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
;
47 struct base_data
*base_cache
;
48 size_t base_cache_used
;
53 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
54 * to memcmp() only the first 20 bytes.
56 #define UNION_BASE_SZ 20
58 #define FLAG_LINK (1u<<20)
59 #define FLAG_CHECKED (1u<<21)
62 union delta_base base
;
66 static struct object_entry
*objects
;
67 static struct delta_entry
*deltas
;
68 static struct thread_local nothread_data
;
69 static int nr_objects
;
71 static int nr_resolved_deltas
;
72 static int nr_threads
;
74 static int from_stdin
;
76 static int do_fsck_object
;
79 static int check_self_contained_and_connected
;
81 static struct progress
*progress
;
83 /* We always read in 4kB chunks. */
84 static unsigned char input_buffer
[4096];
85 static unsigned int input_offset
, input_len
;
86 static off_t consumed_bytes
;
87 static unsigned deepest_delta
;
88 static git_SHA_CTX input_ctx
;
89 static uint32_t input_crc32
;
90 static int input_fd
, output_fd
;
91 static const char *curr_pack
;
95 static struct thread_local
*thread_data
;
96 static int nr_dispatched
;
97 static int threads_active
;
99 static pthread_mutex_t read_mutex
;
100 #define read_lock() lock_mutex(&read_mutex)
101 #define read_unlock() unlock_mutex(&read_mutex)
103 static pthread_mutex_t counter_mutex
;
104 #define counter_lock() lock_mutex(&counter_mutex)
105 #define counter_unlock() unlock_mutex(&counter_mutex)
107 static pthread_mutex_t work_mutex
;
108 #define work_lock() lock_mutex(&work_mutex)
109 #define work_unlock() unlock_mutex(&work_mutex)
111 static pthread_mutex_t deepest_delta_mutex
;
112 #define deepest_delta_lock() lock_mutex(&deepest_delta_mutex)
113 #define deepest_delta_unlock() unlock_mutex(&deepest_delta_mutex)
115 static pthread_mutex_t type_cas_mutex
;
116 #define type_cas_lock() lock_mutex(&type_cas_mutex)
117 #define type_cas_unlock() unlock_mutex(&type_cas_mutex)
119 static pthread_key_t key
;
121 static inline void lock_mutex(pthread_mutex_t
*mutex
)
124 pthread_mutex_lock(mutex
);
127 static inline void unlock_mutex(pthread_mutex_t
*mutex
)
130 pthread_mutex_unlock(mutex
);
134 * Mutex and conditional variable can't be statically-initialized on Windows.
136 static void init_thread(void)
139 init_recursive_mutex(&read_mutex
);
140 pthread_mutex_init(&counter_mutex
, NULL
);
141 pthread_mutex_init(&work_mutex
, NULL
);
142 pthread_mutex_init(&type_cas_mutex
, NULL
);
144 pthread_mutex_init(&deepest_delta_mutex
, NULL
);
145 pthread_key_create(&key
, NULL
);
146 thread_data
= xcalloc(nr_threads
, sizeof(*thread_data
));
147 for (i
= 0; i
< nr_threads
; i
++) {
148 thread_data
[i
].pack_fd
= open(curr_pack
, O_RDONLY
);
149 if (thread_data
[i
].pack_fd
== -1)
150 die_errno(_("unable to open %s"), curr_pack
);
156 static void cleanup_thread(void)
162 pthread_mutex_destroy(&read_mutex
);
163 pthread_mutex_destroy(&counter_mutex
);
164 pthread_mutex_destroy(&work_mutex
);
165 pthread_mutex_destroy(&type_cas_mutex
);
167 pthread_mutex_destroy(&deepest_delta_mutex
);
168 for (i
= 0; i
< nr_threads
; i
++)
169 close(thread_data
[i
].pack_fd
);
170 pthread_key_delete(key
);
177 #define read_unlock()
179 #define counter_lock()
180 #define counter_unlock()
183 #define work_unlock()
185 #define deepest_delta_lock()
186 #define deepest_delta_unlock()
191 static int mark_link(struct object
*obj
, int type
, void *data
)
196 if (type
!= OBJ_ANY
&& obj
->type
!= type
)
197 die(_("object type mismatch at %s"), sha1_to_hex(obj
->sha1
));
199 obj
->flags
|= FLAG_LINK
;
203 /* The content of each linked object must have been checked
204 or it must be already present in the object database */
205 static unsigned check_object(struct object
*obj
)
210 if (!(obj
->flags
& FLAG_LINK
))
213 if (!(obj
->flags
& FLAG_CHECKED
)) {
215 int type
= sha1_object_info(obj
->sha1
, &size
);
217 die(_("did not receive expected object %s"),
218 sha1_to_hex(obj
->sha1
));
219 if (type
!= obj
->type
)
220 die(_("object %s: expected type %s, found %s"),
221 sha1_to_hex(obj
->sha1
),
222 typename(obj
->type
), typename(type
));
223 obj
->flags
|= FLAG_CHECKED
;
230 static unsigned check_objects(void)
232 unsigned i
, max
, foreign_nr
= 0;
234 max
= get_max_object_index();
235 for (i
= 0; i
< max
; i
++)
236 foreign_nr
+= check_object(get_indexed_object(i
));
241 /* Discard current buffer used content. */
242 static void flush(void)
246 write_or_die(output_fd
, input_buffer
, input_offset
);
247 git_SHA1_Update(&input_ctx
, input_buffer
, input_offset
);
248 memmove(input_buffer
, input_buffer
+ input_offset
, input_len
);
254 * Make sure at least "min" bytes are available in the buffer, and
255 * return the pointer to the buffer.
