sha1_file.c: learn about index version 2
[git/gitweb.git] / sha1_file.c
blob927ac0600d17edf0b8d8d05f896cef409ea9fe36
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];
27 static unsigned int sha1_file_open_flag = O_NOATIME;
29 signed char hexval_table[256] = {
30 -1, -1, -1, -1, -1, -1, -1, -1, /* 00-07 */
31 -1, -1, -1, -1, -1, -1, -1, -1, /* 08-0f */
32 -1, -1, -1, -1, -1, -1, -1, -1, /* 10-17 */
33 -1, -1, -1, -1, -1, -1, -1, -1, /* 18-1f */
34 -1, -1, -1, -1, -1, -1, -1, -1, /* 20-27 */
35 -1, -1, -1, -1, -1, -1, -1, -1, /* 28-2f */
36 0, 1, 2, 3, 4, 5, 6, 7, /* 30-37 */
37 8, 9, -1, -1, -1, -1, -1, -1, /* 38-3f */
38 -1, 10, 11, 12, 13, 14, 15, -1, /* 40-47 */
39 -1, -1, -1, -1, -1, -1, -1, -1, /* 48-4f */
40 -1, -1, -1, -1, -1, -1, -1, -1, /* 50-57 */
41 -1, -1, -1, -1, -1, -1, -1, -1, /* 58-5f */
42 -1, 10, 11, 12, 13, 14, 15, -1, /* 60-67 */
43 -1, -1, -1, -1, -1, -1, -1, -1, /* 68-67 */
44 -1, -1, -1, -1, -1, -1, -1, -1, /* 70-77 */
45 -1, -1, -1, -1, -1, -1, -1, -1, /* 78-7f */
46 -1, -1, -1, -1, -1, -1, -1, -1, /* 80-87 */
47 -1, -1, -1, -1, -1, -1, -1, -1, /* 88-8f */
48 -1, -1, -1, -1, -1, -1, -1, -1, /* 90-97 */
49 -1, -1, -1, -1, -1, -1, -1, -1, /* 98-9f */
50 -1, -1, -1, -1, -1, -1, -1, -1, /* a0-a7 */
51 -1, -1, -1, -1, -1, -1, -1, -1, /* a8-af */
52 -1, -1, -1, -1, -1, -1, -1, -1, /* b0-b7 */
53 -1, -1, -1, -1, -1, -1, -1, -1, /* b8-bf */
54 -1, -1, -1, -1, -1, -1, -1, -1, /* c0-c7 */
55 -1, -1, -1, -1, -1, -1, -1, -1, /* c8-cf */
56 -1, -1, -1, -1, -1, -1, -1, -1, /* d0-d7 */
57 -1, -1, -1, -1, -1, -1, -1, -1, /* d8-df */
58 -1, -1, -1, -1, -1, -1, -1, -1, /* e0-e7 */
59 -1, -1, -1, -1, -1, -1, -1, -1, /* e8-ef */
60 -1, -1, -1, -1, -1, -1, -1, -1, /* f0-f7 */
61 -1, -1, -1, -1, -1, -1, -1, -1, /* f8-ff */
64 int get_sha1_hex(const char *hex, unsigned char *sha1)
66 int i;
67 for (i = 0; i < 20; i++) {
68 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
69 if (val & ~0xff)
70 return -1;
71 *sha1++ = val;
72 hex += 2;
74 return 0;
77 int safe_create_leading_directories(char *path)
79 char *pos = path;
80 struct stat st;
82 if (*pos == '/')
83 pos++;
85 while (pos) {
86 pos = strchr(pos, '/');
87 if (!pos)
88 break;
89 *pos = 0;
90 if (!stat(path, &st)) {
91 /* path exists */
92 if (!S_ISDIR(st.st_mode)) {
93 *pos = '/';
94 return -3;
97 else if (mkdir(path, 0777)) {
98 *pos = '/';
99 return -1;
101 else if (adjust_shared_perm(path)) {
102 *pos = '/';
103 return -2;
105 *pos++ = '/';
107 return 0;
110 char * sha1_to_hex(const unsigned char *sha1)
112 static int bufno;
113 static char hexbuffer[4][50];
114 static const char hex[] = "0123456789abcdef";
115 char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
116 int i;
118 for (i = 0; i < 20; i++) {
119 unsigned int val = *sha1++;
120 *buf++ = hex[val >> 4];
121 *buf++ = hex[val & 0xf];
123 *buf = '\0';
125 return buffer;
128 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
130 int i;
131 for (i = 0; i < 20; i++) {
132 static char hex[] = "0123456789abcdef";
133 unsigned int val = sha1[i];
134 char *pos = pathbuf + i*2 + (i > 0);
135 *pos++ = hex[val >> 4];
136 *pos = hex[val & 0xf];
141 * NOTE! This returns a statically allocated buffer, so you have to be
142 * careful about using it. Do a "xstrdup()" if you need to save the
143 * filename.
145 * Also note that this returns the location for creating. Reading
146 * SHA1 file can happen from any alternate directory listed in the
147 * DB_ENVIRONMENT environment variable if it is not found in
148 * the primary object database.
150 char *sha1_file_name(const unsigned char *sha1)
152 static char *name, *base;
154 if (!base) {
155 const char *sha1_file_directory = get_object_directory();
156 int len = strlen(sha1_file_directory);
157 base = xmalloc(len + 60);
158 memcpy(base, sha1_file_directory, len);
159 memset(base+len, 0, 60);
160 base[len] = '/';
161 base[len+3] = '/';
162 name = base + len + 1;
164 fill_sha1_path(name, sha1);
165 return base;
168 char *sha1_pack_name(const unsigned char *sha1)
170 static const char hex[] = "0123456789abcdef";
171 static char *name, *base, *buf;
172 int i;
174 if (!base) {
175 const char *sha1_file_directory = get_object_directory();
176 int len = strlen(sha1_file_directory);
177 base = xmalloc(len + 60);
178 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
179 name = base + len + 11;
182 buf = name;
184 for (i = 0; i < 20; i++) {
185 unsigned int val = *sha1++;
186 *buf++ = hex[val >> 4];
187 *buf++ = hex[val & 0xf];
190 return base;
193 char *sha1_pack_index_name(const unsigned char *sha1)
195 static const char hex[] = "0123456789abcdef";
196 static char *name, *base, *buf;
197 int i;
199 if (!base) {
200 const char *sha1_file_directory = get_object_directory();
201 int len = strlen(sha1_file_directory);
202 base = xmalloc(len + 60);
203 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
204 name = base + len + 11;
207 buf = name;
209 for (i = 0; i < 20; i++) {
210 unsigned int val = *sha1++;
211 *buf++ = hex[val >> 4];
212 *buf++ = hex[val & 0xf];
215 return base;
218 struct alternate_object_database *alt_odb_list;
219 static struct alternate_object_database **alt_odb_tail;
221 static void read_info_alternates(const char * alternates, int depth);
224 * Prepare alternate object database registry.
