[PATCH] Add note about IANA confirmation
[git/jnareb-git.git] / sha1_file.c
blobf4c742eed4736e121448eb8d24697aa812a0d158
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
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"
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
23 static unsigned int sha1_file_open_flag = O_NOATIME;
25 static unsigned hexval(char c)
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;
36 int get_sha1_hex(const char *hex, unsigned char *sha1)
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;
46 return 0;
49 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir,
50 *git_graft_file;
51 static void setup_git_env(void)
53 git_dir = getenv(GIT_DIR_ENVIRONMENT);
54 if (!git_dir)
55 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
56 git_object_dir = getenv(DB_ENVIRONMENT);
57 if (!git_object_dir) {
58 git_object_dir = xmalloc(strlen(git_dir) + 9);
59 sprintf(git_object_dir, "%s/objects", git_dir);
61 git_refs_dir = xmalloc(strlen(git_dir) + 6);
62 sprintf(git_refs_dir, "%s/refs", git_dir);
63 git_index_file = getenv(INDEX_ENVIRONMENT);
64 if (!git_index_file) {
65 git_index_file = xmalloc(strlen(git_dir) + 7);
66 sprintf(git_index_file, "%s/index", git_dir);
68 git_graft_file = getenv(GRAFT_ENVIRONMENT);
69 if (!git_graft_file)
70 git_graft_file = strdup(git_path("info/grafts"));
73 char *get_object_directory(void)
75 if (!git_object_dir)
76 setup_git_env();
77 return git_object_dir;
80 char *get_refs_directory(void)
82 if (!git_refs_dir)
83 setup_git_env();
84 return git_refs_dir;
87 char *get_index_file(void)
89 if (!git_index_file)
90 setup_git_env();
91 return git_index_file;
94 char *get_graft_file(void)
96 if (!git_graft_file)
97 setup_git_env();
98 return git_graft_file;
101 int safe_create_leading_directories(char *path)
103 char *pos = path;
105 while (pos) {
106 pos = strchr(pos, '/');
107 if (!pos)
108 break;
109 *pos = 0;
110 if (mkdir(path, 0777) < 0)
111 if (errno != EEXIST) {
112 *pos = '/';
113 return -1;
115 *pos++ = '/';
117 return 0;
120 char * sha1_to_hex(const unsigned char *sha1)
122 static char buffer[50];
123 static const char hex[] = "0123456789abcdef";
124 char *buf = buffer;
125 int i;
127 for (i = 0; i < 20; i++) {
128 unsigned int val = *sha1++;
129 *buf++ = hex[val >> 4];
130 *buf++ = hex[val & 0xf];
132 return buffer;
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 int i;
138 for (i = 0; i < 20; i++) {
139 static char hex[] = "0123456789abcdef";
140 unsigned int val = sha1[i];
141 char *pos = pathbuf + i*2 + (i > 0);
142 *pos++ = hex[val >> 4];
143 *pos = hex[val & 0xf];
148 * NOTE! This returns a statically allocated buffer, so you have to be
149 * careful about using it. Do a "strdup()" if you need to save the
150 * filename.
152 * Also note that this returns the location for creating. Reading
153 * SHA1 file can happen from any alternate directory listed in the
154 * DB_ENVIRONMENT environment variable if it is not found in
155 * the primary object database.
157 char *sha1_file_name(const unsigned char *sha1)
159 static char *name, *base;
161 if (!base) {
162 const char *sha1_file_directory = get_object_directory();
163 int len = strlen(sha1_file_directory);
164 base = xmalloc(len + 60);
165 memcpy(base, sha1_file_directory, len);
166 memset(base+len, 0, 60);
167 base[len] = '/';
168 base[len+3] = '/';
169 name = base + len + 1;
171 fill_sha1_path(name, sha1);
172 return base;
175 char *sha1_pack_name(const unsigned char *sha1)
177 static const char hex[] = "0123456789abcdef";
178 static char *name, *base, *buf;
179 int i;
181 if (!base) {
182 const char *sha1_file_directory = get_object_directory();
183 int len = strlen(sha1_file_directory);
184 base = xmalloc(len + 60);
185 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186 name = base + len + 11;
189 buf = name;
191 for (i = 0; i < 20; i++) {
192 unsigned int val = *sha1++;
193 *buf++ = hex[val >> 4];
194 *buf++ = hex[val & 0xf];
197 return base;
200 char *sha1_pack_index_name(const unsigned char *sha1)
202 static const char hex[] = "0123456789abcdef";
203 static char *name, *base, *buf;
204 int i;
206 if (!base) {
207 const char *sha1_file_directory = get_object_directory();
208 int len = strlen(sha1_file_directory);
209 base = xmalloc(len + 60);
210 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211 name = base + len + 11;
214 buf = name;
216 for (i = 0; i < 20; i++) {
217 unsigned int val = *sha1++;
218 *buf++ = hex[val >> 4];
219 *buf++ = hex[val & 0xf];
222 return base;
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
229 * Prepare alternate object database registry.
