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];
40 * This is meant to hold a *small* number of objects that you would
41 * want read_sha1_file() to be able to return, but yet you do not want
42 * to write them into the object store (e.g. a browse-only
45 static struct cached_object
{
46 unsigned char sha1
[20];
47 enum object_type type
;
51 static int cached_object_nr
, cached_object_alloc
;
53 static struct cached_object empty_tree
= {
54 EMPTY_TREE_SHA1_BIN_LITERAL
,
61 * A pointer to the last packed_git in which an object was found.
62 * When an object is sought, we look in this packfile first, because
63 * objects that are looked up at similar times are often in the same
64 * packfile as one another.
66 static struct packed_git
*last_found_pack
;
68 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
71 struct cached_object
*co
= cached_objects
;
73 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
74 if (!hashcmp(co
->sha1
, sha1
))
77 if (!hashcmp(sha1
, empty_tree
.sha1
))
82 int mkdir_in_gitdir(const char *path
)
84 if (mkdir(path
, 0777)) {
85 int saved_errno
= errno
;
87 struct strbuf sb
= STRBUF_INIT
;
92 * Are we looking at a path in a symlinked worktree
93 * whose original repository does not yet have it?
94 * e.g. .git/rr-cache pointing at its original
95 * repository in which the user hasn't performed any
96 * conflict resolution yet?
98 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
99 strbuf_readlink(&sb
, path
, st
.st_size
) ||
100 !is_absolute_path(sb
.buf
) ||
101 mkdir(sb
.buf
, 0777)) {
108 return adjust_shared_perm(path
);
111 enum scld_error
safe_create_leading_directories(char *path
)
113 char *next_component
= path
+ offset_1st_component(path
);
114 enum scld_error ret
= SCLD_OK
;
116 while (ret
== SCLD_OK
&& next_component
) {
118 char *slash
= next_component
, slash_character
;
120 while (*slash
&& !is_dir_sep(*slash
))
126 next_component
= slash
+ 1;
127 while (is_dir_sep(*next_component
))
129 if (!*next_component
)
132 slash_character
= *slash
;
134 if (!stat(path
, &st
)) {
136 if (!S_ISDIR(st
.st_mode
))
138 } else if (mkdir(path
, 0777)) {
139 if (errno
== EEXIST
&&
140 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
141 ; /* somebody created it since we checked */
142 else if (errno
== ENOENT
)
144 * Either mkdir() failed because
145 * somebody just pruned the containing
146 * directory, or stat() failed because
147 * the file that was in our way was
148 * just removed. Either way, inform
149 * the caller that it might be worth
155 } else if (adjust_shared_perm(path
)) {
158 *slash
= slash_character
;
163 enum scld_error
safe_create_leading_directories_const(const char *path
)
165 /* path points to cache entries, so xstrdup before messing with it */
166 char *buf
= xstrdup(path
);
167 enum scld_error result
= safe_create_leading_directories(buf
);
172 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
175 for (i
= 0; i
< 20; i
++) {
176 static char hex
[] = "0123456789abcdef";
177 unsigned int val
= sha1
[i
];
178 char *pos
= pathbuf
+ i
*2 + (i
> 0);
179 *pos
++ = hex
[val
>> 4];
180 *pos
= hex
[val
& 0xf];
184 const char *sha1_file_name(const unsigned char *sha1
)
186 static char buf
[PATH_MAX
];
190 objdir
= get_object_directory();
191 len
= strlen(objdir
);
193 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
194 if (len
+ 43 > PATH_MAX
)
195 die("insanely long object directory %s", objdir
);
196 memcpy(buf
, objdir
, len
);
200 fill_sha1_path(buf
+ len
+ 1, sha1
);
205 * Return the name of the pack or index file with the specified sha1
206 * in its filename. *base and *name are scratch space that must be
207 * provided by the caller. which should be "pack" or "idx".
209 static char *sha1_get_pack_name(const unsigned char *sha1
,
210 char **name
, char **base
, const char *which
)
212 static const char hex
[] = "0123456789abcdef";
217 const char *sha1_file_directory
= get_object_directory();
218 int len
= strlen(sha1_file_directory
);
219 *base
= xmalloc(len
+ 60);
220 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
221 sha1_file_directory
, which
);
222 *name
= *base
+ len
+ 11;
227 for (i
= 0; i
< 20; i
++) {
228 unsigned int val
= *sha1
++;
229 *buf
++ = hex
[val
>> 4];
230 *buf
++ = hex
[val
& 0xf];
236 char *sha1_pack_name(const unsigned char *sha1
)
238 static char *name
, *base
;
240 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
243 char *sha1_pack_index_name(const unsigned char *sha1
)
245 static char *name
, *base
;
247 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
250 struct alternate_object_database
*alt_odb_list
;
251 static struct alternate_object_database
**alt_odb_tail
;
254 * Prepare alternate object database registry.
256 * The variable alt_odb_list points at the list of struct
257 * alternate_object_database. The elements on this list come from
258 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
259 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
260 * whose contents is similar to that environment variable but can be
261 * LF separated. Its base points at a statically allocated buffer that
262 * contains "/the/directory/corresponding/to/.git/objects/...", while
263 * its name points just after the slash at the end of ".git/objects/"
264 * in the example above, and has enough space to hold 40-byte hex
265 * SHA1, an extra slash for the first level indirection, and the
268 static int link_alt_odb_entry(const char *entry
, const char *relative_base
, int depth
)
270 const char *objdir
= get_object_directory();
271 struct alternate_object_database
*ent
;
272 struct alternate_object_database
*alt
;
274 struct strbuf pathbuf
= STRBUF_INIT
;
276 if (!is_absolute_path(entry
) && relative_base
) {
277 strbuf_addstr(&pathbuf
, real_path(relative_base
));
278 strbuf_addch(&pathbuf
, '/');
280 strbuf_addstr(&pathbuf
, entry
);
282 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
284 pfxlen
= strlen(pathbuf
.buf
);
287 * The trailing slash after the directory name is given by
288 * this function at the end. Remove duplicates.
290 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
293 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
294 ent
= xmalloc(sizeof(*ent
) + entlen
);
295 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
296 strbuf_release(&pathbuf
);
298 ent
->name
= ent
->base
+ pfxlen
+ 1;
299 ent
->base
[pfxlen
+ 3] = '/';
300 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
302 /* Detect cases where alternate disappeared */
303 if (!is_directory(ent
->base
)) {
304 error("object directory %s does not exist; "
305 "check .git/objects/info/alternates.",
311 /* Prevent the common mistake of listing the same
312 * thing twice, or object directory itself.
