Load core configuration in git-verify-pack.
[git.git] / sha1_file.c
blob548535c8448430457ae9dd4e1507166efbea8551
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 static unsigned int pack_used_ctr;
401 static size_t pack_mapped;
402 static size_t page_size;
403 struct packed_git *packed_git;
405 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
406 void **idx_map_)
408 void *idx_map;
409 unsigned int *index;
410 unsigned long idx_size;
411 int nr, i;
412 int fd = open(path, O_RDONLY);
413 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 index = idx_map;
427 *idx_map_ = idx_map;
428 *idx_size_ = idx_size;
430 /* check index map */
431 if (idx_size < 4*256 + 20 + 20)
432 return error("index file too small");
433 nr = 0;
434 for (i = 0; i < 256; i++) {
435 unsigned int n = ntohl(index[i]);
436 if (n < nr)
437 return error("non-monotonic index");
438 nr = n;
442 * Total size:
443 * - 256 index entries 4 bytes each
444 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
445 * - 20-byte SHA1 of the packfile
446 * - 20-byte SHA1 file checksum
448 if (idx_size != 4*256 + nr * 24 + 20 + 20)
449 return error("wrong index file size");
451 return 0;
454 static int unuse_one_window(void)
456 struct packed_git *p, *lru_p = NULL;
457 struct pack_window *w, *w_l, *lru_w = NULL, *lru_l = NULL;
459 for (p = packed_git; p; p = p->next) {
460 for (w_l = NULL, w = p->windows; w; w = w->next) {
461 if (!w->inuse_cnt) {
462 if (!lru_w || w->last_used < lru_w->last_used) {
463 lru_p = p;
464 lru_w = w;
465 lru_l = w_l;
468 w_l = w;
471 if (lru_p) {
472 munmap(lru_w->base, lru_w->len);
473 pack_mapped -= lru_w->len;
474 if (lru_l)
475 lru_l->next = lru_w->next;
476 else {
477 lru_p->windows = lru_w->next;
478 if (!lru_p->windows) {
479 close(lru_p->pack_fd);
480 lru_p->pack_fd = -1;
483 free(lru_w);
484 return 1;
486 return 0;
489 void unuse_pack(struct pack_window **w_cursor)
491 struct pack_window *w = *w_cursor;
492 if (w) {
493 w->inuse_cnt--;
494 *w_cursor = NULL;
498 static void open_packed_git(struct packed_git *p)
500 struct stat st;
501 struct pack_header hdr;
502 unsigned char sha1[20];
503 unsigned char *idx_sha1;
505 p->pack_fd = open(p->pack_name, O_RDONLY);
506 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
507 die("packfile %s cannot be opened", p->pack_name);
509 /* If we created the struct before we had the pack we lack size. */
510 if (!p->pack_size) {
511 if (!S_ISREG(st.st_mode))
512 die("packfile %s not a regular file", p->pack_name);
513 p->pack_size = st.st_size;
514 } else if (p->pack_size != st.st_size)
515 die("packfile %s size changed", p->pack_name);
517 /* Verify we recognize this pack file format. */
518 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
519 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
520 die("file %s is not a GIT packfile", p->pack_name);
521 if (!pack_version_ok(hdr.hdr_version))
522 die("packfile %s is version %u and not supported"
523 " (try upgrading GIT to a newer version)",
524 p->pack_name, ntohl(hdr.hdr_version));
526 /* Verify the pack matches its index. */
527 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
528 die("packfile %s claims to have %u objects"
529 " while index size indicates %u objects",
530 p->pack_name, ntohl(hdr.hdr_entries),
531 num_packed_objects(p));
532 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
533 die("end of packfile %s is unavailable", p->pack_name);
534 read_or_die(p->pack_fd, sha1, sizeof(sha1));
535 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
536 if (hashcmp(sha1, idx_sha1))
537 die("packfile %s does not match index", p->pack_name);
540 static int in_window(struct pack_window *win, unsigned long offset)
542 /* We must promise at least 20 bytes (one hash) after the
543 * offset is available from this window, otherwise the offset
544 * is not actually in this window and a different window (which
545 * has that one hash excess) must be used. This is to support
546 * the object header and delta base parsing routines below.
548 off_t win_off = win->offset;
549 return win_off <= offset
550 && (offset + 20) <= (win_off + win->len);
553 unsigned char* use_pack(struct packed_git *p,
554 struct pack_window **w_cursor,
555 unsigned long offset,
556 unsigned int *left)
558 struct pack_window *win = *w_cursor;
560 if (p->pack_fd == -1)
561 open_packed_git(p);
563 /* Since packfiles end in a hash of their content and its
564 * pointless to ask for an offset into the middle of that
565 * hash, and the in_window function above wouldn't match
566 * don't allow an offset too close to the end of the file.