231 * The variable alt_odb_list points at the list of struct
232 * alternate_object_database. The elements on this list come from
233 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
234 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
235 * whose contents is exactly in the same format as that environment
236 * variable. Its base points at a statically allocated buffer that
237 * contains "/the/directory/corresponding/to/.git/objects/...", while
238 * its name points just after the slash at the end of ".git/objects/"
239 * in the example above, and has enough space to hold 40-byte hex
240 * SHA1, an extra slash for the first level indirection, and the
241 * terminating NUL.
243 static void link_alt_odb_entries(const char *alt, const char *ep, int sep)
245 const char *cp, *last;
246 struct alternate_object_database *ent;
248 last = alt;
249 while (last < ep) {
250 cp = last;
251 if (cp < ep && *cp == '#') {
252 while (cp < ep && *cp != sep)
253 cp++;
254 last = cp + 1;
255 continue;
257 for ( ; cp < ep && *cp != sep; cp++)
259 if (last != cp) {
260 /* 43 = 40-byte + 2 '/' + terminating NUL */
261 int pfxlen = cp - last;
262 int entlen = pfxlen + 43;
264 ent = xmalloc(sizeof(*ent) + entlen);
265 *alt_odb_tail = ent;
266 alt_odb_tail = &(ent->next);
267 ent->next = NULL;
269 memcpy(ent->base, last, pfxlen);
270 ent->name = ent->base + pfxlen + 1;
271 ent->base[pfxlen] = ent->base[pfxlen + 3] = '/';
272 ent->base[entlen-1] = 0;
274 while (cp < ep && *cp == sep)
275 cp++;
276 last = cp;
280 void prepare_alt_odb(void)
282 char path[PATH_MAX];
283 char *map;
284 int fd;
285 struct stat st;
286 char *alt;
288 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
289 if (!alt) alt = "";
291 sprintf(path, "%s/info/alternates", get_object_directory());
292 if (alt_odb_tail)
293 return;
294 alt_odb_tail = &alt_odb_list;
295 link_alt_odb_entries(alt, alt + strlen(alt), ':');
297 fd = open(path, O_RDONLY);
298 if (fd < 0)
299 return;
300 if (fstat(fd, &st) || (st.st_size == 0)) {
301 close(fd);
302 return;
304 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
305 close(fd);
306 if (map == MAP_FAILED)
307 return;
309 link_alt_odb_entries(map, map + st.st_size, '\n');
310 munmap(map, st.st_size);
313 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
315 char *name = sha1_file_name(sha1);
316 struct alternate_object_database *alt;
318 if (!stat(name, st))
319 return name;
320 prepare_alt_odb();
321 for (alt = alt_odb_list; alt; alt = alt->next) {
322 name = alt->name;
323 fill_sha1_path(name, sha1);
324 if (!stat(alt->base, st))
325 return alt->base;
327 return NULL;
330 #define PACK_MAX_SZ (1<<26)
331 static int pack_used_ctr;
332 static unsigned long pack_mapped;
333 struct packed_git *packed_git;
335 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
336 void **idx_map_)
338 void *idx_map;
339 unsigned int *index;
340 unsigned long idx_size;
341 int nr, i;
342 int fd = open(path, O_RDONLY);
343 struct stat st;
344 if (fd < 0)
345 return -1;
346 if (fstat(fd, &st)) {
347 close(fd);
348 return -1;
350 idx_size = st.st_size;
351 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
352 close(fd);
353 if (idx_map == MAP_FAILED)
354 return -1;
356 index = idx_map;
357 *idx_map_ = idx_map;
358 *idx_size_ = idx_size;
360 /* check index map */
361 if (idx_size < 4*256 + 20 + 20)
362 return error("index file too small");
363 nr = 0;
364 for (i = 0; i < 256; i++) {
365 unsigned int n = ntohl(index[i]);
366 if (n < nr)
367 return error("non-monotonic index");
368 nr = n;
372 * Total size:
373 * - 256 index entries 4 bytes each
374 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
375 * - 20-byte SHA1 of the packfile
376 * - 20-byte SHA1 file checksum
378 if (idx_size != 4*256 + nr * 24 + 20 + 20)
379 return error("wrong index file size");
381 return 0;
384 static int unuse_one_packed_git(void)
386 struct packed_git *p, *lru = NULL;
388 for (p = packed_git; p; p = p->next) {
389 if (p->pack_use_cnt || !p->pack_base)
390 continue;
391 if (!lru || p->pack_last_used < lru->pack_last_used)
392 lru = p;
394 if (!lru)
395 return 0;
396 munmap(lru->pack_base, lru->pack_size);
397 lru->pack_base = NULL;
398 return 1;
401 void unuse_packed_git(struct packed_git *p)
403 p->pack_use_cnt--;
406 int use_packed_git(struct packed_git *p)
408 if (!p->pack_size) {
409 struct stat st;
410 // We created the struct before we had the pack
411 stat(p->pack_name, &st);
412 if (!S_ISREG(st.st_mode))
413 die("packfile %s not a regular file", p->pack_name);
414 p->pack_size = st.st_size;
416 if (!p->pack_base) {
417 int fd;
418 struct stat st;
419 void *map;
421 pack_mapped += p->pack_size;
422 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
423 ; /* nothing */
424 fd = open(p->pack_name, O_RDONLY);
425 if (fd < 0)
426 die("packfile %s cannot be opened", p->pack_name);
427 if (fstat(fd, &st)) {
428 close(fd);
429 die("packfile %s cannot be opened", p->pack_name);
431 if (st.st_size != p->pack_size)
432 die("packfile %s size mismatch.", p->pack_name);
433 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
434 close(fd);
435 if (map == MAP_FAILED)
436 die("packfile %s cannot be mapped.", p->pack_name);
437 p->pack_base = map;
439 /* Check if the pack file matches with the index file.
