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
,
269 int depth
, const char *normalized_objdir
)
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 (pfxlen
== alt
->name
- alt
->base
- 1 &&
316 !memcmp(ent
->base
, alt
->base
, pfxlen
)) {
321 if (!strcmp_icase(ent
->base
, normalized_objdir
)) {
326 /* add the alternate entry */
328 alt_odb_tail
= &(ent
->next
);
331 /* recursively add alternates */
332 read_info_alternates(ent
->base
, depth
+ 1);
334 ent
->base
[pfxlen
] = '/';
339 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
340 const char *relative_base
, int depth
)
342 struct string_list entries
= STRING_LIST_INIT_NODUP
;
345 struct strbuf objdirbuf
= STRBUF_INIT
;
348 error("%s: ignoring alternate object stores, nesting too deep.",
353 strbuf_addstr(&objdirbuf
, absolute_path(get_object_directory()));
354 normalize_path_copy(objdirbuf
.buf
, objdirbuf
.buf
);
356 alt_copy
= xmemdupz(alt
, len
);
357 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
358 for (i
= 0; i
< entries
.nr
; i
++) {
359 const char *entry
= entries
.items
[i
].string
;
360 if (entry
[0] == '\0' || entry
[0] == '#')
362 if (!is_absolute_path(entry
) && depth
) {
363 error("%s: ignoring relative alternate object store %s",
364 relative_base
, entry
);
366 link_alt_odb_entry(entry
, relative_base
, depth
, objdirbuf
.buf
);
369 string_list_clear(&entries
, 0);
371 strbuf_release(&objdirbuf
);
374 void read_info_alternates(const char * relative_base
, int depth
)
379 const char alt_file_name
[] = "info/alternates";
380 /* Given that relative_base is no longer than PATH_MAX,
381 ensure that "path" has enough space to append "/", the
382 file name, "info/alternates", and a trailing NUL. */
383 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
386 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
387 fd
= git_open_noatime(path
);
390 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
394 mapsz
= xsize_t(st
.st_size
);
395 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
398 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
403 void add_to_alternates_file(const char *reference
)
405 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
406 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
407 char *alt
= mkpath("%s\n", reference
);
408 write_or_die(fd
, alt
, strlen(alt
));
409 if (commit_lock_file(lock
))
410 die("could not close alternates file");
412 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
415 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
417 struct alternate_object_database
*ent
;
420 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
425 void prepare_alt_odb(void)
432 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
435 alt_odb_tail
= &alt_odb_list
;
436 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
438 read_info_alternates(get_object_directory(), 0);
441 static int has_loose_object_local(const unsigned char *sha1
)
443 return !access(sha1_file_name(sha1
), F_OK
);
446 int has_loose_object_nonlocal(const unsigned char *sha1
)
448 struct alternate_object_database
*alt
;
450 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
451 fill_sha1_path(alt
->name
, sha1
);
452 if (!access(alt
->base
, F_OK
))
458 static int has_loose_object(const unsigned char *sha1
)
460 return has_loose_object_local(sha1
) ||
461 has_loose_object_nonlocal(sha1
);
464 static unsigned int pack_used_ctr
;
465 static unsigned int pack_mmap_calls
;
466 static unsigned int peak_pack_open_windows
;
467 static unsigned int pack_open_windows
;
468 static unsigned int pack_open_fds
;
469 static unsigned int pack_max_fds
;
470 static size_t peak_pack_mapped
;
471 static size_t pack_mapped
;
472 struct packed_git
*packed_git
;
474 void pack_report(void)
477 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
478 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
479 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
480 sz_fmt(getpagesize()),
481 sz_fmt(packed_git_window_size
),
482 sz_fmt(packed_git_limit
));
484 "pack_report: pack_used_ctr = %10u\n"
485 "pack_report: pack_mmap_calls = %10u\n"
486 "pack_report: pack_open_windows = %10u / %10u\n"
487 "pack_report: pack_mapped = "
488 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
491 pack_open_windows
, peak_pack_open_windows
,
492 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
496 * Open and mmap the index file at path, perform a couple of
497 * consistency checks, then record its information to p. Return 0 on
500 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
503 struct pack_idx_header
*hdr
;
505 uint32_t version
, nr
, i
, *index
;
506 int fd
= git_open_noatime(path
);
511 if (fstat(fd
, &st
)) {
515 idx_size
= xsize_t(st
.st_size
);
516 if (idx_size
< 4 * 256 + 20 + 20) {
518 return error("index file %s is too small", path
);
520 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
524 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
525 version
= ntohl(hdr
->idx_version
);
526 if (version
< 2 || version
> 2) {
527 munmap(idx_map
, idx_size
);
528 return error("index file %s is version %"PRIu32
529 " and is not supported by this binary"
530 " (try upgrading GIT to a newer version)",
539 index
+= 2; /* skip index header */
540 for (i
= 0; i
< 256; i
++) {
541 uint32_t n
= ntohl(index
[i
]);
543 munmap(idx_map
, idx_size
);
544 return error("non-monotonic index %s", path
);
552 * - 256 index entries 4 bytes each
553 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
554 * - 20-byte SHA1 of the packfile
555 * - 20-byte SHA1 file checksum
557 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
558 munmap(idx_map
, idx_size
);
559 return error("wrong index v1 file size in %s", path
);
561 } else if (version
== 2) {
564 * - 8 bytes of header
565 * - 256 index entries 4 bytes each
566 * - 20-byte sha1 entry * nr
567 * - 4-byte crc entry * nr
568 * - 4-byte offset entry * nr
569 * - 20-byte SHA1 of the packfile
570 * - 20-byte SHA1 file checksum
571 * And after the 4-byte offset table might be a
572 * variable sized table containing 8-byte entries
573 * for offsets larger than 2^31.
575 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
576 unsigned long max_size
= min_size
;
578 max_size
+= (nr
- 1)*8;
579 if (idx_size
< min_size
|| idx_size
> max_size
) {
580 munmap(idx_map
, idx_size
);
581 return error("wrong index v2 file size in %s", path
);
583 if (idx_size
!= min_size
&&
585 * make sure we can deal with large pack offsets.
586 * 31-bit signed offset won't be enough, neither
587 * 32-bit unsigned one will be.
589 (sizeof(off_t
) <= 4)) {
590 munmap(idx_map
, idx_size
);
591 return error("pack too large for current definition of off_t in %s", path
);
595 p
->index_version
= version
;
596 p
->index_data
= idx_map
;
597 p
->index_size
= idx_size
;
602 int open_pack_index(struct packed_git
*p
)
610 idx_name
= xstrdup(p
->pack_name
);
611 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
612 ret
= check_packed_git_idx(idx_name
, p
);
617 static void scan_windows(struct packed_git
*p
,
618 struct packed_git
**lru_p
,
619 struct pack_window
**lru_w
,
620 struct pack_window
**lru_l
)
622 struct pack_window
*w
, *w_l
;
624 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
626 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
636 static int unuse_one_window(struct packed_git
*current
)
638 struct packed_git
*p
, *lru_p
= NULL
;
639 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
642 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
643 for (p
= packed_git
; p
; p
= p
->next
)
644 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
646 munmap(lru_w
->base
, lru_w
->len
);
647 pack_mapped
-= lru_w
->len
;
649 lru_l
->next
= lru_w
->next
;
651 lru_p
->windows
= lru_w
->next
;
659 void release_pack_memory(size_t need
)
661 size_t cur
= pack_mapped
;
662 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
666 static void mmap_limit_check(size_t length
)
668 static size_t limit
= 0;
670 limit
= git_env_ulong("GIT_MMAP_LIMIT", 0);
675 die("attempting to mmap %"PRIuMAX
" over limit %"PRIuMAX
,
676 (uintmax_t)length
, (uintmax_t)limit
);
679 void *xmmap(void *start
, size_t length
,
680 int prot
, int flags
, int fd
, off_t offset
)
684 mmap_limit_check(length
);
685 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
686 if (ret
== MAP_FAILED
) {
689 release_pack_memory(length
);
690 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
691 if (ret
== MAP_FAILED
)
692 die_errno("Out of memory? mmap failed");
697 void close_pack_windows(struct packed_git
*p
)
700 struct pack_window
*w
= p
->windows
;
703 die("pack '%s' still has open windows to it",
705 munmap(w
->base
, w
->len
);
706 pack_mapped
-= w
->len
;
708 p
->windows
= w
->next
;
714 * The LRU pack is the one with the oldest MRU window, preferring packs
715 * with no used windows, or the oldest mtime if it has no windows allocated.
