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 enum scld_error
safe_create_leading_directories(char *path
)
110 char *next_component
= path
+ offset_1st_component(path
);
111 enum scld_error ret
= SCLD_OK
;
113 while (ret
== SCLD_OK
&& next_component
) {
115 char *slash
= next_component
, slash_character
;
117 while (*slash
&& !is_dir_sep(*slash
))
123 next_component
= slash
+ 1;
124 while (is_dir_sep(*next_component
))
126 if (!*next_component
)
129 slash_character
= *slash
;
131 if (!stat(path
, &st
)) {
133 if (!S_ISDIR(st
.st_mode
))
135 } else if (mkdir(path
, 0777)) {
136 if (errno
== EEXIST
&&
137 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
138 ; /* somebody created it since we checked */
139 else if (errno
== ENOENT
)
141 * Either mkdir() failed because
142 * somebody just pruned the containing
143 * directory, or stat() failed because
144 * the file that was in our way was
145 * just removed. Either way, inform
146 * the caller that it might be worth
152 } else if (adjust_shared_perm(path
)) {
155 *slash
= slash_character
;
160 enum scld_error
safe_create_leading_directories_const(const char *path
)
162 /* path points to cache entries, so xstrdup before messing with it */
163 char *buf
= xstrdup(path
);
164 enum scld_error result
= safe_create_leading_directories(buf
);
169 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
172 for (i
= 0; i
< 20; i
++) {
173 static char hex
[] = "0123456789abcdef";
174 unsigned int val
= sha1
[i
];
175 char *pos
= pathbuf
+ i
*2 + (i
> 0);
176 *pos
++ = hex
[val
>> 4];
177 *pos
= hex
[val
& 0xf];
182 * NOTE! This returns a statically allocated buffer, so you have to be
183 * careful about using it. Do an "xstrdup()" if you need to save the
186 * Also note that this returns the location for creating. Reading
187 * SHA1 file can happen from any alternate directory listed in the
188 * DB_ENVIRONMENT environment variable if it is not found in
189 * the primary object database.
191 char *sha1_file_name(const unsigned char *sha1
)
193 static char buf
[PATH_MAX
];
197 objdir
= get_object_directory();
198 len
= strlen(objdir
);
200 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
201 if (len
+ 43 > PATH_MAX
)
202 die("insanely long object directory %s", objdir
);
203 memcpy(buf
, objdir
, len
);
207 fill_sha1_path(buf
+ len
+ 1, sha1
);
211 static char *sha1_get_pack_name(const unsigned char *sha1
,
212 char **name
, char **base
, const char *which
)
214 static const char hex
[] = "0123456789abcdef";
219 const char *sha1_file_directory
= get_object_directory();
220 int len
= strlen(sha1_file_directory
);
221 *base
= xmalloc(len
+ 60);
222 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
223 sha1_file_directory
, which
);
224 *name
= *base
+ len
+ 11;
229 for (i
= 0; i
< 20; i
++) {
230 unsigned int val
= *sha1
++;
231 *buf
++ = hex
[val
>> 4];
232 *buf
++ = hex
[val
& 0xf];
238 char *sha1_pack_name(const unsigned char *sha1
)
240 static char *name
, *base
;
242 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
245 char *sha1_pack_index_name(const unsigned char *sha1
)
247 static char *name
, *base
;
249 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
252 struct alternate_object_database
*alt_odb_list
;
253 static struct alternate_object_database
**alt_odb_tail
;
255 static int git_open_noatime(const char *name
);
258 * Prepare alternate object database registry.
260 * The variable alt_odb_list points at the list of struct
261 * alternate_object_database. The elements on this list come from
262 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
263 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
264 * whose contents is similar to that environment variable but can be
265 * LF separated. Its base points at a statically allocated buffer that
266 * contains "/the/directory/corresponding/to/.git/objects/...", while
267 * its name points just after the slash at the end of ".git/objects/"
268 * in the example above, and has enough space to hold 40-byte hex
269 * SHA1, an extra slash for the first level indirection, and the
272 static int link_alt_odb_entry(const char *entry
, const char *relative_base
, int depth
)
274 const char *objdir
= get_object_directory();
275 struct alternate_object_database
*ent
;
276 struct alternate_object_database
*alt
;
278 struct strbuf pathbuf
= STRBUF_INIT
;
280 if (!is_absolute_path(entry
) && relative_base
) {
281 strbuf_addstr(&pathbuf
, real_path(relative_base
));
282 strbuf_addch(&pathbuf
, '/');
284 strbuf_addstr(&pathbuf
, entry
);
286 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
288 pfxlen
= strlen(pathbuf
.buf
);
291 * The trailing slash after the directory name is given by
292 * this function at the end. Remove duplicates.
294 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
297 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
298 ent
= xmalloc(sizeof(*ent
) + entlen
);
299 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
300 strbuf_release(&pathbuf
);
302 ent
->name
= ent
->base
+ pfxlen
+ 1;
303 ent
->base
[pfxlen
+ 3] = '/';
304 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
306 /* Detect cases where alternate disappeared */
307 if (!is_directory(ent
->base
)) {
308 error("object directory %s does not exist; "
309 "check .git/objects/info/alternates.",
315 /* Prevent the common mistake of listing the same
316 * thing twice, or object directory itself.
318 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
319 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
324 if (!strcmp(ent
->base
, objdir
)) {
329 /* add the alternate entry */
331 alt_odb_tail
= &(ent
->next
);
334 /* recursively add alternates */
335 read_info_alternates(ent
->base
, depth
+ 1);
337 ent
->base
[pfxlen
] = '/';
342 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
343 const char *relative_base
, int depth
)
345 struct string_list entries
= STRING_LIST_INIT_NODUP
;
350 error("%s: ignoring alternate object stores, nesting too deep.",
355 alt_copy
= xmemdupz(alt
, len
);
356 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
357 for (i
= 0; i
< entries
.nr
; i
++) {
358 const char *entry
= entries
.items
[i
].string
;
359 if (entry
[0] == '\0' || entry
[0] == '#')
361 if (!is_absolute_path(entry
) && depth
) {
362 error("%s: ignoring relative alternate object store %s",
363 relative_base
, entry
);
365 link_alt_odb_entry(entry
, relative_base
, depth
);
368 string_list_clear(&entries
, 0);
372 void read_info_alternates(const char * relative_base
, int depth
)
377 const char alt_file_name
[] = "info/alternates";
378 /* Given that relative_base is no longer than PATH_MAX,
379 ensure that "path" has enough space to append "/", the
380 file name, "info/alternates", and a trailing NUL. */
381 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
384 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
385 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 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
415 struct alternate_object_database
*ent
;
418 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
423 void prepare_alt_odb(void)
430 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
433 alt_odb_tail
= &alt_odb_list
;
434 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
436 read_info_alternates(get_object_directory(), 0);
439 static int has_loose_object_local(const unsigned char *sha1
)
441 char *name
= sha1_file_name(sha1
);
442 return !access(name
, F_OK
);
445 int has_loose_object_nonlocal(const unsigned char *sha1
)
447 struct alternate_object_database
*alt
;
449 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
450 fill_sha1_path(alt
->name
, sha1
);
451 if (!access(alt
->base
, F_OK
))
457 static int has_loose_object(const unsigned char *sha1
)
459 return has_loose_object_local(sha1
) ||
460 has_loose_object_nonlocal(sha1
);
463 static unsigned int pack_used_ctr
;
464 static unsigned int pack_mmap_calls
;
465 static unsigned int peak_pack_open_windows
;
466 static unsigned int pack_open_windows
;
467 static unsigned int pack_open_fds
;
468 static unsigned int pack_max_fds
;
469 static size_t peak_pack_mapped
;
470 static size_t pack_mapped
;
471 struct packed_git
*packed_git
;
473 void pack_report(void)
476 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
477 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
478 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
479 sz_fmt(getpagesize()),
480 sz_fmt(packed_git_window_size
),
481 sz_fmt(packed_git_limit
));
483 "pack_report: pack_used_ctr = %10u\n"
484 "pack_report: pack_mmap_calls = %10u\n"
485 "pack_report: pack_open_windows = %10u / %10u\n"
486 "pack_report: pack_mapped = "
487 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
490 pack_open_windows
, peak_pack_open_windows
,
491 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
494 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
497 struct pack_idx_header
*hdr
;
499 uint32_t version
, nr
, i
, *index
;
500 int fd
= git_open_noatime(path
);
505 if (fstat(fd
, &st
)) {
509 idx_size
= xsize_t(st
.st_size
);
510 if (idx_size
< 4 * 256 + 20 + 20) {
512 return error("index file %s is too small", path
);
514 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
518 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
519 version
= ntohl(hdr
->idx_version
);
520 if (version
< 2 || version
> 2) {
521 munmap(idx_map
, idx_size
);
522 return error("index file %s is version %"PRIu32
523 " and is not supported by this binary"
524 " (try upgrading GIT to a newer version)",
533 index
+= 2; /* skip index header */
534 for (i
= 0; i
< 256; i
++) {
535 uint32_t n
= ntohl(index
[i
]);
537 munmap(idx_map
, idx_size
);
538 return error("non-monotonic index %s", path
);
546 * - 256 index entries 4 bytes each
547 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
548 * - 20-byte SHA1 of the packfile
549 * - 20-byte SHA1 file checksum
551 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
552 munmap(idx_map
, idx_size
);
553 return error("wrong index v1 file size in %s", path
);
555 } else if (version
== 2) {
558 * - 8 bytes of header
559 * - 256 index entries 4 bytes each
560 * - 20-byte sha1 entry * nr
561 * - 4-byte crc entry * nr
562 * - 4-byte offset entry * nr
563 * - 20-byte SHA1 of the packfile
564 * - 20-byte SHA1 file checksum
565 * And after the 4-byte offset table might be a
566 * variable sized table containing 8-byte entries
567 * for offsets larger than 2^31.