440 * this is cheap.
442 if (memcmp((char*)(p->index_base) + p->index_size - 40,
443 p->pack_base + p->pack_size - 20, 20)) {
445 die("packfile %s does not match index.", p->pack_name);
448 p->pack_last_used = pack_used_ctr++;
449 p->pack_use_cnt++;
450 return 0;
453 struct packed_git *add_packed_git(char *path, int path_len)
455 struct stat st;
456 struct packed_git *p;
457 unsigned long idx_size;
458 void *idx_map;
460 if (check_packed_git_idx(path, &idx_size, &idx_map))
461 return NULL;
463 /* do we have a corresponding .pack file? */
464 strcpy(path + path_len - 4, ".pack");
465 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
466 munmap(idx_map, idx_size);
467 return NULL;
469 /* ok, it looks sane as far as we can check without
470 * actually mapping the pack file.
472 p = xmalloc(sizeof(*p) + path_len + 2);
473 strcpy(p->pack_name, path);
474 p->index_size = idx_size;
475 p->pack_size = st.st_size;
476 p->index_base = idx_map;
477 p->next = NULL;
478 p->pack_base = NULL;
479 p->pack_last_used = 0;
480 p->pack_use_cnt = 0;
481 return p;
484 struct packed_git *parse_pack_index(unsigned char *sha1)
486 char *path = sha1_pack_index_name(sha1);
487 return parse_pack_index_file(sha1, path);
490 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
492 struct packed_git *p;
493 unsigned long idx_size;
494 void *idx_map;
495 char *path;
497 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
498 return NULL;
500 path = sha1_pack_name(sha1);
502 p = xmalloc(sizeof(*p) + strlen(path) + 2);
503 strcpy(p->pack_name, path);
504 p->index_size = idx_size;
505 p->pack_size = 0;
506 p->index_base = idx_map;
507 p->next = NULL;
508 p->pack_base = NULL;
509 p->pack_last_used = 0;
510 p->pack_use_cnt = 0;
511 memcpy(p->sha1, sha1, 20);
512 return p;
515 void install_packed_git(struct packed_git *pack)
517 pack->next = packed_git;
518 packed_git = pack;
521 static void prepare_packed_git_one(char *objdir)
523 char path[PATH_MAX];
524 int len;
525 DIR *dir;
526 struct dirent *de;
528 sprintf(path, "%s/pack", objdir);
529 len = strlen(path);
530 dir = opendir(path);
531 if (!dir)
532 return;
533 path[len++] = '/';
534 while ((de = readdir(dir)) != NULL) {
535 int namelen = strlen(de->d_name);
536 struct packed_git *p;
538 if (strcmp(de->d_name + namelen - 4, ".idx"))
539 continue;
541 /* we have .idx. Is it a file we can map? */
542 strcpy(path + len, de->d_name);
543 p = add_packed_git(path, len + namelen);
544 if (!p)
545 continue;
546 p->next = packed_git;
547 packed_git = p;
549 closedir(dir);
552 void prepare_packed_git(void)
554 static int run_once = 0;
555 struct alternate_object_database *alt;
557 if (run_once)
558 return;
559 prepare_packed_git_one(get_object_directory());
560 prepare_alt_odb();
561 for (alt = alt_odb_list; alt; alt = alt->next) {
562 alt->name[0] = 0;
563 prepare_packed_git_one(alt->base);
565 run_once = 1;
568 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
570 char header[100];
571 unsigned char real_sha1[20];
572 SHA_CTX c;
574 SHA1_Init(&c);
575 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
576 SHA1_Update(&c, map, size);
577 SHA1_Final(real_sha1, &c);
578 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
581 static void *map_sha1_file_internal(const unsigned char *sha1,
582 unsigned long *size)
584 struct stat st;
585 void *map;
586 int fd;
587 char *filename = find_sha1_file(sha1, &st);
589 if (!filename) {
590 return NULL;
593 fd = open(filename, O_RDONLY | sha1_file_open_flag);
594 if (fd < 0) {
595 /* See if it works without O_NOATIME */
596 switch (sha1_file_open_flag) {
597 default:
598 fd = open(filename, O_RDONLY);
599 if (fd >= 0)
600 break;
601 /* Fallthrough */
602 case 0:
603 return NULL;
606 /* If it failed once, it will probably fail again.
