]> git.ipfire.org Git - thirdparty/git.git/blame - builtin/index-pack.c
fsck_object(): allow passing object data separately from the object itself
[thirdparty/git.git] / builtin / index-pack.c
CommitLineData
2ad9d4cb 1#include "builtin.h"
9cf6d335
SV
2#include "delta.h"
3#include "pack.h"
4#include "csum-file.h"
8e440259
PE
5#include "blob.h"
6#include "commit.h"
7#include "tag.h"
8#include "tree.h"
96a02f8f 9#include "progress.h"
0153be05 10#include "fsck.h"
2fb3f6db 11#include "exec_cmd.h"
4614043c 12#include "streaming.h"
b8a2486f 13#include "thread-utils.h"
9cf6d335
SV
14
15static const char index_pack_usage[] =
e337a04d 16"git index-pack [-v] [-o <index-file>] [--keep | --keep=<msg>] [--verify] [--strict] (<pack-file> | --stdin [--fix-thin] [<pack-file>])";
9cf6d335 17
9cba13ca 18struct object_entry {
aa7e44bf 19 struct pack_idx_entry idx;
2d477051
NP
20 unsigned long size;
21 unsigned int hdr_size;
9cf6d335
SV
22 enum object_type type;
23 enum object_type real_type;
38a45561
JH
24 unsigned delta_depth;
25 int base_object_no;
9cf6d335
SV
26};
27
53dda6ff
NP
28union delta_base {
29 unsigned char sha1[20];
d7dd0223 30 off_t offset;
53dda6ff
NP
31};
32
f41aebd4 33struct base_data {
4a438cab
SP
34 struct base_data *base;
35 struct base_data *child;
03993e13 36 struct object_entry *obj;
f41aebd4
SP
37 void *data;
38 unsigned long size;
2baad220
NTND
39 int ref_first, ref_last;
40 int ofs_first, ofs_last;
f41aebd4
SP
41};
42
b8a2486f
NTND
43struct thread_local {
44#ifndef NO_PTHREADS
45 pthread_t thread;
46#endif
47 struct base_data *base_cache;
48 size_t base_cache_used;
39539495 49 int pack_fd;
b8a2486f
NTND
50};
51
3c552873
NP
52/*
53 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
54 * to memcmp() only the first 20 bytes.
55 */
56#define UNION_BASE_SZ 20
57
0153be05
MK
58#define FLAG_LINK (1u<<20)
59#define FLAG_CHECKED (1u<<21)
60
9cba13ca 61struct delta_entry {
53dda6ff 62 union delta_base base;
636171cb 63 int obj_no;
9cf6d335
SV
64};
65
9cf6d335
SV
66static struct object_entry *objects;
67static struct delta_entry *deltas;
b8a2486f 68static struct thread_local nothread_data;
9cf6d335
SV
69static int nr_objects;
70static int nr_deltas;
636171cb 71static int nr_resolved_deltas;
b8a2486f 72static int nr_threads;
9cf6d335 73
e42797f5 74static int from_stdin;
0153be05 75static int strict;
c6807a40 76static int do_fsck_object;
3c9af366 77static int verbose;
3aba2fdd 78static int show_stat;
c6807a40 79static int check_self_contained_and_connected;
3c9af366 80
dc6a0757 81static struct progress *progress;
e42797f5 82
2d477051
NP
83/* We always read in 4kB chunks. */
84static unsigned char input_buffer[4096];
d7dd0223
NP
85static unsigned int input_offset, input_len;
86static off_t consumed_bytes;
d1a0ed18 87static unsigned deepest_delta;
9126f009 88static git_SHA_CTX input_ctx;
ee5743ce 89static uint32_t input_crc32;
39539495
NTND
90static int input_fd, output_fd;
91static const char *curr_pack;
2d477051 92
b8a2486f
NTND
93#ifndef NO_PTHREADS
94
95static struct thread_local *thread_data;
96static int nr_dispatched;
97static int threads_active;
98
99static pthread_mutex_t read_mutex;
100#define read_lock() lock_mutex(&read_mutex)
101#define read_unlock() unlock_mutex(&read_mutex)
102
103static pthread_mutex_t counter_mutex;
104#define counter_lock() lock_mutex(&counter_mutex)
105#define counter_unlock() unlock_mutex(&counter_mutex)
106
107static pthread_mutex_t work_mutex;
108#define work_lock() lock_mutex(&work_mutex)
109#define work_unlock() unlock_mutex(&work_mutex)
110
3aba2fdd
NTND
111static 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)
114
b8a2486f
NTND
115static pthread_key_t key;
116
117static inline void lock_mutex(pthread_mutex_t *mutex)
118{
119 if (threads_active)
120 pthread_mutex_lock(mutex);
121}
122
123static inline void unlock_mutex(pthread_mutex_t *mutex)
124{
125 if (threads_active)
126 pthread_mutex_unlock(mutex);
127}
128
129/*
130 * Mutex and conditional variable can't be statically-initialized on Windows.
131 */
132static void init_thread(void)
133{
39539495 134 int i;
b8a2486f
NTND
135 init_recursive_mutex(&read_mutex);
136 pthread_mutex_init(&counter_mutex, NULL);
137 pthread_mutex_init(&work_mutex, NULL);
3aba2fdd
NTND
138 if (show_stat)
139 pthread_mutex_init(&deepest_delta_mutex, NULL);
b8a2486f
NTND
140 pthread_key_create(&key, NULL);
141 thread_data = xcalloc(nr_threads, sizeof(*thread_data));
39539495
NTND
142 for (i = 0; i < nr_threads; i++) {
143 thread_data[i].pack_fd = open(curr_pack, O_RDONLY);
144 if (thread_data[i].pack_fd == -1)
145 die_errno(_("unable to open %s"), curr_pack);
146 }
147
b8a2486f
NTND
148 threads_active = 1;
149}
150
151static void cleanup_thread(void)
152{
39539495 153 int i;
b8a2486f
NTND
154 if (!threads_active)
155 return;
156 threads_active = 0;
157 pthread_mutex_destroy(&read_mutex);
158 pthread_mutex_destroy(&counter_mutex);
159 pthread_mutex_destroy(&work_mutex);
3aba2fdd
NTND
160 if (show_stat)
161 pthread_mutex_destroy(&deepest_delta_mutex);
39539495
NTND
162 for (i = 0; i < nr_threads; i++)
163 close(thread_data[i].pack_fd);
b8a2486f
NTND
164 pthread_key_delete(key);
165 free(thread_data);
166}
167
168#else
169
170#define read_lock()
171#define read_unlock()
172
173#define counter_lock()
174#define counter_unlock()
175
176#define work_lock()
177#define work_unlock()
178
3aba2fdd
NTND
179#define deepest_delta_lock()
180#define deepest_delta_unlock()
181
b8a2486f
NTND
182#endif
183
184
0153be05
MK
185static int mark_link(struct object *obj, int type, void *data)
186{
187 if (!obj)
188 return -1;
189
190 if (type != OBJ_ANY && obj->type != type)
c2b97ecf 191 die(_("object type mismatch at %s"), sha1_to_hex(obj->sha1));
0153be05
MK
192
193 obj->flags |= FLAG_LINK;
194 return 0;
195}
196
197/* The content of each linked object must have been checked
198 or it must be already present in the object database */
c6807a40 199static unsigned check_object(struct object *obj)
0153be05
MK
200{
201 if (!obj)
c6807a40 202 return 0;
0153be05
MK
203
204 if (!(obj->flags & FLAG_LINK))
c6807a40 205 return 0;
0153be05
MK
206
207 if (!(obj->flags & FLAG_CHECKED)) {
208 unsigned long size;
209 int type = sha1_object_info(obj->sha1, &size);
77583e77
JK
210 if (type <= 0)
211 die(_("did not receive expected object %s"),
212 sha1_to_hex(obj->sha1));
213 if (type != obj->type)
214 die(_("object %s: expected type %s, found %s"),
215 sha1_to_hex(obj->sha1),
216 typename(obj->type), typename(type));
0153be05 217 obj->flags |= FLAG_CHECKED;
c6807a40 218 return 1;
0153be05 219 }
c6807a40
NTND
220
221 return 0;
0153be05
MK
222}
223
c6807a40 224static unsigned check_objects(void)
0153be05 225{
c6807a40 226 unsigned i, max, foreign_nr = 0;
0153be05
MK
227
228 max = get_max_object_index();
229 for (i = 0; i < max; i++)
c6807a40
NTND
230 foreign_nr += check_object(get_indexed_object(i));
231 return foreign_nr;
0153be05
MK
232}
233
234
636171cb 235/* Discard current buffer used content. */
a6e8a767 236static void flush(void)
636171cb
NP
237{
238 if (input_offset) {
239 if (output_fd >= 0)
240 write_or_die(output_fd, input_buffer, input_offset);
9126f009 241 git_SHA1_Update(&input_ctx, input_buffer, input_offset);
554a2636 242 memmove(input_buffer, input_buffer + input_offset, input_len);
636171cb
NP
243 input_offset = 0;
244 }
245}
246
2d477051
NP
247/*
248 * Make sure at least "min" bytes are available in the buffer, and
249 * return the pointer to the buffer.