226 * The variable alt_odb_list points at the list of struct
227 * alternate_object_database. The elements on this list come from
228 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
229 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
230 * whose contents is similar to that environment variable but can be
231 * LF separated. Its base points at a statically allocated buffer that
232 * contains "/the/directory/corresponding/to/.git/objects/...", while
233 * its name points just after the slash at the end of ".git/objects/"
234 * in the example above, and has enough space to hold 40-byte hex
235 * SHA1, an extra slash for the first level indirection, and the
236 * terminating NUL.
238 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
240 struct stat st;
241 const char *objdir = get_object_directory();
242 struct alternate_object_database *ent;
243 struct alternate_object_database *alt;
244 /* 43 = 40-byte + 2 '/' + terminating NUL */
245 int pfxlen = len;
246 int entlen = pfxlen + 43;
247 int base_len = -1;
249 if (*entry != '/' && relative_base) {
250 /* Relative alt-odb */
251 if (base_len < 0)
252 base_len = strlen(relative_base) + 1;
253 entlen += base_len;
254 pfxlen += base_len;
256 ent = xmalloc(sizeof(*ent) + entlen);
258 if (*entry != '/' && relative_base) {
259 memcpy(ent->base, relative_base, base_len - 1);
260 ent->base[base_len - 1] = '/';
261 memcpy(ent->base + base_len, entry, len);
263 else
264 memcpy(ent->base, entry, pfxlen);
266 ent->name = ent->base + pfxlen + 1;
267 ent->base[pfxlen + 3] = '/';
268 ent->base[pfxlen] = ent->base[entlen-1] = 0;
270 /* Detect cases where alternate disappeared */
271 if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
272 error("object directory %s does not exist; "
273 "check .git/objects/info/alternates.",
274 ent->base);
275 free(ent);
276 return -1;
279 /* Prevent the common mistake of listing the same
280 * thing twice, or object directory itself.
282 for (alt = alt_odb_list; alt; alt = alt->next) {
283 if (!memcmp(ent->base, alt->base, pfxlen)) {
284 free(ent);
285 return -1;
288 if (!memcmp(ent->base, objdir, pfxlen)) {
289 free(ent);
290 return -1;
293 /* add the alternate entry */
294 *alt_odb_tail = ent;
295 alt_odb_tail = &(ent->next);
296 ent->next = NULL;
298 /* recursively add alternates */
299 read_info_alternates(ent->base, depth + 1);
301 ent->base[pfxlen] = '/';
303 return 0;
306 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
307 const char *relative_base, int depth)
309 const char *cp, *last;
311 if (depth > 5) {
312 error("%s: ignoring alternate object stores, nesting too deep.",
313 relative_base);
314 return;
317 last = alt;
318 while (last < ep) {
319 cp = last;
320 if (cp < ep && *cp == '#') {
321 while (cp < ep && *cp != sep)
322 cp++;
323 last = cp + 1;
324 continue;
326 while (cp < ep && *cp != sep)
327 cp++;
328 if (last != cp) {
329 if ((*last != '/') && depth) {
330 error("%s: ignoring relative alternate object store %s",
331 relative_base, last);
332 } else {
333 link_alt_odb_entry(last, cp - last,
334 relative_base, depth);
337 while (cp < ep && *cp == sep)
338 cp++;
339 last = cp;
343 static void read_info_alternates(const char * relative_base, int depth)
345 char *map;
346 struct stat st;
347 char path[PATH_MAX];
348 int fd;
350 sprintf(path, "%s/info/alternates", relative_base);
351 fd = open(path, O_RDONLY);
352 if (fd < 0)
353 return;
354 if (fstat(fd, &st) || (st.st_size == 0)) {
355 close(fd);
356 return;
358 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
359 close(fd);
360 if (map == MAP_FAILED)
361 return;
363 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
365 munmap(map, st.st_size);
368 void prepare_alt_odb(void)
370 const char *alt;
372 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
373 if (!alt) alt = "";
375 if (alt_odb_tail)
376 return;
377 alt_odb_tail = &alt_odb_list;
378 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
380 read_info_alternates(get_object_directory(), 0);
383 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
385 char *name = sha1_file_name(sha1);
386 struct alternate_object_database *alt;
388 if (!stat(name, st))
389 return name;
390 prepare_alt_odb();
391 for (alt = alt_odb_list; alt; alt = alt->next) {
392 name = alt->name;
393 fill_sha1_path(name, sha1);
394 if (!stat(alt->base, st))
395 return alt->base;
397 return NULL;
400 #define PACK_MAX_SZ (1<<26)
401 static int pack_used_ctr;
402 static unsigned long pack_mapped;
403 struct packed_git *packed_git;
405 static int check_packed_git_idx(const char *path, struct packed_git *p)
407 void *idx_map;
408 struct pack_idx_header *hdr;
409 unsigned long idx_size;
410 unsigned int version, nr, i, *index;
411 int fd = open(path, O_RDONLY);
412 struct stat st;
414 if (fd < 0)
415 return -1;
416 if (fstat(fd, &st)) {
417 close(fd);
418 return -1;
420 idx_size = st.st_size;
421 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
422 close(fd);
423 if (idx_map == MAP_FAILED)
424 return -1;
426 hdr = idx_map;
427 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
428 version = ntohl(hdr->idx_version);
429 if (version < 2 || version > 2) {
430 munmap(idx_map, idx_size);
431 return error("index file %s is version %d"
432 " and is not supported by this binary"
433 " (try upgrading GIT to a newer version)",
434 path, version);
436 } else
437 version = 1;
439 nr = 0;
440 index = idx_map;
441 if (version > 1)
442 index += 2; /* skip index header */
443 for (i = 0; i < 256; i++) {
444 unsigned int n = ntohl(index[i]);
445 if (n < nr)
446 return error("non-monotonic index");
447 nr = n;
450 if (version == 1) {
452 * Total size:
453 * - 256 index entries 4 bytes each
454 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
455 * - 20-byte SHA1 of the packfile
456 * - 20-byte SHA1 file checksum
458 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
459 munmap(idx_map, idx_size);
460 return error("wrong index file size in %s", path);
462 } else if (version == 2) {
464 * Minimum size:
465 * - 8 bytes of header
466 * - 256 index entries 4 bytes each
467 * - 20-byte sha1 entry * nr
468 * - 4-byte crc entry * nr
469 * - 4-byte offset entry * nr
470 * - 20-byte SHA1 of the packfile
471 * - 20-byte SHA1 file checksum
472 * And after the 4-byte offset table might be a
473 * variable sized table containing 8-byte entries
474 * for offsets larger than 2^31.
