[PATCH] fix git-checkout-cache segfault on parisc
[git/spearce.git] / sha1_file.c
blob8d189d4919f83c243c306973285cefb5b55e2f19
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 = gitenv(GIT_DIR_ENVIRONMENT);
54 if (!git_dir)
55 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
56 git_object_dir = gitenv(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 = gitenv(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 = gitenv(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;
228 * Prepare alternate object database registry.
229 * alt_odb points at an array of struct alternate_object_database.
230 * This array is terminated with an element that has both its base
231 * and name set to NULL. alt_odb[n] comes from n'th non-empty
232 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
233 * variable, and its base points at a statically allocated buffer
234 * that contains "/the/directory/corresponding/to/.git/objects/...",
235 * while its name points just after the slash at the end of
236 * ".git/objects/" in the example above, and has enough space to hold
237 * 40-byte hex SHA1, an extra slash for the first level indirection,
238 * and the terminating NUL.
239 * This function allocates the alt_odb array and all the strings
240 * pointed by base fields of the array elements with one xmalloc();
241 * the string pool immediately follows the array.
243 void prepare_alt_odb(void)
245 int pass, totlen, i;
246 const char *cp, *last;
247 char *op = NULL;
248 const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
250 if (alt_odb)
251 return;
252 /* The first pass counts how large an area to allocate to
253 * hold the entire alt_odb structure, including array of
254 * structs and path buffers for them. The second pass fills
255 * the structure and prepares the path buffers for use by
256 * fill_sha1_path().
258 for (totlen = pass = 0; pass < 2; pass++) {
259 last = alt;
260 i = 0;
261 do {
262 cp = strchr(last, ':') ? : last + strlen(last);
263 if (last != cp) {
264 /* 43 = 40-byte + 2 '/' + terminating NUL */
265 int pfxlen = cp - last;
266 int entlen = pfxlen + 43;
267 if (pass == 0)
268 totlen += entlen;
269 else {
270 alt_odb[i].base = op;
271 alt_odb[i].name = op + pfxlen + 1;
272 memcpy(op, last, pfxlen);
273 op[pfxlen] = op[pfxlen + 3] = '/';
274 op[entlen-1] = 0;
275 op += entlen;
277 i++;
279 while (*cp && *cp == ':')
280 cp++;
281 last = cp;
282 } while (*cp);
283 if (pass)
284 break;
285 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
286 alt_odb[i].base = alt_odb[i].name = NULL;
287 op = (char*)(&alt_odb[i+1]);
291 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
293 int i;
294 char *name = sha1_file_name(sha1);
296 if (!stat(name, st))
297 return name;
298 prepare_alt_odb();
299 for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
300 fill_sha1_path(name, sha1);
301 if (!stat(alt_odb[i].base, st))
302 return alt_odb[i].base;
304 return NULL;
307 #define PACK_MAX_SZ (1<<26)
308 static int pack_used_ctr;
309 static unsigned long pack_mapped;
310 struct packed_git *packed_git;
312 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
313 void **idx_map_)
315 void *idx_map;
316 unsigned int *index;
317 unsigned long idx_size;
318 int nr, i;
319 int fd = open(path, O_RDONLY);
320 struct stat st;
321 if (fd < 0)
322 return -1;
323 if (fstat(fd, &st)) {
324 close(fd);
325 return -1;
327 idx_size = st.st_size;
328 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
329 close(fd);
330 if (idx_map == MAP_FAILED)
331 return -1;
333 index = idx_map;
334 *idx_map_ = idx_map;
335 *idx_size_ = idx_size;
337 /* check index map */
338 if (idx_size < 4*256 + 20 + 20)
339 return error("index file too small");
340 nr = 0;
341 for (i = 0; i < 256; i++) {
342 unsigned int n = ntohl(index[i]);
343 if (n < nr)
344 return error("non-monotonic index");
345 nr = n;
349 * Total size:
350 * - 256 index entries 4 bytes each
351 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
352 * - 20-byte SHA1 of the packfile
353 * - 20-byte SHA1 file checksum
355 if (idx_size != 4*256 + nr * 24 + 20 + 20)