717 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
719 struct pack_window
*w
, *this_mru_w
;
720 int has_windows_inuse
= 0;
723 * Reject this pack if it has windows and the previously selected
724 * one does not. If this pack does not have windows, reject
725 * it if the pack file is newer than the previously selected one.
727 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
730 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
732 * Reject this pack if any of its windows are in use,
733 * but the previously selected pack did not have any
734 * inuse windows. Otherwise, record that this pack
735 * has windows in use.
738 if (*accept_windows_inuse
)
739 has_windows_inuse
= 1;
744 if (w
->last_used
> this_mru_w
->last_used
)
748 * Reject this pack if it has windows that have been
749 * used more recently than the previously selected pack.
750 * If the previously selected pack had windows inuse and
751 * we have not encountered a window in this pack that is
752 * inuse, skip this check since we prefer a pack with no
753 * inuse windows to one that has inuse windows.
755 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
756 this_mru_w
->last_used
> (*mru_w
)->last_used
)
765 *accept_windows_inuse
= has_windows_inuse
;
768 static int close_one_pack(void)
770 struct packed_git
*p
, *lru_p
= NULL
;
771 struct pack_window
*mru_w
= NULL
;
772 int accept_windows_inuse
= 1;
774 for (p
= packed_git
; p
; p
= p
->next
) {
775 if (p
->pack_fd
== -1)
777 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
781 close(lru_p
->pack_fd
);
790 void unuse_pack(struct pack_window
**w_cursor
)
792 struct pack_window
*w
= *w_cursor
;
799 void close_pack_index(struct packed_git
*p
)
802 munmap((void *)p
->index_data
, p
->index_size
);
803 p
->index_data
= NULL
;
808 * This is used by git-repack in case a newly created pack happens to
809 * contain the same set of objects as an existing one. In that case
810 * the resulting file might be different even if its name would be the
811 * same. It is best to close any reference to the old pack before it is
812 * replaced on disk. Of course no index pointers or windows for given pack
813 * must subsist at this point. If ever objects from this pack are requested
814 * again, the new version of the pack will be reinitialized through
815 * reprepare_packed_git().
817 void free_pack_by_name(const char *pack_name
)
819 struct packed_git
*p
, **pp
= &packed_git
;
823 if (strcmp(pack_name
, p
->pack_name
) == 0) {
824 clear_delta_base_cache();
825 close_pack_windows(p
);
826 if (p
->pack_fd
!= -1) {
831 free(p
->bad_object_sha1
);
833 if (last_found_pack
== p
)
834 last_found_pack
= NULL
;
842 static unsigned int get_max_fd_limit(void)
848 if (!getrlimit(RLIMIT_NOFILE
, &lim
))
855 long open_max
= sysconf(_SC_OPEN_MAX
);
859 * Otherwise, we got -1 for one of the two
862 * (1) sysconf() did not understand _SC_OPEN_MAX
863 * and signaled an error with -1; or
864 * (2) sysconf() said there is no limit.
866 * We _could_ clear errno before calling sysconf() to
867 * tell these two cases apart and return a huge number
868 * in the latter case to let the caller cap it to a
869 * value that is not so selfish, but letting the
870 * fallback OPEN_MAX codepath take care of these cases
879 return 1; /* see the caller ;-) */
884 * Do not call this directly as this leaks p->pack_fd on error return;
885 * call open_packed_git() instead.
887 static int open_packed_git_1(struct packed_git
*p
)
890 struct pack_header hdr
;
891 unsigned char sha1
[20];
892 unsigned char *idx_sha1
;
895 if (!p
->index_data
&& open_pack_index(p
))
896 return error("packfile %s index unavailable", p
->pack_name
);
899 unsigned int max_fds
= get_max_fd_limit();
901 /* Save 3 for stdin/stdout/stderr, 22 for work */
903 pack_max_fds
= max_fds
- 25;
908 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
911 p
->pack_fd
= git_open_noatime(p
->pack_name
);
912 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
916 /* If we created the struct before we had the pack we lack size. */
918 if (!S_ISREG(st
.st_mode
))
919 return error("packfile %s not a regular file", p
->pack_name
);
920 p
->pack_size
= st
.st_size
;
921 } else if (p
->pack_size
!= st
.st_size
)
922 return error("packfile %s size changed", p
->pack_name
);
924 /* We leave these file descriptors open with sliding mmap;
925 * there is no point keeping them open across exec(), though.
927 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
929 return error("cannot determine file descriptor flags");
930 fd_flag
|= FD_CLOEXEC
;
931 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
932 return error("cannot set FD_CLOEXEC");
934 /* Verify we recognize this pack file format. */
935 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
936 return error("file %s is far too short to be a packfile", p
->pack_name
);
937 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
938 return error("file %s is not a GIT packfile", p
->pack_name
);
939 if (!pack_version_ok(hdr
.hdr_version
))
940 return error("packfile %s is version %"PRIu32
" and not"
941 " supported (try upgrading GIT to a newer version)",
942 p
->pack_name
, ntohl(hdr
.hdr_version
));
944 /* Verify the pack matches its index. */
945 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
946 return error("packfile %s claims to have %"PRIu32
" objects"
947 " while index indicates %"PRIu32
" objects",
948 p
->pack_name
, ntohl(hdr
.hdr_entries
),
950 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
951 return error("end of packfile %s is unavailable", p
->pack_name
);
952 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
953 return error("packfile %s signature is unavailable", p
->pack_name
);
954 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
955 if (hashcmp(sha1
, idx_sha1
))
956 return error("packfile %s does not match index", p
->pack_name
);
960 static int open_packed_git(struct packed_git
*p
)
962 if (!open_packed_git_1(p
))
964 if (p
->pack_fd
!= -1) {
972 static int in_window(struct pack_window
*win
, off_t offset
)
974 /* We must promise at least 20 bytes (one hash) after the
975 * offset is available from this window, otherwise the offset
976 * is not actually in this window and a different window (which
977 * has that one hash excess) must be used. This is to support
978 * the object header and delta base parsing routines below.
980 off_t win_off
= win
->offset
;
981 return win_off
<= offset
982 && (offset
+ 20) <= (win_off
+ win
->len
);
985 unsigned char *use_pack(struct packed_git
*p
,
986 struct pack_window
**w_cursor
,
990 struct pack_window
*win
= *w_cursor
;
992 /* Since packfiles end in a hash of their content and it's
993 * pointless to ask for an offset into the middle of that
994 * hash, and the in_window function above wouldn't match
995 * don't allow an offset too close to the end of the file.