569 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
570 unsigned long max_size
= min_size
;
572 max_size
+= (nr
- 1)*8;
573 if (idx_size
< min_size
|| idx_size
> max_size
) {
574 munmap(idx_map
, idx_size
);
575 return error("wrong index v2 file size in %s", path
);
577 if (idx_size
!= min_size
&&
579 * make sure we can deal with large pack offsets.
580 * 31-bit signed offset won't be enough, neither
581 * 32-bit unsigned one will be.
583 (sizeof(off_t
) <= 4)) {
584 munmap(idx_map
, idx_size
);
585 return error("pack too large for current definition of off_t in %s", path
);
589 p
->index_version
= version
;
590 p
->index_data
= idx_map
;
591 p
->index_size
= idx_size
;
596 int open_pack_index(struct packed_git
*p
)
604 idx_name
= xstrdup(p
->pack_name
);
605 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
606 ret
= check_packed_git_idx(idx_name
, p
);
611 static void scan_windows(struct packed_git
*p
,
612 struct packed_git
**lru_p
,
613 struct pack_window
**lru_w
,
614 struct pack_window
**lru_l
)
616 struct pack_window
*w
, *w_l
;
618 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
620 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
630 static int unuse_one_window(struct packed_git
*current
)
632 struct packed_git
*p
, *lru_p
= NULL
;
633 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
636 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
637 for (p
= packed_git
; p
; p
= p
->next
)
638 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
640 munmap(lru_w
->base
, lru_w
->len
);
641 pack_mapped
-= lru_w
->len
;
643 lru_l
->next
= lru_w
->next
;
645 lru_p
->windows
= lru_w
->next
;
653 void release_pack_memory(size_t need
)
655 size_t cur
= pack_mapped
;
656 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
660 void *xmmap(void *start
, size_t length
,
661 int prot
, int flags
, int fd
, off_t offset
)
663 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
664 if (ret
== MAP_FAILED
) {
667 release_pack_memory(length
);
668 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
669 if (ret
== MAP_FAILED
)
670 die_errno("Out of memory? mmap failed");
675 void close_pack_windows(struct packed_git
*p
)
678 struct pack_window
*w
= p
->windows
;
681 die("pack '%s' still has open windows to it",
683 munmap(w
->base
, w
->len
);
684 pack_mapped
-= w
->len
;
686 p
->windows
= w
->next
;
692 * The LRU pack is the one with the oldest MRU window, preferring packs
693 * with no used windows, or the oldest mtime if it has no windows allocated.
695 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
697 struct pack_window
*w
, *this_mru_w
;
698 int has_windows_inuse
= 0;
701 * Reject this pack if it has windows and the previously selected
702 * one does not. If this pack does not have windows, reject
703 * it if the pack file is newer than the previously selected one.
705 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
708 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
710 * Reject this pack if any of its windows are in use,
711 * but the previously selected pack did not have any
712 * inuse windows. Otherwise, record that this pack
713 * has windows in use.
716 if (*accept_windows_inuse
)
717 has_windows_inuse
= 1;
722 if (w
->last_used
> this_mru_w
->last_used
)
726 * Reject this pack if it has windows that have been
727 * used more recently than the previously selected pack.
728 * If the previously selected pack had windows inuse and
729 * we have not encountered a window in this pack that is
730 * inuse, skip this check since we prefer a pack with no
731 * inuse windows to one that has inuse windows.
733 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
734 this_mru_w
->last_used
> (*mru_w
)->last_used
)
743 *accept_windows_inuse
= has_windows_inuse
;
746 static int close_one_pack(void)
748 struct packed_git
*p
, *lru_p
= NULL
;
749 struct pack_window
*mru_w
= NULL
;
750 int accept_windows_inuse
= 1;
752 for (p
= packed_git
; p
; p
= p
->next
) {
753 if (p
->pack_fd
== -1)
755 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
759 close(lru_p
->pack_fd
);
768 void unuse_pack(struct pack_window
**w_cursor
)
770 struct pack_window
*w
= *w_cursor
;
777 void close_pack_index(struct packed_git
*p
)
780 munmap((void *)p
->index_data
, p
->index_size
);
781 p
->index_data
= NULL
;
786 * This is used by git-repack in case a newly created pack happens to
787 * contain the same set of objects as an existing one. In that case
788 * the resulting file might be different even if its name would be the
789 * same. It is best to close any reference to the old pack before it is
790 * replaced on disk. Of course no index pointers nor windows for given pack
791 * must subsist at this point. If ever objects from this pack are requested
792 * again, the new version of the pack will be reinitialized through
793 * reprepare_packed_git().
795 void free_pack_by_name(const char *pack_name
)
797 struct packed_git
*p
, **pp
= &packed_git
;
801 if (strcmp(pack_name
, p
->pack_name
) == 0) {
802 clear_delta_base_cache();
803 close_pack_windows(p
);
804 if (p
->pack_fd
!= -1) {
809 free(p
->bad_object_sha1
);
811 if (last_found_pack
== p
)
812 last_found_pack
= NULL
;
820 static unsigned int get_max_fd_limit(void)
825 if (getrlimit(RLIMIT_NOFILE
, &lim
))
826 die_errno("cannot get RLIMIT_NOFILE");
829 #elif defined(_SC_OPEN_MAX)
830 return sysconf(_SC_OPEN_MAX
);
831 #elif defined(OPEN_MAX)
834 return 1; /* see the caller ;-) */
839 * Do not call this directly as this leaks p->pack_fd on error return;
840 * call open_packed_git() instead.
842 static int open_packed_git_1(struct packed_git
*p
)
845 struct pack_header hdr
;
846 unsigned char sha1
[20];
847 unsigned char *idx_sha1
;
850 if (!p
->index_data
&& open_pack_index(p
))
851 return error("packfile %s index unavailable", p
->pack_name
);
854 unsigned int max_fds
= get_max_fd_limit();
856 /* Save 3 for stdin/stdout/stderr, 22 for work */
858 pack_max_fds
= max_fds
- 25;
863 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
866 p
->pack_fd
= git_open_noatime(p
->pack_name
);
867 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
871 /* If we created the struct before we had the pack we lack size. */
873 if (!S_ISREG(st
.st_mode
))
874 return error("packfile %s not a regular file", p
->pack_name
);
875 p
->pack_size
= st
.st_size
;
876 } else if (p
->pack_size
!= st
.st_size
)
877 return error("packfile %s size changed", p
->pack_name
);
879 /* We leave these file descriptors open with sliding mmap;
880 * there is no point keeping them open across exec(), though.