250 */
b89c4e93 251static void *fill(int min)
9cf6d335 252{
2d477051
NP
253 if (min <= input_len)
254 return input_buffer + input_offset;
255 if (min > sizeof(input_buffer))
c2b97ecf
NTND
256 die(Q_("cannot fill %d byte",
257 "cannot fill %d bytes",
258 min),
259 min);
636171cb 260 flush();
2d477051 261 do {
8a912bcb 262 ssize_t ret = xread(input_fd, input_buffer + input_len,
2d477051
NP
263 sizeof(input_buffer) - input_len);
264 if (ret <= 0) {
265 if (!ret)
c2b97ecf
NTND
266 die(_("early EOF"));
267 die_errno(_("read error on input"));
2d477051
NP
268 }
269 input_len += ret;
218558af
NP
270 if (from_stdin)
271 display_throughput(progress, consumed_bytes + input_len);
2d477051
NP
272 } while (input_len < min);
273 return input_buffer;
274}
275
276static void use(int bytes)
277{
278 if (bytes > input_len)
c2b97ecf 279 die(_("used more bytes than were available"));
ee5743ce 280 input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
2d477051
NP
281 input_len -= bytes;
282 input_offset += bytes;
d7dd0223
NP
283
284 /* make sure off_t is sufficiently large not to wrap */
c03c8315 285 if (signed_add_overflows(consumed_bytes, bytes))
c2b97ecf 286 die(_("pack too large for current definition of off_t"));
2d477051
NP
287 consumed_bytes += bytes;
288}
9cf6d335 289
3bb72562 290static const char *open_pack_file(const char *pack_name)
2d477051 291{
e42797f5
NP
292 if (from_stdin) {
293 input_fd = 0;
294 if (!pack_name) {
ab1900a3
ÆAB
295 static char tmp_file[PATH_MAX];
296 output_fd = odb_mkstemp(tmp_file, sizeof(tmp_file),
6e180cdc 297 "pack/tmp_pack_XXXXXX");
ab1900a3 298 pack_name = xstrdup(tmp_file);
e42797f5
NP
299 } else
300 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
301 if (output_fd < 0)
c2b97ecf 302 die_errno(_("unable to create '%s'"), pack_name);
39539495 303 nothread_data.pack_fd = output_fd;
e42797f5
NP
304 } else {
305 input_fd = open(pack_name, O_RDONLY);
306 if (input_fd < 0)
c2b97ecf 307 die_errno(_("cannot open packfile '%s'"), pack_name);
e42797f5 308 output_fd = -1;
39539495 309 nothread_data.pack_fd = input_fd;
e42797f5 310 }
9126f009 311 git_SHA1_Init(&input_ctx);
e42797f5 312 return pack_name;
9cf6d335
SV
313}
314
315static void parse_pack_header(void)
316{
2d477051 317 struct pack_header *hdr = fill(sizeof(struct pack_header));
9cf6d335
SV
318
319 /* Header consistency check */
9cf6d335 320 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
c2b97ecf 321 die(_("pack signature mismatch"));
d60fc1c8 322 if (!pack_version_ok(hdr->hdr_version))
f350df42 323 die(_("pack version %"PRIu32" unsupported"),
6e1c2344 324 ntohl(hdr->hdr_version));
9cf6d335
SV
325
326 nr_objects = ntohl(hdr->hdr_entries);
2d477051 327 use(sizeof(struct pack_header));
9cf6d335
SV
328}
329
a4f3131c
EFL
330static NORETURN void bad_object(unsigned long offset, const char *format,
331 ...) __attribute__((format (printf, 2, 3)));
9cf6d335 332
c2e86add 333static NORETURN void bad_object(unsigned long offset, const char *format, ...)
9cf6d335
SV
334{
335 va_list params;
336 char buf[1024];
337
338 va_start(params, format);
339 vsnprintf(buf, sizeof(buf), format, params);
340 va_end(params);
c2b97ecf 341 die(_("pack has bad object at offset %lu: %s"), offset, buf);
9cf6d335
SV
342}
343
b8a2486f
NTND
344static inline struct thread_local *get_thread_data(void)
345{
346#ifndef NO_PTHREADS
347 if (threads_active)
348 return pthread_getspecific(key);
349 assert(!threads_active &&
350 "This should only be reached when all threads are gone");
351#endif
352 return &nothread_data;
353}
354
355#ifndef NO_PTHREADS
356static void set_thread_data(struct thread_local *data)
357{
358 if (threads_active)
359 pthread_setspecific(key, data);
360}
361#endif
362
2baad220
NTND
363static struct base_data *alloc_base_data(void)
364{
51a60f5b 365 struct base_data *base = xcalloc(1, sizeof(struct base_data));
2baad220
NTND
366 base->ref_last = -1;
367 base->ofs_last = -1;
368 return base;
369}
370
6a87ed97
NP
371static void free_base_data(struct base_data *c)
372{
373 if (c->data) {
374 free(c->data);
375 c->data = NULL;
b8a2486f 376 get_thread_data()->base_cache_used -= c->size;
6a87ed97
NP
377 }
378}
379
92392b4a
SP
380static void prune_base_data(struct base_data *retain)
381{
8e24cbae 382 struct base_data *b;
b8a2486f
NTND
383 struct thread_local *data = get_thread_data();
384 for (b = data->base_cache;
385 data->base_cache_used > delta_base_cache_limit && b;
92392b4a 386 b = b->child) {
6a87ed97
NP
387 if (b->data && b != retain)
388 free_base_data(b);
92392b4a
SP
389 }
390}
391
4a438cab
SP
392static void link_base_data(struct base_data *base, struct base_data *c)
393{
394 if (base)
395 base->child = c;
396 else
b8a2486f 397 get_thread_data()->base_cache = c;
4a438cab
SP
398
399 c->base = base;
400 c->child = NULL;
9441b61d 401 if (c->data)
b8a2486f 402 get_thread_data()->base_cache_used += c->size;
92392b4a 403 prune_base_data(c);
4a438cab
SP
404}
405
406static void unlink_base_data(struct base_data *c)
407{
408 struct base_data *base = c->base;
409 if (base)
410 base->child = NULL;
411 else
b8a2486f 412 get_thread_data()->base_cache = NULL;
6a87ed97 413 free_base_data(c);
4a438cab
SP
414}
415
681b07de
NTND
416static int is_delta_type(enum object_type type)
417{
418 return (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA);
419}
420
421static void *unpack_entry_data(unsigned long offset, unsigned long size,
422 enum object_type type, unsigned char *sha1)
9cf6d335 423{
9ec2dde9 424 static char fixed_buf[8192];
7ce4721a 425 int status;
ef49a7a0 426 git_zstream stream;
9ec2dde9 427 void *buf;
681b07de
NTND
428 git_SHA_CTX c;
429 char hdr[32];
430 int hdrlen;
431
432 if (!is_delta_type(type)) {
433 hdrlen = sprintf(hdr, "%s %lu", typename(type), size) + 1;
434 git_SHA1_Init(&c);
435 git_SHA1_Update(&c, hdr, hdrlen);
436 } else
437 sha1 = NULL;
9ec2dde9
NTND
438 if (type == OBJ_BLOB && size > big_file_threshold)
439 buf = fixed_buf;
440 else
441 buf = xmalloc(size);
9cf6d335
SV
442
443 memset(&stream, 0, sizeof(stream));
7ce4721a 444 git_inflate_init(&stream);
9cf6d335 445 stream.next_out = buf;
9ec2dde9 446 stream.avail_out = buf == fixed_buf ? sizeof(fixed_buf) : size;
9cf6d335 447
7ce4721a 448 do {
681b07de 449 unsigned char *last_out = stream.next_out;
2d477051
NP
450 stream.next_in = fill(1);
451 stream.avail_in = input_len;
7ce4721a
NP
452 status = git_inflate(&stream, 0);
453 use(input_len - stream.avail_in);
681b07de
NTND
454 if (sha1)
455 git_SHA1_Update(&c, last_out, stream.next_out - last_out);
9ec2dde9
NTND
456 if (buf == fixed_buf) {
457 stream.next_out = buf;
458 stream.avail_out = sizeof(fixed_buf);
459 }
7ce4721a
NP
460 } while (status == Z_OK);
