Make git pack files use little-endian size encoding
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include <sys/types.h>
10 #include <dirent.h>
11 #include "cache.h"
12 #include "delta.h"
13 #include "pack.h"
14
15 #ifndef O_NOATIME
16 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
17 #define O_NOATIME 01000000
18 #else
19 #define O_NOATIME 0
20 #endif
21 #endif
22
23 static unsigned int sha1_file_open_flag = O_NOATIME;
24
25 static unsigned hexval(char c)
26 {
27 if (c >= '0' && c <= '9')
28 return c - '0';
29 if (c >= 'a' && c <= 'f')
30 return c - 'a' + 10;
31 if (c >= 'A' && c <= 'F')
32 return c - 'A' + 10;
33 return ~0;
34 }
35
36 int get_sha1_hex(const char *hex, unsigned char *sha1)
37 {
38 int i;
39 for (i = 0; i < 20; i++) {
40 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
41 if (val & ~0xff)
42 return -1;
43 *sha1++ = val;
44 hex += 2;
45 }
46 return 0;
47 }
48
49 static int get_sha1_file(const char *path, unsigned char *result)
50 {
51 char buffer[60];
52 int fd = open(path, O_RDONLY);
53 int len;
54
55 if (fd < 0)
56 return -1;
57 len = read(fd, buffer, sizeof(buffer));
58 close(fd);
59 if (len < 40)
60 return -1;
61 return get_sha1_hex(buffer, result);
62 }
63
64 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir;
65 static void setup_git_env(void)
66 {
67 git_dir = gitenv(GIT_DIR_ENVIRONMENT);
68 if (!git_dir)
69 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
70 git_object_dir = gitenv(DB_ENVIRONMENT);
71 if (!git_object_dir) {
72 git_object_dir = xmalloc(strlen(git_dir) + 9);
73 sprintf(git_object_dir, "%s/objects", git_dir);
74 }
75 git_refs_dir = xmalloc(strlen(git_dir) + 6);
76 sprintf(git_refs_dir, "%s/refs", git_dir);
77 git_index_file = gitenv(INDEX_ENVIRONMENT);
78 if (!git_index_file) {
79 git_index_file = xmalloc(strlen(git_dir) + 7);
80 sprintf(git_index_file, "%s/index", git_dir);
81 }
82 }
83
84 char *get_object_directory(void)
85 {
86 if (!git_object_dir)
87 setup_git_env();
88 return git_object_dir;
89 }
90
91 char *get_refs_directory(void)
92 {
93 if (!git_refs_dir)
94 setup_git_env();
95 return git_refs_dir;
96 }
97
98 char *get_index_file(void)
99 {
100 if (!git_index_file)
101 setup_git_env();
102 return git_index_file;
103 }
104
105 int get_sha1(const char *str, unsigned char *sha1)
106 {
107 static char pathname[PATH_MAX];
108 static const char *prefix[] = {
109 "",
110 "refs",
111 "refs/tags",
112 "refs/heads",
113 "refs/snap",
114 NULL
115 };
116 const char **p;
117
118 if (!get_sha1_hex(str, sha1))
119 return 0;
120
121 if (!git_dir)
122 setup_git_env();
123 for (p = prefix; *p; p++) {
124 snprintf(pathname, sizeof(pathname), "%s/%s/%s",
125 git_dir, *p, str);
126 if (!get_sha1_file(pathname, sha1))
127 return 0;
128 }
129
130 return -1;
131 }
132
133 char * sha1_to_hex(const unsigned char *sha1)
134 {
135 static char buffer[50];
136 static const char hex[] = "0123456789abcdef";
137 char *buf = buffer;
138 int i;
139
140 for (i = 0; i < 20; i++) {
141 unsigned int val = *sha1++;
142 *buf++ = hex[val >> 4];
143 *buf++ = hex[val & 0xf];
144 }
145 return buffer;
146 }
147
148 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
149 {
150 int i;
151 for (i = 0; i < 20; i++) {
152 static char hex[] = "0123456789abcdef";
153 unsigned int val = sha1[i];
154 char *pos = pathbuf + i*2 + (i > 0);
155 *pos++ = hex[val >> 4];
156 *pos = hex[val & 0xf];
157 }
158 }
159
160 /*
161 * NOTE! This returns a statically allocated buffer, so you have to be
162 * careful about using it. Do a "strdup()" if you need to save the
163 * filename.
