Unmap individual windows rather than entire files.
[git/raj.git] / sha1_file.c
blob8e14a5a882c8ff7893b1d548390f7359f60d8fa4
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 int pack_used_ctr;
401 static unsigned long pack_mapped;
402 struct packed_git *packed_git;
404 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
405 void **idx_map_)
407 void *idx_map;
408 unsigned int *index;
409 unsigned long idx_size;
410 int nr, i;
411 int fd = open(path, O_RDONLY);
412 struct stat st;
413 if (fd < 0)
414 return -1;
415 if (fstat(fd, &st)) {
416 close(fd);
417 return -1;
419 idx_size = st.st_size;
420 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
421 close(fd);
422 if (idx_map == MAP_FAILED)
423 return -1;
425 index = idx_map;
426 *idx_map_ = idx_map;
427 *idx_size_ = idx_size;
429 /* check index map */
430 if (idx_size < 4*256 + 20 + 20)
431 return error("index file too small");
432 nr = 0;
433 for (i = 0; i < 256; i++) {
434 unsigned int n = ntohl(index[i]);
435 if (n < nr)
436 return error("non-monotonic index");
437 nr = n;
441 * Total size:
442 * - 256 index entries 4 bytes each
443 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
444 * - 20-byte SHA1 of the packfile
445 * - 20-byte SHA1 file checksum
447 if (idx_size != 4*256 + nr * 24 + 20 + 20)
448 return error("wrong index file size");
450 return 0;
453 static int unuse_one_window(void)
455 struct packed_git *p, *lru_p = NULL;
456 struct pack_window *w, *w_l, *lru_w = NULL, *lru_l = NULL;
458 for (p = packed_git; p; p = p->next) {
459 for (w_l = NULL, w = p->windows; w; w = w->next) {
460 if (!w->inuse_cnt) {
461 if (!lru_w || w->last_used < lru_w->last_used) {
462 lru_p = p;
463 lru_w = w;
464 lru_l = w_l;
467 w_l = w;
470 if (lru_p) {
471 munmap(lru_w->base, lru_w->len);
472 pack_mapped -= lru_w->len;
473 if (lru_l)
474 lru_l->next = lru_w->next;
475 else {
476 lru_p->windows = lru_w->next;
477 if (!lru_p->windows) {
478 close(lru_p->pack_fd);
479 lru_p->pack_fd = -1;
482 free(lru_w);
483 return 1;
485 return 0;
488 void unuse_pack(struct pack_window **w_cursor)
490 struct pack_window *w = *w_cursor;
491 if (w) {
492 w->inuse_cnt--;
493 *w_cursor = NULL;
497 static void open_packed_git(struct packed_git *p)
499 struct stat st;
500 struct pack_header hdr;
501 unsigned char sha1[20];
502 unsigned char *idx_sha1;
504 p->pack_fd = open(p->pack_name, O_RDONLY);
505 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
506 die("packfile %s cannot be opened", p->pack_name);
508 /* If we created the struct before we had the pack we lack size. */
509 if (!p->pack_size) {
510 if (!S_ISREG(st.st_mode))
511 die("packfile %s not a regular file", p->pack_name);
512 p->pack_size = st.st_size;
513 } else if (p->pack_size != st.st_size)
514 die("packfile %s size changed", p->pack_name);
516 /* Verify we recognize this pack file format. */
517 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
518 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
519 die("file %s is not a GIT packfile", p->pack_name);
520 if (!pack_version_ok(hdr.hdr_version))
521 die("packfile %s is version %u and not supported"
522 " (try upgrading GIT to a newer version)",
523 p->pack_name, ntohl(hdr.hdr_version));
525 /* Verify the pack matches its index. */
526 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
527 die("packfile %s claims to have %u objects"
528 " while index size indicates %u objects",
529 p->pack_name, ntohl(hdr.hdr_entries),
530 num_packed_objects(p));
531 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
532 die("end of packfile %s is unavailable", p->pack_name);
533 read_or_die(p->pack_fd, sha1, sizeof(sha1));
534 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
535 if (hashcmp(sha1, idx_sha1))
536 die("packfile %s does not match index", p->pack_name);
539 unsigned char* use_pack(struct packed_git *p,
540 struct pack_window **w_cursor,
541 unsigned long offset,
542 unsigned int *left)
544 struct pack_window *win = p->windows;
546 if (p->pack_fd == -1)
547 open_packed_git(p);
548 if (!win) {
549 pack_mapped += p->pack_size;
550 while (packed_git_limit < pack_mapped && unuse_one_window())
551 ; /* nothing */
552 win = xcalloc(1, sizeof(*win));
553 win->len = p->pack_size;
554 win->base = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, p->pack_fd, 0);
555 if (win->base == MAP_FAILED)
556 die("packfile %s cannot be mapped.", p->pack_name);
557 p->windows = win;
559 if (win != *w_cursor) {
560 win->last_used = pack_used_ctr++;
561 win->inuse_cnt++;
562 *w_cursor = win;
564 if (left)
565 *left = win->len - offset;
566 return win->base + offset;
569 struct packed_git *add_packed_git(char *path, int path_len, int local)
571 struct stat st;
572 struct packed_git *p;
573 unsigned long idx_size;
574 void *idx_map;
575 unsigned char sha1[20];
577 if (check_packed_git_idx(path, &idx_size, &idx_map))
578 return NULL;
580 /* do we have a corresponding .pack file? */
581 strcpy(path + path_len - 4, ".pack");
582 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
583 munmap(idx_map, idx_size);
584 return NULL;
586 /* ok, it looks sane as far as we can check without
587 * actually mapping the pack file.
