]> git.ipfire.org Git - thirdparty/git.git/blame - index-pack.c
make overflow test on delta base offset work regardless of variable size
[thirdparty/git.git] / index-pack.c
CommitLineData
9cf6d335
SV
1#include "cache.h"
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"
9cf6d335
SV
9
10static const char index_pack_usage[] =
b8077709 11"git-index-pack [-v] [-o <index-file>] [{ ---keep | --keep=<msg> }] { <pack-file> | --stdin [--fix-thin] [<pack-file>] }";
9cf6d335
SV
12
13struct object_entry
14{
15 unsigned long offset;
2d477051
NP
16 unsigned long size;
17 unsigned int hdr_size;
9cf6d335
SV
18 enum object_type type;
19 enum object_type real_type;
20 unsigned char sha1[20];
21};
22
53dda6ff
NP
23union delta_base {
24 unsigned char sha1[20];
25 unsigned long offset;
26};
27
3c552873
NP
28/*
29 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
30 * to memcmp() only the first 20 bytes.
31 */
32#define UNION_BASE_SZ 20
33
9cf6d335
SV
34struct delta_entry
35{
53dda6ff 36 union delta_base base;
636171cb 37 int obj_no;
9cf6d335
SV
38};
39
9cf6d335
SV
40static struct object_entry *objects;
41static struct delta_entry *deltas;
42static int nr_objects;
43static int nr_deltas;
636171cb 44static int nr_resolved_deltas;
9cf6d335 45
e42797f5 46static int from_stdin;
3c9af366
NP
47static int verbose;
48
49static volatile sig_atomic_t progress_update;
50
51static void progress_interval(int signum)
52{
53 progress_update = 1;
54}
55
56static void setup_progress_signal(void)
57{
58 struct sigaction sa;
59 struct itimerval v;
60
61 memset(&sa, 0, sizeof(sa));
62 sa.sa_handler = progress_interval;
63 sigemptyset(&sa.sa_mask);
64 sa.sa_flags = SA_RESTART;
65 sigaction(SIGALRM, &sa, NULL);
66
67 v.it_interval.tv_sec = 1;
68 v.it_interval.tv_usec = 0;
69 v.it_value = v.it_interval;
70 setitimer(ITIMER_REAL, &v, NULL);
71
72}
73
74static unsigned display_progress(unsigned n, unsigned total, unsigned last_pc)
75{
76 unsigned percent = n * 100 / total;
77 if (percent != last_pc || progress_update) {
78 fprintf(stderr, "%4u%% (%u/%u) done\r", percent, n, total);
79 progress_update = 0;
80 }
81 return percent;
82}
e42797f5 83
2d477051
NP
84/* We always read in 4kB chunks. */
85static unsigned char input_buffer[4096];
86static unsigned long input_offset, input_len, consumed_bytes;
87static SHA_CTX input_ctx;
6d2fa7f1 88static int input_fd, output_fd, pack_fd;
2d477051 89
636171cb 90/* Discard current buffer used content. */
a6e8a767 91static void flush(void)
636171cb
NP
92{
93 if (input_offset) {
94 if (output_fd >= 0)
95 write_or_die(output_fd, input_buffer, input_offset);
96 SHA1_Update(&input_ctx, input_buffer, input_offset);
554a2636 97 memmove(input_buffer, input_buffer + input_offset, input_len);
636171cb
NP
98 input_offset = 0;
99 }
100}
101
2d477051
NP
102/*
103 * Make sure at least "min" bytes are available in the buffer, and
104 * return the pointer to the buffer.