607 * Stop using O_NOATIME
609 sha1_file_open_flag = 0;
611 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
612 close(fd);
613 if (map == MAP_FAILED)
614 return NULL;
615 *size = st.st_size;
616 return map;
619 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
621 /* Get the data stream */
622 memset(stream, 0, sizeof(*stream));
623 stream->next_in = map;
624 stream->avail_in = mapsize;
625 stream->next_out = buffer;
626 stream->avail_out = size;
628 inflateInit(stream);
629 return inflate(stream, 0);
632 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
634 int bytes = strlen(buffer) + 1;
635 unsigned char *buf = xmalloc(1+size);
637 memcpy(buf, buffer + bytes, stream->total_out - bytes);
638 bytes = stream->total_out - bytes;
639 if (bytes < size) {
640 stream->next_out = buf + bytes;
641 stream->avail_out = size - bytes;
642 while (inflate(stream, Z_FINISH) == Z_OK)
643 /* nothing */;
645 buf[size] = 0;
646 inflateEnd(stream);
647 return buf;
651 * We used to just use "sscanf()", but that's actually way
652 * too permissive for what we want to check. So do an anal
653 * object header parse by hand.
655 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
657 int i;
658 unsigned long size;
661 * The type can be at most ten bytes (including the
662 * terminating '\0' that we add), and is followed by
663 * a space.
665 i = 10;
666 for (;;) {
667 char c = *hdr++;
668 if (c == ' ')
669 break;
670 if (!--i)
671 return -1;
672 *type++ = c;
674 *type = 0;
677 * The length must follow immediately, and be in canonical
678 * decimal format (ie "010" is not valid).
680 size = *hdr++ - '0';
681 if (size > 9)
682 return -1;
683 if (size) {
684 for (;;) {
685 unsigned long c = *hdr - '0';
686 if (c > 9)
687 break;
688 hdr++;
689 size = size * 10 + c;
692 *sizep = size;
695 * The length must be followed by a zero byte
697 return *hdr ? -1 : 0;
700 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
702 int ret;
703 z_stream stream;
704 char hdr[8192];
706 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
707 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
708 return NULL;
710 return unpack_sha1_rest(&stream, hdr, *size);
713 /* forward declaration for a mutually recursive function */
714 static int packed_object_info(struct pack_entry *entry,
715 char *type, unsigned long *sizep);
717 static int packed_delta_info(unsigned char *base_sha1,
718 unsigned long delta_size,
719 unsigned long left,
720 char *type,
721 unsigned long *sizep,
722 struct packed_git *p)
724 struct pack_entry base_ent;
726 if (left < 20)
727 die("truncated pack file");
729 /* The base entry _must_ be in the same pack */
730 if (!find_pack_entry_one(base_sha1, &base_ent, p))
731 die("failed to find delta-pack base object %s",
732 sha1_to_hex(base_sha1));
734 /* We choose to only get the type of the base object and
735 * ignore potentially corrupt pack file that expects the delta
736 * based on a base with a wrong size. This saves tons of
737 * inflate() calls.
740 if (packed_object_info(&base_ent, type, NULL))
741 die("cannot get info for delta-pack base");
743 if (sizep) {
744 const unsigned char *data;
745 unsigned char delta_head[64];
746 unsigned long result_size;
747 z_stream stream;
748 int st;
750 memset(&stream, 0, sizeof(stream));
752 data = stream.next_in = base_sha1 + 20;
753 stream.avail_in = left - 20;
754 stream.next_out = delta_head;
755 stream.avail_out = sizeof(delta_head);
757 inflateInit(&stream);
758 st = inflate(&stream, Z_FINISH);
759 inflateEnd(&stream);
760 if ((st != Z_STREAM_END) &&
761 stream.total_out != sizeof(delta_head))
762 die("delta data unpack-initial failed");
764 /* Examine the initial part of the delta to figure out
765 * the result size.
