]> git.ipfire.org Git - thirdparty/git.git/blob - sha1_file.c
Refactor packed_git to prepare for sliding mmap windows.
[thirdparty/git.git] / sha1_file.c
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"
16
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
24
25 const unsigned char null_sha1[20];
26
27 static unsigned int sha1_file_open_flag = O_NOATIME;
28
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 };
63
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 }
76
77 int safe_create_leading_directories(char *path)
78 {
79 char *pos = path;
80 struct stat st;
81
82 if (*pos == '/')
83 pos++;
84
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 }
109
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;
117
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';
124
125 return buffer;
126 }
127
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 }
139
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;
153
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 }
167
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;
173
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 }
181
182 buf = name;
183
184 for (i = 0; i < 20; i++) {
185 unsigned int val = *sha1++;
186 *buf++ = hex[val >> 4];
187 *buf++ = hex[val & 0xf];
188 }
189
190 return base;
191 }
192
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;
198
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 }
206
207 buf = name;
208
209 for (i = 0; i < 20; i++) {
210 unsigned int val = *sha1++;
211 *buf++ = hex[val >> 4];
212 *buf++ = hex[val & 0xf];
213 }
214
215 return base;
216 }
217
218 struct alternate_object_database *alt_odb_list;
219 static struct alternate_object_database **alt_odb_tail;
220
221 static void read_info_alternates(const char * alternates, int depth);
222
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;
248
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);
257
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);
265
266 ent->name = ent->base + pfxlen + 1;
267 ent->base[pfxlen + 3] = '/';
268 ent->base[pfxlen] = ent->base[entlen-1] = 0;
269
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 }
278
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 }
292
293 /* add the alternate entry */
294 *alt_odb_tail = ent;
295 alt_odb_tail = &(ent->next);
296 ent->next = NULL;
297
298 /* recursively add alternates */
299 read_info_alternates(ent->base, depth + 1);
300
301 ent->base[pfxlen] = '/';
302
303 return 0;
304 }
305
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;
310
311 if (depth > 5) {
312 error("%s: ignoring alternate object stores, nesting too deep.",
313 relative_base);
314 return;
315 }
316
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 }
342
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;
349
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;
362
363 link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
364
365 munmap(map, st.st_size);
366 }
367
368 void prepare_alt_odb(void)
369 {
370 const char *alt;
371
372 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
373 if (!alt) alt = "";
374
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);
379
380 read_info_alternates(get_object_directory(), 0);
381 }
382
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;
387
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 }
399
400 static int pack_used_ctr;
401 static unsigned long pack_mapped;
402 struct packed_git *packed_git;
403
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;
424
425 index = idx_map;
426 *idx_map_ = idx_map;
427 *idx_size_ = idx_size;
428
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 }
439
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");
449
450 return 0;
451 }
452
453 static int unuse_one_packed_git(void)
454 {
455 struct packed_git *p, *lru = NULL;
456
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 return 1;
469 }
470
471 void unuse_packed_git(struct packed_git *p)
472 {
473 if (p->windows)
474 p->windows->inuse_cnt--;
475 }
476
477 int use_packed_git(struct packed_git *p)
478 {
479 if (!p->pack_size) {
480 struct stat st;
481 /* We created the struct before we had the pack */
482 stat(p->pack_name, &st);
483 if (!S_ISREG(st.st_mode))
484 die("packfile %s not a regular file", p->pack_name);
485 p->pack_size = st.st_size;
486 }
487 if (!p->windows) {
488 int fd;
489 struct stat st;
490 struct pack_window *win;
491 struct pack_header *hdr;
492
493 pack_mapped += p->pack_size;
494 while (packed_git_limit < pack_mapped && unuse_one_packed_git())
495 ; /* nothing */
496 fd = open(p->pack_name, O_RDONLY);
497 if (fd < 0)
498 die("packfile %s cannot be opened", p->pack_name);
499 if (fstat(fd, &st)) {
500 close(fd);
501 die("packfile %s cannot be opened", p->pack_name);
502 }
503 if (st.st_size != p->pack_size)
504 die("packfile %s size mismatch.", p->pack_name);
505 win = xcalloc(1, sizeof(*win));
506 win->len = p->pack_size;
507 win->base = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
508 close(fd);
509 if (win->base == MAP_FAILED)
510 die("packfile %s cannot be mapped.", p->pack_name);
511 p->windows = win;
512
513 /* Check if we understand this pack file. If we don't we're
514 * likely too old to handle it.
515 */
516 hdr = (struct pack_header*)win->base;
517 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
518 die("packfile %s isn't actually a pack.", p->pack_name);
519 if (!pack_version_ok(hdr->hdr_version))
520 die("packfile %s is version %i and not supported"
521 " (try upgrading GIT to a newer version)",
522 p->pack_name, ntohl(hdr->hdr_version));
523
524 /* Check if the pack file matches with the index file.
525 * this is cheap.