589 p = xmalloc(sizeof(*p) + path_len + 2);
590 strcpy(p->pack_name, path);
591 p->index_size = idx_size;
592 p->pack_size = st.st_size;
593 p->index_base = idx_map;
594 p->next = NULL;
595 p->windows = NULL;
596 p->pack_fd = -1;
597 p->pack_local = local;
598 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
599 hashcpy(p->sha1, sha1);
600 return p;
603 struct packed_git *parse_pack_index(unsigned char *sha1)
605 char *path = sha1_pack_index_name(sha1);
606 return parse_pack_index_file(sha1, path);
609 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
611 struct packed_git *p;
612 unsigned long idx_size;
613 void *idx_map;
614 char *path;
616 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
617 return NULL;
619 path = sha1_pack_name(sha1);
621 p = xmalloc(sizeof(*p) + strlen(path) + 2);
622 strcpy(p->pack_name, path);
623 p->index_size = idx_size;
624 p->pack_size = 0;
625 p->index_base = idx_map;
626 p->next = NULL;
627 p->windows = NULL;
628 p->pack_fd = -1;
629 hashcpy(p->sha1, sha1);
630 return p;
633 void install_packed_git(struct packed_git *pack)
635 pack->next = packed_git;
636 packed_git = pack;
639 static void prepare_packed_git_one(char *objdir, int local)
641 char path[PATH_MAX];
642 int len;
643 DIR *dir;
644 struct dirent *de;
646 sprintf(path, "%s/pack", objdir);
647 len = strlen(path);
648 dir = opendir(path);
649 if (!dir) {
650 if (errno != ENOENT)
651 error("unable to open object pack directory: %s: %s",
652 path, strerror(errno));
653 return;
655 path[len++] = '/';
656 while ((de = readdir(dir)) != NULL) {
657 int namelen = strlen(de->d_name);
658 struct packed_git *p;
660 if (!has_extension(de->d_name, ".idx"))
661 continue;
663 /* we have .idx. Is it a file we can map? */
664 strcpy(path + len, de->d_name);
665 for (p = packed_git; p; p = p->next) {
666 if (!memcmp(path, p->pack_name, len + namelen - 4))
667 break;
669 if (p)
670 continue;
671 p = add_packed_git(path, len + namelen, local);
672 if (!p)
673 continue;
674 p->next = packed_git;
675 packed_git = p;
677 closedir(dir);
680 static int prepare_packed_git_run_once = 0;
681 void prepare_packed_git(void)
683 struct alternate_object_database *alt;
685 if (prepare_packed_git_run_once)
686 return;
687 prepare_packed_git_one(get_object_directory(), 1);
688 prepare_alt_odb();
689 for (alt = alt_odb_list; alt; alt = alt->next) {
690 alt->name[-1] = 0;
691 prepare_packed_git_one(alt->base, 0);
692 alt->name[-1] = '/';
694 prepare_packed_git_run_once = 1;
697 void reprepare_packed_git(void)
699 prepare_packed_git_run_once = 0;
700 prepare_packed_git();
703 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
705 unsigned char real_sha1[20];
706 hash_sha1_file(map, size, type, real_sha1);
707 return hashcmp(sha1, real_sha1) ? -1 : 0;
710 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
712 struct stat st;
713 void *map;
714 int fd;
715 char *filename = find_sha1_file(sha1, &st);
717 if (!filename) {
718 return NULL;
721 fd = open(filename, O_RDONLY | sha1_file_open_flag);
722 if (fd < 0) {
723 /* See if it works without O_NOATIME */
724 switch (sha1_file_open_flag) {
725 default:
726 fd = open(filename, O_RDONLY);
727 if (fd >= 0)
728 break;
729 /* Fallthrough */
730 case 0:
731 return NULL;
734 /* If it failed once, it will probably fail again.