997 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
998 die("packfile %s cannot be accessed", p
->pack_name
);
999 if (offset
> (p
->pack_size
- 20))
1000 die("offset beyond end of packfile (truncated pack?)");
1002 if (!win
|| !in_window(win
, offset
)) {
1005 for (win
= p
->windows
; win
; win
= win
->next
) {
1006 if (in_window(win
, offset
))
1010 size_t window_align
= packed_git_window_size
/ 2;
1013 if (p
->pack_fd
== -1 && open_packed_git(p
))
1014 die("packfile %s cannot be accessed", p
->pack_name
);
1016 win
= xcalloc(1, sizeof(*win
));
1017 win
->offset
= (offset
/ window_align
) * window_align
;
1018 len
= p
->pack_size
- win
->offset
;
1019 if (len
> packed_git_window_size
)
1020 len
= packed_git_window_size
;
1021 win
->len
= (size_t)len
;
1022 pack_mapped
+= win
->len
;
1023 while (packed_git_limit
< pack_mapped
1024 && unuse_one_window(p
))
1026 win
->base
= xmmap(NULL
, win
->len
,
1027 PROT_READ
, MAP_PRIVATE
,
1028 p
->pack_fd
, win
->offset
);
1029 if (win
->base
== MAP_FAILED
)
1030 die("packfile %s cannot be mapped: %s",
1033 if (!win
->offset
&& win
->len
== p
->pack_size
1034 && !p
->do_not_close
) {
1040 pack_open_windows
++;
1041 if (pack_mapped
> peak_pack_mapped
)
1042 peak_pack_mapped
= pack_mapped
;
1043 if (pack_open_windows
> peak_pack_open_windows
)
1044 peak_pack_open_windows
= pack_open_windows
;
1045 win
->next
= p
->windows
;
1049 if (win
!= *w_cursor
) {
1050 win
->last_used
= pack_used_ctr
++;
1054 offset
-= win
->offset
;
1056 *left
= win
->len
- xsize_t(offset
);
1057 return win
->base
+ offset
;
1060 static struct packed_git
*alloc_packed_git(int extra
)
1062 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1063 memset(p
, 0, sizeof(*p
));
1068 static void try_to_free_pack_memory(size_t size
)
1070 release_pack_memory(size
);
1073 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1075 static int have_set_try_to_free_routine
;
1077 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1079 if (!have_set_try_to_free_routine
) {
1080 have_set_try_to_free_routine
= 1;
1081 set_try_to_free_routine(try_to_free_pack_memory
);
1085 * Make sure a corresponding .pack file exists and that
1086 * the index looks sane.
1088 path_len
-= strlen(".idx");
1093 memcpy(p
->pack_name
, path
, path_len
);
1095 strcpy(p
->pack_name
+ path_len
, ".keep");
1096 if (!access(p
->pack_name
, F_OK
))
1099 strcpy(p
->pack_name
+ path_len
, ".pack");
1100 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1105 /* ok, it looks sane as far as we can check without
1106 * actually mapping the pack file.
1108 p
->pack_size
= st
.st_size
;
1109 p
->pack_local
= local
;
1110 p
->mtime
= st
.st_mtime
;
1111 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1116 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1118 const char *path
= sha1_pack_name(sha1
);
1119 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1121 strcpy(p
->pack_name
, path
);
1122 hashcpy(p
->sha1
, sha1
);
1123 if (check_packed_git_idx(idx_path
, p
)) {
1131 void install_packed_git(struct packed_git
*pack
)
1133 if (pack
->pack_fd
!= -1)
1136 pack
->next
= packed_git
;
1140 void (*report_garbage
)(const char *desc
, const char *path
);
1142 static void report_helper(const struct string_list
*list
,
1143 int seen_bits
, int first
, int last
)
1146 switch (seen_bits
) {
1148 msg
= "no corresponding .idx or .pack";
1151 msg
= "no corresponding .idx";
1154 msg
= "no corresponding .pack";
1159 for (; first
< last
; first
++)
1160 report_garbage(msg
, list
->items
[first
].string
);
1163 static void report_pack_garbage(struct string_list
*list
)
1165 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1167 if (!report_garbage
)
1170 sort_string_list(list
);
1172 for (i
= 0; i
< list
->nr
; i
++) {
1173 const char *path
= list
->items
[i
].string
;
1174 if (baselen
!= -1 &&
1175 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1176 report_helper(list
, seen_bits
, first
, i
);
1180 if (baselen
== -1) {
1181 const char *dot
= strrchr(path
, '.');
1183 report_garbage("garbage found", path
);
1186 baselen
= dot
- path
+ 1;
1189 if (!strcmp(path
+ baselen
, "pack"))
1191 else if (!strcmp(path
+ baselen
, "idx"))
1194 report_helper(list
, seen_bits
, first
, list
->nr
);
1197 static void prepare_packed_git_one(char *objdir
, int local
)
1199 struct strbuf path
= STRBUF_INIT
;
1203 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1205 strbuf_addstr(&path
, objdir
);
1206 strbuf_addstr(&path
, "/pack");
1207 dir
= opendir(path
.buf
);
1209 if (errno
!= ENOENT
)
1210 error("unable to open object pack directory: %s: %s",
1211 path
.buf
, strerror(errno
));
1212 strbuf_release(&path
);
1215 strbuf_addch(&path
, '/');
1216 dirnamelen
= path
.len
;
1217 while ((de
= readdir(dir
)) != NULL
) {
1218 struct packed_git
*p
;
1221 if (is_dot_or_dotdot(de
->d_name
))
1224 strbuf_setlen(&path
, dirnamelen
);
1225 strbuf_addstr(&path
, de
->d_name
);
1227 base_len
= path
.len
;
1228 if (strip_suffix_mem(path
.buf
, &base_len
, ".idx")) {
1229 /* Don't reopen a pack we already have. */
1230 for (p
= packed_git
; p
; p
= p
->next
) {
1232 if (strip_suffix(p
->pack_name
, ".pack", &len
) &&
1234 !memcmp(p
->pack_name
, path
.buf
, len
))
1239 * See if it really is a valid .idx file with
1240 * corresponding .pack file that we can map.
1242 (p
= add_packed_git(path
.buf
, path
.len
, local
)) != NULL
)
1243 install_packed_git(p
);
1246 if (!report_garbage
)
1249 if (ends_with(de
->d_name
, ".idx") ||
1250 ends_with(de
->d_name
, ".pack") ||
1251 ends_with(de
->d_name
, ".bitmap") ||
1252 ends_with(de
->d_name
, ".keep"))
1253 string_list_append(&garbage
, path
.buf
);
1255 report_garbage("garbage found", path
.buf
);
1258 report_pack_garbage(&garbage
);
1259 string_list_clear(&garbage
, 0);
1260 strbuf_release(&path
);
1263 static int sort_pack(const void *a_
, const void *b_
)
1265 struct packed_git
*a
= *((struct packed_git
**)a_
);
1266 struct packed_git
*b
= *((struct packed_git
**)b_
);
1270 * Local packs tend to contain objects specific to our
1271 * variant of the project than remote ones. In addition,
1272 * remote ones could be on a network mounted filesystem.
1273 * Favor local ones for these reasons.
1275 st
= a
->pack_local
- b
->pack_local
;
1280 * Younger packs tend to contain more recent objects,
1281 * and more recent objects tend to get accessed more
1284 if (a
->mtime
< b
->mtime
)
1286 else if (a
->mtime
== b
->mtime
)
1291 static void rearrange_packed_git(void)
1293 struct packed_git
**ary
, *p
;
1296 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1301 /* prepare an array of packed_git for easier sorting */
1302 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1303 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1306 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1308 /* link them back again */
1309 for (i
= 0; i
< n
- 1; i
++)
1310 ary
[i
]->next
= ary
[i
+ 1];
1311 ary
[n
- 1]->next
= NULL
;
1312 packed_git
= ary
[0];
1317 static int prepare_packed_git_run_once
= 0;
1318 void prepare_packed_git(void)
1320 struct alternate_object_database
*alt
;
1322 if (prepare_packed_git_run_once
)
1324 prepare_packed_git_one(get_object_directory(), 1);
1326 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1328 prepare_packed_git_one(alt
->base
, 0);
1329 alt
->name
[-1] = '/';
1331 rearrange_packed_git();
1332 prepare_packed_git_run_once
= 1;
1335 void reprepare_packed_git(void)
1337 prepare_packed_git_run_once
= 0;
1338 prepare_packed_git();
1341 static void mark_bad_packed_object(struct packed_git
*p
,
1342 const unsigned char *sha1
)
1345 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1346 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1348 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1349 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1350 p
->num_bad_objects
++;
1353 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1355 struct packed_git
*p
;
1358 for (p
= packed_git
; p
; p
= p
->next
)
1359 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1360 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1366 * With an in-core object data in "map", rehash it to make sure the
1367 * object name actually matches "sha1" to detect object corruption.
