2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
14 #include "run-command.h"
17 #include "tree-walk.h"
19 #include "pack-revindex.h"
20 #include "sha1-lookup.h"
21 #include "bulk-checkin.h"
22 #include "streaming.h"
25 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
26 #define O_NOATIME 01000000
32 #define SZ_FMT PRIuMAX
33 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
35 const unsigned char null_sha1
[20];
38 * This is meant to hold a *small* number of objects that you would
39 * want read_sha1_file() to be able to return, but yet you do not want
40 * to write them into the object store (e.g. a browse-only
43 static struct cached_object
{
44 unsigned char sha1
[20];
45 enum object_type type
;
49 static int cached_object_nr
, cached_object_alloc
;
51 static struct cached_object empty_tree
= {
52 EMPTY_TREE_SHA1_BIN_LITERAL
,
58 static struct packed_git
*last_found_pack
;
60 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
63 struct cached_object
*co
= cached_objects
;
65 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
66 if (!hashcmp(co
->sha1
, sha1
))
69 if (!hashcmp(sha1
, empty_tree
.sha1
))
74 int mkdir_in_gitdir(const char *path
)
76 if (mkdir(path
, 0777)) {
77 int saved_errno
= errno
;
79 struct strbuf sb
= STRBUF_INIT
;
84 * Are we looking at a path in a symlinked worktree
85 * whose original repository does not yet have it?
86 * e.g. .git/rr-cache pointing at its original
87 * repository in which the user hasn't performed any
88 * conflict resolution yet?
90 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
91 strbuf_readlink(&sb
, path
, st
.st_size
) ||
92 !is_absolute_path(sb
.buf
) ||
93 mkdir(sb
.buf
, 0777)) {
100 return adjust_shared_perm(path
);
103 int safe_create_leading_directories(char *path
)
105 char *pos
= path
+ offset_1st_component(path
);
109 pos
= strchr(pos
, '/');
112 while (*++pos
== '/')
117 if (!stat(path
, &st
)) {
119 if (!S_ISDIR(st
.st_mode
)) {
124 else if (mkdir(path
, 0777)) {
125 if (errno
== EEXIST
&&
126 !stat(path
, &st
) && S_ISDIR(st
.st_mode
)) {
127 ; /* somebody created it since we checked */
133 else if (adjust_shared_perm(path
)) {
142 int safe_create_leading_directories_const(const char *path
)
144 /* path points to cache entries, so xstrdup before messing with it */
145 char *buf
= xstrdup(path
);
146 int result
= safe_create_leading_directories(buf
);
151 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
154 for (i
= 0; i
< 20; i
++) {
155 static char hex
[] = "0123456789abcdef";
156 unsigned int val
= sha1
[i
];
157 char *pos
= pathbuf
+ i
*2 + (i
> 0);
158 *pos
++ = hex
[val
>> 4];
159 *pos
= hex
[val
& 0xf];
164 * NOTE! This returns a statically allocated buffer, so you have to be
165 * careful about using it. Do an "xstrdup()" if you need to save the
168 * Also note that this returns the location for creating. Reading
169 * SHA1 file can happen from any alternate directory listed in the
170 * DB_ENVIRONMENT environment variable if it is not found in
171 * the primary object database.
173 char *sha1_file_name(const unsigned char *sha1
)
175 static char buf
[PATH_MAX
];
179 objdir
= get_object_directory();
180 len
= strlen(objdir
);
182 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
183 if (len
+ 43 > PATH_MAX
)
184 die("insanely long object directory %s", objdir
);
185 memcpy(buf
, objdir
, len
);
189 fill_sha1_path(buf
+ len
+ 1, sha1
);
193 static char *sha1_get_pack_name(const unsigned char *sha1
,
194 char **name
, char **base
, const char *which
)
196 static const char hex
[] = "0123456789abcdef";
201 const char *sha1_file_directory
= get_object_directory();
202 int len
= strlen(sha1_file_directory
);
203 *base
= xmalloc(len
+ 60);
204 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
205 sha1_file_directory
, which
);
206 *name
= *base
+ len
+ 11;
211 for (i
= 0; i
< 20; i
++) {
212 unsigned int val
= *sha1
++;
213 *buf
++ = hex
[val
>> 4];
214 *buf
++ = hex
[val
& 0xf];
220 char *sha1_pack_name(const unsigned char *sha1
)
222 static char *name
, *base
;
224 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
227 char *sha1_pack_index_name(const unsigned char *sha1
)
229 static char *name
, *base
;
231 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
234 struct alternate_object_database
*alt_odb_list
;
235 static struct alternate_object_database
**alt_odb_tail
;
237 static int git_open_noatime(const char *name
);
240 * Prepare alternate object database registry.
242 * The variable alt_odb_list points at the list of struct
243 * alternate_object_database. The elements on this list come from
244 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
245 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
246 * whose contents is similar to that environment variable but can be
247 * LF separated. Its base points at a statically allocated buffer that
248 * contains "/the/directory/corresponding/to/.git/objects/...", while
249 * its name points just after the slash at the end of ".git/objects/"
250 * in the example above, and has enough space to hold 40-byte hex
251 * SHA1, an extra slash for the first level indirection, and the
254 static int link_alt_odb_entry(const char * entry
, int len
, const char * relative_base
, int depth
)
256 const char *objdir
= get_object_directory();
257 struct alternate_object_database
*ent
;
258 struct alternate_object_database
*alt
;
260 struct strbuf pathbuf
= STRBUF_INIT
;
262 if (!is_absolute_path(entry
) && relative_base
) {
263 strbuf_addstr(&pathbuf
, real_path(relative_base
));
264 strbuf_addch(&pathbuf
, '/');
266 strbuf_add(&pathbuf
, entry
, len
);
268 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
270 pfxlen
= strlen(pathbuf
.buf
);
273 * The trailing slash after the directory name is given by
274 * this function at the end. Remove duplicates.
276 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
279 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
280 ent
= xmalloc(sizeof(*ent
) + entlen
);
281 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
282 strbuf_release(&pathbuf
);
284 ent
->name
= ent
->base
+ pfxlen
+ 1;
285 ent
->base
[pfxlen
+ 3] = '/';
286 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
288 /* Detect cases where alternate disappeared */
289 if (!is_directory(ent
->base
)) {
290 error("object directory %s does not exist; "
291 "check .git/objects/info/alternates.",
297 /* Prevent the common mistake of listing the same
298 * thing twice, or object directory itself.