735 * Stop using O_NOATIME
737 sha1_file_open_flag = 0;
739 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
740 close(fd);
741 if (map == MAP_FAILED)
742 return NULL;
743 *size = st.st_size;
744 return map;
747 int legacy_loose_object(unsigned char *map)
749 unsigned int word;
752 * Is it a zlib-compressed buffer? If so, the first byte
753 * must be 0x78 (15-bit window size, deflated), and the
754 * first 16-bit word is evenly divisible by 31
756 word = (map[0] << 8) + map[1];
757 if (map[0] == 0x78 && !(word % 31))
758 return 1;
759 else
760 return 0;
763 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
765 unsigned shift;
766 unsigned char c;
767 unsigned long size;
768 unsigned long used = 0;
770 c = buf[used++];
771 *type = (c >> 4) & 7;
772 size = c & 15;
773 shift = 4;
774 while (c & 0x80) {
775 if (len <= used)
776 return 0;
777 if (sizeof(long) * 8 <= shift)
778 return 0;
779 c = buf[used++];
780 size += (c & 0x7f) << shift;
781 shift += 7;
783 *sizep = size;
784 return used;
787 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
789 unsigned long size, used;
790 static const char valid_loose_object_type[8] = {
791 0, /* OBJ_EXT */
792 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
793 0, /* "delta" and others are invalid in a loose object */
795 enum object_type type;
797 /* Get the data stream */
798 memset(stream, 0, sizeof(*stream));
799 stream->next_in = map;
800 stream->avail_in = mapsize;
801 stream->next_out = buffer;
802 stream->avail_out = bufsiz;
804 if (legacy_loose_object(map)) {
805 inflateInit(stream);
806 return inflate(stream, 0);
809 used = unpack_object_header_gently(map, mapsize, &type, &size);
810 if (!used || !valid_loose_object_type[type])
811 return -1;
812 map += used;
813 mapsize -= used;
815 /* Set up the stream for the rest.. */
816 stream->next_in = map;
817 stream->avail_in = mapsize;
818 inflateInit(stream);
820 /* And generate the fake traditional header */
821 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
822 type_names[type], size);
823 return 0;
826 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
828 int bytes = strlen(buffer) + 1;
829 unsigned char *buf = xmalloc(1+size);
830 unsigned long n;
832 n = stream->total_out - bytes;
833 if (n > size)
834 n = size;
835 memcpy(buf, (char *) buffer + bytes, n);
836 bytes = n;
837 if (bytes < size) {
838 stream->next_out = buf + bytes;
839 stream->avail_out = size - bytes;
840 while (inflate(stream, Z_FINISH) == Z_OK)
841 /* nothing */;
843 buf[size] = 0;
844 inflateEnd(stream);
845 return buf;
849 * We used to just use "sscanf()", but that's actually way
850 * too permissive for what we want to check. So do an anal
851 * object header parse by hand.
853 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
855 int i;
856 unsigned long size;
859 * The type can be at most ten bytes (including the
860 * terminating '\0' that we add), and is followed by
861 * a space.
863 i = 10;
864 for (;;) {
865 char c = *hdr++;
866 if (c == ' ')
867 break;
868 if (!--i)
869 return -1;
870 *type++ = c;
872 *type = 0;
875 * The length must follow immediately, and be in canonical
876 * decimal format (ie "010" is not valid).
878 size = *hdr++ - '0';
879 if (size > 9)
880 return -1;
881 if (size) {
882 for (;;) {
883 unsigned long c = *hdr - '0';
884 if (c > 9)
885 break;
886 hdr++;
887 size = size * 10 + c;
890 *sizep = size;
893 * The length must be followed by a zero byte
895 return *hdr ? -1 : 0;
898 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
900 int ret;
901 z_stream stream;
902 char hdr[8192];
904 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
905 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
906 return NULL;
908 return unpack_sha1_rest(&stream, hdr, *size);
911 static unsigned long get_delta_base(struct packed_git *p,
912 struct pack_window **w_curs,
913 unsigned long offset,
914 enum object_type kind,
915 unsigned long delta_obj_offset,
916 unsigned long *base_obj_offset)
918 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
919 unsigned long base_offset;
921 /* use_pack() assured us we have [base_info, base_info + 20)
922 * as a range that we can look at without walking off the
923 * end of the mapped window. Its actually the hash size
924 * that is assured. An OFS_DELTA longer than the hash size
925 * is stupid, as then a REF_DELTA would be smaller to store.
927 if (kind == OBJ_OFS_DELTA) {
928 unsigned used = 0;
929 unsigned char c = base_info[used++];
930 base_offset = c & 127;
931 while (c & 128) {
932 base_offset += 1;
933 if (!base_offset || base_offset & ~(~0UL >> 7))
934 die("offset value overflow for delta base object");
935 c = base_info[used++];
936 base_offset = (base_offset << 7) + (c & 127);
938 base_offset = delta_obj_offset - base_offset;
939 if (base_offset >= delta_obj_offset)
940 die("delta base offset out of bound");
941 offset += used;
942 } else if (kind == OBJ_REF_DELTA) {
943 /* The base entry _must_ be in the same pack */
944 base_offset = find_pack_entry_one(base_info, p);
945 if (!base_offset)
946 die("failed to find delta-pack base object %s",
947 sha1_to_hex(base_info));
948 offset += 20;
949 } else
950 die("I am totally screwed");
951 *base_obj_offset = base_offset;
952 return offset;
955 /* forward declaration for a mutually recursive function */
956 static int packed_object_info(struct packed_git *p, unsigned long offset,
957 char *type, unsigned long *sizep);
959 static int packed_delta_info(struct packed_git *p,
960 struct pack_window **w_curs,
961 unsigned long offset,
962 enum object_type kind,
963 unsigned long obj_offset,
964 char *type,
965 unsigned long *sizep)
967 unsigned long base_offset;
969 offset = get_delta_base(p, w_curs, offset, kind,
970 obj_offset, &base_offset);
972 /* We choose to only get the type of the base object and
973 * ignore potentially corrupt pack file that expects the delta
974 * based on a base with a wrong size. This saves tons of
975 * inflate() calls.
