2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
16 #include "run-command.h"
19 #include "tree-walk.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
38 const unsigned char null_sha1
[20];
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
46 static struct cached_object
{
47 unsigned char sha1
[20];
48 enum object_type type
;
52 static int cached_object_nr
, cached_object_alloc
;
54 static struct cached_object empty_tree
= {
55 EMPTY_TREE_SHA1_BIN_LITERAL
,
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
67 static struct packed_git
*last_found_pack
;
69 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
72 struct cached_object
*co
= cached_objects
;
74 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
75 if (!hashcmp(co
->sha1
, sha1
))
78 if (!hashcmp(sha1
, empty_tree
.sha1
))
83 int mkdir_in_gitdir(const char *path
)
85 if (mkdir(path
, 0777)) {
86 int saved_errno
= errno
;
88 struct strbuf sb
= STRBUF_INIT
;
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
99 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
100 strbuf_readlink(&sb
, path
, st
.st_size
) ||
101 !is_absolute_path(sb
.buf
) ||
102 mkdir(sb
.buf
, 0777)) {
109 return adjust_shared_perm(path
);
112 enum scld_error
safe_create_leading_directories(char *path
)
114 char *next_component
= path
+ offset_1st_component(path
);
115 enum scld_error ret
= SCLD_OK
;
117 while (ret
== SCLD_OK
&& next_component
) {
119 char *slash
= next_component
, slash_character
;
121 while (*slash
&& !is_dir_sep(*slash
))
127 next_component
= slash
+ 1;
128 while (is_dir_sep(*next_component
))
130 if (!*next_component
)
133 slash_character
= *slash
;
135 if (!stat(path
, &st
)) {
137 if (!S_ISDIR(st
.st_mode
))
139 } else if (mkdir(path
, 0777)) {
140 if (errno
== EEXIST
&&
141 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
142 ; /* somebody created it since we checked */
143 else if (errno
== ENOENT
)
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
156 } else if (adjust_shared_perm(path
)) {
159 *slash
= slash_character
;
164 enum scld_error
safe_create_leading_directories_const(const char *path
)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf
= xstrdup(path
);
168 enum scld_error result
= safe_create_leading_directories(buf
);
173 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
176 for (i
= 0; i
< 20; i
++) {
177 static char hex
[] = "0123456789abcdef";
178 unsigned int val
= sha1
[i
];
179 char *pos
= pathbuf
+ i
*2 + (i
> 0);
180 *pos
++ = hex
[val
>> 4];
181 *pos
= hex
[val
& 0xf];
185 const char *sha1_file_name(const unsigned char *sha1
)
187 static char buf
[PATH_MAX
];
191 objdir
= get_object_directory();
192 len
= strlen(objdir
);
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len
+ 43 > PATH_MAX
)
196 die("insanely long object directory %s", objdir
);
197 memcpy(buf
, objdir
, len
);
201 fill_sha1_path(buf
+ len
+ 1, sha1
);
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
210 static char *sha1_get_pack_name(const unsigned char *sha1
,
211 char **name
, char **base
, const char *which
)
213 static const char hex
[] = "0123456789abcdef";
218 const char *sha1_file_directory
= get_object_directory();
219 int len
= strlen(sha1_file_directory
);
220 *base
= xmalloc(len
+ 60);
221 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory
, which
);
223 *name
= *base
+ len
+ 11;
228 for (i
= 0; i
< 20; i
++) {
229 unsigned int val
= *sha1
++;
230 *buf
++ = hex
[val
>> 4];
231 *buf
++ = hex
[val
& 0xf];
237 char *sha1_pack_name(const unsigned char *sha1
)
239 static char *name
, *base
;
241 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
244 char *sha1_pack_index_name(const unsigned char *sha1
)
246 static char *name
, *base
;
248 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
251 struct alternate_object_database
*alt_odb_list
;
252 static struct alternate_object_database
**alt_odb_tail
;
255 * Prepare alternate object database registry.
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
269 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
270 int depth
, const char *normalized_objdir
)
272 struct alternate_object_database
*ent
;
273 struct alternate_object_database
*alt
;
275 struct strbuf pathbuf
= STRBUF_INIT
;
277 if (!is_absolute_path(entry
) && relative_base
) {
278 strbuf_addstr(&pathbuf
, real_path(relative_base
));
279 strbuf_addch(&pathbuf
, '/');
281 strbuf_addstr(&pathbuf
, entry
);
283 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
285 pfxlen
= strlen(pathbuf
.buf
);
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
291 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
294 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent
= xmalloc(sizeof(*ent
) + entlen
);
296 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
297 strbuf_release(&pathbuf
);
299 ent
->name
= ent
->base
+ pfxlen
+ 1;
300 ent
->base
[pfxlen
+ 3] = '/';
301 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent
->base
)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
315 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
316 if (pfxlen
== alt
->name
- alt
->base
- 1 &&
317 !memcmp(ent
->base
, alt
->base
, pfxlen
)) {
322 if (!strcmp_icase(ent
->base
, normalized_objdir
)) {
327 /* add the alternate entry */
329 alt_odb_tail
= &(ent
->next
);
332 /* recursively add alternates */
333 read_info_alternates(ent
->base
, depth
+ 1);
335 ent
->base
[pfxlen
] = '/';
340 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
341 const char *relative_base
, int depth
)
343 struct string_list entries
= STRING_LIST_INIT_NODUP
;
346 struct strbuf objdirbuf
= STRBUF_INIT
;
349 error("%s: ignoring alternate object stores, nesting too deep.",
354 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
355 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
357 alt_copy
= xmemdupz(alt
, len
);
358 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
359 for (i
= 0; i
< entries
.nr
; i
++) {
360 const char *entry
= entries
.items
[i
].string
;
361 if (entry
[0] == '\0' || entry
[0] == '#')
363 if (!is_absolute_path(entry
) && depth
) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base
, entry
);
367 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
370 string_list_clear(&entries
, 0);
372 strbuf_release(&objdirbuf
);
375 void read_info_alternates(const char * relative_base
, int depth
)
383 path
= xstrfmt("%s/info/alternates", relative_base
);
384 fd
= git_open_noatime(path
);
388 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
392 mapsz
= xsize_t(st
.st_size
);
393 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
396 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
401 void add_to_alternates_file(const char *reference
)
403 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
404 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
405 char *alt
= mkpath("%s\n", reference
);
406 write_or_die(fd
, alt
, strlen(alt
));
407 if (commit_lock_file(lock
))
408 die("could not close alternates file");
410 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
413 int foreach_alt_odb(alt_odb_fn fn
, void *cb
)
415 struct alternate_object_database
*ent
;
419 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
) {
427 void prepare_alt_odb(void)
434 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
437 alt_odb_tail
= &alt_odb_list
;
438 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
440 read_info_alternates(get_object_directory(), 0);
443 /* Returns 1 if we have successfully freshened the file, 0 otherwise. */
444 static int freshen_file(const char *fn
)
447 t
.actime
= t
.modtime
= time(NULL
);
448 return !utime(fn
, &t
);
452 * All of the check_and_freshen functions return 1 if the file exists and was
453 * freshened (if freshening was requested), 0 otherwise. If they return
454 * 0, you should not assume that it is safe to skip a write of the object (it
455 * either does not exist on disk, or has a stale mtime and may be subject to
458 static int check_and_freshen_file(const char *fn
, int freshen
)
460 if (access(fn
, F_OK
))
462 if (freshen
&& !freshen_file(fn
))
467 static int check_and_freshen_local(const unsigned char *sha1
, int freshen
)
469 return check_and_freshen_file(sha1_file_name(sha1
), freshen
);
472 static int check_and_freshen_nonlocal(const unsigned char *sha1
, int freshen
)
474 struct alternate_object_database
*alt
;
476 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
477 fill_sha1_path(alt
->name
, sha1
);
478 if (check_and_freshen_file(alt
->base
, freshen
))
484 static int check_and_freshen(const unsigned char *sha1
, int freshen
)
486 return check_and_freshen_local(sha1
, freshen
) ||
487 check_and_freshen_nonlocal(sha1
, freshen
);
490 int has_loose_object_nonlocal(const unsigned char *sha1
)
492 return check_and_freshen_nonlocal(sha1
, 0);
495 static int has_loose_object(const unsigned char *sha1
)
497 return check_and_freshen(sha1
, 0);
500 static unsigned int pack_used_ctr
;
501 static unsigned int pack_mmap_calls
;
502 static unsigned int peak_pack_open_windows
;
503 static unsigned int pack_open_windows
;
504 static unsigned int pack_open_fds
;
505 static unsigned int pack_max_fds
;
506 static size_t peak_pack_mapped
;
507 static size_t pack_mapped
;
508 struct packed_git
*packed_git
;
510 void pack_report(void)
513 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
514 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
515 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
516 sz_fmt(getpagesize()),
517 sz_fmt(packed_git_window_size
),
518 sz_fmt(packed_git_limit
));
520 "pack_report: pack_used_ctr = %10u\n"
521 "pack_report: pack_mmap_calls = %10u\n"
522 "pack_report: pack_open_windows = %10u / %10u\n"
523 "pack_report: pack_mapped = "
524 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
527 pack_open_windows
, peak_pack_open_windows
,
528 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
532 * Open and mmap the index file at path, perform a couple of
533 * consistency checks, then record its information to p. Return 0 on
536 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
539 struct pack_idx_header
*hdr
;
541 uint32_t version
, nr
, i
, *index
;
542 int fd
= git_open_noatime(path
);
547 if (fstat(fd
, &st
)) {
551 idx_size
= xsize_t(st
.st_size
);
552 if (idx_size
< 4 * 256 + 20 + 20) {
554 return error("index file %s is too small", path
);
556 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
560 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
561 version
= ntohl(hdr
->idx_version
);
562 if (version
< 2 || version
> 2) {
563 munmap(idx_map
, idx_size
);
564 return error("index file %s is version %"PRIu32
565 " and is not supported by this binary"
566 " (try upgrading GIT to a newer version)",
575 index
+= 2; /* skip index header */
576 for (i
= 0; i
< 256; i
++) {
577 uint32_t n
= ntohl(index
[i
]);
579 munmap(idx_map
, idx_size
);
580 return error("non-monotonic index %s", path
);
588 * - 256 index entries 4 bytes each
589 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
590 * - 20-byte SHA1 of the packfile
591 * - 20-byte SHA1 file checksum
593 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
594 munmap(idx_map
, idx_size
);
595 return error("wrong index v1 file size in %s", path
);
597 } else if (version
== 2) {
600 * - 8 bytes of header
601 * - 256 index entries 4 bytes each
602 * - 20-byte sha1 entry * nr
603 * - 4-byte crc entry * nr
604 * - 4-byte offset entry * nr
605 * - 20-byte SHA1 of the packfile
606 * - 20-byte SHA1 file checksum
607 * And after the 4-byte offset table might be a
608 * variable sized table containing 8-byte entries
609 * for offsets larger than 2^31.