461 if (stream.total_out != size || status != Z_STREAM_END)
c2b97ecf 462 bad_object(offset, _("inflate returned %d"), status);
39c68542 463 git_inflate_end(&stream);
681b07de
NTND
464 if (sha1)
465 git_SHA1_Final(sha1, &c);
9ec2dde9 466 return buf == fixed_buf ? NULL : buf;
9cf6d335
SV
467}
468
681b07de
NTND
469static void *unpack_raw_entry(struct object_entry *obj,
470 union delta_base *delta_base,
471 unsigned char *sha1)
9cf6d335 472{
48fb7deb
LT
473 unsigned char *p;
474 unsigned long size, c;
d7dd0223 475 off_t base_offset;
9cf6d335 476 unsigned shift;
ee5743ce 477 void *data;
9cf6d335 478
aa7e44bf 479 obj->idx.offset = consumed_bytes;
1e4cd68c 480 input_crc32 = crc32(0, NULL, 0);
2d477051
NP
481
482 p = fill(1);
483 c = *p;
484 use(1);
485 obj->type = (c >> 4) & 7;
9cf6d335
SV
486 size = (c & 15);
487 shift = 4;
488 while (c & 0x80) {
2d477051
NP
489 p = fill(1);
490 c = *p;
491 use(1);
48fb7deb 492 size += (c & 0x7f) << shift;
9cf6d335
SV
493 shift += 7;
494 }
2d477051 495 obj->size = size;
9cf6d335 496
2d477051 497 switch (obj->type) {
eb32d236 498 case OBJ_REF_DELTA:
2d477051
NP
499 hashcpy(delta_base->sha1, fill(20));
500 use(20);
53dda6ff
NP
501 break;
502 case OBJ_OFS_DELTA:
503 memset(delta_base, 0, sizeof(*delta_base));
2d477051
NP
504 p = fill(1);
505 c = *p;
506 use(1);
53dda6ff
NP
507 base_offset = c & 127;
508 while (c & 128) {
509 base_offset += 1;
8723f216 510 if (!base_offset || MSB(base_offset, 7))
c2b97ecf 511 bad_object(obj->idx.offset, _("offset value overflow for delta base object"));
2d477051
NP
512 p = fill(1);
513 c = *p;
514 use(1);
53dda6ff
NP
515 base_offset = (base_offset << 7) + (c & 127);
516 }
aa7e44bf 517 delta_base->offset = obj->idx.offset - base_offset;
d8f32556 518 if (delta_base->offset <= 0 || delta_base->offset >= obj->idx.offset)
c2b97ecf 519 bad_object(obj->idx.offset, _("delta base offset is out of bound"));
53dda6ff 520 break;
9cf6d335
SV
521 case OBJ_COMMIT:
522 case OBJ_TREE:
523 case OBJ_BLOB:
524 case OBJ_TAG:
9cf6d335
SV
525 break;
526 default:
c2b97ecf 527 bad_object(obj->idx.offset, _("unknown object type %d"), obj->type);
9cf6d335 528 }
aa7e44bf 529 obj->hdr_size = consumed_bytes - obj->idx.offset;
2d477051 530
681b07de 531 data = unpack_entry_data(obj->idx.offset, obj->size, obj->type, sha1);
aa7e44bf 532 obj->idx.crc32 = input_crc32;
ee5743ce 533 return data;
2d477051
NP
534}
535
8a2e163c
NTND
536static void *unpack_data(struct object_entry *obj,
537 int (*consume)(const unsigned char *, unsigned long, void *),
538 void *cb_data)
2d477051 539{
a91ef6e7 540 off_t from = obj[0].idx.offset + obj[0].hdr_size;
aa7e44bf 541 unsigned long len = obj[1].idx.offset - from;
776ea370 542 unsigned char *data, *inbuf;
ef49a7a0 543 git_zstream stream;
776ea370 544 int status;
9cf6d335 545
8a2e163c 546 data = xmalloc(consume ? 64*1024 : obj->size);
776ea370
NP
547 inbuf = xmalloc((len < 64*1024) ? len : 64*1024);
548
2d477051 549 memset(&stream, 0, sizeof(stream));
776ea370 550 git_inflate_init(&stream);
2d477051 551 stream.next_out = data;
8a2e163c 552 stream.avail_out = consume ? 64*1024 : obj->size;
776ea370
NP
553
554 do {
555 ssize_t n = (len < 64*1024) ? len : 64*1024;
53f52cd9 556 n = xpread(get_thread_data()->pack_fd, inbuf, n, from);
776ea370 557 if (n < 0)
c2b97ecf 558 die_errno(_("cannot pread pack file"));
776ea370 559 if (!n)
c2b97ecf
NTND
560 die(Q_("premature end of pack file, %lu byte missing",
561 "premature end of pack file, %lu bytes missing",
562 len),
563 len);
776ea370
NP
564 from += n;
565 len -= n;
566 stream.next_in = inbuf;
567 stream.avail_in = n;
f8b09038
JK
568 if (!consume)
569 status = git_inflate(&stream, 0);
570 else {
571 do {
572 status = git_inflate(&stream, 0);
573 if (consume(data, stream.next_out - data, cb_data)) {
574 free(inbuf);
575 free(data);
576 return NULL;
577 }
578 stream.next_out = data;
579 stream.avail_out = 64*1024;
580 } while (status == Z_OK && stream.avail_in);
8a2e163c 581 }
776ea370
NP
582 } while (len && status == Z_OK && !stream.avail_in);
583
584 /* This has been inflated OK when first encountered, so... */
585 if (status != Z_STREAM_END || stream.total_out != obj->size)
c2b97ecf 586 die(_("serious inflate inconsistency"));
776ea370
NP
587
588 git_inflate_end(&stream);
589 free(inbuf);
8a2e163c
NTND
590 if (consume) {
591 free(data);
592 data = NULL;
593 }
9cf6d335
SV
594 return data;
595}
596
8a2e163c
NTND
597static void *get_data_from_pack(struct object_entry *obj)
598{
599 return unpack_data(obj, NULL, NULL);
600}
601
7218a215
JH
602static int compare_delta_bases(const union delta_base *base1,
603 const union delta_base *base2,
604 enum object_type type1,
605 enum object_type type2)
606{
607 int cmp = type1 - type2;
608 if (cmp)
609 return cmp;
610 return memcmp(base1, base2, UNION_BASE_SZ);
611}
612
613static int find_delta(const union delta_base *base, enum object_type type)
9cf6d335
SV
614{
615 int first = 0, last = nr_deltas;
616
617 while (first < last) {
618 int next = (first + last) / 2;
619 struct delta_entry *delta = &deltas[next];
620 int cmp;
621
7218a215
JH
622 cmp = compare_delta_bases(base, &delta->base,
623 type, objects[delta->obj_no].type);
9cf6d335
SV
624 if (!cmp)
625 return next;
626 if (cmp < 0) {
627 last = next;
628 continue;
629 }
630 first = next+1;
631 }
632 return -first-1;
633}
634
6a87ed97 635static void find_delta_children(const union delta_base *base,
7218a215
JH
636 int *first_index, int *last_index,
637 enum object_type type)
9cf6d335 638{
7218a215 639 int first = find_delta(base, type);
9cf6d335
SV
640 int last = first;
641 int end = nr_deltas - 1;
642
6a87ed97
NP
643 if (first < 0) {
644 *first_index = 0;
645 *last_index = -1;
646 return;
647 }
3c552873 648 while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
9cf6d335 649 --first;
3c552873 650 while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
9cf6d335
SV
651 ++last;
652 *first_index = first;
653 *last_index = last;
9cf6d335
SV
654}
655
4614043c
NTND
656struct compare_data {
657 struct object_entry *entry;
658 struct git_istream *st;
659 unsigned char *buf;
660 unsigned long buf_size;
661};
662
663static int compare_objects(const unsigned char *buf, unsigned long size,
664 void *cb_data)
665{
666 struct compare_data *data = cb_data;
667
668 if (data->buf_size < size) {
669 free(data->buf);
670 data->buf = xmalloc(size);
671 data->buf_size = size;
672 }
673
674 while (size) {
675 ssize_t len = read_istream(data->st, data->buf, size);
676 if (len == 0)
677 die(_("SHA1 COLLISION FOUND WITH %s !"),
678 sha1_to_hex(data->entry->idx.sha1));
679 if (len < 0)
680 die(_("unable to read %s"),
681 sha1_to_hex(data->entry->idx.sha1));
682 if (memcmp(buf, data->buf, len))
