2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
15 #include "run-command.h"
18 #include "tree-walk.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
27 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28 #define O_NOATIME 01000000
34 #define SZ_FMT PRIuMAX
35 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
37 const unsigned char null_sha1
[20];
39 static const char *no_log_pack_access
= "no_log_pack_access";
40 static const char *log_pack_access
;
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
48 static struct cached_object
{
49 unsigned char sha1
[20];
50 enum object_type type
;
54 static int cached_object_nr
, cached_object_alloc
;
56 static struct cached_object empty_tree
= {
57 EMPTY_TREE_SHA1_BIN_LITERAL
,
63 static struct packed_git
*last_found_pack
;
65 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
68 struct cached_object
*co
= cached_objects
;
70 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
71 if (!hashcmp(co
->sha1
, sha1
))
74 if (!hashcmp(sha1
, empty_tree
.sha1
))
79 int mkdir_in_gitdir(const char *path
)
81 if (mkdir(path
, 0777)) {
82 int saved_errno
= errno
;
84 struct strbuf sb
= STRBUF_INIT
;
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
95 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
96 strbuf_readlink(&sb
, path
, st
.st_size
) ||
97 !is_absolute_path(sb
.buf
) ||
98 mkdir(sb
.buf
, 0777)) {
105 return adjust_shared_perm(path
);
108 int safe_create_leading_directories(char *path
)
110 char *pos
= path
+ offset_1st_component(path
);
114 pos
= strchr(pos
, '/');
117 while (*++pos
== '/')
122 if (!stat(path
, &st
)) {
124 if (!S_ISDIR(st
.st_mode
)) {
129 else if (mkdir(path
, 0777)) {
130 if (errno
== EEXIST
&&
131 !stat(path
, &st
) && S_ISDIR(st
.st_mode
)) {
132 ; /* somebody created it since we checked */
138 else if (adjust_shared_perm(path
)) {
147 int safe_create_leading_directories_const(const char *path
)
149 /* path points to cache entries, so xstrdup before messing with it */
150 char *buf
= xstrdup(path
);
151 int result
= safe_create_leading_directories(buf
);
156 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
159 for (i
= 0; i
< 20; i
++) {
160 static char hex
[] = "0123456789abcdef";
161 unsigned int val
= sha1
[i
];
162 char *pos
= pathbuf
+ i
*2 + (i
> 0);
163 *pos
++ = hex
[val
>> 4];
164 *pos
= hex
[val
& 0xf];
169 * NOTE! This returns a statically allocated buffer, so you have to be
170 * careful about using it. Do an "xstrdup()" if you need to save the
173 * Also note that this returns the location for creating. Reading
174 * SHA1 file can happen from any alternate directory listed in the
175 * DB_ENVIRONMENT environment variable if it is not found in
176 * the primary object database.
178 char *sha1_file_name(const unsigned char *sha1
)
180 static char buf
[PATH_MAX
];
184 objdir
= get_object_directory();
185 len
= strlen(objdir
);
187 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
188 if (len
+ 43 > PATH_MAX
)
189 die("insanely long object directory %s", objdir
);
190 memcpy(buf
, objdir
, len
);
194 fill_sha1_path(buf
+ len
+ 1, sha1
);
198 static char *sha1_get_pack_name(const unsigned char *sha1
,
199 char **name
, char **base
, const char *which
)
201 static const char hex
[] = "0123456789abcdef";
206 const char *sha1_file_directory
= get_object_directory();
207 int len
= strlen(sha1_file_directory
);
208 *base
= xmalloc(len
+ 60);
209 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
210 sha1_file_directory
, which
);
211 *name
= *base
+ len
+ 11;
216 for (i
= 0; i
< 20; i
++) {
217 unsigned int val
= *sha1
++;
218 *buf
++ = hex
[val
>> 4];
219 *buf
++ = hex
[val
& 0xf];
225 char *sha1_pack_name(const unsigned char *sha1
)
227 static char *name
, *base
;
229 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
232 char *sha1_pack_index_name(const unsigned char *sha1
)
234 static char *name
, *base
;
236 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
239 struct alternate_object_database
*alt_odb_list
;
240 static struct alternate_object_database
**alt_odb_tail
;
242 static int git_open_noatime(const char *name
);
245 * Prepare alternate object database registry.
247 * The variable alt_odb_list points at the list of struct
248 * alternate_object_database. The elements on this list come from
249 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
250 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
251 * whose contents is similar to that environment variable but can be
252 * LF separated. Its base points at a statically allocated buffer that
253 * contains "/the/directory/corresponding/to/.git/objects/...", while
254 * its name points just after the slash at the end of ".git/objects/"
255 * in the example above, and has enough space to hold 40-byte hex
256 * SHA1, an extra slash for the first level indirection, and the
259 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
260 int depth
, const char *normalized_objdir
)
262 struct alternate_object_database
*ent
;
263 struct alternate_object_database
*alt
;
265 struct strbuf pathbuf
= STRBUF_INIT
;
267 if (!is_absolute_path(entry
) && relative_base
) {
268 strbuf_addstr(&pathbuf
, real_path(relative_base
));
269 strbuf_addch(&pathbuf
, '/');
271 strbuf_addstr(&pathbuf
, entry
);
273 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
275 pfxlen
= strlen(pathbuf
.buf
);
278 * The trailing slash after the directory name is given by
279 * this function at the end. Remove duplicates.
281 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
284 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
285 ent
= xmalloc(sizeof(*ent
) + entlen
);
286 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
287 strbuf_release(&pathbuf
);
289 ent
->name
= ent
->base
+ pfxlen
+ 1;
290 ent
->base
[pfxlen
+ 3] = '/';
291 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
293 /* Detect cases where alternate disappeared */
294 if (!is_directory(ent
->base
)) {
295 error("object directory %s does not exist; "
296 "check .git/objects/info/alternates.",
302 /* Prevent the common mistake of listing the same
303 * thing twice, or object directory itself.
305 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
306 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
311 if (!strcmp_icase(ent
->base
, normalized_objdir
)) {
316 /* add the alternate entry */
318 alt_odb_tail
= &(ent
->next
);
321 /* recursively add alternates */
322 read_info_alternates(ent
->base
, depth
+ 1);
324 ent
->base
[pfxlen
] = '/';
329 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
330 const char *relative_base
, int depth
)
332 struct string_list entries
= STRING_LIST_INIT_NODUP
;
335 struct strbuf objdirbuf
= STRBUF_INIT
;
338 error("%s: ignoring alternate object stores, nesting too deep.",
343 strbuf_addstr(&objdirbuf
, absolute_path(get_object_directory()));
344 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
346 alt_copy
= xmemdupz(alt
, len
);
347 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
348 for (i
= 0; i
< entries
.nr
; i
++) {
349 const char *entry
= entries
.items
[i
].string
;
350 if (entry
[0] == '\0' || entry
[0] == '#')
352 if (!is_absolute_path(entry
) && depth
) {
353 error("%s: ignoring relative alternate object store %s",
354 relative_base
, entry
);
356 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
359 string_list_clear(&entries
, 0);
361 strbuf_release(&objdirbuf
);
364 void read_info_alternates(const char * relative_base
, int depth
)
369 const char alt_file_name
[] = "info/alternates";
370 /* Given that relative_base is no longer than PATH_MAX,
371 ensure that "path" has enough space to append "/", the
372 file name, "info/alternates", and a trailing NUL. */
373 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
376 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
377 fd
= git_open_noatime(path
);
380 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
384 mapsz
= xsize_t(st
.st_size
);
385 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
388 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
393 void add_to_alternates_file(const char *reference
)
395 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
396 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
397 char *alt
= mkpath("%s\n", reference
);
398 write_or_die(fd
, alt
, strlen(alt
));
399 if (commit_lock_file(lock
))
400 die("could not close alternates file");
402 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
405 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
407 struct alternate_object_database
*ent
;
410 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
415 void prepare_alt_odb(void)
422 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
425 alt_odb_tail
= &alt_odb_list
;
426 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
428 read_info_alternates(get_object_directory(), 0);
431 static int has_loose_object_local(const unsigned char *sha1
)
433 char *name
= sha1_file_name(sha1
);
