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
;
83 static struct progress
*progress
;
85 /* We always read in 4kB chunks. */
86 static unsigned char input_buffer
[4096];
87 static unsigned int input_offset
, input_len
;
88 static off_t consumed_bytes
;
89 static unsigned deepest_delta
;
90 static git_SHA_CTX input_ctx
;
91 static uint32_t input_crc32
;
92 static int input_fd
, output_fd
, pack_fd
;
96 static struct thread_local
*thread_data
;
97 static int nr_dispatched
;
98 static int threads_active
;
100 static pthread_mutex_t read_mutex
;
101 #define read_lock() lock_mutex(&read_mutex)
102 #define read_unlock() unlock_mutex(&read_mutex)
104 static pthread_mutex_t counter_mutex
;
105 #define counter_lock() lock_mutex(&counter_mutex)
106 #define counter_unlock() unlock_mutex(&counter_mutex)
108 static pthread_mutex_t work_mutex
;
109 #define work_lock() lock_mutex(&work_mutex)
110 #define work_unlock() unlock_mutex(&work_mutex)
112 static pthread_mutex_t deepest_delta_mutex
;
113 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
114 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
116 static pthread_key_t key
;
118 static inline void lock_mutex(pthread_mutex_t
*mutex
)
121 pthread_mutex_lock(mutex
);
124 static inline void unlock_mutex(pthread_mutex_t
*mutex
)
127 pthread_mutex_unlock(mutex
);
131 * Mutex and conditional variable can't be statically-initialized on Windows.
133 static void init_thread(void)
135 init_recursive_mutex(&read_mutex
);
136 pthread_mutex_init(&counter_mutex
, NULL
);
137 pthread_mutex_init(&work_mutex
, NULL
);
139 pthread_mutex_init(&deepest_delta_mutex
, NULL
);
140 pthread_key_create(&key
, NULL
);
141 thread_data
= xcalloc(nr_threads
, sizeof(*thread_data
));
145 static void cleanup_thread(void)
150 pthread_mutex_destroy(&read_mutex
);
151 pthread_mutex_destroy(&counter_mutex
);
152 pthread_mutex_destroy(&work_mutex
);
154 pthread_mutex_destroy(&deepest_delta_mutex
);
155 pthread_key_delete(key
);
162 #define read_unlock()
164 #define counter_lock()
165 #define counter_unlock()
168 #define work_unlock()
170 #define deepest_delta_lock()
171 #define deepest_delta_unlock()
176 static int mark_link(struct object
*obj
, int type
, void *data
)
181 if (type
!= OBJ_ANY
&& obj
->type
!= type
)
182 die(_("object type mismatch at %s"), sha1_to_hex(obj
->sha1
));
184 obj
->flags
|= FLAG_LINK
;
188 /* The content of each linked object must have been checked
189 or it must be already present in the object database */
190 static void check_object(struct object
*obj
)
195 if (!(obj
->flags
& FLAG_LINK
))
198 if (!(obj
->flags
& FLAG_CHECKED
)) {
200 int type
= sha1_object_info(obj
->sha1
, &size
);
201 if (type
!= obj
->type
|| type
<= 0)
202 die(_("object of unexpected type"));
203 obj
->flags
|= FLAG_CHECKED
;
208 static void check_objects(void)
212 max
= get_max_object_index();
213 for (i
= 0; i
< max
; i
++)
214 check_object(get_indexed_object(i
));
218 /* Discard current buffer used content. */
219 static void flush(void)
223 write_or_die(output_fd
, input_buffer
, input_offset
);
224 git_SHA1_Update(&input_ctx
, input_buffer
, input_offset
);
225 memmove(input_buffer
, input_buffer
+ input_offset
, input_len
);
231 * Make sure at least "min" bytes are available in the buffer, and
232 * return the pointer to the buffer.
234 static void *fill(int min
)
236 if (min
<= input_len
)
237 return input_buffer
+ input_offset
;
238 if (min
> sizeof(input_buffer
))
239 die(Q_("cannot fill %d byte",
240 "cannot fill %d bytes",
245 ssize_t ret
= xread(input_fd
, input_buffer
+ input_len
,
246 sizeof(input_buffer
) - input_len
);
250 die_errno(_("read error on input"));
254 display_throughput(progress
, consumed_bytes
+ input_len
);
255 } while (input_len
< min
);
259 static void use(int bytes
)
261 if (bytes
> input_len
)
262 die(_("used more bytes than were available"));
263 input_crc32
= crc32(input_crc32
, input_buffer
+ input_offset
, bytes
);
265 input_offset
+= bytes
;
267 /* make sure off_t is sufficiently large not to wrap */
268 if (signed_add_overflows(consumed_bytes
, bytes
))
269 die(_("pack too large for current definition of off_t"));
270 consumed_bytes
+= bytes
;
273 static const char *open_pack_file(const char *pack_name
)
278 static char tmp_file
[PATH_MAX
];
279 output_fd
= odb_mkstemp(tmp_file
, sizeof(tmp_file
),
280 "pack/tmp_pack_XXXXXX");
281 pack_name
= xstrdup(tmp_file
);
283 output_fd
= open(pack_name
, O_CREAT
|O_EXCL
|O_RDWR
, 0600);
285 die_errno(_("unable to create '%s'"), pack_name
);
288 input_fd
= open(pack_name
, O_RDONLY
);
290 die_errno(_("cannot open packfile '%s'"), pack_name
);
294 git_SHA1_Init(&input_ctx
);
298 static void parse_pack_header(void)
300 struct pack_header
*hdr
= fill(sizeof(struct pack_header
));
302 /* Header consistency check */
303 if (hdr
->hdr_signature
!= htonl(PACK_SIGNATURE
))
304 die(_("pack signature mismatch"));
305 if (!pack_version_ok(hdr
->hdr_version
))
306 die(_("pack version %"PRIu32
" unsupported"),
307 ntohl(hdr
->hdr_version
));
309 nr_objects
= ntohl(hdr
->hdr_entries
);
310 use(sizeof(struct pack_header
));
313 static NORETURN
void bad_object(unsigned long offset
, const char *format
,
314 ...) __attribute__((format (printf
, 2, 3)));
316 static NORETURN
void bad_object(unsigned long offset
, const char *format
, ...)
