git p4 test: split up big t9800 test
[git/dscho.git] / builtin / index-pack.c
blobdc2cfe6e6f63b6628f8a358f2f3b3e65c14e3b8d
1 #include "builtin.h"
2 #include "delta.h"
3 #include "pack.h"
4 #include "csum-file.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "progress.h"
10 #include "fsck.h"
11 #include "exec_cmd.h"
12 #include "thread-utils.h"
14 static const char index_pack_usage[] =
15 "git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
17 struct object_entry {
18 struct pack_idx_entry idx;
19 unsigned long size;
20 unsigned int hdr_size;
21 enum object_type type;
22 enum object_type real_type;
23 unsigned delta_depth;
24 int base_object_no;
27 union delta_base {
28 unsigned char sha1[20];
29 off_t offset;
32 struct base_data {
33 struct base_data *base;
34 struct base_data *child;
35 struct object_entry *obj;
36 void *data;
37 unsigned long size;
38 int ref_first, ref_last;
39 int ofs_first, ofs_last;
42 #if !defined(NO_PTHREADS) && defined(NO_PREAD)
43 /* NO_PREAD uses compat/pread.c, which is not thread-safe. Disable threading. */
44 #define NO_PTHREADS
45 #endif
47 struct thread_local {
48 #ifndef NO_PTHREADS
49 pthread_t thread;
50 #endif
51 struct base_data *base_cache;
52 size_t base_cache_used;
56 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
57 * to memcmp() only the first 20 bytes.
59 #define UNION_BASE_SZ 20
61 #define FLAG_LINK (1u<<20)
62 #define FLAG_CHECKED (1u<<21)
64 struct delta_entry {
65 union delta_base base;
66 int obj_no;
69 static struct object_entry *objects;
70 static struct delta_entry *deltas;
71 static struct thread_local nothread_data;
72 static int nr_objects;
73 static int nr_deltas;
74 static int nr_resolved_deltas;
75 static int nr_threads;
77 static int from_stdin;
78 static int strict;
79 static int verbose;
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, pack_fd;
92 #ifndef NO_PTHREADS
94 static struct thread_local *thread_data;
95 static int nr_dispatched;
96 static int threads_active;
98 static pthread_mutex_t read_mutex;
99 #define read_lock() lock_mutex(&read_mutex)
100 #define read_unlock() unlock_mutex(&read_mutex)
102 static pthread_mutex_t counter_mutex;
103 #define counter_lock() lock_mutex(&counter_mutex)
104 #define counter_unlock() unlock_mutex(&counter_mutex)
106 static pthread_mutex_t work_mutex;
107 #define work_lock() lock_mutex(&work_mutex)
108 #define work_unlock() unlock_mutex(&work_mutex)
110 static pthread_key_t key;
112 static inline void lock_mutex(pthread_mutex_t *mutex)
114 if (threads_active)
115 pthread_mutex_lock(mutex);
118 static inline void unlock_mutex(pthread_mutex_t *mutex)
120 if (threads_active)
121 pthread_mutex_unlock(mutex);
125 * Mutex and conditional variable can't be statically-initialized on Windows.
127 static void init_thread(void)
129 init_recursive_mutex(&read_mutex);
130 pthread_mutex_init(&counter_mutex, NULL);
131 pthread_mutex_init(&work_mutex, NULL);
132 pthread_key_create(&key, NULL);
133 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
134 threads_active = 1;
137 static void cleanup_thread(void)
139 if (!threads_active)
140 return;
141 threads_active = 0;
142 pthread_mutex_destroy(&read_mutex);
143 pthread_mutex_destroy(&counter_mutex);
144 pthread_mutex_destroy(&work_mutex);
145 pthread_key_delete(key);
146 free(thread_data);
149 #else
151 #define read_lock()
152 #define read_unlock()
154 #define counter_lock()
155 #define counter_unlock()
157 #define work_lock()
158 #define work_unlock()
160 #endif
163 static int mark_link(struct object *obj, int type, void *data)
165 if (!obj)
166 return -1;
168 if (type != OBJ_ANY && obj->type != type)
169 die(_("object type mismatch at %s"), sha1_to_hex(obj->sha1));
171 obj->flags |= FLAG_LINK;
172 return 0;
175 /* The content of each linked object must have been checked
176 or it must be already present in the object database */
177 static void check_object(struct object *obj)
179 if (!obj)
180 return;
182 if (!(obj->flags & FLAG_LINK))
183 return;
185 if (!(obj->flags & FLAG_CHECKED)) {
186 unsigned long size;
187 int type = sha1_object_info(obj->sha1, &size);
188 if (type != obj->type || type <= 0)
189 die(_("object of unexpected type"));
190 obj->flags |= FLAG_CHECKED;
191 return;
195 static void check_objects(void)
197 unsigned i, max;
199 max = get_max_object_index();
200 for (i = 0; i < max; i++)
201 check_object(get_indexed_object(i));
205 /* Discard current buffer used content. */
206 static void flush(void)
208 if (input_offset) {
209 if (output_fd >= 0)
210 write_or_die(output_fd, input_buffer, input_offset);
211 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
212 memmove(input_buffer, input_buffer + input_offset, input_len);
213 input_offset = 0;
218 * Make sure at least "min" bytes are available in the buffer, and
219 * return the pointer to the buffer.