164 *
165 * Also note that this returns the location for creating. Reading
166 * SHA1 file can happen from any alternate directory listed in the
167 * DB_ENVIRONMENT environment variable if it is not found in
168 * the primary object database.
169 */
170 char *sha1_file_name(const unsigned char *sha1)
171 {
172 static char *name, *base;
173
174 if (!base) {
175 const char *sha1_file_directory = get_object_directory();
176 int len = strlen(sha1_file_directory);
177 base = xmalloc(len + 60);
178 memcpy(base, sha1_file_directory, len);
179 memset(base+len, 0, 60);
180 base[len] = '/';
181 base[len+3] = '/';
182 name = base + len + 1;
183 }
184 fill_sha1_path(name, sha1);
185 return base;
186 }
187
188 struct alternate_object_database *alt_odb;
189
190 /*
191 * Prepare alternate object database registry.
192 * alt_odb points at an array of struct alternate_object_database.
193 * This array is terminated with an element that has both its base
194 * and name set to NULL. alt_odb[n] comes from n'th non-empty
195 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
196 * variable, and its base points at a statically allocated buffer
197 * that contains "/the/directory/corresponding/to/.git/objects/...",
198 * while its name points just after the slash at the end of
199 * ".git/objects/" in the example above, and has enough space to hold
200 * 40-byte hex SHA1, an extra slash for the first level indirection,
201 * and the terminating NUL.
202 * This function allocates the alt_odb array and all the strings
203 * pointed by base fields of the array elements with one xmalloc();
204 * the string pool immediately follows the array.
205 */
206 void prepare_alt_odb(void)
207 {
208 int pass, totlen, i;
209 const char *cp, *last;
210 char *op = NULL;
211 const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
212
213 if (alt_odb)
214 return;
215 /* The first pass counts how large an area to allocate to
216 * hold the entire alt_odb structure, including array of
217 * structs and path buffers for them. The second pass fills
218 * the structure and prepares the path buffers for use by
219 * fill_sha1_path().
220 */
221 for (totlen = pass = 0; pass < 2; pass++) {
222 last = alt;
223 i = 0;
224 do {
225 cp = strchr(last, ':') ? : last + strlen(last);
226 if (last != cp) {
227 /* 43 = 40-byte + 2 '/' + terminating NUL */
228 int pfxlen = cp - last;
229 int entlen = pfxlen + 43;
230 if (pass == 0)
231 totlen += entlen;
232 else {
233 alt_odb[i].base = op;
234 alt_odb[i].name = op + pfxlen + 1;
235 memcpy(op, last, pfxlen);
236 op[pfxlen] = op[pfxlen + 3] = '/';
237 op[entlen-1] = 0;
238 op += entlen;
239 }
240 i++;
241 }
242 while (*cp && *cp == ':')
243 cp++;
244 last = cp;
245 } while (*cp);
246 if (pass)
247 break;
248 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
249 alt_odb[i].base = alt_odb[i].name = NULL;
250 op = (char*)(&alt_odb[i+1]);
251 }
252 }
253
254 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
255 {
256 int i;
257 char *name = sha1_file_name(sha1);
258
259 if (!stat(name, st))
260 return name;
261 prepare_alt_odb();
262 for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
263 fill_sha1_path(name, sha1);
264 if (!stat(alt_odb[i].base, st))
265 return alt_odb[i].base;
266 }
267 return NULL;
268 }
269
270 #define PACK_MAX_SZ (1<<26)
271 static int pack_used_ctr;
272 static unsigned long pack_mapped;
273 struct packed_git *packed_git;
274
275 struct pack_entry {
276 unsigned int offset;
277 unsigned char sha1[20];
278 struct packed_git *p;
279 };
280
281 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
282 void **idx_map_)
283 {
284 void *idx_map;
285 unsigned int *index;
286 unsigned long idx_size;
287 int nr, i;
288 int fd = open(path, O_RDONLY);
289 struct stat st;
290 if (fd < 0)
291 return -1;
292 if (fstat(fd, &st)) {
293 close(fd);
294 return -1;
295 }
296 idx_size = st.st_size;
297 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
298 close(fd);
299 if (idx_map == MAP_FAILED)
300 return -1;
301
302 index = idx_map;
303
304 /* check index map */
305 if (idx_size < 4*256 + 20)
306 return error("index file too small");