611 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
612 unsigned long max_size
= min_size
;
614 max_size
+= (nr
- 1)*8;
615 if (idx_size
< min_size
|| idx_size
> max_size
) {
616 munmap(idx_map
, idx_size
);
617 return error("wrong index v2 file size in %s", path
);
619 if (idx_size
!= min_size
&&
621 * make sure we can deal with large pack offsets.
622 * 31-bit signed offset won't be enough, neither
623 * 32-bit unsigned one will be.
625 (sizeof(off_t
) <= 4)) {
626 munmap(idx_map
, idx_size
);
627 return error("pack too large for current definition of off_t in %s", path
);
631 p
->index_version
= version
;
632 p
->index_data
= idx_map
;
633 p
->index_size
= idx_size
;
638 int open_pack_index(struct packed_git
*p
)
646 idx_name
= xstrdup(p
->pack_name
);
647 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
648 ret
= check_packed_git_idx(idx_name
, p
);
653 static void scan_windows(struct packed_git
*p
,
654 struct packed_git
**lru_p
,
655 struct pack_window
**lru_w
,
656 struct pack_window
**lru_l
)
658 struct pack_window
*w
, *w_l
;
660 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
662 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
672 static int unuse_one_window(struct packed_git
*current
)
674 struct packed_git
*p
, *lru_p
= NULL
;
675 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
678 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
679 for (p
= packed_git
; p
; p
= p
->next
)
680 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
682 munmap(lru_w
->base
, lru_w
->len
);
683 pack_mapped
-= lru_w
->len
;
685 lru_l
->next
= lru_w
->next
;
687 lru_p
->windows
= lru_w
->next
;
695 void release_pack_memory(size_t need
)
697 size_t cur
= pack_mapped
;
698 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
702 static void mmap_limit_check(size_t length
)
704 static size_t limit
= 0;
706 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
711 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
712 (uintmax_t)length
, (uintmax_t)limit
);
715 void *xmmap_gently(void *start
, size_t length
,
716 int prot
, int flags
, int fd
, off_t offset
)
720 mmap_limit_check(length
);
721 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
722 if (ret
== MAP_FAILED
) {
725 release_pack_memory(length
);
726 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
731 void *xmmap(void *start
, size_t length
,
732 int prot
, int flags
, int fd
, off_t offset
)
734 void *ret
= xmmap_gently(start
, length
, prot
, flags
, fd
, offset
);
735 if (ret
== MAP_FAILED
)
736 die_errno("mmap failed");
740 void close_pack_windows(struct packed_git
*p
)
743 struct pack_window
*w
= p
->windows
;
746 die("pack '%s' still has open windows to it",
748 munmap(w
->base
, w
->len
);
749 pack_mapped
-= w
->len
;
751 p
->windows
= w
->next
;
757 * The LRU pack is the one with the oldest MRU window, preferring packs
758 * with no used windows, or the oldest mtime if it has no windows allocated.
760 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
762 struct pack_window
*w
, *this_mru_w
;
763 int has_windows_inuse
= 0;
766 * Reject this pack if it has windows and the previously selected
767 * one does not. If this pack does not have windows, reject
768 * it if the pack file is newer than the previously selected one.
770 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
773 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
775 * Reject this pack if any of its windows are in use,
776 * but the previously selected pack did not have any
777 * inuse windows. Otherwise, record that this pack
778 * has windows in use.
781 if (*accept_windows_inuse
)
782 has_windows_inuse
= 1;
787 if (w
->last_used
> this_mru_w
->last_used
)
791 * Reject this pack if it has windows that have been
792 * used more recently than the previously selected pack.
793 * If the previously selected pack had windows inuse and
794 * we have not encountered a window in this pack that is
795 * inuse, skip this check since we prefer a pack with no
796 * inuse windows to one that has inuse windows.
798 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
799 this_mru_w
->last_used
> (*mru_w
)->last_used
)
808 *accept_windows_inuse
= has_windows_inuse
;
811 static int close_one_pack(void)
813 struct packed_git
*p
, *lru_p
= NULL
;
814 struct pack_window
*mru_w
= NULL
;
815 int accept_windows_inuse
= 1;
817 for (p
= packed_git
; p
; p
= p
->next
) {
818 if (p
->pack_fd
== -1)
820 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
824 close(lru_p
->pack_fd
);
833 void unuse_pack(struct pack_window
**w_cursor
)
835 struct pack_window
*w
= *w_cursor
;
842 void close_pack_index(struct packed_git
*p
)
845 munmap((void *)p
->index_data
, p
->index_size
);
846 p
->index_data
= NULL
;
851 * This is used by git-repack in case a newly created pack happens to
852 * contain the same set of objects as an existing one. In that case
853 * the resulting file might be different even if its name would be the
854 * same. It is best to close any reference to the old pack before it is
855 * replaced on disk. Of course no index pointers or windows for given pack
856 * must subsist at this point. If ever objects from this pack are requested
857 * again, the new version of the pack will be reinitialized through
858 * reprepare_packed_git().
860 void free_pack_by_name(const char *pack_name
)
862 struct packed_git
*p
, **pp
= &packed_git
;
866 if (strcmp(pack_name
, p
->pack_name
) == 0) {
867 clear_delta_base_cache();
868 close_pack_windows(p
);
869 if (p
->pack_fd
!= -1) {
874 free(p
->bad_object_sha1
);
876 if (last_found_pack
== p
)
877 last_found_pack
= NULL
;
885 static unsigned int get_max_fd_limit(void)
891 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
898 long open_max
= sysconf(_SC_OPEN_MAX
);
902 * Otherwise, we got -1 for one of the two
905 * (1) sysconf() did not understand _SC_OPEN_MAX
906 * and signaled an error with -1; or
907 * (2) sysconf() said there is no limit.
909 * We _could_ clear errno before calling sysconf() to
910 * tell these two cases apart and return a huge number
911 * in the latter case to let the caller cap it to a
912 * value that is not so selfish, but letting the
913 * fallback OPEN_MAX codepath take care of these cases
922 return 1; /* see the caller ;-) */
927 * Do not call this directly as this leaks p->pack_fd on error return;
928 * call open_packed_git() instead.
930 static int open_packed_git_1(struct packed_git
*p
)
933 struct pack_header hdr
;
934 unsigned char sha1
[20];
935 unsigned char *idx_sha1
;
938 if (!p
->index_data
&& open_pack_index(p
))
939 return error("packfile %s index unavailable", p
->pack_name
);
942 unsigned int max_fds
= get_max_fd_limit();
944 /* Save 3 for stdin/stdout/stderr, 22 for work */
946 pack_max_fds
= max_fds
- 25;
951 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
954 p
->pack_fd
= git_open_noatime(p
->pack_name
);
955 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
959 /* If we created the struct before we had the pack we lack size. */
961 if (!S_ISREG(st
.st_mode
))
962 return error("packfile %s not a regular file", p
->pack_name
);
963 p
->pack_size
= st
.st_size
;
964 } else if (p
->pack_size
!= st
.st_size
)
965 return error("packfile %s size changed", p
->pack_name
);
967 /* We leave these file descriptors open with sliding mmap;
968 * there is no point keeping them open across exec(), though.