683 die(_("SHA1 COLLISION FOUND WITH %s !"),
684 sha1_to_hex(data->entry->idx.sha1));
685 size -= len;
686 buf += len;
687 }
688 return 0;
689}
690
691static int check_collison(struct object_entry *entry)
692{
693 struct compare_data data;
694 enum object_type type;
695 unsigned long size;
696
697 if (entry->size <= big_file_threshold || entry->type != OBJ_BLOB)
698 return -1;
699
700 memset(&data, 0, sizeof(data));
701 data.entry = entry;
702 data.st = open_istream(entry->idx.sha1, &type, &size, NULL);
703 if (!data.st)
704 return -1;
705 if (size != entry->size || type != entry->type)
706 die(_("SHA1 COLLISION FOUND WITH %s !"),
707 sha1_to_hex(entry->idx.sha1));
708 unpack_data(entry, compare_objects, &data);
709 close_istream(data.st);
710 free(data.buf);
711 return 0;
712}
713
9ec2dde9
NTND
714static void sha1_object(const void *data, struct object_entry *obj_entry,
715 unsigned long size, enum object_type type,
716 const unsigned char *sha1)
9cf6d335 717{
9ec2dde9 718 void *new_data = NULL;
4614043c 719 int collision_test_needed;
9ec2dde9
NTND
720
721 assert(data || obj_entry);
722
b8a2486f 723 read_lock();
4614043c
NTND
724 collision_test_needed = has_sha1_file(sha1);
725 read_unlock();
726
727 if (collision_test_needed && !data) {
728 read_lock();
729 if (!check_collison(obj_entry))
730 collision_test_needed = 0;
731 read_unlock();
732 }
733 if (collision_test_needed) {
8685da42
NP
734 void *has_data;
735 enum object_type has_type;
736 unsigned long has_size;
4614043c
NTND
737 read_lock();
738 has_type = sha1_object_info(sha1, &has_size);
739 if (has_type != type || has_size != size)
740 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
8685da42 741 has_data = read_sha1_file(sha1, &has_type, &has_size);
b8a2486f 742 read_unlock();
4614043c
NTND
743 if (!data)
744 data = new_data = get_data_from_pack(obj_entry);
8685da42 745 if (!has_data)
c2b97ecf 746 die(_("cannot read existing object %s"), sha1_to_hex(sha1));
8685da42
NP
747 if (size != has_size || type != has_type ||
748 memcmp(data, has_data, size) != 0)
c2b97ecf 749 die(_("SHA1 COLLISION FOUND WITH %s !"), sha1_to_hex(sha1));
bbf4b41b 750 free(has_data);
4614043c 751 }
b8a2486f 752
0153be05 753 if (strict) {
b8a2486f 754 read_lock();
0153be05
MK
755 if (type == OBJ_BLOB) {
756 struct blob *blob = lookup_blob(sha1);
757 if (blob)
758 blob->object.flags |= FLAG_CHECKED;
759 else
c2b97ecf 760 die(_("invalid blob object %s"), sha1_to_hex(sha1));
0153be05
MK
761 } else {
762 struct object *obj;
763 int eaten;
764 void *buf = (void *) data;
765
920734b0 766 assert(data && "data can only be NULL for large _blobs_");
9ec2dde9 767
0153be05
MK
768 /*
769 * we do not need to free the memory here, as the
770 * buf is deleted by the caller.
771 */
772 obj = parse_object_buffer(sha1, type, size, buf, &eaten);
773 if (!obj)
c2b97ecf 774 die(_("invalid %s"), typename(type));
c6807a40 775 if (do_fsck_object &&
90a398bb
JS
776 fsck_object(obj, buf, size, 1,
777 fsck_error_function))
c2b97ecf 778 die(_("Error in object"));
2af202be 779 if (fsck_walk(obj, mark_link, NULL))
c2b97ecf 780 die(_("Not all child objects of %s are reachable"), sha1_to_hex(obj->sha1));
0153be05
MK
781
782 if (obj->type == OBJ_TREE) {
783 struct tree *item = (struct tree *) obj;
784 item->buffer = NULL;
6e454b9a 785 obj->parsed = 0;
0153be05
MK
786 }
787 if (obj->type == OBJ_COMMIT) {
788 struct commit *commit = (struct commit *) obj;
8597ea3a 789 if (detach_commit_buffer(commit, NULL) != data)
0fb370da 790 die("BUG: parse_object_buffer transmogrified our buffer");
0153be05
MK
791 }
792 obj->flags |= FLAG_CHECKED;
793 }
b8a2486f 794 read_unlock();
0153be05 795 }
9ec2dde9
NTND
796
797 free(new_data);
9cf6d335
SV
798}
799
20e95d0a
NTND
800/*
801 * This function is part of find_unresolved_deltas(). There are two
802 * walkers going in the opposite ways.
803 *
804 * The first one in find_unresolved_deltas() traverses down from
805 * parent node to children, deflating nodes along the way. However,
806 * memory for deflated nodes is limited by delta_base_cache_limit, so
807 * at some point parent node's deflated content may be freed.
808 *
809 * The second walker is this function, which goes from current node up
810 * to top parent if necessary to deflate the node. In normal
811 * situation, its parent node would be already deflated, so it just
812 * needs to apply delta.
813 *
814 * In the worst case scenario, parent node is no longer deflated because
815 * we're running out of delta_base_cache_limit; we need to re-deflate
816 * parents, possibly up to the top base.
817 *
818 * All deflated objects here are subject to be freed if we exceed
819 * delta_base_cache_limit, just like in find_unresolved_deltas(), we
820 * just need to make sure the last node is not freed.
821 */
92392b4a
SP
822static void *get_base_data(struct base_data *c)
823{
824 if (!c->data) {
825 struct object_entry *obj = c->obj;
20e95d0a
NTND
826 struct base_data **delta = NULL;
827 int delta_nr = 0, delta_alloc = 0;
92392b4a 828
20e95d0a
NTND
829 while (is_delta_type(c->obj->type) && !c->data) {
830 ALLOC_GROW(delta, delta_nr + 1, delta_alloc);
831 delta[delta_nr++] = c;
832 c = c->base;
833 }
834 if (!delta_nr) {
835 c->data = get_data_from_pack(obj);
836 c->size = obj->size;
b8a2486f 837 get_thread_data()->base_cache_used += c->size;
20e95d0a
NTND
838 prune_base_data(c);
839 }
840 for (; delta_nr > 0; delta_nr--) {
841 void *base, *raw;
842 c = delta[delta_nr - 1];
843 obj = c->obj;
844 base = get_base_data(c->base);
845 raw = get_data_from_pack(obj);
92392b4a
SP
846 c->data = patch_delta(
847 base, c->base->size,
848 raw, obj->size,
849 &c->size);
850 free(raw);
851 if (!c->data)
c2b97ecf 852 bad_object(obj->idx.offset, _("failed to apply delta"));
b8a2486f 853 get_thread_data()->base_cache_used += c->size;
20e95d0a 854 prune_base_data(c);
9441b61d 855 }
20e95d0a 856 free(delta);
92392b4a
SP
857 }
858 return c->data;
859}
860
f41aebd4 861static void resolve_delta(struct object_entry *delta_obj,
9441b61d 862 struct base_data *base, struct base_data *result)
9cf6d335 863{
ce3f6dc6 864 void *base_data, *delta_data;
9cf6d335 865
ce3f6dc6 866 delta_obj->real_type = base->obj->real_type;
3aba2fdd
NTND
867 if (show_stat) {
868 delta_obj->delta_depth = base->obj->delta_depth + 1;
869 deepest_delta_lock();
870 if (deepest_delta < delta_obj->delta_depth)
871 deepest_delta = delta_obj->delta_depth;
872 deepest_delta_unlock();
873 }
38a45561 874 delta_obj->base_object_no = base->obj - objects;
636171cb 875 delta_data = get_data_from_pack(delta_obj);
ce3f6dc6 876 base_data = get_base_data(base);
9441b61d 877 result->obj = delta_obj;
ce3f6dc6
NP