321 va_start(params
, format
);
322 vsnprintf(buf
, sizeof(buf
), format
, params
);
324 die(_("pack has bad object at offset %lu: %s"), offset
, buf
);
327 static inline struct thread_local
*get_thread_data(void)
331 return pthread_getspecific(key
);
332 assert(!threads_active
&&
333 "This should only be reached when all threads are gone");
335 return ¬hread_data
;
339 static void set_thread_data(struct thread_local
*data
)
342 pthread_setspecific(key
, data
);
346 static struct base_data
*alloc_base_data(void)
348 struct base_data
*base
= xmalloc(sizeof(struct base_data
));
349 memset(base
, 0, sizeof(*base
));
355 static void free_base_data(struct base_data
*c
)
360 get_thread_data()->base_cache_used
-= c
->size
;
364 static void prune_base_data(struct base_data
*retain
)
367 struct thread_local
*data
= get_thread_data();
368 for (b
= data
->base_cache
;
369 data
->base_cache_used
> delta_base_cache_limit
&& b
;
371 if (b
->data
&& b
!= retain
)
376 static void link_base_data(struct base_data
*base
, struct base_data
*c
)
381 get_thread_data()->base_cache
= c
;
386 get_thread_data()->base_cache_used
+= c
->size
;
390 static void unlink_base_data(struct base_data
*c
)
392 struct base_data
*base
= c
->base
;
396 get_thread_data()->base_cache
= NULL
;
400 static int is_delta_type(enum object_type type
)
402 return (type
== OBJ_REF_DELTA
|| type
== OBJ_OFS_DELTA
);
405 static void *unpack_entry_data(unsigned long offset
, unsigned long size
,
406 enum object_type type
, unsigned char *sha1
)
408 static char fixed_buf
[8192];
416 if (!is_delta_type(type
)) {
417 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), size
) + 1;
419 git_SHA1_Update(&c
, hdr
, hdrlen
);
422 if (type
== OBJ_BLOB
&& size
> big_file_threshold
)
427 memset(&stream
, 0, sizeof(stream
));
428 git_inflate_init(&stream
);
429 stream
.next_out
= buf
;
430 stream
.avail_out
= buf
== fixed_buf
? sizeof(fixed_buf
) : size
;
433 unsigned char *last_out
= stream
.next_out
;
434 stream
.next_in
= fill(1);
435 stream
.avail_in
= input_len
;
436 status
= git_inflate(&stream
, 0);
437 use(input_len
- stream
.avail_in
);
439 git_SHA1_Update(&c
, last_out
, stream
.next_out
- last_out
);
440 if (buf
== fixed_buf
) {
441 stream
.next_out
= buf
;
442 stream
.avail_out
= sizeof(fixed_buf
);
444 } while (status
== Z_OK
);
445 if (stream
.total_out
!= size
|| status
!= Z_STREAM_END
)
446 bad_object(offset
, _("inflate returned %d"), status
);
447 git_inflate_end(&stream
);
449 git_SHA1_Final(sha1
, &c
);
450 return buf
== fixed_buf
? NULL
: buf
;
453 static void *unpack_raw_entry(struct object_entry
*obj
,
454 union delta_base
*delta_base
,
458 unsigned long size
, c
;
463 obj
->idx
.offset
= consumed_bytes
;
464 input_crc32
= crc32(0, NULL
, 0);
469 obj
->type
= (c
>> 4) & 7;
476 size
+= (c
& 0x7f) << shift
;
483 hashcpy(delta_base
->sha1
, fill(20));
487 memset(delta_base
, 0, sizeof(*delta_base
));
491 base_offset
= c
& 127;
494 if (!base_offset
|| MSB(base_offset
, 7))
495 bad_object(obj
->idx
.offset
, _("offset value overflow for delta base object"));
499 base_offset
= (base_offset
<< 7) + (c
& 127);
501 delta_base
->offset
= obj
->idx
.offset
- base_offset
;
502 if (delta_base
->offset
<= 0 || delta_base
->offset
>= obj
->idx
.offset
)
503 bad_object(obj
->idx
.offset
, _("delta base offset is out of bound"));
511 bad_object(obj
->idx
.offset
, _("unknown object type %d"), obj
->type
);
513 obj
->hdr_size
= consumed_bytes
- obj
->idx
.offset
;
515 data
= unpack_entry_data(obj
->idx
.offset
, obj
->size
, obj
->type
, sha1
);
516 obj
->idx
.crc32
= input_crc32
;
520 static void *unpack_data(struct object_entry
*obj
,
521 int (*consume
)(const unsigned char *, unsigned long, void *),
524 off_t from
= obj
[0].idx
.offset
+ obj
[0].hdr_size
;
525 unsigned long len
= obj
[1].idx
.offset
- from
;
526 unsigned char *data
, *inbuf
;
530 data
= xmalloc(consume
? 64*1024 : obj
->size
);
531 inbuf
= xmalloc((len
< 64*1024) ? len
: 64*1024);
533 memset(&stream
, 0, sizeof(stream
));
534 git_inflate_init(&stream
);
535 stream
.next_out
= data
;
536 stream
.avail_out
= consume
? 64*1024 : obj
->size
;
539 ssize_t n
= (len
< 64*1024) ? len
: 64*1024;
540 n
= pread(pack_fd
, inbuf
, n
, from
);
542 die_errno(_("cannot pread pack file"));
544 die(Q_("premature end of pack file, %lu byte missing",
545 "premature end of pack file, %lu bytes missing",
550 stream
.next_in
= inbuf
;
553 status
= git_inflate(&stream
, 0);
556 status
= git_inflate(&stream
, 0);
557 if (consume(data
, stream
.next_out
- data
, cb_data
)) {
562 stream
.next_out
= data
;
563 stream
.avail_out
= 64*1024;
564 } while (status