767 data = delta_head;
768 get_delta_hdr_size(&data); /* ignore base size */
770 /* Read the result size */
771 result_size = get_delta_hdr_size(&data);
772 *sizep = result_size;
774 return 0;
777 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
778 enum object_type *type, unsigned long *sizep)
780 unsigned shift;
781 unsigned char *pack, c;
782 unsigned long size;
784 if (offset >= p->pack_size)
785 die("object offset outside of pack file");
787 pack = p->pack_base + offset;
788 c = *pack++;
789 offset++;
790 *type = (c >> 4) & 7;
791 size = c & 15;
792 shift = 4;
793 while (c & 0x80) {
794 if (offset >= p->pack_size)
795 die("object offset outside of pack file");
796 c = *pack++;
797 offset++;
798 size += (c & 0x7f) << shift;
799 shift += 7;
801 *sizep = size;
802 return offset;
805 void packed_object_info_detail(struct pack_entry *e,
806 char *type,
807 unsigned long *size,
808 unsigned long *store_size,
809 int *delta_chain_length,
810 unsigned char *base_sha1)
812 struct packed_git *p = e->p;
813 unsigned long offset, left;
814 unsigned char *pack;
815 enum object_type kind;
817 offset = unpack_object_header(p, e->offset, &kind, size);
818 pack = p->pack_base + offset;
819 left = p->pack_size - offset;
820 if (kind != OBJ_DELTA)
821 *delta_chain_length = 0;
822 else {
823 int chain_length = 0;
824 memcpy(base_sha1, pack, 20);
825 do {
826 struct pack_entry base_ent;
827 unsigned long junk;
829 find_pack_entry_one(pack, &base_ent, p);
830 offset = unpack_object_header(p, base_ent.offset,
831 &kind, &junk);
832 pack = p->pack_base + offset;
833 chain_length++;
834 } while (kind == OBJ_DELTA);
835 *delta_chain_length = chain_length;
837 switch (kind) {
838 case OBJ_COMMIT:
839 strcpy(type, "commit");
840 break;
841 case OBJ_TREE:
842 strcpy(type, "tree");
843 break;
844 case OBJ_BLOB:
845 strcpy(type, "blob");
846 break;
847 case OBJ_TAG:
848 strcpy(type, "tag");
849 break;
850 default:
851 die("corrupted pack file");
853 *store_size = 0; /* notyet */
856 static int packed_object_info(struct pack_entry *entry,
857 char *type, unsigned long *sizep)
859 struct packed_git *p = entry->p;
860 unsigned long offset, size, left;
861 unsigned char *pack;
862 enum object_type kind;
863 int retval;
865 if (use_packed_git(p))
866 die("cannot map packed file");
868 offset = unpack_object_header(p, entry->offset, &kind, &size);
869 pack = p->pack_base + offset;
870 left = p->pack_size - offset;
872 switch (kind) {
873 case OBJ_DELTA:
874 retval = packed_delta_info(pack, size, left, type, sizep, p);
875 unuse_packed_git(p);
876 return retval;
877 case OBJ_COMMIT:
878 strcpy(type, "commit");
879 break;
880 case OBJ_TREE:
881 strcpy(type, "tree");
882 break;
883 case OBJ_BLOB:
884 strcpy(type, "blob");
885 break;
886 case OBJ_TAG:
887 strcpy(type, "tag");
888 break;
889 default:
890 die("corrupted pack file");
892 if (sizep)
893 *sizep = size;
894 unuse_packed_git(p);
895 return 0;
898 /* forward declaration for a mutually recursive function */
899 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
901 static void *unpack_delta_entry(unsigned char *base_sha1,
902 unsigned long delta_size,
903 unsigned long left,
904 char *type,
905 unsigned long *sizep,
906 struct packed_git *p)
908 struct pack_entry base_ent;
909 void *data, *delta_data, *result, *base;
910 unsigned long data_size, result_size, base_size;
911 z_stream stream;
912 int st;
914 if (left < 20)
915 die("truncated pack file");
916 data = base_sha1 + 20;
917 data_size = left - 20;
918 delta_data = xmalloc(delta_size);
920 memset(&stream, 0, sizeof(stream));
922 stream.next_in = data;
923 stream.avail_in = data_size;
924 stream.next_out = delta_data;
925 stream.avail_out = delta_size;
927 inflateInit(&stream);
928 st = inflate(&stream, Z_FINISH);
929 inflateEnd(&stream);
930 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
931 die("delta data unpack failed");
933 /* The base entry _must_ be in the same pack */
934 if (!find_pack_entry_one(base_sha1, &base_ent, p))
935 die("failed to find delta-pack base object %s",
936 sha1_to_hex(base_sha1));
937 base = unpack_entry_gently(&base_ent, type, &base_size);
938 if (!base)
939 die("failed to read delta-pack base object %s",
940 sha1_to_hex(base_sha1));
941 result = patch_delta(base, base_size,
942 delta_data, delta_size,
943 &result_size);
