Only call git-rerere if $GIT_DIR/rr-cache exists.
[git/jnareb-git.git] / sha1_file.c
blob3d11a9bfdc56c00056c2057af5e6455742ab31f5
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"
13 #ifndef O_NOATIME
14 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
15 #define O_NOATIME 01000000
16 #else
17 #define O_NOATIME 0
18 #endif
19 #endif
21 const unsigned char null_sha1[20] = { 0, };
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 int adjust_shared_perm(const char *path)
51 struct stat st;
52 int mode;
54 if (!shared_repository)
55 return 0;
56 if (lstat(path, &st) < 0)
57 return -1;
58 mode = st.st_mode;
59 if (mode & S_IRUSR)
60 mode |= S_IRGRP;
61 if (mode & S_IWUSR)
62 mode |= S_IWGRP;
63 if (mode & S_IXUSR)
64 mode |= S_IXGRP;
65 if (S_ISDIR(mode))
66 mode |= S_ISGID;
67 if (chmod(path, mode) < 0)
68 return -2;
69 return 0;
72 int safe_create_leading_directories(char *path)
74 char *pos = path;
75 struct stat st;
77 if (*pos == '/')
78 pos++;
80 while (pos) {
81 pos = strchr(pos, '/');
82 if (!pos)
83 break;
84 *pos = 0;
85 if (!stat(path, &st)) {
86 /* path exists */
87 if (!S_ISDIR(st.st_mode)) {
88 *pos = '/';
89 return -3;
92 else if (mkdir(path, 0777)) {
93 *pos = '/';
94 return -1;
96 else if (adjust_shared_perm(path)) {
97 *pos = '/';
98 return -2;
100 *pos++ = '/';
102 return 0;
105 char * sha1_to_hex(const unsigned char *sha1)
107 static char buffer[50];
108 static const char hex[] = "0123456789abcdef";
109 char *buf = buffer;
110 int i;
112 for (i = 0; i < 20; i++) {
113 unsigned int val = *sha1++;
114 *buf++ = hex[val >> 4];
115 *buf++ = hex[val & 0xf];
117 *buf = '\0';
119 return buffer;
122 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
124 int i;
125 for (i = 0; i < 20; i++) {
126 static char hex[] = "0123456789abcdef";
127 unsigned int val = sha1[i];
128 char *pos = pathbuf + i*2 + (i > 0);
129 *pos++ = hex[val >> 4];
130 *pos = hex[val & 0xf];
135 * NOTE! This returns a statically allocated buffer, so you have to be
136 * careful about using it. Do a "strdup()" if you need to save the
137 * filename.
139 * Also note that this returns the location for creating. Reading
140 * SHA1 file can happen from any alternate directory listed in the
141 * DB_ENVIRONMENT environment variable if it is not found in
142 * the primary object database.
144 char *sha1_file_name(const unsigned char *sha1)
146 static char *name, *base;
148 if (!base) {
149 const char *sha1_file_directory = get_object_directory();
150 int len = strlen(sha1_file_directory);
151 base = xmalloc(len + 60);
152 memcpy(base, sha1_file_directory, len);
153 memset(base+len, 0, 60);
154 base[len] = '/';
155 base[len+3] = '/';
156 name = base + len + 1;
158 fill_sha1_path(name, sha1);
159 return base;
162 char *sha1_pack_name(const unsigned char *sha1)
164 static const char hex[] = "0123456789abcdef";
165 static char *name, *base, *buf;
166 int i;
168 if (!base) {
169 const char *sha1_file_directory = get_object_directory();
170 int len = strlen(sha1_file_directory);
171 base = xmalloc(len + 60);
172 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
173 name = base + len + 11;
176 buf = name;
178 for (i = 0; i < 20; i++) {
179 unsigned int val = *sha1++;
180 *buf++ = hex[val >> 4];
181 *buf++ = hex[val & 0xf];
184 return base;
187 char *sha1_pack_index_name(const unsigned char *sha1)
189 static const char hex[] = "0123456789abcdef";
190 static char *name, *base, *buf;
191 int i;
193 if (!base) {
194 const char *sha1_file_directory = get_object_directory();
195 int len = strlen(sha1_file_directory);
196 base = xmalloc(len + 60);
197 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
198 name = base + len + 11;
201 buf = name;
203 for (i = 0; i < 20; i++) {
204 unsigned int val = *sha1++;
205 *buf++ = hex[val >> 4];
206 *buf++ = hex[val & 0xf];
209 return base;
212 struct alternate_object_database *alt_odb_list;
213 static struct alternate_object_database **alt_odb_tail;
216 * Prepare alternate object database registry.
218 * The variable alt_odb_list points at the list of struct
219 * alternate_object_database. The elements on this list come from
220 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
221 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
222 * whose contents is similar to that environment variable but can be
223 * LF separated. Its base points at a statically allocated buffer that
224 * contains "/the/directory/corresponding/to/.git/objects/...", while
225 * its name points just after the slash at the end of ".git/objects/"
226 * in the example above, and has enough space to hold 40-byte hex
227 * SHA1, an extra slash for the first level indirection, and the
228 * terminating NUL.