307 nr = 0;
308 for (i = 0; i < 256; i++) {
309 unsigned int n = ntohl(index[i]);
310 if (n < nr)
311 return error("non-monotonic index");
312 nr = n;
313 }
314
315 /*
316 * Total size:
317 * - 256 index entries 4 bytes each
318 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
319 * - 20-byte SHA1 of the packfile
320 * - 20-byte SHA1 file checksum
321 */
322 if (idx_size != 4*256 + nr * 24 + 20 + 20)
323 return error("wrong index file size");
324
325 *idx_map_ = idx_map;
326 *idx_size_ = idx_size;
327 return 0;
328 }
329
330 static void unuse_one_packed_git(void)
331 {
332 /* NOTYET */
333 }
334
335 static int use_packed_git(struct packed_git *p)
336 {
337 if (!p->pack_base) {
338 int fd;
339 struct stat st;
340 void *map;
341
342 pack_mapped += p->pack_size;
343 while (PACK_MAX_SZ < pack_mapped)
344 unuse_one_packed_git();
345 fd = open(p->pack_name, O_RDONLY);
346 if (fd < 0)
347 return -1;
348 if (fstat(fd, &st)) {
349 close(fd);
350 return -1;
351 }
352 if (st.st_size != p->pack_size)
353 return -1;
354 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
355 close(fd);
356 if (map == MAP_FAILED)
357 return -1;
358 p->pack_base = map;
359 }
360 p->pack_last_used = pack_used_ctr++;
361 return 0;
362 }
363
364 static struct packed_git *add_packed_git(char *path, int path_len)
365 {
366 struct stat st;
367 struct packed_git *p;
368 unsigned long idx_size;
369 void *idx_map;
370
371 if (check_packed_git_idx(path, &idx_size, &idx_map))
372 return NULL;
373
374 /* do we have a corresponding .pack file? */
375 strcpy(path + path_len - 4, ".pack");
376 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
377 munmap(idx_map, idx_size);
378 return NULL;
379 }
380 /* ok, it looks sane as far as we can check without
381 * actually mapping the pack file.
382 */
383 p = xmalloc(sizeof(*p) + path_len + 2);
384 strcpy(p->pack_name, path);
385 p->index_size = idx_size;
386 p->pack_size = st.st_size;
387 p->index_base = idx_map;
388 p->next = NULL;
389 p->pack_base = NULL;
390 p->pack_last_used = 0;
391 return p;
392 }
393
394 static void prepare_packed_git_one(char *objdir)
395 {
396 char path[PATH_MAX];
397 int len;
398 DIR *dir;
399 struct dirent *de;
400
401 sprintf(path, "%s/pack", objdir);
402 len = strlen(path);
403 dir = opendir(path);
404 if (!dir)
405 return;
406 path[len++] = '/';
407 while ((de = readdir(dir)) != NULL) {
408 int namelen = strlen(de->d_name);
409 struct packed_git *p;
410
411 if (strcmp(de->d_name + namelen - 4, ".idx"))
412 continue;
413
414 /* we have .idx. Is it a file we can map? */
415 strcpy(path + len, de->d_name);
416 p = add_packed_git(path, len + namelen);
417 if (!p)
418 continue;
419 p->next = packed_git;
420 packed_git = p;
421 }
422 }
423
424 void prepare_packed_git(void)
425 {
426 int i;
427 static int run_once = 0;
428
429 if (run_once++)
430 return;
431
432 prepare_packed_git_one(get_object_directory());
433 prepare_alt_odb();
434 for (i = 0; alt_odb[i].base != NULL; i++) {
435 alt_odb[i].name[0] = 0;
436 prepare_packed_git_one(alt_odb[i].base);
437 }
438 }
439
440 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
441 {
442 char header[100];
443 unsigned char real_sha1[20];
444 SHA_CTX c;
445
446 SHA1_Init(&c);
447 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
448 SHA1_Update(&c, map, size);
449 SHA1_Final(real_sha1, &c);
450 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
451 }
452
453 static void *map_sha1_file_internal(const unsigned char *sha1,
454 unsigned long *size,
455 int say_error)
456 {
457 struct stat st;
458 void *map;
459 int fd;
460 char *filename = find_sha1_file(sha1, &st);
461
462 if (!filename) {
463 if (say_error)
464 error("cannot map sha1 file %s", sha1_to_hex(sha1));
465 return NULL;
466 }
467
468 fd = open(filename, O_RDONLY | sha1_file_open_flag);
469 if (fd < 0) {
470 /* See if it works without O_NOATIME */
471 switch (sha1_file_open_flag) {
472 default:
473 fd = open(filename, O_RDONLY);
474 if (fd >= 0)
475 break;
476 /* Fallthrough */
477 case 0:
478 if (say_error)
479 perror(filename);
480 return NULL;
481 }
482
483 /* If it failed once, it will probably fail again.
