Merge branch 'jc/combine' into next
[git/git-svn.git] / sha1_file.c
blobba8c4f76011c68548e959a17883b9a9d6cfbeeff
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 "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
17 #ifndef O_NOATIME
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
20 #else
21 #define O_NOATIME 0
22 #endif
23 #endif
25 const unsigned char null_sha1[20] = { 0, };
27 static unsigned int sha1_file_open_flag = O_NOATIME;
29 static unsigned hexval(char c)
31 if (c >= '0' && c <= '9')
32 return c - '0';
33 if (c >= 'a' && c <= 'f')
34 return c - 'a' + 10;
35 if (c >= 'A' && c <= 'F')
36 return c - 'A' + 10;
37 return ~0;
40 int get_sha1_hex(const char *hex, unsigned char *sha1)
42 int i;
43 for (i = 0; i < 20; i++) {
44 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
45 if (val & ~0xff)
46 return -1;
47 *sha1++ = val;
48 hex += 2;
50 return 0;
53 int adjust_shared_perm(const char *path)
55 struct stat st;
56 int mode;
58 if (!shared_repository)
59 return 0;
60 if (lstat(path, &st) < 0)
61 return -1;
62 mode = st.st_mode;
63 if (mode & S_IRUSR)
64 mode |= S_IRGRP;
65 if (mode & S_IWUSR)
66 mode |= S_IWGRP;
67 if (mode & S_IXUSR)
68 mode |= S_IXGRP;
69 if (S_ISDIR(mode))
70 mode |= S_ISGID;
71 if (chmod(path, mode) < 0)
72 return -2;
73 return 0;
76 int safe_create_leading_directories(char *path)
78 char *pos = path;
79 struct stat st;
81 if (*pos == '/')
82 pos++;
84 while (pos) {
85 pos = strchr(pos, '/');
86 if (!pos)
87 break;
88 *pos = 0;
89 if (!stat(path, &st)) {
90 /* path exists */
91 if (!S_ISDIR(st.st_mode)) {
92 *pos = '/';
93 return -3;
96 else if (mkdir(path, 0777)) {
97 *pos = '/';
98 return -1;
100 else if (adjust_shared_perm(path)) {
101 *pos = '/';
102 return -2;
104 *pos++ = '/';
106 return 0;
109 char * sha1_to_hex(const unsigned char *sha1)
111 static char buffer[50];
112 static const char hex[] = "0123456789abcdef";
113 char *buf = buffer;
114 int i;
116 for (i = 0; i < 20; i++) {
117 unsigned int val = *sha1++;
118 *buf++ = hex[val >> 4];
119 *buf++ = hex[val & 0xf];
121 *buf = '\0';
123 return buffer;
126 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
128 int i;
129 for (i = 0; i < 20; i++) {
130 static char hex[] = "0123456789abcdef";
131 unsigned int val = sha1[i];
132 char *pos = pathbuf + i*2 + (i > 0);
133 *pos++ = hex[val >> 4];
134 *pos = hex[val & 0xf];
139 * NOTE! This returns a statically allocated buffer, so you have to be
140 * careful about using it. Do a "strdup()" if you need to save the
141 * filename.
143 * Also note that this returns the location for creating. Reading
144 * SHA1 file can happen from any alternate directory listed in the
145 * DB_ENVIRONMENT environment variable if it is not found in
146 * the primary object database.
148 char *sha1_file_name(const unsigned char *sha1)
150 static char *name, *base;
152 if (!base) {
153 const char *sha1_file_directory = get_object_directory();
154 int len = strlen(sha1_file_directory);
155 base = xmalloc(len + 60);
156 memcpy(base, sha1_file_directory, len);
157 memset(base+len, 0, 60);
158 base[len] = '/';
159 base[len+3] = '/';
160 name = base + len + 1;
162 fill_sha1_path(name, sha1);
163 return base;
166 char *sha1_pack_name(const unsigned char *sha1)
168 static const char hex[] = "0123456789abcdef";
169 static char *name, *base, *buf;
170 int i;
172 if (!base) {
173 const char *sha1_file_directory = get_object_directory();
174 int len = strlen(sha1_file_directory);
175 base = xmalloc(len + 60);
176 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
177 name = base + len + 11;
180 buf = name;
182 for (i = 0; i < 20; i++) {
183 unsigned int val = *sha1++;
184 *buf++ = hex[val >> 4];
185 *buf++ = hex[val & 0xf];
188 return base;
191 char *sha1_pack_index_name(const unsigned char *sha1)
193 static const char hex[] = "0123456789abcdef";
194 static char *name, *base, *buf;
195 int i;
197 if (!base) {
198 const char *sha1_file_directory = get_object_directory();
199 int len = strlen(sha1_file_directory);
200 base = xmalloc(len + 60);
201 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
202 name = base + len + 11;
205 buf = name;
207 for (i = 0; i < 20; i++) {
208 unsigned int val = *sha1++;
209 *buf++ = hex[val >> 4];
210 *buf++ = hex[val & 0xf];
213 return base;
216 struct alternate_object_database *alt_odb_list;
217 static struct alternate_object_database **alt_odb_tail;
220 * Prepare alternate object database registry.
222 * The variable alt_odb_list points at the list of struct
223 * alternate_object_database. The elements on this list come from
224 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
225 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
226 * whose contents is similar to that environment variable but can be
227 * LF separated. Its base points at a statically allocated buffer that
228 * contains "/the/directory/corresponding/to/.git/objects/...", while
229 * its name points just after the slash at the end of ".git/objects/"
230 * in the example above, and has enough space to hold 40-byte hex
231 * SHA1, an extra slash for the first level indirection, and the
232 * terminating NUL.