105 */
b89c4e93 106static void *fill(int min)
9cf6d335 107{
2d477051
NP
108 if (min <= input_len)
109 return input_buffer + input_offset;
110 if (min > sizeof(input_buffer))
111 die("cannot fill %d bytes", min);
636171cb 112 flush();
2d477051
NP
113 do {
114 int ret = xread(input_fd, input_buffer + input_len,
115 sizeof(input_buffer) - input_len);
116 if (ret <= 0) {
117 if (!ret)
118 die("early EOF");
119 die("read error on input: %s", strerror(errno));
120 }
121 input_len += ret;
122 } while (input_len < min);
123 return input_buffer;
124}
125
126static void use(int bytes)
127{
128 if (bytes > input_len)
129 die("used more bytes than were available");
130 input_len -= bytes;
131 input_offset += bytes;
132 consumed_bytes += bytes;
133}
9cf6d335 134
b89c4e93 135static const char *open_pack_file(const char *pack_name)
2d477051 136{
e42797f5
NP
137 if (from_stdin) {
138 input_fd = 0;
139 if (!pack_name) {
140 static char tmpfile[PATH_MAX];
141 snprintf(tmpfile, sizeof(tmpfile),
0e55181f 142 "%s/tmp_pack_XXXXXX", get_object_directory());
e42797f5
NP
143 output_fd = mkstemp(tmpfile);
144 pack_name = xstrdup(tmpfile);
145 } else
146 output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
147 if (output_fd < 0)
148 die("unable to create %s: %s\n", pack_name, strerror(errno));
6d2fa7f1 149 pack_fd = output_fd;
e42797f5
NP
150 } else {
151 input_fd = open(pack_name, O_RDONLY);
152 if (input_fd < 0)
153 die("cannot open packfile '%s': %s",
154 pack_name, strerror(errno));
155 output_fd = -1;
6d2fa7f1 156 pack_fd = input_fd;
e42797f5 157 }
2d477051 158 SHA1_Init(&input_ctx);
e42797f5 159 return pack_name;
9cf6d335
SV
160}
161
162static void parse_pack_header(void)
163{
2d477051 164 struct pack_header *hdr = fill(sizeof(struct pack_header));
9cf6d335
SV
165
166 /* Header consistency check */
9cf6d335 167 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
e42797f5 168 die("pack signature mismatch");
d60fc1c8 169 if (!pack_version_ok(hdr->hdr_version))
e42797f5 170 die("pack version %d unsupported", ntohl(hdr->hdr_version));
9cf6d335
SV
171
172 nr_objects = ntohl(hdr->hdr_entries);
2d477051 173 use(sizeof(struct pack_header));
9cf6d335
SV
174}
175
176static void bad_object(unsigned long offset, const char *format,
177 ...) NORETURN __attribute__((format (printf, 2, 3)));
178
179static void bad_object(unsigned long offset, const char *format, ...)
180{
181 va_list params;
182 char buf[1024];
183
184 va_start(params, format);
185 vsnprintf(buf, sizeof(buf), format, params);
186 va_end(params);
e42797f5 187 die("pack has bad object at offset %lu: %s", offset, buf);
9cf6d335
SV
188}
189
2d477051 190static void *unpack_entry_data(unsigned long offset, unsigned long size)
9cf6d335 191{
9cf6d335
SV
192 z_stream stream;
193 void *buf = xmalloc(size);
194
195 memset(&stream, 0, sizeof(stream));
196 stream.next_out = buf;
197 stream.avail_out = size;
2d477051
NP
198 stream.next_in = fill(1);
199 stream.avail_in = input_len;
9cf6d335
SV
200 inflateInit(&stream);
201
202 for (;;) {
203 int ret = inflate(&stream, 0);
2d477051
NP
204 use(input_len - stream.avail_in);
205 if (stream.total_out == size && ret == Z_STREAM_END)
9cf6d335
SV
206 break;
207 if (ret != Z_OK)
208 bad_object(offset, "inflate returned %d", ret);
2d477051
NP
209 stream.next_in = fill(1);
210 stream.avail_in = input_len;
9cf6d335
SV
211 }
212 inflateEnd(&stream);
9cf6d335
SV
213 return buf;
214}
215
2d477051 216static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
9cf6d335 217{
2d477051 218 unsigned char *p, c;
53dda6ff 219 unsigned long size, base_offset;
9cf6d335 220 unsigned shift;
9cf6d335 221
2d477051
NP
222 obj->offset = consumed_bytes;
223
224 p = fill(1);
225 c = *p;
226 use(1);
227 obj->type = (c >> 4) & 7;
9cf6d335
SV
228 size = (c & 15);
229 shift = 4;
230 while (c & 0x80) {
2d477051
NP
231 p = fill(1);
232 c = *p;
233 use(1);
9cf6d335
SV
234 size += (c & 0x7fUL) << shift;
235 shift += 7;
236 }
2d477051 237 obj->size = size;
9cf6d335 238
2d477051 239 switch (obj->type) {
