34669693448b9dd106ed6af66816d9e4096cd0d7
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
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 */
62 };
64 int get_sha1_hex(const char *hex, unsigned char *sha1)
65 {
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;
73 }
74 return 0;
75 }
77 int safe_create_leading_directories(char *path)
78 {
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;
95 }
96 }
97 else if (mkdir(path, 0777)) {
98 *pos = '/';
99 return -1;
100 }
101 else if (adjust_shared_perm(path)) {
102 *pos = '/';
103 return -2;
104 }
105 *pos++ = '/';
106 }
107 return 0;
108 }
110 char * sha1_to_hex(const unsigned char *sha1)
111 {
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];
122 }
123 *buf = '\0';
125 return buffer;
126 }
128 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
129 {
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];
137 }
138 }
140 /*
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.
144 *
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.
149 */
150 char *sha1_file_name(const unsigned char *sha1)
151 {
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;
163 }
164 fill_sha1_path(name, sha1);
165 return base;
166 }
168 char *sha1_pack_name(const unsigned char *sha1)
169 {
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;
180 }
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];
188 }
190 return base;
191 }
193 char *sha1_pack_index_name(const unsigned char *sha1)
194 {
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;
205 }
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];
213 }
215 return base;
216 }
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);
223 /*
224 * Prepare alternate object database registry.
225 *
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.
237 */
238 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
239 {
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;
255 }
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);
262 }
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;
277 }
279 /* Prevent the common mistake of listing the same
280 * thing twice, or object directory itself.
281 */
282 for (alt = alt_odb_list; alt; alt = alt->next) {
283 if (!memcmp(ent->base, alt->base, pfxlen)) {
284 free(ent);
285 return -1;
286 }
287 }
288 if (!memcmp(ent->base, objdir, pfxlen)) {
289 free(ent);
290 return -1;
291 }
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;
304 }
306 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
307 const char *relative_base, int depth)
308 {
309 const char *cp, *last;
311 if (depth > 5) {
312 error("%s: ignoring alternate object stores, nesting too deep.",
313 relative_base);
314 return;
315 }
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;
325 }
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);
335 }
336 }
337 while (cp < ep && *cp == sep)
338 cp++;
339 last = cp;
340 }
341 }
343 static void read_info_alternates(const char * relative_base, int depth)
344 {
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;
357 }
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);
366 }
368 void prepare_alt_odb(void)
369 {
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);
381 }
383 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
384 {
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;
396 }
397 return NULL;
398 }
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_)
406 {
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;
418 }
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;
438 }
440 /*
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
446 */
447 if (idx_size != 4*256 + nr * 24 + 20 + 20)
448 return error("wrong index file size");
450 return 0;
451 }
453 static int unuse_one_packed_git(void)
454 {
455 struct packed_git *p, *lru = NULL;
457 for (p = packed_git; p; p = p->next) {
458 if (!p->windows || p->windows->inuse_cnt)
459 continue;
460 if (!lru || p->windows->last_used < lru->windows->last_used)
461 lru = p;
462 }
463 if (!lru)
464 return 0;
465 munmap(lru->windows->base, lru->windows->len);
466 free(lru->windows);
467 lru->windows = NULL;
468 close(p->pack_fd);
469 p->pack_fd = -1;
470 return 1;
471 }
473 void unuse_pack(struct pack_window **w_cursor)
474 {
475 struct pack_window *w = *w_cursor;
476 if (w) {
477 w->inuse_cnt--;
478 *w_cursor = NULL;
479 }
480 }
482 static void open_packed_git(struct packed_git *p)
483 {
484 struct stat st;
485 struct pack_header hdr;
486 unsigned char sha1[20];
487 unsigned char *idx_sha1;
489 p->pack_fd = open(p->pack_name, O_RDONLY);
490 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
491 die("packfile %s cannot be opened", p->pack_name);
493 /* If we created the struct before we had the pack we lack size. */
494 if (!p->pack_size) {
495 if (!S_ISREG(st.st_mode))
496 die("packfile %s not a regular file", p->pack_name);
497 p->pack_size = st.st_size;
498 } else if (p->pack_size != st.st_size)
499 die("packfile %s size changed", p->pack_name);
501 /* Verify we recognize this pack file format. */
502 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
503 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
504 die("file %s is not a GIT packfile", p->pack_name);
505 if (!pack_version_ok(hdr.hdr_version))
506 die("packfile %s is version %u and not supported"
507 " (try upgrading GIT to a newer version)",
508 p->pack_name, ntohl(hdr.hdr_version));
510 /* Verify the pack matches its index. */
511 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
512 die("packfile %s claims to have %u objects"
513 " while index size indicates %u objects",
514 p->pack_name, ntohl(hdr.hdr_entries),
515 num_packed_objects(p));
516 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
517 die("end of packfile %s is unavailable", p->pack_name);
518 read_or_die(p->pack_fd, sha1, sizeof(sha1));
519 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
520 if (hashcmp(sha1, idx_sha1))
521 die("packfile %s does not match index", p->pack_name);
522 }
524 unsigned char* use_pack(struct packed_git *p,
525 struct pack_window **w_cursor,
526 unsigned long offset,
527 unsigned int *left)
528 {
529 struct pack_window *win = p->windows;
531 if (p->pack_fd == -1)
532 open_packed_git(p);
533 if (!win) {
534 pack_mapped += p->pack_size;
535 while (packed_git_limit < pack_mapped && unuse_one_packed_git())
536 ; /* nothing */
537 win = xcalloc(1, sizeof(*win));
538 win->len = p->pack_size;
539 win->base = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, p->pack_fd, 0);
540 if (win->base == MAP_FAILED)
541 die("packfile %s cannot be mapped.", p->pack_name);
542 p->windows = win;
543 }
544 if (win != *w_cursor) {
545 win->last_used = pack_used_ctr++;
546 win->inuse_cnt++;
547 *w_cursor = win;
548 }
549 if (left)
550 *left = win->len - offset;
551 return win->base + offset;
552 }
554 struct packed_git *add_packed_git(char *path, int path_len, int local)
555 {
556 struct stat st;
557 struct packed_git *p;
558 unsigned long idx_size;
559 void *idx_map;
560 unsigned char sha1[20];
562 if (check_packed_git_idx(path, &idx_size, &idx_map))
563 return NULL;
565 /* do we have a corresponding .pack file? */
566 strcpy(path + path_len - 4, ".pack");
567 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
568 munmap(idx_map, idx_size);
569 return NULL;
570 }
571 /* ok, it looks sane as far as we can check without
572 * actually mapping the pack file.