434 return !access(name
, F_OK
);
437 int has_loose_object_nonlocal(const unsigned char *sha1
)
439 struct alternate_object_database
*alt
;
441 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
442 fill_sha1_path(alt
->name
, sha1
);
443 if (!access(alt
->base
, F_OK
))
449 static int has_loose_object(const unsigned char *sha1
)
451 return has_loose_object_local(sha1
) ||
452 has_loose_object_nonlocal(sha1
);
455 static unsigned int pack_used_ctr
;
456 static unsigned int pack_mmap_calls
;
457 static unsigned int peak_pack_open_windows
;
458 static unsigned int pack_open_windows
;
459 static unsigned int pack_open_fds
;
460 static unsigned int pack_max_fds
;
461 static size_t peak_pack_mapped
;
462 static size_t pack_mapped
;
463 struct packed_git
*packed_git
;
465 void pack_report(void)
468 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
469 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
470 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
471 sz_fmt(getpagesize()),
472 sz_fmt(packed_git_window_size
),
473 sz_fmt(packed_git_limit
));
475 "pack_report: pack_used_ctr = %10u\n"
476 "pack_report: pack_mmap_calls = %10u\n"
477 "pack_report: pack_open_windows = %10u / %10u\n"
478 "pack_report: pack_mapped = "
479 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
482 pack_open_windows
, peak_pack_open_windows
,
483 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
486 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
489 struct pack_idx_header
*hdr
;
491 uint32_t version
, nr
, i
, *index
;
492 int fd
= git_open_noatime(path
);
497 if (fstat(fd
, &st
)) {
501 idx_size
= xsize_t(st
.st_size
);
502 if (idx_size
< 4 * 256 + 20 + 20) {
504 return error("index file %s is too small", path
);
506 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
510 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
511 version
= ntohl(hdr
->idx_version
);
512 if (version
< 2 || version
> 2) {
513 munmap(idx_map
, idx_size
);
514 return error("index file %s is version %"PRIu32
515 " and is not supported by this binary"
516 " (try upgrading GIT to a newer version)",
525 index
+= 2; /* skip index header */
526 for (i
= 0; i
< 256; i
++) {
527 uint32_t n
= ntohl(index
[i
]);
529 munmap(idx_map
, idx_size
);
530 return error("non-monotonic index %s", path
);
538 * - 256 index entries 4 bytes each
539 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
540 * - 20-byte SHA1 of the packfile
541 * - 20-byte SHA1 file checksum
543 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
544 munmap(idx_map
, idx_size
);
545 return error("wrong index v1 file size in %s", path
);
547 } else if (version
== 2) {
550 * - 8 bytes of header
551 * - 256 index entries 4 bytes each
552 * - 20-byte sha1 entry * nr
553 * - 4-byte crc entry * nr
554 * - 4-byte offset entry * nr
555 * - 20-byte SHA1 of the packfile
556 * - 20-byte SHA1 file checksum
557 * And after the 4-byte offset table might be a
558 * variable sized table containing 8-byte entries
559 * for offsets larger than 2^31.
561 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
562 unsigned long max_size
= min_size
;
564 max_size
+= (nr
- 1)*8;
565 if (idx_size
< min_size
|| idx_size
> max_size
) {
566 munmap(idx_map
, idx_size
);
567 return error("wrong index v2 file size in %s", path
);
569 if (idx_size
!= min_size
&&
571 * make sure we can deal with large pack offsets.
572 * 31-bit signed offset won't be enough, neither
573 * 32-bit unsigned one will be.
575 (sizeof(off_t
) <= 4)) {
576 munmap(idx_map
, idx_size
);
577 return error("pack too large for current definition of off_t in %s", path
);
581 p
->index_version
= version
;
582 p
->index_data
= idx_map
;
583 p
->index_size
= idx_size
;
588 int open_pack_index(struct packed_git
*p
)
596 idx_name
= xstrdup(p
->pack_name
);
597 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
598 ret
= check_packed_git_idx(idx_name
, p
);
603 static void scan_windows(struct packed_git
*p
,
604 struct packed_git
**lru_p
,
605 struct pack_window
**lru_w
,
606 struct pack_window
**lru_l
)
608 struct pack_window
*w
, *w_l
;
610 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
612 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
622 static int unuse_one_window(struct packed_git
*current
)
624 struct packed_git
*p
, *lru_p
= NULL
;
625 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
628 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
629 for (p
= packed_git
; p
; p
= p
->next
)
630 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
632 munmap(lru_w
->base
, lru_w
->len
);
633 pack_mapped
-= lru_w
->len
;
635 lru_l
->next
= lru_w
->next
;
637 lru_p
->windows
= lru_w
->next
;
645 void release_pack_memory(size_t need
)
647 size_t cur
= pack_mapped
;
648 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
652 void *xmmap(void *start
, size_t length
,
653 int prot
, int flags
, int fd
, off_t offset
)
655 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
656 if (ret
== MAP_FAILED
) {
659 release_pack_memory(length
);
660 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
661 if (ret
== MAP_FAILED
)
662 die_errno("Out of memory? mmap failed");
667 void close_pack_windows(struct packed_git
*p
)
670 struct pack_window
*w
= p
->windows
;
673 die("pack '%s' still has open windows to it",
675 munmap(w
->base
, w
->len
);
676 pack_mapped
-= w
->len
;
678 p
->windows
= w
->next
;
684 * The LRU pack is the one with the oldest MRU window, preferring packs
685 * with no used windows, or the oldest mtime if it has no windows allocated.
687 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
689 struct pack_window
*w
, *this_mru_w
;
690 int has_windows_inuse
= 0;
693 * Reject this pack if it has windows and the previously selected
694 * one does not. If this pack does not have windows, reject
695 * it if the pack file is newer than the previously selected one.
697 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
700 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
702 * Reject this pack if any of its windows are in use,
703 * but the previously selected pack did not have any
704 * inuse windows. Otherwise, record that this pack
705 * has windows in use.
708 if (*accept_windows_inuse
)
709 has_windows_inuse
= 1;
714 if (w
->last_used
> this_mru_w
->last_used
)
718 * Reject this pack if it has windows that have been
719 * used more recently than the previously selected pack.
720 * If the previously selected pack had windows inuse and
721 * we have not encountered a window in this pack that is
722 * inuse, skip this check since we prefer a pack with no
723 * inuse windows to one that has inuse windows.
725 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
726 this_mru_w
->last_used
> (*mru_w
)->last_used
)
735 *accept_windows_inuse
= has_windows_inuse
;
738 static int close_one_pack(void)
740 struct packed_git
*p
, *lru_p
= NULL
;
741 struct pack_window
*mru_w
= NULL
;
742 int accept_windows_inuse
= 1;
744 for (p
= packed_git
; p
; p
= p
->next
) {
745 if (p
->pack_fd
== -1)
747 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
751 close(lru_p
->pack_fd
);
760 void unuse_pack(struct pack_window
**w_cursor
)
762 struct pack_window
*w
= *w_cursor
;
769 void close_pack_index(struct packed_git
*p
)
772 munmap((void *)p
->index_data
, p
->index_size
);
773 p
->index_data
= NULL
;
778 * This is used by git-repack in case a newly created pack happens to
779 * contain the same set of objects as an existing one. In that case
780 * the resulting file might be different even if its name would be the
781 * same. It is best to close any reference to the old pack before it is
782 * replaced on disk. Of course no index pointers nor windows for given pack
783 * must subsist at this point. If ever objects from this pack are requested
784 * again, the new version of the pack will be reinitialized through
785 * reprepare_packed_git().
787 void free_pack_by_name(const char *pack_name
)
789 struct packed_git
*p
, **pp
= &packed_git
;
793 if (strcmp(pack_name
, p
->pack_name
) == 0) {
794 clear_delta_base_cache();
795 close_pack_windows(p
);
796 if (p
->pack_fd
!= -1) {
801 free(p
->bad_object_sha1
);
803 if (last_found_pack
== p
)
804 last_found_pack
= NULL
;
812 static unsigned int get_max_fd_limit(void)
818 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
825 long open_max
= sysconf(_SC_OPEN_MAX
);
829 * Otherwise, we got -1 for one of the two
832 * (1) sysconf() did not understand _SC_OPEN_MAX
833 * and signaled an error with -1; or
834 * (2) sysconf() said there is no limit.
836 * We _could_ clear errno before calling sysconf() to
837 * tell these two cases apart and return a huge number
838 * in the latter case to let the caller cap it to a
839 * value that is not so selfish, but letting the
840 * fallback OPEN_MAX codepath take care of these cases
849 return 1; /* see the caller ;-) */
854 * Do not call this directly as this leaks p->pack_fd on error return;
855 * call open_packed_git() instead.