356 return error("wrong index file size");
358 return 0;
361 static int unuse_one_packed_git(void)
363 struct packed_git *p, *lru = NULL;
365 for (p = packed_git; p; p = p->next) {
366 if (p->pack_use_cnt || !p->pack_base)
367 continue;
368 if (!lru || p->pack_last_used < lru->pack_last_used)
369 lru = p;
371 if (!lru)
372 return 0;
373 munmap(lru->pack_base, lru->pack_size);
374 lru->pack_base = NULL;
375 return 1;
378 void unuse_packed_git(struct packed_git *p)
380 p->pack_use_cnt--;
383 int use_packed_git(struct packed_git *p)
385 if (!p->pack_size) {
386 struct stat st;
387 // We created the struct before we had the pack
388 stat(p->pack_name, &st);
389 if (!S_ISREG(st.st_mode))
390 die("packfile %s not a regular file", p->pack_name);
391 p->pack_size = st.st_size;
393 if (!p->pack_base) {
394 int fd;
395 struct stat st;
396 void *map;
398 pack_mapped += p->pack_size;
399 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
400 ; /* nothing */
401 fd = open(p->pack_name, O_RDONLY);
402 if (fd < 0)
403 die("packfile %s cannot be opened", p->pack_name);
404 if (fstat(fd, &st)) {
405 close(fd);
406 die("packfile %s cannot be opened", p->pack_name);
408 if (st.st_size != p->pack_size)
409 die("packfile %s size mismatch.", p->pack_name);
410 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
411 close(fd);
412 if (map == MAP_FAILED)
413 die("packfile %s cannot be mapped.", p->pack_name);
414 p->pack_base = map;
416 /* Check if the pack file matches with the index file.
417 * this is cheap.
419 if (memcmp((char*)(p->index_base) + p->index_size - 40,
420 p->pack_base + p->pack_size - 20, 20)) {
422 die("packfile %s does not match index.", p->pack_name);
425 p->pack_last_used = pack_used_ctr++;
426 p->pack_use_cnt++;
427 return 0;
430 struct packed_git *add_packed_git(char *path, int path_len)
432 struct stat st;
433 struct packed_git *p;
434 unsigned long idx_size;
435 void *idx_map;
437 if (check_packed_git_idx(path, &idx_size, &idx_map))
438 return NULL;
440 /* do we have a corresponding .pack file? */
441 strcpy(path + path_len - 4, ".pack");
442 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
443 munmap(idx_map, idx_size);
444 return NULL;
446 /* ok, it looks sane as far as we can check without
447 * actually mapping the pack file.
449 p = xmalloc(sizeof(*p) + path_len + 2);
450 strcpy(p->pack_name, path);
451 p->index_size = idx_size;
452 p->pack_size = st.st_size;
453 p->index_base = idx_map;
454 p->next = NULL;
455 p->pack_base = NULL;
456 p->pack_last_used = 0;
457 p->pack_use_cnt = 0;
458 return p;
461 struct packed_git *parse_pack_index(unsigned char *sha1)
463 struct packed_git *p;
464 unsigned long idx_size;
465 void *idx_map;
466 char *path = sha1_pack_index_name(sha1);
468 if (check_packed_git_idx(path, &idx_size, &idx_map))
469 return NULL;
471 path = sha1_pack_name(sha1);
473 p = xmalloc(sizeof(*p) + strlen(path) + 2);
474 strcpy(p->pack_name, path);
475 p->index_size = idx_size;
476 p->pack_size = 0;
477 p->index_base = idx_map;
478 p->next = NULL;
479 p->pack_base = NULL;
480 p->pack_last_used = 0;
481 p->pack_use_cnt = 0;
482 memcpy(p->sha1, sha1, 20);
483 return p;
486 void install_packed_git(struct packed_git *pack)
488 pack->next = packed_git;
489 packed_git = pack;
492 static void prepare_packed_git_one(char *objdir)
494 char path[PATH_MAX];
495 int len;
496 DIR *dir;
497 struct dirent *de;
499 sprintf(path, "%s/pack", objdir);
500 len = strlen(path);
501 dir = opendir(path);
502 if (!dir)
503 return;
504 path[len++] = '/';
505 while ((de = readdir(dir)) != NULL) {
506 int namelen = strlen(de->d_name);
507 struct packed_git *p;
509 if (strcmp(de->d_name + namelen - 4, ".idx"))
510 continue;
512 /* we have .idx. Is it a file we can map? */
513 strcpy(path + len, de->d_name);
514 p = add_packed_git(path, len + namelen);
515 if (!p)
516 continue;
517 p->next = packed_git;
518 packed_git = p;
520 closedir(dir);
523 void prepare_packed_git(void)
525 int i;
526 static int run_once = 0;
528 if (run_once++)
529 return;