573 */
574 p = xmalloc(sizeof(*p) + path_len + 2);
575 strcpy(p->pack_name, path);
576 p->index_size = idx_size;
577 p->pack_size = st.st_size;
578 p->index_base = idx_map;
579 p->next = NULL;
580 p->windows = NULL;
581 p->pack_fd = -1;
582 p->pack_local = local;
583 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
584 hashcpy(p->sha1, sha1);
585 return p;
586 }
588 struct packed_git *parse_pack_index(unsigned char *sha1)
589 {
590 char *path = sha1_pack_index_name(sha1);
591 return parse_pack_index_file(sha1, path);
592 }
594 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
595 {
596 struct packed_git *p;
597 unsigned long idx_size;
598 void *idx_map;
599 char *path;
601 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
602 return NULL;
604 path = sha1_pack_name(sha1);
606 p = xmalloc(sizeof(*p) + strlen(path) + 2);
607 strcpy(p->pack_name, path);
608 p->index_size = idx_size;
609 p->pack_size = 0;
610 p->index_base = idx_map;
611 p->next = NULL;
612 p->windows = NULL;
613 p->pack_fd = -1;
614 hashcpy(p->sha1, sha1);
615 return p;
616 }
618 void install_packed_git(struct packed_git *pack)
619 {
620 pack->next = packed_git;
621 packed_git = pack;
622 }
624 static void prepare_packed_git_one(char *objdir, int local)
625 {
626 char path[PATH_MAX];
627 int len;
628 DIR *dir;
629 struct dirent *de;
631 sprintf(path, "%s/pack", objdir);
632 len = strlen(path);
633 dir = opendir(path);
634 if (!dir) {
635 if (errno != ENOENT)
636 error("unable to open object pack directory: %s: %s",
637 path, strerror(errno));
638 return;
639 }
640 path[len++] = '/';
641 while ((de = readdir(dir)) != NULL) {
642 int namelen = strlen(de->d_name);
643 struct packed_git *p;
645 if (!has_extension(de->d_name, ".idx"))
646 continue;
648 /* we have .idx. Is it a file we can map? */
649 strcpy(path + len, de->d_name);
650 for (p = packed_git; p; p = p->next) {
651 if (!memcmp(path, p->pack_name, len + namelen - 4))
652 break;
653 }
654 if (p)
655 continue;
656 p = add_packed_git(path, len + namelen, local);
657 if (!p)
658 continue;
659 p->next = packed_git;
660 packed_git = p;
661 }
662 closedir(dir);
663 }
665 static int prepare_packed_git_run_once = 0;
666 void prepare_packed_git(void)
667 {
668 struct alternate_object_database *alt;
670 if (prepare_packed_git_run_once)
671 return;
672 prepare_packed_git_one(get_object_directory(), 1);
673 prepare_alt_odb();
674 for (alt = alt_odb_list; alt; alt = alt->next) {
675 alt->name[-1] = 0;
676 prepare_packed_git_one(alt->base, 0);
677 alt->name[-1] = '/';
678 }
679 prepare_packed_git_run_once = 1;
680 }
682 void reprepare_packed_git(void)
683 {
684 prepare_packed_git_run_once = 0;
685 prepare_packed_git();
686 }
688 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
689 {
690 unsigned char real_sha1[20];
691 hash_sha1_file(map, size, type, real_sha1);
692 return hashcmp(sha1, real_sha1) ? -1 : 0;
693 }
695 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
696 {
697 struct stat st;
698 void *map;
699 int fd;
700 char *filename = find_sha1_file(sha1, &st);
702 if (!filename) {
703 return NULL;
704 }
706 fd = open(filename, O_RDONLY | sha1_file_open_flag);
707 if (fd < 0) {
708 /* See if it works without O_NOATIME */
709 switch (sha1_file_open_flag) {
710 default:
711 fd = open(filename, O_RDONLY);
712 if (fd >= 0)
713 break;
714 /* Fallthrough */
715 case 0:
716 return NULL;
717 }
719 /* If it failed once, it will probably fail again.