878 result->data = patch_delta(base_data, base->size,
879 delta_data, delta_obj->size, &result->size);
9cf6d335 880 free(delta_data);
9441b61d 881 if (!result->data)
c2b97ecf 882 bad_object(delta_obj->idx.offset, _("failed to apply delta"));
681b07de
NTND
883 hash_sha1_file(result->data, result->size,
884 typename(delta_obj->real_type), delta_obj->idx.sha1);
9ec2dde9 885 sha1_object(result->data, NULL, result->size, delta_obj->real_type,
9441b61d 886 delta_obj->idx.sha1);
b8a2486f 887 counter_lock();
636171cb 888 nr_resolved_deltas++;
b8a2486f 889 counter_unlock();
9441b61d
NP
890}
891
2baad220
NTND
892static struct base_data *find_unresolved_deltas_1(struct base_data *base,
893 struct base_data *prev_base)
9441b61d 894{
2baad220 895 if (base->ref_last == -1 && base->ofs_last == -1) {
9441b61d
NP
896 union delta_base base_spec;
897
898 hashcpy(base_spec.sha1, base->obj->idx.sha1);
7218a215 899 find_delta_children(&base_spec,
2baad220 900 &base->ref_first, &base->ref_last, OBJ_REF_DELTA);
53dda6ff 901
9441b61d
NP
902 memset(&base_spec, 0, sizeof(base_spec));
903 base_spec.offset = base->obj->idx.offset;
7218a215 904 find_delta_children(&base_spec,
2baad220 905 &base->ofs_first, &base->ofs_last, OBJ_OFS_DELTA);
4a438cab 906
2baad220
NTND
907 if (base->ref_last == -1 && base->ofs_last == -1) {
908 free(base->data);
909 return NULL;
910 }
53dda6ff 911
2baad220
NTND
912 link_base_data(prev_base, base);
913 }
4a438cab 914
2baad220
NTND
915 if (base->ref_first <= base->ref_last) {
916 struct object_entry *child = objects + deltas[base->ref_first].obj_no;
917 struct base_data *result = alloc_base_data();
7218a215
JH
918
919 assert(child->real_type == OBJ_REF_DELTA);
2baad220
NTND
920 resolve_delta(child, base, result);
921 if (base->ref_first == base->ref_last && base->ofs_last == -1)
7218a215 922 free_base_data(base);
2baad220
NTND
923
924 base->ref_first++;
925 return result;
53dda6ff
NP
926 }
927
2baad220
NTND
928 if (base->ofs_first <= base->ofs_last) {
929 struct object_entry *child = objects + deltas[base->ofs_first].obj_no;
930 struct base_data *result = alloc_base_data();
7218a215
JH
931
932 assert(child->real_type == OBJ_OFS_DELTA);
2baad220
NTND
933 resolve_delta(child, base, result);
934 if (base->ofs_first == base->ofs_last)
7218a215 935 free_base_data(base);
2baad220
NTND
936
937 base->ofs_first++;
938 return result;
9cf6d335 939 }
53dda6ff 940
9441b61d 941 unlink_base_data(base);
2baad220
NTND
942 return NULL;
943}
944
945static void find_unresolved_deltas(struct base_data *base)
946{
947 struct base_data *new_base, *prev_base = NULL;
948 for (;;) {
949 new_base = find_unresolved_deltas_1(base, prev_base);
950
951 if (new_base) {
952 prev_base = base;
953 base = new_base;
954 } else {
955 free(base);
956 base = prev_base;
957 if (!base)
958 return;
959 prev_base = base->base;
960 }
961 }
9cf6d335
SV
962}
963
964static int compare_delta_entry(const void *a, const void *b)
965{
966 const struct delta_entry *delta_a = a;
967 const struct delta_entry *delta_b = b;
7218a215
JH
968
969 /* group by type (ref vs ofs) and then by value (sha-1 or offset) */
970 return compare_delta_bases(&delta_a->base, &delta_b->base,
971 objects[delta_a->obj_no].type,
972 objects[delta_b->obj_no].type);
9cf6d335
SV
973}
974
5272f755
NTND
975static void resolve_base(struct object_entry *obj)
976{
977 struct base_data *base_obj = alloc_base_data();
978 base_obj->obj = obj;
979 base_obj->data = NULL;
980 find_unresolved_deltas(base_obj);
981}
982
b8a2486f
NTND
983#ifndef NO_PTHREADS
984static void *threaded_second_pass(void *data)
985{
986 set_thread_data(data);
987 for (;;) {
988 int i;
8f82aad4 989 counter_lock();
b8a2486f 990 display_progress(progress, nr_resolved_deltas);
8f82aad4
TR
991 counter_unlock();
992 work_lock();
b8a2486f
NTND
993 while (nr_dispatched < nr_objects &&
994 is_delta_type(objects[nr_dispatched].type))
995 nr_dispatched++;
996 if (nr_dispatched >= nr_objects) {
997 work_unlock();
998 break;
999 }
1000 i = nr_dispatched++;
1001 work_unlock();
1002
1003 resolve_base(&objects[i]);
1004 }
1005 return NULL;
1006}
1007#endif
1008
5272f755
NTND
1009/*
1010 * First pass:
1011 * - find locations of all objects;
1012 * - calculate SHA1 of all non-delta objects;
1013 * - remember base (SHA1 or offset) for all deltas.
1014 */
2d477051 1015static void parse_pack_objects(unsigned char *sha1)
9cf6d335 1016{
9ec2dde9 1017 int i, nr_delays = 0;
53dda6ff 1018 struct delta_entry *delta = deltas;
2d477051 1019 struct stat st;
9cf6d335 1020
13aaf148 1021 if (verbose)
29e63ed3 1022 progress = start_progress(
c2b97ecf 1023 from_stdin ? _("Receiving objects") : _("Indexing objects"),
29e63ed3 1024 nr_objects);
9cf6d335
SV
1025 for (i = 0; i < nr_objects; i++) {
1026 struct object_entry *obj = &objects[i];
681b07de 1027 void *data = unpack_raw_entry(obj, &delta->base, obj->idx.sha1);
9cf6d335 1028 obj->real_type = obj->type;
4f8ec74e 1029 if (is_delta_type(obj->type)) {
53dda6ff 1030 nr_deltas++;
636171cb 1031 delta->obj_no = i;
53dda6ff 1032 delta++;
9ec2dde9
NTND
1033 } else if (!data) {
1034 /* large blobs, check later */
1035 obj->real_type = OBJ_BAD;
1036 nr_delays++;
9cf6d335 1037 } else
9ec2dde9 1038 sha1_object(data, NULL, obj->size, obj->type, obj->idx.sha1);
9cf6d335 1039 free(data);
4d4fcc54 1040 display_progress(progress, i+1);
9cf6d335 1041 }
aa7e44bf 1042 objects[i].idx.offset = consumed_bytes;
4d4fcc54 1043 stop_progress(&progress);
2d477051
NP
1044
1045 /* Check pack integrity */
636171cb 1046 flush();
9126f009 1047 git_SHA1_Final(sha1, &input_ctx);
2d477051 1048 if (hashcmp(fill(20), sha1))
c2b97ecf 1049 die(_("pack is corrupted (SHA1 mismatch)"));
9bee2478 1050 use(20);
2d477051
NP
1051
1052 /* If input_fd is a file, we should have reached its end now. */
1053 if (fstat(input_fd, &st))
c2b97ecf 1054 die_errno(_("cannot fstat packfile"));
fa257b05
JS
1055 if (S_ISREG(st.st_mode) &&
1056 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
c2b97ecf 1057 die(_("pack has junk at the end"));
9ec2dde9
NTND
1058
1059 for (i = 0; i < nr_objects; i++) {
1060 struct object_entry *obj = &objects[i];
1061 if (obj->real_type != OBJ_BAD)
1062 continue;
1063 obj->real_type = obj->type;
1064 sha1_object(NULL, obj, obj->size, obj->type, obj->idx.sha1);
1065 nr_delays--;
1066 }
1067 if (nr_delays)
1068 die(_("confusion beyond insanity in parse_pack_objects()"));
5272f755
NTND
1069}
1070
1071/*
1072 * Second pass:
1073 * - for all non-delta objects, look if it is used as a base for
1074 * deltas;
1075 * - if used as a base, uncompress the object and apply all deltas,
1076 * recursively checking if the resulting object is used as a base
1077 * for some more deltas.
