2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
16 #include "run-command.h"
19 #include "tree-walk.h"
21 #include "pack-revindex.h"
22 #include "sha1-lookup.h"
23 #include "bulk-checkin.h"
24 #include "streaming.h"
28 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29 #define O_NOATIME 01000000
35 #define SZ_FMT PRIuMAX
36 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
38 const unsigned char null_sha1
[20];
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
46 static struct cached_object
{
47 unsigned char sha1
[20];
48 enum object_type type
;
52 static int cached_object_nr
, cached_object_alloc
;
54 static struct cached_object empty_tree
= {
55 EMPTY_TREE_SHA1_BIN_LITERAL
,
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
67 static struct packed_git
*last_found_pack
;
69 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
72 struct cached_object
*co
= cached_objects
;
74 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
75 if (!hashcmp(co
->sha1
, sha1
))
78 if (!hashcmp(sha1
, empty_tree
.sha1
))
83 int mkdir_in_gitdir(const char *path
)
85 if (mkdir(path
, 0777)) {
86 int saved_errno
= errno
;
88 struct strbuf sb
= STRBUF_INIT
;
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
99 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
100 strbuf_readlink(&sb
, path
, st
.st_size
) ||
101 !is_absolute_path(sb
.buf
) ||
102 mkdir(sb
.buf
, 0777)) {
109 return adjust_shared_perm(path
);
112 enum scld_error
safe_create_leading_directories(char *path
)
114 char *next_component
= path
+ offset_1st_component(path
);
115 enum scld_error ret
= SCLD_OK
;
117 while (ret
== SCLD_OK
&& next_component
) {
119 char *slash
= next_component
, slash_character
;
121 while (*slash
&& !is_dir_sep(*slash
))
127 next_component
= slash
+ 1;
128 while (is_dir_sep(*next_component
))
130 if (!*next_component
)
133 slash_character
= *slash
;
135 if (!stat(path
, &st
)) {
137 if (!S_ISDIR(st
.st_mode
))
139 } else if (mkdir(path
, 0777)) {
140 if (errno
== EEXIST
&&
141 !stat(path
, &st
) && S_ISDIR(st
.st_mode
))
142 ; /* somebody created it since we checked */
143 else if (errno
== ENOENT
)
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
156 } else if (adjust_shared_perm(path
)) {
159 *slash
= slash_character
;
164 enum scld_error
safe_create_leading_directories_const(const char *path
)
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf
= xstrdup(path
);
168 enum scld_error result
= safe_create_leading_directories(buf
);
173 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
176 for (i
= 0; i
< 20; i
++) {
177 static char hex
[] = "0123456789abcdef";
178 unsigned int val
= sha1
[i
];
179 char *pos
= pathbuf
+ i
*2 + (i
> 0);
180 *pos
++ = hex
[val
>> 4];
181 *pos
= hex
[val
& 0xf];
185 const char *sha1_file_name(const unsigned char *sha1
)
187 static char buf
[PATH_MAX
];
191 objdir
= get_object_directory();
192 len
= strlen(objdir
);
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len
+ 43 > PATH_MAX
)
196 die("insanely long object directory %s", objdir
);
197 memcpy(buf
, objdir
, len
);
201 fill_sha1_path(buf
+ len
+ 1, sha1
);
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
210 static char *sha1_get_pack_name(const unsigned char *sha1
,
211 char **name
, char **base
, const char *which
)
213 static const char hex
[] = "0123456789abcdef";
218 const char *sha1_file_directory
= get_object_directory();
219 int len
= strlen(sha1_file_directory
);
220 *base
= xmalloc(len
+ 60);
221 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
222 sha1_file_directory
, which
);
223 *name
= *base
+ len
+ 11;
228 for (i
= 0; i
< 20; i
++) {
229 unsigned int val
= *sha1
++;
230 *buf
++ = hex
[val
>> 4];
231 *buf
++ = hex
[val
& 0xf];
237 char *sha1_pack_name(const unsigned char *sha1
)
239 static char *name
, *base
;
241 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
244 char *sha1_pack_index_name(const unsigned char *sha1
)
246 static char *name
, *base
;
248 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
251 struct alternate_object_database
*alt_odb_list
;
252 static struct alternate_object_database
**alt_odb_tail
;
255 * Prepare alternate object database registry.
257 * The variable alt_odb_list points at the list of struct
258 * alternate_object_database. The elements on this list come from
259 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
260 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
261 * whose contents is similar to that environment variable but can be
262 * LF separated. Its base points at a statically allocated buffer that
263 * contains "/the/directory/corresponding/to/.git/objects/...", while
264 * its name points just after the slash at the end of ".git/objects/"
265 * in the example above, and has enough space to hold 40-byte hex
266 * SHA1, an extra slash for the first level indirection, and the
269 static int link_alt_odb_entry(const char *entry
, const char *relative_base
,
270 int depth
, const char *normalized_objdir
)
272 struct alternate_object_database
*ent
;
273 struct alternate_object_database
*alt
;
275 struct strbuf pathbuf
= STRBUF_INIT
;
277 if (!is_absolute_path(entry
) && relative_base
) {
278 strbuf_addstr(&pathbuf
, real_path(relative_base
));
279 strbuf_addch(&pathbuf
, '/');
281 strbuf_addstr(&pathbuf
, entry
);
283 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
285 pfxlen
= strlen(pathbuf
.buf
);
288 * The trailing slash after the directory name is given by
289 * this function at the end. Remove duplicates.
291 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
294 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
295 ent
= xmalloc(sizeof(*ent
) + entlen
);
296 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
297 strbuf_release(&pathbuf
);
299 ent
->name
= ent
->base
+ pfxlen
+ 1;
300 ent
->base
[pfxlen
+ 3] = '/';
301 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
303 /* Detect cases where alternate disappeared */
304 if (!is_directory(ent
->base
)) {
305 error("object directory %s does not exist; "
306 "check .git/objects/info/alternates.",
312 /* Prevent the common mistake of listing the same
313 * thing twice, or object directory itself.