970 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
972 return error("cannot determine file descriptor flags");
973 fd_flag
|= FD_CLOEXEC
;
974 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
975 return error("cannot set FD_CLOEXEC");
977 /* Verify we recognize this pack file format. */
978 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
979 return error("file %s is far too short to be a packfile", p
->pack_name
);
980 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
981 return error("file %s is not a GIT packfile", p
->pack_name
);
982 if (!pack_version_ok(hdr
.hdr_version
))
983 return error("packfile %s is version %"PRIu32
" and not"
984 " supported (try upgrading GIT to a newer version)",
985 p
->pack_name
, ntohl(hdr
.hdr_version
));
987 /* Verify the pack matches its index. */
988 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
989 return error("packfile %s claims to have %"PRIu32
" objects"
990 " while index indicates %"PRIu32
" objects",
991 p
->pack_name
, ntohl(hdr
.hdr_entries
),
993 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
994 return error("end of packfile %s is unavailable", p
->pack_name
);
995 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
996 return error("packfile %s signature is unavailable", p
->pack_name
);
997 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
998 if (hashcmp(sha1
, idx_sha1
))
999 return error("packfile %s does not match index", p
->pack_name
);
1003 static int open_packed_git(struct packed_git
*p
)
1005 if (!open_packed_git_1(p
))
1007 if (p
->pack_fd
!= -1) {
1015 static int in_window(struct pack_window
*win
, off_t offset
)
1017 /* We must promise at least 20 bytes (one hash) after the
1018 * offset is available from this window, otherwise the offset
1019 * is not actually in this window and a different window (which
1020 * has that one hash excess) must be used. This is to support
1021 * the object header and delta base parsing routines below.
1023 off_t win_off
= win
->offset
;
1024 return win_off
<= offset
1025 && (offset
+ 20) <= (win_off
+ win
->len
);
1028 unsigned char *use_pack(struct packed_git
*p
,
1029 struct pack_window
**w_cursor
,
1031 unsigned long *left
)
1033 struct pack_window
*win
= *w_cursor
;
1035 /* Since packfiles end in a hash of their content and it's
1036 * pointless to ask for an offset into the middle of that
1037 * hash, and the in_window function above wouldn't match
1038 * don't allow an offset too close to the end of the file.
1040 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
1041 die("packfile %s cannot be accessed", p
->pack_name
);
1042 if (offset
> (p
->pack_size
- 20))
1043 die("offset beyond end of packfile (truncated pack?)");
1045 die("offset before end of packfile (broken .idx?)");
1047 if (!win
|| !in_window(win
, offset
)) {
1050 for (win
= p
->windows
; win
; win
= win
->next
) {
1051 if (in_window(win
, offset
))
1055 size_t window_align
= packed_git_window_size
/ 2;
1058 if (p
->pack_fd
== -1 && open_packed_git(p
))
1059 die("packfile %s cannot be accessed", p
->pack_name
);
1061 win
= xcalloc(1, sizeof(*win
));
1062 win
->offset
= (offset
/ window_align
) * window_align
;
1063 len
= p
->pack_size
- win
->offset
;
1064 if (len
> packed_git_window_size
)
1065 len
= packed_git_window_size
;
1066 win
->len
= (size_t)len
;
1067 pack_mapped
+= win
->len
;
1068 while (packed_git_limit
< pack_mapped
1069 && unuse_one_window(p
))
1071 win
->base
= xmmap(NULL
, win
->len
,
1072 PROT_READ
, MAP_PRIVATE
,
1073 p
->pack_fd
, win
->offset
);
1074 if (win
->base
== MAP_FAILED
)
1075 die("packfile %s cannot be mapped: %s",
1078 if (!win
->offset
&& win
->len
== p
->pack_size
1079 && !p
->do_not_close
) {
1085 pack_open_windows
++;
1086 if (pack_mapped
> peak_pack_mapped
)
1087 peak_pack_mapped
= pack_mapped
;
1088 if (pack_open_windows
> peak_pack_open_windows
)
1089 peak_pack_open_windows
= pack_open_windows
;
1090 win
->next
= p
->windows
;
1094 if (win
!= *w_cursor
) {
1095 win
->last_used
= pack_used_ctr
++;
1099 offset
-= win
->offset
;
1101 *left
= win
->len
- xsize_t(offset
);
1102 return win
->base
+ offset
;
1105 static struct packed_git
*alloc_packed_git(int extra
)
1107 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1108 memset(p
, 0, sizeof(*p
));
1113 static void try_to_free_pack_memory(size_t size
)
1115 release_pack_memory(size
);
1118 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1120 static int have_set_try_to_free_routine
;
1122 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1124 if (!have_set_try_to_free_routine
) {
1125 have_set_try_to_free_routine
= 1;
1126 set_try_to_free_routine(try_to_free_pack_memory
);
1130 * Make sure a corresponding .pack file exists and that
1131 * the index looks sane.
1133 path_len
-= strlen(".idx");
1138 memcpy(p
->pack_name
, path
, path_len
);
1140 strcpy(p
->pack_name
+ path_len
, ".keep");
1141 if (!access(p
->pack_name
, F_OK
))
1144 strcpy(p
->pack_name
+ path_len
, ".pack");
1145 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1150 /* ok, it looks sane as far as we can check without
1151 * actually mapping the pack file.
1153 p
->pack_size
= st
.st_size
;
1154 p
->pack_local
= local
;
1155 p
->mtime
= st
.st_mtime
;
1156 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1161 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1163 const char *path
= sha1_pack_name(sha1
);
1164 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1166 strcpy(p
->pack_name
, path
);
1167 hashcpy(p
->sha1
, sha1
);
1168 if (check_packed_git_idx(idx_path
, p
)) {
1176 void install_packed_git(struct packed_git
*pack
)
1178 if (pack
->pack_fd
!= -1)
1181 pack
->next
= packed_git
;
1185 void (*report_garbage
)(const char *desc
, const char *path
);
1187 static void report_helper(const struct string_list
*list
,
1188 int seen_bits
, int first
, int last
)
1191 switch (seen_bits
) {
1193 msg
= "no corresponding .idx or .pack";
1196 msg
= "no corresponding .idx";
1199 msg
= "no corresponding .pack";
1204 for (; first
< last
; first
++)
1205 report_garbage(msg
, list
->items
[first
].string
);
1208 static void report_pack_garbage(struct string_list
*list
)
1210 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1212 if (!report_garbage
)
1215 string_list_sort(list
);
1217 for (i
= 0; i
< list
->nr
; i
++) {
1218 const char *path
= list
->items
[i
].string
;
1219 if (baselen
!= -1 &&
1220 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1221 report_helper(list
, seen_bits
, first
, i
);
1225 if (baselen
== -1) {
1226 const char *dot
= strrchr(path
, '.');
1228 report_garbage("garbage found", path
);
1231 baselen
= dot
- path
+ 1;
1234 if (!strcmp(path
+ baselen
, "pack"))
1236 else if (!strcmp(path
+ baselen
, "idx"))
1239 report_helper(list
, seen_bits
, first
, list
->nr
);
1242 static void prepare_packed_git_one(char *objdir
, int local
)
1244 struct strbuf path
= STRBUF_INIT
;
1248 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1250 strbuf_addstr(&path
, objdir
);
1251 strbuf_addstr(&path
, "/pack");
1252 dir
= opendir(path
.buf
);
1254 if (errno
!= ENOENT
)
1255 error("unable to open object pack directory: %s: %s",
1256 path
.buf
, strerror(errno
));
1257 strbuf_release(&path
);
1260 strbuf_addch(&path
, '/');
1261 dirnamelen
= path
.len
;
1262 while ((de
= readdir(dir
)) != NULL
) {
1263 struct packed_git
*p
;
1266 if (is_dot_or_dotdot(de
->d_name
))
1269 strbuf_setlen(&path
, dirnamelen
);
1270 strbuf_addstr(&path
, de
->d_name
);
1272 base_len
= path
.len
;
1273 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1274 /* Don't reopen a pack we already have. */
1275 for (p
= packed_git
; p
; p
= p
->next
) {
1277 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1279 !memcmp(p
->pack_name
, path
.buf
, len
))
1284 * See if it really is a valid .idx file with
1285 * corresponding .pack file that we can map.
1287 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1288 install_packed_git(p
);
1291 if (!report_garbage
)
1294 if (ends_with(de
->d_name
, ".idx") ||
1295 ends_with(de
->d_name
, ".pack") ||
1296 ends_with(de
->d_name
, ".bitmap") ||
1297 ends_with(de
->d_name
, ".keep"))
1298 string_list_append(&garbage
, path
.buf
);
1300 report_garbage("garbage found", path
.buf
);
1303 report_pack_garbage(&garbage
);
1304 string_list_clear(&garbage
, 0);
1305 strbuf_release(&path
);
1308 static int sort_pack(const void *a_
, const void *b_
)
1310 struct packed_git
*a
= *((struct packed_git
**)a_
);
1311 struct packed_git
*b
= *((struct packed_git
**)b_
);
1315 * Local packs tend to contain objects specific to our
1316 * variant of the project than remote ones. In addition,
1317 * remote ones could be on a network mounted filesystem.
1318 * Favor local ones for these reasons.
