12 #include "streaming.h"
13 #include "thread-utils.h"
15 static const char index_pack_usage
[] =
16 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
19 struct pack_idx_entry idx
;
21 unsigned int hdr_size
;
22 enum object_type type
;
23 enum object_type real_type
;
29 unsigned char sha1
[20];
34 struct base_data
*base
;
35 struct base_data
*child
;
36 struct object_entry
*obj
;
39 int ref_first
, ref_last
;
40 int ofs_first
, ofs_last
;
43 #if !defined(NO_PTHREADS) && defined(NO_THREAD_SAFE_PREAD)
44 /* pread() emulation is not thread-safe. Disable threading. */
52 struct base_data
*base_cache
;
53 size_t base_cache_used
;
57 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
58 * to memcmp() only the first 20 bytes.
60 #define UNION_BASE_SZ 20
62 #define FLAG_LINK (1u<<20)
63 #define FLAG_CHECKED (1u<<21)
66 union delta_base base
;
70 static struct object_entry
*objects
;
71 static struct delta_entry
*deltas
;
72 static struct thread_local nothread_data
;
73 static int nr_objects
;
75 static int nr_resolved_deltas
;
76 static int nr_threads
;
78 static int from_stdin
;
80 static int do_fsck_object
;
83 static int check_self_contained_and_connected
;
85 static struct progress
*progress
;
87 /* We always read in 4kB chunks. */
88 static unsigned char input_buffer
[4096];
89 static unsigned int input_offset
, input_len
;
90 static off_t consumed_bytes
;
91 static unsigned deepest_delta
;
92 static git_SHA_CTX input_ctx
;
93 static uint32_t input_crc32
;
94 static int input_fd
, output_fd
, pack_fd
;
98 static struct thread_local
*thread_data
;
99 static int nr_dispatched
;
100 static int threads_active
;
102 static pthread_mutex_t read_mutex
;
103 #define read_lock() lock_mutex(&read_mutex)
104 #define read_unlock() unlock_mutex(&read_mutex)
106 static pthread_mutex_t counter_mutex
;
107 #define counter_lock() lock_mutex(&counter_mutex)
108 #define counter_unlock() unlock_mutex(&counter_mutex)
110 static pthread_mutex_t work_mutex
;
111 #define work_lock() lock_mutex(&work_mutex)
112 #define work_unlock() unlock_mutex(&work_mutex)
114 static pthread_mutex_t deepest_delta_mutex
;
115 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
116 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
118 static pthread_key_t key
;
120 static inline void lock_mutex(pthread_mutex_t
*mutex
)
123 pthread_mutex_lock(mutex
);
126 static inline void unlock_mutex(pthread_mutex_t
*mutex
)
129 pthread_mutex_unlock(mutex
);
133 * Mutex and conditional variable can't be statically-initialized on Windows.
135 static void init_thread(void)
137 init_recursive_mutex(&read_mutex
);
138 pthread_mutex_init(&counter_mutex
, NULL
);
139 pthread_mutex_init(&work_mutex
, NULL
);
141 pthread_mutex_init(&deepest_delta_mutex
, NULL
);
142 pthread_key_create(&key
, NULL
);
143 thread_data
= xcalloc(nr_threads
, sizeof(*thread_data
));
147 static void cleanup_thread(void)
152 pthread_mutex_destroy(&read_mutex
);
153 pthread_mutex_destroy(&counter_mutex
);
154 pthread_mutex_destroy(&work_mutex
);
156 pthread_mutex_destroy(&deepest_delta_mutex
);
157 pthread_key_delete(key
);
164 #define read_unlock()
166 #define counter_lock()
167 #define counter_unlock()
170 #define work_unlock()
172 #define deepest_delta_lock()
173 #define deepest_delta_unlock()
178 static int mark_link(struct object
*obj
, int type
, void *data
)
183 if (type
!= OBJ_ANY
&& obj
->type
!= type
)
184 die(_("object type mismatch at %s"), sha1_to_hex(obj
->sha1
));
186 obj
->flags
|= FLAG_LINK
;
190 /* The content of each linked object must have been checked
191 or it must be already present in the object database */
192 static unsigned check_object(struct object
*obj
)
197 if (!(obj
->flags
& FLAG_LINK
))
200 if (!(obj
->flags
& FLAG_CHECKED
)) {
202 int type
= sha1_object_info(obj
->sha1
, &size
);
204 die(_("did not receive expected object %s"),
205 sha1_to_hex(obj
->sha1
));
206 if (type
!= obj
->type
)
207 die(_("object %s: expected type %s, found %s"),
208 sha1_to_hex(obj
->sha1
),
209 typename(obj
->type
), typename(type
));
210 obj
->flags
|= FLAG_CHECKED
;
217 static unsigned check_objects(void)
219 unsigned i
, max
, foreign_nr
= 0;
221 max
= get_max_object_index();
222 for (i
= 0; i
< max
; i
++)
223 foreign_nr
+= check_object(get_indexed_object(i
));
228 /* Discard current buffer used content. */
229 static void flush(void)
233 write_or_die(output_fd
, input_buffer
, input_offset
);
234 git_SHA1_Update(&input_ctx
, input_buffer
, input_offset
);
235 memmove(input_buffer
, input_buffer
+ input_offset
, input_len
);
241 * Make sure at least "min" bytes are available in the buffer, and
242 * return the pointer to the buffer.