568 if (offset > (p->pack_size - 20))
569 die("offset beyond end of packfile (truncated pack?)");
571 if (!win || !in_window(win, offset)) {
572 if (win)
573 win->inuse_cnt--;
574 for (win = p->windows; win; win = win->next) {
575 if (in_window(win, offset))
576 break;
578 if (!win) {
579 if (!page_size)
580 page_size = getpagesize();
581 win = xcalloc(1, sizeof(*win));
582 win->offset = (offset / page_size) * page_size;
583 win->len = p->pack_size - win->offset;
584 if (win->len > packed_git_window_size)
585 win->len = packed_git_window_size;
586 pack_mapped += win->len;
587 while (packed_git_limit < pack_mapped && unuse_one_window())
588 ; /* nothing */
589 win->base = mmap(NULL, win->len,
590 PROT_READ, MAP_PRIVATE,
591 p->pack_fd, win->offset);
592 if (win->base == MAP_FAILED)
593 die("packfile %s cannot be mapped.", p->pack_name);
594 win->next = p->windows;
595 p->windows = win;
598 if (win != *w_cursor) {
599 win->last_used = pack_used_ctr++;
600 win->inuse_cnt++;
601 *w_cursor = win;
603 offset -= win->offset;
604 if (left)
605 *left = win->len - offset;
606 return win->base + offset;
609 struct packed_git *add_packed_git(char *path, int path_len, int local)
611 struct stat st;
612 struct packed_git *p;
613 unsigned long idx_size;
614 void *idx_map;
615 unsigned char sha1[20];
617 if (check_packed_git_idx(path, &idx_size, &idx_map))
618 return NULL;
620 /* do we have a corresponding .pack file? */
621 strcpy(path + path_len - 4, ".pack");
622 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
623 munmap(idx_map, idx_size);
624 return NULL;
626 /* ok, it looks sane as far as we can check without
627 * actually mapping the pack file.
629 p = xmalloc(sizeof(*p) + path_len + 2);
630 strcpy(p->pack_name, path);
631 p->index_size = idx_size;
632 p->pack_size = st.st_size;
633 p->index_base = idx_map;
634 p->next = NULL;
635 p->windows = NULL;
636 p->pack_fd = -1;
637 p->pack_local = local;
638 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
639 hashcpy(p->sha1, sha1);
640 return p;
643 struct packed_git *parse_pack_index(unsigned char *sha1)
645 char *path = sha1_pack_index_name(sha1);
646 return parse_pack_index_file(sha1, path);
649 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
651 struct packed_git *p;
652 unsigned long idx_size;
653 void *idx_map;
654 char *path;
656 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
657 return NULL;
659 path = sha1_pack_name(sha1);
661 p = xmalloc(sizeof(*p) + strlen(path) + 2);
662 strcpy(p->pack_name, path);
663 p->index_size = idx_size;
664 p->pack_size = 0;
665 p->index_base = idx_map;
666 p->next = NULL;
667 p->windows = NULL;
668 p->pack_fd = -1;
669 hashcpy(p->sha1, sha1);
670 return p;
673 void install_packed_git(struct packed_git *pack)
675 pack->next = packed_git;
676 packed_git = pack;
679 static void prepare_packed_git_one(char *objdir, int local)
681 char path[PATH_MAX];
682 int len;
683 DIR *dir;
684 struct dirent *de;
686 sprintf(path, "%s/pack", objdir);
687 len = strlen(path);
688 dir = opendir(path);
689 if (!dir) {
690 if (errno != ENOENT)
691 error("unable to open object pack directory: %s: %s",
692 path, strerror(errno));
693 return;
695 path[len++] = '/';
696 while ((de = readdir(dir)) != NULL) {
697 int namelen = strlen(de->d_name);
698 struct packed_git *p;
700 if (!has_extension(de->d_name, ".idx"))
701 continue;
703 /* we have .idx. Is it a file we can map? */
704 strcpy(path + len, de->d_name);
705 for (p = packed_git; p; p = p->next) {
706 if (!memcmp(path, p->pack_name, len + namelen - 4))
707 break;
709 if (p)
710 continue;
711 p = add_packed_git(path, len + namelen, local);
712 if (!p)
713 continue;
714 p->next = packed_git;
715 packed_git = p;
717 closedir(dir);
720 static int prepare_packed_git_run_once = 0;
721 void prepare_packed_git(void)
723 struct alternate_object_database *alt;
725 if (prepare_packed_git_run_once)
726 return;
727 prepare_packed_git_one(get_object_directory(), 1);
728 prepare_alt_odb();
729 for (alt = alt_odb_list; alt; alt = alt->next) {
730 alt->name[-1] = 0;
731 prepare_packed_git_one(alt->base, 0);
732 alt->name[-1] = '/';
734 prepare_packed_git_run_once = 1;
737 void reprepare_packed_git(void)
739 prepare_packed_git_run_once = 0;
740 prepare_packed_git();
743 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
745 unsigned char real_sha1[20];
746 hash_sha1_file(map, size, type, real_sha1);
747 return hashcmp(sha1, real_sha1) ? -1 : 0;
750 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
752 struct stat st;
753 void *map;
754 int fd;
755 char *filename = find_sha1_file(sha1, &st);
757 if (!filename) {
758 return NULL;
761 fd = open(filename, O_RDONLY | sha1_file_open_flag);
762 if (fd < 0) {
763 /* See if it works without O_NOATIME */
764 switch (sha1_file_open_flag) {
765 default:
766 fd = open(filename, O_RDONLY);
767 if (fd >= 0)
768 break;
769 /* Fallthrough */
770 case 0:
771 return NULL;
774 /* If it failed once, it will probably fail again.