531 prepare_packed_git_one(get_object_directory());
532 prepare_alt_odb();
533 for (i = 0; alt_odb[i].base != NULL; i++) {
534 alt_odb[i].name[0] = 0;
535 prepare_packed_git_one(alt_odb[i].base);
539 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
541 char header[100];
542 unsigned char real_sha1[20];
543 SHA_CTX c;
545 SHA1_Init(&c);
546 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
547 SHA1_Update(&c, map, size);
548 SHA1_Final(real_sha1, &c);
549 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
552 static void *map_sha1_file_internal(const unsigned char *sha1,
553 unsigned long *size)
555 struct stat st;
556 void *map;
557 int fd;
558 char *filename = find_sha1_file(sha1, &st);
560 if (!filename) {
561 return NULL;
564 fd = open(filename, O_RDONLY | sha1_file_open_flag);
565 if (fd < 0) {
566 /* See if it works without O_NOATIME */
567 switch (sha1_file_open_flag) {
568 default:
569 fd = open(filename, O_RDONLY);
570 if (fd >= 0)
571 break;
572 /* Fallthrough */
573 case 0:
574 return NULL;
577 /* If it failed once, it will probably fail again.
578 * Stop using O_NOATIME
580 sha1_file_open_flag = 0;
582 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
583 close(fd);
584 if (map == MAP_FAILED)
585 return NULL;
586 *size = st.st_size;
587 return map;
590 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
592 /* Get the data stream */
593 memset(stream, 0, sizeof(*stream));
594 stream->next_in = map;
595 stream->avail_in = mapsize;
596 stream->next_out = buffer;
597 stream->avail_out = size;
599 inflateInit(stream);
600 return inflate(stream, 0);
603 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
605 int bytes = strlen(buffer) + 1;
606 unsigned char *buf = xmalloc(1+size);
608 memcpy(buf, buffer + bytes, stream->total_out - bytes);
609 bytes = stream->total_out - bytes;
610 if (bytes < size) {
611 stream->next_out = buf + bytes;
612 stream->avail_out = size - bytes;
613 while (inflate(stream, Z_FINISH) == Z_OK)
614 /* nothing */;
616 buf[size] = 0;
617 inflateEnd(stream);
618 return buf;
622 * We used to just use "sscanf()", but that's actually way
623 * too permissive for what we want to check. So do an anal
624 * object header parse by hand.
626 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
628 int i;
629 unsigned long size;
632 * The type can be at most ten bytes (including the
633 * terminating '\0' that we add), and is followed by
634 * a space.
636 i = 10;
637 for (;;) {
638 char c = *hdr++;
639 if (c == ' ')
640 break;
641 if (!--i)
642 return -1;
643 *type++ = c;
645 *type = 0;
648 * The length must follow immediately, and be in canonical
649 * decimal format (ie "010" is not valid).
651 size = *hdr++ - '0';
652 if (size > 9)
653 return -1;
654 if (size) {
655 for (;;) {
656 unsigned long c = *hdr - '0';
657 if (c > 9)
658 break;
659 hdr++;
660 size = size * 10 + c;
663 *sizep = size;
666 * The length must be followed by a zero byte
668 return *hdr ? -1 : 0;
671 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
673 int ret;
674 z_stream stream;
675 char hdr[8192];
677 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
678 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
679 return NULL;
681 return unpack_sha1_rest(&stream, hdr, *size);
684 /* forward declaration for a mutually recursive function */
685 static int packed_object_info(struct pack_entry *entry,
686 char *type, unsigned long *sizep);
688 static int packed_delta_info(unsigned char *base_sha1,
689 unsigned long delta_size,
690 unsigned long left,
691 char *type,
692 unsigned long *sizep,
693 struct packed_git *p)
695 struct pack_entry base_ent;
697 if (left < 20)
698 die("truncated pack file");
700 /* The base entry _must_ be in the same pack */
701 if (!find_pack_entry_one(base_sha1, &base_ent, p))
702 die("failed to find delta-pack base object %s",
703 sha1_to_hex(base_sha1));
705 /* We choose to only get the type of the base object and
706 * ignore potentially corrupt pack file that expects the delta
707 * based on a base with a wrong size. This saves tons of
708 * inflate() calls.