315 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
316 if (pfxlen
== alt
->name
- alt
->base
- 1 &&
317 !memcmp(ent
->base
, alt
->base
, pfxlen
)) {
322 if (!strcmp_icase(ent
->base
, normalized_objdir
)) {
327 /* add the alternate entry */
329 alt_odb_tail
= &(ent
->next
);
332 /* recursively add alternates */
333 read_info_alternates(ent
->base
, depth
+ 1);
335 ent
->base
[pfxlen
] = '/';
340 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
341 const char *relative_base
, int depth
)
343 struct string_list entries
= STRING_LIST_INIT_NODUP
;
346 struct strbuf objdirbuf
= STRBUF_INIT
;
349 error("%s: ignoring alternate object stores, nesting too deep.",
354 strbuf_add_absolute_path(&objdirbuf
, get_object_directory());
355 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
357 alt_copy
= xmemdupz(alt
, len
);
358 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
359 for (i
= 0; i
< entries
.nr
; i
++) {
360 const char *entry
= entries
.items
[i
].string
;
361 if (entry
[0] == '\0' || entry
[0] == '#')
363 if (!is_absolute_path(entry
) && depth
) {
364 error("%s: ignoring relative alternate object store %s",
365 relative_base
, entry
);
367 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
370 string_list_clear(&entries
, 0);
372 strbuf_release(&objdirbuf
);
375 void read_info_alternates(const char * relative_base
, int depth
)
380 const char alt_file_name
[] = "info/alternates";
381 /* Given that relative_base is no longer than PATH_MAX,
382 ensure that "path" has enough space to append "/", the
383 file name, "info/alternates", and a trailing NUL. */
384 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
387 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
388 fd
= git_open_noatime(path
);
391 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
395 mapsz
= xsize_t(st
.st_size
);
396 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
399 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
404 void add_to_alternates_file(const char *reference
)
406 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
407 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
408 char *alt
= mkpath("%s\n", reference
);
409 write_or_die(fd
, alt
, strlen(alt
));
410 if (commit_lock_file(lock
))
411 die("could not close alternates file");
413 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
416 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
418 struct alternate_object_database
*ent
;
421 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
426 void prepare_alt_odb(void)
433 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
436 alt_odb_tail
= &alt_odb_list
;
437 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
439 read_info_alternates(get_object_directory(), 0);
442 static int has_loose_object_local(const unsigned char *sha1
)
444 return !access(sha1_file_name(sha1
), F_OK
);
447 int has_loose_object_nonlocal(const unsigned char *sha1
)
449 struct alternate_object_database
*alt
;
451 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
452 fill_sha1_path(alt
->name
, sha1
);
453 if (!access(alt
->base
, F_OK
))
459 static int has_loose_object(const unsigned char *sha1
)
461 return has_loose_object_local(sha1
) ||
462 has_loose_object_nonlocal(sha1
);
465 static unsigned int pack_used_ctr
;
466 static unsigned int pack_mmap_calls
;
467 static unsigned int peak_pack_open_windows
;
468 static unsigned int pack_open_windows
;
469 static unsigned int pack_open_fds
;
470 static unsigned int pack_max_fds
;
471 static size_t peak_pack_mapped
;
472 static size_t pack_mapped
;
473 struct packed_git
*packed_git
;
475 void pack_report(void)
478 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
479 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
480 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
481 sz_fmt(getpagesize()),
482 sz_fmt(packed_git_window_size
),
483 sz_fmt(packed_git_limit
));
485 "pack_report: pack_used_ctr = %10u\n"
486 "pack_report: pack_mmap_calls = %10u\n"
487 "pack_report: pack_open_windows = %10u / %10u\n"
488 "pack_report: pack_mapped = "
489 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
492 pack_open_windows
, peak_pack_open_windows
,
493 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
497 * Open and mmap the index file at path, perform a couple of
498 * consistency checks, then record its information to p. Return 0 on
501 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
504 struct pack_idx_header
*hdr
;
506 uint32_t version
, nr
, i
, *index
;
507 int fd
= git_open_noatime(path
);
512 if (fstat(fd
, &st
)) {
516 idx_size
= xsize_t(st
.st_size
);
517 if (idx_size
< 4 * 256 + 20 + 20) {
519 return error("index file %s is too small", path
);
521 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
525 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
526 version
= ntohl(hdr
->idx_version
);
527 if (version
< 2 || version
> 2) {
528 munmap(idx_map
, idx_size
);
529 return error("index file %s is version %"PRIu32
530 " and is not supported by this binary"
531 " (try upgrading GIT to a newer version)",
540 index
+= 2; /* skip index header */
541 for (i
= 0; i
< 256; i
++) {
542 uint32_t n
= ntohl(index
[i
]);
544 munmap(idx_map
, idx_size
);
545 return error("non-monotonic index %s", path
);
553 * - 256 index entries 4 bytes each
554 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
555 * - 20-byte SHA1 of the packfile
556 * - 20-byte SHA1 file checksum
558 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
559 munmap(idx_map
, idx_size
);
560 return error("wrong index v1 file size in %s", path
);
562 } else if (version
== 2) {
565 * - 8 bytes of header
566 * - 256 index entries 4 bytes each
567 * - 20-byte sha1 entry * nr
568 * - 4-byte crc entry * nr
569 * - 4-byte offset entry * nr
570 * - 20-byte SHA1 of the packfile
571 * - 20-byte SHA1 file checksum
572 * And after the 4-byte offset table might be a
573 * variable sized table containing 8-byte entries
574 * for offsets larger than 2^31.
576 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
577 unsigned long max_size
= min_size
;
579 max_size
+= (nr
- 1)*8;
580 if (idx_size
< min_size
|| idx_size
> max_size
) {
581 munmap(idx_map
, idx_size
);
582 return error("wrong index v2 file size in %s", path
);
584 if (idx_size
!= min_size
&&
586 * make sure we can deal with large pack offsets.
587 * 31-bit signed offset won't be enough, neither
588 * 32-bit unsigned one will be.
590 (sizeof(off_t
) <= 4)) {
591 munmap(idx_map
, idx_size
);
592 return error("pack too large for current definition of off_t in %s", path
);
596 p
->index_version
= version
;
597 p
->index_data
= idx_map
;
598 p
->index_size
= idx_size
;
603 int open_pack_index(struct packed_git
*p
)
611 idx_name
= xstrdup(p
->pack_name
);
612 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
613 ret
= check_packed_git_idx(idx_name
, p
);
618 static void scan_windows(struct packed_git
*p
,
619 struct packed_git
**lru_p
,
620 struct pack_window
**lru_w
,
621 struct pack_window
**lru_l
)
623 struct pack_window
*w
, *w_l
;
625 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
627 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
637 static int unuse_one_window(struct packed_git
*current
)
639 struct packed_git
*p
, *lru_p
= NULL
;
640 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
643 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
644 for (p
= packed_git
; p
; p
= p
->next
)
645 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
647 munmap(lru_w
->base
, lru_w
->len
);
648 pack_mapped
-= lru_w
->len
;
650 lru_l
->next
= lru_w
->next
;
652 lru_p
->windows
= lru_w
->next
;
660 void release_pack_memory(size_t need
)
662 size_t cur
= pack_mapped
;
663 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
667 void *xmmap(void *start
, size_t length
,
668 int prot
, int flags
, int fd
, off_t offset
)
670 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
671 if (ret
== MAP_FAILED
) {
674 release_pack_memory(length
);
675 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
676 if (ret
== MAP_FAILED
)
677 die_errno("Out of memory? mmap failed");
682 void close_pack_windows(struct packed_git
*p
)
685 struct pack_window
*w
= p
->windows
;
688 die("pack '%s' still has open windows to it",
690 munmap(w
->base
, w
->len
);
691 pack_mapped
-= w
->len
;
693 p
->windows
= w
->next
;
699 * The LRU pack is the one with the oldest MRU window, preferring packs
700 * with no used windows, or the oldest mtime if it has no windows allocated.
702 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
704 struct pack_window
*w
, *this_mru_w
;
705 int has_windows_inuse
= 0;
708 * Reject this pack if it has windows and the previously selected
709 * one does not. If this pack does not have windows, reject
710 * it if the pack file is newer than the previously selected one.
712 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
715 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
717 * Reject this pack if any of its windows are in use,
718 * but the previously selected pack did not have any
719 * inuse windows. Otherwise, record that this pack
720 * has windows in use.
723 if (*accept_windows_inuse
)
724 has_windows_inuse
= 1;
729 if (w
->last_used
> this_mru_w
->last_used
)
733 * Reject this pack if it has windows that have been
734 * used more recently than the previously selected pack.
735 * If the previously selected pack had windows inuse and
736 * we have not encountered a window in this pack that is
737 * inuse, skip this check since we prefer a pack with no
738 * inuse windows to one that has inuse windows.
740 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
741 this_mru_w
->last_used
> (*mru_w
)->last_used
)
750 *accept_windows_inuse
= has_windows_inuse
;
753 static int close_one_pack(void)
755 struct packed_git
*p
, *lru_p
= NULL
;
756 struct pack_window
*mru_w
= NULL
;
757 int accept_windows_inuse
= 1;
759 for (p
= packed_git
; p
; p
= p
->next
) {
760 if (p
->pack_fd
== -1)
762 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
766 close(lru_p
->pack_fd
);
775 void unuse_pack(struct pack_window
**w_cursor
)
777 struct pack_window
*w
= *w_cursor
;
784 void close_pack_index(struct packed_git
*p
)
787 munmap((void *)p
->index_data
, p
->index_size
);
788 p
->index_data
= NULL
;
793 * This is used by git-repack in case a newly created pack happens to
794 * contain the same set of objects as an existing one. In that case
795 * the resulting file might be different even if its name would be the
796 * same. It is best to close any reference to the old pack before it is
797 * replaced on disk. Of course no index pointers or windows for given pack
798 * must subsist at this point. If ever objects from this pack are requested
799 * again, the new version of the pack will be reinitialized through
800 * reprepare_packed_git().
802 void free_pack_by_name(const char *pack_name
)
804 struct packed_git
*p
, **pp
= &packed_git
;
808 if (strcmp(pack_name
, p
->pack_name
) == 0) {
809 clear_delta_base_cache();
810 close_pack_windows(p
);
811 if (p
->pack_fd
!= -1) {
816 free(p
->bad_object_sha1
);
818 if (last_found_pack
== p
)
819 last_found_pack
= NULL
;
827 static unsigned int get_max_fd_limit(void)
833 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
840 long open_max
= sysconf(_SC_OPEN_MAX
);
844 * Otherwise, we got -1 for one of the two
847 * (1) sysconf() did not understand _SC_OPEN_MAX
848 * and signaled an error with -1; or
849 * (2) sysconf() said there is no limit.