526 */
527 if (hashcmp((unsigned char *)(p->index_base) +
528 p->index_size - 40,
529 p->windows->base +
530 p->pack_size - 20)) {
531 die("packfile %s does not match index.", p->pack_name);
532 }
533 }
534 p->windows->last_used = pack_used_ctr++;
535 p->windows->inuse_cnt++;
536 return 0;
537 }
538
539 struct packed_git *add_packed_git(char *path, int path_len, int local)
540 {
541 struct stat st;
542 struct packed_git *p;
543 unsigned long idx_size;
544 void *idx_map;
545 unsigned char sha1[20];
546
547 if (check_packed_git_idx(path, &idx_size, &idx_map))
548 return NULL;
549
550 /* do we have a corresponding .pack file? */
551 strcpy(path + path_len - 4, ".pack");
552 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
553 munmap(idx_map, idx_size);
554 return NULL;
555 }
556 /* ok, it looks sane as far as we can check without
557 * actually mapping the pack file.
558 */
559 p = xmalloc(sizeof(*p) + path_len + 2);
560 strcpy(p->pack_name, path);
561 p->index_size = idx_size;
562 p->pack_size = st.st_size;
563 p->index_base = idx_map;
564 p->next = NULL;
565 p->windows = NULL;
566 p->pack_local = local;
567 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
568 hashcpy(p->sha1, sha1);
569 return p;
570 }
571
572 struct packed_git *parse_pack_index(unsigned char *sha1)
573 {
574 char *path = sha1_pack_index_name(sha1);
575 return parse_pack_index_file(sha1, path);
576 }
577
578 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
579 {
580 struct packed_git *p;
581 unsigned long idx_size;
582 void *idx_map;
583 char *path;
584
585 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
586 return NULL;
587
588 path = sha1_pack_name(sha1);
589
590 p = xmalloc(sizeof(*p) + strlen(path) + 2);
591 strcpy(p->pack_name, path);
592 p->index_size = idx_size;
593 p->pack_size = 0;
594 p->index_base = idx_map;
595 p->next = NULL;
596 p->windows = NULL;
597 hashcpy(p->sha1, sha1);
598 return p;
599 }
600
601 void install_packed_git(struct packed_git *pack)
602 {
603 pack->next = packed_git;
604 packed_git = pack;
605 }
606
607 static void prepare_packed_git_one(char *objdir, int local)
608 {
609 char path[PATH_MAX];
610 int len;
611 DIR *dir;
612 struct dirent *de;
613
614 sprintf(path, "%s/pack", objdir);
615 len = strlen(path);
616 dir = opendir(path);
617 if (!dir) {
618 if (errno != ENOENT)
619 error("unable to open object pack directory: %s: %s",
620 path, strerror(errno));
621 return;
622 }
623 path[len++] = '/';
624 while ((de = readdir(dir)) != NULL) {
625 int namelen = strlen(de->d_name);
626 struct packed_git *p;
627
628 if (!has_extension(de->d_name, ".idx"))
629 continue;
630
631 /* we have .idx. Is it a file we can map? */
632 strcpy(path + len, de->d_name);
633 for (p = packed_git; p; p = p->next) {
634 if (!memcmp(path, p->pack_name, len + namelen - 4))
635 break;
636 }
637 if (p)
638 continue;
639 p = add_packed_git(path, len + namelen, local);
640 if (!p)
641 continue;
642 p->next = packed_git;
643 packed_git = p;
644 }
645 closedir(dir);
646 }
647
648 static int prepare_packed_git_run_once = 0;
649 void prepare_packed_git(void)
650 {
651 struct alternate_object_database *alt;
652
653 if (prepare_packed_git_run_once)
654 return;
655 prepare_packed_git_one(get_object_directory(), 1);
656 prepare_alt_odb();
657 for (alt = alt_odb_list; alt; alt = alt->next) {
658 alt->name[-1] = 0;
659 prepare_packed_git_one(alt->base, 0);
660 alt->name[-1] = '/';
661 }
662 prepare_packed_git_run_once = 1;
663 }
664
665 void reprepare_packed_git(void)
666 {
667 prepare_packed_git_run_once = 0;
668 prepare_packed_git();
669 }
670
671 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
672 {
673 unsigned char real_sha1[20];
674 hash_sha1_file(map, size, type, real_sha1);
675 return hashcmp(sha1, real_sha1) ? -1 : 0;
676 }
677
678 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
679 {
680 struct stat st;
681 void *map;
682 int fd;
683 char *filename = find_sha1_file(sha1, &st);
684
685 if (!filename) {
686 return NULL;
687 }
688
689 fd = open(filename, O_RDONLY | sha1_file_open_flag);
690 if (fd < 0) {
691 /* See if it works without O_NOATIME */
692 switch (sha1_file_open_flag) {
693 default:
694 fd = open(filename, O_RDONLY);
695 if (fd >= 0)
696 break;
697 /* Fallthrough */
698 case 0:
699 return NULL;
700 }
701
702 /* If it failed once, it will probably fail again.