257 static void *fill(int min
)
259 if (min
<= input_len
)
260 return input_buffer
+ input_offset
;
261 if (min
> sizeof(input_buffer
))
262 die(Q_("cannot fill %d byte",
263 "cannot fill %d bytes",
268 ssize_t ret
= xread(input_fd
, input_buffer
+ input_len
,
269 sizeof(input_buffer
) - input_len
);
273 die_errno(_("read error on input"));
277 display_throughput(progress
, consumed_bytes
+ input_len
);
278 } while (input_len
< min
);
282 static void use(int bytes
)
284 if (bytes
> input_len
)
285 die(_("used more bytes than were available"));
286 input_crc32
= crc32(input_crc32
, input_buffer
+ input_offset
, bytes
);
288 input_offset
+= bytes
;
290 /* make sure off_t is sufficiently large not to wrap */
291 if (signed_add_overflows(consumed_bytes
, bytes
))
292 die(_("pack too large for current definition of off_t"));
293 consumed_bytes
+= bytes
;
296 static const char *open_pack_file(const char *pack_name
)
301 static char tmp_file
[PATH_MAX
];
302 output_fd
= odb_mkstemp(tmp_file
, sizeof(tmp_file
),
303 "pack/tmp_pack_XXXXXX");
304 pack_name
= xstrdup(tmp_file
);
306 output_fd
= open(pack_name
, O_CREAT
|O_EXCL
|O_RDWR
, 0600);
308 die_errno(_("unable to create '%s'"), pack_name
);
309 nothread_data
.pack_fd
= output_fd
;
311 input_fd
= open(pack_name
, O_RDONLY
);
313 die_errno(_("cannot open packfile '%s'"), pack_name
);
315 nothread_data
.pack_fd
= input_fd
;
317 git_SHA1_Init(&input_ctx
);
321 static void parse_pack_header(void)
323 struct pack_header
*hdr
= fill(sizeof(struct pack_header
));
325 /* Header consistency check */
326 if (hdr
->hdr_signature
!= htonl(PACK_SIGNATURE
))
327 die(_("pack signature mismatch"));
328 if (!pack_version_ok(hdr
->hdr_version
))
329 die(_("pack version %"PRIu32
" unsupported"),
330 ntohl(hdr
->hdr_version
));
332 nr_objects
= ntohl(hdr
->hdr_entries
);
333 use(sizeof(struct pack_header
));
336 static NORETURN
void bad_object(unsigned long offset
, const char *format
,
337 ...) __attribute__((format (printf
, 2, 3)));
339 static NORETURN
void bad_object(unsigned long offset
, const char *format
, ...)
344 va_start(params
, format
);
345 vsnprintf(buf
, sizeof(buf
), format
, params
);
347 die(_("pack has bad object at offset %lu: %s"), offset
, buf
);
350 static inline struct thread_local
*get_thread_data(void)
354 return pthread_getspecific(key
);
355 assert(!threads_active
&&
356 "This should only be reached when all threads are gone");
358 return ¬hread_data
;
362 static void set_thread_data(struct thread_local
*data
)
365 pthread_setspecific(key
, data
);
369 static struct base_data
*alloc_base_data(void)
371 struct base_data
*base
= xcalloc(1, sizeof(struct base_data
));
377 static void free_base_data(struct base_data
*c
)
382 get_thread_data()->base_cache_used
-= c
->size
;
386 static void prune_base_data(struct base_data
*retain
)
389 struct thread_local
*data
= get_thread_data();
390 for (b
= data
->base_cache
;
391 data
->base_cache_used
> delta_base_cache_limit
&& b
;
393 if (b
->data
&& b
!= retain
)
398 static void link_base_data(struct base_data
*base
, struct base_data
*c
)
403 get_thread_data()->base_cache
= c
;
408 get_thread_data()->base_cache_used
+= c
->size
;
412 static void unlink_base_data(struct base_data
*c
)
414 struct base_data
*base
= c
->base
;
418 get_thread_data()->base_cache
= NULL
;
422 static int is_delta_type(enum object_type type
)
424 return (type
== OBJ_REF_DELTA
|| type
== OBJ_OFS_DELTA
);
427 static void *unpack_entry_data(unsigned long offset
, unsigned long size
,
428 enum object_type type
, unsigned char *sha1
)
430 static char fixed_buf
[8192];
438 if (!is_delta_type(type
)) {
439 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), size
) + 1;
441 git_SHA1_Update(&c
, hdr
, hdrlen
);
444 if (type
== OBJ_BLOB
&& size
> big_file_threshold
)
449 memset(&stream
, 0, sizeof(stream
));
450 git_inflate_init(&stream
);
451 stream
.next_out
= buf
;
452 stream
.avail_out
= buf
== fixed_buf
? sizeof(fixed_buf
) : size
;
455 unsigned char *last_out
= stream
.next_out
;
456 stream
.next_in
= fill(1);
457 stream
.avail_in
= input_len
;
458 status
= git_inflate(&stream
, 0);
459 use(input_len
- stream
.avail_in
);
461 git_SHA1_Update(&c
, last_out
, stream
.next_out
- last_out
);
462 if (buf
== fixed_buf
) {
463 stream
.next_out
= buf
;
464 stream
.avail_out
= sizeof(fixed_buf
);
466 } while (status
== Z_OK
);
467 if (stream
.total_out
!= size
|| status
!= Z_STREAM_END
)
468 bad_object(offset
, _("inflate returned %d"), status
);
469 git_inflate_end(&stream
);
471 git_SHA1_Final(sha1
, &c
);
472 return buf
== fixed_buf
? NULL
: buf
;
475 static void *unpack_raw_entry(struct object_entry
*obj
,
476 union delta_base
*delta_base
,
480 unsigned long size
, c
;
485 obj
->idx
.offset
= consumed_bytes
;
486 input_crc32
= crc32(0, NULL
, 0);
491 obj
->type
= (c
>> 4) & 7;
498 size
+= (c
& 0x7f) << shift
;
505 hashcpy(delta_base
->sha1
, fill(20));
509 memset(delta_base
, 0, sizeof(*delta_base
));
513 base_offset
= c
& 127;
516 if (!base_offset
|| MSB(base_offset
, 7))
517 bad_object(obj
->idx
.offset
, _("offset value overflow for delta base object"));
521 base_offset
= (base_offset
<< 7) + (c
& 127);
523 delta_base
->offset
= obj
->idx
.offset
- base_offset
;
524 if (delta_base
->offset
<= 0 || delta_base
->offset
>= obj
->idx
.offset
)
525 bad_object(obj
->idx