== Z_OK
&& stream
.avail_in
);
566 } while (len
&& status
== Z_OK
&& !stream
.avail_in
);
568 /* This has been inflated OK when first encountered, so... */
569 if (status
!= Z_STREAM_END
|| stream
.total_out
!= obj
->size
)
570 die(_("serious inflate inconsistency"));
572 git_inflate_end(&stream
);
581 static void *get_data_from_pack(struct object_entry
*obj
)
583 return unpack_data(obj
, NULL
, NULL
);
586 static int compare_delta_bases(const union delta_base
*base1
,
587 const union delta_base
*base2
,
588 enum object_type type1
,
589 enum object_type type2
)
591 int cmp
= type1
- type2
;
594 return memcmp(base1
, base2
, UNION_BASE_SZ
);
597 static int find_delta(const union delta_base
*base
, enum object_type type
)
599 int first
= 0, last
= nr_deltas
;
601 while (first
< last
) {
602 int next
= (first
+ last
) / 2;
603 struct delta_entry
*delta
= &deltas
[next
];
606 cmp
= compare_delta_bases(base
, &delta
->base
,
607 type
, objects
[delta
->obj_no
].type
);
619 static void find_delta_children(const union delta_base
*base
,
620 int *first_index
, int *last_index
,
621 enum object_type type
)
623 int first
= find_delta(base
, type
);
625 int end
= nr_deltas
- 1;
632 while (first
> 0 && !memcmp(&deltas
[first
- 1].base
, base
, UNION_BASE_SZ
))
634 while (last
< end
&& !memcmp(&deltas
[last
+ 1].base
, base
, UNION_BASE_SZ
))
636 *first_index
= first
;
640 struct compare_data
{
641 struct object_entry
*entry
;
642 struct git_istream
*st
;
644 unsigned long buf_size
;
647 static int compare_objects(const unsigned char *buf
, unsigned long size
,
650 struct compare_data
*data
= cb_data
;
652 if (data
->buf_size
< size
) {
654 data
->buf
= xmalloc(size
);
655 data
->buf_size
= size
;
659 ssize_t len
= read_istream(data
->st
, data
->buf
, size
);
661 die(_("SHA1 COLLISION FOUND WITH %s !"),
662 sha1_to_hex(data
->entry
->idx
.sha1
));
664 die(_("unable to read %s"),
665 sha1_to_hex(data
->entry
->idx
.sha1
));
666 if (memcmp(buf
, data
->buf
, len
))
667 die(_("SHA1 COLLISION FOUND WITH %s !"),
668 sha1_to_hex(data
->entry
->idx
.sha1
));
675 static int check_collison(struct object_entry
*entry
)
677 struct compare_data data
;
678 enum object_type type
;
681 if (entry
->size
<= big_file_threshold
|| entry
->type
!= OBJ_BLOB
)
684 memset(&data
, 0, sizeof(data
));
686 data
.st
= open_istream(entry
->idx
.sha1
, &type
, &size
, NULL
);
689 if (size
!= entry
->size
|| type
!= entry
->type
)
690 die(_("SHA1 COLLISION FOUND WITH %s !"),
691 sha1_to_hex(entry
->idx
.sha1
));
692 unpack_data(entry
, compare_objects
, &data
);
693 close_istream(data
.st
);
698 static void sha1_object(const void *data
, struct object_entry
*obj_entry
,
699 unsigned long size
, enum object_type type
,
700 const unsigned char *sha1
)
702 void *new_data
= NULL
;
703 int collision_test_needed
;
705 assert(data
|| obj_entry
);
708 collision_test_needed
= has_sha1_file(sha1
);
711 if (collision_test_needed
&& !data
) {
713 if (!check_collison(obj_entry
))
714 collision_test_needed
= 0;
717 if (collision_test_needed
) {
719 enum object_type has_type
;
720 unsigned long has_size
;
722 has_type
= sha1_object_info(sha1
, &has_size
);
723 if (has_type
!= type
|| has_size
!= size
)
724 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1
));
725 has_data
= read_sha1_file(sha1
, &has_type
, &has_size
);
728 data
= new_data
= get_data_from_pack(obj_entry
);
730 die(_("cannot read existing object %s"), sha1_to_hex(sha1
));
731 if (size
!= has_size
|| type
!= has_type
||
732 memcmp(data
, has_data
, size
) != 0)
733 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1
));
739 if (type
== OBJ_BLOB
) {
740 struct blob
*blob
= lookup_blob(sha1
);
742 blob
->object
.flags
|= FLAG_CHECKED
;
744 die(_("invalid blob object %s"), sha1_to_hex(sha1
));
748 void *buf
= (void *) data
;
750 assert(data
&& "data can only be NULL for large _blobs_");
753 * we do not need to free the memory here, as the
754 * buf is deleted by the caller.
756 obj
= parse_object_buffer(sha1
, type
, size
, buf
, &eaten
);
758 die(_("invalid %s"), typename(type
));
759 if (fsck_object(obj
, 1, fsck_error_function
))
760 die(_("Error in object"));
761 if (fsck_walk(obj
, mark_link
, NULL
))
762 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj
->sha1
));
764 if (obj
->type
== OBJ_TREE
) {
765 struct tree
*item
= (struct tree
*) obj
;
768 if (obj
->type
== OBJ_COMMIT
) {
769 struct commit
*commit
= (struct commit
*) obj
;
770 commit
->buffer
= NULL
;
772 obj
->flags
|= FLAG_CHECKED
;
781 * This function is part of find_unresolved_deltas(). There are two
782 * walkers going in the opposite ways.