314 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
315 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
320 if (!strcmp(ent
->base
, objdir
)) {
325 /* add the alternate entry */
327 alt_odb_tail
= &(ent
->next
);
330 /* recursively add alternates */
331 read_info_alternates(ent
->base
, depth
+ 1);
333 ent
->base
[pfxlen
] = '/';
338 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
339 const char *relative_base
, int depth
)
341 struct string_list entries
= STRING_LIST_INIT_NODUP
;
346 error("%s: ignoring alternate object stores, nesting too deep.",
351 alt_copy
= xmemdupz(alt
, len
);
352 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
353 for (i
= 0; i
< entries
.nr
; i
++) {
354 const char *entry
= entries
.items
[i
].string
;
355 if (entry
[0] == '\0' || entry
[0] == '#')
357 if (!is_absolute_path(entry
) && depth
) {
358 error("%s: ignoring relative alternate object store %s",
359 relative_base
, entry
);
361 link_alt_odb_entry(entry
, relative_base
, depth
);
364 string_list_clear(&entries
, 0);
368 void read_info_alternates(const char * relative_base
, int depth
)
373 const char alt_file_name
[] = "info/alternates";
374 /* Given that relative_base is no longer than PATH_MAX,
375 ensure that "path" has enough space to append "/", the
376 file name, "info/alternates", and a trailing NUL. */
377 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
380 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
381 fd
= git_open_noatime(path
);
384 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
388 mapsz
= xsize_t(st
.st_size
);
389 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
392 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
397 void add_to_alternates_file(const char *reference
)
399 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
400 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
401 char *alt
= mkpath("%s\n", reference
);
402 write_or_die(fd
, alt
, strlen(alt
));
403 if (commit_lock_file(lock
))
404 die("could not close alternates file");
406 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
409 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
411 struct alternate_object_database
*ent
;
414 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
419 void prepare_alt_odb(void)
426 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
429 alt_odb_tail
= &alt_odb_list
;
430 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
432 read_info_alternates(get_object_directory(), 0);
435 static int has_loose_object_local(const unsigned char *sha1
)
437 return !access(sha1_file_name(sha1
), F_OK
);
440 int has_loose_object_nonlocal(const unsigned char *sha1
)
442 struct alternate_object_database
*alt
;
444 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
445 fill_sha1_path(alt
->name
, sha1
);
446 if (!access(alt
->base
, F_OK
))
452 static int has_loose_object(const unsigned char *sha1
)
454 return has_loose_object_local(sha1
) ||
455 has_loose_object_nonlocal(sha1
);
458 static unsigned int pack_used_ctr
;
459 static unsigned int pack_mmap_calls
;
460 static unsigned int peak_pack_open_windows
;
461 static unsigned int pack_open_windows
;
462 static unsigned int pack_open_fds
;
463 static unsigned int pack_max_fds
;
464 static size_t peak_pack_mapped
;
465 static size_t pack_mapped
;
466 struct packed_git
*packed_git
;
468 void pack_report(void)
471 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
472 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
473 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
474 sz_fmt(getpagesize()),
475 sz_fmt(packed_git_window_size
),
476 sz_fmt(packed_git_limit
));
478 "pack_report: pack_used_ctr = %10u\n"
479 "pack_report: pack_mmap_calls = %10u\n"
480 "pack_report: pack_open_windows = %10u / %10u\n"
481 "pack_report: pack_mapped = "
482 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
485 pack_open_windows
, peak_pack_open_windows
,
486 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
490 * Open and mmap the index file at path, perform a couple of
491 * consistency checks, then record its information to p. Return 0 on
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 or 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)
826 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
833 long open_max
= sysconf(_SC_OPEN_MAX
);
837 * Otherwise, we got -1 for one of the two
840 * (1) sysconf() did not understand _SC_OPEN_MAX
841 * and signaled an error with -1; or
842 * (2) sysconf() said there is no limit.
844 * We _could_ clear errno before calling sysconf() to
845 * tell these two cases apart and return a huge number
846 * in the latter case to let the caller cap it to a
847 * value that is not so selfish, but letting the
848 * fallback OPEN_MAX codepath take care of these cases
857 return 1; /* see the caller ;-) */
862 * Do not call this directly as this leaks p->pack_fd on error return;
863 * call open_packed_git() instead.
865 static int open_packed_git_1(struct packed_git
*p
)
868 struct pack_header hdr
;
869 unsigned char sha1
[20];
870 unsigned char *idx_sha1
;
873 if (!p
->index_data
&& open_pack_index(p
))
874 return error("packfile %s index unavailable", p
->pack_name
);
877 unsigned int max_fds
= get_max_fd_limit();
879 /* Save 3 for stdin/stdout/stderr, 22 for work */
881 pack_max_fds
= max_fds
- 25;
886 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
889 p
->pack_fd
= git_open_noatime(p
->pack_name
);
890 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
894 /* If we created the struct before we had the pack we lack size. */
896 if (!S_ISREG(st
.st_mode
))
897 return error("packfile %s not a regular file", p
->pack_name
);
898 p
->pack_size
= st
.st_size
;
899 } else if (p
->pack_size
!= st
.st_size
)
900 return error("packfile %s size changed", p
->pack_name
);
902 /* We leave these file descriptors open with sliding mmap;
903 * there is no point keeping them open across exec(), though.
905 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
907 return error("cannot determine file descriptor flags");
908 fd_flag
|= FD_CLOEXEC
;
909 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
910 return error("cannot set FD_CLOEXEC");
912 /* Verify we recognize this pack file format. */
913 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
914 return error("file %s is far too short to be a packfile", p
->pack_name
);
915 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
916 return error("file %s is not a GIT packfile", p
->pack_name
);
917 if (!pack_version_ok(hdr
.hdr_version
))
918 return error("packfile %s is version %"PRIu32
" and not"
919 " supported (try upgrading GIT to a newer version)",
920 p
->pack_name
, ntohl(hdr
.hdr_version
));
922 /* Verify the pack matches its index. */
923 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
924 return error("packfile %s claims to have %"PRIu32
" objects"
925 " while index indicates %"PRIu32
" objects",
926 p
->pack_name
, ntohl(hdr
.hdr_entries
),
928 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
929 return error("end of packfile %s is unavailable", p
->pack_name
);
930 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
931 return error("packfile %s signature is unavailable", p
->pack_name
);
932 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
933 if (hashcmp(sha1
, idx_sha1
))
934 return error("packfile %s does not match index", p
->pack_name
);
938 static int open_packed_git(struct packed_git
*p
)
940 if (!open_packed_git_1(p
))
942 if (p
->pack_fd
!= -1) {
950 static int in_window(struct pack_window
*win
, off_t offset
)
952 /* We must promise at least 20 bytes (one hash) after the
953 * offset is available from this window, otherwise the offset
954 * is not actually in this window and a different window (which
955 * has that one hash excess) must be used. This is to support
956 * the object header and delta base parsing routines below.
958 off_t win_off
= win
->offset
;
959 return win_off
<= offset
960 && (offset
+ 20) <= (win_off
+ win
->len
);
963 unsigned char *use_pack(struct packed_git
*p
,
964 struct pack_window
**w_cursor
,
968 struct pack_window
*win
= *w_cursor
;
970 /* Since packfiles end in a hash of their content and it's
971 * pointless to ask for an offset into the middle of that
972 * hash, and the in_window function above wouldn't match
973 * don't allow an offset too close to the end of the file.