944 if (!result)
945 die("failed to apply delta");
946 free(delta_data);
947 free(base);
948 *sizep = result_size;
949 return result;
952 static void *unpack_non_delta_entry(unsigned char *data,
953 unsigned long size,
954 unsigned long left)
956 int st;
957 z_stream stream;
958 unsigned char *buffer;
960 buffer = xmalloc(size + 1);
961 buffer[size] = 0;
962 memset(&stream, 0, sizeof(stream));
963 stream.next_in = data;
964 stream.avail_in = left;
965 stream.next_out = buffer;
966 stream.avail_out = size;
968 inflateInit(&stream);
969 st = inflate(&stream, Z_FINISH);
970 inflateEnd(&stream);
971 if ((st != Z_STREAM_END) || stream.total_out != size) {
972 free(buffer);
973 return NULL;
976 return buffer;
979 static void *unpack_entry(struct pack_entry *entry,
980 char *type, unsigned long *sizep)
982 struct packed_git *p = entry->p;
983 void *retval;
985 if (use_packed_git(p))
986 die("cannot map packed file");
987 retval = unpack_entry_gently(entry, type, sizep);
988 unuse_packed_git(p);
989 if (!retval)
990 die("corrupted pack file");
991 return retval;
994 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
995 void *unpack_entry_gently(struct pack_entry *entry,
996 char *type, unsigned long *sizep)
998 struct packed_git *p = entry->p;
999 unsigned long offset, size, left;
1000 unsigned char *pack;
1001 enum object_type kind;
1002 void *retval;
1004 offset = unpack_object_header(p, entry->offset, &kind, &size);
1005 pack = p->pack_base + offset;
1006 left = p->pack_size - offset;
1007 switch (kind) {
1008 case OBJ_DELTA:
1009 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1010 return retval;
1011 case OBJ_COMMIT:
1012 strcpy(type, "commit");
1013 break;
1014 case OBJ_TREE:
1015 strcpy(type, "tree");
1016 break;
1017 case OBJ_BLOB:
1018 strcpy(type, "blob");
1019 break;
1020 case OBJ_TAG:
1021 strcpy(type, "tag");
1022 break;
1023 default:
1024 return NULL;
1026 *sizep = size;
1027 retval = unpack_non_delta_entry(pack, size, left);
1028 return retval;
1031 int num_packed_objects(const struct packed_git *p)
1033 /* See check_packed_git_idx() */
1034 return (p->index_size - 20 - 20 - 4*256) / 24;
1037 int nth_packed_object_sha1(const struct packed_git *p, int n,
1038 unsigned char* sha1)
1040 void *index = p->index_base + 256;
1041 if (n < 0 || num_packed_objects(p) <= n)
1042 return -1;
1043 memcpy(sha1, (index + 24 * n + 4), 20);
1044 return 0;
1047 int find_pack_entry_one(const unsigned char *sha1,
1048 struct pack_entry *e, struct packed_git *p)
1050 unsigned int *level1_ofs = p->index_base;
1051 int hi = ntohl(level1_ofs[*sha1]);
1052 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1053 void *index = p->index_base + 256;
1055 do {
1056 int mi = (lo + hi) / 2;
1057 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1058 if (!cmp) {
1059 e->offset = ntohl(*((int*)(index + 24 * mi)));
1060 memcpy(e->sha1, sha1, 20);
1061 e->p = p;
1062 return 1;
1064 if (cmp > 0)
1065 hi = mi;
1066 else
1067 lo = mi+1;
1068 } while (lo < hi);
1069 return 0;
1072 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1074 struct packed_git *p;
1075 prepare_packed_git();
1077 for (p = packed_git; p; p = p->next) {
1078 if (find_pack_entry_one(sha1, e, p))
1079 return 1;
1081 return 0;
1084 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1085 struct packed_git *packs)
1087 struct packed_git *p;
1088 struct pack_entry e;
1090 for (p = packs; p; p = p->next) {
1091 if (find_pack_entry_one(sha1, &e, p))
1092 return p;
1094 return NULL;
1098 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1100 int status;
1101 unsigned long mapsize, size;
1102 void *map;
1103 z_stream stream;
1104 char hdr[128];
1106 map = map_sha1_file_internal(sha1, &mapsize);
1107 if (!map) {
1108 struct pack_entry e;
1110 if (!find_pack_entry(sha1, &e))
1111 return error("unable to find %s", sha1_to_hex(sha1));
1112 return packed_object_info(&e, type, sizep);
1114 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1115 status = error("unable to unpack %s header",
1116 sha1_to_hex(sha1));
1117 if (parse_sha1_header(hdr, type, &size) < 0)
1118 status = error("unable to parse %s header", sha1_to_hex(sha1));
1119 else {
1120 status = 0;
1121 if (sizep)
1122 *sizep = size;
1124 inflateEnd(&stream);
1125 munmap(map, mapsize);
1126 return status;