1320 st
= a
->pack_local
- b
->pack_local
;
1325 * Younger packs tend to contain more recent objects,
1326 * and more recent objects tend to get accessed more
1329 if (a
->mtime
< b
->mtime
)
1331 else if (a
->mtime
== b
->mtime
)
1336 static void rearrange_packed_git(void)
1338 struct packed_git
**ary
, *p
;
1341 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1346 /* prepare an array of packed_git for easier sorting */
1347 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1348 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1351 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1353 /* link them back again */
1354 for (i
= 0; i
< n
- 1; i
++)
1355 ary
[i
]->next
= ary
[i
+ 1];
1356 ary
[n
- 1]->next
= NULL
;
1357 packed_git
= ary
[0];
1362 static int prepare_packed_git_run_once
= 0;
1363 void prepare_packed_git(void)
1365 struct alternate_object_database
*alt
;
1367 if (prepare_packed_git_run_once
)
1369 prepare_packed_git_one(get_object_directory(), 1);
1371 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1373 prepare_packed_git_one(alt
->base
, 0);
1374 alt
->name
[-1] = '/';
1376 rearrange_packed_git();
1377 prepare_packed_git_run_once
= 1;
1380 void reprepare_packed_git(void)
1382 prepare_packed_git_run_once
= 0;
1383 prepare_packed_git();
1386 static void mark_bad_packed_object(struct packed_git
*p
,
1387 const unsigned char *sha1
)
1390 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1391 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1393 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1394 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1395 p
->num_bad_objects
++;
1398 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1400 struct packed_git
*p
;
1403 for (p
= packed_git
; p
; p
= p
->next
)
1404 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1405 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1411 * With an in-core object data in "map", rehash it to make sure the
1412 * object name actually matches "sha1" to detect object corruption.
1413 * With "map" == NULL, try reading the object named with "sha1" using
1414 * the streaming interface and rehash it to do the same.
1416 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1417 unsigned long size
, const char *type
)
1419 unsigned char real_sha1
[20];
1420 enum object_type obj_type
;
1421 struct git_istream
*st
;
1427 hash_sha1_file(map
, size
, type
, real_sha1
);
1428 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1431 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1435 /* Generate the header */
1436 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1440 git_SHA1_Update(&c
, hdr
, hdrlen
);
1442 char buf
[1024 * 16];
1443 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1451 git_SHA1_Update(&c
, buf
, readlen
);
1453 git_SHA1_Final(real_sha1
, &c
);
1455 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1458 int git_open_noatime(const char *name
)
1460 static int sha1_file_open_flag
= O_NOATIME
;
1463 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1467 /* Might the failure be due to O_NOATIME? */
1468 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1469 sha1_file_open_flag
= 0;
1477 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1479 struct alternate_object_database
*alt
;
1481 if (!lstat(sha1_file_name(sha1
), st
))
1486 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1487 fill_sha1_path(alt
->name
, sha1
);
1488 if (!lstat(alt
->base
, st
))
1495 static int open_sha1_file(const unsigned char *sha1
)
1498 struct alternate_object_database
*alt
;
1499 int most_interesting_errno
;
1501 fd
= git_open_noatime(sha1_file_name(sha1
));
1504 most_interesting_errno
= errno
;
1507 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1508 fill_sha1_path(alt
->name
, sha1
);
1509 fd
= git_open_noatime(alt
->base
);
1512 if (most_interesting_errno
== ENOENT
)
1513 most_interesting_errno
= errno
;
1515 errno
= most_interesting_errno
;
1519 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1524 fd
= open_sha1_file(sha1
);
1529 if (!fstat(fd
, &st
)) {
1530 *size
= xsize_t(st
.st_size
);
1532 /* mmap() is forbidden on empty files */
1533 error("object file %s is empty", sha1_file_name(sha1
));
1536 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1543 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1544 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1547 unsigned long size
, c
;
1548 unsigned long used
= 0;
1551 *type
= (c
>> 4) & 7;
1555 if (len
<= used
|| bitsizeof(long) <= shift
) {
1556 error("bad object header");
1561 size
+= (c
& 0x7f) << shift
;
1568 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1570 /* Get the data stream */
1571 memset(stream
, 0, sizeof(*stream
));
1572 stream
->next_in
= map
;
1573 stream
->avail_in
= mapsize
;
1574 stream
->next_out
= buffer
;
1575 stream
->avail_out
= bufsiz
;
1577 git_inflate_init(stream
);
1578 return git_inflate(stream
, 0);
1581 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1583 int bytes
= strlen(buffer
) + 1;
1584 unsigned char *buf
= xmallocz(size
);
1588 n
= stream
->total_out
- bytes
;
1591 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1593 if (bytes
<= size
) {
1595 * The above condition must be (bytes <= size), not
1596 * (bytes < size). In other words, even though we
1597 * expect no more output and set avail_out to zero,
1598 * the input zlib stream may have bytes that express
1599 * "this concludes the stream", and we *do* want to
1602 * Otherwise we would not be able to test that we
1603 * consumed all the input to reach the expected size;
1604 * we also want to check that zlib tells us that all
1605 * went well with status == Z_STREAM_END at the end.
1607 stream
->next_out
= buf
+ bytes
;
1608 stream
->avail_out
= size
- bytes
;
1609 while (status
== Z_OK
)
1610 status
= git_inflate(stream
, Z_FINISH
);
1612 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1613 git_inflate_end(stream
);
1618 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1619 else if (stream
->avail_in
)
1620 error("garbage at end of loose object '%s'",
1627 * We used to just use "sscanf()", but that's actually way
1628 * too permissive for what we want to check. So do an anal
1629 * object header parse by hand.
1631 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1638 * The type can be at most ten bytes (including the
1639 * terminating '\0' that we add), and is followed by
1648 if (i
>= sizeof(type
))
1654 * The length must follow immediately, and be in canonical
1655 * decimal format (ie "010" is not valid).
1657 size
= *hdr
++ - '0';
1662 unsigned long c
= *hdr
- '0';
1666 size
= size
* 10 + c
;
1672 * The length must be followed by a zero byte
1674 return *hdr
? -1 : type_from_string(type
);
1677 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1683 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1684 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1687 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1690 unsigned long get_size_from_delta(struct packed_git
*p
,
1691 struct pack_window
**w_curs
,
1694 const unsigned char *data
;
1695 unsigned char delta_head
[20], *in
;
1699 memset(&stream
, 0, sizeof(stream
));
1700 stream
.next_out
= delta_head
;
1701 stream
.avail_out
= sizeof(delta_head
);
1703 git_inflate_init(&stream
);
1705 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1706 stream
.next_in
= in
;
1707 st
= git_inflate(&stream
, Z_FINISH
);
1708 curpos
+= stream
.next_in
- in
;
1709 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1710 stream
.total_out
< sizeof(delta_head
));
1711 git_inflate_end(&stream
);
1712 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1713 error("delta data unpack-initial failed");
1717 /* Examine the initial part of the delta to figure out
1722 /* ignore base size */
1723 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1725 /* Read the result size */
1726 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1729 static off_t
get_delta_base(struct packed_git
*p
,
1730 struct pack_window
**w_curs
,
1732 enum object_type type
,
1733 off_t delta_obj_offset
)
1735 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1738 /* use_pack() assured us we have [base_info, base_info + 20)
1739 * as a range that we can look at without walking off the
1740 * end of the mapped window. Its actually the hash size
1741 * that is assured. An OFS_DELTA longer than the hash size
1742 * is stupid, as then a REF_DELTA would be smaller to store.
1744 if (type
== OBJ_OFS_DELTA
) {
1746 unsigned char c
= base_info
[used
++];
1747 base_offset
= c
& 127;
1750 if (!base_offset
|| MSB(base_offset
, 7))
1751 return 0; /* overflow */
1752 c
= base_info
[used
++];
1753 base_offset
= (base_offset
<< 7) + (c
& 127);
1755 base_offset
= delta_obj_offset
- base_offset
;
1756 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1757 return 0; /* out of bound */
1759 } else if (type
== OBJ_REF_DELTA
) {
1760 /* The base entry _must_ be in the same pack */
1761 base_offset
= find_pack_entry_one(base_info
, p
);
1764 die("I am totally screwed");
1769 * Like get_delta_base above, but we return the sha1 instead of the pack
1770 * offset. This means it is cheaper for REF deltas (we do not have to do
1771 * the final object lookup), but more expensive for OFS deltas (we
1772 * have to load the revidx to convert the offset back into a sha1).
1774 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1775 struct pack_window
**w_curs
,
1777 enum object_type type
,
1778 off_t delta_obj_offset
)
1780 if (type
== OBJ_REF_DELTA
) {
1781 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1783 } else if (type
== OBJ_OFS_DELTA
) {
1784 struct revindex_entry
*revidx
;
1785 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1786 type
, delta_obj_offset
);
1791 revidx
= find_pack_revindex(p
, base_offset
);
1795 return nth_packed_object_sha1(p
, revidx
->nr
);
1800 int unpack_object_header(struct packed_git
*p
,
1801 struct pack_window
**w_curs
,
1803 unsigned long *sizep
)
1805 unsigned char *base
;
1808 enum object_type type
;
1810 /* use_pack() assures us we have [base, base + 20) available
1811 * as a range that we can look at. (Its actually the hash
1812 * size that is assured.) With our object header encoding
1813 * the maximum deflated object size is 2^137, which is just
1814 * insane, so we know won't exceed what we have been given.