300 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
301 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
306 if (!strcmp(ent
->base
, objdir
)) {
311 /* add the alternate entry */
313 alt_odb_tail
= &(ent
->next
);
316 /* recursively add alternates */
317 read_info_alternates(ent
->base
, depth
+ 1);
319 ent
->base
[pfxlen
] = '/';
324 static void link_alt_odb_entries(const char *alt
, const char *ep
, int sep
,
325 const char *relative_base
, int depth
)
327 const char *cp
, *last
;
330 error("%s: ignoring alternate object stores, nesting too deep.",
338 if (cp
< ep
&& *cp
== '#') {
339 while (cp
< ep
&& *cp
!= sep
)
344 while (cp
< ep
&& *cp
!= sep
)
347 if (!is_absolute_path(last
) && depth
) {
348 error("%s: ignoring relative alternate object store %s",
349 relative_base
, last
);
351 link_alt_odb_entry(last
, cp
- last
,
352 relative_base
, depth
);
355 while (cp
< ep
&& *cp
== sep
)
361 void read_info_alternates(const char * relative_base
, int depth
)
366 const char alt_file_name
[] = "info/alternates";
367 /* Given that relative_base is no longer than PATH_MAX,
368 ensure that "path" has enough space to append "/", the
369 file name, "info/alternates", and a trailing NUL. */
370 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
373 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
374 fd
= git_open_noatime(path
);
377 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
381 mapsz
= xsize_t(st
.st_size
);
382 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
385 link_alt_odb_entries(map
, map
+ mapsz
, '\n', relative_base
, depth
);
390 void add_to_alternates_file(const char *reference
)
392 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
393 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
394 char *alt
= mkpath("%s\n", reference
);
395 write_or_die(fd
, alt
, strlen(alt
));
396 if (commit_lock_file(lock
))
397 die("could not close alternates file");
399 link_alt_odb_entries(alt
, alt
+ strlen(alt
), '\n', NULL
, 0);
402 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
404 struct alternate_object_database
*ent
;
407 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
412 void prepare_alt_odb(void)
419 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
422 alt_odb_tail
= &alt_odb_list
;
423 link_alt_odb_entries(alt
, alt
+ strlen(alt
), PATH_SEP
, NULL
, 0);
425 read_info_alternates(get_object_directory(), 0);
428 static int has_loose_object_local(const unsigned char *sha1
)
430 char *name
= sha1_file_name(sha1
);
431 return !access(name
, F_OK
);
434 int has_loose_object_nonlocal(const unsigned char *sha1
)
436 struct alternate_object_database
*alt
;
438 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
439 fill_sha1_path(alt
->name
, sha1
);
440 if (!access(alt
->base
, F_OK
))
446 static int has_loose_object(const unsigned char *sha1
)
448 return has_loose_object_local(sha1
) ||
449 has_loose_object_nonlocal(sha1
);
452 static unsigned int pack_used_ctr
;
453 static unsigned int pack_mmap_calls
;
454 static unsigned int peak_pack_open_windows
;
455 static unsigned int pack_open_windows
;
456 static unsigned int pack_open_fds
;
457 static unsigned int pack_max_fds
;
458 static size_t peak_pack_mapped
;
459 static size_t pack_mapped
;
460 struct packed_git
*packed_git
;
462 void pack_report(void)
465 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
466 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
467 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
468 sz_fmt(getpagesize()),
469 sz_fmt(packed_git_window_size
),
470 sz_fmt(packed_git_limit
));
472 "pack_report: pack_used_ctr = %10u\n"
473 "pack_report: pack_mmap_calls = %10u\n"
474 "pack_report: pack_open_windows = %10u / %10u\n"
475 "pack_report: pack_mapped = "
476 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
479 pack_open_windows
, peak_pack_open_windows
,
480 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
483 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
486 struct pack_idx_header
*hdr
;
488 uint32_t version
, nr
, i
, *index
;
489 int fd
= git_open_noatime(path
);
494 if (fstat(fd
, &st
)) {
498 idx_size
= xsize_t(st
.st_size
);
499 if (idx_size
< 4 * 256 + 20 + 20) {
501 return error("index file %s is too small", path
);
503 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
507 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
508 version
= ntohl(hdr
->idx_version
);
509 if (version
< 2 || version
> 2) {
510 munmap(idx_map
, idx_size
);
511 return error("index file %s is version %"PRIu32
512 " and is not supported by this binary"
513 " (try upgrading GIT to a newer version)",
522 index
+= 2; /* skip index header */
523 for (i
= 0; i
< 256; i
++) {
524 uint32_t n
= ntohl(index
[i
]);
526 munmap(idx_map
, idx_size
);
527 return error("non-monotonic index %s", path
);
535 * - 256 index entries 4 bytes each
536 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
537 * - 20-byte SHA1 of the packfile
538 * - 20-byte SHA1 file checksum
540 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
541 munmap(idx_map
, idx_size
);
542 return error("wrong index v1 file size in %s", path
);
544 } else if (version
== 2) {
547 * - 8 bytes of header
548 * - 256 index entries 4 bytes each
549 * - 20-byte sha1 entry * nr
550 * - 4-byte crc entry * nr
551 * - 4-byte offset entry * nr
552 * - 20-byte SHA1 of the packfile
553 * - 20-byte SHA1 file checksum
554 * And after the 4-byte offset table might be a
555 * variable sized table containing 8-byte entries
556 * for offsets larger than 2^31.
558 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
559 unsigned long max_size
= min_size
;
561 max_size
+= (nr
- 1)*8;
562 if (idx_size
< min_size
|| idx_size
> max_size
) {
563 munmap(idx_map
, idx_size
);
564 return error("wrong index v2 file size in %s", path
);
566 if (idx_size
!= min_size
&&
568 * make sure we can deal with large pack offsets.
569 * 31-bit signed offset won't be enough, neither
570 * 32-bit unsigned one will be.
572 (sizeof(off_t
) <= 4)) {
573 munmap(idx_map
, idx_size
);
574 return error("pack too large for current definition of off_t in %s", path
);
578 p
->index_version
= version
;
579 p
->index_data
= idx_map
;
580 p
->index_size
= idx_size
;
585 int open_pack_index(struct packed_git
*p
)
593 idx_name
= xstrdup(p
->pack_name
);
594 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
595 ret
= check_packed_git_idx(idx_name
, p
);
600 static void scan_windows(struct packed_git
*p
,
601 struct packed_git
**lru_p
,
602 struct pack_window
**lru_w
,
603 struct pack_window
**lru_l
)
605 struct pack_window
*w
, *w_l
;
607 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
609 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
619 static int unuse_one_window(struct packed_git
*current
, int keep_fd
)
621 struct packed_git
*p
, *lru_p
= NULL
;
622 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
625 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
626 for (p
= packed_git
; p
; p
= p
->next
)
627 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
629 munmap(lru_w
->base
, lru_w
->len
);
630 pack_mapped
-= lru_w
->len
;
632 lru_l
->next
= lru_w
->next
;
634 lru_p
->windows
= lru_w
->next
;
635 if (!lru_p
->windows
&& lru_p
->pack_fd
!= -1
636 && lru_p
->pack_fd
!= keep_fd
) {
637 close(lru_p
->pack_fd
);
649 void release_pack_memory(size_t need
, int fd
)
651 size_t cur
= pack_mapped
;
652 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
, fd
))
656 void *xmmap(void *start
, size_t length
,
657 int prot
, int flags
, int fd
, off_t offset
)
659 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
660 if (ret
== MAP_FAILED
) {
663 release_pack_memory(length
, fd
);
664 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
665 if (ret
== MAP_FAILED
)
666 die_errno("Out of memory? mmap failed");
671 void close_pack_windows(struct packed_git
*p
)
674 struct pack_window
*w
= p
->windows
;
677 die("pack '%s' still has open windows to it",
679 munmap(w
->base
, w
->len
);
680 pack_mapped
-= w
->len
;
682 p
->windows
= w
->next
;
687 void unuse_pack(struct pack_window
**w_cursor
)
689 struct pack_window
*w
= *w_cursor
;
696 void close_pack_index(struct packed_git
*p
)
699 munmap((void *)p
->index_data
, p
->index_size
);
700 p
->index_data
= NULL
;
705 * This is used by git-repack in case a newly created pack happens to
706 * contain the same set of objects as an existing one. In that case
707 * the resulting file might be different even if its name would be the
708 * same. It is best to close any reference to the old pack before it is
709 * replaced on disk. Of course no index pointers nor windows for given pack
710 * must subsist at this point. If ever objects from this pack are requested
711 * again, the new version of the pack will be reinitialized through
712 * reprepare_packed_git().