975 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
976 die("packfile %s cannot be accessed", p
->pack_name
);
977 if (offset
> (p
->pack_size
- 20))
978 die("offset beyond end of packfile (truncated pack?)");
980 if (!win
|| !in_window(win
, offset
)) {
983 for (win
= p
->windows
; win
; win
= win
->next
) {
984 if (in_window(win
, offset
))
988 size_t window_align
= packed_git_window_size
/ 2;
991 if (p
->pack_fd
== -1 && open_packed_git(p
))
992 die("packfile %s cannot be accessed", p
->pack_name
);
994 win
= xcalloc(1, sizeof(*win
));
995 win
->offset
= (offset
/ window_align
) * window_align
;
996 len
= p
->pack_size
- win
->offset
;
997 if (len
> packed_git_window_size
)
998 len
= packed_git_window_size
;
999 win
->len
= (size_t)len
;
1000 pack_mapped
+= win
->len
;
1001 while (packed_git_limit
< pack_mapped
1002 && unuse_one_window(p
))
1004 win
->base
= xmmap(NULL
, win
->len
,
1005 PROT_READ
, MAP_PRIVATE
,
1006 p
->pack_fd
, win
->offset
);
1007 if (win
->base
== MAP_FAILED
)
1008 die("packfile %s cannot be mapped: %s",
1011 if (!win
->offset
&& win
->len
== p
->pack_size
1012 && !p
->do_not_close
) {
1018 pack_open_windows
++;
1019 if (pack_mapped
> peak_pack_mapped
)
1020 peak_pack_mapped
= pack_mapped
;
1021 if (pack_open_windows
> peak_pack_open_windows
)
1022 peak_pack_open_windows
= pack_open_windows
;
1023 win
->next
= p
->windows
;
1027 if (win
!= *w_cursor
) {
1028 win
->last_used
= pack_used_ctr
++;
1032 offset
-= win
->offset
;
1034 *left
= win
->len
- xsize_t(offset
);
1035 return win
->base
+ offset
;
1038 static struct packed_git
*alloc_packed_git(int extra
)
1040 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1041 memset(p
, 0, sizeof(*p
));
1046 static void try_to_free_pack_memory(size_t size
)
1048 release_pack_memory(size
);
1051 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1053 static int have_set_try_to_free_routine
;
1055 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1057 if (!have_set_try_to_free_routine
) {
1058 have_set_try_to_free_routine
= 1;
1059 set_try_to_free_routine(try_to_free_pack_memory
);
1063 * Make sure a corresponding .pack file exists and that
1064 * the index looks sane.
1066 path_len
-= strlen(".idx");
1071 memcpy(p
->pack_name
, path
, path_len
);
1073 strcpy(p
->pack_name
+ path_len
, ".keep");
1074 if (!access(p
->pack_name
, F_OK
))
1077 strcpy(p
->pack_name
+ path_len
, ".pack");
1078 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1083 /* ok, it looks sane as far as we can check without
1084 * actually mapping the pack file.
1086 p
->pack_size
= st
.st_size
;
1087 p
->pack_local
= local
;
1088 p
->mtime
= st
.st_mtime
;
1089 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1094 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1096 const char *path
= sha1_pack_name(sha1
);
1097 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1099 strcpy(p
->pack_name
, path
);
1100 hashcpy(p
->sha1
, sha1
);
1101 if (check_packed_git_idx(idx_path
, p
)) {
1109 void install_packed_git(struct packed_git
*pack
)
1111 if (pack
->pack_fd
!= -1)
1114 pack
->next
= packed_git
;
1118 void (*report_garbage
)(const char *desc
, const char *path
);
1120 static void report_helper(const struct string_list
*list
,
1121 int seen_bits
, int first
, int last
)
1124 switch (seen_bits
) {
1126 msg
= "no corresponding .idx or .pack";
1129 msg
= "no corresponding .idx";
1132 msg
= "no corresponding .pack";
1137 for (; first
< last
; first
++)
1138 report_garbage(msg
, list
->items
[first
].string
);
1141 static void report_pack_garbage(struct string_list
*list
)
1143 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1145 if (!report_garbage
)
1148 sort_string_list(list
);
1150 for (i
= 0; i
< list
->nr
; i
++) {
1151 const char *path
= list
->items
[i
].string
;
1152 if (baselen
!= -1 &&
1153 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1154 report_helper(list
, seen_bits
, first
, i
);
1158 if (baselen
== -1) {
1159 const char *dot
= strrchr(path
, '.');
1161 report_garbage("garbage found", path
);
1164 baselen
= dot
- path
+ 1;
1167 if (!strcmp(path
+ baselen
, "pack"))
1169 else if (!strcmp(path
+ baselen
, "idx"))
1172 report_helper(list
, seen_bits
, first
, list
->nr
);
1175 static void prepare_packed_git_one(char *objdir
, int local
)
1177 /* Ensure that this buffer is large enough so that we can
1178 append "/pack/" without clobbering the stack even if
1179 strlen(objdir) were PATH_MAX. */
1180 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
1184 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1186 sprintf(path
, "%s/pack", objdir
);
1188 dir
= opendir(path
);
1190 if (errno
!= ENOENT
)
1191 error("unable to open object pack directory: %s: %s",
1192 path
, strerror(errno
));
1196 while ((de
= readdir(dir
)) != NULL
) {
1197 int namelen
= strlen(de
->d_name
);
1198 struct packed_git
*p
;
1200 if (len
+ namelen
+ 1 > sizeof(path
)) {
1201 if (report_garbage
) {
1202 struct strbuf sb
= STRBUF_INIT
;
1203 strbuf_addf(&sb
, "%.*s/%s", len
- 1, path
, de
->d_name
);
1204 report_garbage("path too long", sb
.buf
);
1205 strbuf_release(&sb
);
1210 if (is_dot_or_dotdot(de
->d_name
))
1213 strcpy(path
+ len
, de
->d_name
);
1215 if (has_extension(de
->d_name
, ".idx")) {
1216 /* Don't reopen a pack we already have. */
1217 for (p
= packed_git
; p
; p
= p
->next
) {
1218 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1223 * See if it really is a valid .idx file with
1224 * corresponding .pack file that we can map.
1226 (p
= add_packed_git(path
, len
+ namelen
, local
)) != NULL
)
1227 install_packed_git(p
);
1230 if (!report_garbage
)
1233 if (has_extension(de
->d_name
, ".idx") ||
1234 has_extension(de
->d_name
, ".pack") ||
1235 has_extension(de
->d_name
, ".bitmap") ||
1236 has_extension(de
->d_name
, ".keep"))
1237 string_list_append(&garbage
, path
);
1239 report_garbage("garbage found", path
);
1242 report_pack_garbage(&garbage
);
1243 string_list_clear(&garbage
, 0);
1246 static int sort_pack(const void *a_
, const void *b_
)
1248 struct packed_git
*a
= *((struct packed_git
**)a_
);
1249 struct packed_git
*b
= *((struct packed_git
**)b_
);
1253 * Local packs tend to contain objects specific to our
1254 * variant of the project than remote ones. In addition,
1255 * remote ones could be on a network mounted filesystem.
1256 * Favor local ones for these reasons.
1258 st
= a
->pack_local
- b
->pack_local
;
1263 * Younger packs tend to contain more recent objects,
1264 * and more recent objects tend to get accessed more
1267 if (a
->mtime
< b
->mtime
)
1269 else if (a
->mtime
== b
->mtime
)
1274 static void rearrange_packed_git(void)
1276 struct packed_git
**ary
, *p
;
1279 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1284 /* prepare an array of packed_git for easier sorting */
1285 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1286 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1289 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1291 /* link them back again */
1292 for (i
= 0; i
< n
- 1; i
++)
1293 ary
[i
]->next
= ary
[i
+ 1];
1294 ary
[n
- 1]->next
= NULL
;
1295 packed_git
= ary
[0];
1300 static int prepare_packed_git_run_once
= 0;
1301 void prepare_packed_git(void)
1303 struct alternate_object_database
*alt
;
1305 if (prepare_packed_git_run_once
)
1307 prepare_packed_git_one(get_object_directory(), 1);
1309 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1311 prepare_packed_git_one(alt
->base
, 0);
1312 alt
->name
[-1] = '/';
1314 rearrange_packed_git();
1315 prepare_packed_git_run_once
= 1;
1318 void reprepare_packed_git(void)
1320 prepare_packed_git_run_once
= 0;
1321 prepare_packed_git();
1324 static void mark_bad_packed_object(struct packed_git
*p
,
1325 const unsigned char *sha1
)
1328 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1329 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1331 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1332 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1333 p
->num_bad_objects
++;
1336 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1338 struct packed_git
*p
;
1341 for (p
= packed_git
; p
; p
= p
->next
)
1342 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1343 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1349 * With an in-core object data in "map", rehash it to make sure the
1350 * object name actually matches "sha1" to detect object corruption.