882 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
884 return error("cannot determine file descriptor flags");
885 fd_flag
|= FD_CLOEXEC
;
886 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
887 return error("cannot set FD_CLOEXEC");
889 /* Verify we recognize this pack file format. */
890 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
891 return error("file %s is far too short to be a packfile", p
->pack_name
);
892 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
893 return error("file %s is not a GIT packfile", p
->pack_name
);
894 if (!pack_version_ok(hdr
.hdr_version
))
895 return error("packfile %s is version %"PRIu32
" and not"
896 " supported (try upgrading GIT to a newer version)",
897 p
->pack_name
, ntohl(hdr
.hdr_version
));
899 /* Verify the pack matches its index. */
900 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
901 return error("packfile %s claims to have %"PRIu32
" objects"
902 " while index indicates %"PRIu32
" objects",
903 p
->pack_name
, ntohl(hdr
.hdr_entries
),
905 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
906 return error("end of packfile %s is unavailable", p
->pack_name
);
907 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
908 return error("packfile %s signature is unavailable", p
->pack_name
);
909 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
910 if (hashcmp(sha1
, idx_sha1
))
911 return error("packfile %s does not match index", p
->pack_name
);
915 static int open_packed_git(struct packed_git
*p
)
917 if (!open_packed_git_1(p
))
919 if (p
->pack_fd
!= -1) {
927 static int in_window(struct pack_window
*win
, off_t offset
)
929 /* We must promise at least 20 bytes (one hash) after the
930 * offset is available from this window, otherwise the offset
931 * is not actually in this window and a different window (which
932 * has that one hash excess) must be used. This is to support
933 * the object header and delta base parsing routines below.
935 off_t win_off
= win
->offset
;
936 return win_off
<= offset
937 && (offset
+ 20) <= (win_off
+ win
->len
);
940 unsigned char *use_pack(struct packed_git
*p
,
941 struct pack_window
**w_cursor
,
945 struct pack_window
*win
= *w_cursor
;
947 /* Since packfiles end in a hash of their content and it's
948 * pointless to ask for an offset into the middle of that
949 * hash, and the in_window function above wouldn't match
950 * don't allow an offset too close to the end of the file.
952 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
953 die("packfile %s cannot be accessed", p
->pack_name
);
954 if (offset
> (p
->pack_size
- 20))
955 die("offset beyond end of packfile (truncated pack?)");
957 if (!win
|| !in_window(win
, offset
)) {
960 for (win
= p
->windows
; win
; win
= win
->next
) {
961 if (in_window(win
, offset
))
965 size_t window_align
= packed_git_window_size
/ 2;
968 if (p
->pack_fd
== -1 && open_packed_git(p
))
969 die("packfile %s cannot be accessed", p
->pack_name
);
971 win
= xcalloc(1, sizeof(*win
));
972 win
->offset
= (offset
/ window_align
) * window_align
;
973 len
= p
->pack_size
- win
->offset
;
974 if (len
> packed_git_window_size
)
975 len
= packed_git_window_size
;
976 win
->len
= (size_t)len
;
977 pack_mapped
+= win
->len
;
978 while (packed_git_limit
< pack_mapped
979 && unuse_one_window(p
))
981 win
->base
= xmmap(NULL
, win
->len
,
982 PROT_READ
, MAP_PRIVATE
,
983 p
->pack_fd
, win
->offset
);
984 if (win
->base
== MAP_FAILED
)
985 die("packfile %s cannot be mapped: %s",
988 if (!win
->offset
&& win
->len
== p
->pack_size
989 && !p
->do_not_close
) {
996 if (pack_mapped
> peak_pack_mapped
)
997 peak_pack_mapped
= pack_mapped
;
998 if (pack_open_windows
> peak_pack_open_windows
)
999 peak_pack_open_windows
= pack_open_windows
;
1000 win
->next
= p
->windows
;
1004 if (win
!= *w_cursor
) {
1005 win
->last_used
= pack_used_ctr
++;
1009 offset
-= win
->offset
;
1011 *left
= win
->len
- xsize_t(offset
);
1012 return win
->base
+ offset
;
1015 static struct packed_git
*alloc_packed_git(int extra
)
1017 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1018 memset(p
, 0, sizeof(*p
));
1023 static void try_to_free_pack_memory(size_t size
)
1025 release_pack_memory(size
);
1028 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1030 static int have_set_try_to_free_routine
;
1032 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1034 if (!have_set_try_to_free_routine
) {
1035 have_set_try_to_free_routine
= 1;
1036 set_try_to_free_routine(try_to_free_pack_memory
);
1040 * Make sure a corresponding .pack file exists and that
1041 * the index looks sane.
1043 path_len
-= strlen(".idx");
1048 memcpy(p
->pack_name
, path
, path_len
);
1050 strcpy(p
->pack_name
+ path_len
, ".keep");
1051 if (!access(p
->pack_name
, F_OK
))
1054 strcpy(p
->pack_name
+ path_len
, ".pack");
1055 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1060 /* ok, it looks sane as far as we can check without
1061 * actually mapping the pack file.
1063 p
->pack_size
= st
.st_size
;
1064 p
->pack_local
= local
;
1065 p
->mtime
= st
.st_mtime
;
1066 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1071 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1073 const char *path
= sha1_pack_name(sha1
);
1074 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1076 strcpy(p
->pack_name
, path
);
1077 hashcpy(p
->sha1
, sha1
);
1078 if (check_packed_git_idx(idx_path
, p
)) {
1086 void install_packed_git(struct packed_git
*pack
)
1088 if (pack
->pack_fd
!= -1)
1091 pack
->next
= packed_git
;
1095 void (*report_garbage
)(const char *desc
, const char *path
);
1097 static void report_helper(const struct string_list
*list
,
1098 int seen_bits
, int first
, int last
)
1101 switch (seen_bits
) {
1103 msg
= "no corresponding .idx nor .pack";
1106 msg
= "no corresponding .idx";
1109 msg
= "no corresponding .pack";
1114 for (; first
< last
; first
++)
1115 report_garbage(msg
, list
->items
[first
].string
);
1118 static void report_pack_garbage(struct string_list
*list
)
1120 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1122 if (!report_garbage
)
1125 sort_string_list(list
);
1127 for (i
= 0; i
< list
->nr
; i
++) {
1128 const char *path
= list
->items
[i
].string
;
1129 if (baselen
!= -1 &&
1130 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1131 report_helper(list
, seen_bits
, first
, i
);
1135 if (baselen
== -1) {
1136 const char *dot
= strrchr(path
, '.');
1138 report_garbage("garbage found", path
);
1141 baselen
= dot
- path
+ 1;
1144 if (!strcmp(path
+ baselen
, "pack"))
1146 else if (!strcmp(path
+ baselen
, "idx"))
1149 report_helper(list
, seen_bits
, first
, list
->nr
);
1152 static void prepare_packed_git_one(char *objdir
, int local
)
1154 /* Ensure that this buffer is large enough so that we can
1155 append "/pack/" without clobbering the stack even if
1156 strlen(objdir) were PATH_MAX. */
1157 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
1161 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1163 sprintf(path
, "%s/pack", objdir
);
1165 dir
= opendir(path
);
1167 if (errno
!= ENOENT
)
1168 error("unable to open object pack directory: %s: %s",
1169 path
, strerror(errno
));
1173 while ((de
= readdir(dir
)) != NULL
) {
1174 int namelen
= strlen(de
->d_name
);
1175 struct packed_git
*p
;
1177 if (len
+ namelen
+ 1 > sizeof(path
)) {
1178 if (report_garbage
) {
1179 struct strbuf sb
= STRBUF_INIT
;
1180 strbuf_addf(&sb
, "%.*s/%s", len
- 1, path
, de
->d_name
);
1181 report_garbage("path too long", sb
.buf
);
1182 strbuf_release(&sb
);
1187 if (is_dot_or_dotdot(de
->d_name
))
1190 strcpy(path
+ len
, de
->d_name
);
1192 if (has_extension(de
->d_name
, ".idx")) {
1193 /* Don't reopen a pack we already have. */
1194 for (p
= packed_git
; p
; p
= p
->next
) {
1195 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1200 * See if it really is a valid .idx file with
1201 * corresponding .pack file that we can map.