711 if (packed_object_info(&base_ent, type, NULL))
712 die("cannot get info for delta-pack base");
714 if (sizep) {
715 const unsigned char *data;
716 unsigned char delta_head[64];
717 unsigned long result_size;
718 z_stream stream;
719 int st;
721 memset(&stream, 0, sizeof(stream));
723 data = stream.next_in = base_sha1 + 20;
724 stream.avail_in = left - 20;
725 stream.next_out = delta_head;
726 stream.avail_out = sizeof(delta_head);
728 inflateInit(&stream);
729 st = inflate(&stream, Z_FINISH);
730 inflateEnd(&stream);
731 if ((st != Z_STREAM_END) &&
732 stream.total_out != sizeof(delta_head))
733 die("delta data unpack-initial failed");
735 /* Examine the initial part of the delta to figure out
736 * the result size.
738 data = delta_head;
739 get_delta_hdr_size(&data); /* ignore base size */
741 /* Read the result size */
742 result_size = get_delta_hdr_size(&data);
743 *sizep = result_size;
745 return 0;
748 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
749 enum object_type *type, unsigned long *sizep)
751 unsigned shift;
752 unsigned char *pack, c;
753 unsigned long size;
755 if (offset >= p->pack_size)
756 die("object offset outside of pack file");
758 pack = p->pack_base + offset;
759 c = *pack++;
760 offset++;
761 *type = (c >> 4) & 7;
762 size = c & 15;
763 shift = 4;
764 while (c & 0x80) {
765 if (offset >= p->pack_size)
766 die("object offset outside of pack file");
767 c = *pack++;
768 offset++;
769 size += (c & 0x7f) << shift;
770 shift += 7;
772 *sizep = size;
773 return offset;
776 void packed_object_info_detail(struct pack_entry *e,
777 char *type,
778 unsigned long *size,
779 unsigned long *store_size,
780 int *delta_chain_length,
781 unsigned char *base_sha1)
783 struct packed_git *p = e->p;
784 unsigned long offset, left;
785 unsigned char *pack;
786 enum object_type kind;
788 offset = unpack_object_header(p, e->offset, &kind, size);
789 pack = p->pack_base + offset;
790 left = p->pack_size - offset;
791 if (kind != OBJ_DELTA)
792 *delta_chain_length = 0;
793 else {
794 int chain_length = 0;
795 memcpy(base_sha1, pack, 20);
796 do {
797 struct pack_entry base_ent;
798 unsigned long junk;
800 find_pack_entry_one(pack, &base_ent, p);
801 offset = unpack_object_header(p, base_ent.offset,
802 &kind, &junk);
803 pack = p->pack_base + offset;
804 chain_length++;
805 } while (kind == OBJ_DELTA);
806 *delta_chain_length = chain_length;
808 switch (kind) {
809 case OBJ_COMMIT:
810 strcpy(type, "commit");
811 break;
812 case OBJ_TREE:
813 strcpy(type, "tree");
814 break;
815 case OBJ_BLOB:
816 strcpy(type, "blob");
817 break;
818 case OBJ_TAG:
819 strcpy(type, "tag");
820 break;
821 default:
822 die("corrupted pack file");
824 *store_size = 0; /* notyet */
827 static int packed_object_info(struct pack_entry *entry,
828 char *type, unsigned long *sizep)
830 struct packed_git *p = entry->p;
831 unsigned long offset, size, left;
832 unsigned char *pack;
833 enum object_type kind;
834 int retval;
836 if (use_packed_git(p))
837 die("cannot map packed file");
839 offset = unpack_object_header(p, entry->offset, &kind, &size);
840 pack = p->pack_base + offset;
841 left = p->pack_size - offset;
843 switch (kind) {
844 case OBJ_DELTA:
845 retval = packed_delta_info(pack, size, left, type, sizep, p);
846 unuse_packed_git(p);
847 return retval;
848 case OBJ_COMMIT:
849 strcpy(type, "commit");
850 break;
851 case OBJ_TREE:
852 strcpy(type, "tree");
853 break;
854 case OBJ_BLOB:
855 strcpy(type, "blob");
856 break;
857 case OBJ_TAG:
858 strcpy(type, "tag");
859 break;
860 default:
861 die("corrupted pack file");
863 if (sizep)
864 *sizep = size;
865 unuse_packed_git(p);
866 return 0;
869 /* forward declaration for a mutually recursive function */
870 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
872 static void *unpack_delta_entry(unsigned char *base_sha1,
873 unsigned long delta_size,
874 unsigned long left,
875 char *type,
876 unsigned long *sizep,
877 struct packed_git *p)