234 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
235 const char *relative_base)
237 const char *cp, *last;
238 struct alternate_object_database *ent;
239 const char *objdir = get_object_directory();
240 int base_len = -1;
242 last = alt;
243 while (last < ep) {
244 cp = last;
245 if (cp < ep && *cp == '#') {
246 while (cp < ep && *cp != sep)
247 cp++;
248 last = cp + 1;
249 continue;
251 for ( ; cp < ep && *cp != sep; cp++)
253 if (last != cp) {
254 struct stat st;
255 struct alternate_object_database *alt;
256 /* 43 = 40-byte + 2 '/' + terminating NUL */
257 int pfxlen = cp - last;
258 int entlen = pfxlen + 43;
260 if (*last != '/' && relative_base) {
261 /* Relative alt-odb */
262 if (base_len < 0)
263 base_len = strlen(relative_base) + 1;
264 entlen += base_len;
265 pfxlen += base_len;
267 ent = xmalloc(sizeof(*ent) + entlen);
269 if (*last != '/' && relative_base) {
270 memcpy(ent->base, relative_base, base_len - 1);
271 ent->base[base_len - 1] = '/';
272 memcpy(ent->base + base_len,
273 last, cp - last);
275 else
276 memcpy(ent->base, last, pfxlen);
278 ent->name = ent->base + pfxlen + 1;
279 ent->base[pfxlen + 3] = '/';
280 ent->base[pfxlen] = ent->base[entlen-1] = 0;
282 /* Detect cases where alternate disappeared */
283 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
284 error("object directory %s does not exist; "
285 "check .git/objects/info/alternates.",
286 ent->base);
287 goto bad;
289 ent->base[pfxlen] = '/';
291 /* Prevent the common mistake of listing the same
292 * thing twice, or object directory itself.
294 for (alt = alt_odb_list; alt; alt = alt->next)
295 if (!memcmp(ent->base, alt->base, pfxlen))
296 goto bad;
297 if (!memcmp(ent->base, objdir, pfxlen)) {
298 bad:
299 free(ent);
301 else {
302 *alt_odb_tail = ent;
303 alt_odb_tail = &(ent->next);
304 ent->next = NULL;
307 while (cp < ep && *cp == sep)
308 cp++;
309 last = cp;
313 void prepare_alt_odb(void)
315 char path[PATH_MAX];
316 char *map;
317 int fd;
318 struct stat st;
319 char *alt;
321 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
322 if (!alt) alt = "";
324 if (alt_odb_tail)
325 return;
326 alt_odb_tail = &alt_odb_list;
327 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL);
329 sprintf(path, "%s/info/alternates", get_object_directory());
330 fd = open(path, O_RDONLY);
331 if (fd < 0)
332 return;
333 if (fstat(fd, &st) || (st.st_size == 0)) {
334 close(fd);
335 return;
337 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
338 close(fd);
339 if (map == MAP_FAILED)
340 return;
342 link_alt_odb_entries(map, map + st.st_size, '\n',
343 get_object_directory());
344 munmap(map, st.st_size);
347 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
349 char *name = sha1_file_name(sha1);
350 struct alternate_object_database *alt;
352 if (!stat(name, st))
353 return name;
354 prepare_alt_odb();
355 for (alt = alt_odb_list; alt; alt = alt->next) {
356 name = alt->name;
357 fill_sha1_path(name, sha1);
358 if (!stat(alt->base, st))
359 return alt->base;
361 return NULL;
364 #define PACK_MAX_SZ (1<<26)
365 static int pack_used_ctr;
366 static unsigned long pack_mapped;
367 struct packed_git *packed_git;
369 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
370 void **idx_map_)
372 void *idx_map;
373 unsigned int *index;
374 unsigned long idx_size;
375 int nr, i;
376 int fd = open(path, O_RDONLY);
377 struct stat st;
378 if (fd < 0)
379 return -1;
380 if (fstat(fd, &st)) {
381 close(fd);
382 return -1;
384 idx_size = st.st_size;
385 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
386 close(fd);
387 if (idx_map == MAP_FAILED)
388 return -1;
390 index = idx_map;
391 *idx_map_ = idx_map;
392 *idx_size_ = idx_size;
394 /* check index map */
395 if (idx_size < 4*256 + 20 + 20)
396 return error("index file too small");
397 nr = 0;
398 for (i = 0; i < 256; i++) {
399 unsigned int n = ntohl(index[i]);
400 if (n < nr)
401 return error("non-monotonic index");
402 nr = n;
406 * Total size:
407 * - 256 index entries 4 bytes each
408 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
409 * - 20-byte SHA1 of the packfile
410 * - 20-byte SHA1 file checksum
412 if (idx_size != 4*256 + nr * 24 + 20 + 20)
413 return error("wrong index file size");
415 return 0;
418 static int unuse_one_packed_git(void)
420 struct packed_git *p, *lru = NULL;
422 for (p = packed_git; p; p = p->next) {
423 if (p->pack_use_cnt || !p->pack_base)
424 continue;
425 if (!lru || p->pack_last_used < lru->pack_last_used)
426 lru = p;
428 if (!lru)
429 return 0;
430 munmap(lru->pack_base, lru->pack_size);
431 lru->pack_base = NULL;
432 return 1;
435 void unuse_packed_git(struct packed_git *p)
437 p->pack_use_cnt--;
440 int use_packed_git(struct packed_git *p)
442 if (!p->pack_size) {
443 struct stat st;
444 // We created the struct before we had the pack
445 stat(p->pack_name, &st);
446 if (!S_ISREG(st.st_mode))
447 die("packfile %s not a regular file", p->pack_name);
448 p->pack_size = st.st_size;
450 if (!p->pack_base) {
451 int fd;
452 struct stat st;
453 void *map;
455 pack_mapped += p->pack_size;
456 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
457 ; /* nothing */
458 fd = open(p->pack_name, O_RDONLY);
459 if (fd < 0)
460 die("packfile %s cannot be opened", p->pack_name);
461 if (fstat(fd, &st)) {
462 close(fd);
463 die("packfile %s cannot be opened", p->pack_name);
465 if (st.st_size != p->pack_size)
466 die("packfile %s size mismatch.", p->pack_name);
467 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
468 close(fd);
469 if (map == MAP_FAILED)
470 die("packfile %s cannot be mapped.", p->pack_name);
471 p->pack_base = map;
473 /* Check if the pack file matches with the index file.