784 * The first one in find_unresolved_deltas() traverses down from
785 * parent node to children, deflating nodes along the way. However,
786 * memory for deflated nodes is limited by delta_base_cache_limit, so
787 * at some point parent node's deflated content may be freed.
789 * The second walker is this function, which goes from current node up
790 * to top parent if necessary to deflate the node. In normal
791 * situation, its parent node would be already deflated, so it just
792 * needs to apply delta.
794 * In the worst case scenario, parent node is no longer deflated because
795 * we're running out of delta_base_cache_limit; we need to re-deflate
796 * parents, possibly up to the top base.
798 * All deflated objects here are subject to be freed if we exceed
799 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
800 * just need to make sure the last node is not freed.
802 static void *get_base_data(struct base_data
*c
)
805 struct object_entry
*obj
= c
->obj
;
806 struct base_data
**delta
= NULL
;
807 int delta_nr
= 0, delta_alloc
= 0;
809 while (is_delta_type(c
->obj
->type
) && !c
->data
) {
810 ALLOC_GROW(delta
, delta_nr
+ 1, delta_alloc
);
811 delta
[delta_nr
++] = c
;
815 c
->data
= get_data_from_pack(obj
);
817 get_thread_data()->base_cache_used
+= c
->size
;
820 for (; delta_nr
> 0; delta_nr
--) {
822 c
= delta
[delta_nr
- 1];
824 base
= get_base_data(c
->base
);
825 raw
= get_data_from_pack(obj
);
826 c
->data
= patch_delta(
832 bad_object(obj
->idx
.offset
, _("failed to apply delta"));
833 get_thread_data()->base_cache_used
+= c
->size
;
841 static void resolve_delta(struct object_entry
*delta_obj
,
842 struct base_data
*base
, struct base_data
*result
)
844 void *base_data
, *delta_data
;
846 delta_obj
->real_type
= base
->obj
->real_type
;
848 delta_obj
->delta_depth
= base
->obj
->delta_depth
+ 1;
849 deepest_delta_lock();
850 if (deepest_delta
< delta_obj
->delta_depth
)
851 deepest_delta
= delta_obj
->delta_depth
;
852 deepest_delta_unlock();
854 delta_obj
->base_object_no
= base
->obj
- objects
;
855 delta_data
= get_data_from_pack(delta_obj
);
856 base_data
= get_base_data(base
);
857 result
->obj
= delta_obj
;
858 result
->data
= patch_delta(base_data
, base
->size
,
859 delta_data
, delta_obj
->size
, &result
->size
);
862 bad_object(delta_obj
->idx
.offset
, _("failed to apply delta"));
863 hash_sha1_file(result
->data
, result
->size
,
864 typename(delta_obj
->real_type
), delta_obj
->idx
.sha1
);
865 sha1_object(result
->data
, NULL
, result
->size
, delta_obj
->real_type
,
866 delta_obj
->idx
.sha1
);
868 nr_resolved_deltas
++;
872 static struct base_data
*find_unresolved_deltas_1(struct base_data
*base
,
873 struct base_data
*prev_base
)
875 if (base
->ref_last
== -1 && base
->ofs_last
== -1) {
876 union delta_base base_spec
;
878 hashcpy(base_spec
.sha1
, base
->obj
->idx
.sha1
);
879 find_delta_children(&base_spec
,
880 &base
->ref_first
, &base
->ref_last
, OBJ_REF_DELTA
);
882 memset(&base_spec
, 0, sizeof(base_spec
));
883 base_spec
.offset
= base
->obj
->idx
.offset
;
884 find_delta_children(&base_spec
,
885 &base
->ofs_first
, &base
->ofs_last
, OBJ_OFS_DELTA
);
887 if (base
->ref_last
== -1 && base
->ofs_last
== -1) {
892 link_base_data(prev_base
, base
);
895 if (base
->ref_first
<= base
->ref_last
) {
896 struct object_entry
*child
= objects
+ deltas
[base
->ref_first
].obj_no
;
897 struct base_data
*result
= alloc_base_data();
899 assert(child
->real_type
== OBJ_REF_DELTA
);
900 resolve_delta(child
, base
, result
);
901 if (base
->ref_first
== base
->ref_last
&& base
->ofs_last
== -1)
902 free_base_data(base
);
908 if (base
->ofs_first
<= base
->ofs_last
) {
909 struct object_entry
*child
= objects
+ deltas
[base
->ofs_first
].obj_no
;
910 struct base_data
*result
= alloc_base_data();
912 assert(child
->real_type
== OBJ_OFS_DELTA
);
913 resolve_delta(child
, base
, result
);
914 if (base
->ofs_first
== base
->ofs_last
)
915 free_base_data(base
);
921 unlink_base_data(base
);
925 static void find_unresolved_deltas(struct base_data
*base
)
927 struct base_data
*new_base
, *prev_base
= NULL
;
929 new_base
= find_unresolved_deltas_1(base
, prev_base
);
939 prev_base
= base
->base
;
944 static int compare_delta_entry(const void *a
, const void *b
)
946 const struct delta_entry
*delta_a
= a
;
947 const struct delta_entry
*delta_b
= b
;
949 /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
950 return compare_delta_bases(&delta_a
->base
, &delta_b
->base
,
951 objects
[delta_a
->obj_no
].type
,
952 objects
[delta_b
->obj_no
].type
);
955 static void resolve_base(struct object_entry
*obj
)
957 struct base_data
*base_obj
= alloc_base_data();
959 base_obj
->data
= NULL
;
960 find_unresolved_deltas(base_obj
);
964 static void *threaded_second_pass(void *data
)
966 set_thread_data(data
);
970 display_progress(progress
, nr_resolved_deltas
);
973 while (nr_dispatched
< nr_objects
&&
974 is_delta_type(objects
[nr_dispatched
].type
))
976 if (nr_dispatched
>= nr_objects
) {
983 resolve_base(&objects
[i
]);
991 * - find locations of all objects;
992 * - calculate SHA1 of all non-delta objects;
993 * - remember base (SHA1 or offset) for all deltas.