879 struct pack_entry base_ent;
880 void *data, *delta_data, *result, *base;
881 unsigned long data_size, result_size, base_size;
882 z_stream stream;
883 int st;
885 if (left < 20)
886 die("truncated pack file");
887 data = base_sha1 + 20;
888 data_size = left - 20;
889 delta_data = xmalloc(delta_size);
891 memset(&stream, 0, sizeof(stream));
893 stream.next_in = data;
894 stream.avail_in = data_size;
895 stream.next_out = delta_data;
896 stream.avail_out = delta_size;
898 inflateInit(&stream);
899 st = inflate(&stream, Z_FINISH);
900 inflateEnd(&stream);
901 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
902 die("delta data unpack failed");
904 /* The base entry _must_ be in the same pack */
905 if (!find_pack_entry_one(base_sha1, &base_ent, p))
906 die("failed to find delta-pack base object %s",
907 sha1_to_hex(base_sha1));
908 base = unpack_entry_gently(&base_ent, type, &base_size);
909 if (!base)
910 die("failed to read delta-pack base object %s",
911 sha1_to_hex(base_sha1));
912 result = patch_delta(base, base_size,
913 delta_data, delta_size,
914 &result_size);
915 if (!result)
916 die("failed to apply delta");
917 free(delta_data);
918 free(base);
919 *sizep = result_size;
920 return result;
923 static void *unpack_non_delta_entry(unsigned char *data,
924 unsigned long size,
925 unsigned long left)
927 int st;
928 z_stream stream;
929 unsigned char *buffer;
931 buffer = xmalloc(size + 1);
932 buffer[size] = 0;
933 memset(&stream, 0, sizeof(stream));
934 stream.next_in = data;
935 stream.avail_in = left;
936 stream.next_out = buffer;
937 stream.avail_out = size;
939 inflateInit(&stream);
940 st = inflate(&stream, Z_FINISH);
941 inflateEnd(&stream);
942 if ((st != Z_STREAM_END) || stream.total_out != size) {
943 free(buffer);
944 return NULL;
947 return buffer;
950 static void *unpack_entry(struct pack_entry *entry,
951 char *type, unsigned long *sizep)
953 struct packed_git *p = entry->p;
954 void *retval;
956 if (use_packed_git(p))
957 die("cannot map packed file");
958 retval = unpack_entry_gently(entry, type, sizep);
959 unuse_packed_git(p);
960 if (!retval)
961 die("corrupted pack file");
962 return retval;
965 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
966 void *unpack_entry_gently(struct pack_entry *entry,
967 char *type, unsigned long *sizep)
969 struct packed_git *p = entry->p;
970 unsigned long offset, size, left;
971 unsigned char *pack;
972 enum object_type kind;
973 void *retval;
975 offset = unpack_object_header(p, entry->offset, &kind, &size);
976 pack = p->pack_base + offset;
977 left = p->pack_size - offset;
978 switch (kind) {
979 case OBJ_DELTA:
980 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
981 return retval;
982 case OBJ_COMMIT:
983 strcpy(type, "commit");
984 break;
985 case OBJ_TREE:
986 strcpy(type, "tree");
987 break;
988 case OBJ_BLOB:
989 strcpy(type, "blob");
990 break;
991 case OBJ_TAG:
992 strcpy(type, "tag");
993 break;
994 default:
995 return NULL;
997 *sizep = size;
998 retval = unpack_non_delta_entry(pack, size, left);
999 return retval;
1002 int num_packed_objects(const struct packed_git *p)
1004 /* See check_packed_git_idx() */
1005 return (p->index_size - 20 - 20 - 4*256) / 24;
1008 int nth_packed_object_sha1(const struct packed_git *p, int n,
1009 unsigned char* sha1)
1011 void *index = p->index_base + 256;
1012 if (n < 0 || num_packed_objects(p) <= n)
1013 return -1;
1014 memcpy(sha1, (index + 24 * n + 4), 20);
1015 return 0;
1018 int find_pack_entry_one(const unsigned char *sha1,
1019 struct pack_entry *e, struct packed_git *p)
1021 unsigned int *level1_ofs = p->index_base;
1022 int hi = ntohl(level1_ofs[*sha1]);
1023 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1024 void *index = p->index_base + 256;
1026 do {
1027 int mi = (lo + hi) / 2;
1028 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1029 if (!cmp) {
1030 e->offset = ntohl(*((int*)(index + 24 * mi)));
1031 memcpy(e->sha1, sha1, 20);
1032 e->p = p;
1033 return 1;