775 * Stop using O_NOATIME
777 sha1_file_open_flag = 0;
779 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
780 close(fd);
781 if (map == MAP_FAILED)
782 return NULL;
783 *size = st.st_size;
784 return map;
787 int legacy_loose_object(unsigned char *map)
789 unsigned int word;
792 * Is it a zlib-compressed buffer? If so, the first byte
793 * must be 0x78 (15-bit window size, deflated), and the
794 * first 16-bit word is evenly divisible by 31
796 word = (map[0] << 8) + map[1];
797 if (map[0] == 0x78 && !(word % 31))
798 return 1;
799 else
800 return 0;
803 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
805 unsigned shift;
806 unsigned char c;
807 unsigned long size;
808 unsigned long used = 0;
810 c = buf[used++];
811 *type = (c >> 4) & 7;
812 size = c & 15;
813 shift = 4;
814 while (c & 0x80) {
815 if (len <= used)
816 return 0;
817 if (sizeof(long) * 8 <= shift)
818 return 0;
819 c = buf[used++];
820 size += (c & 0x7f) << shift;
821 shift += 7;
823 *sizep = size;
824 return used;
827 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
829 unsigned long size, used;
830 static const char valid_loose_object_type[8] = {
831 0, /* OBJ_EXT */
832 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
833 0, /* "delta" and others are invalid in a loose object */
835 enum object_type type;
837 /* Get the data stream */
838 memset(stream, 0, sizeof(*stream));
839 stream->next_in = map;
840 stream->avail_in = mapsize;
841 stream->next_out = buffer;
842 stream->avail_out = bufsiz;
844 if (legacy_loose_object(map)) {
845 inflateInit(stream);
846 return inflate(stream, 0);
849 used = unpack_object_header_gently(map, mapsize, &type, &size);
850 if (!used || !valid_loose_object_type[type])
851 return -1;
852 map += used;
853 mapsize -= used;
855 /* Set up the stream for the rest.. */
856 stream->next_in = map;
857 stream->avail_in = mapsize;
858 inflateInit(stream);
860 /* And generate the fake traditional header */
861 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
862 type_names[type], size);
863 return 0;
866 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
868 int bytes = strlen(buffer) + 1;
869 unsigned char *buf = xmalloc(1+size);
870 unsigned long n;
872 n = stream->total_out - bytes;
873 if (n > size)
874 n = size;
875 memcpy(buf, (char *) buffer + bytes, n);
876 bytes = n;
877 if (bytes < size) {
878 stream->next_out = buf + bytes;
879 stream->avail_out = size - bytes;
880 while (inflate(stream, Z_FINISH) == Z_OK)
881 /* nothing */;
883 buf[size] = 0;
884 inflateEnd(stream);
885 return buf;
889 * We used to just use "sscanf()", but that's actually way
890 * too permissive for what we want to check. So do an anal
891 * object header parse by hand.
893 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
895 int i;
896 unsigned long size;
899 * The type can be at most ten bytes (including the
900 * terminating '\0' that we add), and is followed by
901 * a space.
903 i = 10;
904 for (;;) {
905 char c = *hdr++;
906 if (c == ' ')
907 break;
908 if (!--i)
909 return -1;
910 *type++ = c;
912 *type = 0;
915 * The length must follow immediately, and be in canonical
916 * decimal format (ie "010" is not valid).
918 size = *hdr++ - '0';
919 if (size > 9)
920 return -1;
921 if (size) {
922 for (;;) {
923 unsigned long c = *hdr - '0';
924 if (c > 9)
925 break;
926 hdr++;
927 size = size * 10 + c;
930 *sizep = size;
933 * The length must be followed by a zero byte
935 return *hdr ? -1 : 0;
938 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
940 int ret;
941 z_stream stream;
942 char hdr[8192];
944 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
945 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
946 return NULL;
948 return unpack_sha1_rest(&stream, hdr, *size);
951 static unsigned long get_delta_base(struct packed_git *p,
952 struct pack_window **w_curs,
953 unsigned long offset,
954 enum object_type kind,
955 unsigned long delta_obj_offset,
956 unsigned long *base_obj_offset)
958 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
959 unsigned long base_offset;
961 /* use_pack() assured us we have [base_info, base_info + 20)
962 * as a range that we can look at without walking off the
963 * end of the mapped window. Its actually the hash size
964 * that is assured. An OFS_DELTA longer than the hash size
965 * is stupid, as then a REF_DELTA would be smaller to store.
967 if (kind == OBJ_OFS_DELTA) {
968 unsigned used = 0;
969 unsigned char c = base_info[used++];
970 base_offset = c & 127;
971 while (c & 128) {
972 base_offset += 1;
973 if (!base_offset || base_offset & ~(~0UL >> 7))
974 die("offset value overflow for delta base object");
975 c = base_info[used++];
976 base_offset = (base_offset << 7) + (c & 127);
978 base_offset = delta_obj_offset - base_offset;
979 if (base_offset >= delta_obj_offset)
980 die("delta base offset out of bound");
981 offset += used;
982 } else if (kind == OBJ_REF_DELTA) {
983 /* The base entry _must_ be in the same pack */
984 base_offset = find_pack_entry_one(base_info, p);
985 if (!base_offset)
986 die("failed to find delta-pack base object %s",
987 sha1_to_hex(base_info));
988 offset += 20;
989 } else
990 die("I am totally screwed");
991 *base_obj_offset = base_offset;
992 return offset;
995 /* forward declaration for a mutually recursive function */
996 static int packed_object_info(struct packed_git *p, unsigned long offset,
997 char *type, unsigned long *sizep);
999 static int packed_delta_info(struct packed_git *p,
1000 struct pack_window **w_curs,
1001 unsigned long offset,
1002 enum object_type kind,
1003 unsigned long obj_offset,
1004 char *type,
1005 unsigned long *sizep)
1007 unsigned long base_offset;
1009 offset = get_delta_base(p, w_curs, offset, kind,
1010 obj_offset, &base_offset);
1012 /* We choose to only get the type of the base object and
1013 * ignore potentially corrupt pack file that expects the delta
1014 * based on a base with a wrong size. This saves tons of
1015 * inflate() calls.