.offset
, _("delta base offset is out of bound"));
533 bad_object(obj
->idx
.offset
, _("unknown object type %d"), obj
->type
);
535 obj
->hdr_size
= consumed_bytes
- obj
->idx
.offset
;
537 data
= unpack_entry_data(obj
->idx
.offset
, obj
->size
, obj
->type
, sha1
);
538 obj
->idx
.crc32
= input_crc32
;
542 static void *unpack_data(struct object_entry
*obj
,
543 int (*consume
)(const unsigned char *, unsigned long, void *),
546 off_t from
= obj
[0].idx
.offset
+ obj
[0].hdr_size
;
547 unsigned long len
= obj
[1].idx
.offset
- from
;
548 unsigned char *data
, *inbuf
;
552 data
= xmalloc(consume
? 64*1024 : obj
->size
);
553 inbuf
= xmalloc((len
< 64*1024) ? len
: 64*1024);
555 memset(&stream
, 0, sizeof(stream
));
556 git_inflate_init(&stream
);
557 stream
.next_out
= data
;
558 stream
.avail_out
= consume
? 64*1024 : obj
->size
;
561 ssize_t n
= (len
< 64*1024) ? len
: 64*1024;
562 n
= xpread(get_thread_data()->pack_fd
, inbuf
, n
, from
);
564 die_errno(_("cannot pread pack file"));
566 die(Q_("premature end of pack file, %lu byte missing",
567 "premature end of pack file, %lu bytes missing",
572 stream
.next_in
= inbuf
;
575 status
= git_inflate(&stream
, 0);
578 status
= git_inflate(&stream
, 0);
579 if (consume(data
, stream
.next_out
- data
, cb_data
)) {
584 stream
.next_out
= data
;
585 stream
.avail_out
= 64*1024;
586 } while (status
== Z_OK
&& stream
.avail_in
);
588 } while (len
&& status
== Z_OK
&& !stream
.avail_in
);
590 /* This has been inflated OK when first encountered, so... */
591 if (status
!= Z_STREAM_END
|| stream
.total_out
!= obj
->size
)
592 die(_("serious inflate inconsistency"));
594 git_inflate_end(&stream
);
603 static void *get_data_from_pack(struct object_entry
*obj
)
605 return unpack_data(obj
, NULL
, NULL
);
608 static int compare_delta_bases(const union delta_base
*base1
,
609 const union delta_base
*base2
,
610 enum object_type type1
,
611 enum object_type type2
)
613 int cmp
= type1
- type2
;
616 return memcmp(base1
, base2
, UNION_BASE_SZ
);
619 static int find_delta(const union delta_base
*base
, enum object_type type
)
621 int first
= 0, last
= nr_deltas
;
623 while (first
< last
) {
624 int next
= (first
+ last
) / 2;
625 struct delta_entry
*delta
= &deltas
[next
];
628 cmp
= compare_delta_bases(base
, &delta
->base
,
629 type
, objects
[delta
->obj_no
].type
);
641 static void find_delta_children(const union delta_base
*base
,
642 int *first_index
, int *last_index
,
643 enum object_type type
)
645 int first
= find_delta(base
, type
);
647 int end
= nr_deltas
- 1;
654 while (first
> 0 && !memcmp(&deltas
[first
- 1].base
, base
, UNION_BASE_SZ
))
656 while (last
< end
&& !memcmp(&deltas
[last
+ 1].base
, base
, UNION_BASE_SZ
))
658 *first_index
= first
;
662 struct compare_data
{
663 struct object_entry
*entry
;
664 struct git_istream
*st
;
666 unsigned long buf_size
;
669 static int compare_objects(const unsigned char *buf
, unsigned long size
,
672 struct compare_data
*data
= cb_data
;
674 if (data
->buf_size
< size
) {
676 data
->buf
= xmalloc(size
);
677 data
->buf_size
= size
;
681 ssize_t len
= read_istream(data
->st
, data
->buf
, size
);
683 die(_("SHA1 COLLISION FOUND WITH %s !"),
684 sha1_to_hex(data
->entry
->idx
.sha1
));
686 die(_("unable to read %s"),
687 sha1_to_hex(data
->entry
->idx
.sha1
));
688 if (memcmp(buf
, data
->buf
, len
))
689 die(_("SHA1 COLLISION FOUND WITH %s !"),
690 sha1_to_hex(data
->entry
->idx
.sha1
));
697 static int check_collison(struct object_entry
*entry
)
699 struct compare_data data
;
700 enum object_type type
;
703 if (entry
->size
<= big_file_threshold
|| entry
->type
!= OBJ_BLOB
)
706 memset(&data
, 0, sizeof(data
));
708 data
.st
= open_istream(entry
->idx
.sha1
, &type
, &size
, NULL
);
711 if (size
!= entry
->size
|| type
!= entry
->type
)
712 die(_("SHA1 COLLISION FOUND WITH %s !"),
713 sha1_to_hex(entry
->idx
.sha1
));
714 unpack_data(entry
, compare_objects
, &data
);
715 close_istream(data
.st
);
720 static void sha1_object(const void *data
, struct object_entry
*obj_entry
,
721 unsigned long size
, enum object_type type
,
722 const unsigned char *sha1
)
724 void *new_data
= NULL
;
725 int collision_test_needed
;
727 assert(data
|| obj_entry
);
730 collision_test_needed
= has_sha1_file(sha1
);
733 if (collision_test_needed
&& !data
) {
735 if (!check_collison(obj_entry
))
736 collision_test_needed
= 0;
739 if (collision_test_needed
) {
741 enum object_type has_type
;
742 unsigned long has_size
;
744 has_type
= sha1_object_info(sha1
, &has_size
);
745 if (has_type
!= type
|| has_size
!= size
)
746 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1
));
747 has_data
= read_sha1_file(sha1
, &has_type
, &has_size
);
750 data
= new_data
= get_data_from_pack(obj_entry
);
752 die(_("cannot read existing object %s"), sha1_to_hex(sha1
));
753 if (size
!= has_size
|| type
!= has_type
||
754 memcmp(data
, has_data
, size
) != 0)
755 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1
));
761 if (type
== OBJ_BLOB
) {
762 struct blob
*blob
= lookup_blob(sha1
);
764 blob
->object
.flags
|= FLAG_CHECKED
;
766 die(_("invalid blob object %s"), sha1_to_hex(sha1
));
770 void *buf
= (void *) data
;
772 assert(data
&& "data can only be NULL for large _blobs_");
775 * we do not need to free the memory here, as the
776 * buf is deleted by the caller.