476 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
477 if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
478 munmap(idx_map, idx_size);
479 return error("wrong index file size in %s", path);
481 if (idx_size != min_size) {
482 /* make sure we can deal with large pack offsets */
483 if (sizeof(unsigned long) <= 4) {
484 munmap(idx_map, idx_size);
485 return error("pack %s too large -- please upgrade your git version", path);
490 p->index_version = version;
491 p->index_data = idx_map;
492 p->index_size = idx_size;
493 p->num_objects = nr;
494 return 0;
497 static int unuse_one_packed_git(void)
499 struct packed_git *p, *lru = NULL;
501 for (p = packed_git; p; p = p->next) {
502 if (p->pack_use_cnt || !p->pack_base)
503 continue;
504 if (!lru || p->pack_last_used < lru->pack_last_used)
505 lru = p;
507 if (!lru)
508 return 0;
509 munmap(lru->pack_base, lru->pack_size);
510 lru->pack_base = NULL;
511 return 1;
514 void unuse_packed_git(struct packed_git *p)
516 p->pack_use_cnt--;
519 int use_packed_git(struct packed_git *p)
521 if (!p->pack_size) {
522 struct stat st;
523 /* We created the struct before we had the pack */
524 stat(p->pack_name, &st);
525 if (!S_ISREG(st.st_mode))
526 die("packfile %s not a regular file", p->pack_name);
527 p->pack_size = st.st_size;
529 if (!p->pack_base) {
530 int fd;
531 struct stat st;
532 void *map;
533 struct pack_header *hdr;
535 pack_mapped += p->pack_size;
536 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
537 ; /* nothing */
538 fd = open(p->pack_name, O_RDONLY);
539 if (fd < 0)
540 die("packfile %s cannot be opened", p->pack_name);
541 if (fstat(fd, &st)) {
542 close(fd);
543 die("packfile %s cannot be opened", p->pack_name);
545 if (st.st_size != p->pack_size)
546 die("packfile %s size mismatch.", p->pack_name);
547 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
548 close(fd);
549 if (map == MAP_FAILED)
550 die("packfile %s cannot be mapped.", p->pack_name);
551 p->pack_base = map;
553 /* Check if we understand this pack file. If we don't we're
554 * likely too old to handle it.
556 hdr = map;
557 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
558 die("packfile %s isn't actually a pack.", p->pack_name);
559 if (!pack_version_ok(hdr->hdr_version))
560 die("packfile %s is version %i and not supported"
561 " (try upgrading GIT to a newer version)",
562 p->pack_name, ntohl(hdr->hdr_version));
564 /* Check if the pack file matches with the index file.
565 * this is cheap.
567 if (hashcmp((unsigned char *)(p->index_data) +
568 p->index_size - 40,
569 (unsigned char *)p->pack_base +
570 p->pack_size - 20)) {
571 die("packfile %s does not match index.", p->pack_name);
574 p->pack_last_used = pack_used_ctr++;
575 p->pack_use_cnt++;
576 return 0;
579 struct packed_git *add_packed_git(char *path, int path_len, int local)
581 struct stat st;
582 struct packed_git *p = xmalloc(sizeof(*p) + path_len + 2);
585 * Make sure a corresponding .pack file exists and that
586 * the index looks sane.
588 path_len -= strlen(".idx");
589 if (path_len < 1)
590 return NULL;
591 memcpy(p->pack_name, path, path_len);
592 strcpy(p->pack_name + path_len, ".pack");
593 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode) ||
594 check_packed_git_idx(path, p)) {
595 free(p);
596 return NULL;
599 /* ok, it looks sane as far as we can check without
600 * actually mapping the pack file.
602 p->pack_size = st.st_size;
603 p->next = NULL;
604 p->pack_base = NULL;
605 p->pack_last_used = 0;
606 p->pack_use_cnt = 0;
607 p->pack_local = local;
608 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
609 hashclr(p->sha1);
610 return p;
613 struct packed_git *parse_pack_index(unsigned char *sha1)
615 char *path = sha1_pack_index_name(sha1);
616 return parse_pack_index_file(sha1, path);
619 struct packed_git *parse_pack_index_file(const unsigned char *sha1,
620 const char *idx_path)
622 const char *path = sha1_pack_name(sha1);
623 struct packed_git *p = xmalloc(sizeof(*p) + strlen(path) + 2);
625 if (check_packed_git_idx(idx_path, p)) {
626 free(p);
627 return NULL;
630 strcpy(p->pack_name, path);
631 p->pack_size = 0;
632 p->next = NULL;
633 p->pack_base = NULL;
634 p->pack_last_used = 0;
635 p->pack_use_cnt = 0;
636 hashcpy(p->sha1, sha1);
637 return p;
640 void install_packed_git(struct packed_git *pack)
642 pack->next = packed_git;
643 packed_git = pack;
646 static void prepare_packed_git_one(char *objdir, int local)
648 char path[PATH_MAX];
649 int len;
650 DIR *dir;
651 struct dirent *de;
653 sprintf(path, "%s/pack", objdir);
654 len = strlen(path);
655 dir = opendir(path);
656 if (!dir) {
657 if (errno != ENOENT)
658 error("unable to open object pack directory: %s: %s",
659 path, strerror(errno));
660 return;
662 path[len++] = '/';
663 while ((de = readdir(dir)) != NULL) {
664 int namelen = strlen(de->d_name);
665 struct packed_git *p;
667 if (!has_extension(de->d_name, ".idx"))
668 continue;
670 /* we have .idx. Is it a file we can map? */
671 strcpy(path + len, de->d_name);
672 for (p = packed_git; p; p = p->next) {
673 if (!memcmp(path, p->pack_name, len + namelen - 4))
674 break;
676 if (p)
677 continue;
678 p = add_packed_git(path, len + namelen, local);
679 if (!p)
680 continue;
681 p->next = packed_git;
682 packed_git = p;
684 closedir(dir);
687 static int prepare_packed_git_run_once = 0;
688 void prepare_packed_git(void)
690 struct alternate_object_database *alt;
692 if (prepare_packed_git_run_once)
693 return;
694 prepare_packed_git_one(get_object_directory(), 1);
695 prepare_alt_odb();
696 for (alt = alt_odb_list; alt; alt = alt->next) {
697 alt->name[-1] = 0;
698 prepare_packed_git_one(alt->base, 0);
699 alt->name[-1] = '/';
701 prepare_packed_git_run_once = 1;
704 void reprepare_packed_git(void)
706 prepare_packed_git_run_once = 0;
707 prepare_packed_git();
710 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
712 unsigned char real_sha1[20];
713 hash_sha1_file(map, size, type, real_sha1);
714 return hashcmp(sha1, real_sha1) ? -1 : 0;
717 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
719 struct stat st;
720 void *map;
721 int fd;
722 char *filename = find_sha1_file(sha1, &st);
724 if (!filename) {
725 return NULL;
728 fd = open(filename, O_RDONLY | sha1_file_open_flag);
729 if (fd < 0) {
730 /* See if it works without O_NOATIME */
731 switch (sha1_file_open_flag) {
732 default:
733 fd = open(filename, O_RDONLY);
734 if (fd >= 0)
735 break;
736 /* Fallthrough */
737 case 0:
738 return NULL;
741 /* If it failed once, it will probably fail again.