851 * We _could_ clear errno before calling sysconf() to
852 * tell these two cases apart and return a huge number
853 * in the latter case to let the caller cap it to a
854 * value that is not so selfish, but letting the
855 * fallback OPEN_MAX codepath take care of these cases
864 return 1; /* see the caller ;-) */
869 * Do not call this directly as this leaks p->pack_fd on error return;
870 * call open_packed_git() instead.
872 static int open_packed_git_1(struct packed_git
*p
)
875 struct pack_header hdr
;
876 unsigned char sha1
[20];
877 unsigned char *idx_sha1
;
880 if (!p
->index_data
&& open_pack_index(p
))
881 return error("packfile %s index unavailable", p
->pack_name
);
884 unsigned int max_fds
= get_max_fd_limit();
886 /* Save 3 for stdin/stdout/stderr, 22 for work */
888 pack_max_fds
= max_fds
- 25;
893 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
896 p
->pack_fd
= git_open_noatime(p
->pack_name
);
897 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
901 /* If we created the struct before we had the pack we lack size. */
903 if (!S_ISREG(st
.st_mode
))
904 return error("packfile %s not a regular file", p
->pack_name
);
905 p
->pack_size
= st
.st_size
;
906 } else if (p
->pack_size
!= st
.st_size
)
907 return error("packfile %s size changed", p
->pack_name
);
909 /* We leave these file descriptors open with sliding mmap;
910 * there is no point keeping them open across exec(), though.
912 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
914 return error("cannot determine file descriptor flags");
915 fd_flag
|= FD_CLOEXEC
;
916 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
917 return error("cannot set FD_CLOEXEC");
919 /* Verify we recognize this pack file format. */
920 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
921 return error("file %s is far too short to be a packfile", p
->pack_name
);
922 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
923 return error("file %s is not a GIT packfile", p
->pack_name
);
924 if (!pack_version_ok(hdr
.hdr_version
))
925 return error("packfile %s is version %"PRIu32
" and not"
926 " supported (try upgrading GIT to a newer version)",
927 p
->pack_name
, ntohl(hdr
.hdr_version
));
929 /* Verify the pack matches its index. */
930 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
931 return error("packfile %s claims to have %"PRIu32
" objects"
932 " while index indicates %"PRIu32
" objects",
933 p
->pack_name
, ntohl(hdr
.hdr_entries
),
935 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
936 return error("end of packfile %s is unavailable", p
->pack_name
);
937 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
938 return error("packfile %s signature is unavailable", p
->pack_name
);
939 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
940 if (hashcmp(sha1
, idx_sha1
))
941 return error("packfile %s does not match index", p
->pack_name
);
945 static int open_packed_git(struct packed_git
*p
)
947 if (!open_packed_git_1(p
))
949 if (p
->pack_fd
!= -1) {
957 static int in_window(struct pack_window
*win
, off_t offset
)
959 /* We must promise at least 20 bytes (one hash) after the
960 * offset is available from this window, otherwise the offset
961 * is not actually in this window and a different window (which
962 * has that one hash excess) must be used. This is to support
963 * the object header and delta base parsing routines below.
965 off_t win_off
= win
->offset
;
966 return win_off
<= offset
967 && (offset
+ 20) <= (win_off
+ win
->len
);
970 unsigned char *use_pack(struct packed_git
*p
,
971 struct pack_window
**w_cursor
,
975 struct pack_window
*win
= *w_cursor
;
977 /* Since packfiles end in a hash of their content and it's
978 * pointless to ask for an offset into the middle of that
979 * hash, and the in_window function above wouldn't match
980 * don't allow an offset too close to the end of the file.
982 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
983 die("packfile %s cannot be accessed", p
->pack_name
);
984 if (offset
> (p
->pack_size
- 20))
985 die("offset beyond end of packfile (truncated pack?)");
987 if (!win
|| !in_window(win
, offset
)) {
990 for (win
= p
->windows
; win
; win
= win
->next
) {
991 if (in_window(win
, offset
))
995 size_t window_align
= packed_git_window_size
/ 2;
998 if (p
->pack_fd
== -1 && open_packed_git(p
))
999 die("packfile %s cannot be accessed", p
->pack_name
);
1001 win
= xcalloc(1, sizeof(*win
));
1002 win
->offset
= (offset
/ window_align
) * window_align
;
1003 len
= p
->pack_size
- win
->offset
;
1004 if (len
> packed_git_window_size
)
1005 len
= packed_git_window_size
;
1006 win
->len
= (size_t)len
;
1007 pack_mapped
+= win
->len
;
1008 while (packed_git_limit
< pack_mapped
1009 && unuse_one_window(p
))
1011 win
->base
= xmmap(NULL
, win
->len
,
1012 PROT_READ
, MAP_PRIVATE
,
1013 p
->pack_fd
, win
->offset
);
1014 if (win
->base
== MAP_FAILED
)
1015 die("packfile %s cannot be mapped: %s",
1018 if (!win
->offset
&& win
->len
== p
->pack_size
1019 && !p
->do_not_close
) {
1025 pack_open_windows
++;
1026 if (pack_mapped
> peak_pack_mapped
)
1027 peak_pack_mapped
= pack_mapped
;
1028 if (pack_open_windows
> peak_pack_open_windows
)
1029 peak_pack_open_windows
= pack_open_windows
;
1030 win
->next
= p
->windows
;
1034 if (win
!= *w_cursor
) {
1035 win
->last_used
= pack_used_ctr
++;
1039 offset
-= win
->offset
;
1041 *left
= win
->len
- xsize_t(offset
);
1042 return win
->base
+ offset
;
1045 static struct packed_git
*alloc_packed_git(int extra
)
1047 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1048 memset(p
, 0, sizeof(*p
));
1053 static void try_to_free_pack_memory(size_t size
)
1055 release_pack_memory(size
);
1058 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1060 static int have_set_try_to_free_routine
;
1062 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1064 if (!have_set_try_to_free_routine
) {
1065 have_set_try_to_free_routine
= 1;
1066 set_try_to_free_routine(try_to_free_pack_memory
);
1070 * Make sure a corresponding .pack file exists and that
1071 * the index looks sane.
1073 path_len
-= strlen(".idx");
1078 memcpy(p
->pack_name
, path
, path_len
);
1080 strcpy(p
->pack_name
+ path_len
, ".keep");
1081 if (!access(p
->pack_name
, F_OK
))
1084 strcpy(p
->pack_name
+ path_len
, ".pack");
1085 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1090 /* ok, it looks sane as far as we can check without
1091 * actually mapping the pack file.
1093 p
->pack_size
= st
.st_size
;
1094 p
->pack_local
= local
;
1095 p
->mtime
= st
.st_mtime
;
1096 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1101 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1103 const char *path
= sha1_pack_name(sha1
);
1104 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1106 strcpy(p
->pack_name
, path
);
1107 hashcpy(p
->sha1
, sha1
);
1108 if (check_packed_git_idx(idx_path
, p
)) {
1116 void install_packed_git(struct packed_git
*pack
)
1118 if (pack
->pack_fd
!= -1)
1121 pack
->next
= packed_git
;
1125 void (*report_garbage
)(const char *desc
, const char *path
);
1127 static void report_helper(const struct string_list
*list
,
1128 int seen_bits
, int first
, int last
)
1131 switch (seen_bits
) {
1133 msg
= "no corresponding .idx or .pack";
1136 msg
= "no corresponding .idx";
1139 msg
= "no corresponding .pack";
1144 for (; first
< last
; first
++)
1145 report_garbage(msg
, list
->items
[first
].string
);
1148 static void report_pack_garbage(struct string_list
*list
)
1150 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1152 if (!report_garbage
)
1155 sort_string_list(list
);
1157 for (i
= 0; i
< list
->nr
; i
++) {
1158 const char *path
= list
->items
[i
].string
;
1159 if (baselen
!= -1 &&
1160 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1161 report_helper(list
, seen_bits
, first
, i
);
1165 if (baselen
== -1) {
1166 const char *dot
= strrchr(path
, '.');
1168 report_garbage("garbage found", path
);
1171 baselen
= dot
- path
+ 1;
1174 if (!strcmp(path
+ baselen
, "pack"))
1176 else if (!strcmp(path
+ baselen
, "idx"))
1179 report_helper(list
, seen_bits
, first
, list
->nr
);
1182 static void prepare_packed_git_one(char *objdir
, int local
)
1184 struct strbuf path
= STRBUF_INIT
;
1188 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1190 strbuf_addstr(&path
, objdir
);
1191 strbuf_addstr(&path
, "/pack");
1192 dir
= opendir(path
.buf
);
1194 if (errno
!= ENOENT
)
1195 error("unable to open object pack directory: %s: %s",
1196 path
.buf
, strerror(errno
));
1197 strbuf_release(&path
);
1200 strbuf_addch(&path
, '/');
1201 dirnamelen
= path
.len
;
1202 while ((de
= readdir(dir
)) != NULL
) {
1203 struct packed_git
*p
;
1206 if (is_dot_or_dotdot(de
->d_name
))
1209 strbuf_setlen(&path
, dirnamelen
);
1210 strbuf_addstr(&path
, de
->d_name
);
1212 base_len
= path
.len
;
1213 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1214 /* Don't reopen a pack we already have. */
1215 for (p
= packed_git
; p
; p
= p
->next
) {
1217 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1219 !memcmp(p
->pack_name
, path
.buf
, len
))
1224 * See if it really is a valid .idx file with
1225 * corresponding .pack file that we can map.