1017 if (packed_object_info(p, base_offset, type, NULL))
1018 die("cannot get info for delta-pack base");
1020 if (sizep) {
1021 const unsigned char *data;
1022 unsigned char delta_head[20], *in;
1023 unsigned long result_size;
1024 z_stream stream;
1025 int st;
1027 memset(&stream, 0, sizeof(stream));
1028 stream.next_out = delta_head;
1029 stream.avail_out = sizeof(delta_head);
1031 inflateInit(&stream);
1032 do {
1033 in = use_pack(p, w_curs, offset, &stream.avail_in);
1034 stream.next_in = in;
1035 st = inflate(&stream, Z_FINISH);
1036 offset += stream.next_in - in;
1037 } while ((st == Z_OK || st == Z_BUF_ERROR)
1038 && stream.total_out < sizeof(delta_head));
1039 inflateEnd(&stream);
1040 if ((st != Z_STREAM_END) &&
1041 stream.total_out != sizeof(delta_head))
1042 die("delta data unpack-initial failed");
1044 /* Examine the initial part of the delta to figure out
1045 * the result size.
1047 data = delta_head;
1049 /* ignore base size */
1050 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1052 /* Read the result size */
1053 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1054 *sizep = result_size;
1056 return 0;
1059 static unsigned long unpack_object_header(struct packed_git *p,
1060 struct pack_window **w_curs,
1061 unsigned long offset,
1062 enum object_type *type,
1063 unsigned long *sizep)
1065 unsigned char *base;
1066 unsigned int left;
1067 unsigned long used;
1069 /* use_pack() assures us we have [base, base + 20) available
1070 * as a range that we can look at at. (Its actually the hash
1071 * size that is assurred.) With our object header encoding
1072 * the maximum deflated object size is 2^137, which is just
1073 * insane, so we know won't exceed what we have been given.
1075 base = use_pack(p, w_curs, offset, &left);
1076 used = unpack_object_header_gently(base, left, type, sizep);
1077 if (!used)
1078 die("object offset outside of pack file");
1080 return offset + used;
1083 void packed_object_info_detail(struct packed_git *p,
1084 unsigned long offset,
1085 char *type,
1086 unsigned long *size,
1087 unsigned long *store_size,
1088 unsigned int *delta_chain_length,
1089 unsigned char *base_sha1)
1091 struct pack_window *w_curs = NULL;
1092 unsigned long obj_offset, val;
1093 unsigned char *next_sha1;
1094 enum object_type kind;
1096 *delta_chain_length = 0;
1097 obj_offset = offset;
1098 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1100 for (;;) {
1101 switch (kind) {
1102 default:
1103 die("pack %s contains unknown object type %d",
1104 p->pack_name, kind);
1105 case OBJ_COMMIT:
1106 case OBJ_TREE:
1107 case OBJ_BLOB:
1108 case OBJ_TAG:
1109 strcpy(type, type_names[kind]);
1110 *store_size = 0; /* notyet */
1111 unuse_pack(&w_curs);
1112 return;
1113 case OBJ_OFS_DELTA:
1114 get_delta_base(p, &w_curs, offset, kind,
1115 obj_offset, &offset);
1116 if (*delta_chain_length == 0) {
1117 /* TODO: find base_sha1 as pointed by offset */
1119 break;
1120 case OBJ_REF_DELTA:
1121 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1122 if (*delta_chain_length == 0)
1123 hashcpy(base_sha1, next_sha1);
1124 offset = find_pack_entry_one(next_sha1, p);
1125 break;
1127 obj_offset = offset;
1128 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1129 (*delta_chain_length)++;
1133 static int packed_object_info(struct packed_git *p, unsigned long offset,
1134 char *type, unsigned long *sizep)
1136 struct pack_window *w_curs = NULL;
1137 unsigned long size, obj_offset = offset;
1138 enum object_type kind;
1139 int r;
1141 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1143 switch (kind) {
1144 case OBJ_OFS_DELTA:
1145 case OBJ_REF_DELTA:
1146 r = packed_delta_info(p, &w_curs, offset, kind,
1147 obj_offset, type, sizep);
1148 unuse_pack(&w_curs);
1149 return r;
1150 case OBJ_COMMIT:
1151 case OBJ_TREE:
1152 case OBJ_BLOB:
1153 case OBJ_TAG:
1154 strcpy(type, type_names[kind]);
1155 unuse_pack(&w_curs);
1156 break;
1157 default:
1158 die("pack %s contains unknown object type %d",
1159 p->pack_name, kind);
1161 if (sizep)
1162 *sizep = size;
1163 return 0;
1166 static void *unpack_compressed_entry(struct packed_git *p,
1167 struct pack_window **w_curs,
1168 unsigned long offset,
1169 unsigned long size)
1171 int st;
1172 z_stream stream;
1173 unsigned char *buffer, *in;
1175 buffer = xmalloc(size + 1);
1176 buffer[size] = 0;
1177 memset(&stream, 0, sizeof(stream));
