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