1816 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1817 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1826 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1829 struct revindex_entry
*revidx
;
1830 const unsigned char *sha1
;
1831 revidx
= find_pack_revindex(p
, obj_offset
);
1834 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1835 mark_bad_packed_object(p
, sha1
);
1836 type
= sha1_object_info(sha1
, NULL
);
1837 if (type
<= OBJ_NONE
)
1842 #define POI_STACK_PREALLOC 64
1844 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1846 enum object_type type
,
1847 struct pack_window
**w_curs
,
1850 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1851 off_t
*poi_stack
= small_poi_stack
;
1852 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1854 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1857 /* Push the object we're going to leave behind */
1858 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1859 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1860 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1861 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1863 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1865 poi_stack
[poi_stack_nr
++] = obj_offset
;
1866 /* If parsing the base offset fails, just unwind */
1867 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1870 curpos
= obj_offset
= base_offset
;
1871 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1872 if (type
<= OBJ_NONE
) {
1873 /* If getting the base itself fails, we first
1874 * retry the base, otherwise unwind */
1875 type
= retry_bad_packed_offset(p
, base_offset
);
1876 if (type
> OBJ_NONE
)
1890 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1891 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1896 if (poi_stack
!= small_poi_stack
)
1901 while (poi_stack_nr
) {
1902 obj_offset
= poi_stack
[--poi_stack_nr
];
1903 type
= retry_bad_packed_offset(p
, obj_offset
);
1904 if (type
> OBJ_NONE
)
1911 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1912 struct object_info
*oi
)
1914 struct pack_window
*w_curs
= NULL
;
1916 off_t curpos
= obj_offset
;
1917 enum object_type type
;
1920 * We always get the representation type, but only convert it to
1921 * a "real" type later if the caller is interested.
1923 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1926 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1927 off_t tmp_pos
= curpos
;
1928 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1934 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1935 if (*oi
->sizep
== 0) {
1944 if (oi
->disk_sizep
) {
1945 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1946 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1950 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1951 if (*oi
->typep
< 0) {
1957 if (oi
->delta_base_sha1
) {
1958 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1959 const unsigned char *base
;
1961 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1968 hashcpy(oi
->delta_base_sha1
, base
);
1970 hashclr(oi
->delta_base_sha1
);
1974 unuse_pack(&w_curs
);
1978 static void *unpack_compressed_entry(struct packed_git
*p
,
1979 struct pack_window
**w_curs
,
1985 unsigned char *buffer
, *in
;
1987 buffer
= xmallocz_gently(size
);
1990 memset(&stream
, 0, sizeof(stream
));
1991 stream
.next_out
= buffer
;
1992 stream
.avail_out
= size
+ 1;
1994 git_inflate_init(&stream
);
1996 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1997 stream
.next_in
= in
;
1998 st
= git_inflate(&stream
, Z_FINISH
);
1999 if (!stream
.avail_out
)
2000 break; /* the payload is larger than it should be */
2001 curpos
+= stream
.next_in
- in
;
2002 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
2003 git_inflate_end(&stream
);
2004 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
2012 #define MAX_DELTA_CACHE (256)
2014 static size_t delta_base_cached
;
2016 static struct delta_base_cache_lru_list
{
2017 struct delta_base_cache_lru_list
*prev
;
2018 struct delta_base_cache_lru_list
*next
;
2019 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
2021 static struct delta_base_cache_entry
{
2022 struct delta_base_cache_lru_list lru
;
2024 struct packed_git
*p
;
2027 enum object_type type
;
2028 } delta_base_cache
[MAX_DELTA_CACHE
];
2030 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
2034 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
2035 hash
+= (hash
>> 8) + (hash
>> 16);
2036 return hash
% MAX_DELTA_CACHE
;
2039 static struct delta_base_cache_entry
*
2040 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
2042 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2043 return delta_base_cache
+ hash
;
2046 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
2047 struct packed_git
*p
, off_t base_offset
)
2049 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
2052 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2054 struct delta_base_cache_entry
*ent
;
2055 ent
= get_delta_base_cache_entry(p
, base_offset
);
2056 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
2059 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2062 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2063 ent
->lru
.prev
->next
= ent
->lru
.next
;
2064 delta_base_cached
-= ent
->size
;
2067 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2068 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2070 struct delta_base_cache_entry
*ent
;
2073 ent
= get_delta_base_cache_entry(p
, base_offset
);
2075 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2076 return unpack_entry(p
, base_offset
, type
, base_size
);
2081 clear_delta_base_cache_entry(ent
);
2083 ret
= xmemdupz(ent
->data
, ent
->size
);
2085 *base_size
= ent
->size
;
2089 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2094 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2095 ent
->lru
.prev
->next
= ent
->lru
.next
;
2096 delta_base_cached
-= ent
->size
;
2100 void clear_delta_base_cache(void)
2103 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2104 release_delta_base_cache(&delta_base_cache
[p
]);
2107 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2108 void *base
, unsigned long base_size
, enum object_type type
)
2110 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2111 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2112 struct delta_base_cache_lru_list
*lru
;
2114 release_delta_base_cache(ent
);
2115 delta_base_cached
+= base_size
;
2117 for (lru
= delta_base_cache_lru
.next
;
2118 delta_base_cached
> delta_base_cache_limit
2119 && lru
!= &delta_base_cache_lru
;
2121 struct delta_base_cache_entry
*f
= (void *)lru
;
2122 if (f
->type
== OBJ_BLOB
)
2123 release_delta_base_cache(f
);
2125 for (lru
= delta_base_cache_lru
.next
;
2126 delta_base_cached
> delta_base_cache_limit
2127 && lru
!= &delta_base_cache_lru
;
2129 struct delta_base_cache_entry
*f
= (void *)lru
;
2130 release_delta_base_cache(f
);
2134 ent
->base_offset
= base_offset
;
2137 ent
->size
= base_size
;
2138 ent
->lru
.next
= &delta_base_cache_lru
;
2139 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2140 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2141 delta_base_cache_lru
.prev
= &ent
->lru
;
2144 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2145 unsigned long *size
);
2147 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2149 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2150 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2151 p
->pack_name
, (uintmax_t)obj_offset
);
2154 int do_check_packed_object_crc
;
2156 #define UNPACK_ENTRY_STACK_PREALLOC 64
2157 struct unpack_entry_stack_ent
{
2163 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2164 enum object_type
*final_type
, unsigned long *final_size
)
2166 struct pack_window
*w_curs
= NULL
;
2167 off_t curpos
= obj_offset
;
2170 enum object_type type
;
2171 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2172 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2173 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2174 int base_from_cache
= 0;
2176 write_pack_access_log(p
, obj_offset
);
2178 /* PHASE 1: drill down to the innermost base object */
2182 struct delta_base_cache_entry
*ent
;
2184 ent
= get_delta_base_cache_entry(p
, curpos
);
2185 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2189 clear_delta_base_cache_entry(ent
);
2190 base_from_cache
= 1;
2194 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2195 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2196 unsigned long len
= revidx
[1].offset
- obj_offset
;
2197 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2198 const unsigned char *sha1
=
2199 nth_packed_object_sha1(p
, revidx
->nr
);
2200 error("bad packed object CRC for %s",
2202 mark_bad_packed_object(p
, sha1
);
2203 unuse_pack(&w_curs
);
2208 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2209 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2212 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2214 error("failed to validate delta base reference "
2215 "at offset %"PRIuMAX
" from %s",
2216 (uintmax_t)curpos
, p
->pack_name
);
2217 /* bail to phase 2, in hopes of recovery */
2222 /* push object, proceed to base */
2223 if (delta_stack_nr
>= delta_stack_alloc
2224 && delta_stack
== small_delta_stack
) {
2225 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2226 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2227 memcpy(delta_stack
, small_delta_stack
,
2228 sizeof(*delta_stack
)*delta_stack_nr
);
2230 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2232 i
= delta_stack_nr
++;
2233 delta_stack
[i
].obj_offset
= obj_offset
;
2234 delta_stack
[i
].curpos
= curpos
;
2235 delta_stack
[i
].size
= size
;
2237 curpos
= obj_offset
= base_offset
;
2240 /* PHASE 2: handle the base */
2245 die("BUG in unpack_entry: left loop at a valid delta");
2251 if (!base_from_cache
)
2252 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2256 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2257 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2260 /* PHASE 3: apply deltas in order */
2263 * 'data' holds the base data, or NULL if there was corruption
2265 while (delta_stack_nr
) {
2268 unsigned long delta_size
, base_size
= size
;
2274 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2278 * We're probably in deep shit, but let's try to fetch
2279 * the required base anyway from another pack or loose.
2280 * This is costly but should happen only in the presence
2281 * of a corrupted pack, and is better than failing outright.