230 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
231 const char *relative_base)
233 const char *cp, *last;
234 struct alternate_object_database *ent;
235 const char *objdir = get_object_directory();
236 int base_len = -1;
238 last = alt;
239 while (last < ep) {
240 cp = last;
241 if (cp < ep && *cp == '#') {
242 while (cp < ep && *cp != sep)
243 cp++;
244 last = cp + 1;
245 continue;
247 for ( ; cp < ep && *cp != sep; cp++)
249 if (last != cp) {
250 struct alternate_object_database *alt;
251 /* 43 = 40-byte + 2 '/' + terminating NUL */
252 int pfxlen = cp - last;
253 int entlen = pfxlen + 43;
255 if (*last != '/' && relative_base) {
256 /* Relative alt-odb */
257 if (base_len < 0)
258 base_len = strlen(relative_base) + 1;
259 entlen += base_len;
260 pfxlen += base_len;
262 ent = xmalloc(sizeof(*ent) + entlen);
264 if (*last != '/' && relative_base) {
265 memcpy(ent->base, relative_base, base_len - 1);
266 ent->base[base_len - 1] = '/';
267 memcpy(ent->base + base_len,
268 last, cp - last);
270 else
271 memcpy(ent->base, last, pfxlen);
272 ent->name = ent->base + pfxlen + 1;
273 ent->base[pfxlen] = ent->base[pfxlen + 3] = '/';
274 ent->base[entlen-1] = 0;
276 /* Prevent the common mistake of listing the same
277 * thing twice, or object directory itself.
279 for (alt = alt_odb_list; alt; alt = alt->next)
280 if (!memcmp(ent->base, alt->base, pfxlen))
281 goto bad;
282 if (!memcmp(ent->base, objdir, pfxlen)) {
283 bad:
284 free(ent);
286 else {
287 *alt_odb_tail = ent;
288 alt_odb_tail = &(ent->next);
289 ent->next = NULL;
292 while (cp < ep && *cp == sep)
293 cp++;
294 last = cp;
298 void prepare_alt_odb(void)
300 char path[PATH_MAX];
301 char *map;
302 int fd;
303 struct stat st;
304 char *alt;
306 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
307 if (!alt) alt = "";
309 if (alt_odb_tail)
310 return;
311 alt_odb_tail = &alt_odb_list;
312 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL);
314 sprintf(path, "%s/info/alternates", get_object_directory());
315 fd = open(path, O_RDONLY);
316 if (fd < 0)
317 return;
318 if (fstat(fd, &st) || (st.st_size == 0)) {
319 close(fd);
320 return;
322 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
323 close(fd);
324 if (map == MAP_FAILED)
325 return;
327 link_alt_odb_entries(map, map + st.st_size, '\n',
328 get_object_directory());
329 munmap(map, st.st_size);
332 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
334 char *name = sha1_file_name(sha1);
335 struct alternate_object_database *alt;
337 if (!stat(name, st))
338 return name;
339 prepare_alt_odb();
340 for (alt = alt_odb_list; alt; alt = alt->next) {
341 name = alt->name;
342 fill_sha1_path(name, sha1);
343 if (!stat(alt->base, st))
344 return alt->base;
346 return NULL;
349 #define PACK_MAX_SZ (1<<26)
350 static int pack_used_ctr;
351 static unsigned long pack_mapped;
352 struct packed_git *packed_git;
354 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
355 void **idx_map_)
357 void *idx_map;
358 unsigned int *index;
359 unsigned long idx_size;
360 int nr, i;
361 int fd = open(path, O_RDONLY);
362 struct stat st;
363 if (fd < 0)
364 return -1;
365 if (fstat(fd, &st)) {
366 close(fd);
367 return -1;
369 idx_size = st.st_size;
370 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
371 close(fd);
372 if (idx_map == MAP_FAILED)
373 return -1;
375 index = idx_map;
376 *idx_map_ = idx_map;
377 *idx_size_ = idx_size;
379 /* check index map */
380 if (idx_size < 4*256 + 20 + 20)
381 return error("index file too small");
382 nr = 0;
383 for (i = 0; i < 256; i++) {
384 unsigned int n = ntohl(index[i]);
385 if (n < nr)
386 return error("non-monotonic index");
387 nr = n;
391 * Total size:
392 * - 256 index entries 4 bytes each
393 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
394 * - 20-byte SHA1 of the packfile
395 * - 20-byte SHA1 file checksum
397 if (idx_size != 4*256 + nr * 24 + 20 + 20)
398 return error("wrong index file size");
400 return 0;
403 static int unuse_one_packed_git(void)
405 struct packed_git *p, *lru = NULL;
407 for (p = packed_git; p; p = p->next) {
408 if (p->pack_use_cnt || !p->pack_base)
409 continue;
410 if (!lru || p->pack_last_used < lru->pack_last_used)
411 lru = p;
413 if (!lru)
414 return 0;
415 munmap(lru->pack_base, lru->pack_size);
416 lru->pack_base = NULL;
417 return 1;
420 void unuse_packed_git(struct packed_git *p)
422 p->pack_use_cnt--;
425 int use_packed_git(struct packed_git *p)
427 if (!p->pack_size) {
428 struct stat st;
429 // We created the struct before we had the pack
430 stat(p->pack_name, &st);
431 if (!S_ISREG(st.st_mode))
432 die("packfile %s not a regular file", p->pack_name);
433 p->pack_size = st.st_size;
435 if (!p->pack_base) {
436 int fd;
437 struct stat st;
438 void *map;
440 pack_mapped += p->pack_size;
441 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
442 ; /* nothing */
443 fd = open(p->pack_name, O_RDONLY);
444 if (fd < 0)
445 die("packfile %s cannot be opened", p->pack_name);
446 if (fstat(fd, &st)) {
447 close(fd);
448 die("packfile %s cannot be opened", p->pack_name);
450 if (st.st_size != p->pack_size)
451 die("packfile %s size mismatch.", p->pack_name);
452 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
453 close(fd);
454 if (map == MAP_FAILED)
455 die("packfile %s cannot be mapped.", p->pack_name);
456 p->pack_base = map;
458 /* Check if the pack file matches with the index file.