742 * Stop using O_NOATIME
744 sha1_file_open_flag = 0;
746 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
747 close(fd);
748 if (map == MAP_FAILED)
749 return NULL;
750 *size = st.st_size;
751 return map;
754 int legacy_loose_object(unsigned char *map)
756 unsigned int word;
759 * Is it a zlib-compressed buffer? If so, the first byte
760 * must be 0x78 (15-bit window size, deflated), and the
761 * first 16-bit word is evenly divisible by 31
763 word = (map[0] << 8) + map[1];
764 if (map[0] == 0x78 && !(word % 31))
765 return 1;
766 else
767 return 0;
770 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
772 unsigned shift;
773 unsigned char c;
774 unsigned long size;
775 unsigned long used = 0;
777 c = buf[used++];
778 *type = (c >> 4) & 7;
779 size = c & 15;
780 shift = 4;
781 while (c & 0x80) {
782 if (len <= used)
783 return 0;
784 if (sizeof(long) * 8 <= shift)
785 return 0;
786 c = buf[used++];
787 size += (c & 0x7f) << shift;
788 shift += 7;
790 *sizep = size;
791 return used;
794 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
796 unsigned long size, used;
797 static const char valid_loose_object_type[8] = {
798 0, /* OBJ_EXT */
799 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
800 0, /* "delta" and others are invalid in a loose object */
802 enum object_type type;
804 /* Get the data stream */
805 memset(stream, 0, sizeof(*stream));
806 stream->next_in = map;
807 stream->avail_in = mapsize;
808 stream->next_out = buffer;
809 stream->avail_out = bufsiz;
811 if (legacy_loose_object(map)) {
812 inflateInit(stream);
813 return inflate(stream, 0);
816 used = unpack_object_header_gently(map, mapsize, &type, &size);
817 if (!used || !valid_loose_object_type[type])
818 return -1;
819 map += used;
820 mapsize -= used;
822 /* Set up the stream for the rest.. */
823 stream->next_in = map;
824 stream->avail_in = mapsize;
825 inflateInit(stream);
827 /* And generate the fake traditional header */
828 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
829 type_names[type], size);
830 return 0;
833 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
835 int bytes = strlen(buffer) + 1;
836 unsigned char *buf = xmalloc(1+size);
837 unsigned long n;
839 n = stream->total_out - bytes;
840 if (n > size)
841 n = size;
842 memcpy(buf, (char *) buffer + bytes, n);
843 bytes = n;
844 if (bytes < size) {
845 stream->next_out = buf + bytes;
846 stream->avail_out = size - bytes;
847 while (inflate(stream, Z_FINISH) == Z_OK)
848 /* nothing */;
850 buf[size] = 0;
851 inflateEnd(stream);
852 return buf;
856 * We used to just use "sscanf()", but that's actually way
857 * too permissive for what we want to check. So do an anal
858 * object header parse by hand.
860 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
862 int i;
863 unsigned long size;
866 * The type can be at most ten bytes (including the
867 * terminating '\0' that we add), and is followed by
868 * a space.
870 i = 10;
871 for (;;) {
872 char c = *hdr++;
873 if (c == ' ')
874 break;
875 if (!--i)
876 return -1;
877 *type++ = c;
879 *type = 0;
882 * The length must follow immediately, and be in canonical
883 * decimal format (ie "010" is not valid).
885 size = *hdr++ - '0';
886 if (size > 9)
887 return -1;
888 if (size) {
889 for (;;) {
890 unsigned long c = *hdr - '0';
891 if (c > 9)
892 break;
893 hdr++;
894 size = size * 10 + c;
897 *sizep = size;
900 * The length must be followed by a zero byte
902 return *hdr ? -1 : 0;
905 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
907 int ret;
908 z_stream stream;
909 char hdr[8192];
911 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
912 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
913 return NULL;
915 return unpack_sha1_rest(&stream, hdr, *size);
918 static unsigned long get_delta_base(struct packed_git *p,
919 unsigned long offset,
920 enum object_type kind,
921 unsigned long delta_obj_offset,
922 unsigned long *base_obj_offset)
924 unsigned char *base_info = (unsigned char *) p->pack_base + offset;
925 unsigned long base_offset;
927 /* there must be at least 20 bytes left regardless of delta type */
928 if (p->pack_size <= offset + 20)
929 die("truncated pack file");
931 if (kind == OBJ_OFS_DELTA) {
932 unsigned used = 0;
933 unsigned char c = base_info[used++];
934 base_offset = c & 127;
935 while (c & 128) {
936 base_offset += 1;
937 if (!base_offset || base_offset & ~(~0UL >> 7))
938 die("offset value overflow for delta base object");
939 c = base_info[used++];
940 base_offset = (base_offset << 7) + (c & 127);
942 base_offset = delta_obj_offset - base_offset;
943 if (base_offset >= delta_obj_offset)
944 die("delta base offset out of bound");
945 offset += used;
946 } else if (kind == OBJ_REF_DELTA) {
947 /* The base entry _must_ be in the same pack */
948 base_offset = find_pack_entry_one(base_info, p);
949 if (!base_offset)
950 die("failed to find delta-pack base object %s",
951 sha1_to_hex(base_info));
952 offset += 20;
953 } else
954 die("I am totally screwed");
955 *base_obj_offset = base_offset;
956 return offset;
959 /* forward declaration for a mutually recursive function */
960 static int packed_object_info(struct packed_git *p, unsigned long offset,
961 char *type, unsigned long *sizep);
963 static int packed_delta_info(struct packed_git *p,
964 unsigned long offset,
965 enum object_type kind,
966 unsigned long obj_offset,
967 char *type,
968 unsigned long *sizep)
970 unsigned long base_offset;
972 offset = get_delta_base(p, offset, kind, obj_offset, &base_offset);
974 /* We choose to only get the type of the base object and
975 * ignore potentially corrupt pack file that expects the delta
976 * based on a base with a wrong size. This saves tons of
977 * inflate() calls.