2283 struct revindex_entry
*revidx
;
2284 const unsigned char *base_sha1
;
2285 revidx
= find_pack_revindex(p
, obj_offset
);
2287 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2288 error("failed to read delta base object %s"
2289 " at offset %"PRIuMAX
" from %s",
2290 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2292 mark_bad_packed_object(p
, base_sha1
);
2293 base
= read_object(base_sha1
, &type
, &base_size
);
2297 i
= --delta_stack_nr
;
2298 obj_offset
= delta_stack
[i
].obj_offset
;
2299 curpos
= delta_stack
[i
].curpos
;
2300 delta_size
= delta_stack
[i
].size
;
2305 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2308 error("failed to unpack compressed delta "
2309 "at offset %"PRIuMAX
" from %s",
2310 (uintmax_t)curpos
, p
->pack_name
);
2315 data
= patch_delta(base
, base_size
,
2316 delta_data
, delta_size
,
2320 * We could not apply the delta; warn the user, but keep going.
2321 * Our failure will be noticed either in the next iteration of
2322 * the loop, or if this is the final delta, in the caller when
2323 * we return NULL. Those code paths will take care of making
2324 * a more explicit warning and retrying with another copy of
2328 error("failed to apply delta");
2336 unuse_pack(&w_curs
);
2338 if (delta_stack
!= small_delta_stack
)
2344 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2347 const unsigned char *index
= p
->index_data
;
2349 if (open_pack_index(p
))
2351 index
= p
->index_data
;
2353 if (n
>= p
->num_objects
)
2356 if (p
->index_version
== 1) {
2357 return index
+ 24 * n
+ 4;
2360 return index
+ 20 * n
;
2364 void check_pack_index_ptr(const struct packed_git
*p
, const void *vptr
)
2366 const unsigned char *ptr
= vptr
;
2367 const unsigned char *start
= p
->index_data
;
2368 const unsigned char *end
= start
+ p
->index_size
;
2370 die("offset before start of pack index for %s (corrupt index?)",
2372 /* No need to check for underflow; .idx files must be at least 8 bytes */
2374 die("offset beyond end of pack index for %s (truncated index?)",
2378 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2380 const unsigned char *index
= p
->index_data
;
2382 if (p
->index_version
== 1) {
2383 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2386 index
+= 8 + p
->num_objects
* (20 + 4);
2387 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2388 if (!(off
& 0x80000000))
2390 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2391 check_pack_index_ptr(p
, index
);
2392 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2393 ntohl(*((uint32_t *)(index
+ 4)));
2397 off_t
find_pack_entry_one(const unsigned char *sha1
,
2398 struct packed_git
*p
)
2400 const uint32_t *level1_ofs
= p
->index_data
;
2401 const unsigned char *index
= p
->index_data
;
2402 unsigned hi
, lo
, stride
;
2403 static int use_lookup
= -1;
2404 static int debug_lookup
= -1;
2406 if (debug_lookup
< 0)
2407 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2410 if (open_pack_index(p
))
2412 level1_ofs
= p
->index_data
;
2413 index
= p
->index_data
;
2415 if (p
->index_version
> 1) {
2420 hi
= ntohl(level1_ofs
[*sha1
]);
2421 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2422 if (p
->index_version
> 1) {
2430 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2431 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2434 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2436 int pos
= sha1_entry_pos(index
, stride
, 0,
2437 lo
, hi
, p
->num_objects
, sha1
);
2440 return nth_packed_object_offset(p
, pos
);
2444 unsigned mi
= (lo
+ hi
) / 2;
2445 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2448 printf("lo %u hi %u rg %u mi %u\n",
2449 lo
, hi
, hi
- lo
, mi
);
2451 return nth_packed_object_offset(p
, mi
);
2460 int is_pack_valid(struct packed_git
*p
)
2462 /* An already open pack is known to be valid. */
2463 if (p
->pack_fd
!= -1)
2466 /* If the pack has one window completely covering the
2467 * file size, the pack is known to be valid even if
2468 * the descriptor is not currently open.
2471 struct pack_window
*w
= p
->windows
;
2473 if (!w
->offset
&& w
->len
== p
->pack_size
)
2477 /* Force the pack to open to prove its valid. */
2478 return !open_packed_git(p
);
2481 static int fill_pack_entry(const unsigned char *sha1
,
2482 struct pack_entry
*e
,
2483 struct packed_git
*p
)
2487 if (p
->num_bad_objects
) {
2489 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2490 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2494 offset
= find_pack_entry_one(sha1
, p
);
2499 * We are about to tell the caller where they can locate the
2500 * requested object. We better make sure the packfile is
2501 * still here and can be accessed before supplying that
2502 * answer, as it may have been deleted since the index was
2505 if (!is_pack_valid(p
))
2509 hashcpy(e
->sha1
, sha1
);
2514 * Iff a pack file contains the object named by sha1, return true and
2515 * store its location to e.
2517 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2519 struct packed_git
*p
;
2521 prepare_packed_git();
2525 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2528 for (p
= packed_git
; p
; p
= p
->next
) {
2529 if (p
== last_found_pack
)
2530 continue; /* we already checked this one */
2532 if (fill_pack_entry(sha1
, e
, p
)) {
2533 last_found_pack
= p
;
2540 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2541 struct packed_git
*packs
)
2543 struct packed_git
*p
;
2545 for (p
= packs
; p
; p
= p
->next
) {
2546 if (find_pack_entry_one(sha1
, p
))
2553 static int sha1_loose_object_info(const unsigned char *sha1
,
2554 struct object_info
*oi
)
2557 unsigned long mapsize
, size
;
2562 if (oi
->delta_base_sha1
)
2563 hashclr(oi
->delta_base_sha1
);
2566 * If we don't care about type or size, then we don't
2567 * need to look inside the object at all. Note that we
2568 * do not optimize out the stat call, even if the
2569 * caller doesn't care about the disk-size, since our
2570 * return value implicitly indicates whether the
2571 * object even exists.
2573 if (!oi
->typep
&& !oi
->sizep
) {
2575 if (stat_sha1_file(sha1
, &st
) < 0)
2578 *oi
->disk_sizep
= st
.st_size
;
2582 map
= map_sha1_file(sha1
, &mapsize
);
2586 *oi
->disk_sizep
= mapsize
;
2587 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2588 status
= error("unable to unpack %s header",
2590 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2591 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2594 git_inflate_end(&stream
);
2595 munmap(map
, mapsize
);
2597 *oi
->typep
= status
;
2601 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2603 struct cached_object
*co
;
2604 struct pack_entry e
;
2606 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2608 co
= find_cached_object(real
);
2611 *(oi
->typep
) = co
->type
;
2613 *(oi
->sizep
) = co
->size
;
2615 *(oi
->disk_sizep
) = 0;
2616 if (oi
->delta_base_sha1
)
2617 hashclr(oi
->delta_base_sha1
);
2618 oi
->whence
= OI_CACHED
;
2622 if (!find_pack_entry(real
, &e
)) {
2623 /* Most likely it's a loose object. */
2624 if (!sha1_loose_object_info(real
, oi
)) {
2625 oi
->whence
= OI_LOOSE
;
2629 /* Not a loose object; someone else may have just packed it. */
2630 reprepare_packed_git();
2631 if (!find_pack_entry(real
, &e
))
2635 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2637 mark_bad_packed_object(e
.p
, real
);
2638 return sha1_object_info_extended(real
, oi
, 0);
2639 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2640 oi
->whence
= OI_DBCACHED
;
2642 oi
->whence
= OI_PACKED
;
2643 oi
->u
.packed
.offset
= e
.offset
;
2644 oi
->u
.packed
.pack
= e
.p
;
2645 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2646 rtype
== OBJ_OFS_DELTA
);
2652 /* returns enum object_type or negative */
2653 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2655 enum object_type type
;
2656 struct object_info oi
= {NULL
};
2660 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2665 static void *read_packed_sha1(const unsigned char *sha1
,
2666 enum object_type
*type
, unsigned long *size
)
2668 struct pack_entry e
;
2671 if (!find_pack_entry(sha1
, &e
))
2673 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2676 * We're probably in deep shit, but let's try to fetch
2677 * the required object anyway from another pack or loose.
2678 * This should happen only in the presence of a corrupted
2679 * pack, and is better than failing outright.
2681 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2682 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2683 mark_bad_packed_object(e
.p
, sha1
);
2684 data
= read_object(sha1
, type
, size
);
2689 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2690 unsigned char *sha1
)
2692 struct cached_object
*co
;
2694 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2695 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2697 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2698 co
= &cached_objects
[cached_object_nr
++];
2701 co
->buf
= xmalloc(len
);
2702 memcpy(co
->buf
, buf
, len
);
2703 hashcpy(co
->sha1
, sha1
);
2707 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2708 unsigned long *size
)
2710 unsigned long mapsize
;
2712 struct cached_object
*co
;
2714 co
= find_cached_object(sha1
);
2718 return xmemdupz(co
->buf
, co
->size
);
2721 buf
= read_packed_sha1(sha1
, type
, size
);
2724 map
= map_sha1_file(sha1
, &mapsize
);
2726 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2727 munmap(map
, mapsize
);
2730 reprepare_packed_git();
2731 return read_packed_sha1(sha1
, type
, size
);
2735 * This function dies on corrupt objects; the callers who want to
2736 * deal with them should arrange to call read_object() and give error
2737 * messages themselves.