714 void free_pack_by_name(const char *pack_name
)
716 struct packed_git
*p
, **pp
= &packed_git
;
720 if (strcmp(pack_name
, p
->pack_name
) == 0) {
721 clear_delta_base_cache();
722 close_pack_windows(p
);
723 if (p
->pack_fd
!= -1) {
728 free(p
->bad_object_sha1
);
730 if (last_found_pack
== p
)
731 last_found_pack
= NULL
;
739 static unsigned int get_max_fd_limit(void)
744 if (getrlimit(RLIMIT_NOFILE
, &lim
))
745 die_errno("cannot get RLIMIT_NOFILE");
748 #elif defined(_SC_OPEN_MAX)
749 return sysconf(_SC_OPEN_MAX
);
750 #elif defined(OPEN_MAX)
753 return 1; /* see the caller ;-) */
758 * Do not call this directly as this leaks p->pack_fd on error return;
759 * call open_packed_git() instead.
761 static int open_packed_git_1(struct packed_git
*p
)
764 struct pack_header hdr
;
765 unsigned char sha1
[20];
766 unsigned char *idx_sha1
;
769 if (!p
->index_data
&& open_pack_index(p
))
770 return error("packfile %s index unavailable", p
->pack_name
);
773 unsigned int max_fds
= get_max_fd_limit();
775 /* Save 3 for stdin/stdout/stderr, 22 for work */
777 pack_max_fds
= max_fds
- 25;
782 while (pack_max_fds
<= pack_open_fds
&& unuse_one_window(NULL
, -1))
785 p
->pack_fd
= git_open_noatime(p
->pack_name
);
786 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
790 /* If we created the struct before we had the pack we lack size. */
792 if (!S_ISREG(st
.st_mode
))
793 return error("packfile %s not a regular file", p
->pack_name
);
794 p
->pack_size
= st
.st_size
;
795 } else if (p
->pack_size
!= st
.st_size
)
796 return error("packfile %s size changed", p
->pack_name
);
798 /* We leave these file descriptors open with sliding mmap;
799 * there is no point keeping them open across exec(), though.
801 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
803 return error("cannot determine file descriptor flags");
804 fd_flag
|= FD_CLOEXEC
;
805 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
806 return error("cannot set FD_CLOEXEC");
808 /* Verify we recognize this pack file format. */
809 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
810 return error("file %s is far too short to be a packfile", p
->pack_name
);
811 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
812 return error("file %s is not a GIT packfile", p
->pack_name
);
813 if (!pack_version_ok(hdr
.hdr_version
))
814 return error("packfile %s is version %"PRIu32
" and not"
815 " supported (try upgrading GIT to a newer version)",
816 p
->pack_name
, ntohl(hdr
.hdr_version
));
818 /* Verify the pack matches its index. */
819 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
820 return error("packfile %s claims to have %"PRIu32
" objects"
821 " while index indicates %"PRIu32
" objects",
822 p
->pack_name
, ntohl(hdr
.hdr_entries
),
824 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
825 return error("end of packfile %s is unavailable", p
->pack_name
);
826 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
827 return error("packfile %s signature is unavailable", p
->pack_name
);
828 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
829 if (hashcmp(sha1
, idx_sha1
))
830 return error("packfile %s does not match index", p
->pack_name
);
834 static int open_packed_git(struct packed_git
*p
)
836 if (!open_packed_git_1(p
))
838 if (p
->pack_fd
!= -1) {
846 static int in_window(struct pack_window
*win
, off_t offset
)
848 /* We must promise at least 20 bytes (one hash) after the
849 * offset is available from this window, otherwise the offset
850 * is not actually in this window and a different window (which
851 * has that one hash excess) must be used. This is to support
852 * the object header and delta base parsing routines below.
854 off_t win_off
= win
->offset
;
855 return win_off
<= offset
856 && (offset
+ 20) <= (win_off
+ win
->len
);
859 unsigned char *use_pack(struct packed_git
*p
,
860 struct pack_window
**w_cursor
,
864 struct pack_window
*win
= *w_cursor
;
866 /* Since packfiles end in a hash of their content and it's
867 * pointless to ask for an offset into the middle of that
868 * hash, and the in_window function above wouldn't match
869 * don't allow an offset too close to the end of the file.
871 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
872 die("packfile %s cannot be accessed", p
->pack_name
);
873 if (offset
> (p
->pack_size
- 20))
874 die("offset beyond end of packfile (truncated pack?)");
876 if (!win
|| !in_window(win
, offset
)) {
879 for (win
= p
->windows
; win
; win
= win
->next
) {
880 if (in_window(win
, offset
))
884 size_t window_align
= packed_git_window_size
/ 2;
887 if (p
->pack_fd
== -1 && open_packed_git(p
))
888 die("packfile %s cannot be accessed", p
->pack_name
);
890 win
= xcalloc(1, sizeof(*win
));
891 win
->offset
= (offset
/ window_align
) * window_align
;
892 len
= p
->pack_size
- win
->offset
;
893 if (len
> packed_git_window_size
)
894 len
= packed_git_window_size
;
895 win
->len
= (size_t)len
;
896 pack_mapped
+= win
->len
;
897 while (packed_git_limit
< pack_mapped
898 && unuse_one_window(p
, p
->pack_fd
))
900 win
->base
= xmmap(NULL
, win
->len
,
901 PROT_READ
, MAP_PRIVATE
,
902 p
->pack_fd
, win
->offset
);
903 if (win
->base
== MAP_FAILED
)
904 die("packfile %s cannot be mapped: %s",
907 if (!win
->offset
&& win
->len
== p
->pack_size
908 && !p
->do_not_close
) {
915 if (pack_mapped
> peak_pack_mapped
)
916 peak_pack_mapped
= pack_mapped
;
917 if (pack_open_windows
> peak_pack_open_windows
)
918 peak_pack_open_windows
= pack_open_windows
;
919 win
->next
= p
->windows
;
923 if (win
!= *w_cursor
) {
924 win
->last_used
= pack_used_ctr
++;
928 offset
-= win
->offset
;
930 *left
= win
->len
- xsize_t(offset
);
931 return win
->base
+ offset
;
934 static struct packed_git
*alloc_packed_git(int extra
)
936 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
937 memset(p
, 0, sizeof(*p
));
942 static void try_to_free_pack_memory(size_t size
)
944 release_pack_memory(size
, -1);
947 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
949 static int have_set_try_to_free_routine
;
951 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
953 if (!have_set_try_to_free_routine
) {
954 have_set_try_to_free_routine
= 1;
955 set_try_to_free_routine(try_to_free_pack_memory
);
959 * Make sure a corresponding .pack file exists and that
960 * the index looks sane.
962 path_len
-= strlen(".idx");
967 memcpy(p
->pack_name
, path
, path_len
);
969 strcpy(p
->pack_name
+ path_len
, ".keep");
970 if (!access(p
->pack_name
, F_OK
))
973 strcpy(p
->pack_name
+ path_len
, ".pack");
974 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
979 /* ok, it looks sane as far as we can check without
980 * actually mapping the pack file.
982 p
->pack_size
= st
.st_size
;
983 p
->pack_local
= local
;
984 p
->mtime
= st
.st_mtime
;
985 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
990 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
992 const char *path
= sha1_pack_name(sha1
);
993 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
995 strcpy(p
->pack_name
, path
);
996 hashcpy(p
->sha1
, sha1
);
997 if (check_packed_git_idx(idx_path
, p
)) {
1005 void install_packed_git(struct packed_git
*pack
)
1007 if (pack
->pack_fd
!= -1)
1010 pack
->next
= packed_git
;
1014 static void prepare_packed_git_one(char *objdir
, int local
)
1016 /* Ensure that this buffer is large enough so that we can
1017 append "/pack/" without clobbering the stack even if
1018 strlen(objdir) were PATH_MAX. */
1019 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
1024 sprintf(path
, "%s/pack", objdir
);
1026 dir
= opendir(path
);
1028 if (errno
!= ENOENT
)
1029 error("unable to open object pack directory: %s: %s",
1030 path
, strerror(errno
));
1034 while ((de
= readdir(dir
)) != NULL
) {
1035 int namelen
= strlen(de
->d_name
);
1036 struct packed_git
*p
;
1038 if (!has_extension(de
->d_name
, ".idx"))
1041 if (len
+ namelen
+ 1 > sizeof(path
))
1044 /* Don't reopen a pack we already have. */
1045 strcpy(path
+ len
, de
->d_name
);
1046 for (p
= packed_git
; p
; p
= p
->next
) {
1047 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1052 /* See if it really is a valid .idx file with corresponding
1053 * .pack file that we can map.