1351 * With "map" == NULL, try reading the object named with "sha1" using
1352 * the streaming interface and rehash it to do the same.
1354 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1355 unsigned long size
, const char *type
)
1357 unsigned char real_sha1
[20];
1358 enum object_type obj_type
;
1359 struct git_istream
*st
;
1365 hash_sha1_file(map
, size
, type
, real_sha1
);
1366 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1369 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1373 /* Generate the header */
1374 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1378 git_SHA1_Update(&c
, hdr
, hdrlen
);
1380 char buf
[1024 * 16];
1381 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1389 git_SHA1_Update(&c
, buf
, readlen
);
1391 git_SHA1_Final(real_sha1
, &c
);
1393 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1396 int git_open_noatime(const char *name
)
1398 static int sha1_file_open_flag
= O_NOATIME
;
1401 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1405 /* Might the failure be due to O_NOATIME? */
1406 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1407 sha1_file_open_flag
= 0;
1415 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1417 struct alternate_object_database
*alt
;
1419 if (!lstat(sha1_file_name(sha1
), st
))
1424 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1425 fill_sha1_path(alt
->name
, sha1
);
1426 if (!lstat(alt
->base
, st
))
1433 static int open_sha1_file(const unsigned char *sha1
)
1436 struct alternate_object_database
*alt
;
1437 int most_interesting_errno
;
1439 fd
= git_open_noatime(sha1_file_name(sha1
));
1442 most_interesting_errno
= errno
;
1445 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1446 fill_sha1_path(alt
->name
, sha1
);
1447 fd
= git_open_noatime(alt
->base
);
1450 if (most_interesting_errno
== ENOENT
)
1451 most_interesting_errno
= errno
;
1453 errno
= most_interesting_errno
;
1457 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1462 fd
= open_sha1_file(sha1
);
1467 if (!fstat(fd
, &st
)) {
1468 *size
= xsize_t(st
.st_size
);
1470 /* mmap() is forbidden on empty files */
1471 error("object file %s is empty", sha1_file_name(sha1
));
1474 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1481 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1482 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1485 unsigned long size
, c
;
1486 unsigned long used
= 0;
1489 *type
= (c
>> 4) & 7;
1493 if (len
<= used
|| bitsizeof(long) <= shift
) {
1494 error("bad object header");
1499 size
+= (c
& 0x7f) << shift
;
1506 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1508 /* Get the data stream */
1509 memset(stream
, 0, sizeof(*stream
));
1510 stream
->next_in
= map
;
1511 stream
->avail_in
= mapsize
;
1512 stream
->next_out
= buffer
;
1513 stream
->avail_out
= bufsiz
;
1515 git_inflate_init(stream
);
1516 return git_inflate(stream
, 0);
1519 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1521 int bytes
= strlen(buffer
) + 1;
1522 unsigned char *buf
= xmallocz(size
);
1526 n
= stream
->total_out
- bytes
;
1529 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1531 if (bytes
<= size
) {
1533 * The above condition must be (bytes <= size), not
1534 * (bytes < size). In other words, even though we
1535 * expect no more output and set avail_out to zero,
1536 * the input zlib stream may have bytes that express
1537 * "this concludes the stream", and we *do* want to
1540 * Otherwise we would not be able to test that we
1541 * consumed all the input to reach the expected size;
1542 * we also want to check that zlib tells us that all
1543 * went well with status == Z_STREAM_END at the end.
1545 stream
->next_out
= buf
+ bytes
;
1546 stream
->avail_out
= size
- bytes
;
1547 while (status
== Z_OK
)
1548 status
= git_inflate(stream
, Z_FINISH
);
1550 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1551 git_inflate_end(stream
);
1556 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1557 else if (stream
->avail_in
)
1558 error("garbage at end of loose object '%s'",
1565 * We used to just use "sscanf()", but that's actually way
1566 * too permissive for what we want to check. So do an anal
1567 * object header parse by hand.
1569 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1576 * The type can be at most ten bytes (including the
1577 * terminating '\0' that we add), and is followed by
1586 if (i
>= sizeof(type
))
1592 * The length must follow immediately, and be in canonical
1593 * decimal format (ie "010" is not valid).
1595 size
= *hdr
++ - '0';
1600 unsigned long c
= *hdr
- '0';
1604 size
= size
* 10 + c
;
1610 * The length must be followed by a zero byte
1612 return *hdr
? -1 : type_from_string(type
);
1615 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1621 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1622 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1625 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1628 unsigned long get_size_from_delta(struct packed_git
*p
,
1629 struct pack_window
**w_curs
,
1632 const unsigned char *data
;
1633 unsigned char delta_head
[20], *in
;
1637 memset(&stream
, 0, sizeof(stream
));
1638 stream
.next_out
= delta_head
;
1639 stream
.avail_out
= sizeof(delta_head
);
1641 git_inflate_init(&stream
);
1643 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1644 stream
.next_in
= in
;
1645 st
= git_inflate(&stream
, Z_FINISH
);
1646 curpos
+= stream
.next_in
- in
;
1647 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1648 stream
.total_out
< sizeof(delta_head
));
1649 git_inflate_end(&stream
);
1650 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1651 error("delta data unpack-initial failed");
1655 /* Examine the initial part of the delta to figure out
1660 /* ignore base size */
1661 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1663 /* Read the result size */
1664 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1667 static off_t
get_delta_base(struct packed_git
*p
,
1668 struct pack_window
**w_curs
,
1670 enum object_type type
,
1671 off_t delta_obj_offset
)
1673 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1676 /* use_pack() assured us we have [base_info, base_info + 20)
1677 * as a range that we can look at without walking off the
1678 * end of the mapped window. Its actually the hash size
1679 * that is assured. An OFS_DELTA longer than the hash size
1680 * is stupid, as then a REF_DELTA would be smaller to store.
1682 if (type
== OBJ_OFS_DELTA
) {
1684 unsigned char c
= base_info
[used
++];
1685 base_offset
= c
& 127;
1688 if (!base_offset
|| MSB(base_offset
, 7))
1689 return 0; /* overflow */
1690 c
= base_info
[used
++];
1691 base_offset
= (base_offset
<< 7) + (c
& 127);
1693 base_offset
= delta_obj_offset
- base_offset
;
1694 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1695 return 0; /* out of bound */
1697 } else if (type
== OBJ_REF_DELTA
) {
1698 /* The base entry _must_ be in the same pack */
1699 base_offset
= find_pack_entry_one(base_info
, p
);
1702 die("I am totally screwed");
1707 * Like get_delta_base above, but we return the sha1 instead of the pack
1708 * offset. This means it is cheaper for REF deltas (we do not have to do
1709 * the final object lookup), but more expensive for OFS deltas (we
1710 * have to load the revidx to convert the offset back into a sha1).
1712 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1713 struct pack_window
**w_curs
,
1715 enum object_type type
,
1716 off_t delta_obj_offset
)
1718 if (type
== OBJ_REF_DELTA
) {
1719 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1721 } else if (type
== OBJ_OFS_DELTA
) {
1722 struct revindex_entry
*revidx
;
1723 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1724 type
, delta_obj_offset
);
1729 revidx
= find_pack_revindex(p
, base_offset
);
1733 return nth_packed_object_sha1(p
, revidx
->nr
);
1738 int unpack_object_header(struct packed_git
*p
,
1739 struct pack_window
**w_curs
,
1741 unsigned long *sizep
)
1743 unsigned char *base
;
1746 enum object_type type
;
1748 /* use_pack() assures us we have [base, base + 20) available
1749 * as a range that we can look at. (Its actually the hash
1750 * size that is assured.) With our object header encoding
1751 * the maximum deflated object size is 2^137, which is just
1752 * insane, so we know won't exceed what we have been given.