474 * this is cheap.
476 if (memcmp((char*)(p->index_base) + p->index_size - 40,
477 p->pack_base + p->pack_size - 20, 20)) {
479 die("packfile %s does not match index.", p->pack_name);
482 p->pack_last_used = pack_used_ctr++;
483 p->pack_use_cnt++;
484 return 0;
487 struct packed_git *add_packed_git(char *path, int path_len, int local)
489 struct stat st;
490 struct packed_git *p;
491 unsigned long idx_size;
492 void *idx_map;
493 unsigned char sha1[20];
495 if (check_packed_git_idx(path, &idx_size, &idx_map))
496 return NULL;
498 /* do we have a corresponding .pack file? */
499 strcpy(path + path_len - 4, ".pack");
500 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
501 munmap(idx_map, idx_size);
502 return NULL;
504 /* ok, it looks sane as far as we can check without
505 * actually mapping the pack file.
507 p = xmalloc(sizeof(*p) + path_len + 2);
508 strcpy(p->pack_name, path);
509 p->index_size = idx_size;
510 p->pack_size = st.st_size;
511 p->index_base = idx_map;
512 p->next = NULL;
513 p->pack_base = NULL;
514 p->pack_last_used = 0;
515 p->pack_use_cnt = 0;
516 p->pack_local = local;
517 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
518 memcpy(p->sha1, sha1, 20);
519 return p;
522 struct packed_git *parse_pack_index(unsigned char *sha1)
524 char *path = sha1_pack_index_name(sha1);
525 return parse_pack_index_file(sha1, path);
528 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
530 struct packed_git *p;
531 unsigned long idx_size;
532 void *idx_map;
533 char *path;
535 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
536 return NULL;
538 path = sha1_pack_name(sha1);
540 p = xmalloc(sizeof(*p) + strlen(path) + 2);
541 strcpy(p->pack_name, path);
542 p->index_size = idx_size;
543 p->pack_size = 0;
544 p->index_base = idx_map;
545 p->next = NULL;
546 p->pack_base = NULL;
547 p->pack_last_used = 0;
548 p->pack_use_cnt = 0;
549 memcpy(p->sha1, sha1, 20);
550 return p;
553 void install_packed_git(struct packed_git *pack)
555 pack->next = packed_git;
556 packed_git = pack;
559 static void prepare_packed_git_one(char *objdir, int local)
561 char path[PATH_MAX];
562 int len;
563 DIR *dir;
564 struct dirent *de;
566 sprintf(path, "%s/pack", objdir);
567 len = strlen(path);
568 dir = opendir(path);
569 if (!dir) {
570 if (errno != ENOENT)
571 error("unable to open object pack directory: %s: %s",
572 path, strerror(errno));
573 return;
575 path[len++] = '/';
576 while ((de = readdir(dir)) != NULL) {
577 int namelen = strlen(de->d_name);
578 struct packed_git *p;
580 if (strcmp(de->d_name + namelen - 4, ".idx"))
581 continue;
583 /* we have .idx. Is it a file we can map? */
584 strcpy(path + len, de->d_name);
585 p = add_packed_git(path, len + namelen, local);
586 if (!p)
587 continue;
588 p->next = packed_git;
589 packed_git = p;
591 closedir(dir);
594 void prepare_packed_git(void)
596 static int run_once = 0;
597 struct alternate_object_database *alt;
599 if (run_once)
600 return;
601 prepare_packed_git_one(get_object_directory(), 1);
602 prepare_alt_odb();
603 for (alt = alt_odb_list; alt; alt = alt->next) {
604 alt->name[-1] = 0;
605 prepare_packed_git_one(alt->base, 0);
606 alt->name[-1] = '/';
608 run_once = 1;
611 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
613 char header[100];
614 unsigned char real_sha1[20];
615 SHA_CTX c;
617 SHA1_Init(&c);
618 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
619 SHA1_Update(&c, map, size);
620 SHA1_Final(real_sha1, &c);
621 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
624 static void *map_sha1_file_internal(const unsigned char *sha1,
625 unsigned long *size)
627 struct stat st;
628 void *map;
629 int fd;
630 char *filename = find_sha1_file(sha1, &st);
632 if (!filename) {
633 return NULL;
636 fd = open(filename, O_RDONLY | sha1_file_open_flag);
637 if (fd < 0) {
638 /* See if it works without O_NOATIME */
639 switch (sha1_file_open_flag) {
640 default:
641 fd = open(filename, O_RDONLY);
642 if (fd >= 0)
643 break;
644 /* Fallthrough */
645 case 0:
646 return NULL;
649 /* If it failed once, it will probably fail again.