2739 void *read_sha1_file_extended(const unsigned char *sha1
,
2740 enum object_type
*type
,
2741 unsigned long *size
,
2745 const struct packed_git
*p
;
2746 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2749 data
= read_object(repl
, type
, size
);
2753 if (errno
&& errno
!= ENOENT
)
2754 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2756 /* die if we replaced an object with one that does not exist */
2758 die("replacement %s not found for %s",
2759 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2761 if (has_loose_object(repl
)) {
2762 const char *path
= sha1_file_name(sha1
);
2764 die("loose object %s (stored in %s) is corrupt",
2765 sha1_to_hex(repl
), path
);
2768 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2769 die("packed object %s (stored in %s) is corrupt",
2770 sha1_to_hex(repl
), p
->pack_name
);
2775 void *read_object_with_reference(const unsigned char *sha1
,
2776 const char *required_type_name
,
2777 unsigned long *size
,
2778 unsigned char *actual_sha1_return
)
2780 enum object_type type
, required_type
;
2782 unsigned long isize
;
2783 unsigned char actual_sha1
[20];
2785 required_type
= type_from_string(required_type_name
);
2786 hashcpy(actual_sha1
, sha1
);
2788 int ref_length
= -1;
2789 const char *ref_type
= NULL
;
2791 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2794 if (type
== required_type
) {
2796 if (actual_sha1_return
)
2797 hashcpy(actual_sha1_return
, actual_sha1
);
2800 /* Handle references */
2801 else if (type
== OBJ_COMMIT
)
2803 else if (type
== OBJ_TAG
)
2804 ref_type
= "object ";
2809 ref_length
= strlen(ref_type
);
2811 if (ref_length
+ 40 > isize
||
2812 memcmp(buffer
, ref_type
, ref_length
) ||
2813 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2818 /* Now we have the ID of the referred-to object in
2819 * actual_sha1. Check again. */
2823 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2824 const char *type
, unsigned char *sha1
,
2825 char *hdr
, int *hdrlen
)
2829 /* Generate the header */
2830 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2834 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2835 git_SHA1_Update(&c
, buf
, len
);
2836 git_SHA1_Final(sha1
, &c
);
2840 * Move the just written object into its final resting place.
2841 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2842 * "moving" is only a part of what it does, when no patch between
2843 * master to pu changes the call sites of this function.
2845 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2849 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2851 else if (link(tmpfile
, filename
))
2855 * Coda hack - coda doesn't like cross-directory links,
2856 * so we fall back to a rename, which will mean that it
2857 * won't be able to check collisions, but that's not a
2860 * The same holds for FAT formatted media.
2862 * When this succeeds, we just return. We have nothing
2865 if (ret
&& ret
!= EEXIST
) {
2867 if (!rename(tmpfile
, filename
))
2871 unlink_or_warn(tmpfile
);
2873 if (ret
!= EEXIST
) {
2874 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2876 /* FIXME!!! Collision check here ? */
2880 if (adjust_shared_perm(filename
))
2881 return error("unable to set permission to '%s'", filename
);
2885 static int write_buffer(int fd
, const void *buf
, size_t len
)
2887 if (write_in_full(fd
, buf
, len
) < 0)
2888 return error("file write error (%s)", strerror(errno
));
2892 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2893 unsigned char *sha1
)
2897 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2901 /* Finalize a file on disk, and close it. */
2902 static void close_sha1_file(int fd
)
2904 if (fsync_object_files
)
2905 fsync_or_die(fd
, "sha1 file");
2907 die_errno("error when closing sha1 file");
2910 /* Size of directory component, including the ending '/' */
2911 static inline int directory_size(const char *filename
)
2913 const char *s
= strrchr(filename
, '/');
2916 return s
- filename
+ 1;
2920 * This creates a temporary file in the same directory as the final
2923 * We want to avoid cross-directory filename renames, because those
2924 * can have problems on various filesystems (FAT, NFS, Coda).
2926 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2928 int fd
, dirlen
= directory_size(filename
);
2930 if (dirlen
+ 20 > bufsiz
) {
2931 errno
= ENAMETOOLONG
;
2934 memcpy(buffer
, filename
, dirlen
);
2935 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2936 fd
= git_mkstemp_mode(buffer
, 0444);
2937 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2938 /* Make sure the directory exists */
2939 memcpy(buffer
, filename
, dirlen
);
2940 buffer
[dirlen
-1] = 0;
2941 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
2943 if (adjust_shared_perm(buffer
))
2947 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2948 fd
= git_mkstemp_mode(buffer
, 0444);
2953 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2954 const void *buf
, unsigned long len
, time_t mtime
)
2957 unsigned char compressed
[4096];
2960 unsigned char parano_sha1
[20];
2961 static char tmp_file
[PATH_MAX
];
2962 const char *filename
= sha1_file_name(sha1
);
2964 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2966 if (errno
== EACCES
)
2967 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2969 return error("unable to create temporary file: %s", strerror(errno
));
2973 git_deflate_init(&stream
, zlib_compression_level
);
2974 stream
.next_out
= compressed
;
2975 stream
.avail_out
= sizeof(compressed
);
2978 /* First header.. */
2979 stream
.next_in
= (unsigned char *)hdr
;
2980 stream
.avail_in
= hdrlen
;
2981 while (git_deflate(&stream
, 0) == Z_OK
)
2983 git_SHA1_Update(&c
, hdr
, hdrlen
);
2985 /* Then the data itself.. */
2986 stream
.next_in
= (void *)buf
;
2987 stream
.avail_in
= len
;
2989 unsigned char *in0
= stream
.next_in
;
2990 ret
= git_deflate(&stream
, Z_FINISH
);
2991 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2992 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2993 die("unable to write sha1 file");
2994 stream
.next_out
= compressed
;
2995 stream
.avail_out
= sizeof(compressed
);
2996 } while (ret
== Z_OK
);
2998 if (ret
!= Z_STREAM_END
)
2999 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
3000 ret
= git_deflate_end_gently(&stream
);
3002 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
3003 git_SHA1_Final(parano_sha1
, &c
);
3004 if (hashcmp(sha1
, parano_sha1
) != 0)
3005 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
3007 close_sha1_file(fd
);
3012 utb
.modtime
= mtime
;
3013 if (utime(tmp_file
, &utb
) < 0)
3014 warning("failed utime() on %s: %s",
3015 tmp_file
, strerror(errno
));
3018 return move_temp_to_file(tmp_file
, filename
);
3021 static int freshen_loose_object(const unsigned char *sha1
)
3023 return check_and_freshen(sha1
, 1);
3026 static int freshen_packed_object(const unsigned char *sha1
)
3028 struct pack_entry e
;
3029 if (!find_pack_entry(sha1
, &e
))
3033 if (!freshen_file(e
.p
->pack_name
))
3039 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *sha1
)
3044 /* Normally if we have it in the pack then we do not bother writing
3045 * it out into .git/objects/??/?{38} file.