eb32d236 240 case OBJ_REF_DELTA:
2d477051
NP
241 hashcpy(delta_base->sha1, fill(20));
242 use(20);
53dda6ff
NP
243 break;
244 case OBJ_OFS_DELTA:
245 memset(delta_base, 0, sizeof(*delta_base));
2d477051
NP
246 p = fill(1);
247 c = *p;
248 use(1);
53dda6ff
NP
249 base_offset = c & 127;
250 while (c & 128) {
251 base_offset += 1;
8723f216 252 if (!base_offset || MSB(base_offset, 7))
2d477051
NP
253 bad_object(obj->offset, "offset value overflow for delta base object");
254 p = fill(1);
255 c = *p;
256 use(1);
53dda6ff
NP
257 base_offset = (base_offset << 7) + (c & 127);
258 }
2d477051
NP
259 delta_base->offset = obj->offset - base_offset;
260 if (delta_base->offset >= obj->offset)
261 bad_object(obj->offset, "delta base offset is out of bound");
53dda6ff 262 break;
9cf6d335
SV
263 case OBJ_COMMIT:
264 case OBJ_TREE:
265 case OBJ_BLOB:
266 case OBJ_TAG:
9cf6d335
SV
267 break;
268 default:
08a19d87 269 bad_object(obj->offset, "unknown object type %d", obj->type);
9cf6d335 270 }
2d477051
NP
271 obj->hdr_size = consumed_bytes - obj->offset;
272
273 return unpack_entry_data(obj->offset, obj->size);
274}
275
b89c4e93 276static void *get_data_from_pack(struct object_entry *obj)
2d477051
NP
277{
278 unsigned long from = obj[0].offset + obj[0].hdr_size;
279 unsigned long len = obj[1].offset - from;
a91d49cd 280 unsigned long rdy = 0;
6d2fa7f1 281 unsigned char *src, *data;
2d477051
NP
282 z_stream stream;
283 int st;
9cf6d335 284
6d2fa7f1 285 src = xmalloc(len);
a91d49cd
SP
286 data = src;
287 do {
288 ssize_t n = pread(pack_fd, data + rdy, len - rdy, from + rdy);
289 if (n <= 0)
290 die("cannot pread pack file: %s", strerror(errno));
291 rdy += n;
292 } while (rdy < len);
2d477051
NP
293 data = xmalloc(obj->size);
294 memset(&stream, 0, sizeof(stream));
295 stream.next_out = data;
296 stream.avail_out = obj->size;
6d2fa7f1 297 stream.next_in = src;
2d477051
NP
298 stream.avail_in = len;
299 inflateInit(&stream);
300 while ((st = inflate(&stream, Z_FINISH)) == Z_OK);
301 inflateEnd(&stream);
302 if (st != Z_STREAM_END || stream.total_out != obj->size)
303 die("serious inflate inconsistency");
6d2fa7f1 304 free(src);
9cf6d335
SV
305 return data;
306}
307
53dda6ff 308static int find_delta(const union delta_base *base)
9cf6d335
SV
309{
310 int first = 0, last = nr_deltas;
311
312 while (first < last) {
313 int next = (first + last) / 2;
314 struct delta_entry *delta = &deltas[next];
315 int cmp;
316
3c552873 317 cmp = memcmp(base, &delta->base, UNION_BASE_SZ);
9cf6d335
SV
318 if (!cmp)
319 return next;
320 if (cmp < 0) {
321 last = next;
322 continue;
323 }
324 first = next+1;
325 }
326 return -first-1;
327}
328
b89c4e93
NP
329static int find_delta_children(const union delta_base *base,
330 int *first_index, int *last_index)
9cf6d335 331{
53dda6ff 332 int first = find_delta(base);
9cf6d335
SV
333 int last = first;
334 int end = nr_deltas - 1;
335
336 if (first < 0)
337 return -1;
3c552873 338 while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
9cf6d335 339 --first;
3c552873 340 while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
9cf6d335
SV
341 ++last;
342 *first_index = first;
343 *last_index = last;
344 return 0;
345}
346
347static void sha1_object(const void *data, unsigned long size,
9096c660 348 enum object_type type, unsigned char *sha1)
9cf6d335 349{
ce9fbf16 350 hash_sha1_file(data, size, typename(type), sha1);
9096c660 351 if (has_sha1_file(sha1)) {
8685da42
NP
352 void *has_data;
353 enum object_type has_type;
354 unsigned long has_size;
355 has_data = read_sha1_file(sha1, &has_type, &has_size);
356 if (!has_data)
357 die("cannot read existing object %s", sha1_to_hex(sha1));
358 if (size != has_size || type != has_type ||
359 memcmp(data, has_data, size) != 0)
360 die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));
bbf4b41b 361 free(has_data);
8685da42 362 }
9cf6d335
SV
363}
364
636171cb 365static void resolve_delta(struct object_entry *delta_obj, void *base_data,
9cf6d335
SV
366 unsigned long base_size, enum object_type type)