1368 * With "map" == NULL, try reading the object named with "sha1" using
1369 * the streaming interface and rehash it to do the same.
1371 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1372 unsigned long size
, const char *type
)
1374 unsigned char real_sha1
[20];
1375 enum object_type obj_type
;
1376 struct git_istream
*st
;
1382 hash_sha1_file(map
, size
, type
, real_sha1
);
1383 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1386 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1390 /* Generate the header */
1391 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1395 git_SHA1_Update(&c
, hdr
, hdrlen
);
1397 char buf
[1024 * 16];
1398 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1406 git_SHA1_Update(&c
, buf
, readlen
);
1408 git_SHA1_Final(real_sha1
, &c
);
1410 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1413 int git_open_noatime(const char *name
)
1415 static int sha1_file_open_flag
= O_NOATIME
;
1418 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1422 /* Might the failure be due to O_NOATIME? */
1423 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1424 sha1_file_open_flag
= 0;
1432 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1434 struct alternate_object_database
*alt
;
1436 if (!lstat(sha1_file_name(sha1
), st
))
1441 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1442 fill_sha1_path(alt
->name
, sha1
);
1443 if (!lstat(alt
->base
, st
))
1450 static int open_sha1_file(const unsigned char *sha1
)
1453 struct alternate_object_database
*alt
;
1454 int most_interesting_errno
;
1456 fd
= git_open_noatime(sha1_file_name(sha1
));
1459 most_interesting_errno
= errno
;
1462 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1463 fill_sha1_path(alt
->name
, sha1
);
1464 fd
= git_open_noatime(alt
->base
);
1467 if (most_interesting_errno
== ENOENT
)
1468 most_interesting_errno
= errno
;
1470 errno
= most_interesting_errno
;
1474 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1479 fd
= open_sha1_file(sha1
);
1484 if (!fstat(fd
, &st
)) {
1485 *size
= xsize_t(st
.st_size
);
1487 /* mmap() is forbidden on empty files */
1488 error("object file %s is empty", sha1_file_name(sha1
));
1491 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1498 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1499 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1502 unsigned long size
, c
;
1503 unsigned long used
= 0;
1506 *type
= (c
>> 4) & 7;
1510 if (len
<= used
|| bitsizeof(long) <= shift
) {
1511 error("bad object header");
1516 size
+= (c
& 0x7f) << shift
;
1523 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1525 /* Get the data stream */
1526 memset(stream
, 0, sizeof(*stream
));
1527 stream
->next_in
= map
;
1528 stream
->avail_in
= mapsize
;
1529 stream
->next_out
= buffer
;
1530 stream
->avail_out
= bufsiz
;
1532 git_inflate_init(stream
);
1533 return git_inflate(stream
, 0);
1536 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1538 int bytes
= strlen(buffer
) + 1;
1539 unsigned char *buf
= xmallocz(size
);
1543 n
= stream
->total_out
- bytes
;
1546 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1548 if (bytes
<= size
) {
1550 * The above condition must be (bytes <= size), not
1551 * (bytes < size). In other words, even though we
1552 * expect no more output and set avail_out to zero,
1553 * the input zlib stream may have bytes that express
1554 * "this concludes the stream", and we *do* want to
1557 * Otherwise we would not be able to test that we
1558 * consumed all the input to reach the expected size;
1559 * we also want to check that zlib tells us that all
1560 * went well with status == Z_STREAM_END at the end.
1562 stream
->next_out
= buf
+ bytes
;
1563 stream
->avail_out
= size
- bytes
;
1564 while (status
== Z_OK
)
1565 status
= git_inflate(stream
, Z_FINISH
);
1567 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1568 git_inflate_end(stream
);
1573 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1574 else if (stream
->avail_in
)
1575 error("garbage at end of loose object '%s'",
1582 * We used to just use "sscanf()", but that's actually way
1583 * too permissive for what we want to check. So do an anal
1584 * object header parse by hand.
1586 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1593 * The type can be at most ten bytes (including the
1594 * terminating '\0' that we add), and is followed by
1603 if (i
>= sizeof(type
))
1609 * The length must follow immediately, and be in canonical
1610 * decimal format (ie "010" is not valid).
1612 size
= *hdr
++ - '0';
1617 unsigned long c
= *hdr
- '0';
1621 size
= size
* 10 + c
;
1627 * The length must be followed by a zero byte
1629 return *hdr
? -1 : type_from_string(type
);
1632 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1638 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1639 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1642 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1645 unsigned long get_size_from_delta(struct packed_git
*p
,
1646 struct pack_window
**w_curs
,
1649 const unsigned char *data
;
1650 unsigned char delta_head
[20], *in
;
1654 memset(&stream
, 0, sizeof(stream
));
1655 stream
.next_out
= delta_head
;
1656 stream
.avail_out
= sizeof(delta_head
);
1658 git_inflate_init(&stream
);
1660 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1661 stream
.next_in
= in
;
1662 st
= git_inflate(&stream
, Z_FINISH
);
1663 curpos
+= stream
.next_in
- in
;
1664 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1665 stream
.total_out
< sizeof(delta_head
));
1666 git_inflate_end(&stream
);
1667 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1668 error("delta data unpack-initial failed");
1672 /* Examine the initial part of the delta to figure out
1677 /* ignore base size */
1678 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1680 /* Read the result size */
1681 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1684 static off_t
get_delta_base(struct packed_git
*p
,
1685 struct pack_window
**w_curs
,
1687 enum object_type type
,
1688 off_t delta_obj_offset
)
1690 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1693 /* use_pack() assured us we have [base_info, base_info + 20)
1694 * as a range that we can look at without walking off the
1695 * end of the mapped window. Its actually the hash size
1696 * that is assured. An OFS_DELTA longer than the hash size
1697 * is stupid, as then a REF_DELTA would be smaller to store.
1699 if (type
== OBJ_OFS_DELTA
) {
1701 unsigned char c
= base_info
[used
++];
1702 base_offset
= c
& 127;
1705 if (!base_offset
|| MSB(base_offset
, 7))
1706 return 0; /* overflow */
1707 c
= base_info
[used
++];
1708 base_offset
= (base_offset
<< 7) + (c
& 127);
1710 base_offset
= delta_obj_offset
- base_offset
;
1711 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1712 return 0; /* out of bound */
1714 } else if (type
== OBJ_REF_DELTA
) {
1715 /* The base entry _must_ be in the same pack */
1716 base_offset
= find_pack_entry_one(base_info
, p
);
1719 die("I am totally screwed");
1724 * Like get_delta_base above, but we return the sha1 instead of the pack
1725 * offset. This means it is cheaper for REF deltas (we do not have to do
1726 * the final object lookup), but more expensive for OFS deltas (we
1727 * have to load the revidx to convert the offset back into a sha1).