1178 stream.next_out = buffer;
1179 stream.avail_out = size;
1181 inflateInit(&stream);
1182 do {
1183 in = use_pack(p, w_curs, offset, &stream.avail_in);
1184 stream.next_in = in;
1185 st = inflate(&stream, Z_FINISH);
1186 offset += stream.next_in - in;
1187 } while (st == Z_OK || st == Z_BUF_ERROR);
1188 inflateEnd(&stream);
1189 if ((st != Z_STREAM_END) || stream.total_out != size) {
1190 free(buffer);
1191 return NULL;
1194 return buffer;
1197 static void *unpack_delta_entry(struct packed_git *p,
1198 struct pack_window **w_curs,
1199 unsigned long offset,
1200 unsigned long delta_size,
1201 enum object_type kind,
1202 unsigned long obj_offset,
1203 char *type,
1204 unsigned long *sizep)
1206 void *delta_data, *result, *base;
1207 unsigned long result_size, base_size, base_offset;
1209 offset = get_delta_base(p, w_curs, offset, kind,
1210 obj_offset, &base_offset);
1211 base = unpack_entry(p, base_offset, type, &base_size);
1212 if (!base)
1213 die("failed to read delta base object at %lu from %s",
1214 base_offset, p->pack_name);
1216 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1217 result = patch_delta(base, base_size,
1218 delta_data, delta_size,
1219 &result_size);
1220 if (!result)
1221 die("failed to apply delta");
1222 free(delta_data);
1223 free(base);
1224 *sizep = result_size;
1225 return result;
1228 void *unpack_entry(struct packed_git *p, unsigned long offset,
1229 char *type, unsigned long *sizep)
1231 struct pack_window *w_curs = NULL;
1232 unsigned long size, obj_offset = offset;
1233 enum object_type kind;
1234 void *retval;
1236 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1237 switch (kind) {
1238 case OBJ_OFS_DELTA:
1239 case OBJ_REF_DELTA:
1240 retval = unpack_delta_entry(p, &w_curs, offset, size,
1241 kind, obj_offset, type, sizep);
1242 break;
1243 case OBJ_COMMIT:
1244 case OBJ_TREE:
1245 case OBJ_BLOB:
1246 case OBJ_TAG:
1247 strcpy(type, type_names[kind]);
1248 *sizep = size;
1249 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1250 break;
1251 default:
1252 die("unknown object type %i in %s", kind, p->pack_name);
1254 unuse_pack(&w_curs);
1255 return retval;
1258 int num_packed_objects(const struct packed_git *p)
1260 /* See check_packed_git_idx() */
1261 return (p->index_size - 20 - 20 - 4*256) / 24;
1264 int nth_packed_object_sha1(const struct packed_git *p, int n,
1265 unsigned char* sha1)
1267 void *index = p->index_base + 256;
1268 if (n < 0 || num_packed_objects(p) <= n)
1269 return -1;
1270 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1271 return 0;
1274 unsigned long find_pack_entry_one(const unsigned char *sha1,
1275 struct packed_git *p)
1277 unsigned int *level1_ofs = p->index_base;
1278 int hi = ntohl(level1_ofs[*sha1]);
1279 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1280 void *index = p->index_base + 256;
1282 do {
1283 int mi = (lo + hi) / 2;
1284 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1285 if (!cmp)
1286 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1287 if (cmp > 0)
1288 hi = mi;
1289 else
1290 lo = mi+1;
1291 } while (lo < hi);
1292 return 0;
1295 static int matches_pack_name(struct packed_git *p, const char *ig)
1297 const char *last_c, *c;
1299 if (!strcmp(p->pack_name, ig))
1300 return 0;
1302 for (c = p->pack_name, last_c = c; *c;)
1303 if (*c == '/')
1304 last_c = ++c;
1305 else
1306 ++c;
1307 if (!strcmp(last_c, ig))
1308 return 0;
1310 return 1;
1313 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1315 struct packed_git *p;
1316 unsigned long offset;
1318 prepare_packed_git();
1320 for (p = packed_git; p; p = p->next) {
1321 if (ignore_packed) {
1322 const char **ig;
1323 for (ig = ignore_packed; *ig; ig++)
1324 if (!matches_pack_name(p, *ig))
1325 break;
1326 if (*ig)
1327 continue;
1329 offset = find_pack_entry_one(sha1, p);
1330 if (offset) {
1331 e->offset = offset;
1332 e->p = p;
1333 hashcpy(e->sha1, sha1);
1334 return 1;
1337 return 0;
1340 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1341 struct packed_git *packs)
1343 struct packed_git *p;
1345 for (p = packs; p; p = p->next) {
1346 if (find_pack_entry_one(sha1, p))
1347 return p;
1349 return NULL;
1353 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1355 int status;
1356 unsigned long mapsize, size;
1357 void *map;
1358 z_stream stream;
1359 char hdr[128];
1361 map = map_sha1_file(sha1, &mapsize);