977 if (packed_object_info(p, base_offset, type, NULL))
978 die("cannot get info for delta-pack base");
980 if (sizep) {
981 const unsigned char *data;
982 unsigned char delta_head[20], *in;
983 unsigned long result_size;
984 z_stream stream;
985 int st;
987 memset(&stream, 0, sizeof(stream));
988 stream.next_out = delta_head;
989 stream.avail_out = sizeof(delta_head);
991 inflateInit(&stream);
992 do {
993 in = use_pack(p, w_curs, offset, &stream.avail_in);
994 stream.next_in = in;
995 st = inflate(&stream, Z_FINISH);
996 offset += stream.next_in - in;
997 } while ((st == Z_OK || st == Z_BUF_ERROR)
998 && stream.total_out < sizeof(delta_head));
999 inflateEnd(&stream);
1000 if ((st != Z_STREAM_END) &&
1001 stream.total_out != sizeof(delta_head))
1002 die("delta data unpack-initial failed");
1004 /* Examine the initial part of the delta to figure out
1005 * the result size.
1007 data = delta_head;
1009 /* ignore base size */
1010 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1012 /* Read the result size */
1013 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1014 *sizep = result_size;
1016 return 0;
1019 static unsigned long unpack_object_header(struct packed_git *p,
1020 struct pack_window **w_curs,
1021 unsigned long offset,
1022 enum object_type *type,
1023 unsigned long *sizep)
1025 unsigned char *base;
1026 unsigned int left;
1027 unsigned long used;
1029 /* use_pack() assures us we have [base, base + 20) available
1030 * as a range that we can look at at. (Its actually the hash
1031 * size that is assurred.) With our object header encoding
1032 * the maximum deflated object size is 2^137, which is just
1033 * insane, so we know won't exceed what we have been given.
1035 base = use_pack(p, w_curs, offset, &left);
1036 used = unpack_object_header_gently(base, left, type, sizep);
1037 if (!used)
1038 die("object offset outside of pack file");
1040 return offset + used;
1043 void packed_object_info_detail(struct packed_git *p,
1044 unsigned long offset,
1045 char *type,
1046 unsigned long *size,
1047 unsigned long *store_size,
1048 unsigned int *delta_chain_length,
1049 unsigned char *base_sha1)
1051 struct pack_window *w_curs = NULL;
1052 unsigned long obj_offset, val;
1053 unsigned char *next_sha1;
1054 enum object_type kind;
1056 *delta_chain_length = 0;
1057 obj_offset = offset;
1058 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1060 for (;;) {
1061 switch (kind) {
1062 default:
1063 die("pack %s contains unknown object type %d",
1064 p->pack_name, kind);
1065 case OBJ_COMMIT:
1066 case OBJ_TREE:
1067 case OBJ_BLOB:
1068 case OBJ_TAG:
1069 strcpy(type, type_names[kind]);
1070 *store_size = 0; /* notyet */
1071 unuse_pack(&w_curs);
1072 return;
1073 case OBJ_OFS_DELTA:
1074 get_delta_base(p, &w_curs, offset, kind,
1075 obj_offset, &offset);
1076 if (*delta_chain_length == 0) {
1077 /* TODO: find base_sha1 as pointed by offset */
1079 break;
1080 case OBJ_REF_DELTA:
1081 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1082 if (*delta_chain_length == 0)
1083 hashcpy(base_sha1, next_sha1);
1084 offset = find_pack_entry_one(next_sha1, p);
1085 break;
1087 obj_offset = offset;
1088 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1089 (*delta_chain_length)++;
1093 static int packed_object_info(struct packed_git *p, unsigned long offset,
1094 char *type, unsigned long *sizep)
1096 struct pack_window *w_curs = NULL;
1097 unsigned long size, obj_offset = offset;
1098 enum object_type kind;
1099 int r;
1101 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1103 switch (kind) {
1104 case OBJ_OFS_DELTA:
1105 case OBJ_REF_DELTA:
1106 r = packed_delta_info(p, &w_curs, offset, kind,
1107 obj_offset, type, sizep);
1108 unuse_pack(&w_curs);
1109 return r;
1110 case OBJ_COMMIT:
1111 case OBJ_TREE:
1112 case OBJ_BLOB:
1113 case OBJ_TAG:
1114 strcpy(type, type_names[kind]);
1115 unuse_pack(&w_curs);
1116 break;
1117 default:
1118 die("pack %s contains unknown object type %d",
1119 p->pack_name, kind);
1121 if (sizep)
1122 *sizep = size;
1123 return 0;
1126 static void *unpack_compressed_entry(struct packed_git *p,
1127 struct pack_window **w_curs,
1128 unsigned long offset,
1129 unsigned long size)
1131 int st;
1132 z_stream stream;
1133 unsigned char *buffer, *in;
1135 buffer = xmalloc(size + 1);
1136 buffer[size] = 0;
1137 memset(&stream, 0, sizeof(stream));