221 static void *fill(int min)
223 if (min <= input_len)
224 return input_buffer + input_offset;
225 if (min > sizeof(input_buffer))
226 die(Q_("cannot fill %d byte",
227 "cannot fill %d bytes",
228 min),
229 min);
230 flush();
231 do {
232 ssize_t ret = xread(input_fd, input_buffer + input_len,
233 sizeof(input_buffer) - input_len);
234 if (ret <= 0) {
235 if (!ret)
236 die(_("early EOF"));
237 die_errno(_("read error on input"));
239 input_len += ret;
240 if (from_stdin)
241 display_throughput(progress, consumed_bytes + input_len);
242 } while (input_len < min);
243 return input_buffer;
246 static void use(int bytes)
248 if (bytes > input_len)
249 die(_("used more bytes than were available"));
250 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
251 input_len -= bytes;
252 input_offset += bytes;
254 /* make sure off_t is sufficiently large not to wrap */
255 if (signed_add_overflows(consumed_bytes, bytes))
256 die(_("pack too large for current definition of off_t"));
257 consumed_bytes += bytes;
260 static const char *open_pack_file(const char *pack_name)
262 if (from_stdin) {
263 input_fd = 0;
264 if (!pack_name) {
265 static char tmp_file[PATH_MAX];
266 output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
267 "pack/tmp_pack_XXXXXX");
268 pack_name = xstrdup(tmp_file);
269 } else
270 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
271 if (output_fd < 0)
272 die_errno(_("unable to create '%s'"), pack_name);
273 pack_fd = output_fd;
274 } else {
275 input_fd = open(pack_name, O_RDONLY);
276 if (input_fd < 0)
277 die_errno(_("cannot open packfile '%s'"), pack_name);
278 output_fd = -1;
279 pack_fd = input_fd;
281 git_SHA1_Init(&input_ctx);
282 return pack_name;
285 static void parse_pack_header(void)
287 struct pack_header *hdr = fill(sizeof(struct pack_header));
289 /* Header consistency check */
290 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
291 die(_("pack signature mismatch"));
292 if (!pack_version_ok(hdr->hdr_version))
293 die("pack version %"PRIu32" unsupported",
294 ntohl(hdr->hdr_version));
296 nr_objects = ntohl(hdr->hdr_entries);
297 use(sizeof(struct pack_header));
300 static NORETURN void bad_object(unsigned long offset, const char *format,
301 ...) __attribute__((format (printf, 2, 3)));
303 static NORETURN void bad_object(unsigned long offset, const char *format, ...)
305 va_list params;
306 char buf[1024];
308 va_start(params, format);
309 vsnprintf(buf, sizeof(buf), format, params);
310 va_end(params);
311 die(_("pack has bad object at offset %lu: %s"), offset, buf);
314 static inline struct thread_local *get_thread_data(void)
316 #ifndef NO_PTHREADS
317 if (threads_active)
318 return pthread_getspecific(key);
319 assert(!threads_active &&
320 "This should only be reached when all threads are gone");
321 #endif
322 return &nothread_data;
325 #ifndef NO_PTHREADS
326 static void set_thread_data(struct thread_local *data)
328 if (threads_active)
329 pthread_setspecific(key, data);
331 #endif
333 static struct base_data *alloc_base_data(void)
335 struct base_data *base = xmalloc(sizeof(struct base_data));
336 memset(base, 0, sizeof(*base));
337 base->ref_last = -1;
338 base->ofs_last = -1;
339 return base;
342 static void free_base_data(struct base_data *c)
344 if (c->data) {
345 free(c->data);
346 c->data = NULL;
347 get_thread_data()->base_cache_used -= c->size;
351 static void prune_base_data(struct base_data *retain)
353 struct base_data *b;
354 struct thread_local *data = get_thread_data();
355 for (b = data->base_cache;
356 data->base_cache_used > delta_base_cache_limit && b;
357 b = b->child) {
358 if (b->data && b != retain)
359 free_base_data(b);
363 static void link_base_data(struct base_data *base, struct base_data *c)
365 if (base)
366 base->child = c;
367 else
368 get_thread_data()->base_cache = c;
370 c->base = base;
371 c->child = NULL;
372 if (c->data)
373 get_thread_data()->base_cache_used += c->size;
374 prune_base_data(c);
377 static void unlink_base_data(struct base_data *c)
379 struct base_data *base = c->base;
380 if (base)
381 base->child = NULL;
382 else
383 get_thread_data()->base_cache = NULL;
384 free_base_data(c);
387 static void *unpack_entry_data(unsigned long offset, unsigned long size)
389 int status;
390 git_zstream stream;
391 void *buf = xmalloc(size);
393 memset(&stream, 0, sizeof(stream));
394 git_inflate_init(&stream);
395 stream.next_out = buf;
396 stream.avail_out = size;
398 do {
399 stream.next_in = fill(1);
400 stream.avail_in = input_len;
401 status = git_inflate(&stream, 0);
402 use(input_len - stream.avail_in);
403 } while (status == Z_OK);
404 if (stream.total_out != size || status != Z_STREAM_END)
405 bad_object(offset, _("inflate returned %d"), status);
406 git_inflate_end(&stream);
407 return buf;
410 static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
412 unsigned char *p;
413 unsigned long size, c;
414 off_t base_offset;
415 unsigned shift;
416 void *data;
418 obj->idx.offset = consumed_bytes;
419 input_crc32 = crc32(0, NULL, 0);
421 p = fill(1);
422 c = *p;
423 use(1);
424 obj->type = (c >> 4) & 7;
425 size = (c & 15);
426 shift = 4;
427 while (c & 0x80) {
428 p = fill(1);
429 c = *p;
430 use(1);
431 size += (c & 0x7f) << shift;
432 shift += 7;