1754 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1755 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1764 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1767 struct revindex_entry
*revidx
;
1768 const unsigned char *sha1
;
1769 revidx
= find_pack_revindex(p
, obj_offset
);
1772 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1773 mark_bad_packed_object(p
, sha1
);
1774 type
= sha1_object_info(sha1
, NULL
);
1775 if (type
<= OBJ_NONE
)
1780 #define POI_STACK_PREALLOC 64
1782 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1784 enum object_type type
,
1785 struct pack_window
**w_curs
,
1788 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1789 off_t
*poi_stack
= small_poi_stack
;
1790 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1792 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1795 /* Push the object we're going to leave behind */
1796 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1797 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1798 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1799 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1801 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1803 poi_stack
[poi_stack_nr
++] = obj_offset
;
1804 /* If parsing the base offset fails, just unwind */
1805 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1808 curpos
= obj_offset
= base_offset
;
1809 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1810 if (type
<= OBJ_NONE
) {
1811 /* If getting the base itself fails, we first
1812 * retry the base, otherwise unwind */
1813 type
= retry_bad_packed_offset(p
, base_offset
);
1814 if (type
> OBJ_NONE
)
1828 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1829 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1834 if (poi_stack
!= small_poi_stack
)
1839 while (poi_stack_nr
) {
1840 obj_offset
= poi_stack
[--poi_stack_nr
];
1841 type
= retry_bad_packed_offset(p
, obj_offset
);
1842 if (type
> OBJ_NONE
)
1849 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1850 struct object_info
*oi
)
1852 struct pack_window
*w_curs
= NULL
;
1854 off_t curpos
= obj_offset
;
1855 enum object_type type
;
1858 * We always get the representation type, but only convert it to
1859 * a "real" type later if the caller is interested.
1861 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1864 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1865 off_t tmp_pos
= curpos
;
1866 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1872 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1873 if (*oi
->sizep
== 0) {
1882 if (oi
->disk_sizep
) {
1883 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1884 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1888 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1889 if (*oi
->typep
< 0) {
1895 if (oi
->delta_base_sha1
) {
1896 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1897 const unsigned char *base
;
1899 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1906 hashcpy(oi
->delta_base_sha1
, base
);
1908 hashclr(oi
->delta_base_sha1
);
1912 unuse_pack(&w_curs
);
1916 static void *unpack_compressed_entry(struct packed_git
*p
,
1917 struct pack_window
**w_curs
,
1923 unsigned char *buffer
, *in
;
1925 buffer
= xmallocz(size
);
1926 memset(&stream
, 0, sizeof(stream
));
1927 stream
.next_out
= buffer
;
1928 stream
.avail_out
= size
+ 1;
1930 git_inflate_init(&stream
);
1932 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1933 stream
.next_in
= in
;
1934 st
= git_inflate(&stream
, Z_FINISH
);
1935 if (!stream
.avail_out
)
1936 break; /* the payload is larger than it should be */
1937 curpos
+= stream
.next_in
- in
;
1938 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1939 git_inflate_end(&stream
);
1940 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1948 #define MAX_DELTA_CACHE (256)
1950 static size_t delta_base_cached
;
1952 static struct delta_base_cache_lru_list
{
1953 struct delta_base_cache_lru_list
*prev
;
1954 struct delta_base_cache_lru_list
*next
;
1955 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1957 static struct delta_base_cache_entry
{
1958 struct delta_base_cache_lru_list lru
;
1960 struct packed_git
*p
;
1963 enum object_type type
;
1964 } delta_base_cache
[MAX_DELTA_CACHE
];
1966 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1970 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1971 hash
+= (hash
>> 8) + (hash
>> 16);
1972 return hash
% MAX_DELTA_CACHE
;
1975 static struct delta_base_cache_entry
*
1976 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1978 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1979 return delta_base_cache
+ hash
;
1982 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
1983 struct packed_git
*p
, off_t base_offset
)
1985 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
1988 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1990 struct delta_base_cache_entry
*ent
;
1991 ent
= get_delta_base_cache_entry(p
, base_offset
);
1992 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
1995 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1998 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1999 ent
->lru
.prev
->next
= ent
->lru
.next
;
2000 delta_base_cached
-= ent
->size
;
2003 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2004 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2006 struct delta_base_cache_entry
*ent
;
2009 ent
= get_delta_base_cache_entry(p
, base_offset
);
2011 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2012 return unpack_entry(p
, base_offset
, type
, base_size
);
2017 clear_delta_base_cache_entry(ent
);
2019 ret
= xmemdupz(ent
->data
, ent
->size
);
2021 *base_size
= ent
->size
;
2025 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2030 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2031 ent
->lru
.prev
->next
= ent
->lru
.next
;
2032 delta_base_cached
-= ent
->size
;
2036 void clear_delta_base_cache(void)
2039 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2040 release_delta_base_cache(&delta_base_cache
[p
]);
2043 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2044 void *base
, unsigned long base_size
, enum object_type type
)
2046 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2047 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2048 struct delta_base_cache_lru_list
*lru
;
2050 release_delta_base_cache(ent
);
2051 delta_base_cached
+= base_size
;
2053 for (lru
= delta_base_cache_lru
.next
;
2054 delta_base_cached
> delta_base_cache_limit
2055 && lru
!= &delta_base_cache_lru
;
2057 struct delta_base_cache_entry
*f
= (void *)lru
;
2058 if (f
->type
== OBJ_BLOB
)
2059 release_delta_base_cache(f
);
2061 for (lru
= delta_base_cache_lru
.next
;
2062 delta_base_cached
> delta_base_cache_limit
2063 && lru
!= &delta_base_cache_lru
;
2065 struct delta_base_cache_entry
*f
= (void *)lru
;
2066 release_delta_base_cache(f
);
2070 ent
->base_offset
= base_offset
;
2073 ent
->size
= base_size
;
2074 ent
->lru
.next
= &delta_base_cache_lru
;
2075 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2076 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2077 delta_base_cache_lru
.prev
= &ent
->lru
;
2080 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2081 unsigned long *size
);
2083 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2085 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2086 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2087 p
->pack_name
, (uintmax_t)obj_offset
);
2090 int do_check_packed_object_crc
;
2092 #define UNPACK_ENTRY_STACK_PREALLOC 64
2093 struct unpack_entry_stack_ent
{
2099 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2100 enum object_type
*final_type
, unsigned long *final_size
)
2102 struct pack_window
*w_curs
= NULL
;
2103 off_t curpos
= obj_offset
;
2106 enum object_type type
;
2107 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2108 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2109 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2110 int base_from_cache
= 0;
2112 write_pack_access_log(p
, obj_offset
);
2114 /* PHASE 1: drill down to the innermost base object */
2118 struct delta_base_cache_entry
*ent
;
2120 ent
= get_delta_base_cache_entry(p
, curpos
);
2121 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2125 clear_delta_base_cache_entry(ent
);
2126 base_from_cache
= 1;
2130 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2131 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2132 unsigned long len
= revidx
[1].offset
- obj_offset
;
2133 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2134 const unsigned char *sha1
=
2135 nth_packed_object_sha1(p
, revidx
->nr
);
2136 error("bad packed object CRC for %s",
2138 mark_bad_packed_object(p
, sha1
);
2139 unuse_pack(&w_curs
);
2144 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2145 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2148 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2150 error("failed to validate delta base reference "
2151 "at offset %"PRIuMAX
" from %s",
2152 (uintmax_t)curpos
, p
->pack_name
);
2153 /* bail to phase 2, in hopes of recovery */
2158 /* push object, proceed to base */
2159 if (delta_stack_nr
>= delta_stack_alloc
2160 && delta_stack
== small_delta_stack
) {
2161 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2162 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2163 memcpy(delta_stack
, small_delta_stack
,
2164 sizeof(*delta_stack
)*delta_stack_nr
);
2166 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2168 i
= delta_stack_nr
++;
2169 delta_stack
[i
].obj_offset
= obj_offset
;
2170 delta_stack
[i
].curpos
= curpos
;
2171 delta_stack
[i
].size
= size
;
2173 curpos
= obj_offset
= base_offset
;
2176 /* PHASE 2: handle the base */
2181 die("BUG in unpack_entry: left loop at a valid delta");
2187 if (!base_from_cache
)
2188 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2192 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2193 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2196 /* PHASE 3: apply deltas in order */
2199 * 'data' holds the base data, or NULL if there was corruption
2201 while (delta_stack_nr
) {
2204 unsigned long delta_size
, base_size
= size
;
2210 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2214 * We're probably in deep shit, but let's try to fetch
2215 * the required base anyway from another pack or loose.