1138 stream.next_out = buffer;
1139 stream.avail_out = size;
1141 inflateInit(&stream);
1142 do {
1143 in = use_pack(p, w_curs, offset, &stream.avail_in);
1144 stream.next_in = in;
1145 st = inflate(&stream, Z_FINISH);
1146 offset += stream.next_in - in;
1147 } while (st == Z_OK || st == Z_BUF_ERROR);
1148 inflateEnd(&stream);
1149 if ((st != Z_STREAM_END) || stream.total_out != size) {
1150 free(buffer);
1151 return NULL;
1154 return buffer;
1157 static void *unpack_delta_entry(struct packed_git *p,
1158 struct pack_window **w_curs,
1159 unsigned long offset,
1160 unsigned long delta_size,
1161 enum object_type kind,
1162 unsigned long obj_offset,
1163 char *type,
1164 unsigned long *sizep)
1166 void *delta_data, *result, *base;
1167 unsigned long result_size, base_size, base_offset;
1169 offset = get_delta_base(p, w_curs, offset, kind,
1170 obj_offset, &base_offset);
1171 base = unpack_entry(p, base_offset, type, &base_size);
1172 if (!base)
1173 die("failed to read delta base object at %lu from %s",
1174 base_offset, p->pack_name);
1176 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1177 result = patch_delta(base, base_size,
1178 delta_data, delta_size,
1179 &result_size);
1180 if (!result)
1181 die("failed to apply delta");
1182 free(delta_data);
1183 free(base);
1184 *sizep = result_size;
1185 return result;
1188 void *unpack_entry(struct packed_git *p, unsigned long offset,
1189 char *type, unsigned long *sizep)
1191 struct pack_window *w_curs = NULL;
1192 unsigned long size, obj_offset = offset;
1193 enum object_type kind;
1194 void *retval;
1196 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1197 switch (kind) {
1198 case OBJ_OFS_DELTA:
1199 case OBJ_REF_DELTA:
1200 retval = unpack_delta_entry(p, &w_curs, offset, size,
1201 kind, obj_offset, type, sizep);
1202 break;
1203 case OBJ_COMMIT:
1204 case OBJ_TREE:
1205 case OBJ_BLOB:
1206 case OBJ_TAG:
1207 strcpy(type, type_names[kind]);
1208 *sizep = size;
1209 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1210 break;
1211 default:
1212 die("unknown object type %i in %s", kind, p->pack_name);
1214 unuse_pack(&w_curs);
1215 return retval;
1218 int num_packed_objects(const struct packed_git *p)
1220 /* See check_packed_git_idx() */
1221 return (p->index_size - 20 - 20 - 4*256) / 24;
1224 int nth_packed_object_sha1(const struct packed_git *p, int n,
1225 unsigned char* sha1)
1227 void *index = p->index_base + 256;
1228 if (n < 0 || num_packed_objects(p) <= n)
1229 return -1;
1230 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1231 return 0;
1234 unsigned long find_pack_entry_one(const unsigned char *sha1,
1235 struct packed_git *p)
1237 unsigned int *level1_ofs = p->index_base;
1238 int hi = ntohl(level1_ofs[*sha1]);
1239 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1240 void *index = p->index_base + 256;
1242 do {
1243 int mi = (lo + hi) / 2;
1244 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1245 if (!cmp)
1246 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1247 if (cmp > 0)
1248 hi = mi;
1249 else
1250 lo = mi+1;
1251 } while (lo < hi);
1252 return 0;
1255 static int matches_pack_name(struct packed_git *p, const char *ig)
1257 const char *last_c, *c;
1259 if (!strcmp(p->pack_name, ig))
1260 return 0;
1262 for (c = p->pack_name, last_c = c; *c;)
1263 if (*c == '/')
1264 last_c = ++c;
1265 else
1266 ++c;
1267 if (!strcmp(last_c, ig))
1268 return 0;
1270 return 1;
1273 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1275 struct packed_git *p;
1276 unsigned long offset;
1278 prepare_packed_git();
1280 for (p = packed_git; p; p = p->next) {
1281 if (ignore_packed) {
1282 const char **ig;
1283 for (ig = ignore_packed; *ig; ig++)
1284 if (!matches_pack_name(p, *ig))
1285 break;
1286 if (*ig)
1287 continue;
1289 offset = find_pack_entry_one(sha1, p);
1290 if (offset) {
1291 e->offset = offset;
1292 e->p = p;
1293 hashcpy(e->sha1, sha1);
1294 return 1;
1297 return 0;
1300 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1301 struct packed_git *packs)
1303 struct packed_git *p;
1305 for (p = packs; p; p = p->next) {
1306 if (find_pack_entry_one(sha1, p))
1307 return p;
1309 return NULL;
1313 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1315 int status;
1316 unsigned long mapsize, size;
1317 void *map;
1318 z_stream stream;
1319 char hdr[128];
1321 map = map_sha1_file(sha1, &mapsize);
1322 if (!map)