434 obj->size = size;
436 switch (obj->type) {
437 case OBJ_REF_DELTA:
438 hashcpy(delta_base->sha1, fill(20));
439 use(20);
440 break;
441 case OBJ_OFS_DELTA:
442 memset(delta_base, 0, sizeof(*delta_base));
443 p = fill(1);
444 c = *p;
445 use(1);
446 base_offset = c & 127;
447 while (c & 128) {
448 base_offset += 1;
449 if (!base_offset || MSB(base_offset, 7))
450 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
451 p = fill(1);
452 c = *p;
453 use(1);
454 base_offset = (base_offset << 7) + (c & 127);
456 delta_base->offset = obj->idx.offset - base_offset;
457 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
458 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
459 break;
460 case OBJ_COMMIT:
461 case OBJ_TREE:
462 case OBJ_BLOB:
463 case OBJ_TAG:
464 break;
465 default:
466 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
468 obj->hdr_size = consumed_bytes - obj->idx.offset;
470 data = unpack_entry_data(obj->idx.offset, obj->size);
471 obj->idx.crc32 = input_crc32;
472 return data;
475 static void *get_data_from_pack(struct object_entry *obj)
477 off_t from = obj[0].idx.offset + obj[0].hdr_size;
478 unsigned long len = obj[1].idx.offset - from;
479 unsigned char *data, *inbuf;
480 git_zstream stream;
481 int status;
483 data = xmalloc(obj->size);
484 inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
486 memset(&stream, 0, sizeof(stream));
487 git_inflate_init(&stream);
488 stream.next_out = data;
489 stream.avail_out = obj->size;
491 do {
492 ssize_t n = (len < 64*1024) ? len : 64*1024;
493 n = pread(pack_fd, inbuf, n, from);
494 if (n < 0)
495 die_errno(_("cannot pread pack file"));
496 if (!n)
497 die(Q_("premature end of pack file, %lu byte missing",
498 "premature end of pack file, %lu bytes missing",
499 len),
500 len);
501 from += n;
502 len -= n;
503 stream.next_in = inbuf;
504 stream.avail_in = n;
505 status = git_inflate(&stream, 0);
506 } while (len && status == Z_OK && !stream.avail_in);
508 /* This has been inflated OK when first encountered, so... */
509 if (status != Z_STREAM_END || stream.total_out != obj->size)
510 die(_("serious inflate inconsistency"));
512 git_inflate_end(&stream);
513 free(inbuf);
514 return data;
517 static int compare_delta_bases(const union delta_base *base1,
518 const union delta_base *base2,
519 enum object_type type1,
520 enum object_type type2)
522 int cmp = type1 - type2;
523 if (cmp)
524 return cmp;
525 return memcmp(base1, base2, UNION_BASE_SZ);
528 static int find_delta(const union delta_base *base, enum object_type type)
530 int first = 0, last = nr_deltas;
532 while (first < last) {
533 int next = (first + last) / 2;
534 struct delta_entry *delta = &deltas[next];
535 int cmp;
537 cmp = compare_delta_bases(base, &delta->base,
538 type, objects[delta->obj_no].type);
539 if (!cmp)
540 return next;
541 if (cmp < 0) {
542 last = next;
543 continue;
545 first = next+1;
547 return -first-1;
550 static void find_delta_children(const union delta_base *base,
551 int *first_index, int *last_index,
552 enum object_type type)
554 int first = find_delta(base, type);
555 int last = first;
556 int end = nr_deltas - 1;
558 if (first < 0) {
559 *first_index = 0;
560 *last_index = -1;
561 return;
563 while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
564 --first;
565 while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
566 ++last;
567 *first_index = first;
568 *last_index = last;
571 static void sha1_object(const void *data, unsigned long size,
572 enum object_type type, unsigned char *sha1)
574 hash_sha1_file(data, size, typename(type), sha1);
575 read_lock();
576 if (has_sha1_file(sha1)) {
577 void *has_data;
578 enum object_type has_type;
579 unsigned long has_size;
580 has_data = read_sha1_file(sha1, &has_type, &has_size);
581 read_unlock();
582 if (!has_data)
583 die(_("cannot read existing object %s"), sha1_to_hex(sha1));
584 if (size != has_size || type != has_type ||
585 memcmp(data, has_data, size) != 0)
586 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
587 free(has_data);
588 } else
589 read_unlock();
591 if (strict) {
592 read_lock();
593 if (type == OBJ_BLOB) {
594 struct blob *blob = lookup_blob(sha1);
595 if (blob)
596 blob->object.flags |= FLAG_CHECKED;
597 else
598 die(_("invalid blob object %s"), sha1_to_hex(sha1));
599 } else {
600 struct object *obj;
601 int eaten;
602 void *buf = (void *) data;
605 * we do not need to free the memory here, as the
606 * buf is deleted by the caller.
608 obj = parse_object_buffer(sha1, type, size, buf, &eaten);
609 if (!obj)
610 die(_("invalid %s"), typename(type));
611 if (fsck_object(obj, 1, fsck_error_function))
612 die(_("Error in object"));
613 if (fsck_walk(obj, mark_link, NULL))
614 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj->sha1));
616 if (obj->type == OBJ_TREE) {
617 struct tree *item = (struct tree *) obj;
618 item->buffer = NULL;
620 if (obj->type == OBJ_COMMIT) {
621 struct commit *commit = (struct commit *) obj;
622 commit->buffer = NULL;
624 obj->flags |= FLAG_CHECKED;
626 read_unlock();
630 static int is_delta_type(enum object_type type)
632 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
636 * This function is part of find_unresolved_deltas(). There are two
637 * walkers going in the opposite ways.