1055 p
= add_packed_git(path
, len
+ namelen
, local
);
1058 install_packed_git(p
);
1063 static int sort_pack(const void *a_
, const void *b_
)
1065 struct packed_git
*a
= *((struct packed_git
**)a_
);
1066 struct packed_git
*b
= *((struct packed_git
**)b_
);
1070 * Local packs tend to contain objects specific to our
1071 * variant of the project than remote ones. In addition,
1072 * remote ones could be on a network mounted filesystem.
1073 * Favor local ones for these reasons.
1075 st
= a
->pack_local
- b
->pack_local
;
1080 * Younger packs tend to contain more recent objects,
1081 * and more recent objects tend to get accessed more
1084 if (a
->mtime
< b
->mtime
)
1086 else if (a
->mtime
== b
->mtime
)
1091 static void rearrange_packed_git(void)
1093 struct packed_git
**ary
, *p
;
1096 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1101 /* prepare an array of packed_git for easier sorting */
1102 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1103 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1106 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1108 /* link them back again */
1109 for (i
= 0; i
< n
- 1; i
++)
1110 ary
[i
]->next
= ary
[i
+ 1];
1111 ary
[n
- 1]->next
= NULL
;
1112 packed_git
= ary
[0];
1117 static int prepare_packed_git_run_once
= 0;
1118 void prepare_packed_git(void)
1120 struct alternate_object_database
*alt
;
1122 if (prepare_packed_git_run_once
)
1124 prepare_packed_git_one(get_object_directory(), 1);
1126 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1128 prepare_packed_git_one(alt
->base
, 0);
1129 alt
->name
[-1] = '/';
1131 rearrange_packed_git();
1132 prepare_packed_git_run_once
= 1;
1135 void reprepare_packed_git(void)
1138 prepare_packed_git_run_once
= 0;
1139 prepare_packed_git();
1142 static void mark_bad_packed_object(struct packed_git
*p
,
1143 const unsigned char *sha1
)
1146 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1147 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1149 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1150 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1151 p
->num_bad_objects
++;
1154 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1156 struct packed_git
*p
;
1159 for (p
= packed_git
; p
; p
= p
->next
)
1160 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1161 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1167 * With an in-core object data in "map", rehash it to make sure the
1168 * object name actually matches "sha1" to detect object corruption.
1169 * With "map" == NULL, try reading the object named with "sha1" using
1170 * the streaming interface and rehash it to do the same.
1172 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1173 unsigned long size
, const char *type
)
1175 unsigned char real_sha1
[20];
1176 enum object_type obj_type
;
1177 struct git_istream
*st
;
1183 hash_sha1_file(map
, size
, type
, real_sha1
);
1184 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1187 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1191 /* Generate the header */
1192 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1196 git_SHA1_Update(&c
, hdr
, hdrlen
);
1198 char buf
[1024 * 16];
1199 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1203 git_SHA1_Update(&c
, buf
, readlen
);
1205 git_SHA1_Final(real_sha1
, &c
);
1207 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1210 static int git_open_noatime(const char *name
)
1212 static int sha1_file_open_flag
= O_NOATIME
;
1215 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1219 /* Might the failure be due to O_NOATIME? */
1220 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1221 sha1_file_open_flag
= 0;
1229 static int open_sha1_file(const unsigned char *sha1
)
1232 char *name
= sha1_file_name(sha1
);
1233 struct alternate_object_database
*alt
;
1235 fd
= git_open_noatime(name
);
1241 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1243 fill_sha1_path(name
, sha1
);
1244 fd
= git_open_noatime(alt
->base
);
1251 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1256 fd
= open_sha1_file(sha1
);
1261 if (!fstat(fd
, &st
)) {
1262 *size
= xsize_t(st
.st_size
);
1264 /* mmap() is forbidden on empty files */
1265 error("object file %s is empty", sha1_file_name(sha1
));
1268 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1276 * There used to be a second loose object header format which
1277 * was meant to mimic the in-pack format, allowing for direct
1278 * copy of the object data. This format turned up not to be
1279 * really worth it and we no longer write loose objects in that
1282 static int experimental_loose_object(unsigned char *map
)
1287 * We must determine if the buffer contains the standard
1288 * zlib-deflated stream or the experimental format based
1289 * on the in-pack object format. Compare the header byte
1292 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1293 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1295 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1296 * in standard loose-object format, UNLESS it is a Git-pack
1297 * format object *exactly* 8 bytes in size when inflated.
1299 * However, RFC1950 also specifies that the 1st 16-bit word
1300 * must be divisible by 31 - this checksum tells us our buffer
1301 * is in the standard format, giving a false positive only if
1302 * the 1st word of the Git-pack format object happens to be
1303 * divisible by 31, ie:
1304 * ((byte0 * 256) + byte1) % 31 = 0
1305 * => 0ttt10000www1000 % 31 = 0
1307 * As it happens, this case can only arise for www=3 & ttt=1
1308 * - ie, a Commit object, which would have to be 8 bytes in
1309 * size. As no Commit can be that small, we find that the
1310 * combination of these two criteria (bitmask & checksum)
1311 * can always correctly determine the buffer format.
1313 word
= (map
[0] << 8) + map
[1];
1314 if ((map
[0] & 0x8F) == 0x08 && !(word
% 31))
1320 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1321 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1324 unsigned long size
, c
;
1325 unsigned long used
= 0;
1328 *type
= (c
>> 4) & 7;
1332 if (len
<= used
|| bitsizeof(long) <= shift
) {
1333 error("bad object header");
1338 size
+= (c
& 0x7f) << shift
;
1345 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1347 unsigned long size
, used
;
1348 static const char valid_loose_object_type
[8] = {
1350 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1351 0, /* "delta" and others are invalid in a loose object */
1353 enum object_type type
;
1355 /* Get the data stream */
1356 memset(stream
, 0, sizeof(*stream
));
1357 stream
->next_in
= map
;
1358 stream
->avail_in
= mapsize
;
1359 stream
->next_out
= buffer
;
1360 stream
->avail_out
= bufsiz
;
1362 if (experimental_loose_object(map
)) {
1364 * The old experimental format we no longer produce;
1365 * we can still read it.
1367 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1368 if (!used
|| !valid_loose_object_type
[type
])
1373 /* Set up the stream for the rest.. */
1374 stream
->next_in
= map
;
1375 stream
->avail_in
= mapsize
;
1376 git_inflate_init(stream
);
1378 /* And generate the fake traditional header */
1379 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1380 typename(type
), size
);
1383 git_inflate_init(stream
);
1384 return git_inflate(stream
, 0);
1387 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1389 int bytes
= strlen(buffer
) + 1;
1390 unsigned char *buf
= xmallocz(size
);
1394 n
= stream
->total_out
- bytes
;
1397 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1399 if (bytes
<= size
) {
1401 * The above condition must be (bytes <= size), not
1402 * (bytes < size). In other words, even though we
1403 * expect no more output and set avail_out to zero,
1404 * the input zlib stream may have bytes that express
1405 * "this concludes the stream", and we *do* want to
1408 * Otherwise we would not be able to test that we
1409 * consumed all the input to reach the expected size;
1410 * we also want to check that zlib tells us that all
1411 * went well with status == Z_STREAM_END at the end.