1203 (p
= add_packed_git(path
, len
+ namelen
, local
)) != NULL
)
1204 install_packed_git(p
);
1207 if (!report_garbage
)
1210 if (has_extension(de
->d_name
, ".idx") ||
1211 has_extension(de
->d_name
, ".pack") ||
1212 has_extension(de
->d_name
, ".keep"))
1213 string_list_append(&garbage
, path
);
1215 report_garbage("garbage found", path
);
1218 report_pack_garbage(&garbage
);
1219 string_list_clear(&garbage
, 0);
1222 static int sort_pack(const void *a_
, const void *b_
)
1224 struct packed_git
*a
= *((struct packed_git
**)a_
);
1225 struct packed_git
*b
= *((struct packed_git
**)b_
);
1229 * Local packs tend to contain objects specific to our
1230 * variant of the project than remote ones. In addition,
1231 * remote ones could be on a network mounted filesystem.
1232 * Favor local ones for these reasons.
1234 st
= a
->pack_local
- b
->pack_local
;
1239 * Younger packs tend to contain more recent objects,
1240 * and more recent objects tend to get accessed more
1243 if (a
->mtime
< b
->mtime
)
1245 else if (a
->mtime
== b
->mtime
)
1250 static void rearrange_packed_git(void)
1252 struct packed_git
**ary
, *p
;
1255 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1260 /* prepare an array of packed_git for easier sorting */
1261 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1262 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1265 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1267 /* link them back again */
1268 for (i
= 0; i
< n
- 1; i
++)
1269 ary
[i
]->next
= ary
[i
+ 1];
1270 ary
[n
- 1]->next
= NULL
;
1271 packed_git
= ary
[0];
1276 static int prepare_packed_git_run_once
= 0;
1277 void prepare_packed_git(void)
1279 struct alternate_object_database
*alt
;
1281 if (prepare_packed_git_run_once
)
1283 prepare_packed_git_one(get_object_directory(), 1);
1285 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1287 prepare_packed_git_one(alt
->base
, 0);
1288 alt
->name
[-1] = '/';
1290 rearrange_packed_git();
1291 prepare_packed_git_run_once
= 1;
1294 void reprepare_packed_git(void)
1297 prepare_packed_git_run_once
= 0;
1298 prepare_packed_git();
1301 static void mark_bad_packed_object(struct packed_git
*p
,
1302 const unsigned char *sha1
)
1305 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1306 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1308 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1309 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1310 p
->num_bad_objects
++;
1313 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1315 struct packed_git
*p
;
1318 for (p
= packed_git
; p
; p
= p
->next
)
1319 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1320 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1326 * With an in-core object data in "map", rehash it to make sure the
1327 * object name actually matches "sha1" to detect object corruption.
1328 * With "map" == NULL, try reading the object named with "sha1" using
1329 * the streaming interface and rehash it to do the same.
1331 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1332 unsigned long size
, const char *type
)
1334 unsigned char real_sha1
[20];
1335 enum object_type obj_type
;
1336 struct git_istream
*st
;
1342 hash_sha1_file(map
, size
, type
, real_sha1
);
1343 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1346 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1350 /* Generate the header */
1351 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1355 git_SHA1_Update(&c
, hdr
, hdrlen
);
1357 char buf
[1024 * 16];
1358 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1366 git_SHA1_Update(&c
, buf
, readlen
);
1368 git_SHA1_Final(real_sha1
, &c
);
1370 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1373 static int git_open_noatime(const char *name
)
1375 static int sha1_file_open_flag
= O_NOATIME
;
1378 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1382 /* Might the failure be due to O_NOATIME? */
1383 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1384 sha1_file_open_flag
= 0;
1392 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1394 char *name
= sha1_file_name(sha1
);
1395 struct alternate_object_database
*alt
;
1397 if (!lstat(name
, st
))
1402 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1404 fill_sha1_path(name
, sha1
);
1405 if (!lstat(alt
->base
, st
))
1412 static int open_sha1_file(const unsigned char *sha1
)
1415 char *name
= sha1_file_name(sha1
);
1416 struct alternate_object_database
*alt
;
1418 fd
= git_open_noatime(name
);
1424 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1426 fill_sha1_path(name
, sha1
);
1427 fd
= git_open_noatime(alt
->base
);
1434 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1439 fd
= open_sha1_file(sha1
);
1444 if (!fstat(fd
, &st
)) {
1445 *size
= xsize_t(st
.st_size
);
1447 /* mmap() is forbidden on empty files */
1448 error("object file %s is empty", sha1_file_name(sha1
));
1451 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1459 * There used to be a second loose object header format which
1460 * was meant to mimic the in-pack format, allowing for direct
1461 * copy of the object data. This format turned up not to be
1462 * really worth it and we no longer write loose objects in that
1465 static int experimental_loose_object(unsigned char *map
)
1470 * We must determine if the buffer contains the standard
1471 * zlib-deflated stream or the experimental format based
1472 * on the in-pack object format. Compare the header byte
1475 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1476 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1478 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1479 * in standard loose-object format, UNLESS it is a Git-pack
1480 * format object *exactly* 8 bytes in size when inflated.
1482 * However, RFC1950 also specifies that the 1st 16-bit word
1483 * must be divisible by 31 - this checksum tells us our buffer
1484 * is in the standard format, giving a false positive only if
1485 * the 1st word of the Git-pack format object happens to be
1486 * divisible by 31, ie:
1487 * ((byte0 * 256) + byte1) % 31 = 0
1488 * => 0ttt10000www1000 % 31 = 0
1490 * As it happens, this case can only arise for www=3 & ttt=1
1491 * - ie, a Commit object, which would have to be 8 bytes in
1492 * size. As no Commit can be that small, we find that the
1493 * combination of these two criteria (bitmask & checksum)
1494 * can always correctly determine the buffer format.
1496 word
= (map
[0] << 8) + map
[1];
1497 if ((map
[0] & 0x8F) == 0x08 && !(word
% 31))
1503 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1504 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1507 unsigned long size
, c
;
1508 unsigned long used
= 0;
1511 *type
= (c
>> 4) & 7;
1515 if (len
<= used
|| bitsizeof(long) <= shift
) {
1516 error("bad object header");
1521 size
+= (c
& 0x7f) << shift
;
1528 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1530 unsigned long size
, used
;
1531 static const char valid_loose_object_type
[8] = {
1533 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1534 0, /* "delta" and others are invalid in a loose object */
1536 enum object_type type
;
1538 /* Get the data stream */
1539 memset(stream
, 0, sizeof(*stream
));
1540 stream
->next_in
= map
;
1541 stream
->avail_in
= mapsize
;
1542 stream
->next_out
= buffer
;
1543 stream
->avail_out
= bufsiz
;
1545 if (experimental_loose_object(map
)) {
1547 * The old experimental format we no longer produce;
1548 * we can still read it.
1550 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1551 if (!used
|| !valid_loose_object_type
[type
])
1556 /* Set up the stream for the rest.. */
1557 stream
->next_in
= map
;
1558 stream
->avail_in
= mapsize
;
1559 git_inflate_init(stream
);
1561 /* And generate the fake traditional header */
1562 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1563 typename(type
), size
);
1566 git_inflate_init(stream
);
1567 return git_inflate(stream
, 0);
1570 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1572 int bytes
= strlen(buffer
) + 1;
1573 unsigned char *buf
= xmallocz(size
);
1577 n
= stream
->total_out
- bytes
;
1580 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1582 if (bytes
<= size
) {
1584 * The above condition must be (bytes <= size), not
1585 * (bytes < size). In other words, even though we
1586 * expect no more output and set avail_out to zero,
1587 * the input zlib stream may have bytes that express
1588 * "this concludes the stream", and we *do* want to
1591 * Otherwise we would not be able to test that we
1592 * consumed all the input to reach the expected size;
1593 * we also want to check that zlib tells us that all
1594 * went well with status == Z_STREAM_END at the end.
1596 stream
->next_out
= buf
+ bytes
;
1597 stream
->avail_out
= size
- bytes
;
1598 while (status
== Z_OK
)
1599 status
= git_inflate(stream
, Z_FINISH
);
1601 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1602 git_inflate_end(stream
);
1607 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1608 else if (stream
->avail_in
)
1609 error("garbage at end of loose object '%s'",
1616 * We used to just use "sscanf()", but that's actually way
1617 * too permissive for what we want to check. So do an anal
1618 * object header parse by hand.