484 * Stop using O_NOATIME
485 */
486 sha1_file_open_flag = 0;
487 }
488 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
489 close(fd);
490 if (-1 == (int)(long)map)
491 return NULL;
492 *size = st.st_size;
493 return map;
494 }
495
496 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
497 {
498 return map_sha1_file_internal(sha1, size, 1);
499 }
500
501 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
502 {
503 /* Get the data stream */
504 memset(stream, 0, sizeof(*stream));
505 stream->next_in = map;
506 stream->avail_in = mapsize;
507 stream->next_out = buffer;
508 stream->avail_out = size;
509
510 inflateInit(stream);
511 return inflate(stream, 0);
512 }
513
514 void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
515 {
516 int bytes = strlen(buffer) + 1;
517 unsigned char *buf = xmalloc(1+size);
518
519 memcpy(buf, buffer + bytes, stream->total_out - bytes);
520 bytes = stream->total_out - bytes;
521 if (bytes < size) {
522 stream->next_out = buf + bytes;
523 stream->avail_out = size - bytes;
524 while (inflate(stream, Z_FINISH) == Z_OK)
525 /* nothing */;
526 }
527 buf[size] = 0;
528 inflateEnd(stream);
529 return buf;
530 }
531
532 /*
533 * We used to just use "sscanf()", but that's actually way
534 * too permissive for what we want to check. So do an anal
535 * object header parse by hand.
536 */
537 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
538 {
539 int i;
540 unsigned long size;
541
542 /*
543 * The type can be at most ten bytes (including the
544 * terminating '\0' that we add), and is followed by
545 * a space.
546 */
547 i = 10;
548 for (;;) {
549 char c = *hdr++;
550 if (c == ' ')
551 break;
552 if (!--i)
553 return -1;
554 *type++ = c;
555 }
556 *type = 0;
557
558 /*
559 * The length must follow immediately, and be in canonical
560 * decimal format (ie "010" is not valid).
561 */
562 size = *hdr++ - '0';
563 if (size > 9)
564 return -1;
565 if (size) {
566 for (;;) {
567 unsigned long c = *hdr - '0';
568 if (c > 9)
569 break;
570 hdr++;
571 size = size * 10 + c;
572 }
573 }
574 *sizep = size;
575
576 /*
577 * The length must be followed by a zero byte
578 */
579 return *hdr ? -1 : 0;
580 }
581
582 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
583 {
584 int ret;
585 z_stream stream;
586 char hdr[8192];
587
588 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
589 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
590 return NULL;
591
592 return unpack_sha1_rest(&stream, hdr, *size);
593 }
594
595 static int packed_delta_info(unsigned char *base_sha1,
596 unsigned long delta_size,
597 unsigned long left,
598 char *type,
599 unsigned long *sizep)
600 {
601 unsigned char *data;
602 unsigned char delta_head[64];
603 int i;
604 unsigned char cmd;
605 unsigned long data_size, result_size, base_size, verify_base_size;
606 z_stream stream;
607 int st;
608
609 if (left < 20)
610 die("truncated pack file");
611 if (sha1_object_info(base_sha1, type, &base_size))
612 die("cannot get info for delta-pack base");
613
614 data = base_sha1 + 20;
615 data_size = left - 20;
616
617 memset(&stream, 0, sizeof(stream));
618
619 stream.next_in = data;
620 stream.avail_in = data_size;
621 stream.next_out = delta_head;
622 stream.avail_out = sizeof(delta_head);
623
624 inflateInit(&stream);
625 st = inflate(&stream, Z_FINISH);
626 inflateEnd(&stream);
627 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
628 die("delta data unpack-initial failed");
629
630 /* Examine the initial part of the delta to figure out
631 * the result size. Verify the base size while we are at it.