703 * Stop using O_NOATIME
704 */
705 sha1_file_open_flag = 0;
706 }
707 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
708 close(fd);
709 if (map == MAP_FAILED)
710 return NULL;
711 *size = st.st_size;
712 return map;
713 }
714
715 int legacy_loose_object(unsigned char *map)
716 {
717 unsigned int word;
718
719 /*
720 * Is it a zlib-compressed buffer? If so, the first byte
721 * must be 0x78 (15-bit window size, deflated), and the
722 * first 16-bit word is evenly divisible by 31
723 */
724 word = (map[0] << 8) + map[1];
725 if (map[0] == 0x78 && !(word % 31))
726 return 1;
727 else
728 return 0;
729 }
730
731 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
732 {
733 unsigned shift;
734 unsigned char c;
735 unsigned long size;
736 unsigned long used = 0;
737
738 c = buf[used++];
739 *type = (c >> 4) & 7;
740 size = c & 15;
741 shift = 4;
742 while (c & 0x80) {
743 if (len <= used)
744 return 0;
745 if (sizeof(long) * 8 <= shift)
746 return 0;
747 c = buf[used++];
748 size += (c & 0x7f) << shift;
749 shift += 7;
750 }
751 *sizep = size;
752 return used;
753 }
754
755 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
756 {
757 unsigned long size, used;
758 static const char valid_loose_object_type[8] = {
759 0, /* OBJ_EXT */
760 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
761 0, /* "delta" and others are invalid in a loose object */
762 };
763 enum object_type type;
764
765 /* Get the data stream */
766 memset(stream, 0, sizeof(*stream));
767 stream->next_in = map;
768 stream->avail_in = mapsize;
769 stream->next_out = buffer;
770 stream->avail_out = bufsiz;
771
772 if (legacy_loose_object(map)) {
773 inflateInit(stream);
774 return inflate(stream, 0);
775 }
776
777 used = unpack_object_header_gently(map, mapsize, &type, &size);
778 if (!used || !valid_loose_object_type[type])
779 return -1;
780 map += used;
781 mapsize -= used;
782
783 /* Set up the stream for the rest.. */
784 stream->next_in = map;
785 stream->avail_in = mapsize;
786 inflateInit(stream);
787
788 /* And generate the fake traditional header */
789 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
790 type_names[type], size);
791 return 0;
792 }
793
794 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
795 {
796 int bytes = strlen(buffer) + 1;
797 unsigned char *buf = xmalloc(1+size);
798 unsigned long n;
799
800 n = stream->total_out - bytes;
801 if (n > size)
802 n = size;
803 memcpy(buf, (char *) buffer + bytes, n);
804 bytes = n;
805 if (bytes < size) {
806 stream->next_out = buf + bytes;
807 stream->avail_out = size - bytes;
808 while (inflate(stream, Z_FINISH) == Z_OK)
809 /* nothing */;
810 }
811 buf[size] = 0;
812 inflateEnd(stream);
813 return buf;
814 }
815
816 /*
817 * We used to just use "sscanf()", but that's actually way
818 * too permissive for what we want to check. So do an anal
819 * object header parse by hand.
820 */
821 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
822 {
823 int i;
824 unsigned long size;
825
826 /*
827 * The type can be at most ten bytes (including the
828 * terminating '\0' that we add), and is followed by
829 * a space.
830 */
831 i = 10;
832 for (;;) {
833 char c = *hdr++;
834 if (c == ' ')
835 break;
836 if (!--i)
837 return -1;
838 *type++ = c;
839 }
840 *type = 0;
841
842 /*
843 * The length must follow immediately, and be in canonical
844 * decimal format (ie "010" is not valid).
845 */
846 size = *hdr++ - '0';
847 if (size > 9)
848 return -1;
849 if (size) {
850 for (;;) {
851 unsigned long c = *hdr - '0';
852 if (c > 9)
853 break;
854 hdr++;
855 size = size * 10 + c;
856 }
857 }
858 *sizep = size;
859
860 /*
861 * The length must be followed by a zero byte
862 */
863 return *hdr ? -1 : 0;
864 }
865
866 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
867 {
868 int ret;
869 z_stream stream;
870 char hdr[8192];
871
872 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
873 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
874 return NULL;
875
876 return unpack_sha1_rest(&stream, hdr, *size);
877 }
878
879 static unsigned long get_delta_base(struct packed_git *p,
880 unsigned long offset,
881 enum object_type kind,
882 unsigned long delta_obj_offset,
883 unsigned long *base_obj_offset)
884 {
885 unsigned char *base_info = p->windows->base + offset;
886 unsigned long base_offset;
887
888 /* there must be at least 20 bytes left regardless of delta type */
889 if (p->pack_size <= offset + 20)
890 die("truncated pack file");
891
892 if (kind == OBJ_OFS_DELTA) {
893 unsigned used = 0;
894 unsigned char c = base_info[used++];
895 base_offset = c & 127;
896 while (c & 128) {
897 base_offset += 1;
898 if (!base_offset || base_offset & ~(~0UL >> 7))
899 die("offset value overflow for delta base object");
900 c = base_info[used++];
901 base_offset = (base_offset << 7) + (c & 127);
902 }
903 base_offset = delta_obj_offset - base_offset;
904 if (base_offset >= delta_obj_offset)
905 die("delta base offset out of bound");
906 offset += used;
907 } else if (kind == OBJ_REF_DELTA) {
908 /* The base entry _must_ be in the same pack */
909 base_offset = find_pack_entry_one(base_info, p);
910 if (!base_offset)
911 die("failed to find delta-pack base object %s",
912 sha1_to_hex(base_info));
913 offset += 20;
914 } else
915 die("I am totally screwed");
916 *base_obj_offset = base_offset;
917 return offset;
918 }
919
920 /* forward declaration for a mutually recursive function */
921 static int packed_object_info(struct packed_git *p, unsigned long offset,
922 char *type, unsigned long *sizep);
923
924 static int packed_delta_info(struct packed_git *p,
925 unsigned long offset,
926 enum object_type kind,
927 unsigned long obj_offset,
928 char *type,
929 unsigned long *sizep)
930 {
931 unsigned long base_offset;
932
933 offset = get_delta_base(p, offset, kind, obj_offset, &base_offset);
934
935 /* We choose to only get the type of the base object and
936 * ignore potentially corrupt pack file that expects the delta
937 * based on a base with a wrong size. This saves tons of
938 * inflate() calls.