367{
9cf6d335
SV
368 void *delta_data;
369 unsigned long delta_size;
370 void *result;
371 unsigned long result_size;
53dda6ff 372 union delta_base delta_base;
9cf6d335
SV
373 int j, first, last;
374
636171cb
NP
375 delta_obj->real_type = type;
376 delta_data = get_data_from_pack(delta_obj);
377 delta_size = delta_obj->size;
9cf6d335
SV
378 result = patch_delta(base_data, base_size, delta_data, delta_size,
379 &result_size);
380 free(delta_data);
381 if (!result)
636171cb 382 bad_object(delta_obj->offset, "failed to apply delta");
9096c660 383 sha1_object(result, result_size, type, delta_obj->sha1);
636171cb 384 nr_resolved_deltas++;
53dda6ff 385
636171cb 386 hashcpy(delta_base.sha1, delta_obj->sha1);
b89c4e93 387 if (!find_delta_children(&delta_base, &first, &last)) {
636171cb
NP
388 for (j = first; j <= last; j++) {
389 struct object_entry *child = objects + deltas[j].obj_no;
390 if (child->real_type == OBJ_REF_DELTA)
391 resolve_delta(child, result, result_size, type);
392 }
53dda6ff
NP
393 }
394
395 memset(&delta_base, 0, sizeof(delta_base));
636171cb 396 delta_base.offset = delta_obj->offset;
b89c4e93 397 if (!find_delta_children(&delta_base, &first, &last)) {
636171cb
NP
398 for (j = first; j <= last; j++) {
399 struct object_entry *child = objects + deltas[j].obj_no;
400 if (child->real_type == OBJ_OFS_DELTA)
401 resolve_delta(child, result, result_size, type);
402 }
9cf6d335 403 }
53dda6ff 404
9cf6d335
SV
405 free(result);
406}
407
408static int compare_delta_entry(const void *a, const void *b)
409{
410 const struct delta_entry *delta_a = a;
411 const struct delta_entry *delta_b = b;
3c552873 412 return memcmp(&delta_a->base, &delta_b->base, UNION_BASE_SZ);
9cf6d335
SV
413}
414
2d477051
NP
415/* Parse all objects and return the pack content SHA1 hash */
416static void parse_pack_objects(unsigned char *sha1)
9cf6d335 417{
3c9af366 418 int i, percent = -1;
53dda6ff 419 struct delta_entry *delta = deltas;
9cf6d335 420 void *data;
2d477051 421 struct stat st;
9cf6d335
SV
422
423 /*
424 * First pass:
425 * - find locations of all objects;
426 * - calculate SHA1 of all non-delta objects;
b89c4e93 427 * - remember base (SHA1 or offset) for all deltas.
9cf6d335 428 */
3c9af366
NP
429 if (verbose)
430 fprintf(stderr, "Indexing %d objects.\n", nr_objects);
9cf6d335
SV
431 for (i = 0; i < nr_objects; i++) {
432 struct object_entry *obj = &objects[i];
2d477051 433 data = unpack_raw_entry(obj, &delta->base);
9cf6d335 434 obj->real_type = obj->type;
53dda6ff
NP
435 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
436 nr_deltas++;
636171cb 437 delta->obj_no = i;
53dda6ff 438 delta++;
9cf6d335 439 } else
9096c660 440 sha1_object(data, obj->size, obj->type, obj->sha1);
9cf6d335 441 free(data);
3c9af366
NP
442 if (verbose)
443 percent = display_progress(i+1, nr_objects, percent);
9cf6d335 444 }
2d477051 445 objects[i].offset = consumed_bytes;
3c9af366
NP
446 if (verbose)
447 fputc('\n', stderr);
2d477051
NP
448
449 /* Check pack integrity */
636171cb 450 flush();
2d477051
NP
451 SHA1_Final(sha1, &input_ctx);
452 if (hashcmp(fill(20), sha1))
e42797f5 453 die("pack is corrupted (SHA1 mismatch)");
9bee2478 454 use(20);
2d477051
NP
455
456 /* If input_fd is a file, we should have reached its end now. */
457 if (fstat(input_fd, &st))
e42797f5 458 die("cannot fstat packfile: %s", strerror(errno));
fa257b05
JS
459 if (S_ISREG(st.st_mode) &&
460 lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
e42797f5 461 die("pack has junk at the end");
9cf6d335 462
3c9af366
NP
463 if (!nr_deltas)
464 return;
465
53dda6ff 466 /* Sort deltas by base SHA1/offset for fast searching */
9cf6d335
SV
467 qsort(deltas, nr_deltas, sizeof(struct delta_entry),
468 compare_delta_entry);
469
470 /*
471 * Second pass:
472 * - for all non-delta objects, look if it is used as a base for
473 * deltas;
474 * - if used as a base, uncompress the object and apply all deltas,
475 * recursively checking if the resulting object is used as a base
476 * for some more deltas.