1729 static const unsigned char *get_delta_base_sha1(struct packed_git
*p
,
1730 struct pack_window
**w_curs
,
1732 enum object_type type
,
1733 off_t delta_obj_offset
)
1735 if (type
== OBJ_REF_DELTA
) {
1736 unsigned char *base
= use_pack(p
, w_curs
, curpos
, NULL
);
1738 } else if (type
== OBJ_OFS_DELTA
) {
1739 struct revindex_entry
*revidx
;
1740 off_t base_offset
= get_delta_base(p
, w_curs
, &curpos
,
1741 type
, delta_obj_offset
);
1746 revidx
= find_pack_revindex(p
, base_offset
);
1750 return nth_packed_object_sha1(p
, revidx
->nr
);
1755 int unpack_object_header(struct packed_git
*p
,
1756 struct pack_window
**w_curs
,
1758 unsigned long *sizep
)
1760 unsigned char *base
;
1763 enum object_type type
;
1765 /* use_pack() assures us we have [base, base + 20) available
1766 * as a range that we can look at. (Its actually the hash
1767 * size that is assured.) With our object header encoding
1768 * the maximum deflated object size is 2^137, which is just
1769 * insane, so we know won't exceed what we have been given.
1771 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1772 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1781 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1784 struct revindex_entry
*revidx
;
1785 const unsigned char *sha1
;
1786 revidx
= find_pack_revindex(p
, obj_offset
);
1789 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1790 mark_bad_packed_object(p
, sha1
);
1791 type
= sha1_object_info(sha1
, NULL
);
1792 if (type
<= OBJ_NONE
)
1797 #define POI_STACK_PREALLOC 64
1799 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1801 enum object_type type
,
1802 struct pack_window
**w_curs
,
1805 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1806 off_t
*poi_stack
= small_poi_stack
;
1807 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1809 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1812 /* Push the object we're going to leave behind */
1813 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1814 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1815 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1816 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1818 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1820 poi_stack
[poi_stack_nr
++] = obj_offset
;
1821 /* If parsing the base offset fails, just unwind */
1822 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1825 curpos
= obj_offset
= base_offset
;
1826 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1827 if (type
<= OBJ_NONE
) {
1828 /* If getting the base itself fails, we first
1829 * retry the base, otherwise unwind */
1830 type
= retry_bad_packed_offset(p
, base_offset
);
1831 if (type
> OBJ_NONE
)
1845 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1846 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1851 if (poi_stack
!= small_poi_stack
)
1856 while (poi_stack_nr
) {
1857 obj_offset
= poi_stack
[--poi_stack_nr
];
1858 type
= retry_bad_packed_offset(p
, obj_offset
);
1859 if (type
> OBJ_NONE
)
1866 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1867 struct object_info
*oi
)
1869 struct pack_window
*w_curs
= NULL
;
1871 off_t curpos
= obj_offset
;
1872 enum object_type type
;
1875 * We always get the representation type, but only convert it to
1876 * a "real" type later if the caller is interested.
1878 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1881 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1882 off_t tmp_pos
= curpos
;
1883 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1889 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1890 if (*oi
->sizep
== 0) {
1899 if (oi
->disk_sizep
) {
1900 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1901 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1905 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1906 if (*oi
->typep
< 0) {
1912 if (oi
->delta_base_sha1
) {
1913 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1914 const unsigned char *base
;
1916 base
= get_delta_base_sha1(p
, &w_curs
, curpos
,
1923 hashcpy(oi
->delta_base_sha1
, base
);
1925 hashclr(oi
->delta_base_sha1
);
1929 unuse_pack(&w_curs
);
1933 static void *unpack_compressed_entry(struct packed_git
*p
,
1934 struct pack_window
**w_curs
,
1940 unsigned char *buffer
, *in
;
1942 buffer
= xmallocz(size
);
1943 memset(&stream
, 0, sizeof(stream
));
1944 stream
.next_out
= buffer
;
1945 stream
.avail_out
= size
+ 1;
1947 git_inflate_init(&stream
);
1949 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1950 stream
.next_in
= in
;
1951 st
= git_inflate(&stream
, Z_FINISH
);
1952 if (!stream
.avail_out
)
1953 break; /* the payload is larger than it should be */
1954 curpos
+= stream
.next_in
- in
;
1955 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1956 git_inflate_end(&stream
);
1957 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1965 #define MAX_DELTA_CACHE (256)
1967 static size_t delta_base_cached
;
1969 static struct delta_base_cache_lru_list
{
1970 struct delta_base_cache_lru_list
*prev
;
1971 struct delta_base_cache_lru_list
*next
;
1972 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1974 static struct delta_base_cache_entry
{
1975 struct delta_base_cache_lru_list lru
;
1977 struct packed_git
*p
;
1980 enum object_type type
;
1981 } delta_base_cache
[MAX_DELTA_CACHE
];
1983 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1987 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1988 hash
+= (hash
>> 8) + (hash
>> 16);
1989 return hash
% MAX_DELTA_CACHE
;
1992 static struct delta_base_cache_entry
*
1993 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1995 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1996 return delta_base_cache
+ hash
;
1999 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
2000 struct packed_git
*p
, off_t base_offset
)
2002 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
2005 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
2007 struct delta_base_cache_entry
*ent
;
2008 ent
= get_delta_base_cache_entry(p
, base_offset
);
2009 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
2012 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
2015 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2016 ent
->lru
.prev
->next
= ent
->lru
.next
;
2017 delta_base_cached
-= ent
->size
;
2020 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
2021 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
2023 struct delta_base_cache_entry
*ent
;
2026 ent
= get_delta_base_cache_entry(p
, base_offset
);
2028 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2029 return unpack_entry(p
, base_offset
, type
, base_size
);
2034 clear_delta_base_cache_entry(ent
);
2036 ret
= xmemdupz(ent
->data
, ent
->size
);
2038 *base_size
= ent
->size
;
2042 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2047 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2048 ent
->lru
.prev
->next
= ent
->lru
.next
;
2049 delta_base_cached
-= ent
->size
;
2053 void clear_delta_base_cache(void)
2056 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2057 release_delta_base_cache(&delta_base_cache
[p
]);
2060 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2061 void *base
, unsigned long base_size
, enum object_type type
)
2063 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2064 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2065 struct delta_base_cache_lru_list
*lru
;
2067 release_delta_base_cache(ent
);
2068 delta_base_cached
+= base_size
;
2070 for (lru
= delta_base_cache_lru
.next
;
2071 delta_base_cached
> delta_base_cache_limit
2072 && lru
!= &delta_base_cache_lru
;
2074 struct delta_base_cache_entry
*f
= (void *)lru
;
2075 if (f
->type
== OBJ_BLOB
)
2076 release_delta_base_cache(f
);
2078 for (lru
= delta_base_cache_lru
.next
;
2079 delta_base_cached
> delta_base_cache_limit
2080 && lru
!= &delta_base_cache_lru
;
2082 struct delta_base_cache_entry
*f
= (void *)lru
;
2083 release_delta_base_cache(f
);
2087 ent
->base_offset
= base_offset
;
2090 ent
->size
= base_size
;
2091 ent
->lru
.next
= &delta_base_cache_lru
;
2092 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2093 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2094 delta_base_cache_lru
.prev
= &ent
->lru
;
2097 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2098 unsigned long *size
);
2100 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2102 static struct trace_key pack_access
= TRACE_KEY_INIT(PACK_ACCESS
);
2103 trace_printf_key(&pack_access
, "%s %"PRIuMAX
"\n",
2104 p
->pack_name
, (uintmax_t)obj_offset
);
2107 int do_check_packed_object_crc
;
2109 #define UNPACK_ENTRY_STACK_PREALLOC 64
2110 struct unpack_entry_stack_ent
{
2116 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2117 enum object_type
*final_type
, unsigned long *final_size
)
2119 struct pack_window
*w_curs
= NULL
;
2120 off_t curpos
= obj_offset
;
2123 enum object_type type
;
2124 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2125 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2126 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2127 int base_from_cache
= 0;
2129 write_pack_access_log(p
, obj_offset
);
2131 /* PHASE 1: drill down to the innermost base object */
2135 struct delta_base_cache_entry
*ent
;
2137 ent
= get_delta_base_cache_entry(p
, curpos
);
2138 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2142 clear_delta_base_cache_entry(ent
);
2143 base_from_cache
= 1;
2147 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2148 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2149 unsigned long len
= revidx
[1].offset
- obj_offset
;
2150 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2151 const unsigned char *sha1
=
2152 nth_packed_object_sha1(p
, revidx
->nr
);
2153 error("bad packed object CRC for %s",
2155 mark_bad_packed_object(p
, sha1
);
2156 unuse_pack(&w_curs
);
2161 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2162 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2165 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2167 error("failed to validate delta base reference "
2168 "at offset %"PRIuMAX
" from %s",
2169 (uintmax_t)curpos
, p
->pack_name
);
2170 /* bail to phase 2, in hopes of recovery */
2175 /* push object, proceed to base */
2176 if (delta_stack_nr
>= delta_stack_alloc
2177 && delta_stack
== small_delta_stack
) {
2178 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2179 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2180 memcpy(delta_stack
, small_delta_stack
,
2181 sizeof(*delta_stack
)*delta_stack_nr
);
2183 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2185 i
= delta_stack_nr
++;
2186 delta_stack
[i
].obj_offset
= obj_offset
;
2187 delta_stack
[i
].curpos
= curpos
;
2188 delta_stack
[i
].size
= size
;
2190 curpos
= obj_offset
= base_offset
;
2193 /* PHASE 2: handle the base */
2198 die("BUG in unpack_entry: left loop at a valid delta");
2204 if (!base_from_cache
)
2205 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2209 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2210 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2213 /* PHASE 3: apply deltas in order */
2216 * 'data' holds the base data, or NULL if there was corruption
2218 while (delta_stack_nr
) {
2221 unsigned long delta_size
, base_size
= size
;
2227 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2231 * We're probably in deep shit, but let's try to fetch
2232 * the required base anyway from another pack or loose.