857 static int open_packed_git_1(struct packed_git
*p
)
860 struct pack_header hdr
;
861 unsigned char sha1
[20];
862 unsigned char *idx_sha1
;
865 if (!p
->index_data
&& open_pack_index(p
))
866 return error("packfile %s index unavailable", p
->pack_name
);
869 unsigned int max_fds
= get_max_fd_limit();
871 /* Save 3 for stdin/stdout/stderr, 22 for work */
873 pack_max_fds
= max_fds
- 25;
878 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
881 p
->pack_fd
= git_open_noatime(p
->pack_name
);
882 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
886 /* If we created the struct before we had the pack we lack size. */
888 if (!S_ISREG(st
.st_mode
))
889 return error("packfile %s not a regular file", p
->pack_name
);
890 p
->pack_size
= st
.st_size
;
891 } else if (p
->pack_size
!= st
.st_size
)
892 return error("packfile %s size changed", p
->pack_name
);
894 /* We leave these file descriptors open with sliding mmap;
895 * there is no point keeping them open across exec(), though.
897 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
899 return error("cannot determine file descriptor flags");
900 fd_flag
|= FD_CLOEXEC
;
901 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
902 return error("cannot set FD_CLOEXEC");
904 /* Verify we recognize this pack file format. */
905 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
906 return error("file %s is far too short to be a packfile", p
->pack_name
);
907 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
908 return error("file %s is not a GIT packfile", p
->pack_name
);
909 if (!pack_version_ok(hdr
.hdr_version
))
910 return error("packfile %s is version %"PRIu32
" and not"
911 " supported (try upgrading GIT to a newer version)",
912 p
->pack_name
, ntohl(hdr
.hdr_version
));
914 /* Verify the pack matches its index. */
915 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
916 return error("packfile %s claims to have %"PRIu32
" objects"
917 " while index indicates %"PRIu32
" objects",
918 p
->pack_name
, ntohl(hdr
.hdr_entries
),
920 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
921 return error("end of packfile %s is unavailable", p
->pack_name
);
922 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
923 return error("packfile %s signature is unavailable", p
->pack_name
);
924 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
925 if (hashcmp(sha1
, idx_sha1
))
926 return error("packfile %s does not match index", p
->pack_name
);
930 static int open_packed_git(struct packed_git
*p
)
932 if (!open_packed_git_1(p
))
934 if (p
->pack_fd
!= -1) {
942 static int in_window(struct pack_window
*win
, off_t offset
)
944 /* We must promise at least 20 bytes (one hash) after the
945 * offset is available from this window, otherwise the offset
946 * is not actually in this window and a different window (which
947 * has that one hash excess) must be used. This is to support
948 * the object header and delta base parsing routines below.
950 off_t win_off
= win
->offset
;
951 return win_off
<= offset
952 && (offset
+ 20) <= (win_off
+ win
->len
);
955 unsigned char *use_pack(struct packed_git
*p
,
956 struct pack_window
**w_cursor
,
960 struct pack_window
*win
= *w_cursor
;
962 /* Since packfiles end in a hash of their content and it's
963 * pointless to ask for an offset into the middle of that
964 * hash, and the in_window function above wouldn't match
965 * don't allow an offset too close to the end of the file.
967 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
968 die("packfile %s cannot be accessed", p
->pack_name
);
969 if (offset
> (p
->pack_size
- 20))
970 die("offset beyond end of packfile (truncated pack?)");
972 if (!win
|| !in_window(win
, offset
)) {
975 for (win
= p
->windows
; win
; win
= win
->next
) {
976 if (in_window(win
, offset
))
980 size_t window_align
= packed_git_window_size
/ 2;
983 if (p
->pack_fd
== -1 && open_packed_git(p
))
984 die("packfile %s cannot be accessed", p
->pack_name
);
986 win
= xcalloc(1, sizeof(*win
));
987 win
->offset
= (offset
/ window_align
) * window_align
;
988 len
= p
->pack_size
- win
->offset
;
989 if (len
> packed_git_window_size
)
990 len
= packed_git_window_size
;
991 win
->len
= (size_t)len
;
992 pack_mapped
+= win
->len
;
993 while (packed_git_limit
< pack_mapped
994 && unuse_one_window(p
))
996 win
->base
= xmmap(NULL
, win
->len
,
997 PROT_READ
, MAP_PRIVATE
,
998 p
->pack_fd
, win
->offset
);
999 if (win
->base
== MAP_FAILED
)
1000 die("packfile %s cannot be mapped: %s",
1003 if (!win
->offset
&& win
->len
== p
->pack_size
1004 && !p
->do_not_close
) {
1010 pack_open_windows
++;
1011 if (pack_mapped
> peak_pack_mapped
)
1012 peak_pack_mapped
= pack_mapped
;
1013 if (pack_open_windows
> peak_pack_open_windows
)
1014 peak_pack_open_windows
= pack_open_windows
;
1015 win
->next
= p
->windows
;
1019 if (win
!= *w_cursor
) {
1020 win
->last_used
= pack_used_ctr
++;
1024 offset
-= win
->offset
;
1026 *left
= win
->len
- xsize_t(offset
);
1027 return win
->base
+ offset
;
1030 static struct packed_git
*alloc_packed_git(int extra
)
1032 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1033 memset(p
, 0, sizeof(*p
));
1038 static void try_to_free_pack_memory(size_t size
)
1040 release_pack_memory(size
);
1043 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1045 static int have_set_try_to_free_routine
;
1047 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1049 if (!have_set_try_to_free_routine
) {
1050 have_set_try_to_free_routine
= 1;
1051 set_try_to_free_routine(try_to_free_pack_memory
);
1055 * Make sure a corresponding .pack file exists and that
1056 * the index looks sane.
1058 path_len
-= strlen(".idx");
1063 memcpy(p
->pack_name
, path
, path_len
);
1065 strcpy(p
->pack_name
+ path_len
, ".keep");
1066 if (!access(p
->pack_name
, F_OK
))
1069 strcpy(p
->pack_name
+ path_len
, ".pack");
1070 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1075 /* ok, it looks sane as far as we can check without
1076 * actually mapping the pack file.
1078 p
->pack_size
= st
.st_size
;
1079 p
->pack_local
= local
;
1080 p
->mtime
= st
.st_mtime
;
1081 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1086 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1088 const char *path
= sha1_pack_name(sha1
);
1089 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1091 strcpy(p
->pack_name
, path
);
1092 hashcpy(p
->sha1
, sha1
);
1093 if (check_packed_git_idx(idx_path
, p
)) {
1101 void install_packed_git(struct packed_git
*pack
)
1103 if (pack
->pack_fd
!= -1)
1106 pack
->next
= packed_git
;
1110 void (*report_garbage
)(const char *desc
, const char *path
);
1112 static void report_helper(const struct string_list
*list
,
1113 int seen_bits
, int first
, int last
)
1116 switch (seen_bits
) {
1118 msg
= "no corresponding .idx nor .pack";
1121 msg
= "no corresponding .idx";
1124 msg
= "no corresponding .pack";
1129 for (; first
< last
; first
++)
1130 report_garbage(msg
, list
->items
[first
].string
);
1133 static void report_pack_garbage(struct string_list
*list
)
1135 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1137 if (!report_garbage
)
1140 sort_string_list(list
);
1142 for (i
= 0; i
< list
->nr
; i
++) {
1143 const char *path
= list
->items
[i
].string
;
1144 if (baselen
!= -1 &&
1145 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1146 report_helper(list
, seen_bits
, first
, i
);
1150 if (baselen
== -1) {
1151 const char *dot
= strrchr(path
, '.');
1153 report_garbage("garbage found", path
);
1156 baselen
= dot
- path
+ 1;
1159 if (!strcmp(path
+ baselen
, "pack"))
1161 else if (!strcmp(path
+ baselen
, "idx"))
1164 report_helper(list
, seen_bits
, first
, list
->nr
);
1167 static void prepare_packed_git_one(char *objdir
, int local
)
1169 /* Ensure that this buffer is large enough so that we can
1170 append "/pack/" without clobbering the stack even if
1171 strlen(objdir) were PATH_MAX. */
1172 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
1176 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1178 sprintf(path
, "%s/pack", objdir
);
1180 dir
= opendir(path
);
1182 if (errno
!= ENOENT
)
1183 error("unable to open object pack directory: %s: %s",
1184 path
, strerror(errno
));
1188 while ((de
= readdir(dir
)) != NULL
) {
1189 int namelen
= strlen(de
->d_name
);
1190 struct packed_git
*p
;
1192 if (len
+ namelen
+ 1 > sizeof(path
)) {
1193 if (report_garbage
) {
1194 struct strbuf sb
= STRBUF_INIT
;
1195 strbuf_addf(&sb
, "%.*s/%s", len
- 1, path
, de
->d_name
);
1196 report_garbage("path too long", sb
.buf
);
1197 strbuf_release(&sb
);
1202 if (is_dot_or_dotdot(de
->d_name
))
1205 strcpy(path
+ len
, de
->d_name
);
1207 if (has_extension(de
->d_name
, ".idx")) {
1208 /* Don't reopen a pack we already have. */
1209 for (p
= packed_git
; p
; p
= p
->next
) {
1210 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1215 * See if it really is a valid .idx file with
1216 * corresponding .pack file that we can map.