939 */
940 if (packed_object_info(p, base_offset, type, NULL))
941 die("cannot get info for delta-pack base");
942
943 if (sizep) {
944 const unsigned char *data;
945 unsigned char delta_head[20];
946 unsigned long result_size;
947 z_stream stream;
948 int st;
949
950 memset(&stream, 0, sizeof(stream));
951
952 stream.next_in = p->windows->base + offset;
953 stream.avail_in = p->pack_size - offset;
954 stream.next_out = delta_head;
955 stream.avail_out = sizeof(delta_head);
956
957 inflateInit(&stream);
958 st = inflate(&stream, Z_FINISH);
959 inflateEnd(&stream);
960 if ((st != Z_STREAM_END) &&
961 stream.total_out != sizeof(delta_head))
962 die("delta data unpack-initial failed");
963
964 /* Examine the initial part of the delta to figure out
965 * the result size.
966 */
967 data = delta_head;
968
969 /* ignore base size */
970 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
971
972 /* Read the result size */
973 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
974 *sizep = result_size;
975 }
976 return 0;
977 }
978
979 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
980 enum object_type *type, unsigned long *sizep)
981 {
982 unsigned long used;
983
984 if (p->pack_size <= offset)
985 die("object offset outside of pack file");
986
987 used = unpack_object_header_gently(p->windows->base + offset,
988 p->pack_size - offset, type, sizep);
989 if (!used)
990 die("object offset outside of pack file");
991
992 return offset + used;
993 }
994
995 void packed_object_info_detail(struct packed_git *p,
996 unsigned long offset,
997 char *type,
998 unsigned long *size,
999 unsigned long *store_size,
1000 unsigned int *delta_chain_length,
1001 unsigned char *base_sha1)
1002 {
1003 unsigned long obj_offset, val;
1004 unsigned char *next_sha1;
1005 enum object_type kind;
1006
1007 *delta_chain_length = 0;
1008 obj_offset = offset;
1009 offset = unpack_object_header(p, offset, &kind, size);
1010
1011 for (;;) {
1012 switch (kind) {
1013 default:
1014 die("pack %s contains unknown object type %d",
1015 p->pack_name, kind);
1016 case OBJ_COMMIT:
1017 case OBJ_TREE:
1018 case OBJ_BLOB:
1019 case OBJ_TAG:
1020 strcpy(type, type_names[kind]);
1021 *store_size = 0; /* notyet */
1022 return;
1023 case OBJ_OFS_DELTA:
1024 get_delta_base(p, offset, kind, obj_offset, &offset);
1025 if (*delta_chain_length == 0) {
1026 /* TODO: find base_sha1 as pointed by offset */
1027 }
1028 break;
1029 case OBJ_REF_DELTA:
1030 if (p->pack_size <= offset + 20)
1031 die("pack file %s records an incomplete delta base",
1032 p->pack_name);
1033 next_sha1 = p->windows->base + offset;
1034 if (*delta_chain_length == 0)
1035 hashcpy(base_sha1, next_sha1);
1036 offset = find_pack_entry_one(next_sha1, p);
1037 break;
1038 }
1039 obj_offset = offset;
1040 offset = unpack_object_header(p, offset, &kind, &val);
1041 (*delta_chain_length)++;
1042 }
1043 }
1044
1045 static int packed_object_info(struct packed_git *p, unsigned long offset,
1046 char *type, unsigned long *sizep)
1047 {
1048 unsigned long size, obj_offset = offset;
1049 enum object_type kind;
1050
1051 offset = unpack_object_header(p, offset, &kind, &size);
1052
1053 switch (kind) {
1054 case OBJ_OFS_DELTA:
1055 case OBJ_REF_DELTA:
1056 return packed_delta_info(p, offset, kind, obj_offset, type, sizep);
1057 case OBJ_COMMIT:
1058 case OBJ_TREE:
1059 case OBJ_BLOB:
1060 case OBJ_TAG:
1061 strcpy(type, type_names[kind]);
1062 break;
1063 default:
1064 die("pack %s contains unknown object type %d",
1065 p->pack_name, kind);
1066 }
1067 if (sizep)
1068 *sizep = size;
1069 return 0;
1070 }
1071
1072 static void *unpack_compressed_entry(struct packed_git *p,
1073 unsigned long offset,
1074 unsigned long size)
1075 {
1076 int st;
1077 z_stream stream;
1078 unsigned char *buffer;
1079
1080 buffer = xmalloc(size + 1);
1081 buffer[size] = 0;
1082 memset(&stream, 0, sizeof(stream));
1083 stream.next_in = p->windows->base + offset;
1084 stream.avail_in = p->pack_size - offset;
1085 stream.next_out = buffer;