2233 * This is costly but should happen only in the presence
2234 * of a corrupted pack, and is better than failing outright.
2236 struct revindex_entry
*revidx
;
2237 const unsigned char *base_sha1
;
2238 revidx
= find_pack_revindex(p
, obj_offset
);
2240 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2241 error("failed to read delta base object %s"
2242 " at offset %"PRIuMAX
" from %s",
2243 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2245 mark_bad_packed_object(p
, base_sha1
);
2246 base
= read_object(base_sha1
, &type
, &base_size
);
2250 i
= --delta_stack_nr
;
2251 obj_offset
= delta_stack
[i
].obj_offset
;
2252 curpos
= delta_stack
[i
].curpos
;
2253 delta_size
= delta_stack
[i
].size
;
2258 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2261 error("failed to unpack compressed delta "
2262 "at offset %"PRIuMAX
" from %s",
2263 (uintmax_t)curpos
, p
->pack_name
);
2268 data
= patch_delta(base
, base_size
,
2269 delta_data
, delta_size
,
2273 * We could not apply the delta; warn the user, but keep going.
2274 * Our failure will be noticed either in the next iteration of
2275 * the loop, or if this is the final delta, in the caller when
2276 * we return NULL. Those code paths will take care of making
2277 * a more explicit warning and retrying with another copy of
2281 error("failed to apply delta");
2289 unuse_pack(&w_curs
);
2291 if (delta_stack
!= small_delta_stack
)
2297 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2300 const unsigned char *index
= p
->index_data
;
2302 if (open_pack_index(p
))
2304 index
= p
->index_data
;
2306 if (n
>= p
->num_objects
)
2309 if (p
->index_version
== 1) {
2310 return index
+ 24 * n
+ 4;
2313 return index
+ 20 * n
;
2317 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2319 const unsigned char *index
= p
->index_data
;
2321 if (p
->index_version
== 1) {
2322 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2325 index
+= 8 + p
->num_objects
* (20 + 4);
2326 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2327 if (!(off
& 0x80000000))
2329 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2330 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2331 ntohl(*((uint32_t *)(index
+ 4)));
2335 off_t
find_pack_entry_one(const unsigned char *sha1
,
2336 struct packed_git
*p
)
2338 const uint32_t *level1_ofs
= p
->index_data
;
2339 const unsigned char *index
= p
->index_data
;
2340 unsigned hi
, lo
, stride
;
2341 static int use_lookup
= -1;
2342 static int debug_lookup
= -1;
2344 if (debug_lookup
< 0)
2345 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2348 if (open_pack_index(p
))
2350 level1_ofs
= p
->index_data
;
2351 index
= p
->index_data
;
2353 if (p
->index_version
> 1) {
2358 hi
= ntohl(level1_ofs
[*sha1
]);
2359 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2360 if (p
->index_version
> 1) {
2368 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2369 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2372 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2374 int pos
= sha1_entry_pos(index
, stride
, 0,
2375 lo
, hi
, p
->num_objects
, sha1
);
2378 return nth_packed_object_offset(p
, pos
);
2382 unsigned mi
= (lo
+ hi
) / 2;
2383 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2386 printf("lo %u hi %u rg %u mi %u\n",
2387 lo
, hi
, hi
- lo
, mi
);
2389 return nth_packed_object_offset(p
, mi
);
2398 int is_pack_valid(struct packed_git
*p
)
2400 /* An already open pack is known to be valid. */
2401 if (p
->pack_fd
!= -1)
2404 /* If the pack has one window completely covering the
2405 * file size, the pack is known to be valid even if
2406 * the descriptor is not currently open.
2409 struct pack_window
*w
= p
->windows
;
2411 if (!w
->offset
&& w
->len
== p
->pack_size
)
2415 /* Force the pack to open to prove its valid. */
2416 return !open_packed_git(p
);
2419 static int fill_pack_entry(const unsigned char *sha1
,
2420 struct pack_entry
*e
,
2421 struct packed_git
*p
)
2425 if (p
->num_bad_objects
) {
2427 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2428 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2432 offset
= find_pack_entry_one(sha1
, p
);
2437 * We are about to tell the caller where they can locate the
2438 * requested object. We better make sure the packfile is
2439 * still here and can be accessed before supplying that
2440 * answer, as it may have been deleted since the index was
2443 if (!is_pack_valid(p
)) {
2444 warning("packfile %s cannot be accessed", p
->pack_name
);
2449 hashcpy(e
->sha1
, sha1
);
2454 * Iff a pack file contains the object named by sha1, return true and
2455 * store its location to e.
2457 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2459 struct packed_git
*p
;
2461 prepare_packed_git();
2465 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2468 for (p
= packed_git
; p
; p
= p
->next
) {
2469 if (p
== last_found_pack
)
2470 continue; /* we already checked this one */
2472 if (fill_pack_entry(sha1
, e
, p
)) {
2473 last_found_pack
= p
;
2480 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2481 struct packed_git
*packs
)
2483 struct packed_git
*p
;
2485 for (p
= packs
; p
; p
= p
->next
) {
2486 if (find_pack_entry_one(sha1
, p
))
2493 static int sha1_loose_object_info(const unsigned char *sha1
,
2494 struct object_info
*oi
)
2497 unsigned long mapsize
, size
;
2502 if (oi
->delta_base_sha1
)
2503 hashclr(oi
->delta_base_sha1
);
2506 * If we don't care about type or size, then we don't
2507 * need to look inside the object at all. Note that we
2508 * do not optimize out the stat call, even if the
2509 * caller doesn't care about the disk-size, since our
2510 * return value implicitly indicates whether the
2511 * object even exists.