1035 if (cmp > 0)
1036 hi = mi;
1037 else
1038 lo = mi+1;
1039 } while (lo < hi);
1040 return 0;
1043 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1045 struct packed_git *p;
1046 prepare_packed_git();
1048 for (p = packed_git; p; p = p->next) {
1049 if (find_pack_entry_one(sha1, e, p))
1050 return 1;
1052 return 0;
1055 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1056 struct packed_git *packs)
1058 struct packed_git *p;
1059 struct pack_entry e;
1061 for (p = packs; p; p = p->next) {
1062 if (find_pack_entry_one(sha1, &e, p))
1063 return p;
1065 return NULL;
1069 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1071 int status;
1072 unsigned long mapsize, size;
1073 void *map;
1074 z_stream stream;
1075 char hdr[128];
1077 map = map_sha1_file_internal(sha1, &mapsize);
1078 if (!map) {
1079 struct pack_entry e;
1081 if (!find_pack_entry(sha1, &e))
1082 return error("unable to find %s", sha1_to_hex(sha1));
1083 return packed_object_info(&e, type, sizep);
1085 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1086 status = error("unable to unpack %s header",
1087 sha1_to_hex(sha1));
1088 if (parse_sha1_header(hdr, type, &size) < 0)
1089 status = error("unable to parse %s header", sha1_to_hex(sha1));
1090 else {
1091 status = 0;
1092 if (sizep)
1093 *sizep = size;
1095 inflateEnd(&stream);
1096 munmap(map, mapsize);
1097 return status;
1100 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1102 struct pack_entry e;
1104 if (!find_pack_entry(sha1, &e)) {
1105 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1106 return NULL;
1108 return unpack_entry(&e, type, size);
1111 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1113 unsigned long mapsize;
1114 void *map, *buf;
1115 struct pack_entry e;
1117 if (find_pack_entry(sha1, &e))
1118 return read_packed_sha1(sha1, type, size);
1119 map = map_sha1_file_internal(sha1, &mapsize);
1120 if (map) {
1121 buf = unpack_sha1_file(map, mapsize, type, size);
1122 munmap(map, mapsize);
1123 return buf;
1125 return NULL;
1128 void *read_object_with_reference(const unsigned char *sha1,
1129 const char *required_type,
1130 unsigned long *size,
1131 unsigned char *actual_sha1_return)
1133 char type[20];
1134 void *buffer;
1135 unsigned long isize;
1136 unsigned char actual_sha1[20];
1138 memcpy(actual_sha1, sha1, 20);
1139 while (1) {
1140 int ref_length = -1;
1141 const char *ref_type = NULL;
1143 buffer = read_sha1_file(actual_sha1, type, &isize);
1144 if (!buffer)
1145 return NULL;
1146 if (!strcmp(type, required_type)) {
1147 *size = isize;
1148 if (actual_sha1_return)
1149 memcpy(actual_sha1_return, actual_sha1, 20);
1150 return buffer;
1152 /* Handle references */
1153 else if (!strcmp(type, "commit"))
1154 ref_type = "tree ";
1155 else if (!strcmp(type, "tag"))
1156 ref_type = "object ";
1157 else {
1158 free(buffer);
1159 return NULL;
1161 ref_length = strlen(ref_type);
1163 if (memcmp(buffer, ref_type, ref_length) ||
1164 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1165 free(buffer);
1166 return NULL;
1168 free(buffer);
1169 /* Now we have the ID of the referred-to object in
1170 * actual_sha1. Check again. */
1174 char *write_sha1_file_prepare(void *buf,
1175 unsigned long len,
1176 const char *type,
1177 unsigned char *sha1,
1178 unsigned char *hdr,
1179 int *hdrlen)
1181 SHA_CTX c;
1183 /* Generate the header */
1184 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1186 /* Sha1.. */
1187 SHA1_Init(&c);
1188 SHA1_Update(&c, hdr, *hdrlen);
1189 SHA1_Update(&c, buf, len);
1190 SHA1_Final(sha1, &c);
1192 return sha1_file_name(sha1);
1195 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1197 int size;
1198 unsigned char *compressed;
1199 z_stream stream;
1200 unsigned char sha1[20];
1201 char *filename;
1202 static char tmpfile[PATH_MAX];
1203 unsigned char hdr[50];
1204 int fd, hdrlen, ret;
1206 /* Normally if we have it in the pack then we do not bother writing
1207 * it out into .git/objects/??/?{38} file.