1086 stream.avail_out = size;
1087
1088 inflateInit(&stream);
1089 st = inflate(&stream, Z_FINISH);
1090 inflateEnd(&stream);
1091 if ((st != Z_STREAM_END) || stream.total_out != size) {
1092 free(buffer);
1093 return NULL;
1094 }
1095
1096 return buffer;
1097 }
1098
1099 static void *unpack_delta_entry(struct packed_git *p,
1100 unsigned long offset,
1101 unsigned long delta_size,
1102 enum object_type kind,
1103 unsigned long obj_offset,
1104 char *type,
1105 unsigned long *sizep)
1106 {
1107 void *delta_data, *result, *base;
1108 unsigned long result_size, base_size, base_offset;
1109
1110 offset = get_delta_base(p, offset, kind, obj_offset, &base_offset);
1111 base = unpack_entry(p, base_offset, type, &base_size);
1112 if (!base)
1113 die("failed to read delta base object at %lu from %s",
1114 base_offset, p->pack_name);
1115
1116 delta_data = unpack_compressed_entry(p, offset, delta_size);
1117 result = patch_delta(base, base_size,
1118 delta_data, delta_size,
1119 &result_size);
1120 if (!result)
1121 die("failed to apply delta");
1122 free(delta_data);
1123 free(base);
1124 *sizep = result_size;
1125 return result;
1126 }
1127
1128 void *unpack_entry(struct packed_git *p, unsigned long offset,
1129 char *type, unsigned long *sizep)
1130 {
1131 unsigned long size, obj_offset = offset;
1132 enum object_type kind;
1133 void *retval;
1134
1135 if (use_packed_git(p))
1136 die("cannot map packed file");
1137 offset = unpack_object_header(p, offset, &kind, &size);
1138 switch (kind) {
1139 case OBJ_OFS_DELTA:
1140 case OBJ_REF_DELTA:
1141 retval = unpack_delta_entry(p, offset, size, kind, obj_offset, type, sizep);
1142 break;
1143 case OBJ_COMMIT:
1144 case OBJ_TREE:
1145 case OBJ_BLOB:
1146 case OBJ_TAG:
1147 strcpy(type, type_names[kind]);
1148 *sizep = size;
1149 retval = unpack_compressed_entry(p, offset, size);
1150 break;
1151 default:
1152 die("unknown object type %i in %s", kind, p->pack_name);
1153 }
1154 unuse_packed_git(p);
1155 return retval;
1156 }
1157
1158 int num_packed_objects(const struct packed_git *p)
1159 {
1160 /* See check_packed_git_idx() */
1161 return (p->index_size - 20 - 20 - 4*256) / 24;
1162 }
1163
1164 int nth_packed_object_sha1(const struct packed_git *p, int n,
1165 unsigned char* sha1)
1166 {
1167 void *index = p->index_base + 256;
1168 if (n < 0 || num_packed_objects(p) <= n)
1169 return -1;
1170 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1171 return 0;
1172 }
1173
1174 unsigned long find_pack_entry_one(const unsigned char *sha1,
1175 struct packed_git *p)
1176 {
1177 unsigned int *level1_ofs = p->index_base;
1178 int hi = ntohl(level1_ofs[*sha1]);
1179 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1180 void *index = p->index_base + 256;
1181
1182 do {
1183 int mi = (lo + hi) / 2;
1184 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1185 if (!cmp)
1186 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1187 if (cmp > 0)
1188 hi = mi;
1189 else
1190 lo = mi+1;
1191 } while (lo < hi);
1192 return 0;
1193 }
1194
1195 static int matches_pack_name(struct packed_git *p, const char *ig)
1196 {
1197 const char *last_c, *c;
1198
1199 if (!strcmp(p->pack_name, ig))
1200 return 0;
1201
1202 for (c = p->pack_name, last_c = c; *c;)
1203 if (*c == '/')
1204 last_c = ++c;
1205 else
1206 ++c;
1207 if (!strcmp(last_c, ig))
1208 return 0;
1209
1210 return 1;
1211 }
1212
1213 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1214 {
1215 struct packed_git *p;
1216 unsigned long offset;
1217
1218 prepare_packed_git();
1219
1220 for (p = packed_git; p; p = p->next) {
1221 if (ignore_packed) {
1222 const char **ig;
1223 for (ig = ignore_packed; *ig; ig++)
1224 if (!matches_pack_name(p, *ig))
1225 break;
1226 if (*ig)
1227 continue;
1228 }
1229 offset = find_pack_entry_one(sha1, p);
1230 if (offset) {
1231 e->offset = offset;
1232 e->p = p;
1233 hashcpy(e->sha1, sha1);
1234 return 1;
1235 }
1236 }
1237 return 0;
1238 }
1239
1240 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1241 struct packed_git *packs)