650 * Stop using O_NOATIME
652 sha1_file_open_flag = 0;
654 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
655 close(fd);
656 if (map == MAP_FAILED)
657 return NULL;
658 *size = st.st_size;
659 return map;
662 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
664 /* Get the data stream */
665 memset(stream, 0, sizeof(*stream));
666 stream->next_in = map;
667 stream->avail_in = mapsize;
668 stream->next_out = buffer;
669 stream->avail_out = size;
671 inflateInit(stream);
672 return inflate(stream, 0);
675 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
677 int bytes = strlen(buffer) + 1;
678 unsigned char *buf = xmalloc(1+size);
680 memcpy(buf, buffer + bytes, stream->total_out - bytes);
681 bytes = stream->total_out - bytes;
682 if (bytes < size) {
683 stream->next_out = buf + bytes;
684 stream->avail_out = size - bytes;
685 while (inflate(stream, Z_FINISH) == Z_OK)
686 /* nothing */;
688 buf[size] = 0;
689 inflateEnd(stream);
690 return buf;
694 * We used to just use "sscanf()", but that's actually way
695 * too permissive for what we want to check. So do an anal
696 * object header parse by hand.
698 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
700 int i;
701 unsigned long size;
704 * The type can be at most ten bytes (including the
705 * terminating '\0' that we add), and is followed by
706 * a space.
708 i = 10;
709 for (;;) {
710 char c = *hdr++;
711 if (c == ' ')
712 break;
713 if (!--i)
714 return -1;
715 *type++ = c;
717 *type = 0;
720 * The length must follow immediately, and be in canonical
721 * decimal format (ie "010" is not valid).
723 size = *hdr++ - '0';
724 if (size > 9)
725 return -1;
726 if (size) {
727 for (;;) {
728 unsigned long c = *hdr - '0';
729 if (c > 9)
730 break;
731 hdr++;
732 size = size * 10 + c;
735 *sizep = size;
738 * The length must be followed by a zero byte
740 return *hdr ? -1 : 0;
743 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
745 int ret;
746 z_stream stream;
747 char hdr[8192];
749 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
750 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
751 return NULL;
753 return unpack_sha1_rest(&stream, hdr, *size);
756 /* forward declaration for a mutually recursive function */
757 static int packed_object_info(struct pack_entry *entry,
758 char *type, unsigned long *sizep);
760 static int packed_delta_info(unsigned char *base_sha1,
761 unsigned long delta_size,
762 unsigned long left,
763 char *type,
764 unsigned long *sizep,
765 struct packed_git *p)
767 struct pack_entry base_ent;
769 if (left < 20)
770 die("truncated pack file");
772 /* The base entry _must_ be in the same pack */
773 if (!find_pack_entry_one(base_sha1, &base_ent, p))
774 die("failed to find delta-pack base object %s",
775 sha1_to_hex(base_sha1));
777 /* We choose to only get the type of the base object and
778 * ignore potentially corrupt pack file that expects the delta
779 * based on a base with a wrong size. This saves tons of
780 * inflate() calls.
783 if (packed_object_info(&base_ent, type, NULL))
784 die("cannot get info for delta-pack base");
786 if (sizep) {
787 const unsigned char *data;
788 unsigned char delta_head[64];
789 unsigned long result_size;
790 z_stream stream;
791 int st;
793 memset(&stream, 0, sizeof(stream));
795 data = stream.next_in = base_sha1 + 20;
796 stream.avail_in = left - 20;
797 stream.next_out = delta_head;
798 stream.avail_out = sizeof(delta_head);
800 inflateInit(&stream);
801 st = inflate(&stream, Z_FINISH);
802 inflateEnd(&stream);
803 if ((st != Z_STREAM_END) &&
804 stream.total_out != sizeof(delta_head))
805 die("delta data unpack-initial failed");
807 /* Examine the initial part of the delta to figure out
808 * the result size.
810 data = delta_head;
811 get_delta_hdr_size(&data); /* ignore base size */
813 /* Read the result size */
814 result_size = get_delta_hdr_size(&data);
815 *sizep = result_size;
817 return 0;
820 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
821 enum object_type *type, unsigned long *sizep)
823 unsigned shift;
824 unsigned char *pack, c;
825 unsigned long size;
827 if (offset >= p->pack_size)
828 die("object offset outside of pack file");
830 pack = p->pack_base + offset;
831 c = *pack++;
832 offset++;
833 *type = (c >> 4) & 7;
834 size = c & 15;
835 shift = 4;
836 while (c & 0x80) {
837 if (offset >= p->pack_size)
838 die("object offset outside of pack file");
839 c = *pack++;
840 offset++;
841 size += (c & 0x7f) << shift;
842 shift += 7;
844 *sizep = size;
845 return offset;
848 int check_reuse_pack_delta(struct packed_git *p, unsigned long offset,
849 unsigned char *base, unsigned long *sizep,
850 enum object_type *kindp)
852 unsigned long ptr;
853 int status = -1;
855 use_packed_git(p);
856 ptr = offset;
857 ptr = unpack_object_header(p, ptr, kindp, sizep);
858 if (*kindp != OBJ_DELTA)
859 goto done;
860 memcpy(base, p->pack_base + ptr, 20);
861 status = 0;
862 done:
863 unuse_packed_git(p);
864 return status;
867 void packed_object_info_detail(struct pack_entry *e,
868 char *type,
869 unsigned long *size,
870 unsigned long *store_size,