477 */
3c9af366
NP
478 if (verbose)
479 fprintf(stderr, "Resolving %d deltas.\n", nr_deltas);
9cf6d335
SV
480 for (i = 0; i < nr_objects; i++) {
481 struct object_entry *obj = &objects[i];
53dda6ff
NP
482 union delta_base base;
483 int j, ref, ref_first, ref_last, ofs, ofs_first, ofs_last;
9cf6d335 484
53dda6ff 485 if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA)
9cf6d335 486 continue;
53dda6ff 487 hashcpy(base.sha1, obj->sha1);
b89c4e93 488 ref = !find_delta_children(&base, &ref_first, &ref_last);
53dda6ff
NP
489 memset(&base, 0, sizeof(base));
490 base.offset = obj->offset;
b89c4e93 491 ofs = !find_delta_children(&base, &ofs_first, &ofs_last);
53dda6ff 492 if (!ref && !ofs)
9cf6d335 493 continue;
2d477051 494 data = get_data_from_pack(obj);
53dda6ff 495 if (ref)
636171cb
NP
496 for (j = ref_first; j <= ref_last; j++) {
497 struct object_entry *child = objects + deltas[j].obj_no;
498 if (child->real_type == OBJ_REF_DELTA)
499 resolve_delta(child, data,
2d477051 500 obj->size, obj->type);
636171cb 501 }
53dda6ff 502 if (ofs)
636171cb
NP
503 for (j = ofs_first; j <= ofs_last; j++) {
504 struct object_entry *child = objects + deltas[j].obj_no;
505 if (child->real_type == OBJ_OFS_DELTA)
506 resolve_delta(child, data,
2d477051 507 obj->size, obj->type);
636171cb 508 }
9cf6d335 509 free(data);
3c9af366
NP
510 if (verbose)
511 percent = display_progress(nr_resolved_deltas,
512 nr_deltas, percent);
9cf6d335 513 }
3c9af366
NP
514 if (verbose && nr_resolved_deltas == nr_deltas)
515 fputc('\n', stderr);
636171cb
NP
516}
517
518static int write_compressed(int fd, void *in, unsigned int size)
519{
520 z_stream stream;
521 unsigned long maxsize;
522 void *out;
523
524 memset(&stream, 0, sizeof(stream));
525 deflateInit(&stream, zlib_compression_level);
526 maxsize = deflateBound(&stream, size);
527 out = xmalloc(maxsize);
528
529 /* Compress it */
530 stream.next_in = in;
531 stream.avail_in = size;
532 stream.next_out = out;
533 stream.avail_out = maxsize;
534 while (deflate(&stream, Z_FINISH) == Z_OK);
535 deflateEnd(&stream);
536
537 size = stream.total_out;
538 write_or_die(fd, out, size);
539 free(out);
540 return size;
541}
542
9096c660 543static void append_obj_to_pack(const unsigned char *sha1, void *buf,
636171cb
NP
544 unsigned long size, enum object_type type)
545{
546 struct object_entry *obj = &objects[nr_objects++];
547 unsigned char header[10];
548 unsigned long s = size;
549 int n = 0;
550 unsigned char c = (type << 4) | (s & 15);
551 s >>= 4;
552 while (s) {
553 header[n++] = c | 0x80;
554 c = s & 0x7f;
555 s >>= 7;
556 }
557 header[n++] = c;
558 write_or_die(output_fd, header, n);
559 obj[1].offset = obj[0].offset + n;
560 obj[1].offset += write_compressed(output_fd, buf, size);
9096c660 561 hashcpy(obj->sha1, sha1);
636171cb
NP
562}
563
564static int delta_pos_compare(const void *_a, const void *_b)
565{
566 struct delta_entry *a = *(struct delta_entry **)_a;
567 struct delta_entry *b = *(struct delta_entry **)_b;
568 return a->obj_no - b->obj_no;
569}
9cf6d335 570
636171cb
NP
571static void fix_unresolved_deltas(int nr_unresolved)
572{
573 struct delta_entry **sorted_by_pos;
3c9af366 574 int i, n = 0, percent = -1;
636171cb
NP
575
576 /*
577 * Since many unresolved deltas may well be themselves base objects
578 * for more unresolved deltas, we really want to include the
579 * smallest number of base objects that would cover as much delta
580 * as possible by picking the
581 * trunc deltas first, allowing for other deltas to resolve without
582 * additional base objects. Since most base objects are to be found
583 * before deltas depending on them, a good heuristic is to start
584 * resolving deltas in the same order as their position in the pack.