244 static void *fill(int min
)
246 if (min
<= input_len
)
247 return input_buffer
+ input_offset
;
248 if (min
> sizeof(input_buffer
))
249 die(Q_("cannot fill %d byte",
250 "cannot fill %d bytes",
255 ssize_t ret
= xread(input_fd
, input_buffer
+ input_len
,
256 sizeof(input_buffer
) - input_len
);
260 die_errno(_("read error on input"));
264 display_throughput(progress
, consumed_bytes
+ input_len
);
265 } while (input_len
< min
);
269 static void use(int bytes
)
271 if (bytes
> input_len
)
272 die(_("used more bytes than were available"));
273 input_crc32
= crc32(input_crc32
, input_buffer
+ input_offset
, bytes
);
275 input_offset
+= bytes
;
277 /* make sure off_t is sufficiently large not to wrap */
278 if (signed_add_overflows(consumed_bytes
, bytes
))
279 die(_("pack too large for current definition of off_t"));
280 consumed_bytes
+= bytes
;
283 static const char *open_pack_file(const char *pack_name
)
288 static char tmp_file
[PATH_MAX
];
289 output_fd
= odb_mkstemp(tmp_file
, sizeof(tmp_file
),
290 "pack/tmp_pack_XXXXXX");
291 pack_name
= xstrdup(tmp_file
);
293 output_fd
= open(pack_name
, O_CREAT
|O_EXCL
|O_RDWR
, 0600);
295 die_errno(_("unable to create '%s'"), pack_name
);
298 input_fd
= open(pack_name
, O_RDONLY
);
300 die_errno(_("cannot open packfile '%s'"), pack_name
);
304 git_SHA1_Init(&input_ctx
);
308 static void parse_pack_header(void)
310 struct pack_header
*hdr
= fill(sizeof(struct pack_header
));
312 /* Header consistency check */
313 if (hdr
->hdr_signature
!= htonl(PACK_SIGNATURE
))
314 die(_("pack signature mismatch"));
315 if (!pack_version_ok(hdr
->hdr_version
))
316 die(_("pack version %"PRIu32
" unsupported"),
317 ntohl(hdr
->hdr_version
));
319 nr_objects
= ntohl(hdr
->hdr_entries
);
320 use(sizeof(struct pack_header
));
323 static NORETURN
void bad_object(unsigned long offset
, const char *format
,
324 ...) __attribute__((format (printf
, 2, 3)));
326 static NORETURN
void bad_object(unsigned long offset
, const char *format
, ...)
331 va_start(params
, format
);
332 vsnprintf(buf
, sizeof(buf
), format
, params
);
334 die(_("pack has bad object at offset %lu: %s"), offset
, buf
);
337 static inline struct thread_local
*get_thread_data(void)
341 return pthread_getspecific(key
);
342 assert(!threads_active
&&
343 "This should only be reached when all threads are gone");
345 return ¬hread_data
;
349 static void set_thread_data(struct thread_local
*data
)
352 pthread_setspecific(key
, data
);
356 static struct base_data
*alloc_base_data(void)
358 struct base_data
*base
= xmalloc(sizeof(struct base_data
));
359 memset(base
, 0, sizeof(*base
));
365 static void free_base_data(struct base_data
*c
)
370 get_thread_data()->base_cache_used
-= c
->size
;
374 static void prune_base_data(struct base_data
*retain
)
377 struct thread_local
*data
= get_thread_data();
378 for (b
= data
->base_cache
;
379 data
->base_cache_used
> delta_base_cache_limit
&& b
;
381 if (b
->data
&& b
!= retain
)
386 static void link_base_data(struct base_data
*base
, struct base_data
*c
)
391 get_thread_data()->base_cache
= c
;
396 get_thread_data()->base_cache_used
+= c
->size
;
400 static void unlink_base_data(struct base_data
*c
)
402 struct base_data
*base
= c
->base
;
406 get_thread_data()->base_cache
= NULL
;
410 static int is_delta_type(enum object_type type
)
412 return (type
== OBJ_REF_DELTA
|| type
== OBJ_OFS_DELTA
);
415 static void *unpack_entry_data(unsigned long offset
, unsigned long size
,
416 enum object_type type
, unsigned char *sha1
)
418 static char fixed_buf
[8192];
426 if (!is_delta_type(type
)) {
427 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), size
) + 1;
429 git_SHA1_Update(&c
, hdr
, hdrlen
);
432 if (type
== OBJ_BLOB
&& size
> big_file_threshold
)
437 memset(&stream
, 0, sizeof(stream
));
438 git_inflate_init(&stream
);
439 stream
.next_out
= buf
;
440 stream
.avail_out
= buf
== fixed_buf
? sizeof(fixed_buf
) : size
;
443 unsigned char *last_out
= stream
.next_out
;
444 stream
.next_in
= fill(1);
445 stream
.avail_in
= input_len
;
446 status
= git_inflate(&stream
, 0);
447 use(input_len
- stream
.avail_in
);
449 git_SHA1_Update(&c
, last_out
, stream
.next_out
- last_out
);
450 if (buf
== fixed_buf
) {
451 stream
.next_out
= buf
;
452 stream
.avail_out
= sizeof(fixed_buf
);
454 } while (status
== Z_OK
);
455 if (stream
.total_out
!= size
|| status
!= Z_STREAM_END
)
456 bad_object(offset
, _("inflate returned %d"), status
);
457 git_inflate_end(&stream
);
459 git_SHA1_Final(sha1
, &c
);
460 return buf
== fixed_buf
? NULL
: buf
;
463 static void *unpack_raw_entry(struct object_entry
*obj
,
464 union delta_base
*delta_base
,
468 unsigned long size
, c
;
473 obj
->idx
.offset
= consumed_bytes
;
474 input_crc32
= crc32(0, NULL
, 0);
479 obj
->type
= (c
>> 4) & 7;
486 size
+= (c
& 0x7f) << shift
;
493 hashcpy(delta_base
->sha1
, fill(20));
497 memset(delta_base
, 0, sizeof(*delta_base
));
501 base_offset
= c
& 127;
504 if (!base_offset
|| MSB(base_offset
, 7))
505 bad_object(obj
->idx
.offset
, _("offset value overflow for delta base object"));
509 base_offset
= (base_offset
<< 7) + (c
& 127);
511 delta_base
->offset
= obj
->idx
.offset
- base_offset
;
512 if (delta_base
->offset
<= 0 || delta_base
->offset
>= obj
->idx
.offset
)
513 bad_object(obj
->idx
.offset