1323 return error("unable to find %s", sha1_to_hex(sha1));
1324 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1325 status = error("unable to unpack %s header",
1326 sha1_to_hex(sha1));
1327 if (parse_sha1_header(hdr, type, &size) < 0)
1328 status = error("unable to parse %s header", sha1_to_hex(sha1));
1329 else {
1330 status = 0;
1331 if (sizep)
1332 *sizep = size;
1334 inflateEnd(&stream);
1335 munmap(map, mapsize);
1336 return status;
1339 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1341 struct pack_entry e;
1343 if (!find_pack_entry(sha1, &e, NULL)) {
1344 reprepare_packed_git();
1345 if (!find_pack_entry(sha1, &e, NULL))
1346 return sha1_loose_object_info(sha1, type, sizep);
1348 return packed_object_info(e.p, e.offset, type, sizep);
1351 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1353 struct pack_entry e;
1355 if (!find_pack_entry(sha1, &e, NULL)) {
1356 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1357 return NULL;
1359 return unpack_entry(e.p, e.offset, type, size);
1362 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1364 unsigned long mapsize;
1365 void *map, *buf;
1366 struct pack_entry e;
1368 if (find_pack_entry(sha1, &e, NULL))
1369 return read_packed_sha1(sha1, type, size);
1370 map = map_sha1_file(sha1, &mapsize);
1371 if (map) {
1372 buf = unpack_sha1_file(map, mapsize, type, size);
1373 munmap(map, mapsize);
1374 return buf;
1376 reprepare_packed_git();
1377 if (find_pack_entry(sha1, &e, NULL))
1378 return read_packed_sha1(sha1, type, size);
1379 return NULL;
1382 void *read_object_with_reference(const unsigned char *sha1,
1383 const char *required_type,
1384 unsigned long *size,
1385 unsigned char *actual_sha1_return)
1387 char type[20];
1388 void *buffer;
1389 unsigned long isize;
1390 unsigned char actual_sha1[20];
1392 hashcpy(actual_sha1, sha1);
1393 while (1) {
1394 int ref_length = -1;
1395 const char *ref_type = NULL;
1397 buffer = read_sha1_file(actual_sha1, type, &isize);
1398 if (!buffer)
1399 return NULL;
1400 if (!strcmp(type, required_type)) {
1401 *size = isize;
1402 if (actual_sha1_return)
1403 hashcpy(actual_sha1_return, actual_sha1);
1404 return buffer;
1406 /* Handle references */
1407 else if (!strcmp(type, commit_type))
1408 ref_type = "tree ";
1409 else if (!strcmp(type, tag_type))
1410 ref_type = "object ";
1411 else {
1412 free(buffer);
1413 return NULL;
1415 ref_length = strlen(ref_type);
1417 if (memcmp(buffer, ref_type, ref_length) ||
1418 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1419 free(buffer);
1420 return NULL;
1422 free(buffer);
1423 /* Now we have the ID of the referred-to object in
1424 * actual_sha1. Check again. */
1428 static void write_sha1_file_prepare(void *buf, unsigned long len,
1429 const char *type, unsigned char *sha1,
1430 unsigned char *hdr, int *hdrlen)
1432 SHA_CTX c;
1434 /* Generate the header */
1435 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1437 /* Sha1.. */
1438 SHA1_Init(&c);
1439 SHA1_Update(&c, hdr, *hdrlen);
1440 SHA1_Update(&c, buf, len);
1441 SHA1_Final(sha1, &c);
1445 * Link the tempfile to the final place, possibly creating the
1446 * last directory level as you do so.
1448 * Returns the errno on failure, 0 on success.
1450 static int link_temp_to_file(const char *tmpfile, const char *filename)
1452 int ret;
1453 char *dir;
1455 if (!link(tmpfile, filename))
1456 return 0;
1459 * Try to mkdir the last path component if that failed.
1461 * Re-try the "link()" regardless of whether the mkdir
1462 * succeeds, since a race might mean that somebody
1463 * else succeeded.
1465 ret = errno;
1466 dir = strrchr(filename, '/');
1467 if (dir) {
1468 *dir = 0;
1469 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1470 *dir = '/';
1471 return -2;
1473 *dir = '/';
1474 if (!link(tmpfile, filename))
1475 return 0;
1476 ret = errno;
1478 return ret;
1482 * Move the just written object into its final resting place
1484 int move_temp_to_file(const char *tmpfile, const char *filename)
1486 int ret = link_temp_to_file(tmpfile, filename);
1489 * Coda hack - coda doesn't like cross-directory links,
1490 * so we fall back to a rename, which will mean that it
1491 * won't be able to check collisions, but that's not a
1492 * big deal.
1494 * The same holds for FAT formatted media.
1496 * When this succeeds, we just return 0. We have nothing
1497 * left to unlink.