1242 {
1243 struct packed_git *p;
1244
1245 for (p = packs; p; p = p->next) {
1246 if (find_pack_entry_one(sha1, p))
1247 return p;
1248 }
1249 return NULL;
1250
1251 }
1252
1253 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1254 {
1255 int status;
1256 unsigned long mapsize, size;
1257 void *map;
1258 z_stream stream;
1259 char hdr[128];
1260
1261 map = map_sha1_file(sha1, &mapsize);
1262 if (!map)
1263 return error("unable to find %s", sha1_to_hex(sha1));
1264 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1265 status = error("unable to unpack %s header",
1266 sha1_to_hex(sha1));
1267 if (parse_sha1_header(hdr, type, &size) < 0)
1268 status = error("unable to parse %s header", sha1_to_hex(sha1));
1269 else {
1270 status = 0;
1271 if (sizep)
1272 *sizep = size;
1273 }
1274 inflateEnd(&stream);
1275 munmap(map, mapsize);
1276 return status;
1277 }
1278
1279 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1280 {
1281 int status;
1282 struct pack_entry e;
1283
1284 if (!find_pack_entry(sha1, &e, NULL)) {
1285 reprepare_packed_git();
1286 if (!find_pack_entry(sha1, &e, NULL))
1287 return sha1_loose_object_info(sha1, type, sizep);
1288 }
1289 if (use_packed_git(e.p))
1290 die("cannot map packed file");
1291 status = packed_object_info(e.p, e.offset, type, sizep);
1292 unuse_packed_git(e.p);
1293 return status;
1294 }
1295
1296 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1297 {
1298 struct pack_entry e;
1299
1300 if (!find_pack_entry(sha1, &e, NULL)) {
1301 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1302 return NULL;
1303 }
1304 return unpack_entry(e.p, e.offset, type, size);
1305 }
1306
1307 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1308 {
1309 unsigned long mapsize;
1310 void *map, *buf;
1311 struct pack_entry e;
1312
1313 if (find_pack_entry(sha1, &e, NULL))
1314 return read_packed_sha1(sha1, type, size);
1315 map = map_sha1_file(sha1, &mapsize);
1316 if (map) {
1317 buf = unpack_sha1_file(map, mapsize, type, size);
1318 munmap(map, mapsize);
1319 return buf;
1320 }
1321 reprepare_packed_git();
1322 if (find_pack_entry(sha1, &e, NULL))
1323 return read_packed_sha1(sha1, type, size);
1324 return NULL;
1325 }
1326
1327 void *read_object_with_reference(const unsigned char *sha1,
1328 const char *required_type,
1329 unsigned long *size,
1330 unsigned char *actual_sha1_return)
1331 {
1332 char type[20];
1333 void *buffer;
1334 unsigned long isize;
1335 unsigned char actual_sha1[20];
1336
1337 hashcpy(actual_sha1, sha1);
1338 while (1) {
1339 int ref_length = -1;
1340 const char *ref_type = NULL;
1341
1342 buffer = read_sha1_file(actual_sha1, type, &isize);
1343 if (!buffer)
1344 return NULL;
1345 if (!strcmp(type, required_type)) {
1346 *size = isize;
1347 if (actual_sha1_return)
1348 hashcpy(actual_sha1_return, actual_sha1);
1349 return buffer;
1350 }
1351 /* Handle references */
1352 else if (!strcmp(type, commit_type))
1353 ref_type = "tree ";
1354 else if (!strcmp(type, tag_type))
1355 ref_type = "object ";
1356 else {
1357 free(buffer);
1358 return NULL;
1359 }
1360 ref_length = strlen(ref_type);
1361
1362 if (memcmp(buffer, ref_type, ref_length) ||
1363 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1364 free(buffer);
1365 return NULL;
1366 }
1367 free(buffer);
1368 /* Now we have the ID of the referred-to object in
1369 * actual_sha1. Check again. */
1370 }
1371 }
1372
1373 static void write_sha1_file_prepare(void *buf, unsigned long len,
1374 const char *type, unsigned char *sha1,
1375 unsigned char *hdr, int *hdrlen)
1376 {
1377 SHA_CTX c;
1378
1379 /* Generate the header */
1380 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1381
1382 /* Sha1.. */
1383 SHA1_Init(&c);
1384 SHA1_Update(&c, hdr, *hdrlen);
1385 SHA1_Update(&c, buf, len);
1386 SHA1_Final(sha1, &c);
1387 }
1388
1389 /*
1390 * Link the tempfile to the final place, possibly creating the
1391 * last directory level as you do so.
1392 *
1393 * Returns the errno on failure, 0 on success.