1227 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1228 install_packed_git(p
);
1231 if (!report_garbage
)
1234 if (ends_with(de
->d_name
, ".idx") ||
1235 ends_with(de
->d_name
, ".pack") ||
1236 ends_with(de
->d_name
, ".bitmap") ||
1237 ends_with(de
->d_name
, ".keep"))
1238 string_list_append(&garbage
, path
.buf
);
1240 report_garbage("garbage found", path
.buf
);
1243 report_pack_garbage(&garbage
);
1244 string_list_clear(&garbage
, 0);
1245 strbuf_release(&path
);
1248 static int sort_pack(const void *a_
, const void *b_
)
1250 struct packed_git
*a
= *((struct packed_git
**)a_
);
1251 struct packed_git
*b
= *((struct packed_git
**)b_
);
1255 * Local packs tend to contain objects specific to our
1256 * variant of the project than remote ones. In addition,
1257 * remote ones could be on a network mounted filesystem.
1258 * Favor local ones for these reasons.
1260 st
= a
->pack_local
- b
->pack_local
;
1265 * Younger packs tend to contain more recent objects,
1266 * and more recent objects tend to get accessed more
1269 if (a
->mtime
< b
->mtime
)
1271 else if (a
->mtime
== b
->mtime
)
1276 static void rearrange_packed_git(void)
1278 struct packed_git
**ary
, *p
;
1281 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1286 /* prepare an array of packed_git for easier sorting */
1287 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1288 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1291 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1293 /* link them back again */
1294 for (i
= 0; i
< n
- 1; i
++)
1295 ary
[i
]->next
= ary
[i
+ 1];
1296 ary
[n
- 1]->next
= NULL
;
1297 packed_git
= ary
[0];
1302 static int prepare_packed_git_run_once
= 0;
1303 void prepare_packed_git(void)
1305 struct alternate_object_database
*alt
;
1307 if (prepare_packed_git_run_once
)
1309 prepare_packed_git_one(get_object_directory(), 1);
1311 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1313 prepare_packed_git_one(alt
->base
, 0);
1314 alt
->name
[-1] = '/';
1316 rearrange_packed_git();
1317 prepare_packed_git_run_once
= 1;
1320 void reprepare_packed_git(void)
1322 prepare_packed_git_run_once
= 0;
1323 prepare_packed_git();
1326 static void mark_bad_packed_object(struct packed_git
*p
,
1327 const unsigned char *sha1
)
1330 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1331 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1333 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1334 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1335 p
->num_bad_objects
++;
1338 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1340 struct packed_git
*p
;
1343 for (p
= packed_git
; p
; p
= p
->next
)
1344 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1345 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1351 * With an in-core object data in "map", rehash it to make sure the
1352 * object name actually matches "sha1" to detect object corruption.
1353 * With "map" == NULL, try reading the object named with "sha1" using
1354 * the streaming interface and rehash it to do the same.
1356 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1357 unsigned long size
, const char *type
)
1359 unsigned char real_sha1
[20];
1360 enum object_type obj_type
;
1361 struct git_istream
*st
;
1367 hash_sha1_file(map
, size
, type
, real_sha1
);
1368 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1371 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1375 /* Generate the header */
1376 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1380 git_SHA1_Update(&c
, hdr
, hdrlen
);
1382 char buf
[1024 * 16];
1383 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1391 git_SHA1_Update(&c
, buf
, readlen
);
1393 git_SHA1_Final(real_sha1
, &c
);
1395 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1398 int git_open_noatime(const char *name
)
1400 static int sha1_file_open_flag
= O_NOATIME
;
1403 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1407 /* Might the failure be due to O_NOATIME? */
1408 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1409 sha1_file_open_flag
= 0;
1417 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1419 struct alternate_object_database
*alt
;
1421 if (!lstat(sha1_file_name(sha1
), st
))
1426 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1427 fill_sha1_path(alt
->name
, sha1
);
1428 if (!lstat(alt
->base
, st
))
1435 static int open_sha1_file(const unsigned char *sha1
)
1438 struct alternate_object_database
*alt
;
1439 int most_interesting_errno
;
1441 fd
= git_open_noatime(sha1_file_name(sha1
));
1444 most_interesting_errno
= errno
;
1447 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1448 fill_sha1_path(alt
->name
, sha1
);
1449 fd
= git_open_noatime(alt
->base
);
1452 if (most_interesting_errno
== ENOENT
)
1453 most_interesting_errno
= errno
;
1455 errno
= most_interesting_errno
;
1459 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1464 fd
= open_sha1_file(sha1
);
1469 if (!fstat(fd
, &st
)) {
1470 *size
= xsize_t(st
.st_size
);
1472 /* mmap() is forbidden on empty files */
1473 error("object file %s is empty", sha1_file_name(sha1
));
1476 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1483 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1484 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1487 unsigned long size
, c
;
1488 unsigned long used
= 0;
1491 *type
= (c
>> 4) & 7;
1495 if (len
<= used
|| bitsizeof(long) <= shift
) {
1496 error("bad object header");
1501 size
+= (c
& 0x7f) << shift
;
1508 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1510 /* Get the data stream */
1511 memset(stream
, 0, sizeof(*stream
));
1512 stream
->next_in
= map
;
1513 stream
->avail_in
= mapsize
;
1514 stream
->next_out
= buffer
;
1515 stream
->avail_out
= bufsiz
;
1517 git_inflate_init(stream
);
1518 return git_inflate(stream
, 0);
1521 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1523 int bytes
= strlen(buffer
) + 1;
1524 unsigned char *buf
= xmallocz(size
);
1528 n
= stream
->total_out
- bytes
;
1531 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1533 if (bytes
<= size
) {
1535 * The above condition must be (bytes <= size), not
1536 * (bytes < size). In other words, even though we
1537 * expect no more output and set avail_out to zero,
1538 * the input zlib stream may have bytes that express
1539 * "this concludes the stream", and we *do* want to
1542 * Otherwise we would not be able to test that we
1543 * consumed all the input to reach the expected size;
1544 * we also want to check that zlib tells us that all
1545 * went well with status == Z_STREAM_END at the end.
1547 stream
->next_out
= buf
+ bytes
;
1548 stream
->avail_out
= size
- bytes
;
1549 while (status
== Z_OK
)
1550 status
= git_inflate(stream
, Z_FINISH
);
1552 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1553 git_inflate_end(stream
);
1558 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1559 else if (stream
->avail_in
)
1560 error("garbage at end of loose object '%s'",
1567 * We used to just use "sscanf()", but that's actually way
1568 * too permissive for what we want to check. So do an anal
1569 * object header parse by hand.
1571 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1578 * The type can be at most ten bytes (including the
1579 * terminating '\0' that we add), and is followed by
1588 if (i
>= sizeof(type
))
1594 * The length must follow immediately, and be in canonical
1595 * decimal format (ie "010" is not valid).