1078 */
1079static void resolve_deltas(void)
1080{
1081 int i;
9cf6d335 1082
3c9af366
NP
1083 if (!nr_deltas)
1084 return;
1085
53dda6ff 1086 /* Sort deltas by base SHA1/offset for fast searching */
9cf6d335
SV
1087 qsort(deltas, nr_deltas, sizeof(struct delta_entry),
1088 compare_delta_entry);
1089
13aaf148 1090 if (verbose)
c2b97ecf 1091 progress = start_progress(_("Resolving deltas"), nr_deltas);
b8a2486f
NTND
1092
1093#ifndef NO_PTHREADS
1094 nr_dispatched = 0;
1095 if (nr_threads > 1 || getenv("GIT_FORCE_THREADS")) {
1096 init_thread();
1097 for (i = 0; i < nr_threads; i++) {
1098 int ret = pthread_create(&thread_data[i].thread, NULL,
1099 threaded_second_pass, thread_data + i);
1100 if (ret)
f350df42
NTND
1101 die(_("unable to create thread: %s"),
1102 strerror(ret));
b8a2486f
NTND
1103 }
1104 for (i = 0; i < nr_threads; i++)
1105 pthread_join(thread_data[i].thread, NULL);
1106 cleanup_thread();
1107 return;
1108 }
1109#endif
1110
9cf6d335
SV
1111 for (i = 0; i < nr_objects; i++) {
1112 struct object_entry *obj = &objects[i];
9cf6d335 1113
4f8ec74e 1114 if (is_delta_type(obj->type))
9cf6d335 1115 continue;
5272f755 1116 resolve_base(obj);
4d4fcc54 1117 display_progress(progress, nr_resolved_deltas);
9cf6d335 1118 }
636171cb
NP
1119}
1120
5272f755
NTND
1121/*
1122 * Third pass:
1123 * - append objects to convert thin pack to full pack if required
1124 * - write the final 20-byte SHA-1
1125 */
1126static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved);
1127static void conclude_pack(int fix_thin_pack, const char *curr_pack, unsigned char *pack_sha1)
1128{
1129 if (nr_deltas == nr_resolved_deltas) {
1130 stop_progress(&progress);
1131 /* Flush remaining pack final 20-byte SHA1. */
1132 flush();
1133 return;
1134 }
1135
1136 if (fix_thin_pack) {
1137 struct sha1file *f;
1138 unsigned char read_sha1[20], tail_sha1[20];
5c3459fc 1139 struct strbuf msg = STRBUF_INIT;
5272f755
NTND
1140 int nr_unresolved = nr_deltas - nr_resolved_deltas;
1141 int nr_objects_initial = nr_objects;
1142 if (nr_unresolved <= 0)
cc13431a 1143 die(_("confusion beyond insanity"));
5272f755
NTND
1144 objects = xrealloc(objects,
1145 (nr_objects + nr_unresolved + 1)
1146 * sizeof(*objects));
57165db0
JK
1147 memset(objects + nr_objects + 1, 0,
1148 nr_unresolved * sizeof(*objects));
5272f755
NTND
1149 f = sha1fd(output_fd, curr_pack);
1150 fix_unresolved_deltas(f, nr_unresolved);
5c3459fc
NTND
1151 strbuf_addf(&msg, _("completed with %d local objects"),
1152 nr_objects - nr_objects_initial);
1153 stop_progress_msg(&progress, msg.buf);
1154 strbuf_release(&msg);
5272f755
NTND
1155 sha1close(f, tail_sha1, 0);
1156 hashcpy(read_sha1, pack_sha1);
1157 fixup_pack_header_footer(output_fd, pack_sha1,
1158 curr_pack, nr_objects,
1159 read_sha1, consumed_bytes-20);
1160 if (hashcmp(read_sha1, tail_sha1) != 0)
f350df42
NTND
1161 die(_("Unexpected tail checksum for %s "
1162 "(disk corruption?)"), curr_pack);
5272f755
NTND
1163 }
1164 if (nr_deltas != nr_resolved_deltas)
cc13431a
JH
1165 die(Q_("pack has %d unresolved delta",
1166 "pack has %d unresolved deltas",
1167 nr_deltas - nr_resolved_deltas),
5272f755
NTND
1168 nr_deltas - nr_resolved_deltas);
1169}
1170
8522148f 1171static int write_compressed(struct sha1file *f, void *in, unsigned int size)
636171cb 1172{
ef49a7a0 1173 git_zstream stream;
7734d7f2
NP
1174 int status;
1175 unsigned char outbuf[4096];
636171cb
NP
1176
1177 memset(&stream, 0, sizeof(stream));
55bb5c91 1178 git_deflate_init(&stream, zlib_compression_level);
636171cb
NP
1179 stream.next_in = in;
1180 stream.avail_in = size;
636171cb 1181
7734d7f2
NP
1182 do {
1183 stream.next_out = outbuf;
1184 stream.avail_out = sizeof(outbuf);
55bb5c91 1185 status = git_deflate(&stream, Z_FINISH);
7734d7f2
NP
1186 sha1write(f, outbuf, sizeof(outbuf) - stream.avail_out);
1187 } while (status == Z_OK);
1188
1189 if (status != Z_STREAM_END)
c2b97ecf 1190 die(_("unable to deflate appended object (%d)"), status);
636171cb 1191 size = stream.total_out;
55bb5c91 1192 git_deflate_end(&stream);
636171cb
NP
1193 return size;
1194}
1195
8522148f 1196static struct object_entry *append_obj_to_pack(struct sha1file *f,
03993e13 1197 const unsigned char *sha1, void *buf,
636171cb
NP
1198 unsigned long size, enum object_type type)
1199{
1200 struct object_entry *obj = &objects[nr_objects++];
1201 unsigned char header[10];
1202 unsigned long s = size;
1203 int n = 0;
1204 unsigned char c = (type << 4) | (s & 15);
1205 s >>= 4;
1206 while (s) {
1207 header[n++] = c | 0x80;
1208 c = s & 0x7f;
1209 s >>= 7;
1210 }
1211 header[n++] = c;
8522148f
NP
1212 crc32_begin(f);
1213 sha1write(f, header, n);
72de2883
BS
1214 obj[0].size = size;
1215 obj[0].hdr_size = n;
1216 obj[0].type = type;
1217 obj[0].real_type = type;
aa7e44bf 1218 obj[1].idx.offset = obj[0].idx.offset + n;
8522148f
NP
1219 obj[1].idx.offset += write_compressed(f, buf, size);
1220 obj[0].idx.crc32 = crc32_end(f);
838cd346 1221 sha1flush(f);
aa7e44bf 1222 hashcpy(obj->idx.sha1, sha1);
03993e13 1223 return obj;
636171cb
NP
1224}
1225
1226static int delta_pos_compare(const void *_a, const void *_b)
1227{
1228 struct delta_entry *a = *(struct delta_entry **)_a;
1229 struct delta_entry *b = *(struct delta_entry **)_b;
1230 return a->obj_no - b->obj_no;
1231}
9cf6d335 1232
8522148f 1233static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
636171cb
NP
1234{
1235 struct delta_entry **sorted_by_pos;
96a02f8f 1236 int i, n = 0;
636171cb
NP
1237
1238 /*
1239 * Since many unresolved deltas may well be themselves base objects
1240 * for more unresolved deltas, we really want to include the
1241 * smallest number of base objects that would cover as much delta
1242 * as possible by picking the
1243 * trunc deltas first, allowing for other deltas to resolve without
1244 * additional base objects. Since most base objects are to be found
1245 * before deltas depending on them, a good heuristic is to start
1246 * resolving deltas in the same order as their position in the pack.
1247 */
1248 sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
9cf6d335 1249 for (i = 0; i < nr_deltas; i++) {
636171cb
NP
1250 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
1251 continue;
1252 sorted_by_pos[n++] = &deltas[i];
9cf6d335 1253 }
636171cb
NP
1254 qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
1255
1256 for (i = 0; i < n; i++) {
1257 struct delta_entry *d = sorted_by_pos[i];
21666f1a 1258 enum object_type type;
2baad220 1259 struct base_data *base_obj = alloc_base_data();
636171cb
NP
1260
1261 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
1262 continue;
2baad220
NTND
1263 base_obj->data = read_sha1_file(d->base.sha1, &type, &base_obj->size);
1264 if (!base_obj->data)
636171cb 1265 continue;
636171cb 1266
2baad220
NTND
1267 if (check_sha1_signature(d->base.sha1, base_obj->data,
1268 base_obj->size, typename(type)))
c2b97ecf 1269 die(_("local object %s is corrupt"), sha1_to_hex(d->base.sha1));
2baad220
NTND
1270 base_obj->obj = append_obj_to_pack(f, d->base.sha1,
1271 base_obj->data, base_obj->size, type);
1272 find_unresolved_deltas(base_obj);
4d4fcc54 1273 display_progress(progress, nr_resolved_deltas);
636171cb
NP
1274 }
1275 free(sorted_by_pos);
1276}
1277
e42797f5
NP
1278static void final(const char *final_pack_name, const char *curr_pack_name,
1279 const char *final_index_name, const char *curr_index_name,
b8077709 1280 const char *keep_name, const char *keep_msg,
e42797f5
NP
1281 unsigned char *sha1)
1282{
3a55602e 1283 const char *report = "pack";
e42797f5
NP
1284 char name[PATH_MAX];
1285 int err;
1286
1287 if (!from_stdin) {
1288 close(input_fd);
1289 } else {
4c81b03e 1290 fsync_or_die(output_fd, curr_pack_name);
e42797f5
NP
1291 err = close(output_fd);
1292 if (err)
c2b97ecf 1293 die_errno(_("error while closing pack file"));
e42797f5
NP
1294 }
1295
b8077709
SP
1296 if (keep_msg) {
1297 int keep_fd, keep_msg_len = strlen(keep_msg);
6e180cdc
JH
1298
1299 if (!keep_name)
1300 keep_fd = odb_pack_keep(name, sizeof(name), sha1);
1301 else
1302 keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
1303
9ca4a201
NP
1304 if (keep_fd < 0) {
1305 if (errno != EEXIST)
c2b97ecf 1306 die_errno(_("cannot write keep file '%s'"),
de983a0a 1307 keep_name ? keep_name : name);
9ca4a201
NP
1308 } else {
1309 if (keep_msg_len > 0) {
1310 write_or_die(keep_fd, keep_msg, keep_msg_len);
1311 write_or_die(keep_fd, "\n", 1);
1312 }
91c8d590 1313 if (close(keep_fd) != 0)
c2b97ecf 1314 die_errno(_("cannot close written keep file '%s'"),
de983a0a 1315 keep_name ? keep_name : name);
576162a4 1316 report = "keep";
b8077709 1317 }
b8077709
SP
1318 }
1319
e42797f5
NP
1320 if (final_pack_name != curr_pack_name) {
1321 if (!final_pack_name) {
1322 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
1323 get_object_directory(), sha1_to_hex(sha1));
1324 final_pack_name = name;
1325 }
1326 if (move_temp_to_file(curr_pack_name, final_pack_name))
c2b97ecf 1327 die(_("cannot store pack file"));
fb8b1936 1328 } else if (from_stdin)
33b65030 1329 chmod(final_pack_name, 0444);
e42797f5 1330
e42797f5
NP
1331 if (final_index_name != curr_index_name) {
1332 if (!final_index_name) {
1333 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
1334 get_object_directory(), sha1_to_hex(sha1));
1335 final_index_name = name;
1336 }
1337 if (move_temp_to_file(curr_index_name, final_index_name))
c2b97ecf 1338 die(_("cannot store index file"));
fb8b1936
JH
1339 } else
1340 chmod(final_index_name, 0444);
576162a4
NP
1341
1342 if (!from_stdin) {
1343 printf("%s\n", sha1_to_hex(sha1));
1344 } else {
1345 char buf[48];
1346 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
1347 report, sha1_to_hex(sha1));
d1b2ddc8 1348 write_or_die(1, buf, len);
576162a4
NP
1349
1350 /*
1351 * Let's just mimic git-unpack-objects here and write
1352 * the last part of the input buffer to stdout.