778 obj
= parse_object_buffer(sha1
, type
, size
, buf
, &eaten
);
780 die(_("invalid %s"), typename(type
));
781 if (do_fsck_object
&&
782 fsck_object(obj
, buf
, size
, 1,
783 fsck_error_function
))
784 die(_("Error in object"));
785 if (fsck_walk(obj
, mark_link
, NULL
))
786 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj
->sha1
));
788 if (obj
->type
== OBJ_TREE
) {
789 struct tree
*item
= (struct tree
*) obj
;
793 if (obj
->type
== OBJ_COMMIT
) {
794 struct commit
*commit
= (struct commit
*) obj
;
795 if (detach_commit_buffer(commit
, NULL
) != data
)
796 die("BUG: parse_object_buffer transmogrified our buffer");
798 obj
->flags
|= FLAG_CHECKED
;
807 * This function is part of find_unresolved_deltas(). There are two
808 * walkers going in the opposite ways.
810 * The first one in find_unresolved_deltas() traverses down from
811 * parent node to children, deflating nodes along the way. However,
812 * memory for deflated nodes is limited by delta_base_cache_limit, so
813 * at some point parent node's deflated content may be freed.
815 * The second walker is this function, which goes from current node up
816 * to top parent if necessary to deflate the node. In normal
817 * situation, its parent node would be already deflated, so it just
818 * needs to apply delta.
820 * In the worst case scenario, parent node is no longer deflated because
821 * we're running out of delta_base_cache_limit; we need to re-deflate
822 * parents, possibly up to the top base.
824 * All deflated objects here are subject to be freed if we exceed
825 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
826 * just need to make sure the last node is not freed.
828 static void *get_base_data(struct base_data
*c
)
831 struct object_entry
*obj
= c
->obj
;
832 struct base_data
**delta
= NULL
;
833 int delta_nr
= 0, delta_alloc
= 0;
835 while (is_delta_type(c
->obj
->type
) && !c
->data
) {
836 ALLOC_GROW(delta
, delta_nr
+ 1, delta_alloc
);
837 delta
[delta_nr
++] = c
;
841 c
->data
= get_data_from_pack(obj
);
843 get_thread_data()->base_cache_used
+= c
->size
;
846 for (; delta_nr
> 0; delta_nr
--) {
848 c
= delta
[delta_nr
- 1];
850 base
= get_base_data(c
->base
);
851 raw
= get_data_from_pack(obj
);
852 c
->data
= patch_delta(
858 bad_object(obj
->idx
.offset
, _("failed to apply delta"));
859 get_thread_data()->base_cache_used
+= c
->size
;
867 static void resolve_delta(struct object_entry
*delta_obj
,
868 struct base_data
*base
, struct base_data
*result
)
870 void *base_data
, *delta_data
;
873 delta_obj
->delta_depth
= base
->obj
->delta_depth
+ 1;
874 deepest_delta_lock();
875 if (deepest_delta
< delta_obj
->delta_depth
)
876 deepest_delta
= delta_obj
->delta_depth
;
877 deepest_delta_unlock();
879 delta_obj
->base_object_no
= base
->obj
- objects
;
880 delta_data
= get_data_from_pack(delta_obj
);
881 base_data
= get_base_data(base
);
882 result
->obj
= delta_obj
;
883 result
->data
= patch_delta(base_data
, base
->size
,
884 delta_data
, delta_obj
->size
, &result
->size
);
887 bad_object(delta_obj
->idx
.offset
, _("failed to apply delta"));
888 hash_sha1_file(result
->data
, result
->size
,
889 typename(delta_obj
->real_type
), delta_obj
->idx
.sha1
);
890 sha1_object(result
->data
, NULL
, result
->size
, delta_obj
->real_type
,
891 delta_obj
->idx
.sha1
);
893 nr_resolved_deltas
++;
898 * Standard boolean compare-and-swap: atomically check whether "*type" is
899 * "want"; if so, swap in "set" and return true. Otherwise, leave it untouched
902 static int compare_and_swap_type(enum object_type
*type
,
903 enum object_type want
,
904 enum object_type set
)
906 enum object_type old
;
917 static struct base_data
*find_unresolved_deltas_1(struct base_data
*base
,
918 struct base_data
*prev_base
)
920 if (base
->ref_last
== -1 && base
->ofs_last
== -1) {
921 union delta_base base_spec
;
923 hashcpy(base_spec
.sha1
, base
->obj
->idx
.sha1
);
924 find_delta_children(&base_spec
,
925 &base
->ref_first
, &base
->ref_last
, OBJ_REF_DELTA
);
927 memset(&base_spec
, 0, sizeof(base_spec
));
928 base_spec
.offset
= base
->obj
->idx
.offset
;
929 find_delta_children(&base_spec
,
930 &base
->ofs_first
, &base
->ofs_last
, OBJ_OFS_DELTA
);
932 if (base
->ref_last
== -1 && base
->ofs_last
== -1) {
937 link_base_data(prev_base
, base
);
940 if (base
->ref_first
<= base
->ref_last
) {
941 struct object_entry
*child
= objects
+ deltas
[base
->ref_first
].obj_no
;
942 struct base_data
*result
= alloc_base_data();
944 if (!compare_and_swap_type(&child
->real_type
, OBJ_REF_DELTA
,
945 base
->obj
->real_type
))
946 die("BUG: child->real_type != OBJ_REF_DELTA");
948 resolve_delta(child
, base
, result
);
949 if (base
->ref_first
== base
->ref_last
&& base
->ofs_last
== -1)
950 free_base_data(base
);
956 if (base
->ofs_first
<= base
->ofs_last
) {
957 struct object_entry
*child
= objects
+ deltas
[base
->ofs_first
].obj_no
;
958 struct base_data
*result
= alloc_base_data();
960 assert(child
->real_type
== OBJ_OFS_DELTA
);
961 child
->real_type
= base
->obj
->real_type
;
962 resolve_delta(child
, base
, result
);
963 if (base
->ofs_first
== base
->ofs_last
)
964 free_base_data(base
);
970 unlink_base_data(base
);
974 static void find_unresolved_deltas(struct base_data
*base
)
976 struct base_data
*new_base
, *prev_base
= NULL
;
978 new_base
= find_unresolved_deltas_1(base
, prev_base
);
988 prev_base
= base
->base
;
993 static int compare_delta_entry(const void *a
, const void *b
)
995 const struct delta_entry
*delta_a
= a
;
996 const struct delta_entry
*delta_b
= b
;
998 /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
999 return compare_delta_bases(&delta_a
->base
, &delta_b
->base
,
1000 objects
[delta_a
->obj_no
].type
,
1001 objects
[delta_b
->obj_no
].type
);
1004 static void resolve_base(struct object_entry
*obj
)
1006 struct base_data
*base_obj
= alloc_base_data();
1007 base_obj
->obj
= obj
;
1008 base_obj
->data
= NULL
;
1009 find_unresolved_deltas(base_obj
);
1013 static void *threaded_second_pass(void *data
)
1015 set_thread_data(data
);
1019 display_progress(progress
, nr_resolved_deltas
);
1022 while (nr_dispatched
< nr_objects
&&
1023 is_delta_type(objects
[nr_dispatched
].type
))
1025 if (nr_dispatched
>= nr_objects
) {
1029 i
= nr_dispatched
++;
1032 resolve_base(&objects
[i
]);
1040 * - find locations of all objects;
1041 * - calculate SHA1 of all non-delta objects;
1042 * - remember base (SHA1 or offset) for all deltas.