459 * this is cheap.
461 if (memcmp((char*)(p->index_base) + p->index_size - 40,
462 p->pack_base + p->pack_size - 20, 20)) {
464 die("packfile %s does not match index.", p->pack_name);
467 p->pack_last_used = pack_used_ctr++;
468 p->pack_use_cnt++;
469 return 0;
472 struct packed_git *add_packed_git(char *path, int path_len, int local)
474 struct stat st;
475 struct packed_git *p;
476 unsigned long idx_size;
477 void *idx_map;
478 unsigned char sha1[20];
480 if (check_packed_git_idx(path, &idx_size, &idx_map))
481 return NULL;
483 /* do we have a corresponding .pack file? */
484 strcpy(path + path_len - 4, ".pack");
485 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
486 munmap(idx_map, idx_size);
487 return NULL;
489 /* ok, it looks sane as far as we can check without
490 * actually mapping the pack file.
492 p = xmalloc(sizeof(*p) + path_len + 2);
493 strcpy(p->pack_name, path);
494 p->index_size = idx_size;
495 p->pack_size = st.st_size;
496 p->index_base = idx_map;
497 p->next = NULL;
498 p->pack_base = NULL;
499 p->pack_last_used = 0;
500 p->pack_use_cnt = 0;
501 p->pack_local = local;
502 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
503 memcpy(p->sha1, sha1, 20);
504 return p;
507 struct packed_git *parse_pack_index(unsigned char *sha1)
509 char *path = sha1_pack_index_name(sha1);
510 return parse_pack_index_file(sha1, path);
513 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
515 struct packed_git *p;
516 unsigned long idx_size;
517 void *idx_map;
518 char *path;
520 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
521 return NULL;
523 path = sha1_pack_name(sha1);
525 p = xmalloc(sizeof(*p) + strlen(path) + 2);
526 strcpy(p->pack_name, path);
527 p->index_size = idx_size;
528 p->pack_size = 0;
529 p->index_base = idx_map;
530 p->next = NULL;
531 p->pack_base = NULL;
532 p->pack_last_used = 0;
533 p->pack_use_cnt = 0;
534 memcpy(p->sha1, sha1, 20);
535 return p;
538 void install_packed_git(struct packed_git *pack)
540 pack->next = packed_git;
541 packed_git = pack;
544 static void prepare_packed_git_one(char *objdir, int local)
546 char path[PATH_MAX];
547 int len;
548 DIR *dir;
549 struct dirent *de;
551 sprintf(path, "%s/pack", objdir);
552 len = strlen(path);
553 dir = opendir(path);
554 if (!dir)
555 return;
556 path[len++] = '/';
557 while ((de = readdir(dir)) != NULL) {
558 int namelen = strlen(de->d_name);
559 struct packed_git *p;
561 if (strcmp(de->d_name + namelen - 4, ".idx"))
562 continue;
564 /* we have .idx. Is it a file we can map? */
565 strcpy(path + len, de->d_name);
566 p = add_packed_git(path, len + namelen, local);
567 if (!p)
568 continue;
569 p->next = packed_git;
570 packed_git = p;
572 closedir(dir);
575 void prepare_packed_git(void)
577 static int run_once = 0;
578 struct alternate_object_database *alt;
580 if (run_once)
581 return;
582 prepare_packed_git_one(get_object_directory(), 1);
583 prepare_alt_odb();
584 for (alt = alt_odb_list; alt; alt = alt->next) {
585 alt->name[-1] = 0;
586 prepare_packed_git_one(alt->base, 0);
587 alt->name[-1] = '/';
589 run_once = 1;
592 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
594 char header[100];
595 unsigned char real_sha1[20];
596 SHA_CTX c;
598 SHA1_Init(&c);
599 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
600 SHA1_Update(&c, map, size);
601 SHA1_Final(real_sha1, &c);
602 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
605 static void *map_sha1_file_internal(const unsigned char *sha1,
606 unsigned long *size)
608 struct stat st;
609 void *map;
610 int fd;
611 char *filename = find_sha1_file(sha1, &st);
613 if (!filename) {
614 return NULL;
617 fd = open(filename, O_RDONLY | sha1_file_open_flag);
618 if (fd < 0) {
619 /* See if it works without O_NOATIME */
620 switch (sha1_file_open_flag) {
621 default:
622 fd = open(filename, O_RDONLY);
623 if (fd >= 0)
624 break;
625 /* Fallthrough */
626 case 0:
627 return NULL;
630 /* If it failed once, it will probably fail again.