979 if (packed_object_info(p, base_offset, type, NULL))
980 die("cannot get info for delta-pack base");
982 if (sizep) {
983 const unsigned char *data;
984 unsigned char delta_head[20];
985 unsigned long result_size;
986 z_stream stream;
987 int st;
989 memset(&stream, 0, sizeof(stream));
991 stream.next_in = (unsigned char *) p->pack_base + offset;
992 stream.avail_in = p->pack_size - offset;
993 stream.next_out = delta_head;
994 stream.avail_out = sizeof(delta_head);
996 inflateInit(&stream);
997 st = inflate(&stream, Z_FINISH);
998 inflateEnd(&stream);
999 if ((st != Z_STREAM_END) &&
1000 stream.total_out != sizeof(delta_head))
1001 die("delta data unpack-initial failed");
1003 /* Examine the initial part of the delta to figure out
1004 * the result size.
1006 data = delta_head;
1008 /* ignore base size */
1009 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1011 /* Read the result size */
1012 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1013 *sizep = result_size;
1015 return 0;
1018 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
1019 enum object_type *type, unsigned long *sizep)
1021 unsigned long used;
1023 if (p->pack_size <= offset)
1024 die("object offset outside of pack file");
1026 used = unpack_object_header_gently((unsigned char *)p->pack_base +
1027 offset,
1028 p->pack_size - offset, type, sizep);
1029 if (!used)
1030 die("object offset outside of pack file");
1032 return offset + used;
1035 void packed_object_info_detail(struct packed_git *p,
1036 unsigned long offset,
1037 char *type,
1038 unsigned long *size,
1039 unsigned long *store_size,
1040 unsigned int *delta_chain_length,
1041 unsigned char *base_sha1)
1043 unsigned long obj_offset, val;
1044 unsigned char *next_sha1;
1045 enum object_type kind;
1047 *delta_chain_length = 0;
1048 obj_offset = offset;
1049 offset = unpack_object_header(p, offset, &kind, size);
1051 for (;;) {
1052 switch (kind) {
1053 default:
1054 die("corrupted pack file %s containing object of kind %d",
1055 p->pack_name, kind);
1056 case OBJ_COMMIT:
1057 case OBJ_TREE:
1058 case OBJ_BLOB:
1059 case OBJ_TAG:
1060 strcpy(type, type_names[kind]);
1061 *store_size = 0; /* notyet */
1062 return;
1063 case OBJ_OFS_DELTA:
1064 get_delta_base(p, offset, kind, obj_offset, &offset);
1065 if (*delta_chain_length == 0) {
1066 /* TODO: find base_sha1 as pointed by offset */
1068 break;
1069 case OBJ_REF_DELTA:
1070 if (p->pack_size <= offset + 20)
1071 die("pack file %s records an incomplete delta base",
1072 p->pack_name);
1073 next_sha1 = (unsigned char *) p->pack_base + offset;
1074 if (*delta_chain_length == 0)
1075 hashcpy(base_sha1, next_sha1);
1076 offset = find_pack_entry_one(next_sha1, p);
1077 break;
1079 obj_offset = offset;
1080 offset = unpack_object_header(p, offset, &kind, &val);
1081 (*delta_chain_length)++;
1085 static int packed_object_info(struct packed_git *p, unsigned long offset,
1086 char *type, unsigned long *sizep)
1088 unsigned long size, obj_offset = offset;
1089 enum object_type kind;
1091 offset = unpack_object_header(p, offset, &kind, &size);
1093 switch (kind) {
1094 case OBJ_OFS_DELTA:
1095 case OBJ_REF_DELTA:
1096 return packed_delta_info(p, offset, kind, obj_offset, type, sizep);
1097 case OBJ_COMMIT:
1098 case OBJ_TREE:
1099 case OBJ_BLOB:
1100 case OBJ_TAG:
1101 strcpy(type, type_names[kind]);
1102 break;
1103 default:
1104 die("corrupted pack file %s containing object of kind %d",
1105 p->pack_name, kind);
1107 if (sizep)
1108 *sizep = size;
1109 return 0;
1112 static void *unpack_compressed_entry(struct packed_git *p,
1113 unsigned long offset,
1114 unsigned long size)
1116 int st;
1117 z_stream stream;
1118 unsigned char *buffer;
1120 buffer = xmalloc(size + 1);
1121 buffer[size] = 0;
1122 memset(&stream, 0, sizeof(stream));
1123 stream.next_in = (unsigned char*)p->pack_base + offset;
1124 stream.avail_in = p->pack_size - offset;
1125 stream.next_out = buffer;
1126 stream.avail_out = size;
1128 inflateInit(&stream);
1129 st = inflate(&stream, Z_FINISH);
1130 inflateEnd(&stream);
1131 if ((st != Z_STREAM_END) || stream.total_out != size) {
1132 free(buffer);
1133 return NULL;
1136 return buffer;
1139 static void *unpack_delta_entry(struct packed_git *p,
1140 unsigned long offset,
1141 unsigned long delta_size,
1142 enum object_type kind,
1143 unsigned long obj_offset,
1144 char *type,
1145 unsigned long *sizep)
1147 void *delta_data, *result, *base;
1148 unsigned long result_size, base_size, base_offset;
1150 offset = get_delta_base(p, offset, kind, obj_offset, &base_offset);
1151 base = unpack_entry_gently(p, base_offset, type, &base_size);
1152 if (!base)
1153 die("failed to read delta base object at %lu from %s",
1154 base_offset, p->pack_name);
1156 delta_data = unpack_compressed_entry(p, offset, delta_size);
1157 result = patch_delta(base, base_size,
1158 delta_data, delta_size,
1159 &result_size);
1160 if (!result)
1161 die("failed to apply delta");
1162 free(delta_data);
1163 free(base);
1164 *sizep = result_size;
1165 return result;
1168 static void *unpack_entry(struct pack_entry *entry,
1169 char *type, unsigned long *sizep)
1171 struct packed_git *p = entry->p;
1172 void *retval;
1174 if (use_packed_git(p))
1175 die("cannot map packed file");
1176 retval = unpack_entry_gently(p, entry->offset, type, sizep);
1177 unuse_packed_git(p);
1178 if (!retval)
1179 die("corrupted pack file %s", p->pack_name);
1180 return retval;
1183 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1184 void *unpack_entry_gently(struct packed_git *p, unsigned long offset,
1185 char *type, unsigned long *sizep)
1187 unsigned long size, obj_offset = offset;
1188 enum object_type kind;
1190 offset = unpack_object_header(p, offset, &kind, &size);
1191 switch (kind) {
1192 case OBJ_OFS_DELTA:
1193 case OBJ_REF_DELTA:
1194 return unpack_delta_entry(p, offset, size, kind, obj_offset, type, sizep);
1195 case OBJ_COMMIT:
1196 case OBJ_TREE:
1197 case OBJ_BLOB:
1198 case OBJ_TAG:
1199 strcpy(type, type_names[kind]);
1200 *sizep = size;
1201 return unpack_compressed_entry(p, offset, size);