1218 (p
= add_packed_git(path
, len
+ namelen
, local
)) != NULL
)
1219 install_packed_git(p
);
1222 if (!report_garbage
)
1225 if (has_extension(de
->d_name
, ".idx") ||
1226 has_extension(de
->d_name
, ".pack") ||
1227 has_extension(de
->d_name
, ".keep"))
1228 string_list_append(&garbage
, path
);
1230 report_garbage("garbage found", path
);
1233 report_pack_garbage(&garbage
);
1234 string_list_clear(&garbage
, 0);
1237 static int sort_pack(const void *a_
, const void *b_
)
1239 struct packed_git
*a
= *((struct packed_git
**)a_
);
1240 struct packed_git
*b
= *((struct packed_git
**)b_
);
1244 * Local packs tend to contain objects specific to our
1245 * variant of the project than remote ones. In addition,
1246 * remote ones could be on a network mounted filesystem.
1247 * Favor local ones for these reasons.
1249 st
= a
->pack_local
- b
->pack_local
;
1254 * Younger packs tend to contain more recent objects,
1255 * and more recent objects tend to get accessed more
1258 if (a
->mtime
< b
->mtime
)
1260 else if (a
->mtime
== b
->mtime
)
1265 static void rearrange_packed_git(void)
1267 struct packed_git
**ary
, *p
;
1270 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1275 /* prepare an array of packed_git for easier sorting */
1276 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1277 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1280 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1282 /* link them back again */
1283 for (i
= 0; i
< n
- 1; i
++)
1284 ary
[i
]->next
= ary
[i
+ 1];
1285 ary
[n
- 1]->next
= NULL
;
1286 packed_git
= ary
[0];
1291 static int prepare_packed_git_run_once
= 0;
1292 void prepare_packed_git(void)
1294 struct alternate_object_database
*alt
;
1296 if (prepare_packed_git_run_once
)
1298 prepare_packed_git_one(get_object_directory(), 1);
1300 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1302 prepare_packed_git_one(alt
->base
, 0);
1303 alt
->name
[-1] = '/';
1305 rearrange_packed_git();
1306 prepare_packed_git_run_once
= 1;
1309 void reprepare_packed_git(void)
1312 prepare_packed_git_run_once
= 0;
1313 prepare_packed_git();
1316 static void mark_bad_packed_object(struct packed_git
*p
,
1317 const unsigned char *sha1
)
1320 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1321 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1323 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1324 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1325 p
->num_bad_objects
++;
1328 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1330 struct packed_git
*p
;
1333 for (p
= packed_git
; p
; p
= p
->next
)
1334 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1335 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1341 * With an in-core object data in "map", rehash it to make sure the
1342 * object name actually matches "sha1" to detect object corruption.
1343 * With "map" == NULL, try reading the object named with "sha1" using
1344 * the streaming interface and rehash it to do the same.
1346 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1347 unsigned long size
, const char *type
)
1349 unsigned char real_sha1
[20];
1350 enum object_type obj_type
;
1351 struct git_istream
*st
;
1357 hash_sha1_file(map
, size
, type
, real_sha1
);
1358 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1361 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1365 /* Generate the header */
1366 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1370 git_SHA1_Update(&c
, hdr
, hdrlen
);
1372 char buf
[1024 * 16];
1373 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1381 git_SHA1_Update(&c
, buf
, readlen
);
1383 git_SHA1_Final(real_sha1
, &c
);
1385 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1388 static int git_open_noatime(const char *name
)
1390 static int sha1_file_open_flag
= O_NOATIME
;
1393 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1397 /* Might the failure be due to O_NOATIME? */
1398 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1399 sha1_file_open_flag
= 0;
1407 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1409 char *name
= sha1_file_name(sha1
);
1410 struct alternate_object_database
*alt
;
1412 if (!lstat(name
, st
))
1417 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1419 fill_sha1_path(name
, sha1
);
1420 if (!lstat(alt
->base
, st
))
1427 static int open_sha1_file(const unsigned char *sha1
)
1430 char *name
= sha1_file_name(sha1
);
1431 struct alternate_object_database
*alt
;
1433 fd
= git_open_noatime(name
);
1439 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1441 fill_sha1_path(name
, sha1
);
1442 fd
= git_open_noatime(alt
->base
);
1449 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1454 fd
= open_sha1_file(sha1
);
1459 if (!fstat(fd
, &st
)) {
1460 *size
= xsize_t(st
.st_size
);
1462 /* mmap() is forbidden on empty files */
1463 error("object file %s is empty", sha1_file_name(sha1
));
1466 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1474 * There used to be a second loose object header format which
1475 * was meant to mimic the in-pack format, allowing for direct
1476 * copy of the object data. This format turned up not to be
1477 * really worth it and we no longer write loose objects in that
1480 static int experimental_loose_object(unsigned char *map
)
1485 * We must determine if the buffer contains the standard
1486 * zlib-deflated stream or the experimental format based
1487 * on the in-pack object format. Compare the header byte
1490 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1491 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1493 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1494 * in standard loose-object format, UNLESS it is a Git-pack
1495 * format object *exactly* 8 bytes in size when inflated.
1497 * However, RFC1950 also specifies that the 1st 16-bit word
1498 * must be divisible by 31 - this checksum tells us our buffer
1499 * is in the standard format, giving a false positive only if
1500 * the 1st word of the Git-pack format object happens to be
1501 * divisible by 31, ie:
1502 * ((byte0 * 256) + byte1) % 31 = 0
1503 * => 0ttt10000www1000 % 31 = 0
1505 * As it happens, this case can only arise for www=3 & ttt=1
1506 * - ie, a Commit object, which would have to be 8 bytes in
1507 * size. As no Commit can be that small, we find that the
1508 * combination of these two criteria (bitmask & checksum)
1509 * can always correctly determine the buffer format.
1511 word
= (map
[0] << 8) + map
[1];
1512 if ((map
[0] & 0x8F) == 0x08 && !(word
% 31))
1518 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1519 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1522 unsigned long size
, c
;
1523 unsigned long used
= 0;
1526 *type
= (c
>> 4) & 7;
1530 if (len
<= used
|| bitsizeof(long) <= shift
) {
1531 error("bad object header");
1536 size
+= (c
& 0x7f) << shift
;
1543 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1545 unsigned long size
, used
;
1546 static const char valid_loose_object_type
[8] = {
1548 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1549 0, /* "delta" and others are invalid in a loose object */
1551 enum object_type type
;
1553 /* Get the data stream */
1554 memset(stream
, 0, sizeof(*stream
));
1555 stream
->next_in
= map
;
1556 stream
->avail_in
= mapsize
;
1557 stream
->next_out
= buffer
;
1558 stream
->avail_out
= bufsiz
;
1560 if (experimental_loose_object(map
)) {
1562 * The old experimental format we no longer produce;
1563 * we can still read it.
1565 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1566 if (!used
|| !valid_loose_object_type
[type
])
1571 /* Set up the stream for the rest.. */
1572 stream
->next_in
= map
;
1573 stream
->avail_in
= mapsize
;
1574 git_inflate_init(stream
);
1576 /* And generate the fake traditional header */
1577 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1578 typename(type
), size
);
1581 git_inflate_init(stream
);
1582 return git_inflate(stream
, 0);
1585 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1587 int bytes
= strlen(buffer
) + 1;
1588 unsigned char *buf
= xmallocz(size
);
1592 n
= stream
->total_out
- bytes
;
1595 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1597 if (bytes
<= size
) {
1599 * The above condition must be (bytes <= size), not
1600 * (bytes < size). In other words, even though we
1601 * expect no more output and set avail_out to zero,
1602 * the input zlib stream may have bytes that express
1603 * "this concludes the stream", and we *do* want to
1606 * Otherwise we would not be able to test that we
1607 * consumed all the input to reach the expected size;
1608 * we also want to check that zlib tells us that all
1609 * went well with status == Z_STREAM_END at the end.