631 * Stop using O_NOATIME
633 sha1_file_open_flag = 0;
635 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
636 close(fd);
637 if (map == MAP_FAILED)
638 return NULL;
639 *size = st.st_size;
640 return map;
643 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
645 /* Get the data stream */
646 memset(stream, 0, sizeof(*stream));
647 stream->next_in = map;
648 stream->avail_in = mapsize;
649 stream->next_out = buffer;
650 stream->avail_out = size;
652 inflateInit(stream);
653 return inflate(stream, 0);
656 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
658 int bytes = strlen(buffer) + 1;
659 unsigned char *buf = xmalloc(1+size);
661 memcpy(buf, buffer + bytes, stream->total_out - bytes);
662 bytes = stream->total_out - bytes;
663 if (bytes < size) {
664 stream->next_out = buf + bytes;
665 stream->avail_out = size - bytes;
666 while (inflate(stream, Z_FINISH) == Z_OK)
667 /* nothing */;
669 buf[size] = 0;
670 inflateEnd(stream);
671 return buf;
675 * We used to just use "sscanf()", but that's actually way
676 * too permissive for what we want to check. So do an anal
677 * object header parse by hand.
679 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
681 int i;
682 unsigned long size;
685 * The type can be at most ten bytes (including the
686 * terminating '\0' that we add), and is followed by
687 * a space.
689 i = 10;
690 for (;;) {
691 char c = *hdr++;
692 if (c == ' ')
693 break;
694 if (!--i)
695 return -1;
696 *type++ = c;
698 *type = 0;
701 * The length must follow immediately, and be in canonical
702 * decimal format (ie "010" is not valid).
704 size = *hdr++ - '0';
705 if (size > 9)
706 return -1;
707 if (size) {
708 for (;;) {
709 unsigned long c = *hdr - '0';
710 if (c > 9)
711 break;
712 hdr++;
713 size = size * 10 + c;
716 *sizep = size;
719 * The length must be followed by a zero byte
721 return *hdr ? -1 : 0;
724 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
726 int ret;
727 z_stream stream;
728 char hdr[8192];
730 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
731 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
732 return NULL;
734 return unpack_sha1_rest(&stream, hdr, *size);
737 /* forward declaration for a mutually recursive function */
738 static int packed_object_info(struct pack_entry *entry,
739 char *type, unsigned long *sizep);
741 static int packed_delta_info(unsigned char *base_sha1,
742 unsigned long delta_size,
743 unsigned long left,
744 char *type,
745 unsigned long *sizep,
746 struct packed_git *p)
748 struct pack_entry base_ent;
750 if (left < 20)
751 die("truncated pack file");
753 /* The base entry _must_ be in the same pack */
754 if (!find_pack_entry_one(base_sha1, &base_ent, p))
755 die("failed to find delta-pack base object %s",
756 sha1_to_hex(base_sha1));
758 /* We choose to only get the type of the base object and
759 * ignore potentially corrupt pack file that expects the delta
760 * based on a base with a wrong size. This saves tons of
761 * inflate() calls.
764 if (packed_object_info(&base_ent, type, NULL))
765 die("cannot get info for delta-pack base");
767 if (sizep) {
768 const unsigned char *data;
769 unsigned char delta_head[64];
770 unsigned long result_size;
771 z_stream stream;
772 int st;
774 memset(&stream, 0, sizeof(stream));
776 data = stream.next_in = base_sha1 + 20;
777 stream.avail_in = left - 20;
778 stream.next_out = delta_head;
779 stream.avail_out = sizeof(delta_head);
781 inflateInit(&stream);
782 st = inflate(&stream, Z_FINISH);
783 inflateEnd(&stream);
784 if ((st != Z_STREAM_END) &&
785 stream.total_out != sizeof(delta_head))
786 die("delta data unpack-initial failed");
788 /* Examine the initial part of the delta to figure out
789 * the result size.
791 data = delta_head;
792 get_delta_hdr_size(&data); /* ignore base size */
794 /* Read the result size */
795 result_size = get_delta_hdr_size(&data);
796 *sizep = result_size;
798 return 0;
801 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
802 enum object_type *type, unsigned long *sizep)
804 unsigned shift;
805 unsigned char *pack, c;
806 unsigned long size;
808 if (offset >= p->pack_size)
809 die("object offset outside of pack file");
811 pack = p->pack_base + offset;
812 c = *pack++;
813 offset++;
814 *type = (c >> 4) & 7;
815 size = c & 15;
816 shift = 4;
817 while (c & 0x80) {
818 if (offset >= p->pack_size)
819 die("object offset outside of pack file");
820 c = *pack++;
821 offset++;
822 size += (c & 0x7f) << shift;
823 shift += 7;
825 *sizep = size;
826 return offset;
829 void packed_object_info_detail(struct pack_entry *e,
830 char *type,
831 unsigned long *size,
832 unsigned long *store_size,
833 int *delta_chain_length,
834 unsigned char *base_sha1)
836 struct packed_git *p = e->p;
837 unsigned long offset, left;
838 unsigned char *pack;
839 enum object_type kind;
841 offset = unpack_object_header(p, e->offset, &kind, size);
842 pack = p->pack_base + offset;
843 left = p->pack_size - offset;
844 if (kind != OBJ_DELTA)
845 *delta_chain_length = 0;
846 else {