1413 stream
->next_out
= buf
+ bytes
;
1414 stream
->avail_out
= size
- bytes
;
1415 while (status
== Z_OK
)
1416 status
= git_inflate(stream
, Z_FINISH
);
1418 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1419 git_inflate_end(stream
);
1424 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1425 else if (stream
->avail_in
)
1426 error("garbage at end of loose object '%s'",
1433 * We used to just use "sscanf()", but that's actually way
1434 * too permissive for what we want to check. So do an anal
1435 * object header parse by hand.
1437 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1444 * The type can be at most ten bytes (including the
1445 * terminating '\0' that we add), and is followed by
1454 if (i
>= sizeof(type
))
1460 * The length must follow immediately, and be in canonical
1461 * decimal format (ie "010" is not valid).
1463 size
= *hdr
++ - '0';
1468 unsigned long c
= *hdr
- '0';
1472 size
= size
* 10 + c
;
1478 * The length must be followed by a zero byte
1480 return *hdr
? -1 : type_from_string(type
);
1483 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1489 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1490 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1493 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1496 unsigned long get_size_from_delta(struct packed_git
*p
,
1497 struct pack_window
**w_curs
,
1500 const unsigned char *data
;
1501 unsigned char delta_head
[20], *in
;
1505 memset(&stream
, 0, sizeof(stream
));
1506 stream
.next_out
= delta_head
;
1507 stream
.avail_out
= sizeof(delta_head
);
1509 git_inflate_init(&stream
);
1511 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1512 stream
.next_in
= in
;
1513 st
= git_inflate(&stream
, Z_FINISH
);
1514 curpos
+= stream
.next_in
- in
;
1515 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1516 stream
.total_out
< sizeof(delta_head
));
1517 git_inflate_end(&stream
);
1518 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1519 error("delta data unpack-initial failed");
1523 /* Examine the initial part of the delta to figure out
1528 /* ignore base size */
1529 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1531 /* Read the result size */
1532 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1535 static off_t
get_delta_base(struct packed_git
*p
,
1536 struct pack_window
**w_curs
,
1538 enum object_type type
,
1539 off_t delta_obj_offset
)
1541 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1544 /* use_pack() assured us we have [base_info, base_info + 20)
1545 * as a range that we can look at without walking off the
1546 * end of the mapped window. Its actually the hash size
1547 * that is assured. An OFS_DELTA longer than the hash size
1548 * is stupid, as then a REF_DELTA would be smaller to store.
1550 if (type
== OBJ_OFS_DELTA
) {
1552 unsigned char c
= base_info
[used
++];
1553 base_offset
= c
& 127;
1556 if (!base_offset
|| MSB(base_offset
, 7))
1557 return 0; /* overflow */
1558 c
= base_info
[used
++];
1559 base_offset
= (base_offset
<< 7) + (c
& 127);
1561 base_offset
= delta_obj_offset
- base_offset
;
1562 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1563 return 0; /* out of bound */
1565 } else if (type
== OBJ_REF_DELTA
) {
1566 /* The base entry _must_ be in the same pack */
1567 base_offset
= find_pack_entry_one(base_info
, p
);
1570 die("I am totally screwed");
1574 /* forward declaration for a mutually recursive function */
1575 static int packed_object_info(struct packed_git
*p
, off_t offset
,
1576 unsigned long *sizep
, int *rtype
);
1578 static int packed_delta_info(struct packed_git
*p
,
1579 struct pack_window
**w_curs
,
1581 enum object_type type
,
1583 unsigned long *sizep
)
1587 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1590 type
= packed_object_info(p
, base_offset
, NULL
, NULL
);
1591 if (type
<= OBJ_NONE
) {
1592 struct revindex_entry
*revidx
;
1593 const unsigned char *base_sha1
;
1594 revidx
= find_pack_revindex(p
, base_offset
);
1597 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1598 mark_bad_packed_object(p
, base_sha1
);
1599 type
= sha1_object_info(base_sha1
, NULL
);
1600 if (type
<= OBJ_NONE
)
1604 /* We choose to only get the type of the base object and
1605 * ignore potentially corrupt pack file that expects the delta
1606 * based on a base with a wrong size. This saves tons of
1610 *sizep
= get_size_from_delta(p
, w_curs
, curpos
);
1618 int unpack_object_header(struct packed_git
*p
,
1619 struct pack_window
**w_curs
,
1621 unsigned long *sizep
)
1623 unsigned char *base
;
1626 enum object_type type
;
1628 /* use_pack() assures us we have [base, base + 20) available
1629 * as a range that we can look at. (Its actually the hash
1630 * size that is assured.) With our object header encoding
1631 * the maximum deflated object size is 2^137, which is just
1632 * insane, so we know won't exceed what we have been given.
1634 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1635 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1644 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1645 unsigned long *sizep
, int *rtype
)
1647 struct pack_window
*w_curs
= NULL
;
1649 off_t curpos
= obj_offset
;
1650 enum object_type type
;
1652 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1654 *rtype
= type
; /* representation type */
1659 type
= packed_delta_info(p
, &w_curs
, curpos
,
1660 type
, obj_offset
, sizep
);
1670 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1671 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1674 unuse_pack(&w_curs
);
1678 static void *unpack_compressed_entry(struct packed_git
*p
,
1679 struct pack_window
**w_curs
,
1685 unsigned char *buffer
, *in
;
1687 buffer
= xmallocz(size
);
1688 memset(&stream
, 0, sizeof(stream
));
1689 stream
.next_out
= buffer
;
1690 stream
.avail_out
= size
+ 1;
1692 git_inflate_init(&stream
);
1694 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1695 stream
.next_in
= in
;
1696 st
= git_inflate(&stream
, Z_FINISH
);
1697 if (!stream
.avail_out
)
1698 break; /* the payload is larger than it should be */
1699 curpos
+= stream
.next_in
- in
;
1700 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1701 git_inflate_end(&stream
);
1702 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1710 #define MAX_DELTA_CACHE (256)
1712 static size_t delta_base_cached
;
1714 static struct delta_base_cache_lru_list
{
1715 struct delta_base_cache_lru_list
*prev
;
1716 struct delta_base_cache_lru_list
*next
;
1717 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1719 static struct delta_base_cache_entry
{
1720 struct delta_base_cache_lru_list lru
;
1722 struct packed_git
*p
;
1725 enum object_type type
;
1726 } delta_base_cache
[MAX_DELTA_CACHE
];
1728 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1732 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1733 hash
+= (hash
>> 8) + (hash
>> 16);
1734 return hash
% MAX_DELTA_CACHE
;
1737 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1739 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1740 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1741 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
1744 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1745 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1748 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1749 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1752 if (!ret
|| ent
->p
!= p
|| ent
->base_offset
!= base_offset
)
1753 return unpack_entry(p
, base_offset
, type
, base_size
);
1757 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1758 ent
->lru
.prev
->next
= ent
->lru
.next
;
1759 delta_base_cached
-= ent
->size
;
1761 ret
= xmemdupz(ent
->data
, ent
->size
);
1764 *base_size
= ent
->size
;
1768 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
1773 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1774 ent
->lru
.prev
->next
= ent
->lru
.next
;
1775 delta_base_cached
-= ent
->size
;
1779 void clear_delta_base_cache(void)
1782 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
1783 release_delta_base_cache(&delta_base_cache
[p
]);
1786 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
1787 void *base
, unsigned long base_size
, enum object_type type
)
1789 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1790 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
1791 struct delta_base_cache_lru_list
*lru
;
1793 release_delta_base_cache(ent
);
1794 delta_base_cached
+= base_size
;
1796 for (lru
= delta_base_cache_lru
.next
;
1797 delta_base_cached
> delta_base_cache_limit
1798 && lru
!= &delta_base_cache_lru
;
1800 struct delta_base_cache_entry
*f
= (void *)lru
;
1801 if (f
->type
== OBJ_BLOB
)
1802 release_delta_base_cache(f
);
1804 for (lru
= delta_base_cache_lru
.next
;
1805 delta_base_cached
> delta_base_cache_limit
1806 && lru
!= &delta_base_cache_lru
;
1808 struct delta_base_cache_entry
*f
= (void *)lru
;
1809 release_delta_base_cache(f
);
1813 ent
->base_offset
= base_offset
;
1816 ent
->size
= base_size
;
1817 ent
->lru
.next
= &delta_base_cache_lru
;
1818 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
1819 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
1820 delta_base_cache_lru
.prev
= &ent
->lru
;
1823 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
1824 unsigned long *size
);
1826 static void *unpack_delta_entry(struct packed_git
*p
,
1827 struct pack_window
**w_curs
,
1829 unsigned long delta_size
,
1831 enum object_type
*type
,
1832 unsigned long *sizep
)
1834 void *delta_data
, *result
, *base
;
1835 unsigned long base_size
;
1838 base_offset
= get_delta_base(p
, w_curs
, &curpos
, *type
, obj_offset
);
1840 error("failed to validate delta base reference "
1841 "at offset %"PRIuMAX
" from %s",
1842 (uintmax_t)curpos
, p
->pack_name
);
1846 base
= cache_or_unpack_entry(p
, base_offset
, &base_size
, type
, 0);
1849 * We're probably in deep shit, but let's try to fetch
1850 * the required base anyway from another pack or loose.