632 */
633 data = delta_head;
634 verify_base_size = i = 0;
635 cmd = *data++;
636 while (cmd) {
637 if (cmd & 1)
638 verify_base_size |= *data++ << i;
639 i += 8;
640 cmd >>= 1;
641 }
642
643 /* Read the result size */
644 result_size = i = 0;
645 cmd = *data++;
646 while (cmd) {
647 if (cmd & 1)
648 result_size |= *data++ << i;
649 i += 8;
650 cmd >>= 1;
651 }
652 if (verify_base_size != base_size)
653 die("delta base size mismatch");
654
655 *sizep = result_size;
656 return 0;
657 }
658
659 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
660 enum object_type *type, unsigned long *sizep)
661 {
662 unsigned shift;
663 unsigned char *pack, c;
664 unsigned long size;
665
666 if (offset >= p->pack_size)
667 die("object offset outside of pack file");
668
669 pack = p->pack_base + offset;
670 c = *pack++;
671 offset++;
672 *type = (c >> 4) & 7;
673 size = c & 15;
674 shift = 4;
675 while (c & 0x80) {
676 if (offset >= p->pack_size)
677 die("object offset outside of pack file");
678 c = *pack++;
679 offset++;
680 size += (c & 0x7f) << shift;
681 shift += 7;
682 }
683 *sizep = size;
684 return offset;
685 }
686
687 static int packed_object_info(struct pack_entry *entry,
688 char *type, unsigned long *sizep)
689 {
690 struct packed_git *p = entry->p;
691 unsigned long offset, size, left;
692 unsigned char *pack;
693 enum object_type kind;
694
695 if (use_packed_git(p))
696 die("cannot map packed file");
697
698 offset = unpack_object_header(p, entry->offset, &kind, &size);
699 pack = p->pack_base + offset;
700 left = p->pack_size - offset;
701
702 switch (kind) {
703 case OBJ_DELTA:
704 return packed_delta_info(pack, size, left, type, sizep);
705 break;
706 case OBJ_COMMIT:
707 strcpy(type, "commit");
708 break;
709 case OBJ_TREE:
710 strcpy(type, "tree");
711 break;
712 case OBJ_BLOB:
713 strcpy(type, "blob");
714 break;
715 case OBJ_TAG:
716 strcpy(type, "tag");
717 break;
718 default:
719 die("corrupted pack file");
720 }
721 *sizep = size;
722 return 0;
723 }
724
725 /* forward declaration for a mutually recursive function */
726 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
727
728 static void *unpack_delta_entry(unsigned char *base_sha1,
729 unsigned long delta_size,
730 unsigned long left,
731 char *type,
732 unsigned long *sizep)
733 {
734 void *data, *delta_data, *result, *base;
735 unsigned long data_size, result_size, base_size;
736 z_stream stream;
737 int st;
738
739 if (left < 20)
740 die("truncated pack file");
741 data = base_sha1 + 20;
742 data_size = left - 20;
743 delta_data = xmalloc(delta_size);
744
745 memset(&stream, 0, sizeof(stream));
746
747 stream.next_in = data;
748 stream.avail_in = data_size;
749 stream.next_out = delta_data;
750 stream.avail_out = delta_size;
751
752 inflateInit(&stream);
753 st = inflate(&stream, Z_FINISH);
754 inflateEnd(&stream);
755 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
756 die("delta data unpack failed");
757
758 /* This may recursively unpack the base, which is what we want */
759 base = read_sha1_file(base_sha1, type, &base_size);
760 if (!base)
761 die("failed to read delta-pack base object %s",
762 sha1_to_hex(base_sha1));
763 result = patch_delta(base, base_size,
764 delta_data, delta_size,
765 &result_size);
766 if (!result)
767 die("failed to apply delta");
768 free(delta_data);
769 free(base);
770 *sizep = result_size;