847 int chain_length = 0;
848 memcpy(base_sha1, pack, 20);
849 do {
850 struct pack_entry base_ent;
851 unsigned long junk;
853 find_pack_entry_one(pack, &base_ent, p);
854 offset = unpack_object_header(p, base_ent.offset,
855 &kind, &junk);
856 pack = p->pack_base + offset;
857 chain_length++;
858 } while (kind == OBJ_DELTA);
859 *delta_chain_length = chain_length;
861 switch (kind) {
862 case OBJ_COMMIT:
863 strcpy(type, "commit");
864 break;
865 case OBJ_TREE:
866 strcpy(type, "tree");
867 break;
868 case OBJ_BLOB:
869 strcpy(type, "blob");
870 break;
871 case OBJ_TAG:
872 strcpy(type, "tag");
873 break;
874 default:
875 die("corrupted pack file %s containing object of kind %d",
876 p->pack_name, kind);
878 *store_size = 0; /* notyet */
881 static int packed_object_info(struct pack_entry *entry,
882 char *type, unsigned long *sizep)
884 struct packed_git *p = entry->p;
885 unsigned long offset, size, left;
886 unsigned char *pack;
887 enum object_type kind;
888 int retval;
890 if (use_packed_git(p))
891 die("cannot map packed file");
893 offset = unpack_object_header(p, entry->offset, &kind, &size);
894 pack = p->pack_base + offset;
895 left = p->pack_size - offset;
897 switch (kind) {
898 case OBJ_DELTA:
899 retval = packed_delta_info(pack, size, left, type, sizep, p);
900 unuse_packed_git(p);
901 return retval;
902 case OBJ_COMMIT:
903 strcpy(type, "commit");
904 break;
905 case OBJ_TREE:
906 strcpy(type, "tree");
907 break;
908 case OBJ_BLOB:
909 strcpy(type, "blob");
910 break;
911 case OBJ_TAG:
912 strcpy(type, "tag");
913 break;
914 default:
915 die("corrupted pack file %s containing object of kind %d",
916 p->pack_name, kind);
918 if (sizep)
919 *sizep = size;
920 unuse_packed_git(p);
921 return 0;
924 /* forward declaration for a mutually recursive function */
925 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
927 static void *unpack_delta_entry(unsigned char *base_sha1,
928 unsigned long delta_size,
929 unsigned long left,
930 char *type,
931 unsigned long *sizep,
932 struct packed_git *p)
934 struct pack_entry base_ent;
935 void *data, *delta_data, *result, *base;
936 unsigned long data_size, result_size, base_size;
937 z_stream stream;
938 int st;
940 if (left < 20)
941 die("truncated pack file");
942 data = base_sha1 + 20;
943 data_size = left - 20;
944 delta_data = xmalloc(delta_size);
946 memset(&stream, 0, sizeof(stream));
948 stream.next_in = data;
949 stream.avail_in = data_size;
950 stream.next_out = delta_data;
951 stream.avail_out = delta_size;
953 inflateInit(&stream);
954 st = inflate(&stream, Z_FINISH);
955 inflateEnd(&stream);
956 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
957 die("delta data unpack failed");
959 /* The base entry _must_ be in the same pack */
960 if (!find_pack_entry_one(base_sha1, &base_ent, p))
961 die("failed to find delta-pack base object %s",
962 sha1_to_hex(base_sha1));
963 base = unpack_entry_gently(&base_ent, type, &base_size);
964 if (!base)
965 die("failed to read delta-pack base object %s",
966 sha1_to_hex(base_sha1));
967 result = patch_delta(base, base_size,
968 delta_data, delta_size,
969 &result_size);
970 if (!result)
971 die("failed to apply delta");
972 free(delta_data);
973 free(base);
974 *sizep = result_size;
975 return result;
978 static void *unpack_non_delta_entry(unsigned char *data,
979 unsigned long size,
980 unsigned long left)
982 int st;
983 z_stream stream;
984 unsigned char *buffer;
986 buffer = xmalloc(size + 1);
987 buffer[size] = 0;
988 memset(&stream, 0, sizeof(stream));
989 stream.next_in = data;
990 stream.avail_in = left;
991 stream.next_out = buffer;
992 stream.avail_out = size;
994 inflateInit(&stream);
995 st = inflate(&stream, Z_FINISH);
996 inflateEnd(&stream);
997 if ((st != Z_STREAM_END) || stream.total_out != size) {
998 free(buffer);
999 return NULL;
1002 return buffer;
1005 static void *unpack_entry(struct pack_entry *entry,
1006 char *type, unsigned long *sizep)
1008 struct packed_git *p = entry->p;
1009 void *retval;
1011 if (use_packed_git(p))
1012 die("cannot map packed file");
1013 retval = unpack_entry_gently(entry, type, sizep);
1014 unuse_packed_git(p);
1015 if (!retval)
1016 die("corrupted pack file %s", p->pack_name);
1017 return retval;
1020 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1021 void *unpack_entry_gently(struct pack_entry *entry,
1022 char *type, unsigned long *sizep)
1024 struct packed_git *p = entry->p;
1025 unsigned long offset, size, left;
1026 unsigned char *pack;
1027 enum object_type kind;
1028 void *retval;
1030 offset = unpack_object_header(p, entry->offset, &kind, &size);
1031 pack = p->pack_base + offset;
1032 left = p->pack_size - offset;
1033 switch (kind) {
1034 case OBJ_DELTA:
1035 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1036 return retval;
1037 case OBJ_COMMIT:
1038 strcpy(type, "commit");
1039 break;
1040 case OBJ_TREE:
1041 strcpy(type, "tree");
1042 break;
1043 case OBJ_BLOB:
1044 strcpy(type, "blob");
1045 break;
1046 case OBJ_TAG:
1047 strcpy(type, "tag");