2216 * This is costly but should happen only in the presence
2217 * of a corrupted pack, and is better than failing outright.
2219 struct revindex_entry
*revidx
;
2220 const unsigned char *base_sha1
;
2221 revidx
= find_pack_revindex(p
, obj_offset
);
2223 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2224 error("failed to read delta base object %s"
2225 " at offset %"PRIuMAX
" from %s",
2226 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2228 mark_bad_packed_object(p
, base_sha1
);
2229 base
= read_object(base_sha1
, &type
, &base_size
);
2233 i
= --delta_stack_nr
;
2234 obj_offset
= delta_stack
[i
].obj_offset
;
2235 curpos
= delta_stack
[i
].curpos
;
2236 delta_size
= delta_stack
[i
].size
;
2241 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2244 error("failed to unpack compressed delta "
2245 "at offset %"PRIuMAX
" from %s",
2246 (uintmax_t)curpos
, p
->pack_name
);
2251 data
= patch_delta(base
, base_size
,
2252 delta_data
, delta_size
,
2256 * We could not apply the delta; warn the user, but keep going.
2257 * Our failure will be noticed either in the next iteration of
2258 * the loop, or if this is the final delta, in the caller when
2259 * we return NULL. Those code paths will take care of making
2260 * a more explicit warning and retrying with another copy of
2264 error("failed to apply delta");
2272 unuse_pack(&w_curs
);
2274 if (delta_stack
!= small_delta_stack
)
2280 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2283 const unsigned char *index
= p
->index_data
;
2285 if (open_pack_index(p
))
2287 index
= p
->index_data
;
2289 if (n
>= p
->num_objects
)
2292 if (p
->index_version
== 1) {
2293 return index
+ 24 * n
+ 4;
2296 return index
+ 20 * n
;
2300 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2302 const unsigned char *index
= p
->index_data
;
2304 if (p
->index_version
== 1) {
2305 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2308 index
+= 8 + p
->num_objects
* (20 + 4);
2309 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2310 if (!(off
& 0x80000000))
2312 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2313 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2314 ntohl(*((uint32_t *)(index
+ 4)));
2318 off_t
find_pack_entry_one(const unsigned char *sha1
,
2319 struct packed_git
*p
)
2321 const uint32_t *level1_ofs
= p
->index_data
;
2322 const unsigned char *index
= p
->index_data
;
2323 unsigned hi
, lo
, stride
;
2324 static int use_lookup
= -1;
2325 static int debug_lookup
= -1;
2327 if (debug_lookup
< 0)
2328 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2331 if (open_pack_index(p
))
2333 level1_ofs
= p
->index_data
;
2334 index
= p
->index_data
;
2336 if (p
->index_version
> 1) {
2341 hi
= ntohl(level1_ofs
[*sha1
]);
2342 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2343 if (p
->index_version
> 1) {
2351 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2352 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2355 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2357 int pos
= sha1_entry_pos(index
, stride
, 0,
2358 lo
, hi
, p
->num_objects
, sha1
);
2361 return nth_packed_object_offset(p
, pos
);
2365 unsigned mi
= (lo
+ hi
) / 2;
2366 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2369 printf("lo %u hi %u rg %u mi %u\n",
2370 lo
, hi
, hi
- lo
, mi
);
2372 return nth_packed_object_offset(p
, mi
);
2381 int is_pack_valid(struct packed_git
*p
)
2383 /* An already open pack is known to be valid. */
2384 if (p
->pack_fd
!= -1)
2387 /* If the pack has one window completely covering the
2388 * file size, the pack is known to be valid even if
2389 * the descriptor is not currently open.
2392 struct pack_window
*w
= p
->windows
;
2394 if (!w
->offset
&& w
->len
== p
->pack_size
)
2398 /* Force the pack to open to prove its valid. */
2399 return !open_packed_git(p
);
2402 static int fill_pack_entry(const unsigned char *sha1
,
2403 struct pack_entry
*e
,
2404 struct packed_git
*p
)
2408 if (p
->num_bad_objects
) {
2410 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2411 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2415 offset
= find_pack_entry_one(sha1
, p
);
2420 * We are about to tell the caller where they can locate the
2421 * requested object. We better make sure the packfile is
2422 * still here and can be accessed before supplying that
2423 * answer, as it may have been deleted since the index was
2426 if (!is_pack_valid(p
)) {
2427 warning("packfile %s cannot be accessed", p
->pack_name
);
2432 hashcpy(e
->sha1
, sha1
);
2437 * Iff a pack file contains the object named by sha1, return true and
2438 * store its location to e.
2440 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2442 struct packed_git
*p
;
2444 prepare_packed_git();
2448 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2451 for (p
= packed_git
; p
; p
= p
->next
) {
2452 if (p
== last_found_pack
)
2453 continue; /* we already checked this one */
2455 if (fill_pack_entry(sha1
, e
, p
)) {
2456 last_found_pack
= p
;
2463 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2464 struct packed_git
*packs
)
2466 struct packed_git
*p
;
2468 for (p
= packs
; p
; p
= p
->next
) {
2469 if (find_pack_entry_one(sha1
, p
))
2476 static int sha1_loose_object_info(const unsigned char *sha1
,
2477 struct object_info
*oi
)
2480 unsigned long mapsize
, size
;
2485 if (oi
->delta_base_sha1
)
2486 hashclr(oi
->delta_base_sha1
);
2489 * If we don't care about type or size, then we don't
2490 * need to look inside the object at all. Note that we
2491 * do not optimize out the stat call, even if the
2492 * caller doesn't care about the disk-size, since our
2493 * return value implicitly indicates whether the
2494 * object even exists.