1362 if (!map)
1363 return error("unable to find %s", sha1_to_hex(sha1));
1364 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1365 status = error("unable to unpack %s header",
1366 sha1_to_hex(sha1));
1367 if (parse_sha1_header(hdr, type, &size) < 0)
1368 status = error("unable to parse %s header", sha1_to_hex(sha1));
1369 else {
1370 status = 0;
1371 if (sizep)
1372 *sizep = size;
1374 inflateEnd(&stream);
1375 munmap(map, mapsize);
1376 return status;
1379 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1381 struct pack_entry e;
1383 if (!find_pack_entry(sha1, &e, NULL)) {
1384 reprepare_packed_git();
1385 if (!find_pack_entry(sha1, &e, NULL))
1386 return sha1_loose_object_info(sha1, type, sizep);
1388 return packed_object_info(e.p, e.offset, type, sizep);
1391 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1393 struct pack_entry e;
1395 if (!find_pack_entry(sha1, &e, NULL)) {
1396 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1397 return NULL;
1399 return unpack_entry(e.p, e.offset, type, size);
1402 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1404 unsigned long mapsize;
1405 void *map, *buf;
1406 struct pack_entry e;
1408 if (find_pack_entry(sha1, &e, NULL))
1409 return read_packed_sha1(sha1, type, size);
1410 map = map_sha1_file(sha1, &mapsize);
1411 if (map) {
1412 buf = unpack_sha1_file(map, mapsize, type, size);
1413 munmap(map, mapsize);
1414 return buf;
1416 reprepare_packed_git();
1417 if (find_pack_entry(sha1, &e, NULL))
1418 return read_packed_sha1(sha1, type, size);
1419 return NULL;
1422 void *read_object_with_reference(const unsigned char *sha1,
1423 const char *required_type,
1424 unsigned long *size,
1425 unsigned char *actual_sha1_return)
1427 char type[20];
1428 void *buffer;
1429 unsigned long isize;
1430 unsigned char actual_sha1[20];
1432 hashcpy(actual_sha1, sha1);
1433 while (1) {
1434 int ref_length = -1;
1435 const char *ref_type = NULL;
1437 buffer = read_sha1_file(actual_sha1, type, &isize);
1438 if (!buffer)
1439 return NULL;
1440 if (!strcmp(type, required_type)) {
1441 *size = isize;
1442 if (actual_sha1_return)
1443 hashcpy(actual_sha1_return, actual_sha1);
1444 return buffer;
1446 /* Handle references */
1447 else if (!strcmp(type, commit_type))
1448 ref_type = "tree ";
1449 else if (!strcmp(type, tag_type))
1450 ref_type = "object ";
1451 else {
1452 free(buffer);
1453 return NULL;
1455 ref_length = strlen(ref_type);
1457 if (memcmp(buffer, ref_type, ref_length) ||
1458 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1459 free(buffer);
1460 return NULL;
1462 free(buffer);
1463 /* Now we have the ID of the referred-to object in
1464 * actual_sha1. Check again. */
1468 static void write_sha1_file_prepare(void *buf, unsigned long len,
1469 const char *type, unsigned char *sha1,
1470 unsigned char *hdr, int *hdrlen)
1472 SHA_CTX c;
1474 /* Generate the header */
1475 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1477 /* Sha1.. */
1478 SHA1_Init(&c);
1479 SHA1_Update(&c, hdr, *hdrlen);
1480 SHA1_Update(&c, buf, len);
1481 SHA1_Final(sha1, &c);
1485 * Link the tempfile to the final place, possibly creating the
1486 * last directory level as you do so.
1488 * Returns the errno on failure, 0 on success.
1490 static int link_temp_to_file(const char *tmpfile, const char *filename)
1492 int ret;
1493 char *dir;
1495 if (!link(tmpfile, filename))
1496 return 0;
1499 * Try to mkdir the last path component if that failed.
1501 * Re-try the "link()" regardless of whether the mkdir
1502 * succeeds, since a race might mean that somebody
1503 * else succeeded.
1505 ret = errno;
1506 dir = strrchr(filename, '/');
1507 if (dir) {
1508 *dir = 0;
1509 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1510 *dir = '/';
1511 return -2;
1513 *dir = '/';
1514 if (!link(tmpfile, filename))
1515 return 0;
1516 ret = errno;
1518 return ret;
1522 * Move the just written object into its final resting place
1524 int move_temp_to_file(const char *tmpfile, const char *filename)
1526 int ret = link_temp_to_file(tmpfile, filename);
1529 * Coda hack - coda doesn't like cross-directory links,
1530 * so we fall back to a rename, which will mean that it
1531 * won't be able to check collisions, but that's not a
1532 * big deal.
1534 * The same holds for FAT formatted media.
1536 * When this succeeds, we just return 0. We have nothing
1537 * left to unlink.