, _("delta base offset is out of bound"));
521 bad_object(obj
->idx
.offset
, _("unknown object type %d"), obj
->type
);
523 obj
->hdr_size
= consumed_bytes
- obj
->idx
.offset
;
525 data
= unpack_entry_data(obj
->idx
.offset
, obj
->size
, obj
->type
, sha1
);
526 obj
->idx
.crc32
= input_crc32
;
530 static void *unpack_data(struct object_entry
*obj
,
531 int (*consume
)(const unsigned char *, unsigned long, void *),
534 off_t from
= obj
[0].idx
.offset
+ obj
[0].hdr_size
;
535 unsigned long len
= obj
[1].idx
.offset
- from
;
536 unsigned char *data
, *inbuf
;
540 data
= xmalloc(consume
? 64*1024 : obj
->size
);
541 inbuf
= xmalloc((len
< 64*1024) ? len
: 64*1024);
543 memset(&stream
, 0, sizeof(stream
));
544 git_inflate_init(&stream
);
545 stream
.next_out
= data
;
546 stream
.avail_out
= consume
? 64*1024 : obj
->size
;
549 ssize_t n
= (len
< 64*1024) ? len
: 64*1024;
550 n
= pread(pack_fd
, inbuf
, n
, from
);
552 die_errno(_("cannot pread pack file"));
554 die(Q_("premature end of pack file, %lu byte missing",
555 "premature end of pack file, %lu bytes missing",
560 stream
.next_in
= inbuf
;
563 status
= git_inflate(&stream
, 0);
566 status
= git_inflate(&stream
, 0);
567 if (consume(data
, stream
.next_out
- data
, cb_data
)) {
572 stream
.next_out
= data
;
573 stream
.avail_out
= 64*1024;
574 } while (status
== Z_OK
&& stream
.avail_in
);
576 } while (len
&& status
== Z_OK
&& !stream
.avail_in
);
578 /* This has been inflated OK when first encountered, so... */
579 if (status
!= Z_STREAM_END
|| stream
.total_out
!= obj
->size
)
580 die(_("serious inflate inconsistency"));
582 git_inflate_end(&stream
);
591 static void *get_data_from_pack(struct object_entry
*obj
)
593 return unpack_data(obj
, NULL
, NULL
);
596 static int compare_delta_bases(const union delta_base
*base1
,
597 const union delta_base
*base2
,
598 enum object_type type1
,
599 enum object_type type2
)
601 int cmp
= type1
- type2
;
604 return memcmp(base1
, base2
, UNION_BASE_SZ
);
607 static int find_delta(const union delta_base
*base
, enum object_type type
)
609 int first
= 0, last
= nr_deltas
;
611 while (first
< last
) {
612 int next
= (first
+ last
) / 2;
613 struct delta_entry
*delta
= &deltas
[next
];
616 cmp
= compare_delta_bases(base
, &delta
->base
,
617 type
, objects
[delta
->obj_no
].type
);
629 static void find_delta_children(const union delta_base
*base
,
630 int *first_index
, int *last_index
,
631 enum object_type type
)
633 int first
= find_delta(base
, type
);
635 int end
= nr_deltas
- 1;
642 while (first
> 0 && !memcmp(&deltas
[first
- 1].base
, base
, UNION_BASE_SZ
))
644 while (last
< end
&& !memcmp(&deltas
[last
+ 1].base
, base
, UNION_BASE_SZ
))
646 *first_index
= first
;
650 struct compare_data
{
651 struct object_entry
*entry
;
652 struct git_istream
*st
;
654 unsigned long buf_size
;
657 static int compare_objects(const unsigned char *buf
, unsigned long size
,
660 struct compare_data
*data
= cb_data
;
662 if (data
->buf_size
< size
) {
664 data
->buf
= xmalloc(size
);
665 data
->buf_size
= size
;
669 ssize_t len
= read_istream(data
->st
, data
->buf
, size
);
671 die(_("SHA1 COLLISION FOUND WITH %s !"),
672 sha1_to_hex(data
->entry
->idx
.sha1
));
674 die(_("unable to read %s"),
675 sha1_to_hex(data
->entry
->idx
.sha1
));
676 if (memcmp(buf
, data
->buf
, len
))
677 die(_("SHA1 COLLISION FOUND WITH %s !"),
678 sha1_to_hex(data
->entry
->idx
.sha1
));
685 static int check_collison(struct object_entry
*entry
)
687 struct compare_data data
;
688 enum object_type type
;
691 if (entry
->size
<= big_file_threshold
|| entry
->type
!= OBJ_BLOB
)
694 memset(&data
, 0, sizeof(data
));
696 data
.st
= open_istream(entry
->idx
.sha1
, &type
, &size
, NULL
);
699 if (size
!= entry
->size
|| type
!= entry
->type
)
700 die(_("SHA1 COLLISION FOUND WITH %s !"),
701 sha1_to_hex(entry
->idx
.sha1
));
702 unpack_data(entry
, compare_objects
, &data
);
703 close_istream(data
.st
);
708 static void sha1_object(const void *data
, struct object_entry
*obj_entry
,
709 unsigned long size
, enum object_type type
,
710 const unsigned char *sha1
)
712 void *new_data
= NULL
;
713 int collision_test_needed
;
715 assert(data
|| obj_entry
);
718 collision_test_needed
= has_sha1_file(sha1
);
721 if (collision_test_needed
&& !data
) {
723 if (!check_collison(obj_entry
))
724 collision_test_needed
= 0;
727 if (collision_test_needed
) {
729 enum object_type has_type
;
730 unsigned long has_size
;
732 has_type
= sha1_object_info(sha1
, &has_size
);
733 if (has_type
!= type
|| has_size
!= size
)
734 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1
));
735 has_data
= read_sha1_file(sha1
, &has_type
, &has_size
);
738 data
= new_data
= get_data_from_pack(obj_entry
);
740 die(_("cannot read existing object %s"), sha1_to_hex(sha1
));
741 if (size
!= has_size
|| type
!= has_type
||
742 memcmp(data
, has_data
, size
) != 0)
743 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1
));
749 if (type
== OBJ_BLOB
) {
750 struct blob
*blob
= lookup_blob(sha1
);
752 blob
->object
.flags
|= FLAG_CHECKED
;
754 die(_("invalid blob object %s"), sha1_to_hex(sha1
));
758 void *buf
= (void *) data
;
760 assert(data
&& "data can only be NULL for large _blobs_");
763 * we do not need to free the memory here, as the
764 * buf is deleted by the caller.