720 * Stop using O_NOATIME
721 */
722 sha1_file_open_flag = 0;
723 }
724 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
725 close(fd);
726 if (map == MAP_FAILED)
727 return NULL;
728 *size = st.st_size;
729 return map;
730 }
732 int legacy_loose_object(unsigned char *map)
733 {
734 unsigned int word;
736 /*
737 * Is it a zlib-compressed buffer? If so, the first byte
738 * must be 0x78 (15-bit window size, deflated), and the
739 * first 16-bit word is evenly divisible by 31
740 */
741 word = (map[0] << 8) + map[1];
742 if (map[0] == 0x78 && !(word % 31))
743 return 1;
744 else
745 return 0;
746 }
748 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
749 {
750 unsigned shift;
751 unsigned char c;
752 unsigned long size;
753 unsigned long used = 0;
755 c = buf[used++];
756 *type = (c >> 4) & 7;
757 size = c & 15;
758 shift = 4;
759 while (c & 0x80) {
760 if (len <= used)
761 return 0;
762 if (sizeof(long) * 8 <= shift)
763 return 0;
764 c = buf[used++];
765 size += (c & 0x7f) << shift;
766 shift += 7;
767 }
768 *sizep = size;
769 return used;
770 }
772 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
773 {
774 unsigned long size, used;
775 static const char valid_loose_object_type[8] = {
776 0, /* OBJ_EXT */
777 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
778 0, /* "delta" and others are invalid in a loose object */
779 };
780 enum object_type type;
782 /* Get the data stream */
783 memset(stream, 0, sizeof(*stream));
784 stream->next_in = map;
785 stream->avail_in = mapsize;
786 stream->next_out = buffer;
787 stream->avail_out = bufsiz;
789 if (legacy_loose_object(map)) {
790 inflateInit(stream);
791 return inflate(stream, 0);
792 }
794 used = unpack_object_header_gently(map, mapsize, &type, &size);
795 if (!used || !valid_loose_object_type[type])
796 return -1;
797 map += used;
798 mapsize -= used;
800 /* Set up the stream for the rest.. */
801 stream->next_in = map;
802 stream->avail_in = mapsize;
803 inflateInit(stream);
805 /* And generate the fake traditional header */
806 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
807 type_names[type], size);
808 return 0;
809 }
811 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
812 {
813 int bytes = strlen(buffer) + 1;
814 unsigned char *buf = xmalloc(1+size);
815 unsigned long n;
817 n = stream->total_out - bytes;
818 if (n > size)
819 n = size;
820 memcpy(buf, (char *) buffer + bytes, n);
821 bytes = n;
822 if (bytes < size) {
823 stream->next_out = buf + bytes;
824 stream->avail_out = size - bytes;
825 while (inflate(stream, Z_FINISH) == Z_OK)
826 /* nothing */;
827 }
828 buf[size] = 0;
829 inflateEnd(stream);
830 return buf;
831 }
833 /*
834 * We used to just use "sscanf()", but that's actually way
835 * too permissive for what we want to check. So do an anal
836 * object header parse by hand.
837 */
838 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
839 {
840 int i;
841 unsigned long size;
843 /*
844 * The type can be at most ten bytes (including the
845 * terminating '\0' that we add), and is followed by
846 * a space.
847 */
848 i = 10;
849 for (;;) {
850 char c = *hdr++;
851 if (c == ' ')
852 break;
853 if (!--i)
854 return -1;
855 *type++ = c;
856 }
857 *type = 0;
859 /*
860 * The length must follow immediately, and be in canonical
861 * decimal format (ie "010" is not valid).
862 */
863 size = *hdr++ - '0';
864 if (size > 9)
865 return -1;
866 if (size) {
867 for (;;) {
868 unsigned long c = *hdr - '0';
869 if (c > 9)
870 break;
871 hdr++;
872 size = size * 10 + c;
873 }
874 }
875 *sizep = size;
877 /*
878 * The length must be followed by a zero byte
879 */
880 return *hdr ? -1 : 0;
881 }
883 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
884 {
885 int ret;
886 z_stream stream;
887 char hdr[8192];
889 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
890 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
891 return NULL;
893 return unpack_sha1_rest(&stream, hdr, *size);
894 }
896 static unsigned long get_delta_base(struct packed_git *p,
897 struct pack_window **w_curs,
898 unsigned long offset,
899 enum object_type kind,
900 unsigned long delta_obj_offset,
901 unsigned long *base_obj_offset)
902 {
903 unsigned char *base_info = use_pack(p, w_curs, offset, NULL);
904 unsigned long base_offset;
906 /* use_pack() assured us we have [base_info, base_info + 20)
907 * as a range that we can look at without walking off the
908 * end of the mapped window. Its actually the hash size
909 * that is assured. An OFS_DELTA longer than the hash size
910 * is stupid, as then a REF_DELTA would be smaller to store.
911 */
912 if (kind == OBJ_OFS_DELTA) {
913 unsigned used = 0;
914 unsigned char c = base_info[used++];
915 base_offset = c & 127;
916 while (c & 128) {
917 base_offset += 1;
918 if (!base_offset || base_offset & ~(~0UL >> 7))
919 die("offset value overflow for delta base object");
920 c = base_info[used++];
921 base_offset = (base_offset << 7) + (c & 127);
922 }
923 base_offset = delta_obj_offset - base_offset;
924 if (base_offset >= delta_obj_offset)
925 die("delta base offset out of bound");
926 offset += used;
927 } else if (kind == OBJ_REF_DELTA) {
928 /* The base entry _must_ be in the same pack */
929 base_offset = find_pack_entry_one(base_info, p);
930 if (!base_offset)
931 die("failed to find delta-pack base object %s",
932 sha1_to_hex(base_info));
933 offset += 20;
934 } else
935 die("I am totally screwed");
936 *base_obj_offset = base_offset;
937 return offset;
938 }
940 /* forward declaration for a mutually recursive function */
941 static int packed_object_info(struct packed_git *p, unsigned long offset,
942 char *type, unsigned long *sizep);
944 static int packed_delta_info(struct packed_git *p,
945 struct pack_window **w_curs,
946 unsigned long offset,
947 enum object_type kind,
948 unsigned long obj_offset,
949 char *type,
950 unsigned long *sizep)
951 {
952 unsigned long base_offset;
954 offset = get_delta_base(p, w_curs, offset, kind,
955 obj_offset, &base_offset);
957 /* We choose to only get the type of the base object and
958 * ignore potentially corrupt pack file that expects the delta
959 * based on a base with a wrong size. This saves tons of
960 * inflate() calls.