1611 stream
->next_out
= buf
+ bytes
;
1612 stream
->avail_out
= size
- bytes
;
1613 while (status
== Z_OK
)
1614 status
= git_inflate(stream
, Z_FINISH
);
1616 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1617 git_inflate_end(stream
);
1622 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1623 else if (stream
->avail_in
)
1624 error("garbage at end of loose object '%s'",
1631 * We used to just use "sscanf()", but that's actually way
1632 * too permissive for what we want to check. So do an anal
1633 * object header parse by hand.
1635 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1642 * The type can be at most ten bytes (including the
1643 * terminating '\0' that we add), and is followed by
1652 if (i
>= sizeof(type
))
1658 * The length must follow immediately, and be in canonical
1659 * decimal format (ie "010" is not valid).
1661 size
= *hdr
++ - '0';
1666 unsigned long c
= *hdr
- '0';
1670 size
= size
* 10 + c
;
1676 * The length must be followed by a zero byte
1678 return *hdr
? -1 : type_from_string(type
);
1681 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1687 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1688 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1691 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1694 unsigned long get_size_from_delta(struct packed_git
*p
,
1695 struct pack_window
**w_curs
,
1698 const unsigned char *data
;
1699 unsigned char delta_head
[20], *in
;
1703 memset(&stream
, 0, sizeof(stream
));
1704 stream
.next_out
= delta_head
;
1705 stream
.avail_out
= sizeof(delta_head
);
1707 git_inflate_init(&stream
);
1709 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1710 stream
.next_in
= in
;
1711 st
= git_inflate(&stream
, Z_FINISH
);
1712 curpos
+= stream
.next_in
- in
;
1713 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1714 stream
.total_out
< sizeof(delta_head
));
1715 git_inflate_end(&stream
);
1716 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1717 error("delta data unpack-initial failed");
1721 /* Examine the initial part of the delta to figure out
1726 /* ignore base size */
1727 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1729 /* Read the result size */
1730 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1733 static off_t
get_delta_base(struct packed_git
*p
,
1734 struct pack_window
**w_curs
,
1736 enum object_type type
,
1737 off_t delta_obj_offset
)
1739 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1742 /* use_pack() assured us we have [base_info, base_info + 20)
1743 * as a range that we can look at without walking off the
1744 * end of the mapped window. Its actually the hash size
1745 * that is assured. An OFS_DELTA longer than the hash size
1746 * is stupid, as then a REF_DELTA would be smaller to store.
1748 if (type
== OBJ_OFS_DELTA
) {
1750 unsigned char c
= base_info
[used
++];
1751 base_offset
= c
& 127;
1754 if (!base_offset
|| MSB(base_offset
, 7))
1755 return 0; /* overflow */
1756 c
= base_info
[used
++];
1757 base_offset
= (base_offset
<< 7) + (c
& 127);
1759 base_offset
= delta_obj_offset
- base_offset
;
1760 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1761 return 0; /* out of bound */
1763 } else if (type
== OBJ_REF_DELTA
) {
1764 /* The base entry _must_ be in the same pack */
1765 base_offset
= find_pack_entry_one(base_info
, p
);
1768 die("I am totally screwed");
1772 int unpack_object_header(struct packed_git
*p
,
1773 struct pack_window
**w_curs
,
1775 unsigned long *sizep
)
1777 unsigned char *base
;
1780 enum object_type type
;
1782 /* use_pack() assures us we have [base, base + 20) available
1783 * as a range that we can look at. (Its actually the hash
1784 * size that is assured.) With our object header encoding
1785 * the maximum deflated object size is 2^137, which is just
1786 * insane, so we know won't exceed what we have been given.
1788 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1789 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1798 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1801 struct revindex_entry
*revidx
;
1802 const unsigned char *sha1
;
1803 revidx
= find_pack_revindex(p
, obj_offset
);
1806 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1807 mark_bad_packed_object(p
, sha1
);
1808 type
= sha1_object_info(sha1
, NULL
);
1809 if (type
<= OBJ_NONE
)
1814 #define POI_STACK_PREALLOC 64
1816 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1818 enum object_type type
,
1819 struct pack_window
**w_curs
,
1822 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1823 off_t
*poi_stack
= small_poi_stack
;
1824 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1826 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1829 /* Push the object we're going to leave behind */
1830 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1831 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1832 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1833 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1835 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1837 poi_stack
[poi_stack_nr
++] = obj_offset
;
1838 /* If parsing the base offset fails, just unwind */
1839 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1842 curpos
= obj_offset
= base_offset
;
1843 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1844 if (type
<= OBJ_NONE
) {
1845 /* If getting the base itself fails, we first
1846 * retry the base, otherwise unwind */
1847 type
= retry_bad_packed_offset(p
, base_offset
);
1848 if (type
> OBJ_NONE
)
1862 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1863 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1868 if (poi_stack
!= small_poi_stack
)
1873 while (poi_stack_nr
) {
1874 obj_offset
= poi_stack
[--poi_stack_nr
];
1875 type
= retry_bad_packed_offset(p
, obj_offset
);
1876 if (type
> OBJ_NONE
)
1883 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1884 struct object_info
*oi
)
1886 struct pack_window
*w_curs
= NULL
;
1888 off_t curpos
= obj_offset
;
1889 enum object_type type
;
1892 * We always get the representation type, but only convert it to
1893 * a "real" type later if the caller is interested.
1895 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1898 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1899 off_t tmp_pos
= curpos
;
1900 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1906 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1907 if (*oi
->sizep
== 0) {
1916 if (oi
->disk_sizep
) {
1917 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1918 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1922 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1923 if (*oi
->typep
< 0) {
1930 unuse_pack(&w_curs
);
1934 static void *unpack_compressed_entry(struct packed_git
*p
,
1935 struct pack_window
**w_curs
,
1941 unsigned char *buffer
, *in
;
1943 buffer
= xmallocz(size
);
1944 memset(&stream
, 0, sizeof(stream
));
1945 stream
.next_out
= buffer
;
1946 stream
.avail_out
= size
+ 1;
1948 git_inflate_init(&stream
);
1950 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1951 stream
.next_in
= in
;
1952 st
= git_inflate(&stream
, Z_FINISH
);
1953 if (!stream
.avail_out
)
1954 break; /* the payload is larger than it should be */
1955 curpos
+= stream
.next_in
- in
;
1956 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1957 git_inflate_end(&stream
);
1958 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1966 #define MAX_DELTA_CACHE (256)
1968 static size_t delta_base_cached
;
1970 static struct delta_base_cache_lru_list
{
1971 struct delta_base_cache_lru_list
*prev
;
1972 struct delta_base_cache_lru_list
*next
;
1973 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1975 static struct delta_base_cache_entry
{
1976 struct delta_base_cache_lru_list lru
;
1978 struct packed_git
*p
;
1981 enum object_type type
;
1982 } delta_base_cache
[MAX_DELTA_CACHE
];
1984 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1988 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1989 hash
+= (hash
>> 8) + (hash
>> 16);
1990 return hash
% MAX_DELTA_CACHE
;
1993 static struct delta_base_cache_entry
*
1994 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1996 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1997 return delta_base_cache
+ hash
;
2000 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
2001 struct packed_git
*p
, off_t base_offset
)
2003 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
2006 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2008 struct delta_base_cache_entry
*ent
;
2009 ent
= get_delta_base_cache_entry(p
, base_offset
);
2010 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
2013 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2016 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2017 ent
->lru
.prev
->next
= ent
->lru
.next
;
2018 delta_base_cached
-= ent
->size
;
2021 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2022 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2024 struct delta_base_cache_entry