995 static void parse_pack_objects(unsigned char *sha1
)
997 int i
, nr_delays
= 0;
998 struct delta_entry
*delta
= deltas
;
1002 progress
= start_progress(
1003 from_stdin
? _("Receiving objects") : _("Indexing objects"),
1005 for (i
= 0; i
< nr_objects
; i
++) {
1006 struct object_entry
*obj
= &objects
[i
];
1007 void *data
= unpack_raw_entry(obj
, &delta
->base
, obj
->idx
.sha1
);
1008 obj
->real_type
= obj
->type
;
1009 if (is_delta_type(obj
->type
)) {
1014 /* large blobs, check later */
1015 obj
->real_type
= OBJ_BAD
;
1018 sha1_object(data
, NULL
, obj
->size
, obj
->type
, obj
->idx
.sha1
);
1020 display_progress(progress
, i
+1);
1022 objects
[i
].idx
.offset
= consumed_bytes
;
1023 stop_progress(&progress
);
1025 /* Check pack integrity */
1027 git_SHA1_Final(sha1
, &input_ctx
);
1028 if (hashcmp(fill(20), sha1
))
1029 die(_("pack is corrupted (SHA1 mismatch)"));
1032 /* If input_fd is a file, we should have reached its end now. */
1033 if (fstat(input_fd
, &st
))
1034 die_errno(_("cannot fstat packfile"));
1035 if (S_ISREG(st
.st_mode
) &&
1036 lseek(input_fd
, 0, SEEK_CUR
) - input_len
!= st
.st_size
)
1037 die(_("pack has junk at the end"));
1039 for (i
= 0; i
< nr_objects
; i
++) {
1040 struct object_entry
*obj
= &objects
[i
];
1041 if (obj
->real_type
!= OBJ_BAD
)
1043 obj
->real_type
= obj
->type
;
1044 sha1_object(NULL
, obj
, obj
->size
, obj
->type
, obj
->idx
.sha1
);
1048 die(_("confusion beyond insanity in parse_pack_objects()"));
1053 * - for all non-delta objects, look if it is used as a base for
1055 * - if used as a base, uncompress the object and apply all deltas,
1056 * recursively checking if the resulting object is used as a base
1057 * for some more deltas.
1059 static void resolve_deltas(void)
1066 /* Sort deltas by base SHA1/offset for fast searching */
1067 qsort(deltas
, nr_deltas
, sizeof(struct delta_entry
),
1068 compare_delta_entry
);
1071 progress
= start_progress(_("Resolving deltas"), nr_deltas
);
1075 if (nr_threads
> 1 || getenv("GIT_FORCE_THREADS")) {
1077 for (i
= 0; i
< nr_threads
; i
++) {
1078 int ret
= pthread_create(&thread_data
[i
].thread
, NULL
,
1079 threaded_second_pass
, thread_data
+ i
);
1081 die(_("unable to create thread: %s"),
1084 for (i
= 0; i
< nr_threads
; i
++)
1085 pthread_join(thread_data
[i
].thread
, NULL
);
1091 for (i
= 0; i
< nr_objects
; i
++) {
1092 struct object_entry
*obj
= &objects
[i
];
1094 if (is_delta_type(obj
->type
))
1097 display_progress(progress
, nr_resolved_deltas
);
1103 * - append objects to convert thin pack to full pack if required
1104 * - write the final 20-byte SHA-1
1106 static void fix_unresolved_deltas(struct sha1file
*f
, int nr_unresolved
);
1107 static void conclude_pack(int fix_thin_pack
, const char *curr_pack
, unsigned char *pack_sha1
)
1109 if (nr_deltas
== nr_resolved_deltas
) {
1110 stop_progress(&progress
);
1111 /* Flush remaining pack final 20-byte SHA1. */
1116 if (fix_thin_pack
) {
1118 unsigned char read_sha1
[20], tail_sha1
[20];
1119 struct strbuf msg
= STRBUF_INIT
;
1120 int nr_unresolved
= nr_deltas
- nr_resolved_deltas
;
1121 int nr_objects_initial
= nr_objects
;
1122 if (nr_unresolved
<= 0)
1123 die(_("confusion beyond insanity"));
1124 objects
= xrealloc(objects
,
1125 (nr_objects
+ nr_unresolved
+ 1)
1126 * sizeof(*objects
));
1127 memset(objects
+ nr_objects
+ 1, 0,
1128 nr_unresolved
* sizeof(*objects
));
1129 f
= sha1fd(output_fd
, curr_pack
);
1130 fix_unresolved_deltas(f
, nr_unresolved
);
1131 strbuf_addf(&msg
, _("completed with %d local objects"),
1132 nr_objects
- nr_objects_initial
);
1133 stop_progress_msg(&progress
, msg
.buf
);
1134 strbuf_release(&msg
);
1135 sha1close(f
, tail_sha1
, 0);
1136 hashcpy(read_sha1
, pack_sha1
);
1137 fixup_pack_header_footer(output_fd
, pack_sha1
,
1138 curr_pack
, nr_objects
,
1139 read_sha1
, consumed_bytes
-20);
1140 if (hashcmp(read_sha1
, tail_sha1
) != 0)
1141 die(_("Unexpected tail checksum for %s "
1142 "(disk corruption?)"), curr_pack
);
1144 if (nr_deltas
!= nr_resolved_deltas
)
1145 die(Q_("pack has %d unresolved delta",
1146 "pack has %d unresolved deltas",
1147 nr_deltas
- nr_resolved_deltas
),
1148 nr_deltas
- nr_resolved_deltas
);
1151 static int write_compressed(struct sha1file
*f
, void *in
, unsigned int size
)
1155 unsigned char outbuf
[4096];
1157 memset(&stream
, 0, sizeof(stream
));
1158 git_deflate_init(&stream
, zlib_compression_level
);
1159 stream
.next_in
= in
;
1160 stream
.avail_in
= size
;
1163 stream
.next_out
= outbuf
;
1164 stream
.avail_out
= sizeof(outbuf
);
1165 status
= git_deflate(&stream
, Z_FINISH
);
1166 sha1write(f
, outbuf
, sizeof(outbuf
) - stream
.avail_out
);