1044 static void parse_pack_objects(unsigned char *sha1
)
1046 int i
, nr_delays
= 0;
1047 struct delta_entry
*delta
= deltas
;
1051 progress
= start_progress(
1052 from_stdin
? _("Receiving objects") : _("Indexing objects"),
1054 for (i
= 0; i
< nr_objects
; i
++) {
1055 struct object_entry
*obj
= &objects
[i
];
1056 void *data
= unpack_raw_entry(obj
, &delta
->base
, obj
->idx
.sha1
);
1057 obj
->real_type
= obj
->type
;
1058 if (is_delta_type(obj
->type
)) {
1063 /* large blobs, check later */
1064 obj
->real_type
= OBJ_BAD
;
1067 sha1_object(data
, NULL
, obj
->size
, obj
->type
, obj
->idx
.sha1
);
1069 display_progress(progress
, i
+1);
1071 objects
[i
].idx
.offset
= consumed_bytes
;
1072 stop_progress(&progress
);
1074 /* Check pack integrity */
1076 git_SHA1_Final(sha1
, &input_ctx
);
1077 if (hashcmp(fill(20), sha1
))
1078 die(_("pack is corrupted (SHA1 mismatch)"));
1081 /* If input_fd is a file, we should have reached its end now. */
1082 if (fstat(input_fd
, &st
))
1083 die_errno(_("cannot fstat packfile"));
1084 if (S_ISREG(st
.st_mode
) &&
1085 lseek(input_fd
, 0, SEEK_CUR
) - input_len
!= st
.st_size
)
1086 die(_("pack has junk at the end"));
1088 for (i
= 0; i
< nr_objects
; i
++) {
1089 struct object_entry
*obj
= &objects
[i
];
1090 if (obj
->real_type
!= OBJ_BAD
)
1092 obj
->real_type
= obj
->type
;
1093 sha1_object(NULL
, obj
, obj
->size
, obj
->type
, obj
->idx
.sha1
);
1097 die(_("confusion beyond insanity in parse_pack_objects()"));
1102 * - for all non-delta objects, look if it is used as a base for
1104 * - if used as a base, uncompress the object and apply all deltas,
1105 * recursively checking if the resulting object is used as a base
1106 * for some more deltas.
1108 static void resolve_deltas(void)
1115 /* Sort deltas by base SHA1/offset for fast searching */
1116 qsort(deltas
, nr_deltas
, sizeof(struct delta_entry
),
1117 compare_delta_entry
);
1120 progress
= start_progress(_("Resolving deltas"), nr_deltas
);
1124 if (nr_threads
> 1 || getenv("GIT_FORCE_THREADS")) {
1126 for (i
= 0; i
< nr_threads
; i
++) {
1127 int ret
= pthread_create(&thread_data
[i
].thread
, NULL
,
1128 threaded_second_pass
, thread_data
+ i
);
1130 die(_("unable to create thread: %s"),
1133 for (i
= 0; i
< nr_threads
; i
++)
1134 pthread_join(thread_data
[i
].thread
, NULL
);
1140 for (i
= 0; i
< nr_objects
; i
++) {
1141 struct object_entry
*obj
= &objects
[i
];
1143 if (is_delta_type(obj
->type
))
1146 display_progress(progress
, nr_resolved_deltas
);
1152 * - append objects to convert thin pack to full pack if required
1153 * - write the final 20-byte SHA-1
1155 static void fix_unresolved_deltas(struct sha1file
*f
, int nr_unresolved
);
1156 static void conclude_pack(int fix_thin_pack
, const char *curr_pack
, unsigned char *pack_sha1
)
1158 if (nr_deltas
== nr_resolved_deltas
) {
1159 stop_progress(&progress
);
1160 /* Flush remaining pack final 20-byte SHA1. */
1165 if (fix_thin_pack
) {
1167 unsigned char read_sha1
[20], tail_sha1
[20];
1168 struct strbuf msg
= STRBUF_INIT
;
1169 int nr_unresolved
= nr_deltas
- nr_resolved_deltas
;
1170 int nr_objects_initial
= nr_objects
;
1171 if (nr_unresolved
<= 0)
1172 die(_("confusion beyond insanity"));
1173 REALLOC_ARRAY(objects
, nr_objects
+ nr_unresolved
+ 1);
1174 memset(objects
+ nr_objects
+ 1, 0,
1175 nr_unresolved
* sizeof(*objects
));
1176 f
= sha1fd(output_fd
, curr_pack
);
1177 fix_unresolved_deltas(f
, nr_unresolved
);
1178 strbuf_addf(&msg
, _("completed with %d local objects"),
1179 nr_objects
- nr_objects_initial
);
1180 stop_progress_msg(&progress
, msg
.buf
);
1181 strbuf_release(&msg
);
1182 sha1close(f
, tail_sha1
, 0);
1183 hashcpy(read_sha1
, pack_sha1
);
1184 fixup_pack_header_footer(output_fd
, pack_sha1
,
1185 curr_pack
, nr_objects
,
1186 read_sha1
, consumed_bytes
-20);
1187 if (hashcmp(read_sha1
, tail_sha1
) != 0)
1188 die(_("Unexpected tail checksum for %s "
1189 "(disk corruption?)"), curr_pack
);
1191 if (nr_deltas
!= nr_resolved_deltas
)
1192 die(Q_("pack has %d unresolved delta",
1193 "pack has %d unresolved deltas",
1194 nr_deltas
- nr_resolved_deltas
),
1195 nr_deltas
- nr_resolved_deltas
);
1198 static int write_compressed(struct sha1file
*f
, void *in
, unsigned int size
)
1202 unsigned char outbuf
[4096];
1204 memset(&stream
, 0, sizeof(stream
));
1205 git_deflate_init(&stream
, zlib_compression_level
);
1206 stream
.next_in
= in
;
1207 stream
.avail_in
= size
;
1210 stream
.next_out
= outbuf
;
1211 stream
.avail_out
= sizeof(outbuf
);
1212 status
= git_deflate(&stream
, Z_FINISH
);
1213 sha1write(f
, outbuf
, sizeof(outbuf
) - stream
.avail_out
);
1214 } while (status
== Z_OK
);
1216 if (status
!= Z_STREAM_END