871 unsigned int *delta_chain_length,
872 unsigned char *base_sha1)
874 struct packed_git *p = e->p;
875 unsigned long offset, left;
876 unsigned char *pack;
877 enum object_type kind;
879 offset = unpack_object_header(p, e->offset, &kind, size);
880 pack = p->pack_base + offset;
881 left = p->pack_size - offset;
882 if (kind != OBJ_DELTA)
883 *delta_chain_length = 0;
884 else {
885 unsigned int chain_length = 0;
886 memcpy(base_sha1, pack, 20);
887 do {
888 struct pack_entry base_ent;
889 unsigned long junk;
891 find_pack_entry_one(pack, &base_ent, p);
892 offset = unpack_object_header(p, base_ent.offset,
893 &kind, &junk);
894 pack = p->pack_base + offset;
895 chain_length++;
896 } while (kind == OBJ_DELTA);
897 *delta_chain_length = chain_length;
899 switch (kind) {
900 case OBJ_COMMIT:
901 strcpy(type, commit_type);
902 break;
903 case OBJ_TREE:
904 strcpy(type, tree_type);
905 break;
906 case OBJ_BLOB:
907 strcpy(type, blob_type);
908 break;
909 case OBJ_TAG:
910 strcpy(type, tag_type);
911 break;
912 default:
913 die("corrupted pack file %s containing object of kind %d",
914 p->pack_name, kind);
916 *store_size = 0; /* notyet */
919 static int packed_object_info(struct pack_entry *entry,
920 char *type, unsigned long *sizep)
922 struct packed_git *p = entry->p;
923 unsigned long offset, size, left;
924 unsigned char *pack;
925 enum object_type kind;
926 int retval;
928 if (use_packed_git(p))
929 die("cannot map packed file");
931 offset = unpack_object_header(p, entry->offset, &kind, &size);
932 pack = p->pack_base + offset;
933 left = p->pack_size - offset;
935 switch (kind) {
936 case OBJ_DELTA:
937 retval = packed_delta_info(pack, size, left, type, sizep, p);
938 unuse_packed_git(p);
939 return retval;
940 case OBJ_COMMIT:
941 strcpy(type, commit_type);
942 break;
943 case OBJ_TREE:
944 strcpy(type, tree_type);
945 break;
946 case OBJ_BLOB:
947 strcpy(type, blob_type);
948 break;
949 case OBJ_TAG:
950 strcpy(type, tag_type);
951 break;
952 default:
953 die("corrupted pack file %s containing object of kind %d",
954 p->pack_name, kind);
956 if (sizep)
957 *sizep = size;
958 unuse_packed_git(p);
959 return 0;
962 /* forward declaration for a mutually recursive function */
963 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
965 static void *unpack_delta_entry(unsigned char *base_sha1,
966 unsigned long delta_size,
967 unsigned long left,
968 char *type,
969 unsigned long *sizep,
970 struct packed_git *p)
972 struct pack_entry base_ent;
973 void *data, *delta_data, *result, *base;
974 unsigned long data_size, result_size, base_size;
975 z_stream stream;
976 int st;
978 if (left < 20)
979 die("truncated pack file");
981 /* The base entry _must_ be in the same pack */
982 if (!find_pack_entry_one(base_sha1, &base_ent, p))
983 die("failed to find delta-pack base object %s",
984 sha1_to_hex(base_sha1));
985 base = unpack_entry_gently(&base_ent, type, &base_size);
986 if (!base)
987 die("failed to read delta-pack base object %s",
988 sha1_to_hex(base_sha1));
990 data = base_sha1 + 20;
991 data_size = left - 20;
992 delta_data = xmalloc(delta_size);
994 memset(&stream, 0, sizeof(stream));
996 stream.next_in = data;
997 stream.avail_in = data_size;
998 stream.next_out = delta_data;
999 stream.avail_out = delta_size;
1001 inflateInit(&stream);
1002 st = inflate(&stream, Z_FINISH);
1003 inflateEnd(&stream);
1004 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
1005 die("delta data unpack failed");
1007 result = patch_delta(base, base_size,
1008 delta_data, delta_size,
1009 &result_size);
1010 if (!result)
1011 die("failed to apply delta");
1012 free(delta_data);
1013 free(base);
1014 *sizep = result_size;
1015 return result;
1018 static void *unpack_non_delta_entry(unsigned char *data,
1019 unsigned long size,
1020 unsigned long left)
1022 int st;
1023 z_stream stream;
1024 unsigned char *buffer;
1026 buffer = xmalloc(size + 1);
1027 buffer[size] = 0;
1028 memset(&stream, 0, sizeof(stream));
1029 stream.next_in = data;
1030 stream.avail_in = left;
1031 stream.next_out = buffer;
1032 stream.avail_out = size;
1034 inflateInit(&stream);
1035 st = inflate(&stream, Z_FINISH);
1036 inflateEnd(&stream);
1037 if ((st != Z_STREAM_END) || stream.total_out != size) {
1038 free(buffer);
1039 return NULL;
1042 return buffer;
1045 static void *unpack_entry(struct pack_entry *entry,
1046 char *type, unsigned long *sizep)
1048 struct packed_git *p = entry->p;
1049 void *retval;
1051 if (use_packed_git(p))
1052 die("cannot map packed file");
1053 retval = unpack_entry_gently(entry, type, sizep);
1054 unuse_packed_git(p);
1055 if (!retval)
1056 die("corrupted pack file %s", p->pack_name);
1057 return retval;
1060 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1061 void *unpack_entry_gently(struct pack_entry *entry,
1062 char *type, unsigned long *sizep)
1064 struct packed_git *p = entry->p;
1065 unsigned long offset, size, left;
1066 unsigned char *pack;
1067 enum object_type kind;
1068 void *retval;
1070 offset = unpack_object_header(p, entry->offset, &kind, &size);
1071 pack = p->pack_base + offset;
1072 left = p->pack_size - offset;