1202 default:
1203 return NULL;
1207 const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
1208 unsigned int n)
1210 const unsigned char *index = p->index_data;
1211 if (n >= p->num_objects)
1212 return NULL;
1213 index += 4 * 256;
1214 if (p->index_version == 1) {
1215 return index + 24 * n + 4;
1216 } else {
1217 index += 8;
1218 return index + 20 * n;
1222 static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1224 const unsigned char *index = p->index_data;
1225 index += 4 * 256;
1226 if (p->index_version == 1) {
1227 return ntohl(*((uint32_t *)(index + 24 * n)));
1228 } else {
1229 uint32_t off;
1230 index += 8 + p->num_objects * (20 + 4);
1231 off = ntohl(*((uint32_t *)(index + 4 * n)));
1232 if (!(off & 0x80000000))
1233 return off;
1234 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1235 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1236 ntohl(*((uint32_t *)(index + 4)));
1240 unsigned long find_pack_entry_one(const unsigned char *sha1,
1241 struct packed_git *p)
1243 const unsigned int *level1_ofs = p->index_data;
1244 const unsigned char *index = p->index_data;
1245 unsigned hi, lo;
1247 if (p->index_version > 1) {
1248 level1_ofs += 2;
1249 index += 8;
1251 index += 4 * 256;
1252 hi = ntohl(level1_ofs[*sha1]);
1253 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1255 do {
1256 unsigned mi = (lo + hi) / 2;
1257 unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
1258 int cmp = hashcmp(index + x, sha1);
1259 if (!cmp)
1260 return nth_packed_object_offset(p, mi);
1261 if (cmp > 0)
1262 hi = mi;
1263 else
1264 lo = mi+1;
1265 } while (lo < hi);
1266 return 0;
1269 static int matches_pack_name(struct packed_git *p, const char *ig)
1271 const char *last_c, *c;
1273 if (!strcmp(p->pack_name, ig))
1274 return 0;
1276 for (c = p->pack_name, last_c = c; *c;)
1277 if (*c == '/')
1278 last_c = ++c;
1279 else
1280 ++c;
1281 if (!strcmp(last_c, ig))
1282 return 0;
1284 return 1;
1287 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1289 struct packed_git *p;
1290 unsigned long offset;
1292 prepare_packed_git();
1294 for (p = packed_git; p; p = p->next) {
1295 if (ignore_packed) {
1296 const char **ig;
1297 for (ig = ignore_packed; *ig; ig++)
1298 if (!matches_pack_name(p, *ig))
1299 break;
1300 if (*ig)
1301 continue;
1303 offset = find_pack_entry_one(sha1, p);
1304 if (offset) {
1305 e->offset = offset;
1306 e->p = p;
1307 hashcpy(e->sha1, sha1);
1308 return 1;
1311 return 0;
1314 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1315 struct packed_git *packs)
1317 struct packed_git *p;
1319 for (p = packs; p; p = p->next) {
1320 if (find_pack_entry_one(sha1, p))
1321 return p;
1323 return NULL;
1327 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1329 int status;
1330 unsigned long mapsize, size;
1331 void *map;
1332 z_stream stream;
1333 char hdr[128];
1335 map = map_sha1_file(sha1, &mapsize);
1336 if (!map) {
1337 struct pack_entry e;
1339 if (!find_pack_entry(sha1, &e, NULL)) {
1340 reprepare_packed_git();
1341 if (!find_pack_entry(sha1, &e, NULL))
1342 return error("unable to find %s", sha1_to_hex(sha1));
1344 if (use_packed_git(e.p))
1345 die("cannot map packed file");
1346 status = packed_object_info(e.p, e.offset, type, sizep);
1347 unuse_packed_git(e.p);
1348 return status;
1350 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1351 status = error("unable to unpack %s header",
1352 sha1_to_hex(sha1));
1353 if (parse_sha1_header(hdr, type, &size) < 0)
1354 status = error("unable to parse %s header", sha1_to_hex(sha1));
1355 else {
1356 status = 0;
1357 if (sizep)
1358 *sizep = size;
1360 inflateEnd(&stream);
1361 munmap(map, mapsize);
1362 return status;
1365 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1367 struct pack_entry e;
1369 if (!find_pack_entry(sha1, &e, NULL)) {
1370 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1371 return NULL;
1373 return unpack_entry(&e, type, size);
1376 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1378 unsigned long mapsize;
1379 void *map, *buf;
1380 struct pack_entry e;
1382 if (find_pack_entry(sha1, &e, NULL))
1383 return read_packed_sha1(sha1, type, size);
1384 map = map_sha1_file(sha1, &mapsize);
1385 if (map) {
1386 buf = unpack_sha1_file(map, mapsize, type, size);
1387 munmap(map, mapsize);
1388 return buf;
1390 reprepare_packed_git();
1391 if (find_pack_entry(sha1, &e, NULL))
1392 return read_packed_sha1(sha1, type, size);
1393 return NULL;
1396 void *read_object_with_reference(const unsigned char *sha1,
1397 const char *required_type,
1398 unsigned long *size,
1399 unsigned char *actual_sha1_return)
1401 char type[20];
1402 void *buffer;
1403 unsigned long isize;
1404 unsigned char actual_sha1[20];
1406 hashcpy(actual_sha1, sha1);
1407 while (1) {
1408 int ref_length = -1;
1409 const char *ref_type = NULL;
1411 buffer = read_sha1_file(actual_sha1, type, &isize);
1412 if (!buffer)
1413 return NULL;
1414 if (!strcmp(type, required_type)) {
1415 *size = isize;
1416 if (actual_sha1_return)
1417 hashcpy(actual_sha1_return, actual_sha1);
1418 return buffer;
1420 /* Handle references */
1421 else if (!strcmp(type, commit_type))
1422 ref_type = "tree ";
1423 else if (!strcmp(type, tag_type))
1424 ref_type = "object ";
1425 else {
1426 free(buffer);
1427 return NULL;
1429 ref_length = strlen(ref_type);
1431 if (memcmp(buffer, ref_type, ref_length) ||
1432 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1433 free(buffer);
1434 return NULL;
1436 free(buffer);
1437 /* Now we have the ID of the referred-to object in
1438 * actual_sha1. Check again. */
1442 static void write_sha1_file_prepare(void *buf, unsigned long len,
1443 const char *type, unsigned char *sha1,
1444 unsigned char *hdr, int *hdrlen)
1446 SHA_CTX c;
1448 /* Generate the header */
1449 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1451 /* Sha1.. */
1452 SHA1_Init(&c);
1453 SHA1_Update(&c, hdr, *hdrlen);
1454 SHA1_Update(&c, buf, len);
1455 SHA1_Final(sha1, &c);
1459 * Link the tempfile to the final place, possibly creating the
1460 * last directory level as you do so.