771 return result;
772 }
773
774 static void *unpack_non_delta_entry(unsigned char *data,
775 unsigned long size,
776 unsigned long left)
777 {
778 int st;
779 z_stream stream;
780 char *buffer;
781
782 buffer = xmalloc(size + 1);
783 buffer[size] = 0;
784 memset(&stream, 0, sizeof(stream));
785 stream.next_in = data;
786 stream.avail_in = left;
787 stream.next_out = buffer;
788 stream.avail_out = size;
789
790 inflateInit(&stream);
791 st = inflate(&stream, Z_FINISH);
792 inflateEnd(&stream);
793 if ((st != Z_STREAM_END) || stream.total_out != size) {
794 free(buffer);
795 return NULL;
796 }
797
798 return buffer;
799 }
800
801 static void *unpack_entry(struct pack_entry *entry,
802 char *type, unsigned long *sizep)
803 {
804 struct packed_git *p = entry->p;
805 unsigned long offset, size, left;
806 unsigned char *pack;
807 enum object_type kind;
808
809 if (use_packed_git(p))
810 die("cannot map packed file");
811
812 offset = unpack_object_header(p, entry->offset, &kind, &size);
813 pack = p->pack_base + offset;
814 left = p->pack_size - offset;
815 switch (kind) {
816 case OBJ_DELTA:
817 return unpack_delta_entry(pack, size, left, type, sizep);
818 case OBJ_COMMIT:
819 strcpy(type, "commit");
820 break;
821 case OBJ_TREE:
822 strcpy(type, "tree");
823 break;
824 case OBJ_BLOB:
825 strcpy(type, "blob");
826 break;
827 case OBJ_TAG:
828 strcpy(type, "tag");
829 break;
830 default:
831 die("corrupted pack file");
832 }
833 *sizep = size;
834 return unpack_non_delta_entry(pack, size, left);
835 }
836
837 int num_packed_objects(const struct packed_git *p)
838 {
839 /* See check_packed_git_idx and pack-objects.c */
840 return (p->index_size - 20 - 20 - 4*256) / 24;
841 }
842
843 int nth_packed_object_sha1(const struct packed_git *p, int n,
844 unsigned char* sha1)
845 {
846 void *index = p->index_base + 256;
847 if (n < 0 || num_packed_objects(p) <= n)
848 return -1;
849 memcpy(sha1, (index + 24 * n + 4), 20);
850 return 0;
851 }
852
853 static int find_pack_entry_1(const unsigned char *sha1,
854 struct pack_entry *e, struct packed_git *p)
855 {
856 int *level1_ofs = p->index_base;
857 int hi = ntohl(level1_ofs[*sha1]);
858 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
859 void *index = p->index_base + 256;
860
861 do {
862 int mi = (lo + hi) / 2;
863 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
864 if (!cmp) {
865 e->offset = ntohl(*((int*)(index + 24 * mi)));
866 memcpy(e->sha1, sha1, 20);
867 e->p = p;
868 return 1;
869 }
870 if (cmp > 0)
871 hi = mi;
872 else
873 lo = mi+1;
874 } while (lo < hi);
875 return 0;
876 }
877
878 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
879 {
880 struct packed_git *p;
881 prepare_packed_git();
882
883 for (p = packed_git; p; p = p->next) {
884 if (find_pack_entry_1(sha1, e, p))
885 return 1;
886 }
887 return 0;
888 }
889
890 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
891 {
892 int status;
893 unsigned long mapsize, size;
894 void *map;
895 z_stream stream;
896 char hdr[128];
897
898 map = map_sha1_file_internal(sha1, &mapsize, 0);
899 if (!map) {
900 struct pack_entry e;
901
902 if (!find_pack_entry(sha1, &e))
903 return error("unable to find %s", sha1_to_hex(sha1));
904 if (!packed_object_info(&e, type, sizep))
905 return 0;
906 /* sheesh */
907 map = unpack_entry(&e, type, sizep);
908 free(map);