639 * The first one in find_unresolved_deltas() traverses down from
640 * parent node to children, deflating nodes along the way. However,
641 * memory for deflated nodes is limited by delta_base_cache_limit, so
642 * at some point parent node's deflated content may be freed.
644 * The second walker is this function, which goes from current node up
645 * to top parent if necessary to deflate the node. In normal
646 * situation, its parent node would be already deflated, so it just
647 * needs to apply delta.
649 * In the worst case scenario, parent node is no longer deflated because
650 * we're running out of delta_base_cache_limit; we need to re-deflate
651 * parents, possibly up to the top base.
653 * All deflated objects here are subject to be freed if we exceed
654 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
655 * just need to make sure the last node is not freed.
657 static void *get_base_data(struct base_data *c)
659 if (!c->data) {
660 struct object_entry *obj = c->obj;
661 struct base_data **delta = NULL;
662 int delta_nr = 0, delta_alloc = 0;
664 while (is_delta_type(c->obj->type) && !c->data) {
665 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
666 delta[delta_nr++] = c;
667 c = c->base;
669 if (!delta_nr) {
670 c->data = get_data_from_pack(obj);
671 c->size = obj->size;
672 get_thread_data()->base_cache_used += c->size;
673 prune_base_data(c);
675 for (; delta_nr > 0; delta_nr--) {
676 void *base, *raw;
677 c = delta[delta_nr - 1];
678 obj = c->obj;
679 base = get_base_data(c->base);
680 raw = get_data_from_pack(obj);
681 c->data = patch_delta(
682 base, c->base->size,
683 raw, obj->size,
684 &c->size);
685 free(raw);
686 if (!c->data)
687 bad_object(obj->idx.offset, _("failed to apply delta"));
688 get_thread_data()->base_cache_used += c->size;
689 prune_base_data(c);
691 free(delta);
693 return c->data;
696 static void resolve_delta(struct object_entry *delta_obj,
697 struct base_data *base, struct base_data *result)
699 void *base_data, *delta_data;
701 delta_obj->real_type = base->obj->real_type;
702 delta_obj->delta_depth = base->obj->delta_depth + 1;
703 if (deepest_delta < delta_obj->delta_depth)
704 deepest_delta = delta_obj->delta_depth;
705 delta_obj->base_object_no = base->obj - objects;
706 delta_data = get_data_from_pack(delta_obj);
707 base_data = get_base_data(base);
708 result->obj = delta_obj;
709 result->data = patch_delta(base_data, base->size,
710 delta_data, delta_obj->size, &result->size);
711 free(delta_data);
712 if (!result->data)
713 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
714 sha1_object(result->data, result->size, delta_obj->real_type,
715 delta_obj->idx.sha1);
716 counter_lock();
717 nr_resolved_deltas++;
718 counter_unlock();
721 static struct base_data *find_unresolved_deltas_1(struct base_data *base,
722 struct base_data *prev_base)
724 if (base->ref_last == -1 && base->ofs_last == -1) {
725 union delta_base base_spec;
727 hashcpy(base_spec.sha1, base->obj->idx.sha1);
728 find_delta_children(&base_spec,
729 &base->ref_first, &base->ref_last, OBJ_REF_DELTA);
731 memset(&base_spec, 0, sizeof(base_spec));
732 base_spec.offset = base->obj->idx.offset;
733 find_delta_children(&base_spec,
734 &base->ofs_first, &base->ofs_last, OBJ_OFS_DELTA);
736 if (base->ref_last == -1 && base->ofs_last == -1) {
737 free(base->data);
738 return NULL;
741 link_base_data(prev_base, base);
744 if (base->ref_first <= base->ref_last) {
745 struct object_entry *child = objects + deltas[base->ref_first].obj_no;
746 struct base_data *result = alloc_base_data();
748 assert(child->real_type == OBJ_REF_DELTA);
749 resolve_delta(child, base, result);
750 if (base->ref_first == base->ref_last && base->ofs_last == -1)
751 free_base_data(base);
753 base->ref_first++;
754 return result;
757 if (base->ofs_first <= base->ofs_last) {
758 struct object_entry *child = objects + deltas[base->ofs_first].obj_no;
759 struct base_data *result = alloc_base_data();
761 assert(child->real_type == OBJ_OFS_DELTA);
762 resolve_delta(child, base, result);
763 if (base->ofs_first == base->ofs_last)
764 free_base_data(base);
766 base->ofs_first++;
767 return result;
770 unlink_base_data(base);
771 return NULL;
774 static void find_unresolved_deltas(struct base_data *base)
776 struct base_data *new_base, *prev_base = NULL;
777 for (;;) {
778 new_base = find_unresolved_deltas_1(base, prev_base);
780 if (new_base) {
781 prev_base = base;
782 base = new_base;
783 } else {
784 free(base);
785 base = prev_base;
786 if (!base)
787 return;
788 prev_base = base->base;
793 static int compare_delta_entry(const void *a, const void *b)
795 const struct delta_entry *delta_a = a;
796 const struct delta_entry *delta_b = b;
798 /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
799 return compare_delta_bases(&delta_a->base, &delta_b->base,
800 objects[delta_a->obj_no].type,
801 objects[delta_b->obj_no].type);
804 static void resolve_base(struct object_entry *obj)
806 struct base_data *base_obj = alloc_base_data();
807 base_obj->obj = obj;
808 base_obj->data = NULL;
809 find_unresolved_deltas(base_obj);
812 #ifndef NO_PTHREADS
813 static void *threaded_second_pass(void *data)
815 set_thread_data(data);
816 for (;;) {
817 int i;
818 work_lock();
819 display_progress(progress, nr_resolved_deltas);
820 while (nr_dispatched < nr_objects &&
821 is_delta_type(objects[nr_dispatched].type))
822 nr_dispatched++;
823 if (nr_dispatched >= nr_objects) {
824 work_unlock();
825 break;
827 i = nr_dispatched++;
828 work_unlock();
830 resolve_base(&objects[i]);
832 return NULL;
834 #endif
837 * First pass:
838 * - find locations of all objects;
839 * - calculate SHA1 of all non-delta objects;
840 * - remember base (SHA1 or offset) for all deltas.