1462 * Returns the errno on failure, 0 on success.
1464 static int link_temp_to_file(const char *tmpfile, const char *filename)
1466 int ret;
1467 char *dir;
1469 if (!link(tmpfile, filename))
1470 return 0;
1473 * Try to mkdir the last path component if that failed.
1475 * Re-try the "link()" regardless of whether the mkdir
1476 * succeeds, since a race might mean that somebody
1477 * else succeeded.
1479 ret = errno;
1480 dir = strrchr(filename, '/');
1481 if (dir) {
1482 *dir = 0;
1483 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1484 *dir = '/';
1485 return -2;
1487 *dir = '/';
1488 if (!link(tmpfile, filename))
1489 return 0;
1490 ret = errno;
1492 return ret;
1496 * Move the just written object into its final resting place
1498 int move_temp_to_file(const char *tmpfile, const char *filename)
1500 int ret = link_temp_to_file(tmpfile, filename);
1503 * Coda hack - coda doesn't like cross-directory links,
1504 * so we fall back to a rename, which will mean that it
1505 * won't be able to check collisions, but that's not a
1506 * big deal.
1508 * The same holds for FAT formatted media.
1510 * When this succeeds, we just return 0. We have nothing
1511 * left to unlink.
1513 if (ret && ret != EEXIST) {
1514 if (!rename(tmpfile, filename))
1515 return 0;
1516 ret = errno;
1518 unlink(tmpfile);
1519 if (ret) {
1520 if (ret != EEXIST) {
1521 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1523 /* FIXME!!! Collision check here ? */
1526 return 0;
1529 static int write_buffer(int fd, const void *buf, size_t len)
1531 while (len) {
1532 ssize_t size;
1534 size = write(fd, buf, len);
1535 if (!size)
1536 return error("file write: disk full");
1537 if (size < 0) {
1538 if (errno == EINTR || errno == EAGAIN)
1539 continue;
1540 return error("file write error (%s)", strerror(errno));
1542 len -= size;
1543 buf = (char *) buf + size;
1545 return 0;
1548 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1550 int hdr_len;
1551 unsigned char c;
1553 c = (type << 4) | (len & 15);
1554 len >>= 4;
1555 hdr_len = 1;
1556 while (len) {
1557 *hdr++ = c | 0x80;
1558 hdr_len++;
1559 c = (len & 0x7f);
1560 len >>= 7;
1562 *hdr = c;
1563 return hdr_len;
1566 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1568 int obj_type, hdr;
1570 if (use_legacy_headers) {
1571 while (deflate(stream, 0) == Z_OK)
1572 /* nothing */;
1573 return;
1575 if (!strcmp(type, blob_type))
1576 obj_type = OBJ_BLOB;
1577 else if (!strcmp(type, tree_type))
1578 obj_type = OBJ_TREE;
1579 else if (!strcmp(type, commit_type))
1580 obj_type = OBJ_COMMIT;
1581 else if (!strcmp(type, tag_type))
1582 obj_type = OBJ_TAG;
1583 else
1584 die("trying to generate bogus object of type '%s'", type);
1585 hdr = write_binary_header(stream->next_out, obj_type, len);
1586 stream->total_out = hdr;
1587 stream->next_out += hdr;
1588 stream->avail_out -= hdr;
1591 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1592 unsigned char *sha1)
1594 unsigned char hdr[50];
1595 int hdrlen;
1596 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1597 return 0;
1600 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1602 int size;
1603 unsigned char *compressed;
1604 z_stream stream;
1605 unsigned char sha1[20];
1606 char *filename;
1607 static char tmpfile[PATH_MAX];
1608 unsigned char hdr[50];
1609 int fd, hdrlen;
1611 /* Normally if we have it in the pack then we do not bother writing
1612 * it out into .git/objects/??/?{38} file.
1614 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1615 filename = sha1_file_name(sha1);
1616 if (returnsha1)
1617 hashcpy(returnsha1, sha1);
1618 if (has_sha1_file(sha1))
1619 return 0;
1620 fd = open(filename, O_RDONLY);
1621 if (fd >= 0) {
1623 * FIXME!!! We might do collision checking here, but we'd
1624 * need to uncompress the old file and check it. Later.
1626 close(fd);
1627 return 0;
1630 if (errno != ENOENT) {
1631 return error("sha1 file %s: %s\n", filename, strerror(errno));
1634 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1636 fd = mkstemp(tmpfile);
1637 if (fd < 0) {
1638 if (errno == EPERM)
1639 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1640 else
1641 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1644 /* Set it up */
1645 memset(&stream, 0, sizeof(stream));
1646 deflateInit(&stream, zlib_compression_level);
1647 size = 8 + deflateBound(&stream, len+hdrlen);
1648 compressed = xmalloc(size);
1650 /* Compress it */
1651 stream.next_out = compressed;
1652 stream.avail_out = size;
1654 /* First header.. */
1655 stream.next_in = hdr;
1656 stream.avail_in = hdrlen;
1657 setup_object_header(&stream, type, len);
1659 /* Then the data itself.. */
1660 stream.next_in = buf;
1661 stream.avail_in = len;
1662 while (deflate(&stream, Z_FINISH) == Z_OK)
1663 /* nothing */;
1664 deflateEnd(&stream);
1665 size = stream.total_out;
1667 if (write_buffer(fd, compressed, size) < 0)
1668 die("unable to write sha1 file");
1669 fchmod(fd, 0444);
1670 close(fd);
1671 free(compressed);
1673 return move_temp_to_file(tmpfile, filename);
1677 * We need to unpack and recompress the object for writing
1678 * it out to a different file.