1851 * This is costly but should happen only in the presence
1852 * of a corrupted pack, and is better than failing outright.
1854 struct revindex_entry
*revidx
;
1855 const unsigned char *base_sha1
;
1856 revidx
= find_pack_revindex(p
, base_offset
);
1859 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1860 error("failed to read delta base object %s"
1861 " at offset %"PRIuMAX
" from %s",
1862 sha1_to_hex(base_sha1
), (uintmax_t)base_offset
,
1864 mark_bad_packed_object(p
, base_sha1
);
1865 base
= read_object(base_sha1
, type
, &base_size
);
1870 delta_data
= unpack_compressed_entry(p
, w_curs
, curpos
, delta_size
);
1872 error("failed to unpack compressed delta "
1873 "at offset %"PRIuMAX
" from %s",
1874 (uintmax_t)curpos
, p
->pack_name
);
1878 result
= patch_delta(base
, base_size
,
1879 delta_data
, delta_size
,
1882 die("failed to apply delta");
1884 add_delta_base_cache(p
, base_offset
, base
, base_size
, *type
);
1888 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
1890 static FILE *log_file
;
1893 log_file
= fopen(log_pack_access
, "w");
1895 error("cannot open pack access log '%s' for writing: %s",
1896 log_pack_access
, strerror(errno
));
1897 log_pack_access
= NULL
;
1901 fprintf(log_file
, "%s %"PRIuMAX
"\n",
1902 p
->pack_name
, (uintmax_t)obj_offset
);
1906 int do_check_packed_object_crc
;
1908 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
1909 enum object_type
*type
, unsigned long *sizep
)
1911 struct pack_window
*w_curs
= NULL
;
1912 off_t curpos
= obj_offset
;
1915 if (log_pack_access
)
1916 write_pack_access_log(p
, obj_offset
);
1918 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
1919 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1920 unsigned long len
= revidx
[1].offset
- obj_offset
;
1921 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
1922 const unsigned char *sha1
=
1923 nth_packed_object_sha1(p
, revidx
->nr
);
1924 error("bad packed object CRC for %s",
1926 mark_bad_packed_object(p
, sha1
);
1927 unuse_pack(&w_curs
);
1932 *type
= unpack_object_header(p
, &w_curs
, &curpos
, sizep
);
1936 data
= unpack_delta_entry(p
, &w_curs
, curpos
, *sizep
,
1937 obj_offset
, type
, sizep
);
1943 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, *sizep
);
1947 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1948 *type
, (uintmax_t)obj_offset
, p
->pack_name
);
1950 unuse_pack(&w_curs
);
1954 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
1957 const unsigned char *index
= p
->index_data
;
1959 if (open_pack_index(p
))
1961 index
= p
->index_data
;
1963 if (n
>= p
->num_objects
)
1966 if (p
->index_version
== 1) {
1967 return index
+ 24 * n
+ 4;
1970 return index
+ 20 * n
;
1974 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
1976 const unsigned char *index
= p
->index_data
;
1978 if (p
->index_version
== 1) {
1979 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
1982 index
+= 8 + p
->num_objects
* (20 + 4);
1983 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
1984 if (!(off
& 0x80000000))
1986 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
1987 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
1988 ntohl(*((uint32_t *)(index
+ 4)));
1992 off_t
find_pack_entry_one(const unsigned char *sha1
,
1993 struct packed_git
*p
)
1995 const uint32_t *level1_ofs
= p
->index_data
;
1996 const unsigned char *index
= p
->index_data
;
1997 unsigned hi
, lo
, stride
;
1998 static int use_lookup
= -1;
1999 static int debug_lookup
= -1;
2001 if (debug_lookup
< 0)
2002 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2005 if (open_pack_index(p
))
2007 level1_ofs
= p
->index_data
;
2008 index
= p
->index_data
;
2010 if (p
->index_version
> 1) {
2015 hi
= ntohl(level1_ofs
[*sha1
]);
2016 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2017 if (p
->index_version
> 1) {
2025 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2026 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2029 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2031 int pos
= sha1_entry_pos(index
, stride
, 0,
2032 lo
, hi
, p
->num_objects
, sha1
);
2035 return nth_packed_object_offset(p
, pos
);
2039 unsigned mi
= (lo
+ hi
) / 2;
2040 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2043 printf("lo %u hi %u rg %u mi %u\n",
2044 lo
, hi
, hi
- lo
, mi
);
2046 return nth_packed_object_offset(p
, mi
);
2055 int is_pack_valid(struct packed_git
*p
)
2057 /* An already open pack is known to be valid. */
2058 if (p
->pack_fd
!= -1)
2061 /* If the pack has one window completely covering the
2062 * file size, the pack is known to be valid even if
2063 * the descriptor is not currently open.