1394 */
1395 static int link_temp_to_file(const char *tmpfile, const char *filename)
1396 {
1397 int ret;
1398 char *dir;
1399
1400 if (!link(tmpfile, filename))
1401 return 0;
1402
1403 /*
1404 * Try to mkdir the last path component if that failed.
1405 *
1406 * Re-try the "link()" regardless of whether the mkdir
1407 * succeeds, since a race might mean that somebody
1408 * else succeeded.
1409 */
1410 ret = errno;
1411 dir = strrchr(filename, '/');
1412 if (dir) {
1413 *dir = 0;
1414 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1415 *dir = '/';
1416 return -2;
1417 }
1418 *dir = '/';
1419 if (!link(tmpfile, filename))
1420 return 0;
1421 ret = errno;
1422 }
1423 return ret;
1424 }
1425
1426 /*
1427 * Move the just written object into its final resting place
1428 */
1429 int move_temp_to_file(const char *tmpfile, const char *filename)
1430 {
1431 int ret = link_temp_to_file(tmpfile, filename);
1432
1433 /*
1434 * Coda hack - coda doesn't like cross-directory links,
1435 * so we fall back to a rename, which will mean that it
1436 * won't be able to check collisions, but that's not a
1437 * big deal.
1438 *
1439 * The same holds for FAT formatted media.
1440 *
1441 * When this succeeds, we just return 0. We have nothing
1442 * left to unlink.
1443 */
1444 if (ret && ret != EEXIST) {
1445 if (!rename(tmpfile, filename))
1446 return 0;
1447 ret = errno;
1448 }
1449 unlink(tmpfile);
1450 if (ret) {
1451 if (ret != EEXIST) {
1452 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1453 }
1454 /* FIXME!!! Collision check here ? */
1455 }
1456
1457 return 0;
1458 }
1459
1460 static int write_buffer(int fd, const void *buf, size_t len)
1461 {
1462 while (len) {
1463 ssize_t size;
1464
1465 size = write(fd, buf, len);
1466 if (!size)
1467 return error("file write: disk full");
1468 if (size < 0) {
1469 if (errno == EINTR || errno == EAGAIN)
1470 continue;
1471 return error("file write error (%s)", strerror(errno));
1472 }
1473 len -= size;
1474 buf = (char *) buf + size;
1475 }
1476 return 0;
1477 }
1478
1479 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1480 {
1481 int hdr_len;
1482 unsigned char c;
1483
1484 c = (type << 4) | (len & 15);
1485 len >>= 4;
1486 hdr_len = 1;
1487 while (len) {
1488 *hdr++ = c | 0x80;
1489 hdr_len++;
1490 c = (len & 0x7f);
1491 len >>= 7;
1492 }
1493 *hdr = c;
1494 return hdr_len;
1495 }
1496
1497 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1498 {
1499 int obj_type, hdr;
1500
1501 if (use_legacy_headers) {
1502 while (deflate(stream, 0) == Z_OK)
1503 /* nothing */;
1504 return;
1505 }
1506 if (!strcmp(type, blob_type))
1507 obj_type = OBJ_BLOB;
1508 else if (!strcmp(type, tree_type))
1509 obj_type = OBJ_TREE;
1510 else if (!strcmp(type, commit_type))
1511 obj_type = OBJ_COMMIT;
1512 else if (!strcmp(type, tag_type))
1513 obj_type = OBJ_TAG;
1514 else
1515 die("trying to generate bogus object of type '%s'", type);
1516 hdr = write_binary_header(stream->next_out, obj_type, len);
1517 stream->total_out = hdr;
1518 stream->next_out += hdr;
1519 stream->avail_out -= hdr;
1520 }
1521
1522 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1523 unsigned char *sha1)
1524 {
1525 unsigned char hdr[50];
1526 int hdrlen;
1527 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1528 return 0;
1529 }
1530
1531 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1532 {
1533 int size;
1534 unsigned char *compressed;
1535 z_stream stream;
1536 unsigned char sha1[20];
1537 char *filename;
1538 static char tmpfile[PATH_MAX];
1539 unsigned char hdr[50];
1540 int fd, hdrlen;
1541
1542 /* Normally if we have it in the pack then we do not bother writing
1543 * it out into .git/objects/??/?{38} file.
1544 */
1545 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1546 filename = sha1_file_name(sha1);
1547 if (returnsha1)
1548 hashcpy(returnsha1, sha1);
1549 if (has_sha1_file(sha1))
1550 return 0;
1551 fd = open(filename, O_RDONLY);
1552 if (fd >= 0) {
1553 /*
1554 * FIXME!!! We might do collision checking here, but we'd
1555 * need to uncompress the old file and check it. Later.