1597 size
= *hdr
++ - '0';
1602 unsigned long c
= *hdr
- '0';
1606 size
= size
* 10 + c
;
1612 * The length must be followed by a zero byte
1614 return *hdr
? -1 : type_from_string(type
);
1617 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1623 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1624 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1627 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1630 unsigned long get_size_from_delta(struct packed_git
*p
,
1631 struct pack_window
**w_curs
,
1634 const unsigned char *data
;
1635 unsigned char delta_head
[20], *in
;
1639 memset(&stream
, 0, sizeof(stream
));
1640 stream
.next_out
= delta_head
;
1641 stream
.avail_out
= sizeof(delta_head
);
1643 git_inflate_init(&stream
);
1645 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1646 stream
.next_in
= in
;
1647 st
= git_inflate(&stream
, Z_FINISH
);
1648 curpos
+= stream
.next_in
- in
;
1649 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1650 stream
.total_out
< sizeof(delta_head
));
1651 git_inflate_end(&stream
);
1652 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1653 error("delta data unpack-initial failed");
1657 /* Examine the initial part of the delta to figure out
1662 /* ignore base size */
1663 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1665 /* Read the result size */
1666 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1669 static off_t
get_delta_base(struct packed_git
*p
,
1670 struct pack_window
**w_curs
,
1672 enum object_type type
,
1673 off_t delta_obj_offset
)
1675 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1678 /* use_pack() assured us we have [base_info, base_info + 20)
1679 * as a range that we can look at without walking off the
1680 * end of the mapped window. Its actually the hash size
1681 * that is assured. An OFS_DELTA longer than the hash size
1682 * is stupid, as then a REF_DELTA would be smaller to store.
1684 if (type
== OBJ_OFS_DELTA
) {
1686 unsigned char c
= base_info
[used
++];
1687 base_offset
= c
& 127;
1690 if (!base_offset
|| MSB(base_offset
, 7))
1691 return 0; /* overflow */
1692 c
= base_info
[used
++];
1693 base_offset
= (base_offset
<< 7) + (c
& 127);
1695 base_offset
= delta_obj_offset
- base_offset
;
1696 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1697 return 0; /* out of bound */
1699 } else if (type
== OBJ_REF_DELTA
) {
1700 /* The base entry _must_ be in the same pack */
1701 base_offset
= find_pack_entry_one(base_info
, p
);
1704 die("I am totally screwed");
1709 * Like get_delta_base above, but we return the sha1 instead of the pack
1710 * offset. This means it is cheaper for REF deltas (we do not have to do
1711 * the final object lookup), but more expensive for OFS deltas (we
1712 * have to load the revidx to convert the offset back into a sha1).
1714 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1715 struct pack_window
**w_curs
,
1717 enum object_type type
,
1718 off_t delta_obj_offset
)
1720 if (type
== OBJ_REF_DELTA
) {
1721 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1723 } else if (type
== OBJ_OFS_DELTA
) {
1724 struct revindex_entry
*revidx
;
1725 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1726 type
, delta_obj_offset
);
1731 revidx
= find_pack_revindex(p
, base_offset
);
1735 return nth_packed_object_sha1(p
, revidx
->nr
);
1740 int unpack_object_header(struct packed_git
*p
,
1741 struct pack_window
**w_curs
,
1743 unsigned long *sizep
)
1745 unsigned char *base
;
1748 enum object_type type
;
1750 /* use_pack() assures us we have [base, base + 20) available
1751 * as a range that we can look at. (Its actually the hash
1752 * size that is assured.) With our object header encoding
1753 * the maximum deflated object size is 2^137, which is just
1754 * insane, so we know won't exceed what we have been given.
1756 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1757 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1766 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1769 struct revindex_entry
*revidx
;
1770 const unsigned char *sha1
;
1771 revidx
= find_pack_revindex(p
, obj_offset
);
1774 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1775 mark_bad_packed_object(p
, sha1
);
1776 type
= sha1_object_info(sha1
, NULL
);
1777 if (type
<= OBJ_NONE
)
1782 #define POI_STACK_PREALLOC 64
1784 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1786 enum object_type type
,
1787 struct pack_window
**w_curs
,
1790 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1791 off_t
*poi_stack
= small_poi_stack
;
1792 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1794 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1797 /* Push the object we're going to leave behind */
1798 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1799 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1800 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1801 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1803 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1805 poi_stack
[poi_stack_nr
++] = obj_offset
;
1806 /* If parsing the base offset fails, just unwind */
1807 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1810 curpos
= obj_offset
= base_offset
;
1811 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1812 if (type
<= OBJ_NONE
) {
1813 /* If getting the base itself fails, we first
1814 * retry the base, otherwise unwind */
1815 type
= retry_bad_packed_offset(p
, base_offset
);
1816 if (type
> OBJ_NONE
)
1830 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1831 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1836 if (poi_stack
!= small_poi_stack
)
1841 while (poi_stack_nr
) {
1842 obj_offset
= poi_stack
[--poi_stack_nr
];
1843 type
= retry_bad_packed_offset(p
, obj_offset
);
1844 if (type
> OBJ_NONE
)
1851 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1852 struct object_info
*oi
)
1854 struct pack_window
*w_curs
= NULL
;
1856 off_t curpos
= obj_offset
;
1857 enum object_type type
;
1860 * We always get the representation type, but only convert it to
1861 * a "real" type later if the caller is interested.
1863 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1866 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1867 off_t tmp_pos
= curpos
;
1868 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1874 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1875 if (*oi
->sizep
== 0) {
1884 if (oi
->disk_sizep
) {
1885 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1886 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1890 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1891 if (*oi
->typep
< 0) {
1897 if (oi
->delta_base_sha1
) {
1898 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1899 const unsigned char *base
;
1901 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1908 hashcpy(oi
->delta_base_sha1
, base
);
1910 hashclr(oi
->delta_base_sha1
);
1914 unuse_pack(&w_curs
);
1918 static void *unpack_compressed_entry(struct packed_git
*p
,
1919 struct pack_window
**w_curs
,
1925 unsigned char *buffer
, *in
;
1927 buffer
= xmallocz_gently(size
);
1930 memset(&stream
, 0, sizeof(stream
));
1931 stream
.next_out
= buffer
;
1932 stream
.avail_out
= size
+ 1;
1934 git_inflate_init(&stream
);
1936 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1937 stream
.next_in
= in
;
1938 st
= git_inflate(&stream
, Z_FINISH
);
1939 if (!stream
.avail_out
)
1940 break; /* the payload is larger than it should be */
1941 curpos
+= stream
.next_in
- in
;
1942 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1943 git_inflate_end(&stream
);
1944 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1952 #define MAX_DELTA_CACHE (256)
1954 static size_t delta_base_cached
;
1956 static struct delta_base_cache_lru_list
{
1957 struct delta_base_cache_lru_list
*prev
;
1958 struct delta_base_cache_lru_list
*next
;
1959 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1961 static struct delta_base_cache_entry
{
1962 struct delta_base_cache_lru_list lru
;
1964 struct packed_git
*p
;
1967 enum object_type type
;
1968 } delta_base_cache
[MAX_DELTA_CACHE
];
1970 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1974 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1975 hash
+= (hash
>> 8) + (hash
>> 16);
1976 return hash
% MAX_DELTA_CACHE
;
1979 static struct delta_base_cache_entry
*
1980 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1982 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1983 return delta_base_cache
+ hash
;
1986 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
1987 struct packed_git
*p
, off_t base_offset
)
1989 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
1992 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1994 struct delta_base_cache_entry
*ent
;
1995 ent
= get_delta_base_cache_entry(p
, base_offset
);
1996 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
1999 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2002 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2003 ent
->lru
.prev
->next
= ent
->lru
.next
;
2004 delta_base_cached
-= ent
->size
;
2007 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2008 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2010 struct delta_base_cache_entry
*ent
;
2013 ent
= get_delta_base_cache_entry(p
, base_offset
);
2015 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2016 return unpack_entry(p
, base_offset
, type
, base_size
);
