 0, /* "delta" and others are invalid in a loose object */
1256 enum object_type type
;
1258 /* Get the data stream */
1259 memset(stream
, 0, sizeof(*stream
));
1260 stream
->next_in
= map
;
1261 stream
->avail_in
= mapsize
;
1262 stream
->next_out
= buffer
;
1263 stream
->avail_out
= bufsiz
;
1265 if (legacy_loose_object(map
)) {
1266 git_inflate_init(stream
);
1267 return git_inflate(stream
, 0);
1272 * There used to be a second loose object header format which
1273 * was meant to mimic the in-pack format, allowing for direct
1274 * copy of the object data. This format turned up not to be
1275 * really worth it and we don't write it any longer. But we
1276 * can still read it.
1278 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1279 if (!used
|| !valid_loose_object_type
[type
])
1284 /* Set up the stream for the rest.. */
1285 stream
->next_in
= map
;
1286 stream
->avail_in
= mapsize
;
1287 git_inflate_init(stream
);
1289 /* And generate the fake traditional header */
1290 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1291 typename(type
), size
);
1295 static void *unpack_sha1_rest(z_stream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1297 int bytes
= strlen(buffer
) + 1;
1298 unsigned char *buf
= xmallocz(size
);
1302 n
= stream
->total_out
- bytes
;
1305 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1307 if (bytes
<= size
) {
1309 * The above condition must be (bytes <= size), not
1310 * (bytes < size). In other words, even though we
1311 * expect no more output and set avail_out to zer0,
1312 * the input zlib stream may have bytes that express
1313 * "this concludes the stream", and we *do* want to
1316 * Otherwise we would not be able to test that we
1317 * consumed all the input to reach the expected size;
1318 * we also want to check that zlib tells us that all
1319 * went well with status == Z_STREAM_END at the end.
1321 stream
->next_out
= buf
+ bytes
;
1322 stream
->avail_out
= size
- bytes
;
1323 while (status
== Z_OK
)
1324 status
= git_inflate(stream
, Z_FINISH
);
1326 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1327 git_inflate_end(stream
);
1332 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1333 else if (stream
->avail_in
)
1334 error("garbage at end of loose object '%s'",
1341 * We used to just use "sscanf()", but that's actually way
1342 * too permissive for what we want to check. So do an anal
1343 * object header parse by hand.
1345 static int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1352 * The type can be at most ten bytes (including the
1353 * terminating '\0' that we add), and is followed by
1362 if (i
>= sizeof(type
))
1368 * The length must follow immediately, and be in canonical
1369 * decimal format (ie "010" is not valid).
1371 size
= *hdr
++ - '0';
1376 unsigned long c
= *hdr
- '0';
1380 size
= size
* 10 + c
;
1386 * The length must be followed by a zero byte
1388 return *hdr
? -1 : type_from_string(type
);
1391 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1397 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1398 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1401 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1404 unsigned long get_size_from_delta(struct packed_git
*p
,
1405 struct pack_window
**w_curs
,
1408 const unsigned char *data
;
1409 unsigned char delta_head
[20], *in
;
1413 memset(&stream
, 0, sizeof(stream
));
1414 stream
.next_out
= delta_head
;
1415 stream
.avail_out
= sizeof(delta_head
);
1417 git_inflate_init(&stream
);
1419 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1420 stream
.next_in
= in
;
1421 st
= git_inflate(&stream
, Z_FINISH
);
1422 curpos
+= stream
.next_in
- in
;
1423 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1424 stream
.total_out
< sizeof(delta_head
));
1425 git_inflate_end(&stream
);
1426 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1427 error("delta data unpack-initial failed");
1431 /* Examine the initial part of the delta to figure out
1436 /* ignore base size */
1437 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1439 /* Read the result size */
1440 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1443 static off_t
get_delta_base(struct packed_git
*p
,
1444 struct pack_window
**w_curs
,
1446 enum object_type type
,
1447 off_t delta_obj_offset
)
1449 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1452 /* use_pack() assured us we have [base_info, base_info + 20)
1453 * as a range that we can look at without walking off the
1454 * end of the mapped window. Its actually the hash size
1455 * that is assured. An OFS_DELTA longer than the hash size
1456 * is stupid, as then a REF_DELTA would be smaller to store.