*ent
;
2027 ent
= get_delta_base_cache_entry(p
, base_offset
);
2029 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2030 return unpack_entry(p
, base_offset
, type
, base_size
);
2035 clear_delta_base_cache_entry(ent
);
2037 ret
= xmemdupz(ent
->data
, ent
->size
);
2039 *base_size
= ent
->size
;
2043 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2048 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2049 ent
->lru
.prev
->next
= ent
->lru
.next
;
2050 delta_base_cached
-= ent
->size
;
2054 void clear_delta_base_cache(void)
2057 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2058 release_delta_base_cache(&delta_base_cache
[p
]);
2061 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2062 void *base
, unsigned long base_size
, enum object_type type
)
2064 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2065 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2066 struct delta_base_cache_lru_list
*lru
;
2068 release_delta_base_cache(ent
);
2069 delta_base_cached
+= base_size
;
2071 for (lru
= delta_base_cache_lru
.next
;
2072 delta_base_cached
> delta_base_cache_limit
2073 && lru
!= &delta_base_cache_lru
;
2075 struct delta_base_cache_entry
*f
= (void *)lru
;
2076 if (f
->type
== OBJ_BLOB
)
2077 release_delta_base_cache(f
);
2079 for (lru
= delta_base_cache_lru
.next
;
2080 delta_base_cached
> delta_base_cache_limit
2081 && lru
!= &delta_base_cache_lru
;
2083 struct delta_base_cache_entry
*f
= (void *)lru
;
2084 release_delta_base_cache(f
);
2088 ent
->base_offset
= base_offset
;
2091 ent
->size
= base_size
;
2092 ent
->lru
.next
= &delta_base_cache_lru
;
2093 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2094 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2095 delta_base_cache_lru
.prev
= &ent
->lru
;
2098 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2099 unsigned long *size
);
2101 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2103 static FILE *log_file
;
2105 if (!log_pack_access
)
2106 log_pack_access
= getenv("GIT_TRACE_PACK_ACCESS");
2107 if (!log_pack_access
)
2108 log_pack_access
= no_log_pack_access
;
2109 if (log_pack_access
== no_log_pack_access
)
2113 log_file
= fopen(log_pack_access
, "w");
2115 error("cannot open pack access log '%s' for writing: %s",
2116 log_pack_access
, strerror(errno
));
2117 log_pack_access
= no_log_pack_access
;
2121 fprintf(log_file
, "%s %"PRIuMAX
"\n",
2122 p
->pack_name
, (uintmax_t)obj_offset
);
2126 int do_check_packed_object_crc
;
2128 #define UNPACK_ENTRY_STACK_PREALLOC 64
2129 struct unpack_entry_stack_ent
{
2135 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2136 enum object_type
*final_type
, unsigned long *final_size
)
2138 struct pack_window
*w_curs
= NULL
;
2139 off_t curpos
= obj_offset
;
2142 enum object_type type
;
2143 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2144 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2145 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2146 int base_from_cache
= 0;
2148 if (log_pack_access
!= no_log_pack_access
)
2149 write_pack_access_log(p
, obj_offset
);
2151 /* PHASE 1: drill down to the innermost base object */
2155 struct delta_base_cache_entry
*ent
;
2157 ent
= get_delta_base_cache_entry(p
, curpos
);
2158 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2162 clear_delta_base_cache_entry(ent
);
2163 base_from_cache
= 1;
2167 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2168 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2169 unsigned long len
= revidx
[1].offset
- obj_offset
;
2170 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2171 const unsigned char *sha1
=
2172 nth_packed_object_sha1(p
, revidx
->nr
);
2173 error("bad packed object CRC for %s",
2175 mark_bad_packed_object(p
, sha1
);
2176 unuse_pack(&w_curs
);
2181 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2182 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2185 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2187 error("failed to validate delta base reference "
2188 "at offset %"PRIuMAX
" from %s",
2189 (uintmax_t)curpos
, p
->pack_name
);
2190 /* bail to phase 2, in hopes of recovery */
2195 /* push object, proceed to base */
2196 if (delta_stack_nr
>= delta_stack_alloc
2197 && delta_stack
== small_delta_stack
) {
2198 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2199 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2200 memcpy(delta_stack
, small_delta_stack
,
2201 sizeof(*delta_stack
)*delta_stack_nr
);
2203 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2205 i
= delta_stack_nr
++;
2206 delta_stack
[i
].obj_offset
= obj_offset
;
2207 delta_stack
[i
].curpos
= curpos
;
2208 delta_stack
[i
].size
= size
;
2210 curpos
= obj_offset
= base_offset
;
2213 /* PHASE 2: handle the base */
2218 die("BUG in unpack_entry: left loop at a valid delta");
2224 if (!base_from_cache
)
2225 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2229 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2230 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2233 /* PHASE 3: apply deltas in order */
2236 * 'data' holds the base data, or NULL if there was corruption
2238 while (delta_stack_nr
) {
2241 unsigned long delta_size
, base_size
= size
;
2247 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2251 * We're probably in deep shit, but let's try to fetch
2252 * the required base anyway from another pack or loose.
2253 * This is costly but should happen only in the presence
2254 * of a corrupted pack, and is better than failing outright.
2256 struct revindex_entry
*revidx
;
2257 const unsigned char *base_sha1
;
2258 revidx
= find_pack_revindex(p
, obj_offset
);
2260 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2261 error("failed to read delta base object %s"
2262 " at offset %"PRIuMAX
" from %s",
2263 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2265 mark_bad_packed_object(p
, base_sha1
);
2266 base
= read_object(base_sha1
, &type
, &base_size
);
2270 i
= --delta_stack_nr
;
2271 obj_offset
= delta_stack
[i
].obj_offset
;
2272 curpos
= delta_stack
[i
].curpos
;
2273 delta_size
= delta_stack
[i
].size
;
2278 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2281 error("failed to unpack compressed delta "
2282 "at offset %"PRIuMAX
" from %s",
2283 (uintmax_t)curpos
, p
->pack_name
);
2288 data
= patch_delta(base
, base_size
,
2289 delta_data
, delta_size
,
2293 * We could not apply the delta; warn the user, but keep going.
2294 * Our failure will be noticed either in the next iteration of
2295 * the loop, or if this is the final delta, in the caller when
2296 * we return NULL. Those code paths will take care of making
2297 * a more explicit warning and retrying with another copy of
2301 error("failed to apply delta");
2309 unuse_pack(&w_curs
);
2313 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2316 const unsigned char *index
= p
->index_data
;
2318 if (open_pack_index(p
))
2320 index
= p
->index_data
;
2322 if (n
>= p
->num_objects
)
2325 if (p
->index_version
== 1) {
2326 return index
+ 24 * n
+ 4;
2329 return index
+ 20 * n
;
2333 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2335 const unsigned char *index
= p
->index_data
;
2337 if (p
->index_version
== 1) {
2338 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2341 index
+= 8 + p
->num_objects
* (20 + 4);
2342 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2343 if (!(off
& 0x80000000))
2345 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2346 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2347 ntohl(*((uint32_t *)(index
+ 4)));
2351 off_t
find_pack_entry_one(const unsigned char *sha1
,
2352 struct packed_git
*p
)
2354 const uint32_t *level1_ofs
= p
->index_data
;
2355 const unsigned char *index
= p
->index_data
;
2356 unsigned hi
, lo
, stride
;
2357 static int use_lookup
= -1;
2358 static int debug_lookup
= -1;
2360 if (debug_lookup
< 0)
2361 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2364 if (open_pack_index(p
))
2366 level1_ofs
= p
->index_data
;
2367 index
= p
->index_data
;
2369 if (p
->index_version
> 1) {
2374 hi
= ntohl(level1_ofs
[*sha1
]);
2375 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2376 if (p
->index_version
> 1) {
2384 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2385 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2388 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2390 int pos
= sha1_entry_pos(index
, stride
, 0,
2391 lo
, hi
, p
->num_objects
, sha1
);
2394 return nth_packed_object_offset(p
, pos
);
2398 unsigned mi
= (lo
+ hi
) / 2;
2399 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2402 printf("lo %u hi %u rg %u mi %u\n",
2403 lo
, hi
, hi
- lo
, mi
);
2405 return nth_packed_object_offset(p
, mi
);
2414 int is_pack_valid(struct packed_git
*p
)
2416 /* An already open pack is known to be valid. */
2417 if (p
->pack_fd
!= -1)
2420 /* If the pack has one window completely covering the
2421 * file size, the pack is known to be valid even if
2422 * the descriptor is not currently open.