2021 clear_delta_base_cache_entry(ent
);
2023 ret
= xmemdupz(ent
->data
, ent
->size
);
2025 *base_size
= ent
->size
;
2029 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2034 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2035 ent
->lru
.prev
->next
= ent
->lru
.next
;
2036 delta_base_cached
-= ent
->size
;
2040 void clear_delta_base_cache(void)
2043 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2044 release_delta_base_cache(&delta_base_cache
[p
]);
2047 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2048 void *base
, unsigned long base_size
, enum object_type type
)
2050 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2051 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2052 struct delta_base_cache_lru_list
*lru
;
2054 release_delta_base_cache(ent
);
2055 delta_base_cached
+= base_size
;
2057 for (lru
= delta_base_cache_lru
.next
;
2058 delta_base_cached
> delta_base_cache_limit
2059 && lru
!= &delta_base_cache_lru
;
2061 struct delta_base_cache_entry
*f
= (void *)lru
;
2062 if (f
->type
== OBJ_BLOB
)
2063 release_delta_base_cache(f
);
2065 for (lru
= delta_base_cache_lru
.next
;
2066 delta_base_cached
> delta_base_cache_limit
2067 && lru
!= &delta_base_cache_lru
;
2069 struct delta_base_cache_entry
*f
= (void *)lru
;
2070 release_delta_base_cache(f
);
2074 ent
->base_offset
= base_offset
;
2077 ent
->size
= base_size
;
2078 ent
->lru
.next
= &delta_base_cache_lru
;
2079 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2080 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2081 delta_base_cache_lru
.prev
= &ent
->lru
;
2084 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2085 unsigned long *size
);
2087 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2089 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2090 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2091 p
->pack_name
, (uintmax_t)obj_offset
);
2094 int do_check_packed_object_crc
;
2096 #define UNPACK_ENTRY_STACK_PREALLOC 64
2097 struct unpack_entry_stack_ent
{
2103 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2104 enum object_type
*final_type
, unsigned long *final_size
)
2106 struct pack_window
*w_curs
= NULL
;
2107 off_t curpos
= obj_offset
;
2110 enum object_type type
;
2111 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2112 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2113 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2114 int base_from_cache
= 0;
2116 write_pack_access_log(p
, obj_offset
);
2118 /* PHASE 1: drill down to the innermost base object */
2122 struct delta_base_cache_entry
*ent
;
2124 ent
= get_delta_base_cache_entry(p
, curpos
);
2125 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2129 clear_delta_base_cache_entry(ent
);
2130 base_from_cache
= 1;
2134 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2135 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2136 unsigned long len
= revidx
[1].offset
- obj_offset
;
2137 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2138 const unsigned char *sha1
=
2139 nth_packed_object_sha1(p
, revidx
->nr
);
2140 error("bad packed object CRC for %s",
2142 mark_bad_packed_object(p
, sha1
);
2143 unuse_pack(&w_curs
);
2148 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2149 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2152 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2154 error("failed to validate delta base reference "
2155 "at offset %"PRIuMAX
" from %s",
2156 (uintmax_t)curpos
, p
->pack_name
);
2157 /* bail to phase 2, in hopes of recovery */
2162 /* push object, proceed to base */
2163 if (delta_stack_nr
>= delta_stack_alloc
2164 && delta_stack
== small_delta_stack
) {
2165 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2166 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2167 memcpy(delta_stack
, small_delta_stack
,
2168 sizeof(*delta_stack
)*delta_stack_nr
);
2170 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2172 i
= delta_stack_nr
++;
2173 delta_stack
[i
].obj_offset
= obj_offset
;
2174 delta_stack
[i
].curpos
= curpos
;
2175 delta_stack
[i
].size
= size
;
2177 curpos
= obj_offset
= base_offset
;
2180 /* PHASE 2: handle the base */
2185 die("BUG in unpack_entry: left loop at a valid delta");
2191 if (!base_from_cache
)
2192 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2196 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2197 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2200 /* PHASE 3: apply deltas in order */
2203 * 'data' holds the base data, or NULL if there was corruption
2205 while (delta_stack_nr
) {
2208 unsigned long delta_size
, base_size
= size
;
2214 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2218 * We're probably in deep shit, but let's try to fetch
2219 * the required base anyway from another pack or loose.
2220 * This is costly but should happen only in the presence
2221 * of a corrupted pack, and is better than failing outright.
2223 struct revindex_entry
*revidx
;
2224 const unsigned char *base_sha1
;
2225 revidx
= find_pack_revindex(p
, obj_offset
);
2227 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2228 error("failed to read delta base object %s"
2229 " at offset %"PRIuMAX
" from %s",
2230 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2232 mark_bad_packed_object(p
, base_sha1
);
2233 base
= read_object(base_sha1
, &type
, &base_size
);
2237 i
= --delta_stack_nr
;
2238 obj_offset
= delta_stack
[i
].obj_offset
;
2239 curpos
= delta_stack
[i
].curpos
;
2240 delta_size
= delta_stack
[i
].size
;
2245 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2248 error("failed to unpack compressed delta "
2249 "at offset %"PRIuMAX
" from %s",
2250 (uintmax_t)curpos
, p
->pack_name
);
2255 data
= patch_delta(base
, base_size
,
2256 delta_data
, delta_size
,
2260 * We could not apply the delta; warn the user, but keep going.
2261 * Our failure will be noticed either in the next iteration of
2262 * the loop, or if this is the final delta, in the caller when
2263 * we return NULL. Those code paths will take care of making
2264 * a more explicit warning and retrying with another copy of
2268 error("failed to apply delta");
2276 unuse_pack(&w_curs
);
2278 if (delta_stack
!= small_delta_stack
)
2284 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2287 const unsigned char *index
= p
->index_data
;
2289 if (open_pack_index(p
))
2291 index
= p
->index_data
;
2293 if (n
>= p
->num_objects
)
2296 if (p
->index_version
== 1) {
2297 return index
+ 24 * n
+ 4;
2300 return index
+ 20 * n
;
2304 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2306 const unsigned char *index
= p
->index_data
;
2308 if (p
->index_version
== 1) {
2309 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2312 index
+= 8 + p
->num_objects
* (20 + 4);
2313 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2314 if (!(off
& 0x80000000))
2316 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2317 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2318 ntohl(*((uint32_t *)(index
+ 4)));
2322 off_t
find_pack_entry_one(const unsigned char *sha1
,
2323 struct packed_git
*p
)
2325 const uint32_t *level1_ofs
= p
->index_data
;
2326 const unsigned char *index
= p
->index_data
;
2327 unsigned hi
, lo
, stride
;
2328 static int use_lookup
= -1;
2329 static int debug_lookup
= -1;
2331 if (debug_lookup
< 0)
2332 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2335 if (open_pack_index(p
))
2337 level1_ofs
= p
->index_data
;
2338 index
= p
->index_data
;
2340 if (p
->index_version
> 1) {
2345 hi
= ntohl(level1_ofs
[*sha1
]);
2346 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2347 if (p
->index_version
> 1) {
2355 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2356 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2359 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2361 int pos
= sha1_entry_pos(index
, stride
, 0,
2362 lo
, hi
, p
->num_objects
, sha1
);
2365 return nth_packed_object_offset(p
, pos
);
2369 unsigned mi
= (lo
+ hi
) / 2;
2370 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2373 printf("lo %u hi %u rg %u mi %u\n",
2374 lo
, hi
, hi
- lo
, mi
);
2376 return nth_packed_object_offset(p
, mi
);
2385 int is_pack_valid(struct packed_git
*p
)
2387 /* An already open pack is known to be valid. */
2388 if (p
->pack_fd
!= -1)
2391 /* If the pack has one window completely covering the
2392 * file size, the pack is known to be valid even if
2393 * the descriptor is not currently open.
2396 struct pack_window
*w
= p
->windows
;
2398 if (!w
->offset
&& w
->len
== p
->pack_size
)
2402 /* Force the pack to open to prove its valid. */
2403 return !open_packed_git(p
);
2406 static int fill_pack_entry(const unsigned char *sha1
,
2407 struct pack_entry
*e
,
2408 struct packed_git
*p
)
2412 if (p
->num_bad_objects
) {
2414 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2415 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2419 offset
= find_pack_entry_one(sha1
, p
);
2424 * We are about to tell the caller where they can locate the
2425 * requested object. We better make sure the packfile is
2426 * still here and can be accessed before supplying that
2427 * answer, as it may have been deleted since the index was
2430 if (!is_pack_valid(p
)) {
2431 warning("packfile %s cannot be accessed", p
->pack_name
);
2436 hashcpy(e
->sha1
, sha1
);
2441 * Iff a pack file contains the object named by sha1, return true and
2442 * store its location to e.