1048 break;
1049 default:
1050 return NULL;
1052 *sizep = size;
1053 retval = unpack_non_delta_entry(pack, size, left);
1054 return retval;
1057 int num_packed_objects(const struct packed_git *p)
1059 /* See check_packed_git_idx() */
1060 return (p->index_size - 20 - 20 - 4*256) / 24;
1063 int nth_packed_object_sha1(const struct packed_git *p, int n,
1064 unsigned char* sha1)
1066 void *index = p->index_base + 256;
1067 if (n < 0 || num_packed_objects(p) <= n)
1068 return -1;
1069 memcpy(sha1, (index + 24 * n + 4), 20);
1070 return 0;
1073 int find_pack_entry_one(const unsigned char *sha1,
1074 struct pack_entry *e, struct packed_git *p)
1076 unsigned int *level1_ofs = p->index_base;
1077 int hi = ntohl(level1_ofs[*sha1]);
1078 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1079 void *index = p->index_base + 256;
1081 do {
1082 int mi = (lo + hi) / 2;
1083 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1084 if (!cmp) {
1085 e->offset = ntohl(*((int*)(index + 24 * mi)));
1086 memcpy(e->sha1, sha1, 20);
1087 e->p = p;
1088 return 1;
1090 if (cmp > 0)
1091 hi = mi;
1092 else
1093 lo = mi+1;
1094 } while (lo < hi);
1095 return 0;
1098 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1100 struct packed_git *p;
1101 prepare_packed_git();
1103 for (p = packed_git; p; p = p->next) {
1104 if (find_pack_entry_one(sha1, e, p))
1105 return 1;
1107 return 0;
1110 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1111 struct packed_git *packs)
1113 struct packed_git *p;
1114 struct pack_entry e;
1116 for (p = packs; p; p = p->next) {
1117 if (find_pack_entry_one(sha1, &e, p))
1118 return p;
1120 return NULL;
1124 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1126 int status;
1127 unsigned long mapsize, size;
1128 void *map;
1129 z_stream stream;
1130 char hdr[128];
1132 map = map_sha1_file_internal(sha1, &mapsize);
1133 if (!map) {
1134 struct pack_entry e;
1136 if (!find_pack_entry(sha1, &e))
1137 return error("unable to find %s", sha1_to_hex(sha1));
1138 return packed_object_info(&e, type, sizep);
1140 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1141 status = error("unable to unpack %s header",
1142 sha1_to_hex(sha1));
1143 if (parse_sha1_header(hdr, type, &size) < 0)
1144 status = error("unable to parse %s header", sha1_to_hex(sha1));
1145 else {
1146 status = 0;
1147 if (sizep)
1148 *sizep = size;
1150 inflateEnd(&stream);
1151 munmap(map, mapsize);
1152 return status;
1155 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1157 struct pack_entry e;
1159 if (!find_pack_entry(sha1, &e)) {
1160 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1161 return NULL;
1163 return unpack_entry(&e, type, size);
1166 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1168 unsigned long mapsize;
1169 void *map, *buf;
1170 struct pack_entry e;
1172 if (find_pack_entry(sha1, &e))
1173 return read_packed_sha1(sha1, type, size);
1174 map = map_sha1_file_internal(sha1, &mapsize);
1175 if (map) {
1176 buf = unpack_sha1_file(map, mapsize, type, size);
1177 munmap(map, mapsize);
1178 return buf;
1180 return NULL;
1183 void *read_object_with_reference(const unsigned char *sha1,
1184 const char *required_type,
1185 unsigned long *size,
1186 unsigned char *actual_sha1_return)
1188 char type[20];
1189 void *buffer;
1190 unsigned long isize;
1191 unsigned char actual_sha1[20];
1193 memcpy(actual_sha1, sha1, 20);
1194 while (1) {
1195 int ref_length = -1;
1196 const char *ref_type = NULL;
1198 buffer = read_sha1_file(actual_sha1, type, &isize);
1199 if (!buffer)
1200 return NULL;
1201 if (!strcmp(type, required_type)) {
1202 *size = isize;
1203 if (actual_sha1_return)
1204 memcpy(actual_sha1_return, actual_sha1, 20);
1205 return buffer;
1207 /* Handle references */
1208 else if (!strcmp(type, "commit"))
1209 ref_type = "tree ";
1210 else if (!strcmp(type, "tag"))
1211 ref_type = "object ";
1212 else {
1213 free(buffer);
1214 return NULL;
1216 ref_length = strlen(ref_type);
1218 if (memcmp(buffer, ref_type, ref_length) ||
1219 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1220 free(buffer);
1221 return NULL;
1223 free(buffer);
1224 /* Now we have the ID of the referred-to object in
1225 * actual_sha1. Check again. */
1229 char *write_sha1_file_prepare(void *buf,
1230 unsigned long len,
1231 const char *type,
1232 unsigned char *sha1,
1233 unsigned char *hdr,
1234 int *hdrlen)
1236 SHA_CTX c;
1238 /* Generate the header */
1239 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1241 /* Sha1.. */
1242 SHA1_Init(&c);
1243 SHA1_Update(&c, hdr, *hdrlen);
1244 SHA1_Update(&c, buf, len);
1245 SHA1_Final(sha1, &c);
1247 return sha1_file_name(sha1);
1251 * Link the tempfile to the final place, possibly creating the
1252 * last directory level as you do so.
1254 * Returns the errno on failure, 0 on success.
1256 static int link_temp_to_file(const char *tmpfile, char *filename)
1258 int ret;
1260 if (!link(tmpfile, filename))
1261 return 0;
1264 * Try to mkdir the last path component if that failed
1265 * with an ENOENT.