2425 struct pack_window
*w
= p
->windows
;
2427 if (!w
->offset
&& w
->len
== p
->pack_size
)
2431 /* Force the pack to open to prove its valid. */
2432 return !open_packed_git(p
);
2435 static int fill_pack_entry(const unsigned char *sha1
,
2436 struct pack_entry
*e
,
2437 struct packed_git
*p
)
2441 if (p
->num_bad_objects
) {
2443 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2444 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2448 offset
= find_pack_entry_one(sha1
, p
);
2453 * We are about to tell the caller where they can locate the
2454 * requested object. We better make sure the packfile is
2455 * still here and can be accessed before supplying that
2456 * answer, as it may have been deleted since the index was
2459 if (!is_pack_valid(p
)) {
2460 warning("packfile %s cannot be accessed", p
->pack_name
);
2465 hashcpy(e
->sha1
, sha1
);
2469 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2471 struct packed_git
*p
;
2473 prepare_packed_git();
2477 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2480 for (p
= packed_git
; p
; p
= p
->next
) {
2481 if (p
== last_found_pack
|| !fill_pack_entry(sha1
, e
, p
))
2484 last_found_pack
= p
;
2490 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2491 struct packed_git
*packs
)
2493 struct packed_git
*p
;
2495 for (p
= packs
; p
; p
= p
->next
) {
2496 if (find_pack_entry_one(sha1
, p
))
2503 static int sha1_loose_object_info(const unsigned char *sha1
,
2504 struct object_info
*oi
)
2507 unsigned long mapsize
, size
;
2513 * If we don't care about type or size, then we don't
2514 * need to look inside the object at all. Note that we
2515 * do not optimize out the stat call, even if the
2516 * caller doesn't care about the disk-size, since our
2517 * return value implicitly indicates whether the
2518 * object even exists.
2520 if (!oi
->typep
&& !oi
->sizep
) {
2522 if (stat_sha1_file(sha1
, &st
) < 0)
2525 *oi
->disk_sizep
= st
.st_size
;
2529 map
= map_sha1_file(sha1
, &mapsize
);
2533 *oi
->disk_sizep
= mapsize
;
2534 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2535 status
= error("unable to unpack %s header",
2537 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2538 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2541 git_inflate_end(&stream
);
2542 munmap(map
, mapsize
);
2544 *oi
->typep
= status
;
2548 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
)
2550 struct cached_object
*co
;
2551 struct pack_entry e
;
2554 co
= find_cached_object(sha1
);
2557 *(oi
->typep
) = co
->type
;
2559 *(oi
->sizep
) = co
->size
;
2561 *(oi
->disk_sizep
) = 0;
2562 oi
->whence
= OI_CACHED
;
2566 if (!find_pack_entry(sha1
, &e
)) {
2567 /* Most likely it's a loose object. */
2568 if (!sha1_loose_object_info(sha1
, oi
)) {
2569 oi
->whence
= OI_LOOSE
;
2573 /* Not a loose object; someone else may have just packed it. */
2574 reprepare_packed_git();
2575 if (!find_pack_entry(sha1
, &e
))
2579 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2581 mark_bad_packed_object(e
.p
, sha1
);
2582 return sha1_object_info_extended(sha1
, oi
);
2583 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2584 oi
->whence
= OI_DBCACHED
;
2586 oi
->whence
= OI_PACKED
;
2587 oi
->u
.packed
.offset
= e
.offset
;
2588 oi
->u
.packed
.pack
= e
.p
;
2589 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2590 rtype
== OBJ_OFS_DELTA
);
2596 /* returns enum object_type or negative */
2597 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2599 enum object_type type
;
2600 struct object_info oi
= {NULL
};
2604 if (sha1_object_info_extended(sha1
, &oi
) < 0)
2609 static void *read_packed_sha1(const unsigned char *sha1
,
2610 enum object_type
*type
, unsigned long *size
)
2612 struct pack_entry e
;
2615 if (!find_pack_entry(sha1
, &e
))
2617 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2620 * We're probably in deep shit, but let's try to fetch
2621 * the required object anyway from another pack or loose.
2622 * This should happen only in the presence of a corrupted
2623 * pack, and is better than failing outright.
2625 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2626 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2627 mark_bad_packed_object(e
.p
, sha1
);
2628 data
= read_object(sha1
, type
, size
);
2633 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2634 unsigned char *sha1
)
2636 struct cached_object
*co
;
2638 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2639 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2641 if (cached_object_alloc
<= cached_object_nr
) {
2642 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2643 cached_objects
= xrealloc(cached_objects
,
2644 sizeof(*cached_objects
) *
2645 cached_object_alloc
);
2647 co
= &cached_objects
[cached_object_nr
++];
2650 co
->buf
= xmalloc(len
);
2651 memcpy(co
->buf
, buf
, len
);
2652 hashcpy(co
->sha1
, sha1
);
2656 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2657 unsigned long *size
)
2659 unsigned long mapsize
;
2661 struct cached_object
*co
;
2663 co
= find_cached_object(sha1
);
2667 return xmemdupz(co
->buf
, co
->size
);
2670 buf
= read_packed_sha1(sha1
, type
, size
);
2673 map
= map_sha1_file(sha1
, &mapsize
);
2675 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2676 munmap(map
, mapsize
);
2679 reprepare_packed_git();
2680 return read_packed_sha1(sha1
, type
, size
);
2684 * This function dies on corrupt objects; the callers who want to
2685 * deal with them should arrange to call read_object() and give error
2686 * messages themselves.
2688 void *read_sha1_file_extended(const unsigned char *sha1
,
2689 enum object_type
*type
,
2690 unsigned long *size
,
2695 const struct packed_git
*p
;
2696 const unsigned char *repl
= (flag
& READ_SHA1_FILE_REPLACE
)
2697 ? lookup_replace_object(sha1
) : sha1
;
2700 data
= read_object(repl
, type
, size
);
2704 if (errno
&& errno
!= ENOENT
)
2705 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2707 /* die if we replaced an object with one that does not exist */
2709 die("replacement %s not found for %s",
2710 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2712 if (has_loose_object(repl
)) {
2713 path
= sha1_file_name(sha1
);
2714 die("loose object %s (stored in %s) is corrupt",
2715 sha1_to_hex(repl
), path
);
2718 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2719 die("packed object %s (stored in %s) is corrupt",
2720 sha1_to_hex(repl
), p
->pack_name
);
2725 void *read_object_with_reference(const unsigned char *sha1
,
2726 const char *required_type_name
,
2727 unsigned long *size
,
2728 unsigned char *actual_sha1_return
)
2730 enum object_type type
, required_type
;
2732 unsigned long isize
;
2733 unsigned char actual_sha1
[20];
2735 required_type
= type_from_string(required_type_name
);
2736 hashcpy(actual_sha1
, sha1
);
2738 int ref_length
= -1;
2739 const char *ref_type
= NULL
;
2741 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2744 if (type
== required_type
) {
2746 if (actual_sha1_return
)
2747 hashcpy(actual_sha1_return
, actual_sha1
);
2750 /* Handle references */
2751 else if (type
== OBJ_COMMIT
)
2753 else if (type
== OBJ_TAG
)
2754 ref_type
= "object ";
2759 ref_length
= strlen(ref_type
);
2761 if (ref_length
+ 40 > isize
||
2762 memcmp(buffer
, ref_type
, ref_length
) ||
2763 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2768 /* Now we have the ID of the referred-to object in
2769 * actual_sha1. Check again. */
2773 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2774 const char *type
, unsigned char *sha1
,
2775 char *hdr
, int *hdrlen
)
2779 /* Generate the header */
2780 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2784 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2785 git_SHA1_Update(&c
, buf
, len
);
2786 git_SHA1_Final(sha1
, &c
);
2790 * Move the just written object into its final resting place.
2791 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2792 * "moving" is only a part of what it does, when no patch between
2793 * master to pu changes the call sites of this function.
2795 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2799 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2801 else if (link(tmpfile
, filename
))
2805 * Coda hack - coda doesn't like cross-directory links,
2806 * so we fall back to a rename, which will mean that it
2807 * won't be able to check collisions, but that's not a
2810 * The same holds for FAT formatted media.
2812 * When this succeeds, we just return. We have nothing
2815 if (ret
&& ret
!= EEXIST
) {
2817 if (!rename(tmpfile
, filename
))
2821 unlink_or_warn(tmpfile
);
2823 if (ret
!= EEXIST
) {
2824 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2826 /* FIXME!!! Collision check here ? */
2830 if (adjust_shared_perm(filename
))
2831 return error("unable to set permission to '%s'", filename
);
2835 static int write_buffer(int fd
, const void *buf
, size_t len
)
2837 if (write_in_full(fd
, buf
, len
) < 0)
2838 return error("file write error (%s)", strerror(errno
));
2842 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2843 unsigned char *sha1
)
2847 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2851 /* Finalize a file on disk, and close it. */
2852 static void close_sha1_file(int fd
)
2854 if (fsync_object_files
)
2855 fsync_or_die(fd
, "sha1 file");
2857 die_errno("error when closing sha1 file");
2860 /* Size of directory component, including the ending '/' */
2861 static inline int directory_size(const char *filename
)
2863 const char *s
= strrchr(filename
, '/');
2866 return s
- filename
+ 1;
2870 * This creates a temporary file in the same directory as the final
2873 * We want to avoid cross-directory filename renames, because those
2874 * can have problems on various filesystems (FAT, NFS, Coda).