1458 if (type
== OBJ_OFS_DELTA
) {
1460 unsigned char c
= base_info
[used
++];
1461 base_offset
= c
& 127;
1464 if (!base_offset
|| MSB(base_offset
, 7))
1465 return 0; /* overflow */
1466 c
= base_info
[used
++];
1467 base_offset
= (base_offset
<< 7) + (c
& 127);
1469 base_offset
= delta_obj_offset
- base_offset
;
1470 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1471 return 0; /* out of bound */
1473 } else if (type
== OBJ_REF_DELTA
) {
1474 /* The base entry _must_ be in the same pack */
1475 base_offset
= find_pack_entry_one(base_info
, p
);
1478 die("I am totally screwed");
1482 /* forward declaration for a mutually recursive function */
1483 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1484 unsigned long *sizep
);
1486 static int packed_delta_info(struct packed_git
*p
,
1487 struct pack_window
**w_curs
,
1489 enum object_type type
,
1491 unsigned long *sizep
)
1495 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1498 type
= packed_object_info(p
, base_offset
, NULL
);
1499 if (type
<= OBJ_NONE
) {
1500 struct revindex_entry
*revidx
;
1501 const unsigned char *base_sha1
;
1502 revidx
= find_pack_revindex(p
, base_offset
);
1505 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1506 mark_bad_packed_object(p
, base_sha1
);
1507 type
= sha1_object_info(base_sha1
, NULL
);
1508 if (type
<= OBJ_NONE
)
1512 /* We choose to only get the type of the base object and
1513 * ignore potentially corrupt pack file that expects the delta
1514 * based on a base with a wrong size. This saves tons of
1518 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1526 static int unpack_object_header(struct packed_git
*p
,
1527 struct pack_window
**w_curs
,
1529 unsigned long *sizep
)
1531 unsigned char *base
;
1534 enum object_type type
;
1536 /* use_pack() assures us we have [base, base + 20) available
1537 * as a range that we can look at at. (Its actually the hash
1538 * size that is assured.) With our object header encoding
1539 * the maximum deflated object size is 2^137, which is just
1540 * insane, so we know won't exceed what we have been given.
1542 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1543 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1552 const char *packed_object_info_detail(struct packed_git
*p
,
1554 unsigned long *size
,
1555 unsigned long *store_size
,
1556 unsigned int *delta_chain_length
,
1557 unsigned char *base_sha1
)
1559 struct pack_window
*w_curs
= NULL
;
1561 unsigned long dummy
;
1562 unsigned char *next_sha1
;
1563 enum object_type type
;
1564 struct revindex_entry
*revidx
;
1566 *delta_chain_length
= 0;
1567 curpos
= obj_offset
;
1568 type
= unpack_object_header(p
, &w_curs
, &curpos
, size
);
1570 revidx
= find_pack_revindex(p
, obj_offset
);
1571 *store_size
= revidx
[1].offset
- obj_offset
;
1576 die("pack %s contains unknown object type %d",
1577 p
->pack_name
, type
);
1582 unuse_pack(&w_curs
);
1583 return typename(type
);
1585 obj_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
1587 die("pack %s contains bad delta base reference of type %s",
1588 p
->pack_name
, typename(type
));
1589 if (*delta_chain_length
== 0) {
1590 revidx
= find_pack_revindex(p
, obj_offset
);
1591 hashcpy(base_sha1
, nth_packed_object_sha1(p
, revidx
->nr
));
1595 next_sha1
= use_pack(p
, &w_curs
, curpos
, NULL
);
1596 if (*delta_chain_length
== 0)
1597 hashcpy(base_sha1
, next_sha1
);
1598 obj_offset
= find_pack_entry_one(next_sha1
, p
);
1601 (*delta_chain_length
)++;
1602 curpos
= obj_offset
;
1603 type
= unpack_object_header(p
, &w_curs
, &curpos
, &dummy
);
1607 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1608 unsigned long *sizep
)
1610 struct pack_window
*w_curs
= NULL
;
1612 off_t curpos
= obj_offset
;
1613 enum object_type type
;
1615 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1620 type
= packed_delta_info(p
, &w_curs
, curpos
,
1621 type
, obj_offset
, sizep
);
1631 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1632 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1635 unuse_pack(&w_curs
);
1639 static void *unpack_compressed_entry(struct packed_git
*p
,
1640 struct pack_window
**w_curs
,
1646 unsigned char *buffer
, *in
;
1648 buffer
= xmallocz(size
);
1649 memset(&stream
, 0, sizeof(stream
));
1650 stream
.next_out
= buffer
;
1651 stream
.avail_out
= size
+ 1;
1653 git_inflate_init(&stream
);