842 static void parse_pack_objects(unsigned char *sha1)
844 int i;
845 struct delta_entry *delta = deltas;
846 struct stat st;
848 if (verbose)
849 progress = start_progress(
850 from_stdin ? _("Receiving objects") : _("Indexing objects"),
851 nr_objects);
852 for (i = 0; i < nr_objects; i++) {
853 struct object_entry *obj = &objects[i];
854 void *data = unpack_raw_entry(obj, &delta->base);
855 obj->real_type = obj->type;
856 if (is_delta_type(obj->type)) {
857 nr_deltas++;
858 delta->obj_no = i;
859 delta++;
860 } else
861 sha1_object(data, obj->size, obj->type, obj->idx.sha1);
862 free(data);
863 display_progress(progress, i+1);
865 objects[i].idx.offset = consumed_bytes;
866 stop_progress(&progress);
868 /* Check pack integrity */
869 flush();
870 git_SHA1_Final(sha1, &input_ctx);
871 if (hashcmp(fill(20), sha1))
872 die(_("pack is corrupted (SHA1 mismatch)"));
873 use(20);
875 /* If input_fd is a file, we should have reached its end now. */
876 if (fstat(input_fd, &st))
877 die_errno(_("cannot fstat packfile"));
878 if (S_ISREG(st.st_mode) &&
879 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
880 die(_("pack has junk at the end"));
884 * Second pass:
885 * - for all non-delta objects, look if it is used as a base for
886 * deltas;
887 * - if used as a base, uncompress the object and apply all deltas,
888 * recursively checking if the resulting object is used as a base
889 * for some more deltas.
891 static void resolve_deltas(void)
893 int i;
895 if (!nr_deltas)
896 return;
898 /* Sort deltas by base SHA1/offset for fast searching */
899 qsort(deltas, nr_deltas, sizeof(struct delta_entry),
900 compare_delta_entry);
902 if (verbose)
903 progress = start_progress(_("Resolving deltas"), nr_deltas);
905 #ifndef NO_PTHREADS
906 nr_dispatched = 0;
907 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
908 init_thread();
909 for (i = 0; i < nr_threads; i++) {
910 int ret = pthread_create(&thread_data[i].thread, NULL,
911 threaded_second_pass, thread_data + i);
912 if (ret)
913 die("unable to create thread: %s", strerror(ret));
915 for (i = 0; i < nr_threads; i++)
916 pthread_join(thread_data[i].thread, NULL);
917 cleanup_thread();
918 return;
920 #endif
922 for (i = 0; i < nr_objects; i++) {
923 struct object_entry *obj = &objects[i];
925 if (is_delta_type(obj->type))
926 continue;
927 resolve_base(obj);
928 display_progress(progress, nr_resolved_deltas);
933 * Third pass:
934 * - append objects to convert thin pack to full pack if required
935 * - write the final 20-byte SHA-1
937 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved);
938 static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
940 if (nr_deltas == nr_resolved_deltas) {
941 stop_progress(&progress);
942 /* Flush remaining pack final 20-byte SHA1. */
943 flush();
944 return;
947 if (fix_thin_pack) {
948 struct sha1file *f;
949 unsigned char read_sha1[20], tail_sha1[20];
950 char msg[48];
951 int nr_unresolved = nr_deltas - nr_resolved_deltas;
952 int nr_objects_initial = nr_objects;
953 if (nr_unresolved <= 0)
954 die(_("confusion beyond insanity"));
955 objects = xrealloc(objects,
956 (nr_objects + nr_unresolved + 1)
957 * sizeof(*objects));
958 f = sha1fd(output_fd, curr_pack);
959 fix_unresolved_deltas(f, nr_unresolved);
960 sprintf(msg, "completed with %d local objects",
961 nr_objects - nr_objects_initial);
962 stop_progress_msg(&progress, msg);
963 sha1close(f, tail_sha1, 0);
964 hashcpy(read_sha1, pack_sha1);
965 fixup_pack_header_footer(output_fd, pack_sha1,
966 curr_pack, nr_objects,
967 read_sha1, consumed_bytes-20);
968 if (hashcmp(read_sha1, tail_sha1) != 0)
969 die("Unexpected tail checksum for %s "
970 "(disk corruption?)", curr_pack);
972 if (nr_deltas != nr_resolved_deltas)
973 die(Q_("pack has %d unresolved delta",
974 "pack has %d unresolved deltas",
975 nr_deltas - nr_resolved_deltas),
976 nr_deltas - nr_resolved_deltas);
979 static int write_compressed(struct sha1file *f, void *in, unsigned int size)
981 git_zstream stream;
982 int status;
983 unsigned char outbuf[4096];
985 memset(&stream, 0, sizeof(stream));
986 git_deflate_init(&stream, zlib_compression_level);
987 stream.next_in = in;
988 stream.avail_in = size;
990 do {
991 stream.next_out = outbuf;
992 stream.avail_out = sizeof(outbuf);
993 status = git_deflate(&stream, Z_FINISH);
994 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
995 } while (status == Z_OK);
997 if (status != Z_STREAM_END)
998 die(_("unable to deflate appended object (%d)"), status);
999 size = stream.total_out;
1000 git_deflate_end(&stream);
1001 return size;
1004 static struct object_entry *append_obj_to_pack(struct sha1file *f,
1005 const unsigned char *sha1, void *buf,
1006 unsigned long size, enum object_type type)