1620 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1627 * The type can be at most ten bytes (including the
1628 * terminating '\0' that we add), and is followed by
1637 if (i
>= sizeof(type
))
1643 * The length must follow immediately, and be in canonical
1644 * decimal format (ie "010" is not valid).
1646 size
= *hdr
++ - '0';
1651 unsigned long c
= *hdr
- '0';
1655 size
= size
* 10 + c
;
1661 * The length must be followed by a zero byte
1663 return *hdr
? -1 : type_from_string(type
);
1666 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1672 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1673 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1676 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1679 unsigned long get_size_from_delta(struct packed_git
*p
,
1680 struct pack_window
**w_curs
,
1683 const unsigned char *data
;
1684 unsigned char delta_head
[20], *in
;
1688 memset(&stream
, 0, sizeof(stream
));
1689 stream
.next_out
= delta_head
;
1690 stream
.avail_out
= sizeof(delta_head
);
1692 git_inflate_init(&stream
);
1694 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1695 stream
.next_in
= in
;
1696 st
= git_inflate(&stream
, Z_FINISH
);
1697 curpos
+= stream
.next_in
- in
;
1698 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1699 stream
.total_out
< sizeof(delta_head
));
1700 git_inflate_end(&stream
);
1701 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1702 error("delta data unpack-initial failed");
1706 /* Examine the initial part of the delta to figure out
1711 /* ignore base size */
1712 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1714 /* Read the result size */
1715 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1718 static off_t
get_delta_base(struct packed_git
*p
,
1719 struct pack_window
**w_curs
,
1721 enum object_type type
,
1722 off_t delta_obj_offset
)
1724 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1727 /* use_pack() assured us we have [base_info, base_info + 20)
1728 * as a range that we can look at without walking off the
1729 * end of the mapped window. Its actually the hash size
1730 * that is assured. An OFS_DELTA longer than the hash size
1731 * is stupid, as then a REF_DELTA would be smaller to store.
1733 if (type
== OBJ_OFS_DELTA
) {
1735 unsigned char c
= base_info
[used
++];
1736 base_offset
= c
& 127;
1739 if (!base_offset
|| MSB(base_offset
, 7))
1740 return 0; /* overflow */
1741 c
= base_info
[used
++];
1742 base_offset
= (base_offset
<< 7) + (c
& 127);
1744 base_offset
= delta_obj_offset
- base_offset
;
1745 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1746 return 0; /* out of bound */
1748 } else if (type
== OBJ_REF_DELTA
) {
1749 /* The base entry _must_ be in the same pack */
1750 base_offset
= find_pack_entry_one(base_info
, p
);
1753 die("I am totally screwed");
1757 int unpack_object_header(struct packed_git
*p
,
1758 struct pack_window
**w_curs
,
1760 unsigned long *sizep
)
1762 unsigned char *base
;
1765 enum object_type type
;
1767 /* use_pack() assures us we have [base, base + 20) available
1768 * as a range that we can look at. (Its actually the hash
1769 * size that is assured.) With our object header encoding
1770 * the maximum deflated object size is 2^137, which is just
1771 * insane, so we know won't exceed what we have been given.
1773 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1774 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1783 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1786 struct revindex_entry
*revidx
;
1787 const unsigned char *sha1
;
1788 revidx
= find_pack_revindex(p
, obj_offset
);
1791 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1792 mark_bad_packed_object(p
, sha1
);
1793 type
= sha1_object_info(sha1
, NULL
);
1794 if (type
<= OBJ_NONE
)
1799 #define POI_STACK_PREALLOC 64
1801 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1803 enum object_type type
,
1804 struct pack_window
**w_curs
,
1807 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1808 off_t
*poi_stack
= small_poi_stack
;
1809 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1811 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1814 /* Push the object we're going to leave behind */
1815 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1816 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1817 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1818 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1820 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1822 poi_stack
[poi_stack_nr
++] = obj_offset
;
1823 /* If parsing the base offset fails, just unwind */
1824 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1827 curpos
= obj_offset
= base_offset
;
1828 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1829 if (type
<= OBJ_NONE
) {
1830 /* If getting the base itself fails, we first
1831 * retry the base, otherwise unwind */
1832 type
= retry_bad_packed_offset(p
, base_offset
);
1833 if (type
> OBJ_NONE
)
1847 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1848 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1853 if (poi_stack
!= small_poi_stack
)
1858 while (poi_stack_nr
) {
1859 obj_offset
= poi_stack
[--poi_stack_nr
];
1860 type
= retry_bad_packed_offset(p
, obj_offset
);
1861 if (type
> OBJ_NONE
)
1868 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1869 struct object_info
*oi
)
1871 struct pack_window
*w_curs
= NULL
;
1873 off_t curpos
= obj_offset
;
1874 enum object_type type
;
1877 * We always get the representation type, but only convert it to
1878 * a "real" type later if the caller is interested.
1880 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1883 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1884 off_t tmp_pos
= curpos
;
1885 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1891 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1892 if (*oi
->sizep
== 0) {
1901 if (oi
->disk_sizep
) {
1902 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1903 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1907 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1908 if (*oi
->typep
< 0) {
1915 unuse_pack(&w_curs
);
1919 static void *unpack_compressed_entry(struct packed_git
*p
,
1920 struct pack_window
**w_curs
,
1926 unsigned char *buffer
, *in
;
1928 buffer
= xmallocz(size
);
1929 memset(&stream
, 0, sizeof(stream
));
1930 stream
.next_out
= buffer
;
1931 stream
.avail_out
= size
+ 1;
1933 git_inflate_init(&stream
);
1935 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1936 stream
.next_in
= in
;
1937 st
= git_inflate(&stream
, Z_FINISH
);
1938 if (!stream
.avail_out
)
1939 break; /* the payload is larger than it should be */
1940 curpos
+= stream
.next_in
- in
;
1941 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1942 git_inflate_end(&stream
);
1943 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1951 #define MAX_DELTA_CACHE (256)
1953 static size_t delta_base_cached
;
1955 static struct delta_base_cache_lru_list
{
1956 struct delta_base_cache_lru_list
*prev
;
1957 struct delta_base_cache_lru_list
*next
;
1958 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1960 static struct delta_base_cache_entry
{
1961 struct delta_base_cache_lru_list lru
;
1963 struct packed_git
*p
;
1966 enum object_type type
;
1967 } delta_base_cache
[MAX_DELTA_CACHE
];
1969 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1973 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1974 hash
+= (hash
>> 8) + (hash
>> 16);
1975 return hash
% MAX_DELTA_CACHE
;
1978 static struct delta_base_cache_entry
*
1979 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1981 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1982 return delta_base_cache
+ hash
;
1985 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
1986 struct packed_git
*p
, off_t base_offset
)
1988 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
1991 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1993 struct delta_base_cache_entry
*ent
;
1994 ent
= get_delta_base_cache_entry(p
, base_offset
);
1995 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
1998 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2001 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2002 ent
->lru
.prev
->next
= ent
->lru
.next
;
2003 delta_base_cached
-= ent
->size
;
2006 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2007 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2009 struct delta_base_cache_entry
*ent
;
2012 ent
= get_delta_base_cache_entry(p
, base_offset
);
2014 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2015 return unpack_entry(p
, base_offset
, type
, base_size
);
2020 clear_delta_base_cache_entry(ent
);
2022 ret
= xmemdupz(ent
->data
, ent
->size
);
2024 *base_size
= ent
->size
;
2028 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2033 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2034 ent
->lru
.prev
->next
= ent
->lru
.next
;
2035 delta_base_cached
-= ent
->size
;
2039 void clear_delta_base_cache(void)
2042 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2043 release_delta_base_cache(&delta_base_cache
[p
]);
2046 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2047 void *base
, unsigned long base_size
, enum object_type type
)
2049 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2050 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2051 struct delta_base_cache_lru_list
*lru
;
2053 release_delta_base_cache(ent
);
2054 delta_base_cached
+= base_size
;
2056 for (lru
= delta_base_cache_lru
.next
;
2057 delta_base_cached
> delta_base_cache_limit
2058 && lru
!= &delta_base_cache_lru
;
2060 struct delta_base_cache_entry