1655 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1656 stream
.next_in
= in
;
1657 st
= git_inflate(&stream
, Z_FINISH
);
1658 if (!stream
.avail_out
)
1659 break; /* the payload is larger than it should be */
1660 curpos
+= stream
.next_in
- in
;
1661 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1662 git_inflate_end(&stream
);
1663 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1671 static size_t delta_base_cached
;
1672 static size_t delta_base_hash_mask
;
1673 static size_t delta_base_cache_count
;
1675 struct delta_base_cache_entry
{
1676 struct delta_base_cache_entry
*hash_next
;
1677 struct delta_base_cache_entry
*hash_prev
;
1678 struct delta_base_cache_entry
*lru_prev
;
1679 struct delta_base_cache_entry
*lru_next
;
1681 struct packed_git
*p
;
1684 enum object_type type
;
1687 static struct delta_base_cache_entry
*initial_empty_delta_base_cache
;
1688 static struct delta_base_cache_entry
**delta_base_cache
= &initial_empty_delta_base_cache
;
1689 static struct delta_base_cache_entry
*free_delta_base_cache_entry
;
1691 struct delta_base_cache_entry delta_base_cache_lru
= {
1692 .lru_next
= &delta_base_cache_lru
,
1693 .lru_prev
= &delta_base_cache_lru
,
1696 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1700 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1701 hash
+= (hash
>> 8) + (hash
>> 16);
1702 return hash
& delta_base_hash_mask
;
1705 static struct delta_base_cache_entry
*find_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1707 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1708 struct delta_base_cache_entry
*ent
= delta_base_cache
[hash
];
1709 for (ent
= delta_base_cache
[hash
]; ent
; ent
= ent
->hash_next
) {
1710 if (ent
->p
== p
&& ent
->base_offset
== base_offset
)
1716 static void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1719 ent
->lru_next
->lru_prev
= ent
->lru_prev
;
1720 ent
->lru_prev
->lru_next
= ent
->lru_next
;
1722 ent
->hash_next
->hash_prev
= ent
->hash_prev
;
1723 ent
->hash_prev
->hash_next
= ent
->hash_next
;
1724 delta_base_cached
-= ent
->size
;
1725 delta_base_cache_count
--;
1726 ent
->hash_next
= free_delta_base_cache_entry
;
1727 free_delta_base_cache_entry
= ent
;
1730 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1731 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1734 struct delta_base_cache_entry
*ent
= find_delta_base_cache(p
, base_offset
);
1737 return unpack_entry(p
, base_offset
, type
, base_size
);
1741 ent
->data
= NULL
; // Move the data to caller instead of freeing
1742 release_delta_base_cache(ent
);
1744 ret
= xmemdupz(ent
->data
, ent
->size
);
1747 *base_size
= ent
->size
;
1751 static void limit_delta_base_cache(void)
1753 while (delta_base_cached
> delta_base_cache_limit
&&
1754 delta_base_cache_lru
.lru_next
!= &delta_base_cache_lru
) {
1755 int random
= delta_base_cache_lru
.lru_next
->base_offset
;
1756 release_delta_base_cache(delta_base_cache_lru
.lru_next
);
1757 delta_base_cache_lru
.lru_next
->lru_prev
= delta_base_cache_lru
.lru_prev
;
1758 delta_base_cache_lru
.lru_prev
->lru_next
= delta_base_cache_lru
.lru_next
;
1759 struct delta_base_cache_entry
*skip
= delta_base_cache_lru
.lru_next
;
1761 skip
= skip
->lru_next
;
1762 delta_base_cache_lru
.lru_prev
= skip
;
1763 delta_base_cache_lru
.lru_next
= skip
->lru_next
;
1764 delta_base_cache_lru
.lru_prev
->lru_next
= &delta_base_cache_lru
;
1765 delta_base_cache_lru
.lru_next
->lru_prev
= &delta_base_cache_lru
;
1769 void clear_delta_base_cache(void)
1771 while (delta_base_cache_lru
.lru_next
!= &delta_base_cache_lru
)
1772 release_delta_base_cache(delta_base_cache_lru
.lru_next
);
1775 static void insert_delta_base_cache_hash(struct delta_base_cache_entry
*ent
)
1777 unsigned long hash
= pack_entry_hash(ent
->p
, ent
->base_offset
);
1778 ent
->hash_prev
= delta_base_cache
+ hash
;
1779 ent
->hash_next
= delta_base_cache
[hash
];
1781 ent
->hash_next
->hash_prev
= ent
;
1782 delta_base_cache
[hash
] = ent
;
1783 delta_base_cache_count
++;
1786 static void grow_delta_base_cache(void)
1788 if (delta_base_cache_count
* 2 < delta_base_hash_mask
)
1790 if (delta_base_cache
!= &initial_empty_delta_base_cache
)
1791 free(delta_base_cache
);
1792 delta_base_cache_count
= 0;
1793 delta_base_hash_mask
= delta_base_hash_mask
* 2 + 1;
1794 delta_base_cache
= xcalloc(delta_base_hash_mask
+ 1, sizeof(struct delta_base_cache_entry