766 obj
= parse_object_buffer(sha1
, type
, size
, buf
, &eaten
);
768 die(_("invalid %s"), typename(type
));
769 if (do_fsck_object
&&
770 fsck_object(obj
, 1, fsck_error_function
))
771 die(_("Error in object"));
772 if (fsck_walk(obj
, mark_link
, NULL
))
773 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj
->sha1
));
775 if (obj
->type
== OBJ_TREE
) {
776 struct tree
*item
= (struct tree
*) obj
;
780 if (obj
->type
== OBJ_COMMIT
) {
781 struct commit
*commit
= (struct commit
*) obj
;
782 commit
->buffer
= NULL
;
784 obj
->flags
|= FLAG_CHECKED
;
793 * This function is part of find_unresolved_deltas(). There are two
794 * walkers going in the opposite ways.
796 * The first one in find_unresolved_deltas() traverses down from
797 * parent node to children, deflating nodes along the way. However,
798 * memory for deflated nodes is limited by delta_base_cache_limit, so
799 * at some point parent node's deflated content may be freed.
801 * The second walker is this function, which goes from current node up
802 * to top parent if necessary to deflate the node. In normal
803 * situation, its parent node would be already deflated, so it just
804 * needs to apply delta.
806 * In the worst case scenario, parent node is no longer deflated because
807 * we're running out of delta_base_cache_limit; we need to re-deflate
808 * parents, possibly up to the top base.
810 * All deflated objects here are subject to be freed if we exceed
811 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
812 * just need to make sure the last node is not freed.
814 static void *get_base_data(struct base_data
*c
)
817 struct object_entry
*obj
= c
->obj
;
818 struct base_data
**delta
= NULL
;
819 int delta_nr
= 0, delta_alloc
= 0;
821 while (is_delta_type(c
->obj
->type
) && !c
->data
) {
822 ALLOC_GROW(delta
, delta_nr
+ 1, delta_alloc
);
823 delta
[delta_nr
++] = c
;
827 c
->data
= get_data_from_pack(obj
);
829 get_thread_data()->base_cache_used
+= c
->size
;
832 for (; delta_nr
> 0; delta_nr
--) {
834 c
= delta
[delta_nr
- 1];
836 base
= get_base_data(c
->base
);
837 raw
= get_data_from_pack(obj
);
838 c
->data
= patch_delta(
844 bad_object(obj
->idx
.offset
, _("failed to apply delta"));
845 get_thread_data()->base_cache_used
+= c
->size
;
853 static void resolve_delta(struct object_entry
*delta_obj
,
854 struct base_data
*base
, struct base_data
*result
)
856 void *base_data
, *delta_data
;
858 delta_obj
->real_type
= base
->obj
->real_type
;
860 delta_obj
->delta_depth
= base
->obj
->delta_depth
+ 1;
861 deepest_delta_lock();
862 if (deepest_delta
< delta_obj
->delta_depth
)
863 deepest_delta
= delta_obj
->delta_depth
;
864 deepest_delta_unlock();
866 delta_obj
->base_object_no
= base
->obj
- objects
;
867 delta_data
= get_data_from_pack(delta_obj
);
868 base_data
= get_base_data(base
);
869 result
->obj
= delta_obj
;
870 result
->data
= patch_delta(base_data
, base
->size
,
871 delta_data
, delta_obj
->size
, &result
->size
);
874 bad_object(delta_obj
->idx
.offset
, _("failed to apply delta"));
875 hash_sha1_file(result
->data
, result
->size
,
876 typename(delta_obj
->real_type
), delta_obj
->idx
.sha1
);
877 sha1_object(result
->data
, NULL
, result
->size
, delta_obj
->real_type
,
878 delta_obj
->idx
.sha1
);
880 nr_resolved_deltas
++;
884 static struct base_data
*find_unresolved_deltas_1(struct base_data
*base
,
885 struct base_data
*prev_base
)
887 if (base
->ref_last
== -1 && base
->ofs_last
== -1) {
888 union delta_base base_spec
;
890 hashcpy(base_spec
.sha1
, base
->obj
->idx
.sha1
);
891 find_delta_children(&base_spec
,
892 &base
->ref_first
, &base
->ref_last
, OBJ_REF_DELTA
);
894 memset(&base_spec
, 0, sizeof(base_spec
));
895 base_spec
.offset
= base
->obj
->idx
.offset
;
896 find_delta_children(&base_spec
,
897 &base
->ofs_first
, &base
->ofs_last
, OBJ_OFS_DELTA
);
899 if (base
->ref_last
== -1 && base
->ofs_last
== -1) {
904 link_base_data(prev_base
, base
);
907 if (base
->ref_first
<= base
->ref_last
) {
908 struct object_entry
*child
= objects
+ deltas
[base
->ref_first
].obj_no
;
909 struct base_data
*result
= alloc_base_data();
911 assert(child
->real_type
== OBJ_REF_DELTA
);
912 resolve_delta(child
, base
, result
);
913 if (base
->ref_first
== base
->ref_last
&& base
->ofs_last
== -1)
914 free_base_data(base
);
920 if (base
->ofs_first
<= base
->ofs_last
) {
921 struct object_entry
*child
= objects
+ deltas
[base
->ofs_first
].obj_no
;
922 struct base_data
*result
= alloc_base_data();
924 assert(child
->real_type
== OBJ_OFS_DELTA
);
925 resolve_delta(child
, base
, result
);
926 if (base
->ofs_first
== base
->ofs_last
)
927 free_base_data(base
);
933 unlink_base_data(base
);
937 static void find_unresolved_deltas(struct base_data
*base
)
939 struct base_data
*new_base
, *prev_base
= NULL
;
941 new_base
= find_unresolved_deltas_1(base
, prev_base
);
951 prev_base
= base
->base
;
956 static int compare_delta_entry(const void *a
, const void *b
)
958 const struct delta_entry
*delta_a
= a
;
959 const struct delta_entry
*delta_b
= b
;
961 /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
962 return compare_delta_bases(&delta_a
->base
, &delta_b
->base
,
963 objects
[delta_a
->obj_no
].type
,
964 objects
[delta_b
->obj_no
].type
);
967 static void resolve_base(struct object_entry
*obj
)
969 struct base_data
*base_obj
= alloc_base_data();
971 base_obj
->data
= NULL
;
972 find_unresolved_deltas(base_obj
);
976 static void *threaded_second_pass(void *data
)
978 set_thread_data(data
);
982 display_progress(progress
, nr_resolved_deltas
);
985 while (nr_dispatched
< nr_objects
&&
986 is_delta_type(objects
[nr_dispatched
].type
))
988 if (nr_dispatched
>= nr_objects
) {
995 resolve_base(&objects
[i
]);
1003 * - find locations of all objects;
1004 * - calculate SHA1 of all non-delta objects;
1005 * - remember base (SHA1 or offset) for all deltas.