585 */
586 sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
9cf6d335 587 for (i = 0; i < nr_deltas; i++) {
636171cb
NP
588 if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
589 continue;
590 sorted_by_pos[n++] = &deltas[i];
9cf6d335 591 }
636171cb
NP
592 qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
593
594 for (i = 0; i < n; i++) {
595 struct delta_entry *d = sorted_by_pos[i];
596 void *data;
597 unsigned long size;
21666f1a 598 enum object_type type;
636171cb
NP
599 int j, first, last;
600
601 if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
602 continue;
21666f1a 603 data = read_sha1_file(d->base.sha1, &type, &size);
636171cb
NP
604 if (!data)
605 continue;
636171cb 606
b89c4e93 607 find_delta_children(&d->base, &first, &last);
636171cb
NP
608 for (j = first; j <= last; j++) {
609 struct object_entry *child = objects + deltas[j].obj_no;
610 if (child->real_type == OBJ_REF_DELTA)
21666f1a 611 resolve_delta(child, data, size, type);
636171cb
NP
612 }
613
9096c660
NP
614 if (check_sha1_signature(d->base.sha1, data, size, typename(type)))
615 die("local object %s is corrupt", sha1_to_hex(d->base.sha1));
616 append_obj_to_pack(d->base.sha1, data, size, type);
636171cb 617 free(data);
3c9af366
NP
618 if (verbose)
619 percent = display_progress(nr_resolved_deltas,
620 nr_deltas, percent);
636171cb
NP
621 }
622 free(sorted_by_pos);
3c9af366
NP
623 if (verbose)
624 fputc('\n', stderr);
636171cb
NP
625}
626
627static void readjust_pack_header_and_sha1(unsigned char *sha1)
628{
629 struct pack_header hdr;
630 SHA_CTX ctx;
631 int size;
632
633 /* Rewrite pack header with updated object number */
634 if (lseek(output_fd, 0, SEEK_SET) != 0)
635 die("cannot seek back: %s", strerror(errno));
93d26e4c 636 if (read_in_full(output_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
636171cb
NP
637 die("cannot read pack header back: %s", strerror(errno));
638 hdr.hdr_entries = htonl(nr_objects);
639 if (lseek(output_fd, 0, SEEK_SET) != 0)
640 die("cannot seek back: %s", strerror(errno));
641 write_or_die(output_fd, &hdr, sizeof(hdr));
642 if (lseek(output_fd, 0, SEEK_SET) != 0)
643 die("cannot seek back: %s", strerror(errno));
644
645 /* Recompute and store the new pack's SHA1 */
646 SHA1_Init(&ctx);
647 do {
648 unsigned char *buf[4096];
649 size = xread(output_fd, buf, sizeof(buf));
650 if (size < 0)
651 die("cannot read pack data back: %s", strerror(errno));
652 SHA1_Update(&ctx, buf, size);
653 } while (size > 0);
654 SHA1_Final(sha1, &ctx);
655 write_or_die(output_fd, sha1, 20);
9cf6d335
SV
656}
657
658static int sha1_compare(const void *_a, const void *_b)
659{
660 struct object_entry *a = *(struct object_entry **)_a;
661 struct object_entry *b = *(struct object_entry **)_b;
a89fccd2 662 return hashcmp(a->sha1, b->sha1);
9cf6d335
SV
663}
664
2d477051
NP
665/*
666 * On entry *sha1 contains the pack content SHA1 hash, on exit it is
667 * the SHA1 hash of sorted object names.
668 */
b89c4e93 669static const char *write_index_file(const char *index_name, unsigned char *sha1)
9cf6d335
SV
670{
671 struct sha1file *f;
7e4a2a84 672 struct object_entry **sorted_by_sha, **list, **last;
9cf6d335 673 unsigned int array[256];
e42797f5 674 int i, fd;
84c8d8ae 675 SHA_CTX ctx;
9cf6d335 676
7e4a2a84
JH
677 if (nr_objects) {
678 sorted_by_sha =
679 xcalloc(nr_objects, sizeof(struct object_entry *));
680 list = sorted_by_sha;
681 last = sorted_by_sha + nr_objects;
682 for (i = 0; i < nr_objects; ++i)
683 sorted_by_sha[i] = &objects[i];
684 qsort(sorted_by_sha, nr_objects, sizeof(sorted_by_sha[0]),
685 sha1_compare);
686
687 }
688 else
689 sorted_by_sha = list = last = NULL;
9cf6d335 690
e42797f5
NP
691 if (!index_name) {
692 static char tmpfile[PATH_MAX];
693 snprintf(tmpfile, sizeof(tmpfile),
0e55181f 694 "%s/tmp_idx_XXXXXX", get_object_directory());
e42797f5
NP
695 fd = mkstemp(tmpfile);
696 index_name = xstrdup(tmpfile);
697 } else {
698 unlink(index_name);
699 fd = open(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);
700 }
701 if (fd < 0)
702 die("unable to create %s: %s", index_name, strerror(errno));
703 f = sha1fd(fd, index_name);
9cf6d335
SV
704
705 /*
706 * Write the first-level table (the list is sorted,
707 * but we use a 256-entry lookup to be able to avoid
708 * having to do eight extra binary search iterations).