1209 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1210 if (returnsha1)
1211 memcpy(returnsha1, sha1, 20);
1212 if (has_sha1_file(sha1))
1213 return 0;
1214 fd = open(filename, O_RDONLY);
1215 if (fd >= 0) {
1217 * FIXME!!! We might do collision checking here, but we'd
1218 * need to uncompress the old file and check it. Later.
1220 close(fd);
1221 return 0;
1224 if (errno != ENOENT) {
1225 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1226 return -1;
1229 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1231 fd = mkstemp(tmpfile);
1232 if (fd < 0) {
1233 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1234 return -1;
1237 /* Set it up */
1238 memset(&stream, 0, sizeof(stream));
1239 deflateInit(&stream, Z_BEST_COMPRESSION);
1240 size = deflateBound(&stream, len+hdrlen);
1241 compressed = xmalloc(size);
1243 /* Compress it */
1244 stream.next_out = compressed;
1245 stream.avail_out = size;
1247 /* First header.. */
1248 stream.next_in = hdr;
1249 stream.avail_in = hdrlen;
1250 while (deflate(&stream, 0) == Z_OK)
1251 /* nothing */;
1253 /* Then the data itself.. */
1254 stream.next_in = buf;
1255 stream.avail_in = len;
1256 while (deflate(&stream, Z_FINISH) == Z_OK)
1257 /* nothing */;
1258 deflateEnd(&stream);
1259 size = stream.total_out;
1261 if (write(fd, compressed, size) != size)
1262 die("unable to write file");
1263 fchmod(fd, 0444);
1264 close(fd);
1265 free(compressed);
1267 ret = link(tmpfile, filename);
1268 if (ret < 0) {
1269 ret = errno;
1272 * Coda hack - coda doesn't like cross-directory links,
1273 * so we fall back to a rename, which will mean that it
1274 * won't be able to check collisions, but that's not a
1275 * big deal.
1277 * When this succeeds, we just return 0. We have nothing
1278 * left to unlink.