1680 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1682 size_t size;
1683 z_stream stream;
1684 unsigned char *unpacked;
1685 unsigned long len;
1686 char type[20];
1687 char hdr[50];
1688 int hdrlen;
1689 void *buf;
1691 /* need to unpack and recompress it by itself */
1692 unpacked = read_packed_sha1(sha1, type, &len);
1694 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1696 /* Set it up */
1697 memset(&stream, 0, sizeof(stream));
1698 deflateInit(&stream, zlib_compression_level);
1699 size = deflateBound(&stream, len + hdrlen);
1700 buf = xmalloc(size);
1702 /* Compress it */
1703 stream.next_out = buf;
1704 stream.avail_out = size;
1706 /* First header.. */
1707 stream.next_in = (void *)hdr;
1708 stream.avail_in = hdrlen;
1709 while (deflate(&stream, 0) == Z_OK)
1710 /* nothing */;
1712 /* Then the data itself.. */
1713 stream.next_in = unpacked;
1714 stream.avail_in = len;
1715 while (deflate(&stream, Z_FINISH) == Z_OK)
1716 /* nothing */;
1717 deflateEnd(&stream);
1718 free(unpacked);
1720 *objsize = stream.total_out;
1721 return buf;
1724 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1726 int retval;
1727 unsigned long objsize;
1728 void *buf = map_sha1_file(sha1, &objsize);
1730 if (buf) {
1731 retval = write_buffer(fd, buf, objsize);
1732 munmap(buf, objsize);
1733 return retval;
1736 buf = repack_object(sha1, &objsize);
1737 retval = write_buffer(fd, buf, objsize);
1738 free(buf);
1739 return retval;
1742 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1743 size_t bufsize, size_t *bufposn)
1745 char tmpfile[PATH_MAX];
1746 int local;
1747 z_stream stream;
1748 unsigned char real_sha1[20];
1749 unsigned char discard[4096];
1750 int ret;
1751 SHA_CTX c;
1753 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1755 local = mkstemp(tmpfile);
1756 if (local < 0) {
1757 if (errno == EPERM)
1758 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1759 else
1760 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1763 memset(&stream, 0, sizeof(stream));
1765 inflateInit(&stream);
1767 SHA1_Init(&c);
1769 do {
1770 ssize_t size;
1771 if (*bufposn) {
1772 stream.avail_in = *bufposn;
1773 stream.next_in = (unsigned char *) buffer;
1774 do {
1775 stream.next_out = discard;
1776 stream.avail_out = sizeof(discard);
1777 ret = inflate(&stream, Z_SYNC_FLUSH);
1778 SHA1_Update(&c, discard, sizeof(discard) -
1779 stream.avail_out);
1780 } while (stream.avail_in && ret == Z_OK);
1781 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1782 die("unable to write sha1 file");
1783 memmove(buffer, buffer + *bufposn - stream.avail_in,
1784 stream.avail_in);
1785 *bufposn = stream.avail_in;
1786 if (ret != Z_OK)
1787 break;
1789 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1790 if (size <= 0) {
1791 close(local);
1792 unlink(tmpfile);
1793 if (!size)
1794 return error("Connection closed?");
1795 perror("Reading from connection");
1796 return -1;
1798 *bufposn += size;
1799 } while (1);
1800 inflateEnd(&stream);
1802 close(local);
1803 SHA1_Final(real_sha1, &c);
1804 if (ret != Z_STREAM_END) {
1805 unlink(tmpfile);
1806 return error("File %s corrupted", sha1_to_hex(sha1));
1808 if (hashcmp(sha1, real_sha1)) {
1809 unlink(tmpfile);
1810 return error("File %s has bad hash", sha1_to_hex(sha1));
1813 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1816 int has_pack_index(const unsigned char *sha1)
1818 struct stat st;
1819 if (stat(sha1_pack_index_name(sha1), &st))
1820 return 0;
1821 return 1;
1824 int has_pack_file(const unsigned char *sha1)
1826 struct stat st;
1827 if (stat(sha1_pack_name(sha1), &st))
1828 return 0;
1829 return 1;
1832 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1834 struct pack_entry e;
1835 return find_pack_entry(sha1, &e, ignore_packed);
1838 int has_sha1_file(const unsigned char *sha1)
1840 struct stat st;
1841 struct pack_entry e;
1843 if (find_pack_entry(sha1, &e, NULL))
1844 return 1;
1845 return find_sha1_file(sha1, &st) ? 1 : 0;
1849 * reads from fd as long as possible into a supplied buffer of size bytes.
1850 * If necessary the buffer's size is increased using realloc()
1852 * returns 0 if anything went fine and -1 otherwise
1854 * NOTE: both buf and size may change, but even when -1 is returned
1855 * you still have to free() it yourself.
1857 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1859 char* buf = *return_buf;
1860 unsigned long size = *return_size;
1861 int iret;
1862 unsigned long off = 0;
1864 do {
1865 iret = xread(fd, buf + off, size - off);
1866 if (iret > 0) {
1867 off += iret;
1868 if (off == size) {
1869 size *= 2;
1870 buf = xrealloc(buf, size);
1873 } while (iret > 0);
1875 *return_buf = buf;
1876 *return_size = off;
1878 if (iret < 0)
1879 return -1;
1880 return 0;
1883 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1885 unsigned long size = 4096;
1886 char *buf = xmalloc(size);
1887 int ret;
1889 if (read_pipe(fd, &buf, &size)) {
1890 free(buf);
1891 return -1;
1894 if (!type)
1895 type = blob_type;
1896 if (write_object)
1897 ret = write_sha1_file(buf, size, type, sha1);
1898 else
1899 ret = hash_sha1_file(buf, size, type, sha1);
1900 free(buf);
1901 return ret;
1904 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1906 unsigned long size = st->st_size;
1907 void *buf;
1908 int ret;
1910 buf = "";
1911 if (size)
1912 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1913 close(fd);
1914 if (buf == MAP_FAILED)
1915 return -1;
1917 if (!type)
1918 type = blob_type;
1919 if (write_object)
1920 ret = write_sha1_file(buf, size, type, sha1);
1921 else
1922 ret = hash_sha1_file(buf, size, type, sha1);
1923 if (size)
1924 munmap(buf, size);
1925 return ret;
1928 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1930 int fd;
1931 char *target;
1933 switch (st->st_mode & S_IFMT) {
1934 case S_IFREG:
1935 fd = open(path, O_RDONLY);
1936 if (fd < 0)
1937 return error("open(\"%s\"): %s", path,
1938 strerror(errno));
1939 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1940 return error("%s: failed to insert into database",
1941 path);
1942 break;
1943 case S_IFLNK:
1944 target = xmalloc(st->st_size+1);
1945 if (readlink(path, target, st->st_size+1) != st->st_size) {
1946 char *errstr = strerror(errno);
1947 free(target);
1948 return error("readlink(\"%s\"): %s", path,
1949 errstr);
1951 if (!write_object)
1952 hash_sha1_file(target, st->st_size, blob_type, sha1);
1953 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1954 return error("%s: failed to insert into database",
1955 path);
1956 free(target);
1957 break;
1958 default:
1959 return error("%s: unsupported file type", path);
1961 return 0;