909 return (map == NULL) ? 0 : -1;
910 }
911 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
912 status = error("unable to unpack %s header",
913 sha1_to_hex(sha1));
914 if (parse_sha1_header(hdr, type, &size) < 0)
915 status = error("unable to parse %s header", sha1_to_hex(sha1));
916 else {
917 status = 0;
918 *sizep = size;
919 }
920 inflateEnd(&stream);
921 munmap(map, mapsize);
922 return status;
923 }
924
925 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
926 {
927 struct pack_entry e;
928
929 if (!find_pack_entry(sha1, &e)) {
930 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
931 return NULL;
932 }
933 return unpack_entry(&e, type, size);
934 }
935
936 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
937 {
938 unsigned long mapsize;
939 void *map, *buf;
940
941 map = map_sha1_file_internal(sha1, &mapsize, 0);
942 if (map) {
943 buf = unpack_sha1_file(map, mapsize, type, size);
944 munmap(map, mapsize);
945 return buf;
946 }
947 return read_packed_sha1(sha1, type, size);
948 }
949
950 void *read_object_with_reference(const unsigned char *sha1,
951 const char *required_type,
952 unsigned long *size,
953 unsigned char *actual_sha1_return)
954 {
955 char type[20];
956 void *buffer;
957 unsigned long isize;
958 unsigned char actual_sha1[20];
959
960 memcpy(actual_sha1, sha1, 20);
961 while (1) {
962 int ref_length = -1;
963 const char *ref_type = NULL;
964
965 buffer = read_sha1_file(actual_sha1, type, &isize);
966 if (!buffer)
967 return NULL;
968 if (!strcmp(type, required_type)) {
969 *size = isize;
970 if (actual_sha1_return)
971 memcpy(actual_sha1_return, actual_sha1, 20);
972 return buffer;
973 }
974 /* Handle references */
975 else if (!strcmp(type, "commit"))
976 ref_type = "tree ";
977 else if (!strcmp(type, "tag"))
978 ref_type = "object ";
979 else {
980 free(buffer);
981 return NULL;
982 }
983 ref_length = strlen(ref_type);
984
985 if (memcmp(buffer, ref_type, ref_length) ||
986 get_sha1_hex(buffer + ref_length, actual_sha1)) {
987 free(buffer);
988 return NULL;
989 }
990 /* Now we have the ID of the referred-to object in
991 * actual_sha1. Check again. */
992 }
993 }
994
995 static char *write_sha1_file_prepare(void *buf,
996 unsigned long len,
997 const char *type,
998 unsigned char *sha1,
999 unsigned char *hdr,
1000 int *hdrlen)
1001 {
1002 SHA_CTX c;
1003
1004 /* Generate the header */
1005 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1006
1007 /* Sha1.. */
1008 SHA1_Init(&c);
1009 SHA1_Update(&c, hdr, *hdrlen);
1010 SHA1_Update(&c, buf, len);
1011 SHA1_Final(sha1, &c);
1012
1013 return sha1_file_name(sha1);
1014 }
1015
1016 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1017 {
1018 int size;
1019 unsigned char *compressed;
1020 z_stream stream;
1021 unsigned char sha1[20];
1022 char *filename;
1023 static char tmpfile[PATH_MAX];
1024 unsigned char hdr[50];
1025 int fd, hdrlen, ret;
1026
1027 /* Normally if we have it in the pack then we do not bother writing
1028 * it out into .git/objects/??/?{38} file.
1029 */
1030 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1031 if (returnsha1)
1032 memcpy(returnsha1, sha1, 20);
1033 if (has_sha1_file(sha1))
1034 return 0;
1035 fd = open(filename, O_RDONLY);
1036 if (fd >= 0) {
1037 /*
1038 * FIXME!!! We might do collision checking here, but we'd
1039 * need to uncompress the old file and check it. Later.