961 */
962 if (packed_object_info(p, base_offset, type, NULL))
963 die("cannot get info for delta-pack base");
965 if (sizep) {
966 const unsigned char *data;
967 unsigned char delta_head[20], *in;
968 unsigned long result_size;
969 z_stream stream;
970 int st;
972 memset(&stream, 0, sizeof(stream));
973 stream.next_out = delta_head;
974 stream.avail_out = sizeof(delta_head);
976 inflateInit(&stream);
977 do {
978 in = use_pack(p, w_curs, offset, &stream.avail_in);
979 stream.next_in = in;
980 st = inflate(&stream, Z_FINISH);
981 offset += stream.next_in - in;
982 } while ((st == Z_OK || st == Z_BUF_ERROR)
983 && stream.total_out < sizeof(delta_head));
984 inflateEnd(&stream);
985 if ((st != Z_STREAM_END) &&
986 stream.total_out != sizeof(delta_head))
987 die("delta data unpack-initial failed");
989 /* Examine the initial part of the delta to figure out
990 * the result size.
991 */
992 data = delta_head;
994 /* ignore base size */
995 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
997 /* Read the result size */
998 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
999 *sizep = result_size;
1000 }
1001 return 0;
1002 }
1004 static unsigned long unpack_object_header(struct packed_git *p,
1005 struct pack_window **w_curs,
1006 unsigned long offset,
1007 enum object_type *type,
1008 unsigned long *sizep)
1009 {
1010 unsigned char *base;
1011 unsigned int left;
1012 unsigned long used;
1014 /* use_pack() assures us we have [base, base + 20) available
1015 * as a range that we can look at at. (Its actually the hash
1016 * size that is assurred.) With our object header encoding
1017 * the maximum deflated object size is 2^137, which is just
1018 * insane, so we know won't exceed what we have been given.
1019 */
1020 base = use_pack(p, w_curs, offset, &left);
1021 used = unpack_object_header_gently(base, left, type, sizep);
1022 if (!used)
1023 die("object offset outside of pack file");
1025 return offset + used;
1026 }
1028 void packed_object_info_detail(struct packed_git *p,
1029 unsigned long offset,
1030 char *type,
1031 unsigned long *size,
1032 unsigned long *store_size,
1033 unsigned int *delta_chain_length,
1034 unsigned char *base_sha1)
1035 {
1036 struct pack_window *w_curs = NULL;
1037 unsigned long obj_offset, val;
1038 unsigned char *next_sha1;
1039 enum object_type kind;
1041 *delta_chain_length = 0;
1042 obj_offset = offset;
1043 offset = unpack_object_header(p, &w_curs, offset, &kind, size);
1045 for (;;) {
1046 switch (kind) {
1047 default:
1048 die("pack %s contains unknown object type %d",
1049 p->pack_name, kind);
1050 case OBJ_COMMIT:
1051 case OBJ_TREE:
1052 case OBJ_BLOB:
1053 case OBJ_TAG:
1054 strcpy(type, type_names[kind]);
1055 *store_size = 0; /* notyet */
1056 unuse_pack(&w_curs);
1057 return;
1058 case OBJ_OFS_DELTA:
1059 get_delta_base(p, &w_curs, offset, kind,
1060 obj_offset, &offset);
1061 if (*delta_chain_length == 0) {
1062 /* TODO: find base_sha1 as pointed by offset */
1063 }
1064 break;
1065 case OBJ_REF_DELTA:
1066 next_sha1 = use_pack(p, &w_curs, offset, NULL);
1067 if (*delta_chain_length == 0)
1068 hashcpy(base_sha1, next_sha1);
1069 offset = find_pack_entry_one(next_sha1, p);
1070 break;
1071 }
1072 obj_offset = offset;
1073 offset = unpack_object_header(p, &w_curs, offset, &kind, &val);
1074 (*delta_chain_length)++;
1075 }
1076 }
1078 static int packed_object_info(struct packed_git *p, unsigned long offset,
1079 char *type, unsigned long *sizep)
1080 {
1081 struct pack_window *w_curs = NULL;
1082 unsigned long size, obj_offset = offset;
1083 enum object_type kind;
1084 int r;
1086 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1088 switch (kind) {
1089 case OBJ_OFS_DELTA:
1090 case OBJ_REF_DELTA:
1091 r = packed_delta_info(p, &w_curs, offset, kind,
1092 obj_offset, type, sizep);
1093 unuse_pack(&w_curs);
1094 return r;
1095 case OBJ_COMMIT:
1096 case OBJ_TREE:
1097 case OBJ_BLOB:
1098 case OBJ_TAG:
1099 strcpy(type, type_names[kind]);
1100 unuse_pack(&w_curs);
1101 break;
1102 default:
1103 die("pack %s contains unknown object type %d",
1104 p->pack_name, kind);
1105 }
1106 if (sizep)
1107 *sizep = size;
1108 return 0;
1109 }
1111 static void *unpack_compressed_entry(struct packed_git *p,
1112 struct pack_window **w_curs,
1113 unsigned long offset,
1114 unsigned long size)
1115 {
1116 int st;
1117 z_stream stream;
1118 unsigned char *buffer, *in;
1120 buffer = xmalloc(size + 1);
1121 buffer[size] = 0;
1122 memset(&stream, 0, sizeof(stream));
1123 stream.next_out = buffer;
1124 stream.avail_out = size;
1126 inflateInit(&stream);