1129 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1131 struct pack_entry e;
1133 if (!find_pack_entry(sha1, &e)) {
1134 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1135 return NULL;
1137 return unpack_entry(&e, type, size);
1140 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1142 unsigned long mapsize;
1143 void *map, *buf;
1144 struct pack_entry e;
1146 if (find_pack_entry(sha1, &e))
1147 return read_packed_sha1(sha1, type, size);
1148 map = map_sha1_file_internal(sha1, &mapsize);
1149 if (map) {
1150 buf = unpack_sha1_file(map, mapsize, type, size);
1151 munmap(map, mapsize);
1152 return buf;
1154 return NULL;
1157 void *read_object_with_reference(const unsigned char *sha1,
1158 const char *required_type,
1159 unsigned long *size,
1160 unsigned char *actual_sha1_return)
1162 char type[20];
1163 void *buffer;
1164 unsigned long isize;
1165 unsigned char actual_sha1[20];
1167 memcpy(actual_sha1, sha1, 20);
1168 while (1) {
1169 int ref_length = -1;
1170 const char *ref_type = NULL;
1172 buffer = read_sha1_file(actual_sha1, type, &isize);
1173 if (!buffer)
1174 return NULL;
1175 if (!strcmp(type, required_type)) {
1176 *size = isize;
1177 if (actual_sha1_return)
1178 memcpy(actual_sha1_return, actual_sha1, 20);
1179 return buffer;
1181 /* Handle references */
1182 else if (!strcmp(type, "commit"))
1183 ref_type = "tree ";
1184 else if (!strcmp(type, "tag"))
1185 ref_type = "object ";
1186 else {
1187 free(buffer);
1188 return NULL;
1190 ref_length = strlen(ref_type);
1192 if (memcmp(buffer, ref_type, ref_length) ||
1193 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1194 free(buffer);
1195 return NULL;
1197 free(buffer);
1198 /* Now we have the ID of the referred-to object in
1199 * actual_sha1. Check again. */
1203 char *write_sha1_file_prepare(void *buf,
1204 unsigned long len,
1205 const char *type,
1206 unsigned char *sha1,
1207 unsigned char *hdr,
1208 int *hdrlen)
1210 SHA_CTX c;
1212 /* Generate the header */
1213 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1215 /* Sha1.. */
1216 SHA1_Init(&c);
1217 SHA1_Update(&c, hdr, *hdrlen);
1218 SHA1_Update(&c, buf, len);
1219 SHA1_Final(sha1, &c);
1221 return sha1_file_name(sha1);
1224 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1226 int size;
1227 unsigned char *compressed;
1228 z_stream stream;
1229 unsigned char sha1[20];
1230 char *filename;
1231 static char tmpfile[PATH_MAX];
1232 unsigned char hdr[50];
1233 int fd, hdrlen, ret;
1235 /* Normally if we have it in the pack then we do not bother writing
1236 * it out into .git/objects/??/?{38} file.
1238 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1239 if (returnsha1)
1240 memcpy(returnsha1, sha1, 20);
1241 if (has_sha1_file(sha1))
1242 return 0;
1243 fd = open(filename, O_RDONLY);
1244 if (fd >= 0) {
1246 * FIXME!!! We might do collision checking here, but we'd
1247 * need to uncompress the old file and check it. Later.
1249 close(fd);
1250 return 0;
1253 if (errno != ENOENT) {
1254 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1255 return -1;
1258 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1260 fd = mkstemp(tmpfile);
1261 if (fd < 0) {
1262 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1263 return -1;
1266 /* Set it up */
1267 memset(&stream, 0, sizeof(stream));
1268 deflateInit(&stream, Z_BEST_COMPRESSION);
1269 size = deflateBound(&stream, len+hdrlen);
1270 compressed = xmalloc(size);
1272 /* Compress it */
1273 stream.next_out = compressed;
1274 stream.avail_out = size;
1276 /* First header.. */
1277 stream.next_in = hdr;
1278 stream.avail_in = hdrlen;
1279 while (deflate(&stream, 0) == Z_OK)
1280 /* nothing */;
1282 /* Then the data itself.. */
1283 stream.next_in = buf;
1284 stream.avail_in = len;
1285 while (deflate(&stream, Z_FINISH) == Z_OK)
1286 /* nothing */;
1287 deflateEnd(&stream);
1288 size = stream.total_out;
1290 if (write(fd, compressed, size) != size)
1291 die("unable to write file");
1292 fchmod(fd, 0444);
1293 close(fd);
1294 free(compressed);
1296 ret = link(tmpfile, filename);
1297 if (ret < 0) {
1298 ret = errno;
1301 * Coda hack - coda doesn't like cross-directory links,
1302 * so we fall back to a rename, which will mean that it
1303 * won't be able to check collisions, but that's not a
1304 * big deal.
1306 * When this succeeds, we just return 0. We have nothing
1307 * left to unlink.