1007 static void parse_pack_objects(unsigned char *sha1
)
1009 int i
, nr_delays
= 0;
1010 struct delta_entry
*delta
= deltas
;
1014 progress
= start_progress(
1015 from_stdin
? _("Receiving objects") : _("Indexing objects"),
1017 for (i
= 0; i
< nr_objects
; i
++) {
1018 struct object_entry
*obj
= &objects
[i
];
1019 void *data
= unpack_raw_entry(obj
, &delta
->base
, obj
->idx
.sha1
);
1020 obj
->real_type
= obj
->type
;
1021 if (is_delta_type(obj
->type
)) {
1026 /* large blobs, check later */
1027 obj
->real_type
= OBJ_BAD
;
1030 sha1_object(data
, NULL
, obj
->size
, obj
->type
, obj
->idx
.sha1
);
1032 display_progress(progress
, i
+1);
1034 objects
[i
].idx
.offset
= consumed_bytes
;
1035 stop_progress(&progress
);
1037 /* Check pack integrity */
1039 git_SHA1_Final(sha1
, &input_ctx
);
1040 if (hashcmp(fill(20), sha1
))
1041 die(_("pack is corrupted (SHA1 mismatch)"));
1044 /* If input_fd is a file, we should have reached its end now. */
1045 if (fstat(input_fd
, &st
))
1046 die_errno(_("cannot fstat packfile"));
1047 if (S_ISREG(st
.st_mode
) &&
1048 lseek(input_fd
, 0, SEEK_CUR
) - input_len
!= st
.st_size
)
1049 die(_("pack has junk at the end"));
1051 for (i
= 0; i
< nr_objects
; i
++) {
1052 struct object_entry
*obj
= &objects
[i
];
1053 if (obj
->real_type
!= OBJ_BAD
)
1055 obj
->real_type
= obj
->type
;
1056 sha1_object(NULL
, obj
, obj
->size
, obj
->type
, obj
->idx
.sha1
);
1060 die(_("confusion beyond insanity in parse_pack_objects()"));
1065 * - for all non-delta objects, look if it is used as a base for
1067 * - if used as a base, uncompress the object and apply all deltas,
1068 * recursively checking if the resulting object is used as a base
1069 * for some more deltas.
1071 static void resolve_deltas(void)
1078 /* Sort deltas by base SHA1/offset for fast searching */
1079 qsort(deltas
, nr_deltas
, sizeof(struct delta_entry
),
1080 compare_delta_entry
);
1083 progress
= start_progress(_("Resolving deltas"), nr_deltas
);
1087 if (nr_threads
> 1 || getenv("GIT_FORCE_THREADS")) {
1089 for (i
= 0; i
< nr_threads
; i
++) {
1090 int ret
= pthread_create(&thread_data
[i
].thread
, NULL
,
1091 threaded_second_pass
, thread_data
+ i
);
1093 die(_("unable to create thread: %s"),
1096 for (i
= 0; i
< nr_threads
; i
++)
1097 pthread_join(thread_data
[i
].thread
, NULL
);
1103 for (i
= 0; i
< nr_objects
; i
++) {
1104 struct object_entry
*obj
= &objects
[i
];
1106 if (is_delta_type(obj
->type
))
1109 display_progress(progress
, nr_resolved_deltas
);
1115 * - append objects to convert thin pack to full pack if required
1116 * - write the final 20-byte SHA-1
1118 static void fix_unresolved_deltas(struct sha1file
*f
, int nr_unresolved
);
1119 static void conclude_pack(int fix_thin_pack
, const char *curr_pack
, unsigned char *pack_sha1
)
1121 if (nr_deltas
== nr_resolved_deltas
) {
1122 stop_progress(&progress
);
1123 /* Flush remaining pack final 20-byte SHA1. */
1128 if (fix_thin_pack
) {
1130 unsigned char read_sha1
[20], tail_sha1
[20];
1131 struct strbuf msg
= STRBUF_INIT
;
1132 int nr_unresolved
= nr_deltas
- nr_resolved_deltas
;
1133 int nr_objects_initial
= nr_objects
;
1134 if (nr_unresolved
<= 0)
1135 die(_("confusion beyond insanity"));
1136 objects
= xrealloc(objects
,
1137 (nr_objects
+ nr_unresolved
+ 1)
1138 * sizeof(*objects
));
1139 memset(objects
+ nr_objects
+ 1, 0,
1140 nr_unresolved
* sizeof(*objects
));
1141 f
= sha1fd(output_fd
, curr_pack
);
1142 fix_unresolved_deltas(f
, nr_unresolved
);
1143 strbuf_addf(&msg
, _("completed with %d local objects"),
1144 nr_objects
- nr_objects_initial
);
1145 stop_progress_msg(&progress
, msg
.buf
);
1146 strbuf_release(&msg
);
1147 sha1close(f
, tail_sha1
, 0);
1148 hashcpy(read_sha1
, pack_sha1
);
1149 fixup_pack_header_footer(output_fd
, pack_sha1
,
1150 curr_pack
, nr_objects
,
1151 read_sha1
, consumed_bytes
-20);
1152 if (hashcmp(read_sha1
, tail_sha1
) != 0)
1153 die(_("Unexpected tail checksum for %s "
1154 "(disk corruption?)"), curr_pack
);
1156 if (nr_deltas
!= nr_resolved_deltas
)
1157 die(Q_("pack has %d unresolved delta",
1158 "pack has %d unresolved deltas",
1159 nr_deltas
- nr_resolved_deltas
),
1160 nr_deltas
- nr_resolved_deltas
);
1163 static int write_compressed(struct sha1file
*f
, void *in
, unsigned int size
)
1167 unsigned char outbuf
[4096];
1169 memset(&stream
, 0, sizeof(stream
));
1170 git_deflate_init(&stream
, zlib_compression_level
);
1171 stream
.next_in
= in
;
1172 stream
.avail_in
= size
;
1175 stream
.next_out
= outbuf
;
1176 stream
.avail_out
= sizeof(outbuf
);
1177 status
= git_deflate(&stream
, Z_FINISH
);
1178 sha1write(f
, outbuf
, sizeof(outbuf
) - stream
.avail_out
);
1179 } while (status
== Z_OK
);
1181 if (status