1127 do {
1128 in = use_pack(p, w_curs, offset, &stream.avail_in);
1129 stream.next_in = in;
1130 st = inflate(&stream, Z_FINISH);
1131 offset += stream.next_in - in;
1132 } while (st == Z_OK || st == Z_BUF_ERROR);
1133 inflateEnd(&stream);
1134 if ((st != Z_STREAM_END) || stream.total_out != size) {
1135 free(buffer);
1136 return NULL;
1137 }
1139 return buffer;
1140 }
1142 static void *unpack_delta_entry(struct packed_git *p,
1143 struct pack_window **w_curs,
1144 unsigned long offset,
1145 unsigned long delta_size,
1146 enum object_type kind,
1147 unsigned long obj_offset,
1148 char *type,
1149 unsigned long *sizep)
1150 {
1151 void *delta_data, *result, *base;
1152 unsigned long result_size, base_size, base_offset;
1154 offset = get_delta_base(p, w_curs, offset, kind,
1155 obj_offset, &base_offset);
1156 base = unpack_entry(p, base_offset, type, &base_size);
1157 if (!base)
1158 die("failed to read delta base object at %lu from %s",
1159 base_offset, p->pack_name);
1161 delta_data = unpack_compressed_entry(p, w_curs, offset, delta_size);
1162 result = patch_delta(base, base_size,
1163 delta_data, delta_size,
1164 &result_size);
1165 if (!result)
1166 die("failed to apply delta");
1167 free(delta_data);
1168 free(base);
1169 *sizep = result_size;
1170 return result;
1171 }
1173 void *unpack_entry(struct packed_git *p, unsigned long offset,
1174 char *type, unsigned long *sizep)
1175 {
1176 struct pack_window *w_curs = NULL;
1177 unsigned long size, obj_offset = offset;
1178 enum object_type kind;
1179 void *retval;
1181 offset = unpack_object_header(p, &w_curs, offset, &kind, &size);
1182 switch (kind) {
1183 case OBJ_OFS_DELTA:
1184 case OBJ_REF_DELTA:
1185 retval = unpack_delta_entry(p, &w_curs, offset, size,
1186 kind, obj_offset, type, sizep);
1187 break;
1188 case OBJ_COMMIT:
1189 case OBJ_TREE:
1190 case OBJ_BLOB:
1191 case OBJ_TAG:
1192 strcpy(type, type_names[kind]);
1193 *sizep = size;
1194 retval = unpack_compressed_entry(p, &w_curs, offset, size);
1195 break;
1196 default:
1197 die("unknown object type %i in %s", kind, p->pack_name);
1198 }
1199 unuse_pack(&w_curs);
1200 return retval;
1201 }
1203 int num_packed_objects(const struct packed_git *p)
1204 {
1205 /* See check_packed_git_idx() */
1206 return (p->index_size - 20 - 20 - 4*256) / 24;
1207 }
1209 int nth_packed_object_sha1(const struct packed_git *p, int n,
1210 unsigned char* sha1)
1211 {
1212 void *index = p->index_base + 256;
1213 if (n < 0 || num_packed_objects(p) <= n)
1214 return -1;
1215 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1216 return 0;
1217 }
1219 unsigned long find_pack_entry_one(const unsigned char *sha1,
1220 struct packed_git *p)
1221 {
1222 unsigned int *level1_ofs = p->index_base;
1223 int hi = ntohl(level1_ofs[*sha1]);
1224 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1225 void *index = p->index_base + 256;
1227 do {
1228 int mi = (lo + hi) / 2;
1229 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1230 if (!cmp)
1231 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1232 if (cmp > 0)
1233 hi = mi;
1234 else
1235 lo = mi+1;
1236 } while (lo < hi);
1237 return 0;
1238 }
1240 static int matches_pack_name(struct packed_git *p, const char *ig)
1241 {
1242 const char *last_c, *c;
1244 if (!strcmp(p->pack_name, ig))
1245 return 0;
1247 for (c = p->pack_name, last_c = c; *c;)
1248 if (*c == '/')
1249 last_c = ++c;
1250 else
1251 ++c;
1252 if (!strcmp(last_c, ig))
1253 return 0;
1255 return 1;
1256 }
1258 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1259 {
1260 struct packed_git *p;
1261 unsigned long offset;
1263 prepare_packed_git();
1265 for (p = packed_git; p; p = p->next) {
1266 if (ignore_packed) {
1267 const char **ig;
1268 for (ig = ignore_packed; *ig; ig++)
1269 if (!matches_pack_name(p, *ig))
1270 break;
1271 if (*ig)
1272 continue;
1273 }
1274 offset = find_pack_entry_one(sha1, p);
1275 if (offset) {
1276 e->offset = offset;
1277 e->p = p;
1278 hashcpy(e->sha1, sha1);
1279 return 1;
1280 }
1281 }
1282 return 0;
1283 }
1285 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1286 struct packed_git *packs)
1287 {
1288 struct packed_git *p;
1290 for (p = packs; p; p = p->next) {
1291 if (find_pack_entry_one(sha1, p))
1292 return p;
1293 }
1294 return NULL;
1296 }
1298 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1299 {
1300 int status;
1301 unsigned long mapsize, size;
1302 void *map;
1303 z_stream stream;
1304 char hdr[128];
1306 map = map_sha1_file(sha1, &mapsize);
1307 if (!map)
1308 return error("unable to find %s", sha1_to_hex(sha1));