2444 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2446 struct packed_git
*p
;
2448 prepare_packed_git();
2452 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2455 for (p
= packed_git
; p
; p
= p
->next
) {
2456 if (p
== last_found_pack
)
2457 continue; /* we already checked this one */
2459 if (fill_pack_entry(sha1
, e
, p
)) {
2460 last_found_pack
= p
;
2467 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2468 struct packed_git
*packs
)
2470 struct packed_git
*p
;
2472 for (p
= packs
; p
; p
= p
->next
) {
2473 if (find_pack_entry_one(sha1
, p
))
2480 static int sha1_loose_object_info(const unsigned char *sha1
,
2481 struct object_info
*oi
)
2484 unsigned long mapsize
, size
;
2489 if (oi
->delta_base_sha1
)
2490 hashclr(oi
->delta_base_sha1
);
2493 * If we don't care about type or size, then we don't
2494 * need to look inside the object at all. Note that we
2495 * do not optimize out the stat call, even if the
2496 * caller doesn't care about the disk-size, since our
2497 * return value implicitly indicates whether the
2498 * object even exists.
2500 if (!oi
->typep
&& !oi
->sizep
) {
2502 if (stat_sha1_file(sha1
, &st
) < 0)
2505 *oi
->disk_sizep
= st
.st_size
;
2509 map
= map_sha1_file(sha1
, &mapsize
);
2513 *oi
->disk_sizep
= mapsize
;
2514 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2515 status
= error("unable to unpack %s header",
2517 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2518 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2521 git_inflate_end(&stream
);
2522 munmap(map
, mapsize
);
2524 *oi
->typep
= status
;
2528 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2530 struct cached_object
*co
;
2531 struct pack_entry e
;
2533 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2535 co
= find_cached_object(real
);
2538 *(oi
->typep
) = co
->type
;
2540 *(oi
->sizep
) = co
->size
;
2542 *(oi
->disk_sizep
) = 0;
2543 if (oi
->delta_base_sha1
)
2544 hashclr(oi
->delta_base_sha1
);
2545 oi
->whence
= OI_CACHED
;
2549 if (!find_pack_entry(real
, &e
)) {
2550 /* Most likely it's a loose object. */
2551 if (!sha1_loose_object_info(real
, oi
)) {
2552 oi
->whence
= OI_LOOSE
;
2556 /* Not a loose object; someone else may have just packed it. */
2557 reprepare_packed_git();
2558 if (!find_pack_entry(real
, &e
))
2562 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2564 mark_bad_packed_object(e
.p
, real
);
2565 return sha1_object_info_extended(real
, oi
, 0);
2566 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2567 oi
->whence
= OI_DBCACHED
;
2569 oi
->whence
= OI_PACKED
;
2570 oi
->u
.packed
.offset
= e
.offset
;
2571 oi
->u
.packed
.pack
= e
.p
;
2572 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2573 rtype
== OBJ_OFS_DELTA
);
2579 /* returns enum object_type or negative */
2580 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2582 enum object_type type
;
2583 struct object_info oi
= {NULL
};
2587 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2592 static void *read_packed_sha1(const unsigned char *sha1
,
2593 enum object_type
*type
, unsigned long *size
)
2595 struct pack_entry e
;
2598 if (!find_pack_entry(sha1
, &e
))
2600 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2603 * We're probably in deep shit, but let's try to fetch
2604 * the required object anyway from another pack or loose.
2605 * This should happen only in the presence of a corrupted
2606 * pack, and is better than failing outright.
2608 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2609 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2610 mark_bad_packed_object(e
.p
, sha1
);
2611 data
= read_object(sha1
, type
, size
);
2616 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2617 unsigned char *sha1
)
2619 struct cached_object
*co
;
2621 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2622 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2624 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2625 co
= &cached_objects
[cached_object_nr
++];
2628 co
->buf
= xmalloc(len
);
2629 memcpy(co
->buf
, buf
, len
);
2630 hashcpy(co
->sha1
, sha1
);
2634 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2635 unsigned long *size
)
2637 unsigned long mapsize
;
2639 struct cached_object
*co
;
2641 co
= find_cached_object(sha1
);
2645 return xmemdupz(co
->buf
, co
->size
);
2648 buf
= read_packed_sha1(sha1
, type
, size
);
2651 map
= map_sha1_file(sha1
, &mapsize
);
2653 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2654 munmap(map
, mapsize
);
2657 reprepare_packed_git();
2658 return read_packed_sha1(sha1
, type
, size
);
2662 * This function dies on corrupt objects; the callers who want to
2663 * deal with them should arrange to call read_object() and give error
2664 * messages themselves.
2666 void *read_sha1_file_extended(const unsigned char *sha1
,
2667 enum object_type
*type
,
2668 unsigned long *size
,
2672 const struct packed_git
*p
;
2673 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2676 data
= read_object(repl
, type
, size
);
2680 if (errno
&& errno
!= ENOENT
)
2681 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2683 /* die if we replaced an object with one that does not exist */
2685 die("replacement %s not found for %s",
2686 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2688 if (has_loose_object(repl
)) {
2689 const char *path
= sha1_file_name(sha1
);
2691 die("loose object %s (stored in %s) is corrupt",
2692 sha1_to_hex(repl
), path
);
2695 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2696 die("packed object %s (stored in %s) is corrupt",
2697 sha1_to_hex(repl
), p
->pack_name
);
2702 void *read_object_with_reference(const unsigned char *sha1
,
2703 const char *required_type_name
,
2704 unsigned long *size
,
2705 unsigned char *actual_sha1_return
)
2707 enum object_type type
, required_type
;
2709 unsigned long isize
;
2710 unsigned char actual_sha1
[20];
2712 required_type
= type_from_string(required_type_name
);
2713 hashcpy(actual_sha1
, sha1
);
2715 int ref_length
= -1;
2716 const char *ref_type
= NULL
;
2718 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2721 if (type
== required_type
) {
2723 if (actual_sha1_return
)
2724 hashcpy(actual_sha1_return
, actual_sha1
);
2727 /* Handle references */
2728 else if (type
== OBJ_COMMIT
)
2730 else if (type
== OBJ_TAG
)
2731 ref_type
= "object ";
2736 ref_length
= strlen(ref_type
);
2738 if (ref_length
+ 40 > isize
||
2739 memcmp(buffer
, ref_type
, ref_length
) ||
2740 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2745 /* Now we have the ID of the referred-to object in
2746 * actual_sha1. Check again. */
2750 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2751 const char *type
, unsigned char *sha1
,
2752 char *hdr
, int *hdrlen
)
2756 /* Generate the header */
2757 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2761 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2762 git_SHA1_Update(&c
, buf
, len
);
2763 git_SHA1_Final(sha1
, &c
);
2767 * Move the just written object into its final resting place.
2768 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2769 * "moving" is only a part of what it does, when no patch between
2770 * master to pu changes the call sites of this function.
2772 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2776 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2778 else if (link(tmpfile
, filename
))
2782 * Coda hack - coda doesn't like cross-directory links,
2783 * so we fall back to a rename, which will mean that it
2784 * won't be able to check collisions, but that's not a
2787 * The same holds for FAT formatted media.
2789 * When this succeeds, we just return. We have nothing
2792 if (ret
&& ret
!= EEXIST
) {
2794 if (!rename(tmpfile
, filename
))
2798 unlink_or_warn(tmpfile
);
2800 if (ret
!= EEXIST
) {
2801 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2803 /* FIXME!!! Collision check here ? */
2807 if (adjust_shared_perm(filename
))
2808 return error("unable to set permission to '%s'", filename
);
2812 static int write_buffer(int fd
, const void *buf
, size_t len
)
2814 if (write_in_full(fd
, buf
, len
) < 0)
2815 return error("file write error (%s)", strerror(errno
));
2819 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2820 unsigned char *sha1
)
2824 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2828 /* Finalize a file on disk, and close it. */
2829 static void close_sha1_file(int fd
)
2831 if (fsync_object_files
)
2832 fsync_or_die(fd
, "sha1 file");
2834 die_errno("error when closing sha1 file");
2837 /* Size of directory component, including the ending '/' */
2838 static inline int directory_size(const char *filename
)
2840 const char *s
= strrchr(filename
, '/');
2843 return s
- filename
+ 1;
2847 * This creates a temporary file in the same directory as the final
2850 * We want to avoid cross-directory filename renames, because those
2851 * can have problems on various filesystems (FAT, NFS, Coda).