)
1217 die(_("unable to deflate appended object (%d)"), status
);
1218 size
= stream
.total_out
;
1219 git_deflate_end(&stream
);
1223 static struct object_entry
*append_obj_to_pack(struct sha1file
*f
,
1224 const unsigned char *sha1
, void *buf
,
1225 unsigned long size
, enum object_type type
)
1227 struct object_entry
*obj
= &objects
[nr_objects
++];
1228 unsigned char header
[10];
1229 unsigned long s
= size
;
1231 unsigned char c
= (type
<< 4) | (s
& 15);
1234 header
[n
++] = c
| 0x80;
1240 sha1write(f
, header
, n
);
1242 obj
[0].hdr_size
= n
;
1244 obj
[0].real_type
= type
;
1245 obj
[1].idx
.offset
= obj
[0].idx
.offset
+ n
;
1246 obj
[1].idx
.offset
+= write_compressed(f
, buf
, size
);
1247 obj
[0].idx
.crc32
= crc32_end(f
);
1249 hashcpy(obj
->idx
.sha1
, sha1
);
1253 static int delta_pos_compare(const void *_a
, const void *_b
)
1255 struct delta_entry
*a
= *(struct delta_entry
**)_a
;
1256 struct delta_entry
*b
= *(struct delta_entry
**)_b
;
1257 return a
->obj_no
- b
->obj_no
;
1260 static void fix_unresolved_deltas(struct sha1file
*f
, int nr_unresolved
)
1262 struct delta_entry
**sorted_by_pos
;
1266 * Since many unresolved deltas may well be themselves base objects
1267 * for more unresolved deltas, we really want to include the
1268 * smallest number of base objects that would cover as much delta
1269 * as possible by picking the
1270 * trunc deltas first, allowing for other deltas to resolve without
1271 * additional base objects. Since most base objects are to be found
1272 * before deltas depending on them, a good heuristic is to start
1273 * resolving deltas in the same order as their position in the pack.
1275 sorted_by_pos
= xmalloc(nr_unresolved
* sizeof(*sorted_by_pos
));
1276 for (i
= 0; i
< nr_deltas
; i
++) {
1277 if (objects
[deltas
[i
].obj_no
].real_type
!= OBJ_REF_DELTA
)
1279 sorted_by_pos
[n
++] = &deltas
[i
];
1281 qsort(sorted_by_pos
, n
, sizeof(*sorted_by_pos
), delta_pos_compare
);
1283 for (i
= 0; i
< n
; i
++) {
1284 struct delta_entry
*d
= sorted_by_pos
[i
];
1285 enum object_type type
;
1286 struct base_data
*base_obj
= alloc_base_data();
1288 if (objects
[d
->obj_no
].real_type
!= OBJ_REF_DELTA
)
1290 base_obj
->data
= read_sha1_file(d
->base
.sha1
, &type
, &base_obj
->size
);
1291 if (!base_obj
->data
)
1294 if (check_sha1_signature(d
->base
.sha1
, base_obj
->data
,
1295 base_obj
->size
, typename(type
)))
1296 die(_("local object %s is corrupt"), sha1_to_hex(d
->base
.sha1
));
1297 base_obj
->obj
= append_obj_to_pack(f
, d
->base
.sha1
,
1298 base_obj
->data
, base_obj
->size
, type
);
1299 find_unresolved_deltas(base_obj
);
1300 display_progress(progress
, nr_resolved_deltas
);
1302 free(sorted_by_pos
);
1305 static void final(const char *final_pack_name
, const char *curr_pack_name
,
1306 const char *final_index_name
, const char *curr_index_name
,
1307 const char *keep_name
, const char *keep_msg
,
1308 unsigned char *sha1
)
1310 const char *report
= "pack";
1311 char name
[PATH_MAX
];
1317 fsync_or_die(output_fd
, curr_pack_name
);
1318 err
= close(output_fd
);
1320 die_errno(_("error while closing pack file"));
1324 int keep_fd
, keep_msg_len
= strlen(keep_msg
);
1327 keep_fd
= odb_pack_keep(name
, sizeof(name
), sha1
);
1329 keep_fd
= open(keep_name
, O_RDWR
|O_CREAT
|O_EXCL
, 0600);
1332 if (errno
!= EEXIST
)
1333 die_errno(_("cannot write keep file '%s'"),
1334 keep_name
? keep_name
: name
);
1336 if (keep_msg_len
> 0) {
1337 write_or_die(keep_fd
, keep_msg
, keep_msg_len
);
1338 write_or_die(keep_fd
, "\n", 1);
1340 if (close(keep_fd
) != 0)
1341 die_errno(_("cannot close written keep file '%s'"),
1342 keep_name
? keep_name
: name
);
1347 if (final_pack_name
!= curr_pack_name
) {
1348 if (!final_pack_name
) {
1349 snprintf(name
, sizeof(name
), "%s/pack/pack-%s.pack",
1350 get_object_directory(), sha1_to_hex(sha1
));
1351 final_pack_name
= name
;
1353 if (move_temp_to_file(curr_pack_name
, final_pack_name
))
1354 die(_("cannot store pack file"));
1355 } else if (from_stdin
)
1356 chmod(final_pack_name
, 0444);
1358 if (final_index_name
!= curr_index_name
) {
1359 if (!final_index_name
) {
1360 snprintf(name
, sizeof(name
), "%s/pack/pack-%s.idx",
1361 get_object_directory(), sha1_to_hex(sha1
));
1362 final_index_name
= name
;
1364 if (move_temp_to_file(curr_index_name
, final_index_name
))
1365 die(_("cannot store index file"));
1367 chmod(final_index_name
, 0444);
1370 printf("%s\n", sha1_to_hex(sha1
));
1373 int len
= snprintf(buf
, sizeof(buf
), "%s\t%s\n",
1374 report
, sha1_to_hex(sha1
));
1375 write_or_die(1, buf
, len
);
1378 * Let's just mimic git-unpack-objects here and write
1379 * the last part of the input buffer to stdout.