1008 struct object_entry *obj = &objects[nr_objects++];
1009 unsigned char header[10];
1010 unsigned long s = size;
1011 int n = 0;
1012 unsigned char c = (type << 4) | (s & 15);
1013 s >>= 4;
1014 while (s) {
1015 header[n++] = c | 0x80;
1016 c = s & 0x7f;
1017 s >>= 7;
1019 header[n++] = c;
1020 crc32_begin(f);
1021 sha1write(f, header, n);
1022 obj[0].size = size;
1023 obj[0].hdr_size = n;
1024 obj[0].type = type;
1025 obj[0].real_type = type;
1026 obj[1].idx.offset = obj[0].idx.offset + n;
1027 obj[1].idx.offset += write_compressed(f, buf, size);
1028 obj[0].idx.crc32 = crc32_end(f);
1029 sha1flush(f);
1030 hashcpy(obj->idx.sha1, sha1);
1031 return obj;
1034 static int delta_pos_compare(const void *_a, const void *_b)
1036 struct delta_entry *a = *(struct delta_entry **)_a;
1037 struct delta_entry *b = *(struct delta_entry **)_b;
1038 return a->obj_no - b->obj_no;
1041 static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
1043 struct delta_entry **sorted_by_pos;
1044 int i, n = 0;
1047 * Since many unresolved deltas may well be themselves base objects
1048 * for more unresolved deltas, we really want to include the
1049 * smallest number of base objects that would cover as much delta
1050 * as possible by picking the
1051 * trunc deltas first, allowing for other deltas to resolve without
1052 * additional base objects. Since most base objects are to be found
1053 * before deltas depending on them, a good heuristic is to start
1054 * resolving deltas in the same order as their position in the pack.
1056 sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
1057 for (i = 0; i < nr_deltas; i++) {
1058 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
1059 continue;
1060 sorted_by_pos[n++] = &deltas[i];
1062 qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
1064 for (i = 0; i < n; i++) {
1065 struct delta_entry *d = sorted_by_pos[i];
1066 enum object_type type;
1067 struct base_data *base_obj = alloc_base_data();
1069 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1070 continue;
1071 base_obj->data = read_sha1_file(d->base.sha1, &type, &base_obj->size);
1072 if (!base_obj->data)
1073 continue;
1075 if (check_sha1_signature(d->base.sha1, base_obj->data,
1076 base_obj->size, typename(type)))
1077 die(_("local object %s is corrupt"), sha1_to_hex(d->base.sha1));
1078 base_obj->obj = append_obj_to_pack(f, d->base.sha1,
1079 base_obj->data, base_obj->size, type);
1080 find_unresolved_deltas(base_obj);
1081 display_progress(progress, nr_resolved_deltas);
1083 free(sorted_by_pos);
1086 static void final(const char *final_pack_name, const char *curr_pack_name,
1087 const char *final_index_name, const char *curr_index_name,
1088 const char *keep_name, const char *keep_msg,
1089 unsigned char *sha1)
1091 const char *report = "pack";
1092 char name[PATH_MAX];
1093 int err;
1095 if (!from_stdin) {
1096 close(input_fd);
1097 } else {
1098 fsync_or_die(output_fd, curr_pack_name);
1099 err = close(output_fd);
1100 if (err)
1101 die_errno(_("error while closing pack file"));
1104 if (keep_msg) {
1105 int keep_fd, keep_msg_len = strlen(keep_msg);
1107 if (!keep_name)
1108 keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1109 else
1110 keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1112 if (keep_fd < 0) {
1113 if (errno != EEXIST)
1114 die_errno(_("cannot write keep file '%s'"),
1115 keep_name);
1116 } else {
1117 if (keep_msg_len > 0) {
1118 write_or_die(keep_fd, keep_msg, keep_msg_len);
1119 write_or_die(keep_fd, "\n", 1);
1121 if (close(keep_fd) != 0)
1122 die_errno(_("cannot close written keep file '%s'"),
1123 keep_name);
1124 report = "keep";
1128 if (final_pack_name != curr_pack_name) {
1129 if (!final_pack_name) {
1130 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1131 get_object_directory(), sha1_to_hex(sha1));
1132 final_pack_name = name;
1134 if (move_temp_to_file(curr_pack_name, final_pack_name))
1135 die(_("cannot store pack file"));
1136 } else if (from_stdin)
1137 chmod(final_pack_name, 0444);
1139 if (final_index_name != curr_index_name) {
1140 if (!final_index_name) {
1141 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1142 get_object_directory(), sha1_to_hex(sha1));
1143 final_index_name = name;
1145 if (move_temp_to_file(curr_index_name, final_index_name))
1146 die(_("cannot store index file"));
1147 } else
1148 chmod(final_index_name, 0444);
1150 if (!from_stdin) {
1151 printf("%s\n", sha1_to_hex(sha1));
1152 } else {
1153 char buf[48];
1154 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1155 report, sha1_to_hex(sha1));
1156 write_or_die(1, buf, len);
1159 * Let's just mimic git-unpack-objects here and write
1160 * the last part of the input buffer to stdout.