2513 if (!oi
->typep
&& !oi
->sizep
) {
2515 if (stat_sha1_file(sha1
, &st
) < 0)
2518 *oi
->disk_sizep
= st
.st_size
;
2522 map
= map_sha1_file(sha1
, &mapsize
);
2526 *oi
->disk_sizep
= mapsize
;
2527 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2528 status
= error("unable to unpack %s header",
2530 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2531 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2534 git_inflate_end(&stream
);
2535 munmap(map
, mapsize
);
2537 *oi
->typep
= status
;
2541 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
, unsigned flags
)
2543 struct cached_object
*co
;
2544 struct pack_entry e
;
2546 const unsigned char *real
= lookup_replace_object_extended(sha1
, flags
);
2548 co
= find_cached_object(real
);
2551 *(oi
->typep
) = co
->type
;
2553 *(oi
->sizep
) = co
->size
;
2555 *(oi
->disk_sizep
) = 0;
2556 if (oi
->delta_base_sha1
)
2557 hashclr(oi
->delta_base_sha1
);
2558 oi
->whence
= OI_CACHED
;
2562 if (!find_pack_entry(real
, &e
)) {
2563 /* Most likely it's a loose object. */
2564 if (!sha1_loose_object_info(real
, oi
)) {
2565 oi
->whence
= OI_LOOSE
;
2569 /* Not a loose object; someone else may have just packed it. */
2570 reprepare_packed_git();
2571 if (!find_pack_entry(real
, &e
))
2575 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2577 mark_bad_packed_object(e
.p
, real
);
2578 return sha1_object_info_extended(real
, oi
, 0);
2579 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2580 oi
->whence
= OI_DBCACHED
;
2582 oi
->whence
= OI_PACKED
;
2583 oi
->u
.packed
.offset
= e
.offset
;
2584 oi
->u
.packed
.pack
= e
.p
;
2585 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2586 rtype
== OBJ_OFS_DELTA
);
2592 /* returns enum object_type or negative */
2593 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2595 enum object_type type
;
2596 struct object_info oi
= {NULL
};
2600 if (sha1_object_info_extended(sha1
, &oi
, LOOKUP_REPLACE_OBJECT
) < 0)
2605 static void *read_packed_sha1(const unsigned char *sha1
,
2606 enum object_type
*type
, unsigned long *size
)
2608 struct pack_entry e
;
2611 if (!find_pack_entry(sha1
, &e
))
2613 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2616 * We're probably in deep shit, but let's try to fetch
2617 * the required object anyway from another pack or loose.
2618 * This should happen only in the presence of a corrupted
2619 * pack, and is better than failing outright.
2621 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2622 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2623 mark_bad_packed_object(e
.p
, sha1
);
2624 data
= read_object(sha1
, type
, size
);
2629 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2630 unsigned char *sha1
)
2632 struct cached_object
*co
;
2634 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2635 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2637 ALLOC_GROW(cached_objects
, cached_object_nr
+ 1, cached_object_alloc
);
2638 co
= &cached_objects
[cached_object_nr
++];
2641 co
->buf
= xmalloc(len
);
2642 memcpy(co
->buf
, buf
, len
);
2643 hashcpy(co
->sha1
, sha1
);
2647 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2648 unsigned long *size
)
2650 unsigned long mapsize
;
2652 struct cached_object
*co
;
2654 co
= find_cached_object(sha1
);
2658 return xmemdupz(co
->buf
, co
->size
);
2661 buf
= read_packed_sha1(sha1
, type
, size
);
2664 map
= map_sha1_file(sha1
, &mapsize
);
2666 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2667 munmap(map
, mapsize
);
2670 reprepare_packed_git();
2671 return read_packed_sha1(sha1
, type
, size
);
2675 * This function dies on corrupt objects; the callers who want to
2676 * deal with them should arrange to call read_object() and give error
2677 * messages themselves.
2679 void *read_sha1_file_extended(const unsigned char *sha1
,
2680 enum object_type
*type
,
2681 unsigned long *size
,
2685 const struct packed_git
*p
;
2686 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2689 data
= read_object(repl
, type
, size
);
2693 if (errno
&& errno
!= ENOENT
)
2694 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2696 /* die if we replaced an object with one that does not exist */
2698 die("replacement %s not found for %s",
2699 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2701 if (has_loose_object(repl
)) {
2702 const char *path
= sha1_file_name(sha1
);
2704 die("loose object %s (stored in %s) is corrupt",
2705 sha1_to_hex(repl
), path
);
2708 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2709 die("packed object %s (stored in %s) is corrupt",
2710 sha1_to_hex(repl
), p
->pack_name
);
2715 void *read_object_with_reference(const unsigned char *sha1
,
2716 const char *required_type_name
,
2717 unsigned long *size
,
2718 unsigned char *actual_sha1_return
)
2720 enum object_type type
, required_type
;
2722 unsigned long isize
;
2723 unsigned char actual_sha1
[20];
2725 required_type
= type_from_string(required_type_name
);
2726 hashcpy(actual_sha1
, sha1
);
2728 int ref_length
= -1;
2729 const char *ref_type
= NULL
;
2731 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2734 if (type
== required_type
) {
2736 if (actual_sha1_return
)
2737 hashcpy(actual_sha1_return
, actual_sha1
);
2740 /* Handle references */
2741 else if (type
== OBJ_COMMIT
)
2743 else if (type
== OBJ_TAG
)
2744 ref_type
= "object ";
2749 ref_length
= strlen(ref_type
);
2751 if (ref_length
+ 40 > isize
||
2752 memcmp(buffer
, ref_type
, ref_length
) ||
2753 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2758 /* Now we have the ID of the referred-to object in
2759 * actual_sha1. Check again. */
2763 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2764 const char *type
, unsigned char *sha1
,
2765 char *hdr
, int *hdrlen
)
2769 /* Generate the header */
2770 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2774 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2775 git_SHA1_Update(&c
, buf
, len
);
2776 git_SHA1_Final(sha1
, &c
);
2780 * Move the just written object into its final resting place.
2781 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2782 * "moving" is only a part of what it does, when no patch between
2783 * master to pu changes the call sites of this function.
2785 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2789 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2791 else if (link(tmpfile
, filename
))
2795 * Coda hack - coda doesn't like cross-directory links,
2796 * so we fall back to a rename, which will mean that it
2797 * won't be able to check collisions, but that's not a
2800 * The same holds for FAT formatted media.
2802 * When this succeeds, we just return. We have nothing
2805 if (ret
&& ret
!= EEXIST
) {
2807 if (!rename(tmpfile
, filename
))
2811 unlink_or_warn(tmpfile
);
2813 if (ret
!= EEXIST
) {
2814 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2816 /* FIXME!!! Collision check here ? */
2820 if (adjust_shared_perm(filename
))
2821 return error("unable to set permission to '%s'", filename
);
2825 static int write_buffer(int fd
, const void *buf
, size_t len
)
2827 if (write_in_full(fd
, buf
, len
) < 0)
2828 return error("file write error (%s)", strerror(errno
));
2832 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2833 unsigned char *sha1
)
2837 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2841 /* Finalize a file on disk, and close it. */
2842 static void close_sha1_file(int fd
)
2844 if (fsync_object_files
)
2845 fsync_or_die(fd
, "sha1 file");
2847 die_errno("error when closing sha1 file");
2850 /* Size of directory component, including the ending '/' */
2851 static inline int directory_size(const char *filename
)
2853 const char *s
= strrchr(filename
, '/');
2856 return s
- filename
+ 1;
2860 * This creates a temporary file in the same directory as the final
2863 * We want to avoid cross-directory filename renames, because those
2864 * can have problems on various filesystems (FAT, NFS, Coda).