!= Z_STREAM_END
)
1182 die(_("unable to deflate appended object (%d)"), status
);
1183 size
= stream
.total_out
;
1184 git_deflate_end(&stream
);
1188 static struct object_entry
*append_obj_to_pack(struct sha1file
*f
,
1189 const unsigned char *sha1
, void *buf
,
1190 unsigned long size
, enum object_type type
)
1192 struct object_entry
*obj
= &objects
[nr_objects
++];
1193 unsigned char header
[10];
1194 unsigned long s
= size
;
1196 unsigned char c
= (type
<< 4) | (s
& 15);
1199 header
[n
++] = c
| 0x80;
1205 sha1write(f
, header
, n
);
1207 obj
[0].hdr_size
= n
;
1209 obj
[0].real_type
= type
;
1210 obj
[1].idx
.offset
= obj
[0].idx
.offset
+ n
;
1211 obj
[1].idx
.offset
+= write_compressed(f
, buf
, size
);
1212 obj
[0].idx
.crc32
= crc32_end(f
);
1214 hashcpy(obj
->idx
.sha1
, sha1
);
1218 static int delta_pos_compare(const void *_a
, const void *_b
)
1220 struct delta_entry
*a
= *(struct delta_entry
**)_a
;
1221 struct delta_entry
*b
= *(struct delta_entry
**)_b
;
1222 return a
->obj_no
- b
->obj_no
;
1225 static void fix_unresolved_deltas(struct sha1file
*f
, int nr_unresolved
)
1227 struct delta_entry
**sorted_by_pos
;
1231 * Since many unresolved deltas may well be themselves base objects
1232 * for more unresolved deltas, we really want to include the
1233 * smallest number of base objects that would cover as much delta
1234 * as possible by picking the
1235 * trunc deltas first, allowing for other deltas to resolve without
1236 * additional base objects. Since most base objects are to be found
1237 * before deltas depending on them, a good heuristic is to start
1238 * resolving deltas in the same order as their position in the pack.
1240 sorted_by_pos
= xmalloc(nr_unresolved
* sizeof(*sorted_by_pos
));
1241 for (i
= 0; i
< nr_deltas
; i
++) {
1242 if (objects
[deltas
[i
].obj_no
].real_type
!= OBJ_REF_DELTA
)
1244 sorted_by_pos
[n
++] = &deltas
[i
];
1246 qsort(sorted_by_pos
, n
, sizeof(*sorted_by_pos
), delta_pos_compare
);
1248 for (i
= 0; i
< n
; i
++) {
1249 struct delta_entry
*d
= sorted_by_pos
[i
];
1250 enum object_type type
;
1251 struct base_data
*base_obj
= alloc_base_data();
1253 if (objects
[d
->obj_no
].real_type
!= OBJ_REF_DELTA
)
1255 base_obj
->data
= read_sha1_file(d
->base
.sha1
, &type
, &base_obj
->size
);
1256 if (!base_obj
->data
)
1259 if (check_sha1_signature(d
->base
.sha1
, base_obj
->data
,
1260 base_obj
->size
, typename(type
)))
1261 die(_("local object %s is corrupt"), sha1_to_hex(d
->base
.sha1
));
1262 base_obj
->obj
= append_obj_to_pack(f
, d
->base
.sha1
,
1263 base_obj
->data
, base_obj
->size
, type
);
1264 find_unresolved_deltas(base_obj
);
1265 display_progress(progress
, nr_resolved_deltas
);
1267 free(sorted_by_pos
);
1270 static void final(const char *final_pack_name
, const char *curr_pack_name
,
1271 const char *final_index_name
, const char *curr_index_name
,
1272 const char *keep_name
, const char *keep_msg
,
1273 unsigned char *sha1
)
1275 const char *report
= "pack";
1276 char name
[PATH_MAX
];
1282 fsync_or_die(output_fd
, curr_pack_name
);
1283 err
= close(output_fd
);
1285 die_errno(_("error while closing pack file"));
1289 int keep_fd
, keep_msg_len
= strlen(keep_msg
);
1292 keep_fd
= odb_pack_keep(name
, sizeof(name
), sha1
);
1294 keep_fd
= open(keep_name
, O_RDWR
|O_CREAT
|O_EXCL
, 0600);
1297 if (errno
!= EEXIST
)
1298 die_errno(_("cannot write keep file '%s'"),
1299 keep_name
? keep_name
: name
);
1301 if (keep_msg_len
> 0) {
1302 write_or_die(keep_fd
, keep_msg
, keep_msg_len
);
1303 write_or_die(keep_fd
, "\n", 1);
1305 if (close(keep_fd
) != 0)
1306 die_errno(_("cannot close written keep file '%s'"),
1307 keep_name
? keep_name
: name
);
1312 if (final_pack_name
!= curr_pack_name
) {
1313 if (!final_pack_name
) {
1314 snprintf(name
, sizeof(name
), "%s/pack/pack-%s.pack",
1315 get_object_directory(), sha1_to_hex(sha1
));
1316 final_pack_name
= name
;
1318 if (move_temp_to_file(curr_pack_name
, final_pack_name
))
1319 die(_("cannot store pack file"));
1320 } else if (from_stdin
)
1321 chmod(final_pack_name
, 0444);
1323 if (final_index_name
!= curr_index_name
) {
1324 if (!final_index_name
) {
1325 snprintf(name
, sizeof(name
), "%s/pack/pack-%s.idx",
1326 get_object_directory(), sha1_to_hex(sha1
));
1327 final_index_name
= name
;
1329 if (move_temp_to_file(curr_index_name
, final_index_name
))
1330 die(_("cannot store index file"));
1332 chmod(final_index_name
, 0444);
1335 printf("%s\n", sha1_to_hex(sha1
));
1338 int len
= snprintf(buf
, sizeof(buf
), "%s\t%s\n",
1339 report
, sha1_to_hex(sha1
));
1340 write_or_die(1, buf
, len
);
1343 * Let's just mimic git-unpack-objects here and write
1344 * the last part of the input buffer to stdout.