1162 while (input_len) {
1163 err = xwrite(1, input_buffer + input_offset, input_len);
1164 if (err <= 0)
1165 break;
1166 input_len -= err;
1167 input_offset += err;
1172 static int git_index_pack_config(const char *k, const char *v, void *cb)
1174 struct pack_idx_option *opts = cb;
1176 if (!strcmp(k, "pack.indexversion")) {
1177 opts->version = git_config_int(k, v);
1178 if (opts->version > 2)
1179 die("bad pack.indexversion=%"PRIu32, opts->version);
1180 return 0;
1182 if (!strcmp(k, "pack.threads")) {
1183 nr_threads = git_config_int(k, v);
1184 if (nr_threads < 0)
1185 die("invalid number of threads specified (%d)",
1186 nr_threads);
1187 #ifdef NO_PTHREADS
1188 if (nr_threads != 1)
1189 warning("no threads support, ignoring %s", k);
1190 nr_threads = 1;
1191 #endif
1192 return 0;
1194 return git_default_config(k, v, cb);
1197 static int cmp_uint32(const void *a_, const void *b_)
1199 uint32_t a = *((uint32_t *)a_);
1200 uint32_t b = *((uint32_t *)b_);
1202 return (a < b) ? -1 : (a != b);
1205 static void read_v2_anomalous_offsets(struct packed_git *p,
1206 struct pack_idx_option *opts)
1208 const uint32_t *idx1, *idx2;
1209 uint32_t i;
1211 /* The address of the 4-byte offset table */
1212 idx1 = (((const uint32_t *)p->index_data)
1213 + 2 /* 8-byte header */
1214 + 256 /* fan out */
1215 + 5 * p->num_objects /* 20-byte SHA-1 table */
1216 + p->num_objects /* CRC32 table */
1219 /* The address of the 8-byte offset table */
1220 idx2 = idx1 + p->num_objects;
1222 for (i = 0; i < p->num_objects; i++) {
1223 uint32_t off = ntohl(idx1[i]);
1224 if (!(off & 0x80000000))
1225 continue;
1226 off = off & 0x7fffffff;
1227 if (idx2[off * 2])
1228 continue;
1230 * The real offset is ntohl(idx2[off * 2]) in high 4
1231 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1232 * octets. But idx2[off * 2] is Zero!!!
1234 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1235 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1238 if (1 < opts->anomaly_nr)
1239 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1242 static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1244 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1246 if (!p)
1247 die(_("Cannot open existing pack file '%s'"), pack_name);
1248 if (open_pack_index(p))
1249 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
1251 /* Read the attributes from the existing idx file */
1252 opts->version = p->index_version;
1254 if (opts->version == 2)
1255 read_v2_anomalous_offsets(p, opts);
1258 * Get rid of the idx file as we do not need it anymore.
1259 * NEEDSWORK: extract this bit from free_pack_by_name() in
1260 * sha1_file.c, perhaps? It shouldn't matter very much as we
1261 * know we haven't installed this pack (hence we never have
1262 * read anything from it).
1264 close_pack_index(p);
1265 free(p);
1268 static void show_pack_info(int stat_only)
1270 int i, baseobjects = nr_objects - nr_deltas;
1271 unsigned long *chain_histogram = NULL;
1273 if (deepest_delta)
1274 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1276 for (i = 0; i < nr_objects; i++) {
1277 struct object_entry *obj = &objects[i];
1279 if (is_delta_type(obj->type))
1280 chain_histogram[obj->delta_depth - 1]++;
1281 if (stat_only)
1282 continue;
1283 printf("%s %-6s %lu %lu %"PRIuMAX,
1284 sha1_to_hex(obj->idx.sha1),
1285 typename(obj->real_type), obj->size,
1286 (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1287 (uintmax_t)obj->idx.offset);
1288 if (is_delta_type(obj->type)) {
1289 struct object_entry *bobj = &objects[obj->base_object_no];
1290 printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
1292 putchar('\n');
1295 if (baseobjects)
1296 printf_ln(Q_("non delta: %d object",
1297 "non delta: %d objects",
1298 baseobjects),
1299 baseobjects);
1300 for (i = 0; i < deepest_delta; i++) {
1301 if (!chain_histogram[i])
1302 continue;
1303 printf_ln(Q_("chain length = %d: %lu object",
1304 "chain length = %d: %lu objects",
1305 chain_histogram[i]),
1306 i + 1,
1307 chain_histogram[i]);
1311 int cmd_index_pack(int argc, const char **argv, const char *prefix)
1313 int i, fix_thin_pack = 0, verify = 0, stat_only = 0, stat = 0;
1314 const char *curr_pack, *curr_index;
1315 const char *index_name = NULL, *pack_name = NULL;
1316 const char *keep_name = NULL, *keep_msg = NULL;
1317 char *index_name_buf = NULL, *keep_name_buf = NULL;
1318 struct pack_idx_entry **idx_objects;
1319 struct pack_idx_option opts;
1320 unsigned char pack_sha1[20];
1322 if (argc == 2 && !strcmp(argv[1], "-h"))
1323 usage(index_pack_usage);
1325 read_replace_refs = 0;