1353 */
1354 while (input_len) {
1355 err = xwrite(1, input_buffer + input_offset, input_len);
1356 if (err <= 0)
1357 break;
1358 input_len -= err;
1359 input_offset += err;
1360 }
1361 }
9cf6d335
SV
1362}
1363
ef90d6d4 1364static int git_index_pack_config(const char *k, const char *v, void *cb)
4d00bda2 1365{
ebcfb379
JH
1366 struct pack_idx_option *opts = cb;
1367
4d00bda2 1368 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
1369 opts->version = git_config_int(k, v);
1370 if (opts->version > 2)
f350df42 1371 die(_("bad pack.indexversion=%"PRIu32), opts->version);
4d00bda2
NP
1372 return 0;
1373 }
b8a2486f
NTND
1374 if (!strcmp(k, "pack.threads")) {
1375 nr_threads = git_config_int(k, v);
1376 if (nr_threads < 0)
f350df42 1377 die(_("invalid number of threads specified (%d)"),
b8a2486f
NTND
1378 nr_threads);
1379#ifdef NO_PTHREADS
1380 if (nr_threads != 1)
f350df42 1381 warning(_("no threads support, ignoring %s"), k);
b8a2486f
NTND
1382 nr_threads = 1;
1383#endif
1384 return 0;
1385 }
ef90d6d4 1386 return git_default_config(k, v, cb);
4d00bda2
NP
1387}
1388
3c9fc074
JH
1389static int cmp_uint32(const void *a_, const void *b_)
1390{
1391 uint32_t a = *((uint32_t *)a_);
1392 uint32_t b = *((uint32_t *)b_);
1393
1394 return (a < b) ? -1 : (a != b);
1395}
1396
1397static void read_v2_anomalous_offsets(struct packed_git *p,
1398 struct pack_idx_option *opts)
1399{
1400 const uint32_t *idx1, *idx2;
1401 uint32_t i;
1402
1403 /* The address of the 4-byte offset table */
1404 idx1 = (((const uint32_t *)p->index_data)
1405 + 2 /* 8-byte header */
1406 + 256 /* fan out */
1407 + 5 * p->num_objects /* 20-byte SHA-1 table */
1408 + p->num_objects /* CRC32 table */
1409 );
1410
1411 /* The address of the 8-byte offset table */
1412 idx2 = idx1 + p->num_objects;
1413
1414 for (i = 0; i < p->num_objects; i++) {
1415 uint32_t off = ntohl(idx1[i]);
1416 if (!(off & 0x80000000))
1417 continue;
1418 off = off & 0x7fffffff;
1419 if (idx2[off * 2])
1420 continue;
1421 /*
1422 * The real offset is ntohl(idx2[off * 2]) in high 4
1423 * octets, and ntohl(idx2[off * 2 + 1]) in low 4
1424 * octets. But idx2[off * 2] is Zero!!!
1425 */
1426 ALLOC_GROW(opts->anomaly, opts->anomaly_nr + 1, opts->anomaly_alloc);
1427 opts->anomaly[opts->anomaly_nr++] = ntohl(idx2[off * 2 + 1]);
1428 }
1429
1430 if (1 < opts->anomaly_nr)
1431 qsort(opts->anomaly, opts->anomaly_nr, sizeof(uint32_t), cmp_uint32);
1432}
1433
e337a04d
JH
1434static void read_idx_option(struct pack_idx_option *opts, const char *pack_name)
1435{
1436 struct packed_git *p = add_packed_git(pack_name, strlen(pack_name), 1);
1437
1438 if (!p)
c2b97ecf 1439 die(_("Cannot open existing pack file '%s'"), pack_name);
e337a04d 1440 if (open_pack_index(p))
c2b97ecf 1441 die(_("Cannot open existing pack idx file for '%s'"), pack_name);
e337a04d
JH
1442
1443 /* Read the attributes from the existing idx file */
1444 opts->version = p->index_version;
1445
3c9fc074
JH
1446 if (opts->version == 2)
1447 read_v2_anomalous_offsets(p, opts);
1448
e337a04d
JH
1449 /*
1450 * Get rid of the idx file as we do not need it anymore.
1451 * NEEDSWORK: extract this bit from free_pack_by_name() in
1452 * sha1_file.c, perhaps? It shouldn't matter very much as we
1453 * know we haven't installed this pack (hence we never have
1454 * read anything from it).