1382 err
= xwrite(1, input_buffer
+ input_offset
, input_len
);
1386 input_offset
+= err
;
1391 static int git_index_pack_config(const char *k
, const char *v
, void *cb
)
1393 struct pack_idx_option
*opts
= cb
;
1395 if (!strcmp(k
, "pack.indexversion")) {
1396 opts
->version
= git_config_int(k
, v
);
1397 if (opts
->version
> 2)
1398 die(_("bad pack.indexversion=%"PRIu32
), opts
->version
);
1401 if (!strcmp(k
, "pack.threads")) {
1402 nr_threads
= git_config_int(k
, v
);
1404 die(_("invalid number of threads specified (%d)"),
1407 if (nr_threads
!= 1)
1408 warning(_("no threads support, ignoring %s"), k
);
1413 return git_default_config(k
, v
, cb
);
1416 static int cmp_uint32(const void *a_
, const void *b_
)
1418 uint32_t a
= *((uint32_t *)a_
);
1419 uint32_t b
= *((uint32_t *)b_
);
1421 return (a
< b
) ? -1 : (a
!= b
);
1424 static void read_v2_anomalous_offsets(struct packed_git
*p
,
1425 struct pack_idx_option
*opts
)
1427 const uint32_t *idx1
, *idx2
;
1430 /* The address of the 4-byte offset table */
1431 idx1
= (((const uint32_t *)p
->index_data
)
1432 + 2 /* 8-byte header */
1434 + 5 * p
->num_objects
/* 20-byte SHA-1 table */
1435 + p
->num_objects
/* CRC32 table */
1438 /* The address of the 8-byte offset table */
1439 idx2
= idx1
+ p
->num_objects
;
1441 for (i
= 0; i
< p
->num_objects
; i
++) {
1442 uint32_t off
= ntohl(idx1
[i
]);
1443 if (!(off
& 0x80000000))
1445 off
= off
& 0x7fffffff;
1449 * The real offset is ntohl(idx2[off * 2]) in high 4
1450 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1451 * octets. But idx2[off * 2] is Zero!!!
1453 ALLOC_GROW(opts
->anomaly
, opts
->anomaly_nr
+ 1, opts
->anomaly_alloc
);
1454 opts
->anomaly
[opts
->anomaly_nr
++] = ntohl(idx2
[off
* 2 + 1]);
1457 if (1 < opts
->anomaly_nr
)
1458 qsort(opts
->anomaly
, opts
->anomaly_nr
, sizeof(uint32_t), cmp_uint32
);
1461 static void read_idx_option(struct pack_idx_option
*opts
, const char *pack_name
)
1463 struct packed_git
*p
= add_packed_git(pack_name
, strlen(pack_name
), 1);
1466 die(_("Cannot open existing pack file '%s'"), pack_name
);
1467 if (open_pack_index(p
))
1468 die(_("Cannot open existing pack idx file for '%s'"), pack_name
);
1470 /* Read the attributes from the existing idx file */
1471 opts
->version
= p
->index_version
;
1473 if (opts
->version
== 2)
1474 read_v2_anomalous_offsets(p
, opts
);
1477 * Get rid of the idx file as we do not need it anymore.
1478 * NEEDSWORK: extract this bit from free_pack_by_name() in
1479 * sha1_file.c, perhaps? It shouldn't matter very much as we
1480 * know we haven't installed this pack (hence we never have
1481 * read anything from it).