2066 struct pack_window
*w
= p
->windows
;
2068 if (!w
->offset
&& w
->len
== p
->pack_size
)
2072 /* Force the pack to open to prove its valid. */
2073 return !open_packed_git(p
);
2076 static int fill_pack_entry(const unsigned char *sha1
,
2077 struct pack_entry
*e
,
2078 struct packed_git
*p
)
2082 if (p
->num_bad_objects
) {
2084 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2085 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2089 offset
= find_pack_entry_one(sha1
, p
);
2094 * We are about to tell the caller where they can locate the
2095 * requested object. We better make sure the packfile is
2096 * still here and can be accessed before supplying that
2097 * answer, as it may have been deleted since the index was
2100 if (!is_pack_valid(p
)) {
2101 warning("packfile %s cannot be accessed", p
->pack_name
);
2106 hashcpy(e
->sha1
, sha1
);
2110 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2112 struct packed_git
*p
;
2114 prepare_packed_git();
2118 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2121 for (p
= packed_git
; p
; p
= p
->next
) {
2122 if (p
== last_found_pack
|| !fill_pack_entry(sha1
, e
, p
))
2125 last_found_pack
= p
;
2131 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2132 struct packed_git
*packs
)
2134 struct packed_git
*p
;
2136 for (p
= packs
; p
; p
= p
->next
) {
2137 if (find_pack_entry_one(sha1
, p
))
2144 static int sha1_loose_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2147 unsigned long mapsize
, size
;
2152 map
= map_sha1_file(sha1
, &mapsize
);
2154 return error("unable to find %s", sha1_to_hex(sha1
));
2155 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2156 status
= error("unable to unpack %s header",
2158 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2159 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2162 git_inflate_end(&stream
);
2163 munmap(map
, mapsize
);
2167 /* returns enum object_type or negative */
2168 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
)
2170 struct cached_object
*co
;
2171 struct pack_entry e
;
2174 co
= find_cached_object(sha1
);
2177 *(oi
->sizep
) = co
->size
;
2178 oi
->whence
= OI_CACHED
;
2182 if (!find_pack_entry(sha1
, &e
)) {
2183 /* Most likely it's a loose object. */
2184 status
= sha1_loose_object_info(sha1
, oi
->sizep
);
2186 oi
->whence
= OI_LOOSE
;
2190 /* Not a loose object; someone else may have just packed it. */
2191 reprepare_packed_git();
2192 if (!find_pack_entry(sha1
, &e
))
2196 status
= packed_object_info(e
.p
, e
.offset
, oi
->sizep
, &rtype
);
2198 mark_bad_packed_object(e
.p
, sha1
);
2199 status
= sha1_object_info_extended(sha1
, oi
);
2200 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2201 oi
->whence
= OI_DBCACHED
;
2203 oi
->whence
= OI_PACKED
;
2204 oi
->u
.packed
.offset
= e
.offset
;
2205 oi
->u
.packed
.pack
= e
.p
;
2206 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2207 rtype
== OBJ_OFS_DELTA
);
2213 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2215 struct object_info oi
;
2218 return sha1_object_info_extended(sha1
, &oi
);
2221 static void *read_packed_sha1(const unsigned char *sha1
,
2222 enum object_type
*type
, unsigned long *size
)
2224 struct pack_entry e
;
2227 if (!find_pack_entry(sha1
, &e
))
2229 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2232 * We're probably in deep shit, but let's try to fetch
2233 * the required object anyway from another pack or loose.
2234 * This should happen only in the presence of a corrupted
2235 * pack, and is better than failing outright.
2237 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2238 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2239 mark_bad_packed_object(e
.p
, sha1
);
2240 data
= read_object(sha1
, type
, size
);
2245 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2246 unsigned char *sha1
)
2248 struct cached_object
*co
;
2250 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2251 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2253 if (cached_object_alloc
<= cached_object_nr
) {
2254 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2255 cached_objects
= xrealloc(cached_objects
,
2256 sizeof(*cached_objects
) *
2257 cached_object_alloc
);
2259 co
= &cached_objects
[cached_object_nr
++];
2262 co
->buf
= xmalloc(len
);
2263 memcpy(co
->buf
, buf
, len
);
2264 hashcpy(co
->sha1
, sha1
);
2268 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2269 unsigned long *size
)
2271 unsigned long mapsize
;
2273 struct cached_object
*co
;
2275 co
= find_cached_object(sha1
);
2279 return xmemdupz(co
->buf
, co
->size
);
2282 buf
= read_packed_sha1(sha1
, type
, size
);
2285 map
= map_sha1_file(sha1
, &mapsize
);
2287 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2288 munmap(map
, mapsize
);
2291 reprepare_packed_git();
2292 return read_packed_sha1(sha1
, type
, size
);
2296 * This function dies on corrupt objects; the callers who want to
2297 * deal with them should arrange to call read_object() and give error
2298 * messages themselves.
2300 void *read_sha1_file_extended(const unsigned char *sha1
,
2301 enum object_type
*type
,
2302 unsigned long *size
,
2307 const struct packed_git
*p
;
2308 const unsigned char *repl
= (flag
& READ_SHA1_FILE_REPLACE
)
2309 ? lookup_replace_object(sha1
) : sha1
;
2312 data
= read_object(repl
, type
, size
);
2316 if (errno
&& errno
!= ENOENT
)
2317 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2319 /* die if we replaced an object with one that does not exist */
2321 die("replacement %s not found for %s",
2322 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2324 if (has_loose_object(repl
)) {
2325 path
= sha1_file_name(sha1
);
2326 die("loose object %s (stored in %s) is corrupt",
2327 sha1_to_hex(repl
), path
);
2330 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2331 die("packed object %s (stored in %s) is corrupt",
2332 sha1_to_hex(repl
), p
->pack_name
);
2337 void *read_object_with_reference(const unsigned char *sha1
,
2338 const char *required_type_name
,
2339 unsigned long *size
,
2340 unsigned char *actual_sha1_return
)
2342 enum object_type type
, required_type
;
2344 unsigned long isize
;
2345 unsigned char actual_sha1
[20];
2347 required_type
= type_from_string(required_type_name
);
2348 hashcpy(actual_sha1
, sha1
);
2350 int ref_length
= -1;
2351 const char *ref_type
= NULL
;
2353 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2356 if (type
== required_type
) {
2358 if (actual_sha1_return
)
2359 hashcpy(actual_sha1_return
, actual_sha1
);
2362 /* Handle references */
2363 else if (type
== OBJ_COMMIT
)
2365 else if (type
== OBJ_TAG
)
2366 ref_type
= "object ";
2371 ref_length
= strlen(ref_type
);
2373 if (ref_length
+ 40 > isize
||
2374 memcmp(buffer
, ref_type
, ref_length
) ||
2375 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2380 /* Now we have the ID of the referred-to object in
2381 * actual_sha1. Check again. */
2385 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2386 const char *type
, unsigned char *sha1
,
2387 char *hdr
, int *hdrlen
)
2391 /* Generate the header */
2392 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2396 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2397 git_SHA1_Update(&c
, buf
, len
);
2398 git_SHA1_Final(sha1
, &c
);
2402 * Move the just written object into its final resting place.
2403 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2404 * "moving" is only a part of what it does, when no patch between
2405 * master to pu changes the call sites of this function.
2407 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2411 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2413 else if (link(tmpfile
, filename
))
2417 * Coda hack - coda doesn't like cross-directory links,
2418 * so we fall back to a rename, which will mean that it
2419 * won't be able to check collisions, but that's not a
2422 * The same holds for FAT formatted media.
2424 * When this succeeds, we just return. We have nothing
2427 if (ret
&& ret
!= EEXIST
) {
2429 if (!rename(tmpfile
, filename
))
2433 unlink_or_warn(tmpfile
);
2435 if (ret
!= EEXIST
) {
2436 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2438 /* FIXME!!! Collision check here ? */
2442 if (adjust_shared_perm(filename
))
2443 return error("unable to set permission to '%s'", filename
);
2447 static int write_buffer(int fd
, const void *buf
, size_t len
)
2449 if (write_in_full(fd
, buf
, len
) < 0)
2450 return error("file write error (%s)", strerror(errno
));
2454 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2455 unsigned char *sha1
)
2459 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2463 /* Finalize a file on disk, and close it. */
2464 static void close_sha1_file(int fd
)
2466 if (fsync_object_files
)
2467 fsync_or_die(fd
, "sha1 file");
2469 die_errno("error when closing sha1 file");
2472 /* Size of directory component, including the ending '/' */
2473 static inline int directory_size(const char *filename
)
2475 const char *s
= strrchr(filename
, '/');
2478 return s
- filename
+ 1;
2482 * This creates a temporary file in the same directory as the final
2485 * We want to avoid cross-directory filename renames, because those
2486 * can have problems on various filesystems (FAT, NFS, Coda).