1539 if (ret && ret != EEXIST) {
1540 if (!rename(tmpfile, filename))
1541 return 0;
1542 ret = errno;
1544 unlink(tmpfile);
1545 if (ret) {
1546 if (ret != EEXIST) {
1547 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1549 /* FIXME!!! Collision check here ? */
1552 return 0;
1555 static int write_buffer(int fd, const void *buf, size_t len)
1557 while (len) {
1558 ssize_t size;
1560 size = write(fd, buf, len);
1561 if (!size)
1562 return error("file write: disk full");
1563 if (size < 0) {
1564 if (errno == EINTR || errno == EAGAIN)
1565 continue;
1566 return error("file write error (%s)", strerror(errno));
1568 len -= size;
1569 buf = (char *) buf + size;
1571 return 0;
1574 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1576 int hdr_len;
1577 unsigned char c;
1579 c = (type << 4) | (len & 15);
1580 len >>= 4;
1581 hdr_len = 1;
1582 while (len) {
1583 *hdr++ = c | 0x80;
1584 hdr_len++;
1585 c = (len & 0x7f);
1586 len >>= 7;
1588 *hdr = c;
1589 return hdr_len;
1592 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1594 int obj_type, hdr;
1596 if (use_legacy_headers) {
1597 while (deflate(stream, 0) == Z_OK)
1598 /* nothing */;
1599 return;
1601 if (!strcmp(type, blob_type))
1602 obj_type = OBJ_BLOB;
1603 else if (!strcmp(type, tree_type))
1604 obj_type = OBJ_TREE;
1605 else if (!strcmp(type, commit_type))
1606 obj_type = OBJ_COMMIT;
1607 else if (!strcmp(type, tag_type))
1608 obj_type = OBJ_TAG;
1609 else
1610 die("trying to generate bogus object of type '%s'", type);
1611 hdr = write_binary_header(stream->next_out, obj_type, len);
1612 stream->total_out = hdr;
1613 stream->next_out += hdr;
1614 stream->avail_out -= hdr;
1617 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1618 unsigned char *sha1)
1620 unsigned char hdr[50];
1621 int hdrlen;
1622 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1623 return 0;
1626 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1628 int size;
1629 unsigned char *compressed;
1630 z_stream stream;
1631 unsigned char sha1[20];
1632 char *filename;
1633 static char tmpfile[PATH_MAX];
1634 unsigned char hdr[50];
1635 int fd, hdrlen;
1637 /* Normally if we have it in the pack then we do not bother writing
1638 * it out into .git/objects/??/?{38} file.
1640 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1641 filename = sha1_file_name(sha1);
1642 if (returnsha1)
1643 hashcpy(returnsha1, sha1);
1644 if (has_sha1_file(sha1))
1645 return 0;
1646 fd = open(filename, O_RDONLY);
1647 if (fd >= 0) {
1649 * FIXME!!! We might do collision checking here, but we'd
1650 * need to uncompress the old file and check it. Later.
1652 close(fd);
1653 return 0;
1656 if (errno != ENOENT) {
1657 return error("sha1 file %s: %s\n", filename, strerror(errno));
1660 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1662 fd = mkstemp(tmpfile);
1663 if (fd < 0) {
1664 if (errno == EPERM)
1665 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1666 else
1667 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1670 /* Set it up */
1671 memset(&stream, 0, sizeof(stream));
1672 deflateInit(&stream, zlib_compression_level);
1673 size = 8 + deflateBound(&stream, len+hdrlen);
1674 compressed = xmalloc(size);
1676 /* Compress it */
1677 stream.next_out = compressed;
1678 stream.avail_out = size;
1680 /* First header.. */
1681 stream.next_in = hdr;
1682 stream.avail_in = hdrlen;
1683 setup_object_header(&stream, type, len);
1685 /* Then the data itself.. */
1686 stream.next_in = buf;
1687 stream.avail_in = len;
1688 while (deflate(&stream, Z_FINISH) == Z_OK)
1689 /* nothing */;
1690 deflateEnd(&stream);
1691 size = stream.total_out;
1693 if (write_buffer(fd, compressed, size) < 0)
1694 die("unable to write sha1 file");
1695 fchmod(fd, 0444);
1696 close(fd);
1697 free(compressed);
1699 return move_temp_to_file(tmpfile, filename);
1703 * We need to unpack and recompress the object for writing
1704 * it out to a different file.
1706 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1708 size_t size;
1709 z_stream stream;
1710 unsigned char *unpacked;
1711 unsigned long len;
1712 char type[20];
1713 char hdr[50];
1714 int hdrlen;
1715 void *buf;
1717 /* need to unpack and recompress it by itself */
1718 unpacked = read_packed_sha1(sha1, type, &len);
1720 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1722 /* Set it up */
1723 memset(&stream, 0, sizeof(stream));
1724 deflateInit(&stream, zlib_compression_level);
1725 size = deflateBound(&stream, len + hdrlen);
1726 buf = xmalloc(size);
1728 /* Compress it */
1729 stream.next_out = buf;
1730 stream.avail_out = size;
1732 /* First header.. */
1733 stream.next_in = (void *)hdr;
1734 stream.avail_in = hdrlen;
1735 while (deflate(&stream, 0) == Z_OK)
1736 /* nothing */;
1738 /* Then the data itself.. */
1739 stream.next_in = unpacked;
1740 stream.avail_in = len;
1741 while (deflate(&stream, Z_FINISH) == Z_OK)
1742 /* nothing */;
1743 deflateEnd(&stream);
1744 free(unpacked);
1746 *objsize = stream.total_out;
1747 return buf;
1750 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1752 int retval;
1753 unsigned long objsize;
1754 void *buf = map_sha1_file(sha1, &objsize);
1756 if (buf) {
1757 retval = write_buffer(fd, buf, objsize);
1758 munmap(buf, objsize);
1759 return retval;