1073 switch (kind) {
1074 case OBJ_DELTA:
1075 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1076 return retval;
1077 case OBJ_COMMIT:
1078 strcpy(type, commit_type);
1079 break;
1080 case OBJ_TREE:
1081 strcpy(type, tree_type);
1082 break;
1083 case OBJ_BLOB:
1084 strcpy(type, blob_type);
1085 break;
1086 case OBJ_TAG:
1087 strcpy(type, tag_type);
1088 break;
1089 default:
1090 return NULL;
1092 *sizep = size;
1093 retval = unpack_non_delta_entry(pack, size, left);
1094 return retval;
1097 int num_packed_objects(const struct packed_git *p)
1099 /* See check_packed_git_idx() */
1100 return (p->index_size - 20 - 20 - 4*256) / 24;
1103 int nth_packed_object_sha1(const struct packed_git *p, int n,
1104 unsigned char* sha1)
1106 void *index = p->index_base + 256;
1107 if (n < 0 || num_packed_objects(p) <= n)
1108 return -1;
1109 memcpy(sha1, (index + 24 * n + 4), 20);
1110 return 0;
1113 int find_pack_entry_one(const unsigned char *sha1,
1114 struct pack_entry *e, struct packed_git *p)
1116 unsigned int *level1_ofs = p->index_base;
1117 int hi = ntohl(level1_ofs[*sha1]);
1118 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1119 void *index = p->index_base + 256;
1121 do {
1122 int mi = (lo + hi) / 2;
1123 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1124 if (!cmp) {
1125 e->offset = ntohl(*((int*)(index + 24 * mi)));
1126 memcpy(e->sha1, sha1, 20);
1127 e->p = p;
1128 return 1;
1130 if (cmp > 0)
1131 hi = mi;
1132 else
1133 lo = mi+1;
1134 } while (lo < hi);
1135 return 0;
1138 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1140 struct packed_git *p;
1141 prepare_packed_git();
1143 for (p = packed_git; p; p = p->next) {
1144 if (find_pack_entry_one(sha1, e, p))
1145 return 1;
1147 return 0;
1150 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1151 struct packed_git *packs)
1153 struct packed_git *p;
1154 struct pack_entry e;
1156 for (p = packs; p; p = p->next) {
1157 if (find_pack_entry_one(sha1, &e, p))
1158 return p;
1160 return NULL;
1164 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1166 int status;
1167 unsigned long mapsize, size;
1168 void *map;
1169 z_stream stream;
1170 char hdr[128];
1172 map = map_sha1_file_internal(sha1, &mapsize);
1173 if (!map) {
1174 struct pack_entry e;
1176 if (!find_pack_entry(sha1, &e))
1177 return error("unable to find %s", sha1_to_hex(sha1));
1178 return packed_object_info(&e, type, sizep);
1180 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1181 status = error("unable to unpack %s header",
1182 sha1_to_hex(sha1));
1183 if (parse_sha1_header(hdr, type, &size) < 0)
1184 status = error("unable to parse %s header", sha1_to_hex(sha1));
1185 else {
1186 status = 0;
1187 if (sizep)
1188 *sizep = size;
1190 inflateEnd(&stream);
1191 munmap(map, mapsize);
1192 return status;
1195 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1197 struct pack_entry e;
1199 if (!find_pack_entry(sha1, &e)) {
1200 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1201 return NULL;
1203 return unpack_entry(&e, type, size);
1206 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1208 unsigned long mapsize;
1209 void *map, *buf;
1210 struct pack_entry e;
1212 if (find_pack_entry(sha1, &e))
1213 return read_packed_sha1(sha1, type, size);
1214 map = map_sha1_file_internal(sha1, &mapsize);
1215 if (map) {
1216 buf = unpack_sha1_file(map, mapsize, type, size);
1217 munmap(map, mapsize);
1218 return buf;
1220 return NULL;
1223 void *read_object_with_reference(const unsigned char *sha1,
1224 const char *required_type,
1225 unsigned long *size,
1226 unsigned char *actual_sha1_return)
1228 char type[20];
1229 void *buffer;
1230 unsigned long isize;
1231 unsigned char actual_sha1[20];
1233 memcpy(actual_sha1, sha1, 20);
1234 while (1) {
1235 int ref_length = -1;
1236 const char *ref_type = NULL;
1238 buffer = read_sha1_file(actual_sha1, type, &isize);
1239 if (!buffer)
1240 return NULL;
1241 if (!strcmp(type, required_type)) {
1242 *size = isize;
1243 if (actual_sha1_return)
1244 memcpy(actual_sha1_return, actual_sha1, 20);
1245 return buffer;
1247 /* Handle references */
1248 else if (!strcmp(type, commit_type))
1249 ref_type = "tree ";
1250 else if (!strcmp(type, tag_type))
1251 ref_type = "object ";
1252 else {
1253 free(buffer);
1254 return NULL;
1256 ref_length = strlen(ref_type);
1258 if (memcmp(buffer, ref_type, ref_length) ||
1259 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1260 free(buffer);
1261 return NULL;
1263 free(buffer);
1264 /* Now we have the ID of the referred-to object in
1265 * actual_sha1. Check again. */
1269 char *write_sha1_file_prepare(void *buf,
1270 unsigned long len,
1271 const char *type,
1272 unsigned char *sha1,
1273 unsigned char *hdr,
1274 int *hdrlen)
1276 SHA_CTX c;
1278 /* Generate the header */
1279 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1281 /* Sha1.. */
1282 SHA1_Init(&c);
1283 SHA1_Update(&c, hdr, *hdrlen);
1284 SHA1_Update(&c, buf, len);
1285 SHA1_Final(sha1, &c);
1287 return sha1_file_name(sha1);
1291 * Link the tempfile to the final place, possibly creating the
1292 * last directory level as you do so.