1455 */
1456 close_pack_index(p);
1457 free(p);
1458}
1459
38a45561
JH
1460static void show_pack_info(int stat_only)
1461{
d1a0ed18
JH
1462 int i, baseobjects = nr_objects - nr_deltas;
1463 unsigned long *chain_histogram = NULL;
1464
1465 if (deepest_delta)
1466 chain_histogram = xcalloc(deepest_delta, sizeof(unsigned long));
1467
38a45561
JH
1468 for (i = 0; i < nr_objects; i++) {
1469 struct object_entry *obj = &objects[i];
1470
d1a0ed18
JH
1471 if (is_delta_type(obj->type))
1472 chain_histogram[obj->delta_depth - 1]++;
38a45561
JH
1473 if (stat_only)
1474 continue;
1475 printf("%s %-6s %lu %lu %"PRIuMAX,
1476 sha1_to_hex(obj->idx.sha1),
1477 typename(obj->real_type), obj->size,
1478 (unsigned long)(obj[1].idx.offset - obj->idx.offset),
1479 (uintmax_t)obj->idx.offset);
1480 if (is_delta_type(obj->type)) {
1481 struct object_entry *bobj = &objects[obj->base_object_no];
1482 printf(" %u %s", obj->delta_depth, sha1_to_hex(bobj->idx.sha1));
1483 }
1484 putchar('\n');
1485 }
d1a0ed18
JH
1486
1487 if (baseobjects)
c2b97ecf
NTND
1488 printf_ln(Q_("non delta: %d object",
1489 "non delta: %d objects",
1490 baseobjects),
1491 baseobjects);
d1a0ed18
JH
1492 for (i = 0; i < deepest_delta; i++) {
1493 if (!chain_histogram[i])
1494 continue;
c2b97ecf
NTND
1495 printf_ln(Q_("chain length = %d: %lu object",
1496 "chain length = %d: %lu objects",
1497 chain_histogram[i]),
1498 i + 1,
1499 chain_histogram[i]);
d1a0ed18 1500 }
38a45561
JH
1501}
1502
3bb72562 1503int cmd_index_pack(int argc, const char **argv, const char *prefix)
9cf6d335 1504{
3aba2fdd 1505 int i, fix_thin_pack = 0, verify = 0, stat_only = 0;
39539495 1506 const char *curr_index;
3bb72562 1507 const char *index_name = NULL, *pack_name = NULL;
b8077709 1508 const char *keep_name = NULL, *keep_msg = NULL;
592ce208
JK
1509 struct strbuf index_name_buf = STRBUF_INIT,
1510 keep_name_buf = STRBUF_INIT;
aa7e44bf 1511 struct pack_idx_entry **idx_objects;
ebcfb379 1512 struct pack_idx_option opts;
8522148f 1513 unsigned char pack_sha1[20];
c6807a40 1514 unsigned foreign_nr = 1; /* zero is a "good" value, assume bad */
9cf6d335 1515
99caeed0
JN
1516 if (argc == 2 && !strcmp(argv[1], "-h"))
1517 usage(index_pack_usage);
1518
afc711b8 1519 check_replace_refs = 0;
6e2a09d2 1520
ebcfb379
JH
1521 reset_pack_idx_option(&opts);
1522 git_config(git_index_pack_config, &opts);
3f8099fc 1523 if (prefix && chdir(prefix))
c2b97ecf 1524 die(_("Cannot come back to cwd"));
4d00bda2 1525
9cf6d335 1526 for (i = 1; i < argc; i++) {
3bb72562 1527 const char *arg = argv[i];
9cf6d335
SV
1528
1529 if (*arg == '-') {
e42797f5
NP
1530 if (!strcmp(arg, "--stdin")) {
1531 from_stdin = 1;
636171cb
NP
1532 } else if (!strcmp(arg, "--fix-thin")) {
1533 fix_thin_pack = 1;
0153be05
MK
1534 } else if (!strcmp(arg, "--strict")) {
1535 strict = 1;
c6807a40
NTND
1536 do_fsck_object = 1;
1537 } else if (!strcmp(arg, "--check-self-contained-and-connected")) {
1538 strict = 1;
1539 check_self_contained_and_connected = 1;
e337a04d
JH
1540 } else if (!strcmp(arg, "--verify")) {
1541 verify = 1;
38a45561
JH
1542 } else if (!strcmp(arg, "--verify-stat")) {
1543 verify = 1;
3aba2fdd 1544 show_stat = 1;
38a45561
JH
1545 } else if (!strcmp(arg, "--verify-stat-only")) {
1546 verify = 1;
3aba2fdd 1547 show_stat = 1;
38a45561 1548 stat_only = 1;
b8077709
SP
1549 } else if (!strcmp(arg, "--keep")) {
1550 keep_msg = "";
59556548 1551 } else if (starts_with(arg, "--keep=")) {
b8077709 1552 keep_msg = arg + 7;
59556548 1553 } else if (starts_with(arg, "--threads=")) {
b8a2486f
NTND
1554 char *end;
1555 nr_threads = strtoul(arg+10, &end, 0);
1556 if (!arg[10] || *end || nr_threads < 0)
1557 usage(index_pack_usage);
1558#ifdef NO_PTHREADS
1559 if (nr_threads != 1)
f350df42
NTND
1560 warning(_("no threads support, "
1561 "ignoring %s"), arg);
b8a2486f
NTND
1562 nr_threads = 1;
1563#endif
59556548 1564 } else if (starts_with(arg, "--pack_header=")) {
bed006fb
NP
1565 struct pack_header *hdr;
1566 char *c;
1567
1568 hdr = (struct pack_header *)input_buffer;
1569 hdr->hdr_signature = htonl(PACK_SIGNATURE);
1570 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
1571 if (*c != ',')
c2b97ecf 1572 die(_("bad %s"), arg);
bed006fb
NP
1573 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
1574 if (*c)
c2b97ecf 1575 die(_("bad %s"), arg);
bed006fb 1576 input_len = sizeof(*hdr);
3c9af366
NP
1577 } else if (!strcmp(arg, "-v")) {
1578 verbose = 1;
e42797f5 1579 } else if (!strcmp(arg, "-o")) {
9cf6d335
SV
1580 if (index_name || (i+1) >= argc)
1581 usage(index_pack_usage);
1582 index_name = argv[++i];
59556548 1583 } else if (starts_with(arg, "--index-version=")) {
4ba7d711 1584 char *c;
ebcfb379
JH
1585 opts.version = strtoul(arg + 16, &c, 10);
1586 if (opts.version > 2)
c2b97ecf 1587 die(_("bad %s"), arg);
4ba7d711 1588 if (*c == ',')
ebcfb379
JH
1589 opts.off32_limit = strtoul(c+1, &c, 0);
1590 if (*c || opts.off32_limit & 0x80000000)
c2b97ecf 1591 die(_("bad %s"), arg);
9cf6d335
SV
1592 } else
1593 usage(index_pack_usage);
1594 continue;
1595 }
1596
1597 if (pack_name)
1598 usage(index_pack_usage);
1599 pack_name = arg;
1600 }
1601
e42797f5 1602 if (!pack_name && !from_stdin)
9cf6d335 1603 usage(index_pack_usage);
636171cb 1604 if (fix_thin_pack && !from_stdin)
c2b97ecf 1605 die(_("--fix-thin cannot be used without --stdin"));
e42797f5 1606 if (!index_name && pack_name) {
592ce208
JK
1607 size_t len;
1608 if (!strip_suffix(pack_name, ".pack", &len))
c2b97ecf 1609 die(_("packfile name '%s' does not end with '.pack'"),
9cf6d335 1610 pack_name);
592ce208
JK
1611 strbuf_add(&index_name_buf, pack_name, len);
1612 strbuf_addstr(&index_name_buf, ".idx");
1613 index_name = index_name_buf.buf;
9cf6d335 1614 }
b8077709 1615 if (keep_msg && !keep_name && pack_name) {
592ce208
JK
1616 size_t len;
1617 if (!strip_suffix(pack_name, ".pack", &len))
c2b97ecf 1618 die(_("packfile name '%s' does not end with '.pack'"),
b8077709 1619 pack_name);
592ce208
JK
1620 strbuf_add(&keep_name_buf, pack_name, len);
1621 strbuf_addstr(&keep_name_buf, ".idx");
1622 keep_name = keep_name_buf.buf;
b8077709 1623 }
e337a04d
JH
1624 if (verify) {
1625 if (!index_name)
c2b97ecf 1626 die(_("--verify with no packfile name given"));
e337a04d 1627 read_idx_option(&opts, index_name);
68be2fea 1628 opts.flags |= WRITE_IDX_VERIFY | WRITE_IDX_STRICT;
e337a04d 1629 }
68be2fea
JH
1630 if (strict)
1631 opts.flags |= WRITE_IDX_STRICT;
9cf6d335 1632
b8a2486f
NTND
1633#ifndef NO_PTHREADS
1634 if (!nr_threads) {
1635 nr_threads = online_cpus();
1636 /* An experiment showed that more threads does not mean faster */
1637 if (nr_threads > 3)
1638 nr_threads = 3;
1639 }
1640#endif
1641
e42797f5 1642 curr_pack = open_pack_file(pack_name);
9cf6d335 1643 parse_pack_header();
38a45561
JH
1644 objects = xcalloc(nr_objects + 1, sizeof(struct object_entry));
1645 deltas = xcalloc(nr_objects, sizeof(struct delta_entry));
8522148f 1646 parse_pack_objects(pack_sha1);
5272f755
NTND
1647 resolve_deltas();
1648 conclude_pack(fix_thin_pack, curr_pack, pack_sha1);
9cf6d335 1649 free(deltas);
0153be05 1650 if (strict)
c6807a40 1651 foreign_nr = check_objects();
aa7e44bf 1652
3aba2fdd 1653 if (show_stat)
38a45561
JH
1654 show_pack_info(stat_only);
1655
aa7e44bf
GB
1656 idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1657 for (i = 0; i < nr_objects; i++)
1658 idx_objects[i] = &objects[i].idx;
ebcfb379 1659 curr_index = write_idx_file(index_name, idx_objects, nr_objects, &opts, pack_sha1);
aa7e44bf
GB
1660 free(idx_objects);
1661
e337a04d
JH
1662 if (!verify)
1663 final(pack_name, curr_pack,
1664 index_name, curr_index,
1665 keep_name, keep_msg,
1666 pack_sha1);
1667 else
1668 close(input_fd);
9cf6d335 1669 free(objects);
592ce208
JK
1670 strbuf_release(&index_name_buf);
1671 strbuf_release(&keep_name_buf);
c85228ed 1672 if (pack_name == NULL)
3bb72562 1673 free((void *) curr_pack);
c85228ed 1674 if (index_name == NULL)
3bb72562 1675 free((void *) curr_index);
9cf6d335 1676
c6807a40
NTND
1677 /*
1678 * Let the caller know this pack is not self contained
1679 */
1680 if (check_self_contained_and_connected && foreign_nr)
1681 return 1;
1682
9cf6d335
SV
1683 return 0;
1684}