1167 } while (status
== Z_OK
);
1169 if (status
!= Z_STREAM_END
)
1170 die(_("unable to deflate appended object (%d)"), status
);
1171 size
= stream
.total_out
;
1172 git_deflate_end(&stream
);
1176 static struct object_entry
*append_obj_to_pack(struct sha1file
*f
,
1177 const unsigned char *sha1
, void *buf
,
1178 unsigned long size
, enum object_type type
)
1180 struct object_entry
*obj
= &objects
[nr_objects
++];
1181 unsigned char header
[10];
1182 unsigned long s
= size
;
1184 unsigned char c
= (type
<< 4) | (s
& 15);
1187 header
[n
++] = c
| 0x80;
1193 sha1write(f
, header
, n
);
1195 obj
[0].hdr_size
= n
;
1197 obj
[0].real_type
= type
;
1198 obj
[1].idx
.offset
= obj
[0].idx
.offset
+ n
;
1199 obj
[1].idx
.offset
+= write_compressed(f
, buf
, size
);
1200 obj
[0].idx
.crc32
= crc32_end(f
);
1202 hashcpy(obj
->idx
.sha1
, sha1
);
1206 static int delta_pos_compare(const void *_a
, const void *_b
)
1208 struct delta_entry
*a
= *(struct delta_entry
**)_a
;
1209 struct delta_entry
*b
= *(struct delta_entry
**)_b
;
1210 return a
->obj_no
- b
->obj_no
;
1213 static void fix_unresolved_deltas(struct sha1file
*f
, int nr_unresolved
)
1215 struct delta_entry
**sorted_by_pos
;
1219 * Since many unresolved deltas may well be themselves base objects
1220 * for more unresolved deltas, we really want to include the
1221 * smallest number of base objects that would cover as much delta
1222 * as possible by picking the
1223 * trunc deltas first, allowing for other deltas to resolve without
1224 * additional base objects. Since most base objects are to be found
1225 * before deltas depending on them, a good heuristic is to start
1226 * resolving deltas in the same order as their position in the pack.
1228 sorted_by_pos
= xmalloc(nr_unresolved
* sizeof(*sorted_by_pos
));
1229 for (i
= 0; i
< nr_deltas
; i
++) {
1230 if (objects
[deltas
[i
].obj_no
].real_type
!= OBJ_REF_DELTA
)
1232 sorted_by_pos
[n
++] = &deltas
[i
];
1234 qsort(sorted_by_pos
, n
, sizeof(*sorted_by_pos
), delta_pos_compare
);
1236 for (i
= 0; i
< n
; i
++) {
1237 struct delta_entry
*d
= sorted_by_pos
[i
];
1238 enum object_type type
;
1239 struct base_data
*base_obj
= alloc_base_data();
1241 if (objects
[d
->obj_no
].real_type
!= OBJ_REF_DELTA
)
1243 base_obj
->data
= read_sha1_file(d
->base
.sha1
, &type
, &base_obj
->size
);
1244 if (!base_obj
->data
)
1247 if (check_sha1_signature(d
->base
.sha1
, base_obj
->data
,
1248 base_obj
->size
, typename(type
)))
1249 die(_("local object %s is corrupt"), sha1_to_hex(d
->base
.sha1
));
1250 base_obj
->obj
= append_obj_to_pack(f
, d
->base
.sha1
,
1251 base_obj
->data
, base_obj
->size
, type
);
1252 find_unresolved_deltas(base_obj
);
1253 display_progress(progress
, nr_resolved_deltas
);
1255 free(sorted_by_pos
);
1258 static void final(const char *final_pack_name
, const char *curr_pack_name
,
1259 const char *final_index_name
, const char *curr_index_name
,
1260 const char *keep_name
, const char *keep_msg
,
1261 unsigned char *sha1
)
1263 const char *report
= "pack";
1264 char name
[PATH_MAX
];
1270 fsync_or_die(output_fd
, curr_pack_name
);
1271 err
= close(output_fd
);
1273 die_errno(_("error while closing pack file"));
1277 int keep_fd
, keep_msg_len
= strlen(keep_msg
);
1280 keep_fd
= odb_pack_keep(name
, sizeof(name
), sha1
);
1282 keep_fd
= open(keep_name
, O_RDWR
|O_CREAT
|O_EXCL
, 0600);
1285 if (errno
!= EEXIST
)
1286 die_errno(_("cannot write keep file '%s'"),
1289 if (keep_msg_len
> 0) {
1290 write_or_die(keep_fd
, keep_msg
, keep_msg_len
);
1291 write_or_die(keep_fd
, "\n", 1);
1293 if (close(keep_fd
) != 0)
1294 die_errno(_("cannot close written keep file '%s'"),
1300 if (final_pack_name
!= curr_pack_name
) {
1301 if (!final_pack_name
) {
1302 snprintf(name
, sizeof(name
), "%s/pack/pack-%s.pack",
1303 get_object_directory(), sha1_to_hex(sha1
));
1304 final_pack_name
= name
;
1306 if (move_temp_to_file(curr_pack_name
, final_pack_name
))
1307 die(_("cannot store pack file"));
1308 } else if (from_stdin
)
1309 chmod(final_pack_name
, 0444);
1311 if (final_index_name
!= curr_index_name
) {
1312 if (!final_index_name
) {
1313 snprintf(name
, sizeof(name
), "%s/pack/pack-%s.idx",
1314 get_object_directory(), sha1_to_hex(sha1
));
1315 final_index_name
= name
;
1317 if (move_temp_to_file(curr_index_name
, final_index_name
))
1318 die(_("cannot store index file"));
1320 chmod(final_index_name
, 0444);
1323 printf("%s\n", sha1_to_hex(sha1
));
1326 int len
= snprintf(buf
, sizeof(buf
), "%s\t%s\n",
1327 report
, sha1_to_hex(sha1
));
1328 write_or_die(1, buf
, len
);
1331 * Let's just mimic git-unpack-objects here and write
1332 * the last part of the input buffer to stdout.