1309 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1310 status = error("unable to unpack %s header",
1311 sha1_to_hex(sha1));
1312 if (parse_sha1_header(hdr, type, &size) < 0)
1313 status = error("unable to parse %s header", sha1_to_hex(sha1));
1314 else {
1315 status = 0;
1316 if (sizep)
1317 *sizep = size;
1318 }
1319 inflateEnd(&stream);
1320 munmap(map, mapsize);
1321 return status;
1322 }
1324 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1325 {
1326 struct pack_entry e;
1328 if (!find_pack_entry(sha1, &e, NULL)) {
1329 reprepare_packed_git();
1330 if (!find_pack_entry(sha1, &e, NULL))
1331 return sha1_loose_object_info(sha1, type, sizep);
1332 }
1333 return packed_object_info(e.p, e.offset, type, sizep);
1334 }
1336 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1337 {
1338 struct pack_entry e;
1340 if (!find_pack_entry(sha1, &e, NULL)) {
1341 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1342 return NULL;
1343 }
1344 return unpack_entry(e.p, e.offset, type, size);
1345 }
1347 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1348 {
1349 unsigned long mapsize;
1350 void *map, *buf;
1351 struct pack_entry e;
1353 if (find_pack_entry(sha1, &e, NULL))
1354 return read_packed_sha1(sha1, type, size);
1355 map = map_sha1_file(sha1, &mapsize);
1356 if (map) {
1357 buf = unpack_sha1_file(map, mapsize, type, size);
1358 munmap(map, mapsize);
1359 return buf;
1360 }
1361 reprepare_packed_git();
1362 if (find_pack_entry(sha1, &e, NULL))
1363 return read_packed_sha1(sha1, type, size);
1364 return NULL;
1365 }
1367 void *read_object_with_reference(const unsigned char *sha1,
1368 const char *required_type,
1369 unsigned long *size,
1370 unsigned char *actual_sha1_return)
1371 {
1372 char type[20];
1373 void *buffer;
1374 unsigned long isize;
1375 unsigned char actual_sha1[20];
1377 hashcpy(actual_sha1, sha1);
1378 while (1) {
1379 int ref_length = -1;
1380 const char *ref_type = NULL;
1382 buffer = read_sha1_file(actual_sha1, type, &isize);
1383 if (!buffer)
1384 return NULL;
1385 if (!strcmp(type, required_type)) {
1386 *size = isize;
1387 if (actual_sha1_return)
1388 hashcpy(actual_sha1_return, actual_sha1);
1389 return buffer;
1390 }
1391 /* Handle references */
1392 else if (!strcmp(type, commit_type))
1393 ref_type = "tree ";
1394 else if (!strcmp(type, tag_type))
1395 ref_type = "object ";
1396 else {
1397 free(buffer);
1398 return NULL;
1399 }
1400 ref_length = strlen(ref_type);
1402 if (memcmp(buffer, ref_type, ref_length) ||
1403 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1404 free(buffer);
1405 return NULL;
1406 }
1407 free(buffer);
1408 /* Now we have the ID of the referred-to object in
1409 * actual_sha1. Check again. */
1410 }
1411 }
1413 static void write_sha1_file_prepare(void *buf, unsigned long len,
1414 const char *type, unsigned char *sha1,
1415 unsigned char *hdr, int *hdrlen)
1416 {
1417 SHA_CTX c;
1419 /* Generate the header */
1420 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1422 /* Sha1.. */
1423 SHA1_Init(&c);
1424 SHA1_Update(&c, hdr, *hdrlen);
1425 SHA1_Update(&c, buf, len);
1426 SHA1_Final(sha1, &c);
1427 }
1429 /*
1430 * Link the tempfile to the final place, possibly creating the
1431 * last directory level as you do so.
1432 *
1433 * Returns the errno on failure, 0 on success.
1434 */
1435 static int link_temp_to_file(const char *tmpfile, const char *filename)
1436 {
1437 int ret;
1438 char *dir;
1440 if (!link(tmpfile, filename))
1441 return 0;
1443 /*
1444 * Try to mkdir the last path component if that failed.
1445 *
1446 * Re-try the "link()" regardless of whether the mkdir
1447 * succeeds, since a race might mean that somebody
1448 * else succeeded.
1449 */
1450 ret = errno;
1451 dir = strrchr(filename, '/');
1452 if (dir) {
1453 *dir = 0;
1454 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1455 *dir = '/';
1456 return -2;
1457 }
1458 *dir = '/';
1459 if (!link(tmpfile, filename))
1460 return 0;
1461 ret = errno;
1462 }
1463 return ret;
1464 }
1466 /*
1467 * Move the just written object into its final resting place
1468 */
1469 int move_temp_to_file(const char *tmpfile, const char *filename)
1470 {
1471 int ret = link_temp_to_file(tmpfile, filename);
1473 /*
1474 * Coda hack - coda doesn't like cross-directory links,
1475 * so we fall back to a rename, which will mean that it
1476 * won't be able to check collisions, but that's not a
1477 * big deal.
1478 *
1479 * The same holds for FAT formatted media.
1480 *
1481 * When this succeeds, we just return 0. We have nothing
1482 * left to unlink.