*f
= (void *)lru
;
2061 if (f
->type
== OBJ_BLOB
)
2062 release_delta_base_cache(f
);
2064 for (lru
= delta_base_cache_lru
.next
;
2065 delta_base_cached
> delta_base_cache_limit
2066 && lru
!= &delta_base_cache_lru
;
2068 struct delta_base_cache_entry
*f
= (void *)lru
;
2069 release_delta_base_cache(f
);
2073 ent
->base_offset
= base_offset
;
2076 ent
->size
= base_size
;
2077 ent
->lru
.next
= &delta_base_cache_lru
;
2078 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2079 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2080 delta_base_cache_lru
.prev
= &ent
->lru
;
2083 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2084 unsigned long *size
);
2086 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2088 static FILE *log_file
;
2090 if (!log_pack_access
)
2091 log_pack_access
= getenv("GIT_TRACE_PACK_ACCESS");
2092 if (!log_pack_access
)
2093 log_pack_access
= no_log_pack_access
;
2094 if (log_pack_access
== no_log_pack_access
)
2098 log_file
= fopen(log_pack_access
, "w");
2100 error("cannot open pack access log '%s' for writing: %s",
2101 log_pack_access
, strerror(errno
));
2102 log_pack_access
= no_log_pack_access
;
2106 fprintf(log_file
, "%s %"PRIuMAX
"\n",
2107 p
->pack_name
, (uintmax_t)obj_offset
);
2111 int do_check_packed_object_crc
;
2113 #define UNPACK_ENTRY_STACK_PREALLOC 64
2114 struct unpack_entry_stack_ent
{
2120 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2121 enum object_type
*final_type
, unsigned long *final_size
)
2123 struct pack_window
*w_curs
= NULL
;
2124 off_t curpos
= obj_offset
;
2127 enum object_type type
;
2128 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2129 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2130 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2131 int base_from_cache
= 0;
2133 if (log_pack_access
!= no_log_pack_access
)
2134 write_pack_access_log(p
, obj_offset
);
2136 /* PHASE 1: drill down to the innermost base object */
2140 struct delta_base_cache_entry
*ent
;
2142 ent
= get_delta_base_cache_entry(p
, curpos
);
2143 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2147 clear_delta_base_cache_entry(ent
);
2148 base_from_cache
= 1;
2152 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2153 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2154 unsigned long len
= revidx
[1].offset
- obj_offset
;
2155 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2156 const unsigned char *sha1
=
2157 nth_packed_object_sha1(p
, revidx
->nr
);
2158 error("bad packed object CRC for %s",
2160 mark_bad_packed_object(p
, sha1
);
2161 unuse_pack(&w_curs
);
2166 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2167 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2170 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2172 error("failed to validate delta base reference "
2173 "at offset %"PRIuMAX
" from %s",
2174 (uintmax_t)curpos
, p
->pack_name
);
2175 /* bail to phase 2, in hopes of recovery */
2180 /* push object, proceed to base */
2181 if (delta_stack_nr
>= delta_stack_alloc
2182 && delta_stack
== small_delta_stack
) {
2183 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2184 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2185 memcpy(delta_stack
, small_delta_stack
,
2186 sizeof(*delta_stack
)*delta_stack_nr
);
2188 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2190 i
= delta_stack_nr
++;
2191 delta_stack
[i
].obj_offset
= obj_offset
;
2192 delta_stack
[i
].curpos
= curpos
;
2193 delta_stack
[i
].size
= size
;
2195 curpos
= obj_offset
= base_offset
;
2198 /* PHASE 2: handle the base */
2203 die("BUG in unpack_entry: left loop at a valid delta");
2209 if (!base_from_cache
)
2210 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2214 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2215 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2218 /* PHASE 3: apply deltas in order */
2221 * 'data' holds the base data, or NULL if there was corruption
2223 while (delta_stack_nr
) {
2226 unsigned long delta_size
, base_size
= size
;
2232 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2236 * We're probably in deep shit, but let's try to fetch
2237 * the required base anyway from another pack or loose.
2238 * This is costly but should happen only in the presence
2239 * of a corrupted pack, and is better than failing outright.
2241 struct revindex_entry
*revidx
;
2242 const unsigned char *base_sha1
;
2243 revidx
= find_pack_revindex(p
, obj_offset
);
2245 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2246 error("failed to read delta base object %s"
2247 " at offset %"PRIuMAX
" from %s",
2248 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2250 mark_bad_packed_object(p
, base_sha1
);
2251 base
= read_object(base_sha1
, &type
, &base_size
);
2255 i
= --delta_stack_nr
;
2256 obj_offset
= delta_stack
[i
].obj_offset
;
2257 curpos
= delta_stack
[i
].curpos
;
2258 delta_size
= delta_stack
[i
].size
;
2263 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2266 error("failed to unpack compressed delta "
2267 "at offset %"PRIuMAX
" from %s",
2268 (uintmax_t)curpos
, p
->pack_name
);
2273 data
= patch_delta(base
, base_size
,
2274 delta_data
, delta_size
,
2278 * We could not apply the delta; warn the user, but keep going.
2279 * Our failure will be noticed either in the next iteration of
2280 * the loop, or if this is the final delta, in the caller when
2281 * we return NULL. Those code paths will take care of making
2282 * a more explicit warning and retrying with another copy of
2286 error("failed to apply delta");
2294 unuse_pack(&w_curs
);
2298 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2301 const unsigned char *index
= p
->index_data
;
2303 if (open_pack_index(p
))
2305 index
= p
->index_data
;
2307 if (n
>= p
->num_objects
)
2310 if (p
->index_version
== 1) {
2311 return index
+ 24 * n
+ 4;
2314 return index
+ 20 * n
;
2318 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2320 const unsigned char *index
= p
->index_data
;
2322 if (p
->index_version
== 1) {
2323 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2326 index
+= 8 + p
->num_objects
* (20 + 4);
2327 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2328 if (!(off
& 0x80000000))
2330 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2331 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2332 ntohl(*((uint32_t *)(index
+ 4)));
2336 off_t
find_pack_entry_one(const unsigned char *sha1
,
2337 struct packed_git
*p
)
2339 const uint32_t *level1_ofs
= p
->index_data
;
2340 const unsigned char *index
= p
->index_data
;
2341 unsigned hi
, lo
, stride
;
2342 static int use_lookup
= -1;
2343 static int debug_lookup
= -1;
2345 if (debug_lookup
< 0)
2346 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2349 if (open_pack_index(p
))
2351 level1_ofs
= p
->index_data
;
2352 index
= p
->index_data
;
2354 if (p
->index_version
> 1) {
2359 hi
= ntohl(level1_ofs
[*sha1
]);
2360 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2361 if (p
->index_version
> 1) {
2369 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2370 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2373 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2375 int pos
= sha1_entry_pos(index
, stride
, 0,
2376 lo
, hi
, p
->num_objects
, sha1
);
2379 return nth_packed_object_offset(p
, pos
);
2383 unsigned mi
= (lo
+ hi
) / 2;
2384 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2387 printf("lo %u hi %u rg %u mi %u\n",
2388 lo
, hi
, hi
- lo
, mi
);
2390 return nth_packed_object_offset(p
, mi
);
2399 int is_pack_valid(struct packed_git
*p
)
2401 /* An already open pack is known to be valid. */
2402 if (p
->pack_fd
!= -1)
2405 /* If the pack has one window completely covering the
2406 * file size, the pack is known to be valid even if
2407 * the descriptor is not currently open.
2410 struct pack_window
*w
= p
->windows
;
2412 if (!w
->offset
&& w
->len
== p
->pack_size
)
2416 /* Force the pack to open to prove its valid. */
2417 return !open_packed_git(p
);
2420 static int fill_pack_entry(const unsigned char *sha1
,
2421 struct pack_entry
*e
,
2422 struct packed_git
*p
)
2426 if (p
->num_bad_objects
) {
2428 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2429 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2433 offset
= find_pack_entry_one(sha1
, p
);
2438 * We are about to tell the caller where they can locate the
2439 * requested object. We better make sure the packfile is
2440 * still here and can be accessed before supplying that
2441 * answer, as it may have been deleted since the index was
2444 if (!is_pack_valid(p
)) {
2445 warning("packfile %s cannot be accessed", p
->pack_name
);
2450 hashcpy(e
->sha1
, sha1
);
2454 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2456 struct packed_git
*p
;
2458 prepare_packed_git();
2462 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2465 for (p
= packed_git
; p
; p
= p
->next
) {
2466 if (p
== last_found_pack
|| !fill_pack_entry(sha1
, e
, p
))
2469 last_found_pack
= p
;
2475 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2476 struct packed_git
*packs
)
2478 struct packed_git
*p
;
2480 for (p
= packs
; p
; p
= p
->next
) {
2481 if (find_pack_entry_one(sha1
, p
))
2488 static int sha1_loose_object_info(const unsigned char *sha1
,
2489 struct object_info
*oi
)
2492 unsigned long mapsize
, size
;
2498 * If we don't care about type or size, then we don't
2499 * need to look inside the object at all. Note that we
2500 * do not optimize out the stat call, even if the
2501 * caller doesn't care about the disk-size, since our
2502 * return value implicitly indicates whether the
2503 * object even exists.