1280 if (ret == EXDEV && !rename(tmpfile, filename))
1281 return 0;
1283 unlink(tmpfile);
1284 if (ret) {
1285 if (ret != EEXIST) {
1286 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1287 return -1;
1289 /* FIXME!!! Collision check here ? */
1292 return 0;
1295 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1297 ssize_t size;
1298 unsigned long objsize;
1299 int posn = 0;
1300 void *map = map_sha1_file_internal(sha1, &objsize);
1301 void *buf = map;
1302 void *temp_obj = NULL;
1303 z_stream stream;
1305 if (!buf) {
1306 unsigned char *unpacked;
1307 unsigned long len;
1308 char type[20];
1309 char hdr[50];
1310 int hdrlen;
1311 // need to unpack and recompress it by itself
1312 unpacked = read_packed_sha1(sha1, type, &len);
1314 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1316 /* Set it up */
1317 memset(&stream, 0, sizeof(stream));
1318 deflateInit(&stream, Z_BEST_COMPRESSION);
1319 size = deflateBound(&stream, len + hdrlen);
1320 temp_obj = buf = xmalloc(size);
1322 /* Compress it */
1323 stream.next_out = buf;
1324 stream.avail_out = size;
1326 /* First header.. */
1327 stream.next_in = (void *)hdr;
1328 stream.avail_in = hdrlen;
1329 while (deflate(&stream, 0) == Z_OK)
1330 /* nothing */;
1332 /* Then the data itself.. */
1333 stream.next_in = unpacked;
1334 stream.avail_in = len;
1335 while (deflate(&stream, Z_FINISH) == Z_OK)
1336 /* nothing */;
1337 deflateEnd(&stream);
1338 free(unpacked);
1340 objsize = stream.total_out;
1343 do {
1344 size = write(fd, buf + posn, objsize - posn);
1345 if (size <= 0) {
1346 if (!size) {
1347 fprintf(stderr, "write closed");
1348 } else {
1349 perror("write ");
1351 return -1;
1353 posn += size;
1354 } while (posn < objsize);
1356 if (map)
1357 munmap(map, objsize);
1358 if (temp_obj)
1359 free(temp_obj);
1361 return 0;
1364 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1365 size_t bufsize, size_t *bufposn)
1367 char *filename = sha1_file_name(sha1);
1369 int local;
1370 z_stream stream;
1371 unsigned char real_sha1[20];
1372 unsigned char discard[4096];
1373 int ret;
1374 SHA_CTX c;
1376 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1378 if (local < 0)
1379 return error("Couldn't open %s\n", filename);
1381 memset(&stream, 0, sizeof(stream));
1383 inflateInit(&stream);
1385 SHA1_Init(&c);
1387 do {
1388 ssize_t size;
1389 if (*bufposn) {
1390 stream.avail_in = *bufposn;
1391 stream.next_in = (unsigned char *) buffer;
1392 do {
1393 stream.next_out = discard;
1394 stream.avail_out = sizeof(discard);
1395 ret = inflate(&stream, Z_SYNC_FLUSH);
1396 SHA1_Update(&c, discard, sizeof(discard) -
1397 stream.avail_out);
1398 } while (stream.avail_in && ret == Z_OK);
1399 write(local, buffer, *bufposn - stream.avail_in);
1400 memmove(buffer, buffer + *bufposn - stream.avail_in,
1401 stream.avail_in);
1402 *bufposn = stream.avail_in;
1403 if (ret != Z_OK)
1404 break;
1406 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1407 if (size <= 0) {
1408 close(local);
1409 unlink(filename);
1410 if (!size)
1411 return error("Connection closed?");
1412 perror("Reading from connection");
1413 return -1;
1415 *bufposn += size;
1416 } while (1);
1417 inflateEnd(&stream);
1419 close(local);
1420 SHA1_Final(real_sha1, &c);
1421 if (ret != Z_STREAM_END) {
1422 unlink(filename);
1423 return error("File %s corrupted", sha1_to_hex(sha1));
1425 if (memcmp(sha1, real_sha1, 20)) {
1426 unlink(filename);
1427 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1430 return 0;
1433 int has_pack_index(const unsigned char *sha1)
1435 struct stat st;
1436 if (stat(sha1_pack_index_name(sha1), &st))
1437 return 0;
1438 return 1;
1441 int has_pack_file(const unsigned char *sha1)
1443 struct stat st;
1444 if (stat(sha1_pack_name(sha1), &st))
1445 return 0;
1446 return 1;
1449 int has_sha1_pack(const unsigned char *sha1)
1451 struct pack_entry e;
1452 return find_pack_entry(sha1, &e);
1455 int has_sha1_file(const unsigned char *sha1)
1457 struct stat st;
1458 struct pack_entry e;
1460 if (find_pack_entry(sha1, &e))
1461 return 1;
1462 return find_sha1_file(sha1, &st) ? 1 : 0;
1465 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1467 unsigned long size = st->st_size;
1468 void *buf;
1469 int ret;
1470 unsigned char hdr[50];
1471 int hdrlen;
1473 buf = "";
1474 if (size)
1475 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1476 close(fd);
1477 if (buf == MAP_FAILED)
1478 return -1;
1480 if (!type)
1481 type = "blob";
1482 if (write_object)
1483 ret = write_sha1_file(buf, size, type, sha1);
1484 else {
1485 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1486 ret = 0;
1488 if (size)
1489 munmap(buf, size);
1490 return ret;