1483 */
1484 if (ret && ret != EEXIST) {
1485 if (!rename(tmpfile, filename))
1486 return 0;
1487 ret = errno;
1488 }
1489 unlink(tmpfile);
1490 if (ret) {
1491 if (ret != EEXIST) {
1492 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1493 }
1494 /* FIXME!!! Collision check here ? */
1495 }
1497 return 0;
1498 }
1500 static int write_buffer(int fd, const void *buf, size_t len)
1501 {
1502 while (len) {
1503 ssize_t size;
1505 size = write(fd, buf, len);
1506 if (!size)
1507 return error("file write: disk full");
1508 if (size < 0) {
1509 if (errno == EINTR || errno == EAGAIN)
1510 continue;
1511 return error("file write error (%s)", strerror(errno));
1512 }
1513 len -= size;
1514 buf = (char *) buf + size;
1515 }
1516 return 0;
1517 }
1519 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1520 {
1521 int hdr_len;
1522 unsigned char c;
1524 c = (type << 4) | (len & 15);
1525 len >>= 4;
1526 hdr_len = 1;
1527 while (len) {
1528 *hdr++ = c | 0x80;
1529 hdr_len++;
1530 c = (len & 0x7f);
1531 len >>= 7;
1532 }
1533 *hdr = c;
1534 return hdr_len;
1535 }
1537 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1538 {
1539 int obj_type, hdr;
1541 if (use_legacy_headers) {
1542 while (deflate(stream, 0) == Z_OK)
1543 /* nothing */;
1544 return;
1545 }
1546 if (!strcmp(type, blob_type))
1547 obj_type = OBJ_BLOB;
1548 else if (!strcmp(type, tree_type))
1549 obj_type = OBJ_TREE;
1550 else if (!strcmp(type, commit_type))
1551 obj_type = OBJ_COMMIT;
1552 else if (!strcmp(type, tag_type))
1553 obj_type = OBJ_TAG;
1554 else
1555 die("trying to generate bogus object of type '%s'", type);
1556 hdr = write_binary_header(stream->next_out, obj_type, len);
1557 stream->total_out = hdr;
1558 stream->next_out += hdr;
1559 stream->avail_out -= hdr;
1560 }
1562 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1563 unsigned char *sha1)
1564 {
1565 unsigned char hdr[50];
1566 int hdrlen;
1567 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1568 return 0;
1569 }
1571 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1572 {
1573 int size;
1574 unsigned char *compressed;
1575 z_stream stream;
1576 unsigned char sha1[20];
1577 char *filename;
1578 static char tmpfile[PATH_MAX];
1579 unsigned char hdr[50];
1580 int fd, hdrlen;
1582 /* Normally if we have it in the pack then we do not bother writing
1583 * it out into .git/objects/??/?{38} file.
1584 */
1585 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1586 filename = sha1_file_name(sha1);
1587 if (returnsha1)
1588 hashcpy(returnsha1, sha1);
1589 if (has_sha1_file(sha1))
1590 return 0;
1591 fd = open(filename, O_RDONLY);
1592 if (fd >= 0) {
1593 /*
1594 * FIXME!!! We might do collision checking here, but we'd
1595 * need to uncompress the old file and check it. Later.
1596 */
1597 close(fd);
1598 return 0;
1599 }
1601 if (errno != ENOENT) {
1602 return error("sha1 file %s: %s\n", filename, strerror(errno));
1603 }
1605 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1607 fd = mkstemp(tmpfile);
1608 if (fd < 0) {
1609 if (errno == EPERM)
1610 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1611 else
1612 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1613 }
1615 /* Set it up */
1616 memset(&stream, 0, sizeof(stream));
1617 deflateInit(&stream, zlib_compression_level);
1618 size = 8 + deflateBound(&stream, len+hdrlen);
1619 compressed = xmalloc(size);
1621 /* Compress it */
1622 stream.next_out = compressed;
1623 stream.avail_out = size;
1625 /* First header.. */
1626 stream.next_in = hdr;
1627 stream.avail_in = hdrlen;
1628 setup_object_header(&stream, type, len);
1630 /* Then the data itself.. */
1631 stream.next_in = buf;
1632 stream.avail_in = len;
1633 while (deflate(&stream, Z_FINISH) == Z_OK)
1634 /* nothing */;
1635 deflateEnd(&stream);
1636 size = stream.total_out;
1638 if (write_buffer(fd, compressed, size) < 0)
1639 die("unable to write sha1 file");
1640 fchmod(fd, 0444);
1641 close(fd);
1642 free(compressed);
1644 return move_temp_to_file(tmpfile, filename);
1645 }
1647 /*
1648 * We need to unpack and recompress the object for writing
1649 * it out to a different file.
1650 */
1651 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1652 {
1653 size_t size;
1654 z_stream stream;
1655 unsigned char *unpacked;
1656 unsigned long len;
1657 char type[20];
1658 char hdr[50];
1659 int hdrlen;
1660 void *buf;
1662 /* need to unpack and recompress it by itself */
1663 unpacked = read_packed_sha1(sha1, type, &len);
1665 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1667 /* Set it up */
1668 memset(&stream, 0, sizeof(stream));
1669 deflateInit(&stream, zlib_compression_level);
1670 size = deflateBound(&stream, len + hdrlen);
1671 buf = xmalloc(size);
1673 /* Compress it */
1674 stream.next_out = buf;
1675 stream.avail_out = size;
1677 /* First header.. */
1678 stream.next_in = (void *)hdr;
1679 stream.avail_in = hdrlen;
1680 while (deflate(&stream, 0) == Z_OK)
1681 /* nothing */;
1683 /* Then the data itself.. */
1684 stream.next_in = unpacked;
1685 stream.avail_in = len;
1686 while (deflate(&stream, Z_FINISH) == Z_OK)
1687 /* nothing */;
1688 deflateEnd(&stream);
1689 free(unpacked);
1691 *objsize = stream.total_out;
1692 return buf;
1693 }
1695 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1696 {
1697 int retval;
1698 unsigned long objsize;
1699 void *buf = map_sha1_file(sha1, &objsize);
1701 if (buf) {
1702 retval = write_buffer(fd, buf, objsize);
1703 munmap(buf, objsize);
1704 return retval;
1705 }
1707 buf = repack_object(sha1, &objsize);