2505 if (!oi
->typep
&& !oi
->sizep
) {
2507 if (stat_sha1_file(sha1
, &st
) < 0)
2510 *oi
->disk_sizep
= st
.st_size
;
2514 map
= map_sha1_file(sha1
, &mapsize
);
2518 *oi
->disk_sizep
= mapsize
;
2519 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2520 status
= error("unable to unpack %s header",
2522 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2523 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2526 git_inflate_end(&stream
);
2527 munmap(map
, mapsize
);
2529 *oi
->typep
= status
;
2533 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
)
2535 struct cached_object
*co
;
2536 struct pack_entry e
;
2539 co
= find_cached_object(sha1
);
2542 *(oi
->typep
) = co
->type
;
2544 *(oi
->sizep
) = co
->size
;
2546 *(oi
->disk_sizep
) = 0;
2547 oi
->whence
= OI_CACHED
;
2551 if (!find_pack_entry(sha1
, &e
)) {
2552 /* Most likely it's a loose object. */
2553 if (!sha1_loose_object_info(sha1
, oi
)) {
2554 oi
->whence
= OI_LOOSE
;
2558 /* Not a loose object; someone else may have just packed it. */
2559 reprepare_packed_git();
2560 if (!find_pack_entry(sha1
, &e
))
2564 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2566 mark_bad_packed_object(e
.p
, sha1
);
2567 return sha1_object_info_extended(sha1
, oi
);
2568 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2569 oi
->whence
= OI_DBCACHED
;
2571 oi
->whence
= OI_PACKED
;
2572 oi
->u
.packed
.offset
= e
.offset
;
2573 oi
->u
.packed
.pack
= e
.p
;
2574 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2575 rtype
== OBJ_OFS_DELTA
);
2581 /* returns enum object_type or negative */
2582 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2584 enum object_type type
;
2585 struct object_info oi
= {NULL
};
2589 if (sha1_object_info_extended(sha1
, &oi
) < 0)
2594 static void *read_packed_sha1(const unsigned char *sha1
,
2595 enum object_type
*type
, unsigned long *size
)
2597 struct pack_entry e
;
2600 if (!find_pack_entry(sha1
, &e
))
2602 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2605 * We're probably in deep shit, but let's try to fetch
2606 * the required object anyway from another pack or loose.
2607 * This should happen only in the presence of a corrupted
2608 * pack, and is better than failing outright.
2610 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2611 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2612 mark_bad_packed_object(e
.p
, sha1
);
2613 data
= read_object(sha1
, type
, size
);
2618 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2619 unsigned char *sha1
)
2621 struct cached_object
*co
;
2623 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2624 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2626 if (cached_object_alloc
<= cached_object_nr
) {
2627 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2628 cached_objects
= xrealloc(cached_objects
,
2629 sizeof(*cached_objects
) *
2630 cached_object_alloc
);
2632 co
= &cached_objects
[cached_object_nr
++];
2635 co
->buf
= xmalloc(len
);
2636 memcpy(co
->buf
, buf
, len
);
2637 hashcpy(co
->sha1
, sha1
);
2641 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2642 unsigned long *size
)
2644 unsigned long mapsize
;
2646 struct cached_object
*co
;
2648 co
= find_cached_object(sha1
);
2652 return xmemdupz(co
->buf
, co
->size
);
2655 buf
= read_packed_sha1(sha1
, type
, size
);
2658 map
= map_sha1_file(sha1
, &mapsize
);
2660 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2661 munmap(map
, mapsize
);
2664 reprepare_packed_git();
2665 return read_packed_sha1(sha1
, type
, size
);
2669 * This function dies on corrupt objects; the callers who want to
2670 * deal with them should arrange to call read_object() and give error
2671 * messages themselves.
2673 void *read_sha1_file_extended(const unsigned char *sha1
,
2674 enum object_type
*type
,
2675 unsigned long *size
,
2680 const struct packed_git
*p
;
2681 const unsigned char *repl
= (flag
& READ_SHA1_FILE_REPLACE
)
2682 ? lookup_replace_object(sha1
) : sha1
;
2685 data
= read_object(repl
, type
, size
);
2689 if (errno
&& errno
!= ENOENT
)
2690 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2692 /* die if we replaced an object with one that does not exist */
2694 die("replacement %s not found for %s",
2695 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2697 if (has_loose_object(repl
)) {
2698 path
= sha1_file_name(sha1
);
2699 die("loose object %s (stored in %s) is corrupt",
2700 sha1_to_hex(repl
), path
);
2703 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2704 die("packed object %s (stored in %s) is corrupt",
2705 sha1_to_hex(repl
), p
->pack_name
);
2710 void *read_object_with_reference(const unsigned char *sha1
,
2711 const char *required_type_name
,
2712 unsigned long *size
,
2713 unsigned char *actual_sha1_return
)
2715 enum object_type type
, required_type
;
2717 unsigned long isize
;
2718 unsigned char actual_sha1
[20];
2720 required_type
= type_from_string(required_type_name
);
2721 hashcpy(actual_sha1
, sha1
);
2723 int ref_length
= -1;
2724 const char *ref_type
= NULL
;
2726 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2729 if (type
== required_type
) {
2731 if (actual_sha1_return
)
2732 hashcpy(actual_sha1_return
, actual_sha1
);
2735 /* Handle references */
2736 else if (type
== OBJ_COMMIT
)
2738 else if (type
== OBJ_TAG
)
2739 ref_type
= "object ";
2744 ref_length
= strlen(ref_type
);
2746 if (ref_length
+ 40 > isize
||
2747 memcmp(buffer
, ref_type
, ref_length
) ||
2748 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2753 /* Now we have the ID of the referred-to object in
2754 * actual_sha1. Check again. */
2758 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2759 const char *type
, unsigned char *sha1
,
2760 char *hdr
, int *hdrlen
)
2764 /* Generate the header */
2765 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2769 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2770 git_SHA1_Update(&c
, buf
, len
);
2771 git_SHA1_Final(sha1
, &c
);
2775 * Move the just written object into its final resting place.
2776 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2777 * "moving" is only a part of what it does, when no patch between
2778 * master to pu changes the call sites of this function.
2780 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2784 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2786 else if (link(tmpfile
, filename
))
2790 * Coda hack - coda doesn't like cross-directory links,
2791 * so we fall back to a rename, which will mean that it
2792 * won't be able to check collisions, but that's not a
2795 * The same holds for FAT formatted media.
2797 * When this succeeds, we just return. We have nothing
2800 if (ret
&& ret
!= EEXIST
) {
2802 if (!rename(tmpfile
, filename
))
2806 unlink_or_warn(tmpfile
);
2808 if (ret
!= EEXIST
) {
2809 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2811 /* FIXME!!! Collision check here ? */
2815 if (adjust_shared_perm(filename
))
2816 return error("unable to set permission to '%s'", filename
);
2820 static int write_buffer(int fd
, const void *buf
, size_t len
)
2822 if (write_in_full(fd
, buf
, len
) < 0)
2823 return error("file write error (%s)", strerror(errno
));
2827 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2828 unsigned char *sha1
)
2832 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2836 /* Finalize a file on disk, and close it. */
2837 static void close_sha1_file(int fd
)
2839 if (fsync_object_files
)
2840 fsync_or_die(fd
, "sha1 file");
2842 die_errno("error when closing sha1 file");
2845 /* Size of directory component, including the ending '/' */
2846 static inline int directory_size(const char *filename
)
2848 const char *s
= strrchr(filename
, '/');
2851 return s
- filename
+ 1;
2855 * This creates a temporary file in the same directory as the final
2858 * We want to avoid cross-directory filename renames, because those
2859 * can have problems on various filesystems (FAT, NFS, Coda).