1040 */
1041 close(fd);
1042 return 0;
1043 }
1044
1045 if (errno != ENOENT) {
1046 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1047 return -1;
1048 }
1049
1050 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1051
1052 fd = mkstemp(tmpfile);
1053 if (fd < 0) {
1054 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1055 return -1;
1056 }
1057
1058 /* Set it up */
1059 memset(&stream, 0, sizeof(stream));
1060 deflateInit(&stream, Z_BEST_COMPRESSION);
1061 size = deflateBound(&stream, len+hdrlen);
1062 compressed = xmalloc(size);
1063
1064 /* Compress it */
1065 stream.next_out = compressed;
1066 stream.avail_out = size;
1067
1068 /* First header.. */
1069 stream.next_in = hdr;
1070 stream.avail_in = hdrlen;
1071 while (deflate(&stream, 0) == Z_OK)
1072 /* nothing */;
1073
1074 /* Then the data itself.. */
1075 stream.next_in = buf;
1076 stream.avail_in = len;
1077 while (deflate(&stream, Z_FINISH) == Z_OK)
1078 /* nothing */;
1079 deflateEnd(&stream);
1080 size = stream.total_out;
1081
1082 if (write(fd, compressed, size) != size)
1083 die("unable to write file");
1084 fchmod(fd, 0444);
1085 close(fd);
1086 free(compressed);
1087
1088 ret = link(tmpfile, filename);
1089 if (ret < 0) {
1090 ret = errno;
1091
1092 /*
1093 * Coda hack - coda doesn't like cross-directory links,
1094 * so we fall back to a rename, which will mean that it
1095 * won't be able to check collisions, but that's not a
1096 * big deal.
1097 *
1098 * When this succeeds, we just return 0. We have nothing
1099 * left to unlink.
1100 */
1101 if (ret == EXDEV && !rename(tmpfile, filename))
1102 return 0;
1103 }
1104 unlink(tmpfile);
1105 if (ret) {
1106 if (ret != EEXIST) {
1107 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1108 return -1;
1109 }
1110 /* FIXME!!! Collision check here ? */
1111 }
1112
1113 return 0;
1114 }
1115
1116 int write_sha1_from_fd(const unsigned char *sha1, int fd)
1117 {
1118 char *filename = sha1_file_name(sha1);
1119
1120 int local;
1121 z_stream stream;
1122 unsigned char real_sha1[20];
1123 unsigned char buf[4096];
1124 unsigned char discard[4096];
1125 int ret;
1126 SHA_CTX c;
1127
1128 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1129
1130 if (local < 0)
1131 return error("Couldn't open %s\n", filename);
1132
1133 memset(&stream, 0, sizeof(stream));
1134
1135 inflateInit(&stream);
1136
1137 SHA1_Init(&c);
1138
1139 do {
1140 ssize_t size;
1141 size = read(fd, buf, 4096);
1142 if (size <= 0) {
1143 close(local);
1144 unlink(filename);
1145 if (!size)
1146 return error("Connection closed?");
1147 perror("Reading from connection");
1148 return -1;
1149 }
1150 write(local, buf, size);
1151 stream.avail_in = size;
1152 stream.next_in = buf;
1153 do {
1154 stream.next_out = discard;
1155 stream.avail_out = sizeof(discard);
1156 ret = inflate(&stream, Z_SYNC_FLUSH);
1157 SHA1_Update(&c, discard, sizeof(discard) -
1158 stream.avail_out);
1159 } while (stream.avail_in && ret == Z_OK);
1160
1161 } while (ret == Z_OK);
1162 inflateEnd(&stream);
1163
1164 close(local);
1165 SHA1_Final(real_sha1, &c);
1166 if (ret != Z_STREAM_END) {
1167 unlink(filename);
1168 return error("File %s corrupted", sha1_to_hex(sha1));
1169 }
1170 if (memcmp(sha1, real_sha1, 20)) {
1171 unlink(filename);
1172 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1173 }
1174
1175 return 0;
1176 }
1177
1178 int has_sha1_file(const unsigned char *sha1)
1179 {
1180 struct stat st;
1181 struct pack_entry e;
1182
1183 if (find_sha1_file(sha1, &st))
1184 return 1;
1185 return find_pack_entry(sha1, &e);
1186 }
1187
1188 int index_fd(unsigned char *sha1, int fd, struct stat *st)
1189 {
1190 unsigned long size = st->st_size;
1191 void *buf;
1192 int ret;
1193
1194 buf = "";
1195 if (size)
1196 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1197 close(fd);
1198 if ((int)(long)buf == -1)
1199 return -1;
1200
1201 ret = write_sha1_file(buf, size, "blob", sha1);
1202 if (size)
1203 munmap(buf, size);
1204 return ret;
1205 }