1708 retval = write_buffer(fd, buf, objsize);
1709 free(buf);
1710 return retval;
1711 }
1713 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1714 size_t bufsize, size_t *bufposn)
1715 {
1716 char tmpfile[PATH_MAX];
1717 int local;
1718 z_stream stream;
1719 unsigned char real_sha1[20];
1720 unsigned char discard[4096];
1721 int ret;
1722 SHA_CTX c;
1724 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1726 local = mkstemp(tmpfile);
1727 if (local < 0) {
1728 if (errno == EPERM)
1729 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1730 else
1731 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1732 }
1734 memset(&stream, 0, sizeof(stream));
1736 inflateInit(&stream);
1738 SHA1_Init(&c);
1740 do {
1741 ssize_t size;
1742 if (*bufposn) {
1743 stream.avail_in = *bufposn;
1744 stream.next_in = (unsigned char *) buffer;
1745 do {
1746 stream.next_out = discard;
1747 stream.avail_out = sizeof(discard);
1748 ret = inflate(&stream, Z_SYNC_FLUSH);
1749 SHA1_Update(&c, discard, sizeof(discard) -
1750 stream.avail_out);
1751 } while (stream.avail_in && ret == Z_OK);
1752 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1753 die("unable to write sha1 file");
1754 memmove(buffer, buffer + *bufposn - stream.avail_in,
1755 stream.avail_in);
1756 *bufposn = stream.avail_in;
1757 if (ret != Z_OK)
1758 break;
1759 }
1760 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1761 if (size <= 0) {
1762 close(local);
1763 unlink(tmpfile);
1764 if (!size)
1765 return error("Connection closed?");
1766 perror("Reading from connection");
1767 return -1;
1768 }
1769 *bufposn += size;
1770 } while (1);
1771 inflateEnd(&stream);
1773 close(local);
1774 SHA1_Final(real_sha1, &c);
1775 if (ret != Z_STREAM_END) {
1776 unlink(tmpfile);
1777 return error("File %s corrupted", sha1_to_hex(sha1));
1778 }
1779 if (hashcmp(sha1, real_sha1)) {
1780 unlink(tmpfile);
1781 return error("File %s has bad hash", sha1_to_hex(sha1));
1782 }
1784 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1785 }
1787 int has_pack_index(const unsigned char *sha1)
1788 {
1789 struct stat st;
1790 if (stat(sha1_pack_index_name(sha1), &st))
1791 return 0;
1792 return 1;
1793 }
1795 int has_pack_file(const unsigned char *sha1)
1796 {
1797 struct stat st;
1798 if (stat(sha1_pack_name(sha1), &st))
1799 return 0;
1800 return 1;
1801 }
1803 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1804 {
1805 struct pack_entry e;
1806 return find_pack_entry(sha1, &e, ignore_packed);
1807 }
1809 int has_sha1_file(const unsigned char *sha1)
1810 {
1811 struct stat st;
1812 struct pack_entry e;
1814 if (find_pack_entry(sha1, &e, NULL))
1815 return 1;
1816 return find_sha1_file(sha1, &st) ? 1 : 0;
1817 }
1819 /*
1820 * reads from fd as long as possible into a supplied buffer of size bytes.
1821 * If necessary the buffer's size is increased using realloc()
1822 *
1823 * returns 0 if anything went fine and -1 otherwise
1824 *
1825 * NOTE: both buf and size may change, but even when -1 is returned
1826 * you still have to free() it yourself.
1827 */
1828 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1829 {
1830 char* buf = *return_buf;
1831 unsigned long size = *return_size;
1832 int iret;
1833 unsigned long off = 0;
1835 do {
1836 iret = xread(fd, buf + off, size - off);
1837 if (iret > 0) {
1838 off += iret;
1839 if (off == size) {
1840 size *= 2;
1841 buf = xrealloc(buf, size);
1842 }
1843 }
1844 } while (iret > 0);
1846 *return_buf = buf;
1847 *return_size = off;
1849 if (iret < 0)
1850 return -1;
1851 return 0;
1852 }
1854 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1855 {
1856 unsigned long size = 4096;
1857 char *buf = xmalloc(size);
1858 int ret;
1860 if (read_pipe(fd, &buf, &size)) {
1861 free(buf);
1862 return -1;
1863 }
1865 if (!type)
1866 type = blob_type;
1867 if (write_object)
1868 ret = write_sha1_file(buf, size, type, sha1);
1869 else
1870 ret = hash_sha1_file(buf, size, type, sha1);
1871 free(buf);
1872 return ret;
1873 }
1875 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1876 {
1877 unsigned long size = st->st_size;
1878 void *buf;
1879 int ret;
1881 buf = "";
1882 if (size)
1883 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1884 close(fd);
1885 if (buf == MAP_FAILED)
1886 return -1;
1888 if (!type)
1889 type = blob_type;
1890 if (write_object)
1891 ret = write_sha1_file(buf, size, type, sha1);
1892 else
1893 ret = hash_sha1_file(buf, size, type, sha1);
1894 if (size)
1895 munmap(buf, size);
1896 return ret;
1897 }
1899 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1900 {
1901 int fd;
1902 char *target;
1904 switch (st->st_mode & S_IFMT) {
1905 case S_IFREG:
1906 fd = open(path, O_RDONLY);
1907 if (fd < 0)
1908 return error("open(\"%s\"): %s", path,
1909 strerror(errno));
1910 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1911 return error("%s: failed to insert into database",
1912 path);
1913 break;
1914 case S_IFLNK:
1915 target = xmalloc(st->st_size+1);
1916 if (readlink(path, target, st->st_size+1) != st->st_size) {
1917 char *errstr = strerror(errno);
1918 free(target);
1919 return error("readlink(\"%s\"): %s", path,
1920 errstr);
1921 }
1922 if (!write_object)
1923 hash_sha1_file(target, st->st_size, blob_type, sha1);
1924 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1925 return error("%s: failed to insert into database",
1926 path);
1927 free(target);
1928 break;
1929 default:
1930 return error("%s: unsupported file type", path);
1931 }
1932 return 0;
1933 }