1309 if (ret == EXDEV && !rename(tmpfile, filename))
1310 return 0;
1312 unlink(tmpfile);
1313 if (ret) {
1314 if (ret != EEXIST) {
1315 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1316 return -1;
1318 /* FIXME!!! Collision check here ? */
1321 return 0;
1324 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1326 ssize_t size;
1327 unsigned long objsize;
1328 int posn = 0;
1329 void *map = map_sha1_file_internal(sha1, &objsize);
1330 void *buf = map;
1331 void *temp_obj = NULL;
1332 z_stream stream;
1334 if (!buf) {
1335 unsigned char *unpacked;
1336 unsigned long len;
1337 char type[20];
1338 char hdr[50];
1339 int hdrlen;
1340 // need to unpack and recompress it by itself
1341 unpacked = read_packed_sha1(sha1, type, &len);
1343 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1345 /* Set it up */
1346 memset(&stream, 0, sizeof(stream));
1347 deflateInit(&stream, Z_BEST_COMPRESSION);
1348 size = deflateBound(&stream, len + hdrlen);
1349 temp_obj = buf = xmalloc(size);
1351 /* Compress it */
1352 stream.next_out = buf;
1353 stream.avail_out = size;
1355 /* First header.. */
1356 stream.next_in = (void *)hdr;
1357 stream.avail_in = hdrlen;
1358 while (deflate(&stream, 0) == Z_OK)
1359 /* nothing */;
1361 /* Then the data itself.. */
1362 stream.next_in = unpacked;
1363 stream.avail_in = len;
1364 while (deflate(&stream, Z_FINISH) == Z_OK)
1365 /* nothing */;
1366 deflateEnd(&stream);
1367 free(unpacked);
1369 objsize = stream.total_out;
1372 do {
1373 size = write(fd, buf + posn, objsize - posn);
1374 if (size <= 0) {
1375 if (!size) {
1376 fprintf(stderr, "write closed");
1377 } else {
1378 perror("write ");
1380 return -1;
1382 posn += size;
1383 } while (posn < objsize);
1385 if (map)
1386 munmap(map, objsize);
1387 if (temp_obj)
1388 free(temp_obj);
1390 return 0;
1393 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1394 size_t bufsize, size_t *bufposn)
1396 char *filename = sha1_file_name(sha1);
1398 int local;
1399 z_stream stream;
1400 unsigned char real_sha1[20];
1401 unsigned char discard[4096];
1402 int ret;
1403 SHA_CTX c;
1405 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1407 if (local < 0)
1408 return error("Couldn't open %s\n", filename);
1410 memset(&stream, 0, sizeof(stream));
1412 inflateInit(&stream);
1414 SHA1_Init(&c);
1416 do {
1417 ssize_t size;
1418 if (*bufposn) {
1419 stream.avail_in = *bufposn;
1420 stream.next_in = (unsigned char *) buffer;
1421 do {
1422 stream.next_out = discard;
1423 stream.avail_out = sizeof(discard);
1424 ret = inflate(&stream, Z_SYNC_FLUSH);
1425 SHA1_Update(&c, discard, sizeof(discard) -
1426 stream.avail_out);
1427 } while (stream.avail_in && ret == Z_OK);
1428 write(local, buffer, *bufposn - stream.avail_in);
1429 memmove(buffer, buffer + *bufposn - stream.avail_in,
1430 stream.avail_in);
1431 *bufposn = stream.avail_in;
1432 if (ret != Z_OK)
1433 break;
1435 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1436 if (size <= 0) {
1437 close(local);
1438 unlink(filename);
1439 if (!size)
1440 return error("Connection closed?");
1441 perror("Reading from connection");
1442 return -1;
1444 *bufposn += size;
1445 } while (1);
1446 inflateEnd(&stream);
1448 close(local);
1449 SHA1_Final(real_sha1, &c);
1450 if (ret != Z_STREAM_END) {
1451 unlink(filename);
1452 return error("File %s corrupted", sha1_to_hex(sha1));
1454 if (memcmp(sha1, real_sha1, 20)) {
1455 unlink(filename);
1456 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1459 return 0;
1462 int has_pack_index(const unsigned char *sha1)
1464 struct stat st;
1465 if (stat(sha1_pack_index_name(sha1), &st))
1466 return 0;
1467 return 1;
1470 int has_pack_file(const unsigned char *sha1)
1472 struct stat st;
1473 if (stat(sha1_pack_name(sha1), &st))
1474 return 0;
1475 return 1;
1478 int has_sha1_pack(const unsigned char *sha1)
1480 struct pack_entry e;
1481 return find_pack_entry(sha1, &e);
1484 int has_sha1_file(const unsigned char *sha1)
1486 struct stat st;
1487 struct pack_entry e;
1489 if (find_pack_entry(sha1, &e))
1490 return 1;
1491 return find_sha1_file(sha1, &st) ? 1 : 0;
1494 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1496 unsigned long size = st->st_size;
1497 void *buf;
1498 int ret;
1499 unsigned char hdr[50];
1500 int hdrlen;
1502 buf = "";
1503 if (size)
1504 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1505 close(fd);
1506 if (buf == MAP_FAILED)
1507 return -1;
1509 if (!type)
1510 type = "blob";
1511 if (write_object)
1512 ret = write_sha1_file(buf, size, type, sha1);
1513 else {
1514 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1515 ret = 0;
1517 if (size)
1518 munmap(buf, size);
1519 return ret;