2861 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2863 int fd
, dirlen
= directory_size(filename
);
2865 if (dirlen
+ 20 > bufsiz
) {
2866 errno
= ENAMETOOLONG
;
2869 memcpy(buffer
, filename
, dirlen
);
2870 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2871 fd
= git_mkstemp_mode(buffer
, 0444);
2872 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2873 /* Make sure the directory exists */
2874 memcpy(buffer
, filename
, dirlen
);
2875 buffer
[dirlen
-1] = 0;
2876 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
2878 if (adjust_shared_perm(buffer
))
2882 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2883 fd
= git_mkstemp_mode(buffer
, 0444);
2888 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2889 const void *buf
, unsigned long len
, time_t mtime
)
2892 unsigned char compressed
[4096];
2895 unsigned char parano_sha1
[20];
2897 static char tmp_file
[PATH_MAX
];
2899 filename
= sha1_file_name(sha1
);
2900 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2902 if (errno
== EACCES
)
2903 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2905 return error("unable to create temporary file: %s", strerror(errno
));
2909 memset(&stream
, 0, sizeof(stream
));
2910 git_deflate_init(&stream
, zlib_compression_level
);
2911 stream
.next_out
= compressed
;
2912 stream
.avail_out
= sizeof(compressed
);
2915 /* First header.. */
2916 stream
.next_in
= (unsigned char *)hdr
;
2917 stream
.avail_in
= hdrlen
;
2918 while (git_deflate(&stream
, 0) == Z_OK
)
2920 git_SHA1_Update(&c
, hdr
, hdrlen
);
2922 /* Then the data itself.. */
2923 stream
.next_in
= (void *)buf
;
2924 stream
.avail_in
= len
;
2926 unsigned char *in0
= stream
.next_in
;
2927 ret
= git_deflate(&stream
, Z_FINISH
);
2928 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2929 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2930 die("unable to write sha1 file");
2931 stream
.next_out
= compressed
;
2932 stream
.avail_out
= sizeof(compressed
);
2933 } while (ret
== Z_OK
);
2935 if (ret
!= Z_STREAM_END
)
2936 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2937 ret
= git_deflate_end_gently(&stream
);
2939 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2940 git_SHA1_Final(parano_sha1
, &c
);
2941 if (hashcmp(sha1
, parano_sha1
) != 0)
2942 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2944 close_sha1_file(fd
);
2949 utb
.modtime
= mtime
;
2950 if (utime(tmp_file
, &utb
) < 0)
2951 warning("failed utime() on %s: %s",
2952 tmp_file
, strerror(errno
));
2955 return move_temp_to_file(tmp_file
, filename
);
2958 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2960 unsigned char sha1
[20];
2964 /* Normally if we have it in the pack then we do not bother writing
2965 * it out into .git/objects/??/?{38} file.
2967 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2969 hashcpy(returnsha1
, sha1
);
2970 if (has_sha1_file(sha1
))
2972 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2975 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2979 enum object_type type
;
2984 if (has_loose_object(sha1
))
2986 buf
= read_packed_sha1(sha1
, &type
, &len
);
2988 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2989 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2990 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2996 int has_pack_index(const unsigned char *sha1
)
2999 if (stat(sha1_pack_index_name(sha1
), &st
))
3004 int has_sha1_pack(const unsigned char *sha1
)
3006 struct pack_entry e
;
3007 return find_pack_entry(sha1
, &e
);
3010 int has_sha1_file(const unsigned char *sha1
)
3012 struct pack_entry e
;
3014 if (find_pack_entry(sha1
, &e
))
3016 if (has_loose_object(sha1
))
3018 reprepare_packed_git();
3019 return find_pack_entry(sha1
, &e
);
3022 static void check_tree(const void *buf
, size_t size
)
3024 struct tree_desc desc
;
3025 struct name_entry entry
;
3027 init_tree_desc(&desc
, buf
, size
);
3028 while (tree_entry(&desc
, &entry
))
3030 * tree_entry() will die() on malformed entries */
3034 static void check_commit(const void *buf
, size_t size
)
3037 memset(&c
, 0, sizeof(c
));
3038 if (parse_commit_buffer(&c
, buf
, size
))
3039 die("corrupt commit");
3042 static void check_tag(const void *buf
, size_t size
)
3045 memset(&t
, 0, sizeof(t
));
3046 if (parse_tag_buffer(&t
, buf
, size
))
3050 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3051 enum object_type type
,
3052 const char *path
, unsigned flags
)
3054 int ret
, re_allocated
= 0;
3055 int write_object
= flags
& HASH_WRITE_OBJECT
;
3061 * Convert blobs to git internal format
3063 if ((type
== OBJ_BLOB
) && path
) {
3064 struct strbuf nbuf
= STRBUF_INIT
;
3065 if (convert_to_git(path
, buf
, size
, &nbuf
,
3066 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3067 buf
= strbuf_detach(&nbuf
, &size
);
3071 if (flags
& HASH_FORMAT_CHECK
) {
3072 if (type
== OBJ_TREE
)
3073 check_tree(buf
, size
);
3074 if (type
== OBJ_COMMIT
)
3075 check_commit(buf
, size
);
3076 if (type
== OBJ_TAG
)
3077 check_tag(buf
, size
);
3081 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3083 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3089 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3090 const char *path
, unsigned flags
)
3092 struct strbuf sbuf
= STRBUF_INIT
;
3095 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3096 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3099 strbuf_release(&sbuf
);
3103 #define SMALL_FILE_SIZE (32*1024)
3105 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3106 enum object_type type
, const char *path
,
3112 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
3113 } else if (size
<= SMALL_FILE_SIZE
) {
3114 char *buf
= xmalloc(size
);
3115 if (size
== read_in_full(fd
, buf
, size
))
3116 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3118 ret
= error("short read %s", strerror(errno
));
3121 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3122 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3129 * This creates one packfile per large blob unless bulk-checkin
3130 * machinery is "plugged".
3132 * This also bypasses the usual "convert-to-git" dance, and that is on
3133 * purpose. We could write a streaming version of the converting
3134 * functions and insert that before feeding the data to fast-import
3135 * (or equivalent in-core API described above). However, that is
3136 * somewhat complicated, as we do not know the size of the filter
3137 * result, which we need to know beforehand when writing a git object.
3138 * Since the primary motivation for trying to stream from the working
3139 * tree file and to avoid mmaping it in core is to deal with large
3140 * binary blobs, they generally do not want to get any conversion, and
3141 * callers should avoid this code path when filters are requested.
3143 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3144 enum object_type type
, const char *path
,
3147 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3150 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3151 enum object_type type
, const char *path
, unsigned flags
)
3154 size_t size
= xsize_t(st
->st_size
);
3156 if (!S_ISREG(st
->st_mode
))
3157 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3158 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3159 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
3160 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
3162 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
3167 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3170 struct strbuf sb
= STRBUF_INIT
;
3172 switch (st
->st_mode
& S_IFMT
) {
3174 fd
= open(path
, O_RDONLY
);
3176 return error("open(\"%s\"): %s", path
,
3178 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3179 return error("%s: failed to insert into database",
3183 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3184 char *errstr
= strerror(errno
);
3185 return error("readlink(\"%s\"): %s", path
,
3188 if (!(flags
& HASH_WRITE_OBJECT
))
3189 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3190 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3191 return error("%s: failed to insert into database",
3193 strbuf_release(&sb
);
3196 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3198 return error("%s: unsupported file type", path
);
3203 int read_pack_header(int fd
, struct pack_header
*header
)
3205 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3206 /* "eof before pack header was fully read" */
3207 return PH_ERROR_EOF
;
3209 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3210 /* "protocol error (pack signature mismatch detected)" */
3211 return PH_ERROR_PACK_SIGNATURE
;
3212 if (!pack_version_ok(header
->hdr_version
))
3213 /* "protocol error (pack version unsupported)" */
3214 return PH_ERROR_PROTOCOL
;
3218 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3220 enum object_type type
= sha1_object_info(sha1
, NULL
);
3222 die("%s is not a valid object", sha1_to_hex(sha1
));
3224 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),