1499 if (ret && ret != EEXIST) {
1500 if (!rename(tmpfile, filename))
1501 return 0;
1502 ret = errno;
1504 unlink(tmpfile);
1505 if (ret) {
1506 if (ret != EEXIST) {
1507 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1509 /* FIXME!!! Collision check here ? */
1512 return 0;
1515 static int write_buffer(int fd, const void *buf, size_t len)
1517 while (len) {
1518 ssize_t size;
1520 size = write(fd, buf, len);
1521 if (!size)
1522 return error("file write: disk full");
1523 if (size < 0) {
1524 if (errno == EINTR || errno == EAGAIN)
1525 continue;
1526 return error("file write error (%s)", strerror(errno));
1528 len -= size;
1529 buf = (char *) buf + size;
1531 return 0;
1534 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1536 int hdr_len;
1537 unsigned char c;
1539 c = (type << 4) | (len & 15);
1540 len >>= 4;
1541 hdr_len = 1;
1542 while (len) {
1543 *hdr++ = c | 0x80;
1544 hdr_len++;
1545 c = (len & 0x7f);
1546 len >>= 7;
1548 *hdr = c;
1549 return hdr_len;
1552 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1554 int obj_type, hdr;
1556 if (use_legacy_headers) {
1557 while (deflate(stream, 0) == Z_OK)
1558 /* nothing */;
1559 return;
1561 if (!strcmp(type, blob_type))
1562 obj_type = OBJ_BLOB;
1563 else if (!strcmp(type, tree_type))
1564 obj_type = OBJ_TREE;
1565 else if (!strcmp(type, commit_type))
1566 obj_type = OBJ_COMMIT;
1567 else if (!strcmp(type, tag_type))
1568 obj_type = OBJ_TAG;
1569 else
1570 die("trying to generate bogus object of type '%s'", type);
1571 hdr = write_binary_header(stream->next_out, obj_type, len);
1572 stream->total_out = hdr;
1573 stream->next_out += hdr;
1574 stream->avail_out -= hdr;
1577 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1578 unsigned char *sha1)
1580 unsigned char hdr[50];
1581 int hdrlen;
1582 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1583 return 0;
1586 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1588 int size;
1589 unsigned char *compressed;
1590 z_stream stream;
1591 unsigned char sha1[20];
1592 char *filename;
1593 static char tmpfile[PATH_MAX];
1594 unsigned char hdr[50];
1595 int fd, hdrlen;
1597 /* Normally if we have it in the pack then we do not bother writing
1598 * it out into .git/objects/??/?{38} file.
1600 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1601 filename = sha1_file_name(sha1);
1602 if (returnsha1)
1603 hashcpy(returnsha1, sha1);
1604 if (has_sha1_file(sha1))
1605 return 0;
1606 fd = open(filename, O_RDONLY);
1607 if (fd >= 0) {
1609 * FIXME!!! We might do collision checking here, but we'd
1610 * need to uncompress the old file and check it. Later.
1612 close(fd);
1613 return 0;
1616 if (errno != ENOENT) {
1617 return error("sha1 file %s: %s\n", filename, strerror(errno));
1620 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1622 fd = mkstemp(tmpfile);
1623 if (fd < 0) {
1624 if (errno == EPERM)
1625 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1626 else
1627 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1630 /* Set it up */
1631 memset(&stream, 0, sizeof(stream));
1632 deflateInit(&stream, zlib_compression_level);
1633 size = 8 + deflateBound(&stream, len+hdrlen);
1634 compressed = xmalloc(size);
1636 /* Compress it */
1637 stream.next_out = compressed;
1638 stream.avail_out = size;
1640 /* First header.. */
1641 stream.next_in = hdr;
1642 stream.avail_in = hdrlen;
1643 setup_object_header(&stream, type, len);
1645 /* Then the data itself.. */
1646 stream.next_in = buf;
1647 stream.avail_in = len;
1648 while (deflate(&stream, Z_FINISH) == Z_OK)
1649 /* nothing */;
1650 deflateEnd(&stream);
1651 size = stream.total_out;
1653 if (write_buffer(fd, compressed, size) < 0)
1654 die("unable to write sha1 file");
1655 fchmod(fd, 0444);
1656 close(fd);
1657 free(compressed);
1659 return move_temp_to_file(tmpfile, filename);
1663 * We need to unpack and recompress the object for writing
1664 * it out to a different file.
1666 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1668 size_t size;
1669 z_stream stream;
1670 unsigned char *unpacked;
1671 unsigned long len;
1672 char type[20];
1673 char hdr[50];
1674 int hdrlen;
1675 void *buf;
1677 /* need to unpack and recompress it by itself */
1678 unpacked = read_packed_sha1(sha1, type, &len);
1680 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1682 /* Set it up */
1683 memset(&stream, 0, sizeof(stream));
1684 deflateInit(&stream, zlib_compression_level);
1685 size = deflateBound(&stream, len + hdrlen);
1686 buf = xmalloc(size);
1688 /* Compress it */
1689 stream.next_out = buf;
1690 stream.avail_out = size;
1692 /* First header.. */
1693 stream.next_in = (void *)hdr;
1694 stream.avail_in = hdrlen;
1695 while (deflate(&stream, 0) == Z_OK)
1696 /* nothing */;
1698 /* Then the data itself.. */
1699 stream.next_in = unpacked;
1700 stream.avail_in = len;
1701 while (deflate(&stream, Z_FINISH) == Z_OK)
1702 /* nothing */;
1703 deflateEnd(&stream);
1704 free(unpacked);
1706 *objsize = stream.total_out;
1707 return buf;
1710 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1712 int retval;
1713 unsigned long objsize;
1714 void *buf = map_sha1_file(sha1, &objsize);
1716 if (buf) {
1717 retval = write_buffer(fd, buf, objsize);
1718 munmap(buf, objsize);
1719 return retval;