*));
1795 struct delta_base_cache_entry
*lru
;
1796 for (lru
= delta_base_cache_lru
.lru_prev
; lru
!= &delta_base_cache_lru
; lru
= lru
->lru_prev
)
1797 insert_delta_base_cache_hash(lru
);
1800 static struct delta_base_cache_entry
*alloc_delta_base_cache_entry(void)
1802 struct delta_base_cache_entry
*ent
= free_delta_base_cache_entry
;
1804 free_delta_base_cache_entry
= ent
->hash_next
;
1807 ent
= xmalloc(sizeof(*ent
) * 50);
1809 for (i
= 1; i
< 49; i
++)
1810 ent
[i
].hash_next
= ent
+ i
+ 1;
1811 ent
[49].hash_next
= NULL
;
1812 free_delta_base_cache_entry
= ent
+ 1;
1816 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1817 void *base
, unsigned long base_size
, enum object_type type
)
1819 delta_base_cached
+= base_size
;
1820 limit_delta_base_cache();
1821 grow_delta_base_cache();
1822 struct delta_base_cache_entry
*ent
= alloc_delta_base_cache_entry();
1824 ent
->base_offset
= base_offset
;
1827 ent
->size
= base_size
;
1828 insert_delta_base_cache_hash(ent
);
1829 ent
->lru_next
= &delta_base_cache_lru
;
1830 ent
->lru_prev
= delta_base_cache_lru
.lru_prev
;
1831 delta_base_cache_lru
.lru_prev
->lru_next
= ent
;
1832 delta_base_cache_lru
.lru_prev
= ent
;
1835 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1836 unsigned long *size
);
1838 static void *unpack_delta_entry(struct packed_git
*p
,
1839 struct pack_window
**w_curs
,
1841 unsigned long delta_size
,
1843 enum object_type
*type
,
1844 unsigned long *sizep
)
1846 void *delta_data
, *result
, *base
;
1847 unsigned long base_size
;
1850 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1852 error("failed to validate delta base reference "
1853 "at offset %"PRIuMAX
" from %s",
1854 (uintmax_t)curpos
, p
->pack_name
);
1858 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1861 * We're probably in deep shit, but let's try to fetch
1862 * the required base anyway from another pack or loose.
1863 * This is costly but should happen only in the presence
1864 * of a corrupted pack, and is better than failing outright.
1866 struct revindex_entry
*revidx
;
1867 const unsigned char *base_sha1
;
1868 revidx
= find_pack_revindex(p
, base_offset
);
1871 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1872 error("failed to read delta base object %s"
1873 " at offset %"PRIuMAX
" from %s",
1874 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1876 mark_bad_packed_object(p
, base_sha1
);
1877 base
= read_object(base_sha1
, type
, &base_size
);
1882 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1884 error("failed to unpack compressed delta "
1885 "at offset %"PRIuMAX
" from %s",
1886 (uintmax_t)curpos
, p
->pack_name
);
1890 result
= patch_delta(base
, base_size
,
1891 delta_data
, delta_size
,
1894 die("failed to apply delta");
1896 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1900 int do_check_packed_object_crc
;
1902 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1903 enum object_type
*type
, unsigned long *sizep
)
1905 struct pack_window
*w_curs
= NULL
;
1906 off_t curpos
= obj_offset
;
1909 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1910 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1911 unsigned long len
= revidx
[1].offset
- obj_offset
;
1912 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1913 const unsigned char *sha1
=
1914 nth_packed_object_sha1(p
, revidx
->nr
);
1915 error("bad packed object CRC for %s",
1917 mark_bad_packed_object(p
, sha1
);
1918 unuse_pack(&w_curs
);
1923 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1927 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1928 obj_offset
, type
, sizep
);
1934 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1938 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1939 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1941 unuse_pack(&w_curs
);
1945 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1948 const unsigned char *index
= p
->index_data
;
1950 if (open_pack_index(p
))
1952 index
= p
->index_data
;
1954 if (n
>= p
->num_objects
)
1957 if (p
->index_version
== 1) {
1958 return index
+ 24 * n
+ 4;
1961 return index
+ 20 * n
;
1965 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1967 const unsigned char *index
= p
->index_data
;
1969 if (p
->index_version
== 1) {
1970 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1973 index
+= 8 + p
->num_objects
* (20 + 4);