1347 err
= xwrite(1, input_buffer
+ input_offset
, input_len
);
1351 input_offset
+= err
;
1356 static int git_index_pack_config(const char *k
, const char *v
, void *cb
)
1358 struct pack_idx_option
*opts
= cb
;
1360 if (!strcmp(k
, "pack.indexversion")) {
1361 opts
->version
= git_config_int(k
, v
);
1362 if (opts
->version
> 2)
1363 die(_("bad pack.indexversion=%"PRIu32
), opts
->version
);
1366 if (!strcmp(k
, "pack.threads")) {
1367 nr_threads
= git_config_int(k
, v
);
1369 die(_("invalid number of threads specified (%d)"),
1372 if (nr_threads
!= 1)
1373 warning(_("no threads support, ignoring %s"), k
);
1378 return git_default_config(k
, v
, cb
);
1381 static int cmp_uint32(const void *a_
, const void *b_
)
1383 uint32_t a
= *((uint32_t *)a_
);
1384 uint32_t b
= *((uint32_t *)b_
);
1386 return (a
< b
) ? -1 : (a
!= b
);
1389 static void read_v2_anomalous_offsets(struct packed_git
*p
,
1390 struct pack_idx_option
*opts
)
1392 const uint32_t *idx1
, *idx2
;
1395 /* The address of the 4-byte offset table */
1396 idx1
= (((const uint32_t *)p
->index_data
)
1397 + 2 /* 8-byte header */
1399 + 5 * p
->num_objects
/* 20-byte SHA-1 table */
1400 + p
->num_objects
/* CRC32 table */
1403 /* The address of the 8-byte offset table */
1404 idx2
= idx1
+ p
->num_objects
;
1406 for (i
= 0; i
< p
->num_objects
; i
++) {
1407 uint32_t off
= ntohl(idx1
[i
]);
1408 if (!(off
& 0x80000000))
1410 off
= off
& 0x7fffffff;
1414 * The real offset is ntohl(idx2[off * 2]) in high 4
1415 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1416 * octets. But idx2[off * 2] is Zero!!!
1418 ALLOC_GROW(opts
->anomaly
, opts
->anomaly_nr
+ 1, opts
->anomaly_alloc
);
1419 opts
->anomaly
[opts
->anomaly_nr
++] = ntohl(idx2
[off
* 2 + 1]);
1422 if (1 < opts
->anomaly_nr
)
1423 qsort(opts
->anomaly
, opts
->anomaly_nr
, sizeof(uint32_t), cmp_uint32
);
1426 static void read_idx_option(struct pack_idx_option
*opts
, const char *pack_name
)
1428 struct packed_git
*p
= add_packed_git(pack_name
, strlen(pack_name
), 1);
1431 die(_("Cannot open existing pack file '%s'"), pack_name
);
1432 if (open_pack_index(p
))
1433 die(_("Cannot open existing pack idx file for '%s'"), pack_name
);
1435 /* Read the attributes from the existing idx file */
1436 opts
->version
= p
->index_version
;
1438 if (opts
->version
== 2)
1439 read_v2_anomalous_offsets(p
, opts
);
1442 * Get rid of the idx file as we do not need it anymore.
1443 * NEEDSWORK: extract this bit from free_pack_by_name() in
1444 * sha1_file.c, perhaps? It shouldn't matter very much as we
1445 * know we haven't installed this pack (hence we never have
1446 * read anything from it).
1448 close_pack_index(p
);
1452 static void show_pack_info(int stat_only
)
1454 int i
, baseobjects
= nr_objects
- nr_deltas
;
1455 unsigned long *chain_histogram
= NULL
;
1458 chain_histogram
= xcalloc(deepest_delta
, sizeof(unsigned long));
1460 for (i
= 0; i
< nr_objects
; i
++) {
1461 struct object_entry
*obj
= &objects
[i
];
1463 if (is_delta_type(obj
->type
))
1464 chain_histogram
[obj
->delta_depth
- 1]++;
1467 printf("%s %-6s %lu %lu %"PRIuMAX
,
1468 sha1_to_hex(obj
->idx
.sha1
),
1469 typename(obj
->real_type
), obj
->size
,
1470 (unsigned long)(obj
[1].idx
.offset
- obj
->idx
.offset
),
1471 (uintmax_t)obj
->idx
.offset
);
1472 if (is_delta_type(obj
->type
)) {
1473 struct object_entry
*bobj
= &objects
[obj
->base_object_no
];
1474 printf(" %u %s", obj
->delta_depth
, sha1_to_hex(bobj
->idx
.sha1
));
1480 printf_ln(Q_("non delta: %d object",
1481 "non delta: %d objects",
1484 for (i
= 0; i
< deepest_delta
; i
++) {
1485 if (!chain_histogram
[i
])
1487 printf_ln(Q_("chain length = %d: %lu object",
1488 "chain length = %d: %lu objects",
1489 chain_histogram
[i
]),
1491 chain_histogram
[i
]);
1495 int cmd_index_pack(int argc
, const char **argv
, const char *prefix
)
1497 int i
, fix_thin_pack
= 0, verify
= 0, stat_only
= 0;
1498 const char *curr_pack
, *curr_index
;
1499 const char *index_name
= NULL
, *pack_name
= NULL
;
1500 const char *keep_name
= NULL
, *keep_msg
= NULL
;
1501 char *index_name_buf
= NULL
, *keep_name_buf
= NULL
;
1502 struct pack_idx_entry
**idx_objects
;
1503 struct pack_idx_option opts
;
1504 unsigned char pack_sha1
[20];
1505 unsigned foreign_nr