1267 * Re-try the "link()" regardless of whether the mkdir
1268 * succeeds, since a race might mean that somebody
1269 * else succeeded.
1271 ret = errno;
1272 if (ret == ENOENT) {
1273 char *dir = strrchr(filename, '/');
1274 if (dir) {
1275 *dir = 0;
1276 mkdir(filename, 0777);
1277 if (adjust_shared_perm(filename))
1278 return -2;
1279 *dir = '/';
1280 if (!link(tmpfile, filename))
1281 return 0;
1282 ret = errno;
1285 return ret;
1289 * Move the just written object into its final resting place
1291 int move_temp_to_file(const char *tmpfile, char *filename)
1293 int ret = link_temp_to_file(tmpfile, filename);
1296 * Coda hack - coda doesn't like cross-directory links,
1297 * so we fall back to a rename, which will mean that it
1298 * won't be able to check collisions, but that's not a
1299 * big deal.
1301 * The same holds for FAT formatted media.
1303 * When this succeeds, we just return 0. We have nothing
1304 * left to unlink.
1306 if (ret && ret != EEXIST) {
1307 if (!rename(tmpfile, filename))
1308 return 0;
1309 ret = errno;
1311 unlink(tmpfile);
1312 if (ret) {
1313 if (ret != EEXIST) {
1314 fprintf(stderr, "unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1315 return -1;
1317 /* FIXME!!! Collision check here ? */
1320 return 0;
1323 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1325 int size;
1326 unsigned char *compressed;
1327 z_stream stream;
1328 unsigned char sha1[20];
1329 char *filename;
1330 static char tmpfile[PATH_MAX];
1331 unsigned char hdr[50];
1332 int fd, hdrlen;
1334 /* Normally if we have it in the pack then we do not bother writing
1335 * it out into .git/objects/??/?{38} file.
1337 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1338 if (returnsha1)
1339 memcpy(returnsha1, sha1, 20);
1340 if (has_sha1_file(sha1))
1341 return 0;
1342 fd = open(filename, O_RDONLY);
1343 if (fd >= 0) {
1345 * FIXME!!! We might do collision checking here, but we'd
1346 * need to uncompress the old file and check it. Later.
1348 close(fd);
1349 return 0;
1352 if (errno != ENOENT) {
1353 fprintf(stderr, "sha1 file %s: %s\n", filename, strerror(errno));
1354 return -1;
1357 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1359 fd = mkstemp(tmpfile);
1360 if (fd < 0) {
1361 fprintf(stderr, "unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1362 return -1;
1365 /* Set it up */
1366 memset(&stream, 0, sizeof(stream));
1367 deflateInit(&stream, Z_BEST_COMPRESSION);
1368 size = deflateBound(&stream, len+hdrlen);
1369 compressed = xmalloc(size);
1371 /* Compress it */
1372 stream.next_out = compressed;
1373 stream.avail_out = size;
1375 /* First header.. */
1376 stream.next_in = hdr;
1377 stream.avail_in = hdrlen;
1378 while (deflate(&stream, 0) == Z_OK)
1379 /* nothing */;
1381 /* Then the data itself.. */
1382 stream.next_in = buf;
1383 stream.avail_in = len;
1384 while (deflate(&stream, Z_FINISH) == Z_OK)
1385 /* nothing */;
1386 deflateEnd(&stream);
1387 size = stream.total_out;
1389 if (write(fd, compressed, size) != size)
1390 die("unable to write file");
1391 fchmod(fd, 0444);
1392 close(fd);
1393 free(compressed);
1395 return move_temp_to_file(tmpfile, filename);
1398 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1400 ssize_t size;
1401 unsigned long objsize;
1402 int posn = 0;
1403 void *map = map_sha1_file_internal(sha1, &objsize);
1404 void *buf = map;
1405 void *temp_obj = NULL;
1406 z_stream stream;
1408 if (!buf) {
1409 unsigned char *unpacked;
1410 unsigned long len;
1411 char type[20];
1412 char hdr[50];
1413 int hdrlen;
1414 // need to unpack and recompress it by itself
1415 unpacked = read_packed_sha1(sha1, type, &len);
1417 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1419 /* Set it up */
1420 memset(&stream, 0, sizeof(stream));
1421 deflateInit(&stream, Z_BEST_COMPRESSION);
1422 size = deflateBound(&stream, len + hdrlen);
1423 temp_obj = buf = xmalloc(size);
1425 /* Compress it */
1426 stream.next_out = buf;
1427 stream.avail_out = size;
1429 /* First header.. */
1430 stream.next_in = (void *)hdr;
1431 stream.avail_in = hdrlen;
1432 while (deflate(&stream, 0) == Z_OK)
1433 /* nothing */;
1435 /* Then the data itself.. */
1436 stream.next_in = unpacked;
1437 stream.avail_in = len;
1438 while (deflate(&stream, Z_FINISH) == Z_OK)
1439 /* nothing */;
1440 deflateEnd(&stream);
1441 free(unpacked);
1443 objsize = stream.total_out;
1446 do {
1447 size = write(fd, buf + posn, objsize - posn);
1448 if (size <= 0) {
1449 if (!size) {
1450 fprintf(stderr, "write closed\n");
1451 } else {
1452 perror("write ");
1454 return -1;
1456 posn += size;
1457 } while (posn < objsize);
1459 if (map)
1460 munmap(map, objsize);
1461 if (temp_obj)
1462 free(temp_obj);