1294 * Returns the errno on failure, 0 on success.
1296 static int link_temp_to_file(const char *tmpfile, char *filename)
1298 int ret;
1300 if (!link(tmpfile, filename))
1301 return 0;
1304 * Try to mkdir the last path component if that failed
1305 * with an ENOENT.
1307 * Re-try the "link()" regardless of whether the mkdir
1308 * succeeds, since a race might mean that somebody
1309 * else succeeded.
1311 ret = errno;
1312 if (ret == ENOENT) {
1313 char *dir = strrchr(filename, '/');
1314 if (dir) {
1315 *dir = 0;
1316 mkdir(filename, 0777);
1317 if (adjust_shared_perm(filename))
1318 return -2;
1319 *dir = '/';
1320 if (!link(tmpfile, filename))
1321 return 0;
1322 ret = errno;
1325 return ret;
1329 * Move the just written object into its final resting place
1331 int move_temp_to_file(const char *tmpfile, char *filename)
1333 int ret = link_temp_to_file(tmpfile, filename);
1336 * Coda hack - coda doesn't like cross-directory links,
1337 * so we fall back to a rename, which will mean that it
1338 * won't be able to check collisions, but that's not a
1339 * big deal.
1341 * The same holds for FAT formatted media.
1343 * When this succeeds, we just return 0. We have nothing
1344 * left to unlink.
1346 if (ret && ret != EEXIST) {
1347 if (!rename(tmpfile, filename))
1348 return 0;
1349 ret = errno;
1351 unlink(tmpfile);
1352 if (ret) {
1353 if (ret != EEXIST) {
1354 fprintf(stderr, "unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1355 return -1;
1357 /* FIXME!!! Collision check here ? */
1360 return 0;
1363 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1365 int size;
1366 unsigned char *compressed;
1367 z_stream stream;
1368 unsigned char sha1[20];
1369 char *filename;
1370 static char tmpfile[PATH_MAX];
1371 unsigned char hdr[50];
1372 int fd, hdrlen;
1374 /* Normally if we have it in the pack then we do not bother writing
1375 * it out into .git/objects/??/?{38} file.
1377 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1378 if (returnsha1)
1379 memcpy(returnsha1, sha1, 20);
1380 if (has_sha1_file(sha1))
1381 return 0;
1382 fd = open(filename, O_RDONLY);
1383 if (fd >= 0) {
1385 * FIXME!!! We might do collision checking here, but we'd
1386 * need to uncompress the old file and check it. Later.
1388 close(fd);
1389 return 0;
1392 if (errno != ENOENT) {
1393 fprintf(stderr, "sha1 file %s: %s\n", filename, strerror(errno));
1394 return -1;
1397 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1399 fd = mkstemp(tmpfile);
1400 if (fd < 0) {
1401 fprintf(stderr, "unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1402 return -1;
1405 /* Set it up */
1406 memset(&stream, 0, sizeof(stream));
1407 deflateInit(&stream, Z_BEST_COMPRESSION);
1408 size = deflateBound(&stream, len+hdrlen);
1409 compressed = xmalloc(size);
1411 /* Compress it */
1412 stream.next_out = compressed;
1413 stream.avail_out = size;
1415 /* First header.. */
1416 stream.next_in = hdr;
1417 stream.avail_in = hdrlen;
1418 while (deflate(&stream, 0) == Z_OK)
1419 /* nothing */;
1421 /* Then the data itself.. */
1422 stream.next_in = buf;
1423 stream.avail_in = len;
1424 while (deflate(&stream, Z_FINISH) == Z_OK)
1425 /* nothing */;
1426 deflateEnd(&stream);
1427 size = stream.total_out;
1429 if (write(fd, compressed, size) != size)
1430 die("unable to write file");
1431 fchmod(fd, 0444);
1432 close(fd);
1433 free(compressed);
1435 return move_temp_to_file(tmpfile, filename);
1438 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1440 ssize_t size;
1441 unsigned long objsize;
1442 int posn = 0;
1443 void *map = map_sha1_file_internal(sha1, &objsize);
1444 void *buf = map;
1445 void *temp_obj = NULL;
1446 z_stream stream;
1448 if (!buf) {
1449 unsigned char *unpacked;
1450 unsigned long len;
1451 char type[20];
1452 char hdr[50];
1453 int hdrlen;
1454 // need to unpack and recompress it by itself
1455 unpacked = read_packed_sha1(sha1, type, &len);
1457 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1459 /* Set it up */
1460 memset(&stream, 0, sizeof(stream));
1461 deflateInit(&stream, Z_BEST_COMPRESSION);
1462 size = deflateBound(&stream, len + hdrlen);
1463 temp_obj = buf = xmalloc(size);
1465 /* Compress it */
1466 stream.next_out = buf;
1467 stream.avail_out = size;
1469 /* First header.. */
1470 stream.next_in = (void *)hdr;
1471 stream.avail_in = hdrlen;
1472 while (deflate(&stream, 0) == Z_OK)
1473 /* nothing */;
1475 /* Then the data itself.. */
1476 stream.next_in = unpacked;
1477 stream.avail_in = len;
1478 while (deflate(&stream, Z_FINISH) == Z_OK)
1479 /* nothing */;
1480 deflateEnd(&stream);
1481 free(unpacked);
1483 objsize = stream.total_out;
1486 do {
1487 size = write(fd, buf + posn, objsize - posn);
1488 if (size <= 0) {
1489 if (!size) {
1490 fprintf(stderr, "write closed\n");
1491 } else {
1492 perror("write ");
1494 return -1;