3047 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
3048 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3050 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
3053 int hash_sha1_file_literally(const void *buf
, unsigned long len
, const char *type
,
3054 unsigned char *sha1
, unsigned flags
)
3057 int hdrlen
, status
= 0;
3059 /* type string, SP, %lu of the length plus NUL must fit this */
3060 header
= xmalloc(strlen(type
) + 32);
3061 write_sha1_file_prepare(buf
, len
, type
, sha1
, header
, &hdrlen
);
3063 if (!(flags
& HASH_WRITE_OBJECT
))
3065 if (freshen_packed_object(sha1
) || freshen_loose_object(sha1
))
3067 status
= write_loose_object(sha1
, header
, hdrlen
, buf
, len
, 0);
3074 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
3078 enum object_type type
;
3083 if (has_loose_object(sha1
))
3085 buf
= read_packed_sha1(sha1
, &type
, &len
);
3087 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
3088 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
3089 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3095 int has_pack_index(const unsigned char *sha1
)
3098 if (stat(sha1_pack_index_name(sha1
), &st
))
3103 int has_sha1_pack(const unsigned char *sha1
)
3105 struct pack_entry e
;
3106 return find_pack_entry(sha1
, &e
);
3109 int has_sha1_file_with_flags(const unsigned char *sha1
, int flags
)
3111 struct pack_entry e
;
3113 if (find_pack_entry(sha1
, &e
))
3115 if (has_loose_object(sha1
))
3117 if (flags
& HAS_SHA1_QUICK
)
3119 reprepare_packed_git();
3120 return find_pack_entry(sha1
, &e
);
3123 static void check_tree(const void *buf
, size_t size
)
3125 struct tree_desc desc
;
3126 struct name_entry entry
;
3128 init_tree_desc(&desc
, buf
, size
);
3129 while (tree_entry(&desc
, &entry
))
3131 * tree_entry() will die() on malformed entries */
3135 static void check_commit(const void *buf
, size_t size
)
3138 memset(&c
, 0, sizeof(c
));
3139 if (parse_commit_buffer(&c
, buf
, size
))
3140 die("corrupt commit");
3143 static void check_tag(const void *buf
, size_t size
)
3146 memset(&t
, 0, sizeof(t
));
3147 if (parse_tag_buffer(&t
, buf
, size
))
3151 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3152 enum object_type type
,
3153 const char *path
, unsigned flags
)
3155 int ret
, re_allocated
= 0;
3156 int write_object
= flags
& HASH_WRITE_OBJECT
;
3162 * Convert blobs to git internal format
3164 if ((type
== OBJ_BLOB
) && path
) {
3165 struct strbuf nbuf
= STRBUF_INIT
;
3166 if (convert_to_git(path
, buf
, size
, &nbuf
,
3167 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3168 buf
= strbuf_detach(&nbuf
, &size
);
3172 if (flags
& HASH_FORMAT_CHECK
) {
3173 if (type
== OBJ_TREE
)
3174 check_tree(buf
, size
);
3175 if (type
== OBJ_COMMIT
)
3176 check_commit(buf
, size
);
3177 if (type
== OBJ_TAG
)
3178 check_tag(buf
, size
);
3182 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3184 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3190 static int index_stream_convert_blob(unsigned char *sha1
, int fd
,
3191 const char *path
, unsigned flags
)
3194 const int write_object
= flags
& HASH_WRITE_OBJECT
;
3195 struct strbuf sbuf
= STRBUF_INIT
;
3198 assert(would_convert_to_git_filter_fd(path
));
3200 convert_to_git_filter_fd(path
, fd
, &sbuf
,
3201 write_object
? safe_crlf
: SAFE_CRLF_FALSE
);
3204 ret
= write_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3207 ret
= hash_sha1_file(sbuf
.buf
, sbuf
.len
, typename(OBJ_BLOB
),
3209 strbuf_release(&sbuf
);
3213 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3214 const char *path
, unsigned flags
)
3216 struct strbuf sbuf
= STRBUF_INIT
;
3219 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3220 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3223 strbuf_release(&sbuf
);
3227 #define SMALL_FILE_SIZE (32*1024)
3229 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3230 enum object_type type
, const char *path
,
3236 ret
= index_mem(sha1
, "", size
, type
, path
, flags
);
3237 } else if (size
<= SMALL_FILE_SIZE
) {
3238 char *buf
= xmalloc(size
);
3239 if (size
== read_in_full(fd
, buf
, size
))
3240 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3242 ret
= error("short read %s", strerror(errno
));
3245 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3246 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3253 * This creates one packfile per large blob unless bulk-checkin
3254 * machinery is "plugged".
3256 * This also bypasses the usual "convert-to-git" dance, and that is on
3257 * purpose. We could write a streaming version of the converting
3258 * functions and insert that before feeding the data to fast-import
3259 * (or equivalent in-core API described above). However, that is
3260 * somewhat complicated, as we do not know the size of the filter
3261 * result, which we need to know beforehand when writing a git object.
3262 * Since the primary motivation for trying to stream from the working
3263 * tree file and to avoid mmaping it in core is to deal with large
3264 * binary blobs, they generally do not want to get any conversion, and
3265 * callers should avoid this code path when filters are requested.
3267 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3268 enum object_type type
, const char *path
,
3271 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3274 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3275 enum object_type type
, const char *path
, unsigned flags
)
3280 * Call xsize_t() only when needed to avoid potentially unnecessary
3281 * die() for large files.
3283 if (type
== OBJ_BLOB
&& path
&& would_convert_to_git_filter_fd(path
))
3284 ret
= index_stream_convert_blob(sha1
, fd
, path
, flags
);
3285 else if (!S_ISREG(st
->st_mode
))
3286 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3287 else if (st
->st_size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3288 (path
&& would_convert_to_git(path
)))
3289 ret
= index_core(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3292 ret
= index_stream(sha1
, fd
, xsize_t(st
->st_size
), type
, path
,
3298 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3301 struct strbuf sb
= STRBUF_INIT
;
3303 switch (st
->st_mode
& S_IFMT
) {
3305 fd
= open(path
, O_RDONLY
);
3307 return error("open(\"%s\"): %s", path
,
3309 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3310 return error("%s: failed to insert into database",
3314 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3315 char *errstr
= strerror(errno
);
3316 return error("readlink(\"%s\"): %s", path
,
3319 if (!(flags
& HASH_WRITE_OBJECT
))
3320 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3321 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3322 return error("%s: failed to insert into database",
3324 strbuf_release(&sb
);
3327 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3329 return error("%s: unsupported file type", path
);
3334 int read_pack_header(int fd
, struct pack_header
*header
)
3336 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3337 /* "eof before pack header was fully read" */
3338 return PH_ERROR_EOF
;
3340 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3341 /* "protocol error (pack signature mismatch detected)" */
3342 return PH_ERROR_PACK_SIGNATURE
;
3343 if (!pack_version_ok(header
->hdr_version
))
3344 /* "protocol error (pack version unsupported)" */
3345 return PH_ERROR_PROTOCOL
;
3349 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3351 enum object_type type
= sha1_object_info(sha1
, NULL
);
3353 die("%s is not a valid object", sha1_to_hex(sha1
));
3355 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),
3359 static int for_each_file_in_obj_subdir(int subdir_nr
,
3360 struct strbuf
*path
,
3361 each_loose_object_fn obj_cb
,
3362 each_loose_cruft_fn cruft_cb
,
3363 each_loose_subdir_fn subdir_cb
,
3366 size_t baselen
= path
->len
;
3367 DIR *dir
= opendir(path
->buf
);
3372 if (errno
== ENOENT
)
3374 return error("unable to open %s: %s", path
->buf
, strerror(errno
));
3377 while ((de
= readdir(dir
))) {
3378 if (is_dot_or_dotdot(de
->d_name
))
3381 strbuf_setlen(path
, baselen
);
3382 strbuf_addf(path
, "/%s", de
->d_name
);
3384 if (strlen(de
->d_name
) == 38) {
3386 unsigned char sha1
[20];
3388 snprintf(hex
, sizeof(hex
), "%02x%s",
3389 subdir_nr
, de
->d_name
);
3390 if (!get_sha1_hex(hex
, sha1
)) {
3392 r
= obj_cb(sha1
, path
->buf
, data
);
3401 r
= cruft_cb(de
->d_name
, path
->buf
, data
);
3406 strbuf_setlen(path
, baselen
);
3408 if (!r
&& subdir_cb
)
3409 r
= subdir_cb(subdir_nr
, path
->buf
, data
);
3415 int for_each_loose_file_in_objdir_buf(struct strbuf
*path
,
3416 each_loose_object_fn obj_cb
,
3417 each_loose_cruft_fn cruft_cb
,
3418 each_loose_subdir_fn subdir_cb
,
3421 size_t baselen
= path
->len
;
3425 for (i
= 0; i
< 256; i
++) {
3426 strbuf_addf(path
, "/%02x", i
);
3427 r
= for_each_file_in_obj_subdir(i
, path
, obj_cb
, cruft_cb
,
3429 strbuf_setlen(path
, baselen
);
3437 int for_each_loose_file_in_objdir(const char *path
,
3438 each_loose_object_fn obj_cb
,
3439 each_loose_cruft_fn cruft_cb
,
3440 each_loose_subdir_fn subdir_cb
,
3443 struct strbuf buf
= STRBUF_INIT
;
3446 strbuf_addstr(&buf
, path
);
3447 r
= for_each_loose_file_in_objdir_buf(&buf
, obj_cb
, cruft_cb
,
3449 strbuf_release(&buf
);
3454 struct loose_alt_odb_data
{
3455 each_loose_object_fn
*cb
;
3459 static int loose_from_alt_odb(struct alternate_object_database
*alt
,
3462 struct loose_alt_odb_data
*data
= vdata
;
3463 struct strbuf buf
= STRBUF_INIT
;
3466 /* copy base not including trailing '/' */
3467 strbuf_add(&buf
, alt
->base
, alt
->name
- alt
->base
- 1);
3468 r
= for_each_loose_file_in_objdir_buf(&buf
,
3469 data
->cb
, NULL
, NULL
,
3471 strbuf_release(&buf
);
3475 int for_each_loose_object(each_loose_object_fn cb
, void *data
, unsigned flags
)
3477 struct loose_alt_odb_data alt
;
3480 r
= for_each_loose_file_in_objdir(get_object_directory(),
3481 cb
, NULL
, NULL
, data
);
3485 if (flags
& FOR_EACH_OBJECT_LOCAL_ONLY
)
3490 return foreach_alt_odb(loose_from_alt_odb
, &alt
);
3493 static int for_each_object_in_pack(struct packed_git
*p
, each_packed_object_fn cb
, void *data
)
3498 for (i
= 0; i
< p
->num_objects
; i
++) {
3499 const unsigned char *sha1
= nth_packed_object_sha1(p
, i
);
3502 return error("unable to get sha1 of object %u in %s",
3505 r
= cb(sha1
, p
, i
, data
);
3512 int for_each_packed_object(each_packed_object_fn cb
, void *data
, unsigned flags
)
3514 struct packed_git
*p
;
3517 prepare_packed_git();
3518 for (p
= packed_git
; p
; p
= p
->next
) {
3519 if ((flags
& FOR_EACH_OBJECT_LOCAL_ONLY
) && !p
->pack_local
)
3521 r
= for_each_object_in_pack(p
, cb
, data
);