1327 reset_pack_idx_option(&opts);
1328 git_config(git_index_pack_config, &opts);
1329 if (prefix && chdir(prefix))
1330 die(_("Cannot come back to cwd"));
1332 for (i = 1; i < argc; i++) {
1333 const char *arg = argv[i];
1335 if (*arg == '-') {
1336 if (!strcmp(arg, "--stdin")) {
1337 from_stdin = 1;
1338 } else if (!strcmp(arg, "--fix-thin")) {
1339 fix_thin_pack = 1;
1340 } else if (!strcmp(arg, "--strict")) {
1341 strict = 1;
1342 } else if (!strcmp(arg, "--verify")) {
1343 verify = 1;
1344 } else if (!strcmp(arg, "--verify-stat")) {
1345 verify = 1;
1346 stat = 1;
1347 } else if (!strcmp(arg, "--verify-stat-only")) {
1348 verify = 1;
1349 stat = 1;
1350 stat_only = 1;
1351 } else if (!strcmp(arg, "--keep")) {
1352 keep_msg = "";
1353 } else if (!prefixcmp(arg, "--keep=")) {
1354 keep_msg = arg + 7;
1355 } else if (!prefixcmp(arg, "--threads=")) {
1356 char *end;
1357 nr_threads = strtoul(arg+10, &end, 0);
1358 if (!arg[10] || *end || nr_threads < 0)
1359 usage(index_pack_usage);
1360 #ifdef NO_PTHREADS
1361 if (nr_threads != 1)
1362 warning("no threads support, "
1363 "ignoring %s", arg);
1364 nr_threads = 1;
1365 #endif
1366 } else if (!prefixcmp(arg, "--pack_header=")) {
1367 struct pack_header *hdr;
1368 char *c;
1370 hdr = (struct pack_header *)input_buffer;
1371 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1372 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1373 if (*c != ',')
1374 die(_("bad %s"), arg);
1375 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1376 if (*c)
1377 die(_("bad %s"), arg);
1378 input_len = sizeof(*hdr);
1379 } else if (!strcmp(arg, "-v")) {
1380 verbose = 1;
1381 } else if (!strcmp(arg, "-o")) {
1382 if (index_name || (i+1) >= argc)
1383 usage(index_pack_usage);
1384 index_name = argv[++i];
1385 } else if (!prefixcmp(arg, "--index-version=")) {
1386 char *c;
1387 opts.version = strtoul(arg + 16, &c, 10);
1388 if (opts.version > 2)
1389 die(_("bad %s"), arg);
1390 if (*c == ',')
1391 opts.off32_limit = strtoul(c+1, &c, 0);
1392 if (*c || opts.off32_limit & 0x80000000)
1393 die(_("bad %s"), arg);
1394 } else
1395 usage(index_pack_usage);
1396 continue;
1399 if (pack_name)
1400 usage(index_pack_usage);
1401 pack_name = arg;
1404 if (!pack_name && !from_stdin)
1405 usage(index_pack_usage);
1406 if (fix_thin_pack && !from_stdin)
1407 die(_("--fix-thin cannot be used without --stdin"));
1408 if (!index_name && pack_name) {
1409 int len = strlen(pack_name);
1410 if (!has_extension(pack_name, ".pack"))
1411 die(_("packfile name '%s' does not end with '.pack'"),
1412 pack_name);
1413 index_name_buf = xmalloc(len);
1414 memcpy(index_name_buf, pack_name, len - 5);
1415 strcpy(index_name_buf + len - 5, ".idx");
1416 index_name = index_name_buf;
1418 if (keep_msg && !keep_name && pack_name) {
1419 int len = strlen(pack_name);
1420 if (!has_extension(pack_name, ".pack"))
1421 die(_("packfile name '%s' does not end with '.pack'"),
1422 pack_name);
1423 keep_name_buf = xmalloc(len);
1424 memcpy(keep_name_buf, pack_name, len - 5);
1425 strcpy(keep_name_buf + len - 5, ".keep");
1426 keep_name = keep_name_buf;
1428 if (verify) {
1429 if (!index_name)
1430 die(_("--verify with no packfile name given"));
1431 read_idx_option(&opts, index_name);
1432 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
1434 if (strict)
1435 opts.flags |= WRITE_IDX_STRICT;
1437 #ifndef NO_PTHREADS
1438 if (!nr_threads) {
1439 nr_threads = online_cpus();
1440 /* An experiment showed that more threads does not mean faster */
1441 if (nr_threads > 3)
1442 nr_threads = 3;
1444 #endif
1446 curr_pack = open_pack_file(pack_name);
1447 parse_pack_header();
1448 objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1449 deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
1450 parse_pack_objects(pack_sha1);
1451 resolve_deltas();
1452 conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
1453 free(deltas);
1454 if (strict)
1455 check_objects();
1457 if (stat)
1458 show_pack_info(stat_only);
1460 idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1461 for (i = 0; i < nr_objects; i++)
1462 idx_objects[i] = &objects[i].idx;
1463 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
1464 free(idx_objects);
1466 if (!verify)
1467 final(pack_name, curr_pack,
1468 index_name, curr_index,
1469 keep_name, keep_msg,
1470 pack_sha1);
1471 else
1472 close(input_fd);
1473 free(objects);
1474 free(index_name_buf);
1475 free(keep_name_buf);
1476 if (pack_name == NULL)
1477 free((void *) curr_pack);
1478 if (index_name == NULL)
1479 free((void *) curr_index);
1481 return 0;