1762 buf = repack_object(sha1, &objsize);
1763 retval = write_buffer(fd, buf, objsize);
1764 free(buf);
1765 return retval;
1768 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1769 size_t bufsize, size_t *bufposn)
1771 char tmpfile[PATH_MAX];
1772 int local;
1773 z_stream stream;
1774 unsigned char real_sha1[20];
1775 unsigned char discard[4096];
1776 int ret;
1777 SHA_CTX c;
1779 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1781 local = mkstemp(tmpfile);
1782 if (local < 0) {
1783 if (errno == EPERM)
1784 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1785 else
1786 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1789 memset(&stream, 0, sizeof(stream));
1791 inflateInit(&stream);
1793 SHA1_Init(&c);
1795 do {
1796 ssize_t size;
1797 if (*bufposn) {
1798 stream.avail_in = *bufposn;
1799 stream.next_in = (unsigned char *) buffer;
1800 do {
1801 stream.next_out = discard;
1802 stream.avail_out = sizeof(discard);
1803 ret = inflate(&stream, Z_SYNC_FLUSH);
1804 SHA1_Update(&c, discard, sizeof(discard) -
1805 stream.avail_out);
1806 } while (stream.avail_in && ret == Z_OK);
1807 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1808 die("unable to write sha1 file");
1809 memmove(buffer, buffer + *bufposn - stream.avail_in,
1810 stream.avail_in);
1811 *bufposn = stream.avail_in;
1812 if (ret != Z_OK)
1813 break;
1815 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1816 if (size <= 0) {
1817 close(local);
1818 unlink(tmpfile);
1819 if (!size)
1820 return error("Connection closed?");
1821 perror("Reading from connection");
1822 return -1;
1824 *bufposn += size;
1825 } while (1);
1826 inflateEnd(&stream);
1828 close(local);
1829 SHA1_Final(real_sha1, &c);
1830 if (ret != Z_STREAM_END) {
1831 unlink(tmpfile);
1832 return error("File %s corrupted", sha1_to_hex(sha1));
1834 if (hashcmp(sha1, real_sha1)) {
1835 unlink(tmpfile);
1836 return error("File %s has bad hash", sha1_to_hex(sha1));
1839 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1842 int has_pack_index(const unsigned char *sha1)
1844 struct stat st;
1845 if (stat(sha1_pack_index_name(sha1), &st))
1846 return 0;
1847 return 1;
1850 int has_pack_file(const unsigned char *sha1)
1852 struct stat st;
1853 if (stat(sha1_pack_name(sha1), &st))
1854 return 0;
1855 return 1;
1858 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1860 struct pack_entry e;
1861 return find_pack_entry(sha1, &e, ignore_packed);
1864 int has_sha1_file(const unsigned char *sha1)
1866 struct stat st;
1867 struct pack_entry e;
1869 if (find_pack_entry(sha1, &e, NULL))
1870 return 1;
1871 return find_sha1_file(sha1, &st) ? 1 : 0;
1875 * reads from fd as long as possible into a supplied buffer of size bytes.
1876 * If necessary the buffer's size is increased using realloc()
1878 * returns 0 if anything went fine and -1 otherwise
1880 * NOTE: both buf and size may change, but even when -1 is returned
1881 * you still have to free() it yourself.
1883 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1885 char* buf = *return_buf;
1886 unsigned long size = *return_size;
1887 int iret;
1888 unsigned long off = 0;
1890 do {
1891 iret = xread(fd, buf + off, size - off);
1892 if (iret > 0) {
1893 off += iret;
1894 if (off == size) {
1895 size *= 2;
1896 buf = xrealloc(buf, size);
1899 } while (iret > 0);
1901 *return_buf = buf;
1902 *return_size = off;
1904 if (iret < 0)
1905 return -1;
1906 return 0;
1909 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1911 unsigned long size = 4096;
1912 char *buf = xmalloc(size);
1913 int ret;
1915 if (read_pipe(fd, &buf, &size)) {
1916 free(buf);
1917 return -1;
1920 if (!type)
1921 type = blob_type;
1922 if (write_object)
1923 ret = write_sha1_file(buf, size, type, sha1);
1924 else
1925 ret = hash_sha1_file(buf, size, type, sha1);
1926 free(buf);
1927 return ret;
1930 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1932 unsigned long size = st->st_size;
1933 void *buf;
1934 int ret;
1936 buf = "";
1937 if (size)
1938 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1939 close(fd);
1940 if (buf == MAP_FAILED)
1941 return -1;
1943 if (!type)
1944 type = blob_type;
1945 if (write_object)
1946 ret = write_sha1_file(buf, size, type, sha1);
1947 else
1948 ret = hash_sha1_file(buf, size, type, sha1);
1949 if (size)
1950 munmap(buf, size);
1951 return ret;
1954 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1956 int fd;
1957 char *target;
1959 switch (st->st_mode & S_IFMT) {
1960 case S_IFREG:
1961 fd = open(path, O_RDONLY);
1962 if (fd < 0)
1963 return error("open(\"%s\"): %s", path,
1964 strerror(errno));
1965 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1966 return error("%s: failed to insert into database",
1967 path);
1968 break;
1969 case S_IFLNK:
1970 target = xmalloc(st->st_size+1);
1971 if (readlink(path, target, st->st_size+1) != st->st_size) {
1972 char *errstr = strerror(errno);
1973 free(target);
1974 return error("readlink(\"%s\"): %s", path,
1975 errstr);
1977 if (!write_object)
1978 hash_sha1_file(target, st->st_size, blob_type, sha1);
1979 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1980 return error("%s: failed to insert into database",
1981 path);
1982 free(target);
1983 break;
1984 default:
1985 return error("%s: unsupported file type", path);
1987 return 0;