2853 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2855 int fd
, dirlen
= directory_size(filename
);
2857 if (dirlen
+ 20 > bufsiz
) {
2858 errno
= ENAMETOOLONG
;
2861 memcpy(buffer
, filename
, dirlen
);
2862 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2863 fd
= git_mkstemp_mode(buffer
, 0444);
2864 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2865 /* Make sure the directory exists */
2866 memcpy(buffer
, filename
, dirlen
);
2867 buffer
[dirlen
-1] = 0;
2868 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
2870 if (adjust_shared_perm(buffer
))
2874 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2875 fd
= git_mkstemp_mode(buffer
, 0444);
2880 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2881 const void *buf
, unsigned long len
, time_t mtime
)
2884 unsigned char compressed
[4096];
2887 unsigned char parano_sha1
[20];
2888 static char tmp_file
[PATH_MAX
];
2889 const char *filename
= sha1_file_name(sha1
);
2891 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2893 if (errno
== EACCES
)
2894 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2896 return error("unable to create temporary file: %s", strerror(errno
));
2900 memset(&stream
, 0, sizeof(stream
));
2901 git_deflate_init(&stream
, zlib_compression_level
);
2902 stream
.next_out
= compressed
;
2903 stream
.avail_out
= sizeof(compressed
);
2906 /* First header.. */
2907 stream
.next_in
= (unsigned char *)hdr
;
2908 stream
.avail_in
= hdrlen
;
2909 while (git_deflate(&stream
, 0) == Z_OK
)
2911 git_SHA1_Update(&c
, hdr
, hdrlen
);
2913 /* Then the data itself.. */
2914 stream
.next_in
= (void *)buf
;
2915 stream
.avail_in
= len
;
2917 unsigned char *in0
= stream
.next_in
;
2918 ret
= git_deflate(&stream
, Z_FINISH
);
2919 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2920 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2921 die("unable to write sha1 file");
2922 stream
.next_out
= compressed
;
2923 stream
.avail_out
= sizeof(compressed
);
2924 } while (ret
== Z_OK
);
2926 if (ret
!= Z_STREAM_END
)
2927 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2928 ret
= git_deflate_end_gently(&stream
);
2930 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2931 git_SHA1_Final(parano_sha1
, &c
);
2932 if (hashcmp(sha1
, parano_sha1
) != 0)
2933 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2935 close_sha1_file(fd
);
2940 utb
.modtime
= mtime
;
2941 if (utime(tmp_file
, &utb
) < 0)
2942 warning("failed utime() on %s: %s",
2943 tmp_file
, strerror(errno
));
2946 return move_temp_to_file(tmp_file
, filename
);
2949 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2951 unsigned char sha1
[20];
2955 /* Normally if we have it in the pack then we do not bother writing
2956 * it out into .git/objects/??/?{38} file.
2958 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2960 hashcpy(returnsha1
, sha1
);
2961 if (has_sha1_file(sha1
))
2963 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2966 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2970 enum object_type type
;
2975 if (has_loose_object(sha1
))
2977 buf
= read_packed_sha1(sha1
, &type
, &len
);
2979 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2980 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2981 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2987 int has_pack_index(const unsigned char *sha1
)
2990 if (stat(sha1_pack_index_name(sha1
), &st
))
2995 int has_sha1_pack(const unsigned char *sha1
)
2997 struct pack_entry e
;
2998 return find_pack_entry(sha1
, &e
);
3001 int has_sha1_file(const unsigned char *sha1
)
3003 struct pack_entry e
;
3005 if (find_pack_entry(sha1
, &e
))
3007 if (has_loose_object(sha1
))
3009 reprepare_packed_git();
3010 return find_pack_entry(sha1
, &e
);
3013 static void check_tree(const void *buf
, size_t size
)
3015 struct tree_desc desc
;
3016 struct name_entry entry
;
3018 init_tree_desc(&desc
, buf
, size
);
3019 while (tree_entry(&desc
, &entry
))
3021 * tree_entry() will die() on malformed entries */
3025 static void check_commit(const void *buf
, size_t size
)
3028 memset(&c
, 0, sizeof(c
));
3029 if (parse_commit_buffer(&c
, buf
, size
))
3030 die("corrupt commit");
3033 static void check_tag(const void *buf
, size_t size
)
3036 memset(&t
, 0, sizeof(t
));
3037 if (parse_tag_buffer(&t
, buf
, size
))
3041 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3042 enum object_type type
,
3043 const char *path
, unsigned flags
)
3045 int ret
, re_allocated
= 0;
3046 int write_object
= flags
& HASH_WRITE_OBJECT
;
3052 * Convert blobs to git internal format
3054 if ((type
== OBJ_BLOB
) && path
) {
3055 struct strbuf nbuf
= STRBUF_INIT
;
3056 if (convert_to_git(path
, buf
, size
, &nbuf
,
3057 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3058 buf
= strbuf_detach(&nbuf
, &size
);
3062 if (flags
& HASH_FORMAT_CHECK
) {
3063 if (type
== OBJ_TREE
)
3064 check_tree(buf
, size
);
3065 if (type
== OBJ_COMMIT
)
3066 check_commit(buf
, size
);
3067 if (type
== OBJ_TAG
)
3068 check_tag(buf
, size
);
3072 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3074 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3080 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3081 const char *path
, unsigned flags
)
3083 struct strbuf sbuf
= STRBUF_INIT
;
3086 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3087 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3090 strbuf_release(&sbuf
);
3094 #define SMALL_FILE_SIZE (32*1024)
3096 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3097 enum object_type type
, const char *path
,
3103 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
3104 } else if (size
<= SMALL_FILE_SIZE
) {
3105 char *buf
= xmalloc(size
);
3106 if (size
== read_in_full(fd
, buf
, size
))
3107 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3109 ret
= error("short read %s", strerror(errno
));
3112 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3113 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3120 * This creates one packfile per large blob unless bulk-checkin
3121 * machinery is "plugged".
3123 * This also bypasses the usual "convert-to-git" dance, and that is on
3124 * purpose. We could write a streaming version of the converting
3125 * functions and insert that before feeding the data to fast-import
3126 * (or equivalent in-core API described above). However, that is
3127 * somewhat complicated, as we do not know the size of the filter
3128 * result, which we need to know beforehand when writing a git object.
3129 * Since the primary motivation for trying to stream from the working
3130 * tree file and to avoid mmaping it in core is to deal with large
3131 * binary blobs, they generally do not want to get any conversion, and
3132 * callers should avoid this code path when filters are requested.
3134 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3135 enum object_type type
, const char *path
,
3138 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3141 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3142 enum object_type type
, const char *path
, unsigned flags
)
3145 size_t size
= xsize_t(st
->st_size
);
3147 if (!S_ISREG(st
->st_mode
))
3148 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3149 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3150 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
3151 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
3153 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
3158 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3161 struct strbuf sb
= STRBUF_INIT
;
3163 switch (st
->st_mode
& S_IFMT
) {
3165 fd
= open(path
, O_RDONLY
);
3167 return error("open(\"%s\"): %s", path
,
3169 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3170 return error("%s: failed to insert into database",
3174 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3175 char *errstr
= strerror(errno
);
3176 return error("readlink(\"%s\"): %s", path
,
3179 if (!(flags
& HASH_WRITE_OBJECT
))
3180 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3181 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3182 return error("%s: failed to insert into database",
3184 strbuf_release(&sb
);
3187 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3189 return error("%s: unsupported file type", path
);
3194 int read_pack_header(int fd
, struct pack_header
*header
)
3196 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3197 /* "eof before pack header was fully read" */
3198 return PH_ERROR_EOF
;
3200 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3201 /* "protocol error (pack signature mismatch detected)" */
3202 return PH_ERROR_PACK_SIGNATURE
;
3203 if (!pack_version_ok(header
->hdr_version
))
3204 /* "protocol error (pack version unsupported)" */
3205 return PH_ERROR_PROTOCOL
;
3209 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3211 enum object_type type
= sha1_object_info(sha1
, NULL
);
3213 die("%s is not a valid object", sha1_to_hex(sha1
));
3215 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),