2488 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2490 int fd
, dirlen
= directory_size(filename
);
2492 if (dirlen
+ 20 > bufsiz
) {
2493 errno
= ENAMETOOLONG
;
2496 memcpy(buffer
, filename
, dirlen
);
2497 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2498 fd
= git_mkstemp_mode(buffer
, 0444);
2499 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2500 /* Make sure the directory exists */
2501 memcpy(buffer
, filename
, dirlen
);
2502 buffer
[dirlen
-1] = 0;
2503 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2507 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2508 fd
= git_mkstemp_mode(buffer
, 0444);
2513 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2514 const void *buf
, unsigned long len
, time_t mtime
)
2517 unsigned char compressed
[4096];
2520 unsigned char parano_sha1
[20];
2522 static char tmp_file
[PATH_MAX
];
2524 filename
= sha1_file_name(sha1
);
2525 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2527 if (errno
== EACCES
)
2528 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2530 return error("unable to create temporary file: %s", strerror(errno
));
2534 memset(&stream
, 0, sizeof(stream
));
2535 git_deflate_init(&stream
, zlib_compression_level
);
2536 stream
.next_out
= compressed
;
2537 stream
.avail_out
= sizeof(compressed
);
2540 /* First header.. */
2541 stream
.next_in
= (unsigned char *)hdr
;
2542 stream
.avail_in
= hdrlen
;
2543 while (git_deflate(&stream
, 0) == Z_OK
)
2545 git_SHA1_Update(&c
, hdr
, hdrlen
);
2547 /* Then the data itself.. */
2548 stream
.next_in
= (void *)buf
;
2549 stream
.avail_in
= len
;
2551 unsigned char *in0
= stream
.next_in
;
2552 ret
= git_deflate(&stream
, Z_FINISH
);
2553 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2554 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2555 die("unable to write sha1 file");
2556 stream
.next_out
= compressed
;
2557 stream
.avail_out
= sizeof(compressed
);
2558 } while (ret
== Z_OK
);
2560 if (ret
!= Z_STREAM_END
)
2561 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2562 ret
= git_deflate_end_gently(&stream
);
2564 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2565 git_SHA1_Final(parano_sha1
, &c
);
2566 if (hashcmp(sha1
, parano_sha1
) != 0)
2567 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2569 close_sha1_file(fd
);
2574 utb
.modtime
= mtime
;
2575 if (utime(tmp_file
, &utb
) < 0)
2576 warning("failed utime() on %s: %s",
2577 tmp_file
, strerror(errno
));
2580 return move_temp_to_file(tmp_file
, filename
);
2583 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2585 unsigned char sha1
[20];
2589 /* Normally if we have it in the pack then we do not bother writing
2590 * it out into .git/objects/??/?{38} file.
2592 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2594 hashcpy(returnsha1
, sha1
);
2595 if (has_sha1_file(sha1
))
2597 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2600 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2604 enum object_type type
;
2609 if (has_loose_object(sha1
))
2611 buf
= read_packed_sha1(sha1
, &type
, &len
);
2613 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2614 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2615 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2621 int has_pack_index(const unsigned char *sha1
)
2624 if (stat(sha1_pack_index_name(sha1
), &st
))
2629 int has_sha1_pack(const unsigned char *sha1
)
2631 struct pack_entry e
;
2632 return find_pack_entry(sha1
, &e
);
2635 int has_sha1_file(const unsigned char *sha1
)
2637 struct pack_entry e
;
2639 if (find_pack_entry(sha1
, &e
))
2641 return has_loose_object(sha1
);
2644 static void check_tree(const void *buf
, size_t size
)
2646 struct tree_desc desc
;
2647 struct name_entry entry
;
2649 init_tree_desc(&desc
, buf
, size
);
2650 while (tree_entry(&desc
, &entry
))
2652 * tree_entry() will die() on malformed entries */
2656 static void check_commit(const void *buf
, size_t size
)
2659 memset(&c
, 0, sizeof(c
));
2660 if (parse_commit_buffer(&c
, buf
, size
))
2661 die("corrupt commit");
2664 static void check_tag(const void *buf
, size_t size
)
2667 memset(&t
, 0, sizeof(t
));
2668 if (parse_tag_buffer(&t
, buf
, size
))
2672 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
2673 enum object_type type
,
2674 const char *path
, unsigned flags
)
2676 int ret
, re_allocated
= 0;
2677 int write_object
= flags
& HASH_WRITE_OBJECT
;
2683 * Convert blobs to git internal format
2685 if ((type
== OBJ_BLOB
) && path
) {
2686 struct strbuf nbuf
= STRBUF_INIT
;
2687 if (convert_to_git(path
, buf
, size
, &nbuf
,
2688 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
2689 buf
= strbuf_detach(&nbuf
, &size
);
2693 if (flags
& HASH_FORMAT_CHECK
) {
2694 if (type
== OBJ_TREE
)
2695 check_tree(buf
, size
);
2696 if (type
== OBJ_COMMIT
)
2697 check_commit(buf
, size
);
2698 if (type
== OBJ_TAG
)
2699 check_tag(buf
, size
);
2703 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
2705 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
2711 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
2712 const char *path
, unsigned flags
)
2714 struct strbuf sbuf
= STRBUF_INIT
;
2717 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
2718 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
2721 strbuf_release(&sbuf
);
2725 #define SMALL_FILE_SIZE (32*1024)
2727 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
2728 enum object_type type
, const char *path
,
2734 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
2735 } else if (size
<= SMALL_FILE_SIZE
) {
2736 char *buf
= xmalloc(size
);
2737 if (size
== read_in_full(fd
, buf
, size
))
2738 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
2740 ret
= error("short read %s", strerror(errno
));
2743 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2744 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
2751 * This creates one packfile per large blob unless bulk-checkin
2752 * machinery is "plugged".
2754 * This also bypasses the usual "convert-to-git" dance, and that is on
2755 * purpose. We could write a streaming version of the converting
2756 * functions and insert that before feeding the data to fast-import
2757 * (or equivalent in-core API described above). However, that is
2758 * somewhat complicated, as we do not know the size of the filter
2759 * result, which we need to know beforehand when writing a git object.
2760 * Since the primary motivation for trying to stream from the working
2761 * tree file and to avoid mmaping it in core is to deal with large
2762 * binary blobs, they generally do not want to get any conversion, and
2763 * callers should avoid this code path when filters are requested.
2765 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
2766 enum object_type type
, const char *path
,
2769 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
2772 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
2773 enum object_type type
, const char *path
, unsigned flags
)
2776 size_t size
= xsize_t(st
->st_size
);
2778 if (!S_ISREG(st
->st_mode
))
2779 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
2780 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
2781 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
2782 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
2784 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
2789 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
2792 struct strbuf sb
= STRBUF_INIT
;
2794 switch (st
->st_mode
& S_IFMT
) {
2796 fd
= open(path
, O_RDONLY
);
2798 return error("open(\"%s\"): %s", path
,
2800 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
2801 return error("%s: failed to insert into database",
2805 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
2806 char *errstr
= strerror(errno
);
2807 return error("readlink(\"%s\"): %s", path
,
2810 if (!(flags
& HASH_WRITE_OBJECT
))
2811 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
2812 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
2813 return error("%s: failed to insert into database",
2815 strbuf_release(&sb
);
2818 return resolve_gitlink_ref(path
, "HEAD", sha1
);
2820 return error("%s: unsupported file type", path
);
2825 int read_pack_header(int fd
, struct pack_header
*header
)
2827 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
2828 /* "eof before pack header was fully read" */
2829 return PH_ERROR_EOF
;
2831 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
2832 /* "protocol error (pack signature mismatch detected)" */
2833 return PH_ERROR_PACK_SIGNATURE
;
2834 if (!pack_version_ok(header
->hdr_version
))
2835 /* "protocol error (pack version unsupported)" */
2836 return PH_ERROR_PROTOCOL
;
2840 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
2842 enum object_type type
= sha1_object_info(sha1
, NULL
);
2844 die("%s is not a valid object", sha1_to_hex(sha1
));
2846 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),