2496 if (!oi
->typep
&& !oi
->sizep
) {
2498 if (stat_sha1_file(sha1
, &st
) < 0)
2501 *oi
->disk_sizep
= st
.st_size
;
2505 map
= map_sha1_file(sha1
, &mapsize
);
2509 *oi
->disk_sizep
= mapsize
;
2510 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2511 status
= error("unable to unpack %s header",
2513 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2514 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2517 git_inflate_end(&stream
);
2518 munmap(map
, mapsize
);
2520 *oi
->typep
= status
;
2524 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2526 struct cached_object
*co
;
2527 struct pack_entry e
;
2529 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2531 co
= find_cached_object(real
);
2534 *(oi
->typep
) = co
->type
;
2536 *(oi
->sizep
) = co
->size
;
2538 *(oi
->disk_sizep
) = 0;
2539 if (oi
->delta_base_sha1
)
2540 hashclr(oi
->delta_base_sha1
);
2541 oi
->whence
= OI_CACHED
;
2545 if (!find_pack_entry(real
, &e
)) {
2546 /* Most likely it's a loose object. */
2547 if (!sha1_loose_object_info(real
, oi
)) {
2548 oi
->whence
= OI_LOOSE
;
2552 /* Not a loose object; someone else may have just packed it. */
2553 reprepare_packed_git();
2554 if (!find_pack_entry(real
, &e
))
2558 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2560 mark_bad_packed_object(e
.p
, real
);
2561 return sha1_object_info_extended(real
, oi
, 0);
2562 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2563 oi
->whence
= OI_DBCACHED
;
2565 oi
->whence
= OI_PACKED
;
2566 oi
->u
.packed
.offset
= e
.offset
;
2567 oi
->u
.packed
.pack
= e
.p
;
2568 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2569 rtype
== OBJ_OFS_DELTA
);
2575 /* returns enum object_type or negative */
2576 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2578 enum object_type type
;
2579 struct object_info oi
= {NULL
};
2583 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2588 static void *read_packed_sha1(const unsigned char *sha1
,
2589 enum object_type
*type
, unsigned long *size
)
2591 struct pack_entry e
;
2594 if (!find_pack_entry(sha1
, &e
))
2596 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2599 * We're probably in deep shit, but let's try to fetch
2600 * the required object anyway from another pack or loose.
2601 * This should happen only in the presence of a corrupted
2602 * pack, and is better than failing outright.
2604 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2605 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2606 mark_bad_packed_object(e
.p
, sha1
);
2607 data
= read_object(sha1
, type
, size
);
2612 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2613 unsigned char *sha1
)
2615 struct cached_object
*co
;
2617 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2618 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2620 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2621 co
= &cached_objects
[cached_object_nr
++];
2624 co
->buf
= xmalloc(len
);
2625 memcpy(co
->buf
, buf
, len
);
2626 hashcpy(co
->sha1
, sha1
);
2630 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2631 unsigned long *size
)
2633 unsigned long mapsize
;
2635 struct cached_object
*co
;
2637 co
= find_cached_object(sha1
);
2641 return xmemdupz(co
->buf
, co
->size
);
2644 buf
= read_packed_sha1(sha1
, type
, size
);
2647 map
= map_sha1_file(sha1
, &mapsize
);
2649 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2650 munmap(map
, mapsize
);
2653 reprepare_packed_git();
2654 return read_packed_sha1(sha1
, type
, size
);
2658 * This function dies on corrupt objects; the callers who want to
2659 * deal with them should arrange to call read_object() and give error
2660 * messages themselves.
2662 void *read_sha1_file_extended(const unsigned char *sha1
,
2663 enum object_type
*type
,
2664 unsigned long *size
,
2668 const struct packed_git
*p
;
2669 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2672 data
= read_object(repl
, type
, size
);
2676 if (errno
&& errno
!= ENOENT
)
2677 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2679 /* die if we replaced an object with one that does not exist */
2681 die("replacement %s not found for %s",
2682 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2684 if (has_loose_object(repl
)) {
2685 const char *path
= sha1_file_name(sha1
);
2687 die("loose object %s (stored in %s) is corrupt",
2688 sha1_to_hex(repl
), path
);
2691 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2692 die("packed object %s (stored in %s) is corrupt",
2693 sha1_to_hex(repl
), p
->pack_name
);
2698 void *read_object_with_reference(const unsigned char *sha1
,
2699 const char *required_type_name
,
2700 unsigned long *size
,
2701 unsigned char *actual_sha1_return
)
2703 enum object_type type
, required_type
;
2705 unsigned long isize
;
2706 unsigned char actual_sha1
[20];
2708 required_type
= type_from_string(required_type_name
);
2709 hashcpy(actual_sha1
, sha1
);
2711 int ref_length
= -1;
2712 const char *ref_type
= NULL
;
2714 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2717 if (type
== required_type
) {
2719 if (actual_sha1_return
)
2720 hashcpy(actual_sha1_return
, actual_sha1
);
2723 /* Handle references */
2724 else if (type
== OBJ_COMMIT
)
2726 else if (type
== OBJ_TAG
)
2727 ref_type
= "object ";
2732 ref_length
= strlen(ref_type
);
2734 if (ref_length
+ 40 > isize
||
2735 memcmp(buffer
, ref_type
, ref_length
) ||
2736 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2741 /* Now we have the ID of the referred-to object in
2742 * actual_sha1. Check again. */
2746 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2747 const char *type
, unsigned char *sha1
,
2748 char *hdr
, int *hdrlen
)
2752 /* Generate the header */
2753 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2757 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2758 git_SHA1_Update(&c
, buf
, len
);
2759 git_SHA1_Final(sha1
, &c
);
2763 * Move the just written object into its final resting place.
2764 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2765 * "moving" is only a part of what it does, when no patch between
2766 * master to pu changes the call sites of this function.
2768 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2772 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2774 else if (link(tmpfile
, filename
))
2778 * Coda hack - coda doesn't like cross-directory links,
2779 * so we fall back to a rename, which will mean that it
2780 * won't be able to check collisions, but that's not a
2783 * The same holds for FAT formatted media.
2785 * When this succeeds, we just return. We have nothing
2788 if (ret
&& ret
!= EEXIST
) {
2790 if (!rename(tmpfile
, filename
))
2794 unlink_or_warn(tmpfile
);
2796 if (ret
!= EEXIST
) {
2797 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2799 /* FIXME!!! Collision check here ? */
2803 if (adjust_shared_perm(filename
))
2804 return error("unable to set permission to '%s'", filename
);
2808 static int write_buffer(int fd
, const void *buf
, size_t len
)
2810 if (write_in_full(fd
, buf
, len
) < 0)
2811 return error("file write error (%s)", strerror(errno
));
2815 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2816 unsigned char *sha1
)
2820 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2824 /* Finalize a file on disk, and close it. */
2825 static void close_sha1_file(int fd
)
2827 if (fsync_object_files
)
2828 fsync_or_die(fd
, "sha1 file");
2830 die_errno("error when closing sha1 file");
2833 /* Size of directory component, including the ending '/' */
2834 static inline int directory_size(const char *filename
)
2836 const char *s
= strrchr(filename
, '/');
2839 return s
- filename
+ 1;
2843 * This creates a temporary file in the same directory as the final
2846 * We want to avoid cross-directory filename renames, because those
2847 * can have problems on various filesystems (FAT, NFS, Coda).