709 */
710 for (i = 0; i < 256; i++) {
711 struct object_entry **next = list;
712 while (next < last) {
713 struct object_entry *obj = *next;
714 if (obj->sha1[0] != i)
715 break;
716 next++;
717 }
718 array[i] = htonl(next - sorted_by_sha);
719 list = next;
720 }
721 sha1write(f, array, 256 * sizeof(int));
722
84c8d8ae
JH
723 /* recompute the SHA1 hash of sorted object names.
724 * currently pack-objects does not do this, but that
725 * can be fixed.
726 */
727 SHA1_Init(&ctx);
9cf6d335
SV
728 /*
729 * Write the actual SHA1 entries..
730 */
731 list = sorted_by_sha;
732 for (i = 0; i < nr_objects; i++) {
733 struct object_entry *obj = *list++;
734 unsigned int offset = htonl(obj->offset);
735 sha1write(f, &offset, 4);
736 sha1write(f, obj->sha1, 20);
84c8d8ae 737 SHA1_Update(&ctx, obj->sha1, 20);
9cf6d335 738 }
2d477051 739 sha1write(f, sha1, 20);
9cf6d335
SV
740 sha1close(f, NULL, 1);
741 free(sorted_by_sha);
84c8d8ae 742 SHA1_Final(sha1, &ctx);
e42797f5
NP
743 return index_name;
744}
745
746static void final(const char *final_pack_name, const char *curr_pack_name,
747 const char *final_index_name, const char *curr_index_name,
b8077709 748 const char *keep_name, const char *keep_msg,
e42797f5
NP
749 unsigned char *sha1)
750{
3a55602e 751 const char *report = "pack";
e42797f5
NP
752 char name[PATH_MAX];
753 int err;
754
755 if (!from_stdin) {
756 close(input_fd);
757 } else {
758 err = close(output_fd);
759 if (err)
760 die("error while closing pack file: %s", strerror(errno));
761 chmod(curr_pack_name, 0444);
762 }
763
b8077709
SP
764 if (keep_msg) {
765 int keep_fd, keep_msg_len = strlen(keep_msg);
766 if (!keep_name) {
767 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
768 get_object_directory(), sha1_to_hex(sha1));
769 keep_name = name;
770 }
9ca4a201
NP
771 keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
772 if (keep_fd < 0) {
773 if (errno != EEXIST)
774 die("cannot write keep file");
775 } else {
776 if (keep_msg_len > 0) {
777 write_or_die(keep_fd, keep_msg, keep_msg_len);
778 write_or_die(keep_fd, "\n", 1);
779 }
780 close(keep_fd);
576162a4 781 report = "keep";
b8077709 782 }
b8077709
SP
783 }
784
e42797f5
NP
785 if (final_pack_name != curr_pack_name) {
786 if (!final_pack_name) {
787 snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
788 get_object_directory(), sha1_to_hex(sha1));
789 final_pack_name = name;
790 }
791 if (move_temp_to_file(curr_pack_name, final_pack_name))
792 die("cannot store pack file");
793 }
794
795 chmod(curr_index_name, 0444);
796 if (final_index_name != curr_index_name) {
797 if (!final_index_name) {
798 snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
799 get_object_directory(), sha1_to_hex(sha1));
800 final_index_name = name;
801 }
802 if (move_temp_to_file(curr_index_name, final_index_name))
803 die("cannot store index file");
804 }
576162a4
NP
805
806 if (!from_stdin) {
807 printf("%s\n", sha1_to_hex(sha1));
808 } else {
809 char buf[48];
810 int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
811 report, sha1_to_hex(sha1));
d1b2ddc8 812 write_or_die(1, buf, len);
576162a4
NP
813
814 /*
815 * Let's just mimic git-unpack-objects here and write
816 * the last part of the input buffer to stdout.