1335 err
= xwrite(1, input_buffer
+ input_offset
, input_len
);
1339 input_offset
+= err
;
1344 static int git_index_pack_config(const char *k
, const char *v
, void *cb
)
1346 struct pack_idx_option
*opts
= cb
;
1348 if (!strcmp(k
, "pack.indexversion")) {
1349 opts
->version
= git_config_int(k
, v
);
1350 if (opts
->version
> 2)
1351 die(_("bad pack.indexversion=%"PRIu32
), opts
->version
);
1354 if (!strcmp(k
, "pack.threads")) {
1355 nr_threads
= git_config_int(k
, v
);
1357 die(_("invalid number of threads specified (%d)"),
1360 if (nr_threads
!= 1)
1361 warning(_("no threads support, ignoring %s"), k
);
1366 return git_default_config(k
, v
, cb
);
1369 static int cmp_uint32(const void *a_
, const void *b_
)
1371 uint32_t a
= *((uint32_t *)a_
);
1372 uint32_t b
= *((uint32_t *)b_
);
1374 return (a
< b
) ? -1 : (a
!= b
);
1377 static void read_v2_anomalous_offsets(struct packed_git
*p
,
1378 struct pack_idx_option
*opts
)
1380 const uint32_t *idx1
, *idx2
;
1383 /* The address of the 4-byte offset table */
1384 idx1
= (((const uint32_t *)p
->index_data
)
1385 + 2 /* 8-byte header */
1387 + 5 * p
->num_objects
/* 20-byte SHA-1 table */
1388 + p
->num_objects
/* CRC32 table */
1391 /* The address of the 8-byte offset table */
1392 idx2
= idx1
+ p
->num_objects
;
1394 for (i
= 0; i
< p
->num_objects
; i
++) {
1395 uint32_t off
= ntohl(idx1
[i
]);
1396 if (!(off
& 0x80000000))
1398 off
= off
& 0x7fffffff;
1402 * The real offset is ntohl(idx2[off * 2]) in high 4
1403 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1404 * octets. But idx2[off * 2] is Zero!!!
1406 ALLOC_GROW(opts
->anomaly
, opts
->anomaly_nr
+ 1, opts
->anomaly_alloc
);
1407 opts
->anomaly
[opts
->anomaly_nr
++] = ntohl(idx2
[off
* 2 + 1]);
1410 if (1 < opts
->anomaly_nr
)
1411 qsort(opts
->anomaly
, opts
->anomaly_nr
, sizeof(uint32_t), cmp_uint32
);
1414 static void read_idx_option(struct pack_idx_option
*opts
, const char *pack_name
)
1416 struct packed_git
*p
= add_packed_git(pack_name
, strlen(pack_name
), 1);
1419 die(_("Cannot open existing pack file '%s'"), pack_name
);
1420 if (open_pack_index(p
))
1421 die(_("Cannot open existing pack idx file for '%s'"), pack_name
);
1423 /* Read the attributes from the existing idx file */
1424 opts
->version
= p
->index_version
;
1426 if (opts
->version
== 2)
1427 read_v2_anomalous_offsets(p
, opts
);
1430 * Get rid of the idx file as we do not need it anymore.
1431 * NEEDSWORK: extract this bit from free_pack_by_name() in
1432 * sha1_file.c, perhaps? It shouldn't matter very much as we
1433 * know we haven't installed this pack (hence we never have
1434 * read anything from it).
1436 close_pack_index(p
);
1440 static void show_pack_info(int stat_only
)
1442 int i
, baseobjects
= nr_objects
- nr_deltas
;
1443 unsigned long *chain_histogram
= NULL
;
1446 chain_histogram
= xcalloc(deepest_delta
, sizeof(unsigned long));
1448 for (i
= 0; i
< nr_objects
; i
++) {
1449 struct object_entry
*obj
= &objects
[i
];
1451 if (is_delta_type(obj
->type
))
1452 chain_histogram
[obj
->delta_depth
- 1]++;
1455 printf("%s %-6s %lu %lu %"PRIuMAX
,
1456 sha1_to_hex(obj
->idx
.sha1
),
1457 typename(obj
->real_type
), obj
->size
,
1458 (unsigned long)(obj
[1].idx
.offset
- obj
->idx
.offset
),
1459 (uintmax_t)obj
->idx
.offset
);
1460 if (is_delta_type(obj
->type
)) {
1461 struct object_entry
*bobj
= &objects
[obj
->base_object_no
];
1462 printf(" %u %s", obj
->delta_depth
, sha1_to_hex(bobj
->idx
.sha1
));
1468 printf_ln(Q_("non delta: %d object",
1469 "non delta: %d objects",
1472 for (i
= 0; i
< deepest_delta
; i
++) {
1473 if (!chain_histogram
[i
])
1475 printf_ln(Q_("chain length = %d: %lu object",
1476 "chain length = %d: %lu objects",
1477 chain_histogram
[i
]),
1479 chain_histogram
[i
]);
1483 int cmd_index_pack(int argc
, const char **argv
, const char *prefix
)
1485 int i
, fix_thin_pack
= 0, verify
= 0, stat_only
= 0;
1486 const char *curr_pack
, *curr_index
;
1487 const char *index_name
= NULL
, *pack_name
= NULL
;
1488 const char *keep_name
= NULL
, *keep_msg
= NULL
;
1489 char *index_name_buf