1974 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1975 if (!(off
& 0x80000000))
1977 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1978 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1979 ntohl(*((uint32_t *)(index
+ 4)));
1983 off_t
find_pack_entry_one(const unsigned char *sha1
,
1984 struct packed_git
*p
)
1986 const uint32_t *level1_ofs
= p
->index_data
;
1987 const unsigned char *index
= p
->index_data
;
1988 unsigned hi
, lo
, stride
;
1989 static int use_lookup
= -1;
1990 static int debug_lookup
= -1;
1992 if (debug_lookup
< 0)
1993 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
1996 if (open_pack_index(p
))
1998 level1_ofs
= p
->index_data
;
1999 index
= p
->index_data
;
2001 if (p
->index_version
> 1) {
2006 hi
= ntohl(level1_ofs
[*sha1
]);
2007 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2008 if (p
->index_version
> 1) {
2016 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2017 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2020 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2022 int pos
= sha1_entry_pos(index
, stride
, 0,
2023 lo
, hi
, p
->num_objects
, sha1
);
2026 return nth_packed_object_offset(p
, pos
);
2030 unsigned mi
= (lo
+ hi
) / 2;
2031 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2034 printf("lo %u hi %u rg %u mi %u\n",
2035 lo
, hi
, hi
- lo
, mi
);
2037 return nth_packed_object_offset(p
, mi
);
2046 static int is_pack_valid(struct packed_git
*p
)
2048 /* An already open pack is known to be valid. */
2049 if (p
->pack_fd
!= -1)
2052 /* If the pack has one window completely covering the
2053 * file size, the pack is known to be valid even if
2054 * the descriptor is not currently open.
2057 struct pack_window
*w
= p
->windows
;
2059 if (!w
->offset
&& w
->len
== p
->pack_size
)
2063 /* Force the pack to open to prove its valid. */
2064 return !open_packed_git(p
);
2067 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2069 static struct packed_git
*last_found
= (void *)1;
2070 struct packed_git
*p
;
2073 prepare_packed_git();
2076 p
= (last_found
== (void *)1) ? packed_git
: last_found
;
2079 if (p
->num_bad_objects
) {
2081 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2082 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2086 offset
= find_pack_entry_one(sha1
, p
);
2089 * We are about to tell the caller where they can
2090 * locate the requested object. We better make
2091 * sure the packfile is still here and can be
2092 * accessed before supplying that answer, as
2093 * it may have been deleted since the index
2096 if (!is_pack_valid(p
)) {
2097 error("packfile %s cannot be accessed", p
->pack_name
);
2102 hashcpy(e
->sha1
, sha1
);
2108 if (p
== last_found
)
2112 if (p
== last_found
)
2118 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2119 struct packed_git
*packs
)
2121 struct packed_git
*p
;
2123 for (p
= packs
; p
; p
= p
->next
) {
2124 if (find_pack_entry_one(sha1
, p
))
2131 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2134 unsigned long mapsize
, size
;
2139 map
= map_sha1_file(sha1
, &mapsize
);
2141 return error("unable to find %s", sha1_to_hex(sha1
));
2142 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2143 status
= error("unable to unpack %s header",
2145 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2146 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2149 git_inflate_end(&stream
);
2150 munmap(map
, mapsize
);
2154 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2156 struct cached_object
*co
;
2157 struct pack_entry e
;
2160 co
= find_cached_object(sha1
);
2167 if (!find_pack_entry(sha1
, &e
)) {
2168 /* Most likely it's a loose object. */
2169 status
= sha1_loose_object_info(sha1
, sizep
);
2173 /* Not a loose object; someone else may have just packed it. */
2174 reprepare_packed_git();
2175 if (!find_pack_entry(sha1
, &e
))
2179 status
= packed_object_info(e
.p
, e
.offset
, sizep
);
2181 mark_bad_packed_object(e
.p
, sha1
);
2182 status
= sha1_object_info(sha1
, sizep
);
2188 static void *read_packed_sha1(const unsigned char *sha1
,
2189 enum object_type
*type
, unsigned long *size
)
2191 struct pack_entry e
;
2194 if (!find_pack_entry(sha1
, &e
))
2196 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2199 * We're probably in deep shit, but let's try to fetch
2200 * the required object anyway from another pack or loose.