817 */
818 while (input_len) {
819 err = xwrite(1, input_buffer + input_offset, input_len);
820 if (err <= 0)
821 break;
822 input_len -= err;
823 input_offset += err;
824 }
825 }
9cf6d335
SV
826}
827
828int main(int argc, char **argv)
829{
636171cb 830 int i, fix_thin_pack = 0;
e42797f5
NP
831 const char *curr_pack, *pack_name = NULL;
832 const char *curr_index, *index_name = NULL;
b8077709
SP
833 const char *keep_name = NULL, *keep_msg = NULL;
834 char *index_name_buf = NULL, *keep_name_buf = NULL;
84c8d8ae 835 unsigned char sha1[20];
9cf6d335
SV
836
837 for (i = 1; i < argc; i++) {
838 const char *arg = argv[i];
839
840 if (*arg == '-') {
e42797f5
NP
841 if (!strcmp(arg, "--stdin")) {
842 from_stdin = 1;
636171cb
NP
843 } else if (!strcmp(arg, "--fix-thin")) {
844 fix_thin_pack = 1;
b8077709
SP
845 } else if (!strcmp(arg, "--keep")) {
846 keep_msg = "";
cc44c765 847 } else if (!prefixcmp(arg, "--keep=")) {
b8077709 848 keep_msg = arg + 7;
cc44c765 849 } else if (!prefixcmp(arg, "--pack_header=")) {
bed006fb
NP
850 struct pack_header *hdr;
851 char *c;
852
853 hdr = (struct pack_header *)input_buffer;
854 hdr->hdr_signature = htonl(PACK_SIGNATURE);
855 hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
856 if (*c != ',')
857 die("bad %s", arg);
858 hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
859 if (*c)
860 die("bad %s", arg);
861 input_len = sizeof(*hdr);
3c9af366
NP
862 } else if (!strcmp(arg, "-v")) {
863 verbose = 1;
e42797f5 864 } else if (!strcmp(arg, "-o")) {
9cf6d335
SV
865 if (index_name || (i+1) >= argc)
866 usage(index_pack_usage);
867 index_name = argv[++i];
868 } else
869 usage(index_pack_usage);
870 continue;
871 }
872
873 if (pack_name)
874 usage(index_pack_usage);
875 pack_name = arg;
876 }
877
e42797f5 878 if (!pack_name && !from_stdin)
9cf6d335 879 usage(index_pack_usage);
636171cb
NP
880 if (fix_thin_pack && !from_stdin)
881 die("--fix-thin cannot be used without --stdin");
e42797f5 882 if (!index_name && pack_name) {
9cf6d335 883 int len = strlen(pack_name);
5bb1cda5 884 if (!has_extension(pack_name, ".pack"))
9cf6d335
SV
885 die("packfile name '%s' does not end with '.pack'",
886 pack_name);
6689f087 887 index_name_buf = xmalloc(len);
9cf6d335
SV
888 memcpy(index_name_buf, pack_name, len - 5);
889 strcpy(index_name_buf + len - 5, ".idx");
890 index_name = index_name_buf;
891 }
b8077709
SP
892 if (keep_msg && !keep_name && pack_name) {
893 int len = strlen(pack_name);
894 if (!has_extension(pack_name, ".pack"))
895 die("packfile name '%s' does not end with '.pack'",
896 pack_name);
897 keep_name_buf = xmalloc(len);
898 memcpy(keep_name_buf, pack_name, len - 5);
899 strcpy(keep_name_buf + len - 5, ".keep");
900 keep_name = keep_name_buf;
901 }
9cf6d335 902
e42797f5 903 curr_pack = open_pack_file(pack_name);
9cf6d335 904 parse_pack_header();
636171cb
NP
905 objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry));
906 deltas = xmalloc(nr_objects * sizeof(struct delta_entry));
3c9af366
NP
907 if (verbose)
908 setup_progress_signal();
2d477051 909 parse_pack_objects(sha1);
636171cb
NP
910 if (nr_deltas != nr_resolved_deltas) {
911 if (fix_thin_pack) {
912 int nr_unresolved = nr_deltas - nr_resolved_deltas;
3c9af366 913 int nr_objects_initial = nr_objects;
636171cb
NP
914 if (nr_unresolved <= 0)
915 die("confusion beyond insanity");
916 objects = xrealloc(objects,
917 (nr_objects + nr_unresolved + 1)
918 * sizeof(*objects));
919 fix_unresolved_deltas(nr_unresolved);
3c9af366
NP
920 if (verbose)
921 fprintf(stderr, "%d objects were added to complete this thin pack.\n",
922 nr_objects - nr_objects_initial);
636171cb
NP
923 readjust_pack_header_and_sha1(sha1);
924 }
925 if (nr_deltas != nr_resolved_deltas)
926 die("pack has %d unresolved deltas",
927 nr_deltas - nr_resolved_deltas);
928 } else {
929 /* Flush remaining pack final 20-byte SHA1. */
636171cb
NP
930 flush();
931 }
9cf6d335 932 free(deltas);
e42797f5 933 curr_index = write_index_file(index_name, sha1);
b8077709
SP
934 final(pack_name, curr_pack,
935 index_name, curr_index,
936 keep_name, keep_msg,
937 sha1);
9cf6d335
SV
938 free(objects);
939 free(index_name_buf);
b8077709 940 free(keep_name_buf);
9cf6d335
SV
941
942 return 0;
943}