= 1; /* zero is a "good" value, assume bad */
1507 if (argc
== 2 && !strcmp(argv
[1], "-h"))
1508 usage(index_pack_usage
);
1510 read_replace_refs
= 0;
1512 reset_pack_idx_option(&opts
);
1513 git_config(git_index_pack_config
, &opts
);
1514 if (prefix
&& chdir(prefix
))
1515 die(_("Cannot come back to cwd"));
1517 for (i
= 1; i
< argc
; i
++) {
1518 const char *arg
= argv
[i
];
1521 if (!strcmp(arg
, "--stdin")) {
1523 } else if (!strcmp(arg
, "--fix-thin")) {
1525 } else if (!strcmp(arg
, "--strict")) {
1528 } else if (!strcmp(arg
, "--check-self-contained-and-connected")) {
1530 check_self_contained_and_connected
= 1;
1531 } else if (!strcmp(arg
, "--verify")) {
1533 } else if (!strcmp(arg
, "--verify-stat")) {
1536 } else if (!strcmp(arg
, "--verify-stat-only")) {
1540 } else if (!strcmp(arg
, "--keep")) {
1542 } else if (starts_with(arg
, "--keep=")) {
1544 } else if (starts_with(arg
, "--threads=")) {
1546 nr_threads
= strtoul(arg
+10, &end
, 0);
1547 if (!arg
[10] || *end
|| nr_threads
< 0)
1548 usage(index_pack_usage
);
1550 if (nr_threads
!= 1)
1551 warning(_("no threads support, "
1552 "ignoring %s"), arg
);
1555 } else if (starts_with(arg
, "--pack_header=")) {
1556 struct pack_header
*hdr
;
1559 hdr
= (struct pack_header
*)input_buffer
;
1560 hdr
->hdr_signature
= htonl(PACK_SIGNATURE
);
1561 hdr
->hdr_version
= htonl(strtoul(arg
+ 14, &c
, 10));
1563 die(_("bad %s"), arg
);
1564 hdr
->hdr_entries
= htonl(strtoul(c
+ 1, &c
, 10));
1566 die(_("bad %s"), arg
);
1567 input_len
= sizeof(*hdr
);
1568 } else if (!strcmp(arg
, "-v")) {
1570 } else if (!strcmp(arg
, "-o")) {
1571 if (index_name
|| (i
+1) >= argc
)
1572 usage(index_pack_usage
);
1573 index_name
= argv
[++i
];
1574 } else if (starts_with(arg
, "--index-version=")) {
1576 opts
.version
= strtoul(arg
+ 16, &c
, 10);
1577 if (opts
.version
> 2)
1578 die(_("bad %s"), arg
);
1580 opts
.off32_limit
= strtoul(c
+1, &c
, 0);
1581 if (*c
|| opts
.off32_limit
& 0x80000000)
1582 die(_("bad %s"), arg
);
1584 usage(index_pack_usage
);
1589 usage(index_pack_usage
);
1593 if (!pack_name
&& !from_stdin
)
1594 usage(index_pack_usage
);
1595 if (fix_thin_pack
&& !from_stdin
)
1596 die(_("--fix-thin cannot be used without --stdin"));
1597 if (!index_name
&& pack_name
) {
1598 int len
= strlen(pack_name
);
1599 if (!has_extension(pack_name
, ".pack"))
1600 die(_("packfile name '%s' does not end with '.pack'"),
1602 index_name_buf
= xmalloc(len
);
1603 memcpy(index_name_buf
, pack_name
, len
- 5);
1604 strcpy(index_name_buf
+ len
- 5, ".idx");
1605 index_name
= index_name_buf
;
1607 if (keep_msg
&& !keep_name
&& pack_name
) {
1608 int len
= strlen(pack_name
);
1609 if (!has_extension(pack_name
, ".pack"))
1610 die(_("packfile name '%s' does not end with '.pack'"),
1612 keep_name_buf
= xmalloc(len
);
1613 memcpy(keep_name_buf
, pack_name
, len
- 5);
1614 strcpy(keep_name_buf
+ len
- 5, ".keep");
1615 keep_name
= keep_name_buf
;
1619 die(_("--verify with no packfile name given"));
1620 read_idx_option(&opts
, index_name
);
1621 opts
.flags
|= WRITE_IDX_VERIFY
| WRITE_IDX_STRICT
;
1624 opts
.flags
|= WRITE_IDX_STRICT
;
1628 nr_threads
= online_cpus();
1629 /* An experiment showed that more threads does not mean faster */
1635 curr_pack
= open_pack_file(pack_name
);
1636 parse_pack_header();
1637 objects
= xcalloc(nr_objects
+ 1, sizeof(struct object_entry
));
1638 deltas
= xcalloc(nr_objects
, sizeof(struct delta_entry
));
1639 parse_pack_objects(pack_sha1
);
1641 conclude_pack(fix_thin_pack
, curr_pack
, pack_sha1
);
1644 foreign_nr
= check_objects();
1647 show_pack_info(stat_only
);
1649 idx_objects
= xmalloc((nr_objects
) * sizeof(struct pack_idx_entry
*));
1650 for (i
= 0; i
< nr_objects
; i
++)
1651 idx_objects
[i
] = &objects
[i
].idx
;
1652 curr_index
= write_idx_file(index_name
, idx_objects
, nr_objects
, &opts
, pack_sha1
);
1656 final(pack_name
, curr_pack
,
1657 index_name
, curr_index
,
1658 keep_name
, keep_msg
,
1663 free(index_name_buf
);
1664 free(keep_name_buf
);
1665 if (pack_name
== NULL
)
1666 free((void *) curr_pack
);
1667 if (index_name
== NULL
)
1668 free((void *) curr_index
);
1671 * Let the caller know this pack is not self contained
1673 if (check_self_contained_and_connected
&& foreign_nr
)