2201 * This should happen only in the presence of a corrupted
2202 * pack, and is better than failing outright.
2204 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2205 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2206 mark_bad_packed_object(e
.p
, sha1
);
2207 data
= read_object(sha1
, type
, size
);
2212 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2213 unsigned char *sha1
)
2215 struct cached_object
*co
;
2217 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2218 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2220 if (cached_object_alloc
<= cached_object_nr
) {
2221 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2222 cached_objects
= xrealloc(cached_objects
,
2223 sizeof(*cached_objects
) *
2224 cached_object_alloc
);
2226 co
= &cached_objects
[cached_object_nr
++];
2229 co
->buf
= xmalloc(len
);
2230 memcpy(co
->buf
, buf
, len
);
2231 hashcpy(co
->sha1
, sha1
);
2235 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2236 unsigned long *size
)
2238 unsigned long mapsize
;
2240 struct cached_object
*co
;
2242 co
= find_cached_object(sha1
);
2246 return xmemdupz(co
->buf
, co
->size
);
2249 buf
= read_packed_sha1(sha1
, type
, size
);
2252 map
= map_sha1_file(sha1
, &mapsize
);
2254 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2255 munmap(map
, mapsize
);
2258 reprepare_packed_git();
2259 return read_packed_sha1(sha1
, type
, size
);
2263 * This function dies on corrupt objects; the callers who want to
2264 * deal with them should arrange to call read_object() and give error
2265 * messages themselves.
2267 void *read_sha1_file_repl(const unsigned char *sha1
,
2268 enum object_type
*type
,
2269 unsigned long *size
,
2270 const unsigned char **replacement
)
2272 const unsigned char *repl
= lookup_replace_object(sha1
);
2275 const struct packed_git
*p
;
2278 data
= read_object(repl
, type
, size
);
2281 *replacement
= repl
;
2285 if (errno
&& errno
!= ENOENT
)
2286 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2288 /* die if we replaced an object with one that does not exist */
2290 die("replacement %s not found for %s",
2291 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2293 if (has_loose_object(repl
)) {
2294 path
= sha1_file_name(sha1
);
2295 die("loose object %s (stored in %s) is corrupt",
2296 sha1_to_hex(repl
), path
);
2299 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2300 die("packed object %s (stored in %s) is corrupt",
2301 sha1_to_hex(repl
), p
->pack_name
);
2306 void *read_object_with_reference(const unsigned char *sha1
,
2307 const char *required_type_name
,
2308 unsigned long *size
,
2309 unsigned char *actual_sha1_return
)
2311 enum object_type type
, required_type
;
2313 unsigned long isize
;
2314 unsigned char actual_sha1
[20];
2316 required_type
= type_from_string(required_type_name
);
2317 hashcpy(actual_sha1
, sha1
);
2319 int ref_length
= -1;
2320 const char *ref_type
= NULL
;
2322 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2325 if (type
== required_type
) {
2327 if (actual_sha1_return
)
2328 hashcpy(actual_sha1_return
, actual_sha1
);
2331 /* Handle references */
2332 else if (type
== OBJ_COMMIT
)
2334 else if (type
== OBJ_TAG
)
2335 ref_type
= "object ";
2340 ref_length
= strlen(ref_type
);
2342 if (ref_length
+ 40 > isize
||
2343 memcmp(buffer
, ref_type
, ref_length
) ||
2344 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2349 /* Now we have the ID of the referred-to object in
2350 * actual_sha1. Check again. */
2354 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2355 const char *type
, unsigned char *sha1
,
2356 char *hdr
, int *hdrlen
)
2360 /* Generate the header */
2361 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2365 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2366 git_SHA1_Update(&c
, buf
, len
);
2367 git_SHA1_Final(sha1
, &c
);
2371 * Move the just written object into its final resting place.
2372 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2373 * "moving" is only a part of what it does, when no patch between
2374 * master to pu changes the call sites of this function.
2376 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2380 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2382 else if (link(tmpfile
, filename
))
2386 * Coda hack - coda doesn't like cross-directory links,
2387 * so we fall back to a rename, which will mean that it
2388 * won't be able to check collisions, but that's not a
2391 * The same holds for FAT formatted media.