2849 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2851 int fd
, dirlen
= directory_size(filename
);
2853 if (dirlen
+ 20 > bufsiz
) {
2854 errno
= ENAMETOOLONG
;
2857 memcpy(buffer
, filename
, dirlen
);
2858 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2859 fd
= git_mkstemp_mode(buffer
, 0444);
2860 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2861 /* Make sure the directory exists */
2862 memcpy(buffer
, filename
, dirlen
);
2863 buffer
[dirlen
-1] = 0;
2864 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
2866 if (adjust_shared_perm(buffer
))
2870 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2871 fd
= git_mkstemp_mode(buffer
, 0444);
2876 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2877 const void *buf
, unsigned long len
, time_t mtime
)
2880 unsigned char compressed
[4096];
2883 unsigned char parano_sha1
[20];
2884 static char tmp_file
[PATH_MAX
];
2885 const char *filename
= sha1_file_name(sha1
);
2887 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2889 if (errno
== EACCES
)
2890 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2892 return error("unable to create temporary file: %s", strerror(errno
));
2896 memset(&stream
, 0, sizeof(stream
));
2897 git_deflate_init(&stream
, zlib_compression_level
);
2898 stream
.next_out
= compressed
;
2899 stream
.avail_out
= sizeof(compressed
);
2902 /* First header.. */
2903 stream
.next_in
= (unsigned char *)hdr
;
2904 stream
.avail_in
= hdrlen
;
2905 while (git_deflate(&stream
, 0) == Z_OK
)
2907 git_SHA1_Update(&c
, hdr
, hdrlen
);
2909 /* Then the data itself.. */
2910 stream
.next_in
= (void *)buf
;
2911 stream
.avail_in
= len
;
2913 unsigned char *in0
= stream
.next_in
;
2914 ret
= git_deflate(&stream
, Z_FINISH
);
2915 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2916 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2917 die("unable to write sha1 file");
2918 stream
.next_out
= compressed
;
2919 stream
.avail_out
= sizeof(compressed
);
2920 } while (ret
== Z_OK
);
2922 if (ret
!= Z_STREAM_END
)
2923 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2924 ret
= git_deflate_end_gently(&stream
);
2926 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2927 git_SHA1_Final(parano_sha1
, &c
);
2928 if (hashcmp(sha1
, parano_sha1
) != 0)
2929 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2931 close_sha1_file(fd
);
2936 utb
.modtime
= mtime
;
2937 if (utime(tmp_file
, &utb
) < 0)
2938 warning("failed utime() on %s: %s",
2939 tmp_file
, strerror(errno
));
2942 return move_temp_to_file(tmp_file
, filename
);
2945 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2947 unsigned char sha1
[20];
2951 /* Normally if we have it in the pack then we do not bother writing
2952 * it out into .git/objects/??/?{38} file.
2954 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2956 hashcpy(returnsha1
, sha1
);
2957 if (has_sha1_file(sha1
))
2959 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2962 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2966 enum object_type type
;
2971 if (has_loose_object(sha1
))
2973 buf
= read_packed_sha1(sha1
, &type
, &len
);
2975 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2976 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2977 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2983 int has_pack_index(const unsigned char *sha1
)
2986 if (stat(sha1_pack_index_name(sha1
), &st
))
2991 int has_sha1_pack(const unsigned char *sha1
)
2993 struct pack_entry e
;
2994 return find_pack_entry(sha1
, &e
);
2997 int has_sha1_file(const unsigned char *sha1
)
2999 struct pack_entry e
;
3001 if (find_pack_entry(sha1
, &e
))
3003 if (has_loose_object(sha1
))
3005 reprepare_packed_git();
3006 return find_pack_entry(sha1
, &e
);
3009 static void check_tree(const void *buf
, size_t size
)
3011 struct tree_desc desc
;
3012 struct name_entry entry
;
3014 init_tree_desc(&desc
, buf
, size
);
3015 while (tree_entry(&desc
, &entry
))
3017 * tree_entry() will die() on malformed entries */
3021 static void check_commit(const void *buf
, size_t size
)
3024 memset(&c
, 0, sizeof(c
));
3025 if (parse_commit_buffer(&c
, buf
, size
))
3026 die("corrupt commit");
3029 static void check_tag(const void *buf
, size_t size
)
3032 memset(&t
, 0, sizeof(t
));
3033 if (parse_tag_buffer(&t
, buf
, size
))
3037 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3038 enum object_type type
,
3039 const char *path
, unsigned flags
)
3041 int ret
, re_allocated
= 0;
3042 int write_object
= flags
& HASH_WRITE_OBJECT
;
3048 * Convert blobs to git internal format
3050 if ((type
== OBJ_BLOB
) && path
) {
3051 struct strbuf nbuf
= STRBUF_INIT
;
3052 if (convert_to_git(path
, buf
, size
, &nbuf
,
3053 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3054 buf
= strbuf_detach(&nbuf
, &size
);
3058 if (flags
& HASH_FORMAT_CHECK
) {
3059 if (type
== OBJ_TREE
)
3060 check_tree(buf
, size
);
3061 if (type
== OBJ_COMMIT
)
3062 check_commit(buf
, size
);
3063 if (type
== OBJ_TAG
)
3064 check_tag(buf
, size
);
3068 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3070 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3076 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3077 const char *path
, unsigned flags
)
3079 struct strbuf sbuf
= STRBUF_INIT
;
3082 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3083 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3086 strbuf_release(&sbuf
);
3090 #define SMALL_FILE_SIZE (32*1024)
3092 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3093 enum object_type type
, const char *path
,
3099 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
3100 } else if (size
<= SMALL_FILE_SIZE
) {
3101 char *buf
= xmalloc(size
);
3102 if (size
== read_in_full(fd
, buf
, size
))
3103 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3105 ret
= error("short read %s", strerror(errno
));
3108 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3109 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3116 * This creates one packfile per large blob unless bulk-checkin
3117 * machinery is "plugged".
3119 * This also bypasses the usual "convert-to-git" dance, and that is on
3120 * purpose. We could write a streaming version of the converting
3121 * functions and insert that before feeding the data to fast-import
3122 * (or equivalent in-core API described above). However, that is
3123 * somewhat complicated, as we do not know the size of the filter
3124 * result, which we need to know beforehand when writing a git object.
3125 * Since the primary motivation for trying to stream from the working
3126 * tree file and to avoid mmaping it in core is to deal with large
3127 * binary blobs, they generally do not want to get any conversion, and
3128 * callers should avoid this code path when filters are requested.
3130 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3131 enum object_type type
, const char *path
,
3134 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3137 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3138 enum object_type type
, const char *path
, unsigned flags
)
3141 size_t size
= xsize_t(st
->st_size
);
3143 if (!S_ISREG(st
->st_mode
))
3144 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3145 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3146 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
3147 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
3149 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
3154 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3157 struct strbuf sb
= STRBUF_INIT
;
3159 switch (st
->st_mode
& S_IFMT
) {
3161 fd
= open(path
, O_RDONLY
);
3163 return error("open(\"%s\"): %s", path
,
3165 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3166 return error("%s: failed to insert into database",
3170 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3171 char *errstr
= strerror(errno
);
3172 return error("readlink(\"%s\"): %s", path
,
3175 if (!(flags
& HASH_WRITE_OBJECT
))
3176 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3177 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3178 return error("%s: failed to insert into database",
3180 strbuf_release(&sb
);
3183 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3185 return error("%s: unsupported file type", path
);
3190 int read_pack_header(int fd
, struct pack_header
*header
)
3192 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3193 /* "eof before pack header was fully read" */
3194 return PH_ERROR_EOF
;
3196 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3197 /* "protocol error (pack signature mismatch detected)" */
3198 return PH_ERROR_PACK_SIGNATURE
;
3199 if (!pack_version_ok(header
->hdr_version
))
3200 /* "protocol error (pack version unsupported)" */
3201 return PH_ERROR_PROTOCOL
;
3205 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3207 enum object_type type
= sha1_object_info(sha1
, NULL
);
3209 die("%s is not a valid object", sha1_to_hex(sha1
));
3211 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),