2866 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2868 int fd
, dirlen
= directory_size(filename
);
2870 if (dirlen
+ 20 > bufsiz
) {
2871 errno
= ENAMETOOLONG
;
2874 memcpy(buffer
, filename
, dirlen
);
2875 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2876 fd
= git_mkstemp_mode(buffer
, 0444);
2877 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2878 /* Make sure the directory exists */
2879 memcpy(buffer
, filename
, dirlen
);
2880 buffer
[dirlen
-1] = 0;
2881 if (mkdir(buffer
, 0777) && errno
!= EEXIST
)
2883 if (adjust_shared_perm(buffer
))
2887 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2888 fd
= git_mkstemp_mode(buffer
, 0444);
2893 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2894 const void *buf
, unsigned long len
, time_t mtime
)
2897 unsigned char compressed
[4096];
2900 unsigned char parano_sha1
[20];
2901 static char tmp_file
[PATH_MAX
];
2902 const char *filename
= sha1_file_name(sha1
);
2904 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2906 if (errno
== EACCES
)
2907 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2909 return error("unable to create temporary file: %s", strerror(errno
));
2913 memset(&stream
, 0, sizeof(stream
));
2914 git_deflate_init(&stream
, zlib_compression_level
);
2915 stream
.next_out
= compressed
;
2916 stream
.avail_out
= sizeof(compressed
);
2919 /* First header.. */
2920 stream
.next_in
= (unsigned char *)hdr
;
2921 stream
.avail_in
= hdrlen
;
2922 while (git_deflate(&stream
, 0) == Z_OK
)
2924 git_SHA1_Update(&c
, hdr
, hdrlen
);
2926 /* Then the data itself.. */
2927 stream
.next_in
= (void *)buf
;
2928 stream
.avail_in
= len
;
2930 unsigned char *in0
= stream
.next_in
;
2931 ret
= git_deflate(&stream
, Z_FINISH
);
2932 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2933 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2934 die("unable to write sha1 file");
2935 stream
.next_out
= compressed
;
2936 stream
.avail_out
= sizeof(compressed
);
2937 } while (ret
== Z_OK
);
2939 if (ret
!= Z_STREAM_END
)
2940 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2941 ret
= git_deflate_end_gently(&stream
);
2943 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2944 git_SHA1_Final(parano_sha1
, &c
);
2945 if (hashcmp(sha1
, parano_sha1
) != 0)
2946 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2948 close_sha1_file(fd
);
2953 utb
.modtime
= mtime
;
2954 if (utime(tmp_file
, &utb
) < 0)
2955 warning("failed utime() on %s: %s",
2956 tmp_file
, strerror(errno
));
2959 return move_temp_to_file(tmp_file
, filename
);
2962 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2964 unsigned char sha1
[20];
2968 /* Normally if we have it in the pack then we do not bother writing
2969 * it out into .git/objects/??/?{38} file.
2971 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2973 hashcpy(returnsha1
, sha1
);
2974 if (has_sha1_file(sha1
))
2976 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2979 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2983 enum object_type type
;
2988 if (has_loose_object(sha1
))
2990 buf
= read_packed_sha1(sha1
, &type
, &len
);
2992 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2993 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2994 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
3000 int has_pack_index(const unsigned char *sha1
)
3003 if (stat(sha1_pack_index_name(sha1
), &st
))
3008 int has_sha1_pack(const unsigned char *sha1
)
3010 struct pack_entry e
;
3011 return find_pack_entry(sha1
, &e
);
3014 int has_sha1_file(const unsigned char *sha1
)
3016 struct pack_entry e
;
3018 if (find_pack_entry(sha1
, &e
))
3020 if (has_loose_object(sha1
))
3022 reprepare_packed_git();
3023 return find_pack_entry(sha1
, &e
);
3026 static void check_tree(const void *buf
, size_t size
)
3028 struct tree_desc desc
;
3029 struct name_entry entry
;
3031 init_tree_desc(&desc
, buf
, size
);
3032 while (tree_entry(&desc
, &entry
))
3034 * tree_entry() will die() on malformed entries */
3038 static void check_commit(const void *buf
, size_t size
)
3041 memset(&c
, 0, sizeof(c
));
3042 if (parse_commit_buffer(&c
, buf
, size
))
3043 die("corrupt commit");
3046 static void check_tag(const void *buf
, size_t size
)
3049 memset(&t
, 0, sizeof(t
));
3050 if (parse_tag_buffer(&t
, buf
, size
))
3054 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3055 enum object_type type
,
3056 const char *path
, unsigned flags
)
3058 int ret
, re_allocated
= 0;
3059 int write_object
= flags
& HASH_WRITE_OBJECT
;
3065 * Convert blobs to git internal format
3067 if ((type
== OBJ_BLOB
) && path
) {
3068 struct strbuf nbuf
= STRBUF_INIT
;
3069 if (convert_to_git(path
, buf
, size
, &nbuf
,
3070 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3071 buf
= strbuf_detach(&nbuf
, &size
);
3075 if (flags
& HASH_FORMAT_CHECK
) {
3076 if (type
== OBJ_TREE
)
3077 check_tree(buf
, size
);
3078 if (type
== OBJ_COMMIT
)
3079 check_commit(buf
, size
);
3080 if (type
== OBJ_TAG
)
3081 check_tag(buf
, size
);
3085 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3087 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3093 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3094 const char *path
, unsigned flags
)
3096 struct strbuf sbuf
= STRBUF_INIT
;
3099 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3100 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3103 strbuf_release(&sbuf
);
3107 #define SMALL_FILE_SIZE (32*1024)
3109 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3110 enum object_type type
, const char *path
,
3116 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
3117 } else if (size
<= SMALL_FILE_SIZE
) {
3118 char *buf
= xmalloc(size
);
3119 if (size
== read_in_full(fd
, buf
, size
))
3120 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3122 ret
= error("short read %s", strerror(errno
));
3125 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3126 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3133 * This creates one packfile per large blob unless bulk-checkin
3134 * machinery is "plugged".
3136 * This also bypasses the usual "convert-to-git" dance, and that is on
3137 * purpose. We could write a streaming version of the converting
3138 * functions and insert that before feeding the data to fast-import
3139 * (or equivalent in-core API described above). However, that is
3140 * somewhat complicated, as we do not know the size of the filter
3141 * result, which we need to know beforehand when writing a git object.
3142 * Since the primary motivation for trying to stream from the working
3143 * tree file and to avoid mmaping it in core is to deal with large
3144 * binary blobs, they generally do not want to get any conversion, and
3145 * callers should avoid this code path when filters are requested.
3147 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3148 enum object_type type
, const char *path
,
3151 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3154 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3155 enum object_type type
, const char *path
, unsigned flags
)
3158 size_t size
= xsize_t(st
->st_size
);
3160 if (!S_ISREG(st
->st_mode
))
3161 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3162 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3163 (path
&& would_convert_to_git(path
)))
3164 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
3166 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
3171 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3174 struct strbuf sb
= STRBUF_INIT
;
3176 switch (st
->st_mode
& S_IFMT
) {
3178 fd
= open(path
, O_RDONLY
);
3180 return error("open(\"%s\"): %s", path
,
3182 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3183 return error("%s: failed to insert into database",
3187 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3188 char *errstr
= strerror(errno
);
3189 return error("readlink(\"%s\"): %s", path
,
3192 if (!(flags
& HASH_WRITE_OBJECT
))
3193 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3194 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3195 return error("%s: failed to insert into database",
3197 strbuf_release(&sb
);
3200 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3202 return error("%s: unsupported file type", path
);
3207 int read_pack_header(int fd
, struct pack_header
*header
)
3209 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3210 /* "eof before pack header was fully read" */
3211 return PH_ERROR_EOF
;
3213 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3214 /* "protocol error (pack signature mismatch detected)" */
3215 return PH_ERROR_PACK_SIGNATURE
;
3216 if (!pack_version_ok(header
->hdr_version
))
3217 /* "protocol error (pack version unsupported)" */
3218 return PH_ERROR_PROTOCOL
;
3222 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3224 enum object_type type
= sha1_object_info(sha1
, NULL
);
3226 die("%s is not a valid object", sha1_to_hex(sha1
));
3228 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),