2876 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2878 int fd
, dirlen
= directory_size(filename
);
2880 if (dirlen
+ 20 > bufsiz
) {
2881 errno
= ENAMETOOLONG
;
2884 memcpy(buffer
, filename
, dirlen
);
2885 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2886 fd
= git_mkstemp_mode(buffer
, 0444);
2887 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2888 /* Make sure the directory exists */
2889 memcpy(buffer
, filename
, dirlen
);
2890 buffer
[dirlen
-1] = 0;
2891 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
2893 if (adjust_shared_perm(buffer
))
2897 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2898 fd
= git_mkstemp_mode(buffer
, 0444);
2903 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2904 const void *buf
, unsigned long len
, time_t mtime
)
2907 unsigned char compressed
[4096];
2910 unsigned char parano_sha1
[20];
2912 static char tmp_file
[PATH_MAX
];
2914 filename
= sha1_file_name(sha1
);
2915 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2917 if (errno
== EACCES
)
2918 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2920 return error("unable to create temporary file: %s", strerror(errno
));
2924 memset(&stream
, 0, sizeof(stream
));
2925 git_deflate_init(&stream
, zlib_compression_level
);
2926 stream
.next_out
= compressed
;
2927 stream
.avail_out
= sizeof(compressed
);
2930 /* First header.. */
2931 stream
.next_in
= (unsigned char *)hdr
;
2932 stream
.avail_in
= hdrlen
;
2933 while (git_deflate(&stream
, 0) == Z_OK
)
2935 git_SHA1_Update(&c
, hdr
, hdrlen
);
2937 /* Then the data itself.. */
2938 stream
.next_in
= (void *)buf
;
2939 stream
.avail_in
= len
;
2941 unsigned char *in0
= stream
.next_in
;
2942 ret
= git_deflate(&stream
, Z_FINISH
);
2943 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2944 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2945 die("unable to write sha1 file");
2946 stream
.next_out
= compressed
;
2947 stream
.avail_out
= sizeof(compressed
);
2948 } while (ret
== Z_OK
);
2950 if (ret
!= Z_STREAM_END
)
2951 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2952 ret
= git_deflate_end_gently(&stream
);
2954 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2955 git_SHA1_Final(parano_sha1
, &c
);
2956 if (hashcmp(sha1
, parano_sha1
) != 0)
2957 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2959 close_sha1_file(fd
);
2964 utb
.modtime
= mtime
;
2965 if (utime(tmp_file
, &utb
) < 0)
2966 warning("failed utime() on %s: %s",
2967 tmp_file
, strerror(errno
));
2970 return move_temp_to_file(tmp_file
, filename
);
2973 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2975 unsigned char sha1
[20];
2979 /* Normally if we have it in the pack then we do not bother writing
2980 * it out into .git/objects/??/?{38} file.
2982 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2984 hashcpy(returnsha1
, sha1
);
2985 if (has_sha1_file(sha1
))
2987 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2990 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2994 enum object_type type
;
2999 if (has_loose_object(sha1
))
3001 buf
= read_packed_sha1(sha1
, &type
, &len
);
3003 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3004 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
3005 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3011 int has_pack_index(const unsigned char *sha1
)
3014 if (stat(sha1_pack_index_name(sha1
), &st
))
3019 int has_sha1_pack(const unsigned char *sha1
)
3021 struct pack_entry e
;
3022 return find_pack_entry(sha1
, &e
);
3025 int has_sha1_file(const unsigned char *sha1
)
3027 struct pack_entry e
;
3029 if (find_pack_entry(sha1
, &e
))
3031 if (has_loose_object(sha1
))
3033 reprepare_packed_git();
3034 return find_pack_entry(sha1
, &e
);
3037 static void check_tree(const void *buf
, size_t size
)
3039 struct tree_desc desc
;
3040 struct name_entry entry
;
3042 init_tree_desc(&desc
, buf
, size
);
3043 while (tree_entry(&desc
, &entry
))
3045 * tree_entry() will die() on malformed entries */
3049 static void check_commit(const void *buf
, size_t size
)
3052 memset(&c
, 0, sizeof(c
));
3053 if (parse_commit_buffer(&c
, buf
, size
))
3054 die("corrupt commit");
3057 static void check_tag(const void *buf
, size_t size
)
3060 memset(&t
, 0, sizeof(t
));
3061 if (parse_tag_buffer(&t
, buf
, size
))
3065 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3066 enum object_type type
,
3067 const char *path
, unsigned flags
)
3069 int ret
, re_allocated
= 0;
3070 int write_object
= flags
& HASH_WRITE_OBJECT
;
3076 * Convert blobs to git internal format
3078 if ((type
== OBJ_BLOB
) && path
) {
3079 struct strbuf nbuf
= STRBUF_INIT
;
3080 if (convert_to_git(path
, buf
, size
, &nbuf
,
3081 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3082 buf
= strbuf_detach(&nbuf
, &size
);
3086 if (flags
& HASH_FORMAT_CHECK
) {
3087 if (type
== OBJ_TREE
)
3088 check_tree(buf
, size
);
3089 if (type
== OBJ_COMMIT
)
3090 check_commit(buf
, size
);
3091 if (type
== OBJ_TAG
)
3092 check_tag(buf
, size
);
3096 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3098 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3104 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3105 const char *path
, unsigned flags
)
3107 struct strbuf sbuf
= STRBUF_INIT
;
3110 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3111 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3114 strbuf_release(&sbuf
);
3118 #define SMALL_FILE_SIZE (32*1024)
3120 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3121 enum object_type type
, const char *path
,
3127 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
3128 } else if (size
<= SMALL_FILE_SIZE
) {
3129 char *buf
= xmalloc(size
);
3130 if (size
== read_in_full(fd
, buf
, size
))
3131 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3133 ret
= error("short read %s", strerror(errno
));
3136 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3137 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3144 * This creates one packfile per large blob unless bulk-checkin
3145 * machinery is "plugged".
3147 * This also bypasses the usual "convert-to-git" dance, and that is on
3148 * purpose. We could write a streaming version of the converting
3149 * functions and insert that before feeding the data to fast-import
3150 * (or equivalent in-core API described above). However, that is
3151 * somewhat complicated, as we do not know the size of the filter
3152 * result, which we need to know beforehand when writing a git object.
3153 * Since the primary motivation for trying to stream from the working
3154 * tree file and to avoid mmaping it in core is to deal with large
3155 * binary blobs, they generally do not want to get any conversion, and
3156 * callers should avoid this code path when filters are requested.
3158 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3159 enum object_type type
, const char *path
,
3162 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3165 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3166 enum object_type type
, const char *path
, unsigned flags
)
3169 size_t size
= xsize_t(st
->st_size
);
3171 if (!S_ISREG(st
->st_mode
))
3172 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3173 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3174 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
3175 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
3177 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
3182 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3185 struct strbuf sb
= STRBUF_INIT
;
3187 switch (st
->st_mode
& S_IFMT
) {
3189 fd
= open(path
, O_RDONLY
);
3191 return error("open(\"%s\"): %s", path
,
3193 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3194 return error("%s: failed to insert into database",
3198 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3199 char *errstr
= strerror(errno
);
3200 return error("readlink(\"%s\"): %s", path
,
3203 if (!(flags
& HASH_WRITE_OBJECT
))
3204 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3205 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3206 return error("%s: failed to insert into database",
3208 strbuf_release(&sb
);
3211 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3213 return error("%s: unsupported file type", path
);
3218 int read_pack_header(int fd
, struct pack_header
*header
)
3220 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3221 /* "eof before pack header was fully read" */
3222 return PH_ERROR_EOF
;
3224 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3225 /* "protocol error (pack signature mismatch detected)" */
3226 return PH_ERROR_PACK_SIGNATURE
;
3227 if (!pack_version_ok(header
->hdr_version
))
3228 /* "protocol error (pack version unsupported)" */
3229 return PH_ERROR_PROTOCOL
;
3233 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3235 enum object_type type
= sha1_object_info(sha1
, NULL
);
3237 die("%s is not a valid object", sha1_to_hex(sha1
));
3239 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),