2393 * When this succeeds, we just return. We have nothing
2396 if (ret
&& ret
!= EEXIST
) {
2398 if (!rename(tmpfile
, filename
))
2402 unlink_or_warn(tmpfile
);
2404 if (ret
!= EEXIST
) {
2405 return error("unable to write sha1 filename %s: %s\n", filename
, strerror(ret
));
2407 /* FIXME!!! Collision check here ? */
2411 if (adjust_shared_perm(filename
))
2412 return error("unable to set permission to '%s'", filename
);
2416 static int write_buffer(int fd
, const void *buf
, size_t len
)
2418 if (write_in_full(fd
, buf
, len
) < 0)
2419 return error("file write error (%s)", strerror(errno
));
2423 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2424 unsigned char *sha1
)
2428 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2432 /* Finalize a file on disk, and close it. */
2433 static void close_sha1_file(int fd
)
2435 if (fsync_object_files
)
2436 fsync_or_die(fd
, "sha1 file");
2438 die_errno("error when closing sha1 file");
2441 /* Size of directory component, including the ending '/' */
2442 static inline int directory_size(const char *filename
)
2444 const char *s
= strrchr(filename
, '/');
2447 return s
- filename
+ 1;
2451 * This creates a temporary file in the same directory as the final
2454 * We want to avoid cross-directory filename renames, because those
2455 * can have problems on various filesystems (FAT, NFS, Coda).
2457 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2459 int fd
, dirlen
= directory_size(filename
);
2461 if (dirlen
+ 20 > bufsiz
) {
2462 errno
= ENAMETOOLONG
;
2465 memcpy(buffer
, filename
, dirlen
);
2466 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2467 fd
= git_mkstemp_mode(buffer
, 0444);
2468 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2469 /* Make sure the directory exists */
2470 memcpy(buffer
, filename
, dirlen
);
2471 buffer
[dirlen
-1] = 0;
2472 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2476 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2477 fd
= git_mkstemp_mode(buffer
, 0444);
2482 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2483 const void *buf
, unsigned long len
, time_t mtime
)
2486 unsigned char compressed
[4096];
2489 unsigned char parano_sha1
[20];
2491 static char tmpfile
[PATH_MAX
];
2493 filename
= sha1_file_name(sha1
);
2494 fd
= create_tmpfile(tmpfile
, sizeof(tmpfile
), filename
);
2496 if (errno
== EACCES
)
2497 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2499 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile
, strerror(errno
));
2503 memset(&stream
, 0, sizeof(stream
));
2504 deflateInit(&stream
, zlib_compression_level
);
2505 stream
.next_out
= compressed
;
2506 stream
.avail_out
= sizeof(compressed
);
2509 /* First header.. */
2510 stream
.next_in
= (unsigned char *)hdr
;
2511 stream
.avail_in
= hdrlen
;
2512 while (deflate(&stream
, 0) == Z_OK
)
2514 git_SHA1_Update(&c
, hdr
, hdrlen
);
2516 /* Then the data itself.. */
2517 stream
.next_in
= (void *)buf
;
2518 stream
.avail_in
= len
;
2520 unsigned char *in0
= stream
.next_in
;
2521 ret
= deflate(&stream
, Z_FINISH
);
2522 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2523 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2524 die("unable to write sha1 file");
2525 stream
.next_out
= compressed
;
2526 stream
.avail_out
= sizeof(compressed
);
2527 } while (ret
== Z_OK
);
2529 if (ret
!= Z_STREAM_END
)
2530 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2531 ret
= deflateEnd(&stream
);
2533 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2534 git_SHA1_Final(parano_sha1
, &c
);
2535 if (hashcmp(sha1
, parano_sha1
) != 0)
2536 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2538 close_sha1_file(fd
);
2543 utb
.modtime
= mtime
;
2544 if (utime(tmpfile
, &utb
) < 0)
2545 warning("failed utime() on %s: %s",
2546 tmpfile
, strerror(errno
));
2549 return move_temp_to_file(tmpfile
, filename
);
2552 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2554 unsigned char sha1
[20];
2558 /* Normally if we have it in the pack then we do not bother writing
2559 * it out into .git/objects/??/?{38} file.