1464 return 0;
1467 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1468 size_t bufsize, size_t *bufposn)
1470 char tmpfile[PATH_MAX];
1471 int local;
1472 z_stream stream;
1473 unsigned char real_sha1[20];
1474 unsigned char discard[4096];
1475 int ret;
1476 SHA_CTX c;
1478 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1480 local = mkstemp(tmpfile);
1481 if (local < 0)
1482 return error("Couldn't open %s for %s\n", tmpfile, sha1_to_hex(sha1));
1484 memset(&stream, 0, sizeof(stream));
1486 inflateInit(&stream);
1488 SHA1_Init(&c);
1490 do {
1491 ssize_t size;
1492 if (*bufposn) {
1493 stream.avail_in = *bufposn;
1494 stream.next_in = (unsigned char *) buffer;
1495 do {
1496 stream.next_out = discard;
1497 stream.avail_out = sizeof(discard);
1498 ret = inflate(&stream, Z_SYNC_FLUSH);
1499 SHA1_Update(&c, discard, sizeof(discard) -
1500 stream.avail_out);
1501 } while (stream.avail_in && ret == Z_OK);
1502 write(local, buffer, *bufposn - stream.avail_in);
1503 memmove(buffer, buffer + *bufposn - stream.avail_in,
1504 stream.avail_in);
1505 *bufposn = stream.avail_in;
1506 if (ret != Z_OK)
1507 break;
1509 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1510 if (size <= 0) {
1511 close(local);
1512 unlink(tmpfile);
1513 if (!size)
1514 return error("Connection closed?");
1515 perror("Reading from connection");
1516 return -1;
1518 *bufposn += size;
1519 } while (1);
1520 inflateEnd(&stream);
1522 close(local);
1523 SHA1_Final(real_sha1, &c);
1524 if (ret != Z_STREAM_END) {
1525 unlink(tmpfile);
1526 return error("File %s corrupted", sha1_to_hex(sha1));
1528 if (memcmp(sha1, real_sha1, 20)) {
1529 unlink(tmpfile);
1530 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1533 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1536 int has_pack_index(const unsigned char *sha1)
1538 struct stat st;
1539 if (stat(sha1_pack_index_name(sha1), &st))
1540 return 0;
1541 return 1;
1544 int has_pack_file(const unsigned char *sha1)
1546 struct stat st;
1547 if (stat(sha1_pack_name(sha1), &st))
1548 return 0;
1549 return 1;
1552 int has_sha1_pack(const unsigned char *sha1)
1554 struct pack_entry e;
1555 return find_pack_entry(sha1, &e);
1558 int has_sha1_file(const unsigned char *sha1)
1560 struct stat st;
1561 struct pack_entry e;
1563 if (find_pack_entry(sha1, &e))
1564 return 1;
1565 return find_sha1_file(sha1, &st) ? 1 : 0;
1568 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1570 unsigned long size = 4096;
1571 char *buf = malloc(size);
1572 int iret, ret;
1573 unsigned long off = 0;
1574 unsigned char hdr[50];
1575 int hdrlen;
1576 do {
1577 iret = read(fd, buf + off, size - off);
1578 if (iret > 0) {
1579 off += iret;
1580 if (off == size) {
1581 size *= 2;
1582 buf = realloc(buf, size);
1585 } while (iret > 0);
1586 if (iret < 0) {
1587 free(buf);
1588 return -1;
1590 if (!type)
1591 type = "blob";
1592 if (write_object)
1593 ret = write_sha1_file(buf, off, type, sha1);
1594 else {
1595 write_sha1_file_prepare(buf, off, type, sha1, hdr, &hdrlen);
1596 ret = 0;
1598 free(buf);
1599 return ret;
1602 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1604 unsigned long size = st->st_size;
1605 void *buf;
1606 int ret;
1607 unsigned char hdr[50];
1608 int hdrlen;
1610 buf = "";
1611 if (size)
1612 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1613 close(fd);
1614 if (buf == MAP_FAILED)
1615 return -1;
1617 if (!type)
1618 type = "blob";
1619 if (write_object)
1620 ret = write_sha1_file(buf, size, type, sha1);
1621 else {
1622 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1623 ret = 0;
1625 if (size)
1626 munmap(buf, size);
1627 return ret;
1630 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1632 int fd;
1633 char *target;
1635 switch (st->st_mode & S_IFMT) {
1636 case S_IFREG:
1637 fd = open(path, O_RDONLY);
1638 if (fd < 0)
1639 return error("open(\"%s\"): %s", path,
1640 strerror(errno));
1641 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1642 return error("%s: failed to insert into database",
1643 path);
1644 break;
1645 case S_IFLNK:
1646 target = xmalloc(st->st_size+1);
1647 if (readlink(path, target, st->st_size+1) != st->st_size) {
1648 char *errstr = strerror(errno);
1649 free(target);
1650 return error("readlink(\"%s\"): %s", path,
1651 errstr);
1653 if (!write_object) {
1654 unsigned char hdr[50];
1655 int hdrlen;
1656 write_sha1_file_prepare(target, st->st_size, "blob",
1657 sha1, hdr, &hdrlen);
1658 } else if (write_sha1_file(target, st->st_size, "blob", sha1))
1659 return error("%s: failed to insert into database",
1660 path);
1661 free(target);
1662 break;
1663 default:
1664 return error("%s: unsupported file type", path);
1666 return 0;