1496 posn += size;
1497 } while (posn < objsize);
1499 if (map)
1500 munmap(map, objsize);
1501 if (temp_obj)
1502 free(temp_obj);
1504 return 0;
1507 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1508 size_t bufsize, size_t *bufposn)
1510 char tmpfile[PATH_MAX];
1511 int local;
1512 z_stream stream;
1513 unsigned char real_sha1[20];
1514 unsigned char discard[4096];
1515 int ret;
1516 SHA_CTX c;
1518 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1520 local = mkstemp(tmpfile);
1521 if (local < 0)
1522 return error("Couldn't open %s for %s",
1523 tmpfile, sha1_to_hex(sha1));
1525 memset(&stream, 0, sizeof(stream));
1527 inflateInit(&stream);
1529 SHA1_Init(&c);
1531 do {
1532 ssize_t size;
1533 if (*bufposn) {
1534 stream.avail_in = *bufposn;
1535 stream.next_in = (unsigned char *) buffer;
1536 do {
1537 stream.next_out = discard;
1538 stream.avail_out = sizeof(discard);
1539 ret = inflate(&stream, Z_SYNC_FLUSH);
1540 SHA1_Update(&c, discard, sizeof(discard) -
1541 stream.avail_out);
1542 } while (stream.avail_in && ret == Z_OK);
1543 write(local, buffer, *bufposn - stream.avail_in);
1544 memmove(buffer, buffer + *bufposn - stream.avail_in,
1545 stream.avail_in);
1546 *bufposn = stream.avail_in;
1547 if (ret != Z_OK)
1548 break;
1550 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1551 if (size <= 0) {
1552 close(local);
1553 unlink(tmpfile);
1554 if (!size)
1555 return error("Connection closed?");
1556 perror("Reading from connection");
1557 return -1;
1559 *bufposn += size;
1560 } while (1);
1561 inflateEnd(&stream);
1563 close(local);
1564 SHA1_Final(real_sha1, &c);
1565 if (ret != Z_STREAM_END) {
1566 unlink(tmpfile);
1567 return error("File %s corrupted", sha1_to_hex(sha1));
1569 if (memcmp(sha1, real_sha1, 20)) {
1570 unlink(tmpfile);
1571 return error("File %s has bad hash", sha1_to_hex(sha1));
1574 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1577 int has_pack_index(const unsigned char *sha1)
1579 struct stat st;
1580 if (stat(sha1_pack_index_name(sha1), &st))
1581 return 0;
1582 return 1;
1585 int has_pack_file(const unsigned char *sha1)
1587 struct stat st;
1588 if (stat(sha1_pack_name(sha1), &st))
1589 return 0;
1590 return 1;
1593 int has_sha1_pack(const unsigned char *sha1)
1595 struct pack_entry e;
1596 return find_pack_entry(sha1, &e);
1599 int has_sha1_file(const unsigned char *sha1)
1601 struct stat st;
1602 struct pack_entry e;
1604 if (find_pack_entry(sha1, &e))
1605 return 1;
1606 return find_sha1_file(sha1, &st) ? 1 : 0;
1609 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1611 unsigned long size = 4096;
1612 char *buf = malloc(size);
1613 int iret, ret;
1614 unsigned long off = 0;
1615 unsigned char hdr[50];
1616 int hdrlen;
1617 do {
1618 iret = read(fd, buf + off, size - off);
1619 if (iret > 0) {
1620 off += iret;
1621 if (off == size) {
1622 size *= 2;
1623 buf = realloc(buf, size);
1626 } while (iret > 0);
1627 if (iret < 0) {
1628 free(buf);
1629 return -1;
1631 if (!type)
1632 type = blob_type;
1633 if (write_object)
1634 ret = write_sha1_file(buf, off, type, sha1);
1635 else {
1636 write_sha1_file_prepare(buf, off, type, sha1, hdr, &hdrlen);
1637 ret = 0;
1639 free(buf);
1640 return ret;
1643 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1645 unsigned long size = st->st_size;
1646 void *buf;
1647 int ret;
1648 unsigned char hdr[50];
1649 int hdrlen;
1651 buf = "";
1652 if (size)
1653 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1654 close(fd);
1655 if (buf == MAP_FAILED)
1656 return -1;
1658 if (!type)
1659 type = blob_type;
1660 if (write_object)
1661 ret = write_sha1_file(buf, size, type, sha1);
1662 else {
1663 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1664 ret = 0;
1666 if (size)
1667 munmap(buf, size);
1668 return ret;
1671 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1673 int fd;
1674 char *target;
1676 switch (st->st_mode & S_IFMT) {
1677 case S_IFREG:
1678 fd = open(path, O_RDONLY);
1679 if (fd < 0)
1680 return error("open(\"%s\"): %s", path,
1681 strerror(errno));
1682 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1683 return error("%s: failed to insert into database",
1684 path);
1685 break;
1686 case S_IFLNK:
1687 target = xmalloc(st->st_size+1);
1688 if (readlink(path, target, st->st_size+1) != st->st_size) {
1689 char *errstr = strerror(errno);
1690 free(target);
1691 return error("readlink(\"%s\"): %s", path,
1692 errstr);
1694 if (!write_object) {
1695 unsigned char hdr[50];
1696 int hdrlen;
1697 write_sha1_file_prepare(target, st->st_size, blob_type,
1698 sha1, hdr, &hdrlen);
1699 } else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1700 return error("%s: failed to insert into database",
1701 path);
1702 free(target);
1703 break;
1704 default:
1705 return error("%s: unsupported file type", path);
1707 return 0;