2561 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2563 hashcpy(returnsha1
, sha1
);
2564 if (has_sha1_file(sha1
))
2566 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2569 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2573 enum object_type type
;
2578 if (has_loose_object(sha1
))
2580 buf
= read_packed_sha1(sha1
, &type
, &len
);
2582 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2583 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2584 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2590 int has_pack_index(const unsigned char *sha1
)
2593 if (stat(sha1_pack_index_name(sha1
), &st
))
2598 int has_sha1_pack(const unsigned char *sha1
)
2600 struct pack_entry e
;
2601 return find_pack_entry(sha1
, &e
);
2604 int has_sha1_file(const unsigned char *sha1
)
2606 struct pack_entry e
;
2608 if (find_pack_entry(sha1
, &e
))
2610 return has_loose_object(sha1
);
2613 static void check_tree(const void *buf
, size_t size
)
2615 struct tree_desc desc
;
2616 struct name_entry entry
;
2618 init_tree_desc(&desc
, buf
, size
);
2619 while (tree_entry(&desc
, &entry
))
2621 * tree_entry() will die() on malformed entries */
2625 static void check_commit(const void *buf
, size_t size
)
2628 memset(&c
, 0, sizeof(c
));
2629 if (parse_commit_buffer(&c
, buf
, size
))
2630 die("corrupt commit");
2633 static void check_tag(const void *buf
, size_t size
)
2636 memset(&t
, 0, sizeof(t
));
2637 if (parse_tag_buffer(&t
, buf
, size
))
2641 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2642 int write_object
, enum object_type type
,
2643 const char *path
, int format_check
)
2645 int ret
, re_allocated
= 0;
2651 * Convert blobs to git internal format
2653 if ((type
== OBJ_BLOB
) && path
) {
2654 struct strbuf nbuf
= STRBUF_INIT
;
2655 if (convert_to_git(path
, buf
, size
, &nbuf
,
2656 write_object
? safe_crlf
: 0)) {
2657 buf
= strbuf_detach(&nbuf
, &size
);
2662 if (type
== OBJ_TREE
)
2663 check_tree(buf
, size
);
2664 if (type
== OBJ_COMMIT
)
2665 check_commit(buf
, size
);
2666 if (type
== OBJ_TAG
)
2667 check_tag(buf
, size
);
2671 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2673 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2679 #define SMALL_FILE_SIZE (32*1024)
2681 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
, int write_object
,
2682 enum object_type type
, const char *path
, int format_check
)
2685 size_t size
= xsize_t(st
->st_size
);
2687 if (!S_ISREG(st
->st_mode
)) {
2688 struct strbuf sbuf
= STRBUF_INIT
;
2689 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2690 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, write_object
,
2691 type
, path
, format_check
);
2694 strbuf_release(&sbuf
);
2696 ret
= index_mem(sha1
, NULL
, size
, write_object
, type
, path
,
2698 } else if (size
<= SMALL_FILE_SIZE
) {
2699 char *buf
= xmalloc(size
);
2700 if (size
== read_in_full(fd
, buf
, size
))
2701 ret
= index_mem(sha1
, buf
, size
, write_object
, type
,
2702 path
, format_check
);
2704 ret
= error("short read %s", strerror(errno
));
2707 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2708 ret
= index_mem(sha1
, buf
, size
, write_object
, type
, path
,
2716 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, int write_object
)
2719 struct strbuf sb
= STRBUF_INIT
;
2721 switch (st
->st_mode
& S_IFMT
) {
2723 fd
= open(path
, O_RDONLY
);
2725 return error("open(\"%s\"): %s", path
,
2727 if (index_fd(sha1
, fd
, st
, write_object
, OBJ_BLOB
, path
, 0) < 0)
2728 return error("%s: failed to insert into database",
2732 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2733 char *errstr
= strerror(errno
);
2734 return error("readlink(\"%s\"): %s", path
,
2738 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2739 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2740 return error("%s: failed to insert into database",
2742 strbuf_release(&sb
);
2745 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2747 return error("%s: unsupported file type", path
);
2752 int read_pack_header(int fd
, struct pack_header
*header
)
2754 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2755 /* "eof before pack header was fully read" */
2756 return PH_ERROR_EOF
;
2758 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2759 /* "protocol error (pack signature mismatch detected)" */
2760 return PH_ERROR_PACK_SIGNATURE
;
2761 if (!pack_version_ok(header
->hdr_version
))
2762 /* "protocol error (pack version unsupported)" */
2763 return PH_ERROR_PROTOCOL
;
2767 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2769 enum object_type type
= sha1_object_info(sha1
, NULL
);
2771 die("%s is not a valid object", sha1_to_hex(sha1
));
2773 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),