1483 close_pack_index(p
);
1487 static void show_pack_info(int stat_only
)
1489 int i
, baseobjects
= nr_objects
- nr_deltas
;
1490 unsigned long *chain_histogram
= NULL
;
1493 chain_histogram
= xcalloc(deepest_delta
, sizeof(unsigned long));
1495 for (i
= 0; i
< nr_objects
; i
++) {
1496 struct object_entry
*obj
= &objects
[i
];
1498 if (is_delta_type(obj
->type
))
1499 chain_histogram
[obj
->delta_depth
- 1]++;
1502 printf("%s %-6s %lu %lu %"PRIuMAX
,
1503 sha1_to_hex(obj
->idx
.sha1
),
1504 typename(obj
->real_type
), obj
->size
,
1505 (unsigned long)(obj
[1].idx
.offset
- obj
->idx
.offset
),
1506 (uintmax_t)obj
->idx
.offset
);
1507 if (is_delta_type(obj
->type
)) {
1508 struct object_entry
*bobj
= &objects
[obj
->base_object_no
];
1509 printf(" %u %s", obj
->delta_depth
, sha1_to_hex(bobj
->idx
.sha1
));
1515 printf_ln(Q_("non delta: %d object",
1516 "non delta: %d objects",
1519 for (i
= 0; i
< deepest_delta
; i
++) {
1520 if (!chain_histogram
[i
])
1522 printf_ln(Q_("chain length = %d: %lu object",
1523 "chain length = %d: %lu objects",
1524 chain_histogram
[i
]),
1526 chain_histogram
[i
]);
1530 int cmd_index_pack(int argc
, const char **argv
, const char *prefix
)
1532 int i
, fix_thin_pack
= 0, verify
= 0, stat_only
= 0;
1533 const char *curr_index
;
1534 const char *index_name
= NULL
, *pack_name
= NULL
;
1535 const char *keep_name
= NULL
, *keep_msg
= NULL
;
1536 struct strbuf index_name_buf
= STRBUF_INIT
,
1537 keep_name_buf
= STRBUF_INIT
;
1538 struct pack_idx_entry
**idx_objects
;
1539 struct pack_idx_option opts
;
1540 unsigned char pack_sha1
[20];
1541 unsigned foreign_nr
= 1; /* zero is a "good" value, assume bad */
1543 if (argc
== 2 && !strcmp(argv
[1], "-h"))
1544 usage(index_pack_usage
);
1546 check_replace_refs
= 0;
1548 reset_pack_idx_option(&opts
);
1549 git_config(git_index_pack_config
, &opts
);
1550 if (prefix
&& chdir(prefix
))
1551 die(_("Cannot come back to cwd"));
1553 for (i
= 1; i
< argc
; i
++) {
1554 const char *arg
= argv
[i
];
1557 if (!strcmp(arg
, "--stdin")) {
1559 } else if (!strcmp(arg
, "--fix-thin")) {
1561 } else if (!strcmp(arg
, "--strict")) {
1564 } else if (!strcmp(arg
, "--check-self-contained-and-connected")) {
1566 check_self_contained_and_connected
= 1;
1567 } else if (!strcmp(arg
, "--verify")) {
1569 } else if (!strcmp(arg
, "--verify-stat")) {
1572 } else if (!strcmp(arg
, "--verify-stat-only")) {
1576 } else if (!strcmp(arg
, "--keep")) {
1578 } else if (starts_with(arg
, "--keep=")) {
1580 } else if (starts_with(arg
, "--threads=")) {
1582 nr_threads
= strtoul(arg
+10, &end
, 0);
1583 if (!arg
[10] || *end
|| nr_threads
< 0)
1584 usage(index_pack_usage
);
1586 if (nr_threads
!= 1)
1587 warning(_("no threads support, "
1588 "ignoring %s"), arg
);
1591 } else if (starts_with(arg
, "--pack_header=")) {
1592 struct pack_header
*hdr
;
1595 hdr
= (struct pack_header
*)input_buffer
;
1596 hdr
->hdr_signature
= htonl(PACK_SIGNATURE
);
1597 hdr
->hdr_version
= htonl(strtoul(arg
+ 14, &c
, 10));
1599 die(_("bad %s"), arg
);
1600 hdr
->hdr_entries
= htonl(strtoul(c
+ 1, &c
, 10));
1602 die(_("bad %s"), arg
);
1603 input_len
= sizeof(*hdr
);
1604 } else if (!strcmp(arg
, "-v")) {
1606 } else if (!strcmp(arg
, "-o")) {
1607 if (index_name
|| (i
+1) >= argc
)
1608 usage(index_pack_usage
);
1609 index_name
= argv
[++i
];
1610 } else if (starts_with(arg
, "--index-version=")) {
1612 opts
.version
= strtoul(arg
+ 16, &c
, 10);
1613 if (opts
.version
> 2)
1614 die(_("bad %s"), arg
);
1616 opts
.off32_limit
= strtoul(c
+1, &c
, 0);
1617 if (*c
|| opts
.off32_limit
& 0x80000000)
1618 die(_("bad %s"), arg
);
1620 usage(index_pack_usage
);
1625 usage(index_pack_usage
);
1629 if (!pack_name
&& !from_stdin
)
1630 usage(index_pack_usage
);
1631 if (fix_thin_pack
&& !from_stdin
)
1632 die(_("--fix-thin cannot be used without --stdin"));
1633 if (!index_name
&& pack_name
) {
1635 if (!strip_suffix(pack_name
, ".pack", &len
))
1636 die(_("packfile name '%s' does not end with '.pack'"),
1638 strbuf_add(&index_name_buf
, pack_name
, len
);
1639 strbuf_addstr(&index_name_buf
, ".idx");
1640 index_name
= index_name_buf
.buf
;
1642 if (keep_msg
&& !keep_name
&& pack_name
) {
1644 if (!strip_suffix(pack_name
, ".pack", &len
))
1645 die(_("packfile name '%s' does not end with '.pack'"),
1647 strbuf_add(&keep_name_buf
, pack_name
, len
);
1648 strbuf_addstr(&keep_name_buf
, ".idx");
1649 keep_name
= keep_name_buf
.buf
;
1653 die(_("--verify with no packfile name given"));
1654 read_idx_option(&opts
, index_name
);
1655 opts
.flags
|= WRITE_IDX_VERIFY
| WRITE_IDX_STRICT
;
1658 opts
.flags
|= WRITE_IDX_STRICT
;
1662 nr_threads
= online_cpus();
1663 /* An experiment showed that more threads does not mean faster */
1669 curr_pack
= open_pack_file(pack_name
);
1670 parse_pack_header();
1671 objects
= xcalloc(nr_objects
+ 1, sizeof(struct object_entry
));
1672 deltas
= xcalloc(nr_objects
, sizeof(struct delta_entry
));
1673 parse_pack_objects(pack_sha1
);
1675 conclude_pack(fix_thin_pack
, curr_pack
, pack_sha1
);
1678 foreign_nr
= check_objects();
1681 show_pack_info(stat_only
);
1683 idx_objects
= xmalloc((nr_objects
) * sizeof(struct pack_idx_entry
*));
1684 for (i
= 0; i
< nr_objects
; i
++)
1685 idx_objects
[i
] = &objects
[i
].idx
;
1686 curr_index
= write_idx_file(index_name
, idx_objects
, nr_objects
, &opts
, pack_sha1
);
1690 final(pack_name
, curr_pack
,
1691 index_name
, curr_index
,
1692 keep_name
, keep_msg
,
1697 strbuf_release(&index_name_buf
);
1698 strbuf_release(&keep_name_buf
);
1699 if (pack_name
== NULL
)
1700 free((void *) curr_pack
);
1701 if (index_name
== NULL
)
1702 free((void *) curr_index
);
1705 * Let the caller know this pack is not self contained
1707 if (check_self_contained_and_connected
&& foreign_nr
)