1722 buf = repack_object(sha1, &objsize);
1723 retval = write_buffer(fd, buf, objsize);
1724 free(buf);
1725 return retval;
1728 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1729 size_t bufsize, size_t *bufposn)
1731 char tmpfile[PATH_MAX];
1732 int local;
1733 z_stream stream;
1734 unsigned char real_sha1[20];
1735 unsigned char discard[4096];
1736 int ret;
1737 SHA_CTX c;
1739 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1741 local = mkstemp(tmpfile);
1742 if (local < 0) {
1743 if (errno == EPERM)
1744 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1745 else
1746 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1749 memset(&stream, 0, sizeof(stream));
1751 inflateInit(&stream);
1753 SHA1_Init(&c);
1755 do {
1756 ssize_t size;
1757 if (*bufposn) {
1758 stream.avail_in = *bufposn;
1759 stream.next_in = (unsigned char *) buffer;
1760 do {
1761 stream.next_out = discard;
1762 stream.avail_out = sizeof(discard);
1763 ret = inflate(&stream, Z_SYNC_FLUSH);
1764 SHA1_Update(&c, discard, sizeof(discard) -
1765 stream.avail_out);
1766 } while (stream.avail_in && ret == Z_OK);
1767 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1768 die("unable to write sha1 file");
1769 memmove(buffer, buffer + *bufposn - stream.avail_in,
1770 stream.avail_in);
1771 *bufposn = stream.avail_in;
1772 if (ret != Z_OK)
1773 break;
1775 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1776 if (size <= 0) {
1777 close(local);
1778 unlink(tmpfile);
1779 if (!size)
1780 return error("Connection closed?");
1781 perror("Reading from connection");
1782 return -1;
1784 *bufposn += size;
1785 } while (1);
1786 inflateEnd(&stream);
1788 close(local);
1789 SHA1_Final(real_sha1, &c);
1790 if (ret != Z_STREAM_END) {
1791 unlink(tmpfile);
1792 return error("File %s corrupted", sha1_to_hex(sha1));
1794 if (hashcmp(sha1, real_sha1)) {
1795 unlink(tmpfile);
1796 return error("File %s has bad hash", sha1_to_hex(sha1));
1799 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1802 int has_pack_index(const unsigned char *sha1)
1804 struct stat st;
1805 if (stat(sha1_pack_index_name(sha1), &st))
1806 return 0;
1807 return 1;
1810 int has_pack_file(const unsigned char *sha1)
1812 struct stat st;
1813 if (stat(sha1_pack_name(sha1), &st))
1814 return 0;
1815 return 1;
1818 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1820 struct pack_entry e;
1821 return find_pack_entry(sha1, &e, ignore_packed);
1824 int has_sha1_file(const unsigned char *sha1)
1826 struct stat st;
1827 struct pack_entry e;
1829 if (find_pack_entry(sha1, &e, NULL))
1830 return 1;
1831 return find_sha1_file(sha1, &st) ? 1 : 0;
1835 * reads from fd as long as possible into a supplied buffer of size bytes.
1836 * If necessary the buffer's size is increased using realloc()
1838 * returns 0 if anything went fine and -1 otherwise
1840 * NOTE: both buf and size may change, but even when -1 is returned
1841 * you still have to free() it yourself.
1843 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1845 char* buf = *return_buf;
1846 unsigned long size = *return_size;
1847 int iret;
1848 unsigned long off = 0;
1850 do {
1851 iret = xread(fd, buf + off, size - off);
1852 if (iret > 0) {
1853 off += iret;
1854 if (off == size) {
1855 size *= 2;
1856 buf = xrealloc(buf, size);
1859 } while (iret > 0);
1861 *return_buf = buf;
1862 *return_size = off;
1864 if (iret < 0)
1865 return -1;
1866 return 0;
1869 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1871 unsigned long size = 4096;
1872 char *buf = xmalloc(size);
1873 int ret;
1875 if (read_pipe(fd, &buf, &size)) {
1876 free(buf);
1877 return -1;
1880 if (!type)
1881 type = blob_type;
1882 if (write_object)
1883 ret = write_sha1_file(buf, size, type, sha1);
1884 else
1885 ret = hash_sha1_file(buf, size, type, sha1);
1886 free(buf);
1887 return ret;
1890 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1892 unsigned long size = st->st_size;
1893 void *buf;
1894 int ret;
1896 buf = "";
1897 if (size)
1898 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1899 close(fd);
1900 if (buf == MAP_FAILED)
1901 return -1;
1903 if (!type)
1904 type = blob_type;
1905 if (write_object)
1906 ret = write_sha1_file(buf, size, type, sha1);
1907 else
1908 ret = hash_sha1_file(buf, size, type, sha1);
1909 if (size)
1910 munmap(buf, size);
1911 return ret;
1914 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1916 int fd;
1917 char *target;
1919 switch (st->st_mode & S_IFMT) {
1920 case S_IFREG:
1921 fd = open(path, O_RDONLY);
1922 if (fd < 0)
1923 return error("open(\"%s\"): %s", path,
1924 strerror(errno));
1925 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1926 return error("%s: failed to insert into database",
1927 path);
1928 break;
1929 case S_IFLNK:
1930 target = xmalloc(st->st_size+1);
1931 if (readlink(path, target, st->st_size+1) != st->st_size) {
1932 char *errstr = strerror(errno);
1933 free(target);
1934 return error("readlink(\"%s\"): %s", path,
1935 errstr);
1937 if (!write_object)
1938 hash_sha1_file(target, st->st_size, blob_type, sha1);
1939 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1940 return error("%s: failed to insert into database",
1941 path);
1942 free(target);
1943 break;
1944 default:
1945 return error("%s: unsupported file type", path);
1947 return 0;