= NULL
, *keep_name_buf
= NULL
;
1490 struct pack_idx_entry
**idx_objects
;
1491 struct pack_idx_option opts
;
1492 unsigned char pack_sha1
[20];
1494 if (argc
== 2 && !strcmp(argv
[1], "-h"))
1495 usage(index_pack_usage
);
1497 read_replace_refs
= 0;
1499 reset_pack_idx_option(&opts
);
1500 git_config(git_index_pack_config
, &opts
);
1501 if (prefix
&& chdir(prefix
))
1502 die(_("Cannot come back to cwd"));
1504 for (i
= 1; i
< argc
; i
++) {
1505 const char *arg
= argv
[i
];
1508 if (!strcmp(arg
, "--stdin")) {
1510 } else if (!strcmp(arg
, "--fix-thin")) {
1512 } else if (!strcmp(arg
, "--strict")) {
1514 } else if (!strcmp(arg
, "--verify")) {
1516 } else if (!strcmp(arg
, "--verify-stat")) {
1519 } else if (!strcmp(arg
, "--verify-stat-only")) {
1523 } else if (!strcmp(arg
, "--keep")) {
1525 } else if (!prefixcmp(arg
, "--keep=")) {
1527 } else if (!prefixcmp(arg
, "--threads=")) {
1529 nr_threads
= strtoul(arg
+10, &end
, 0);
1530 if (!arg
[10] || *end
|| nr_threads
< 0)
1531 usage(index_pack_usage
);
1533 if (nr_threads
!= 1)
1534 warning(_("no threads support, "
1535 "ignoring %s"), arg
);
1538 } else if (!prefixcmp(arg
, "--pack_header=")) {
1539 struct pack_header
*hdr
;
1542 hdr
= (struct pack_header
*)input_buffer
;
1543 hdr
->hdr_signature
= htonl(PACK_SIGNATURE
);
1544 hdr
->hdr_version
= htonl(strtoul(arg
+ 14, &c
, 10));
1546 die(_("bad %s"), arg
);
1547 hdr
->hdr_entries
= htonl(strtoul(c
+ 1, &c
, 10));
1549 die(_("bad %s"), arg
);
1550 input_len
= sizeof(*hdr
);
1551 } else if (!strcmp(arg
, "-v")) {
1553 } else if (!strcmp(arg
, "-o")) {
1554 if (index_name
|| (i
+1) >= argc
)
1555 usage(index_pack_usage
);
1556 index_name
= argv
[++i
];
1557 } else if (!prefixcmp(arg
, "--index-version=")) {
1559 opts
.version
= strtoul(arg
+ 16, &c
, 10);
1560 if (opts
.version
> 2)
1561 die(_("bad %s"), arg
);
1563 opts
.off32_limit
= strtoul(c
+1, &c
, 0);
1564 if (*c
|| opts
.off32_limit
& 0x80000000)
1565 die(_("bad %s"), arg
);
1567 usage(index_pack_usage
);
1572 usage(index_pack_usage
);
1576 if (!pack_name
&& !from_stdin
)
1577 usage(index_pack_usage
);
1578 if (fix_thin_pack
&& !from_stdin
)
1579 die(_("--fix-thin cannot be used without --stdin"));
1580 if (!index_name
&& pack_name
) {
1581 int len
= strlen(pack_name
);
1582 if (!has_extension(pack_name
, ".pack"))
1583 die(_("packfile name '%s' does not end with '.pack'"),
1585 index_name_buf
= xmalloc(len
);
1586 memcpy(index_name_buf
, pack_name
, len
- 5);
1587 strcpy(index_name_buf
+ len
- 5, ".idx");
1588 index_name
= index_name_buf
;
1590 if (keep_msg
&& !keep_name
&& pack_name
) {
1591 int len
= strlen(pack_name
);
1592 if (!has_extension(pack_name
, ".pack"))
1593 die(_("packfile name '%s' does not end with '.pack'"),
1595 keep_name_buf
= xmalloc(len
);
1596 memcpy(keep_name_buf
, pack_name
, len
- 5);
1597 strcpy(keep_name_buf
+ len
- 5, ".keep");
1598 keep_name
= keep_name_buf
;
1602 die(_("--verify with no packfile name given"));
1603 read_idx_option(&opts
, index_name
);
1604 opts
.flags
|= WRITE_IDX_VERIFY
| WRITE_IDX_STRICT
;
1607 opts
.flags
|= WRITE_IDX_STRICT
;
1611 nr_threads
= online_cpus();
1612 /* An experiment showed that more threads does not mean faster */
1618 curr_pack
= open_pack_file(pack_name
);
1619 parse_pack_header();
1620 objects
= xcalloc(nr_objects
+ 1, sizeof(struct object_entry
));
1621 deltas
= xcalloc(nr_objects
, sizeof(struct delta_entry
));
1622 parse_pack_objects(pack_sha1
);
1624 conclude_pack(fix_thin_pack
, curr_pack
, pack_sha1
);
1630 show_pack_info(stat_only
);
1632 idx_objects
= xmalloc((nr_objects
) * sizeof(struct pack_idx_entry
*));
1633 for (i
= 0; i
< nr_objects
; i
++)
1634 idx_objects
[i
] = &objects
[i
].idx
;
1635 curr_index
= write_idx_file(index_name
, idx_objects
, nr_objects
, &opts
, pack_sha1
);
1639 final(pack_name
, curr_pack
,
1640 index_name
, curr_index
,
1641 keep_name
, keep_msg
,
1646 free(index_name_buf
);
1647 free(keep_name_buf
);
1648 if (pack_name
== NULL
)
1649 free((void *) curr_pack
);
1650 if (index_name
== NULL
)
1651 free((void *) curr_index
);