1556 */
1557 close(fd);
1558 return 0;
1559 }
1560
1561 if (errno != ENOENT) {
1562 return error("sha1 file %s: %s\n", filename, strerror(errno));
1563 }
1564
1565 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1566
1567 fd = mkstemp(tmpfile);
1568 if (fd < 0) {
1569 if (errno == EPERM)
1570 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1571 else
1572 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1573 }
1574
1575 /* Set it up */
1576 memset(&stream, 0, sizeof(stream));
1577 deflateInit(&stream, zlib_compression_level);
1578 size = 8 + deflateBound(&stream, len+hdrlen);
1579 compressed = xmalloc(size);
1580
1581 /* Compress it */
1582 stream.next_out = compressed;
1583 stream.avail_out = size;
1584
1585 /* First header.. */
1586 stream.next_in = hdr;
1587 stream.avail_in = hdrlen;
1588 setup_object_header(&stream, type, len);
1589
1590 /* Then the data itself.. */
1591 stream.next_in = buf;
1592 stream.avail_in = len;
1593 while (deflate(&stream, Z_FINISH) == Z_OK)
1594 /* nothing */;
1595 deflateEnd(&stream);
1596 size = stream.total_out;
1597
1598 if (write_buffer(fd, compressed, size) < 0)
1599 die("unable to write sha1 file");
1600 fchmod(fd, 0444);
1601 close(fd);
1602 free(compressed);
1603
1604 return move_temp_to_file(tmpfile, filename);
1605 }
1606
1607 /*
1608 * We need to unpack and recompress the object for writing
1609 * it out to a different file.
1610 */
1611 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1612 {
1613 size_t size;
1614 z_stream stream;
1615 unsigned char *unpacked;
1616 unsigned long len;
1617 char type[20];
1618 char hdr[50];
1619 int hdrlen;
1620 void *buf;
1621
1622 /* need to unpack and recompress it by itself */
1623 unpacked = read_packed_sha1(sha1, type, &len);
1624
1625 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1626
1627 /* Set it up */
1628 memset(&stream, 0, sizeof(stream));
1629 deflateInit(&stream, zlib_compression_level);
1630 size = deflateBound(&stream, len + hdrlen);
1631 buf = xmalloc(size);
1632
1633 /* Compress it */
1634 stream.next_out = buf;
1635 stream.avail_out = size;
1636
1637 /* First header.. */
1638 stream.next_in = (void *)hdr;
1639 stream.avail_in = hdrlen;
1640 while (deflate(&stream, 0) == Z_OK)
1641 /* nothing */;
1642
1643 /* Then the data itself.. */
1644 stream.next_in = unpacked;
1645 stream.avail_in = len;
1646 while (deflate(&stream, Z_FINISH) == Z_OK)
1647 /* nothing */;
1648 deflateEnd(&stream);
1649 free(unpacked);
1650
1651 *objsize = stream.total_out;
1652 return buf;
1653 }
1654
1655 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1656 {
1657 int retval;
1658 unsigned long objsize;
1659 void *buf = map_sha1_file(sha1, &objsize);
1660
1661 if (buf) {
1662 retval = write_buffer(fd, buf, objsize);
1663 munmap(buf, objsize);
1664 return retval;
1665 }
1666
1667 buf = repack_object(sha1, &objsize);
1668 retval = write_buffer(fd, buf, objsize);
1669 free(buf);
1670 return retval;
1671 }
1672
1673 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1674 size_t bufsize, size_t *bufposn)
1675 {
1676 char tmpfile[PATH_MAX];
1677 int local;
1678 z_stream stream;
1679 unsigned char real_sha1[20];
1680 unsigned char discard[4096];
1681 int ret;
1682 SHA_CTX c;
1683
1684 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1685
1686 local = mkstemp(tmpfile);
1687 if (local < 0) {
1688 if (errno == EPERM)
1689 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1690 else
1691 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1692 }
1693
1694 memset(&stream, 0, sizeof(stream));
1695
1696 inflateInit(&stream);
1697
1698 SHA1_Init(&c);
1699
1700 do {
1701 ssize_t size;
1702 if (*bufposn) {
1703 stream.avail_in = *bufposn;
1704 stream.next_in = (unsigned char *) buffer;
1705 do {
1706 stream.next_out = discard;
1707 stream.avail_out = sizeof(discard);
1708 ret = inflate(&stream, Z_SYNC_FLUSH);
1709 SHA1_Update(&c, discard, sizeof(discard) -
1710 stream.avail_out);
1711 } while (stream.avail_in && ret == Z_OK);
1712 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1713 die("unable to write sha1 file");
1714 memmove(buffer, buffer + *bufposn - stream.avail_in,
1715 stream.avail_in);
1716 *bufposn = stream.avail_in;
1717 if (ret != Z_OK)
1718 break;
1719 }
1720 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1721 if (size <= 0) {
1722 close(local);
1723 unlink(tmpfile);
1724 if (!size)
1725 return error("Connection closed?");
1726 perror("Reading from connection");
1727 return -1;
1728 }
1729 *bufposn += size;
1730 } while (1);
1731 inflateEnd(&stream);
1732
1733 close(local);
1734 SHA1_Final(real_sha1, &c);
1735 if (ret != Z_STREAM_END) {
1736 unlink(tmpfile);
1737 return error("File %s corrupted", sha1_to_hex(sha1));
1738 }
1739 if (hashcmp(sha1, real_sha1)) {
1740 unlink(tmpfile);
1741 return error("File %s has bad hash", sha1_to_hex(sha1));
1742 }
1743
1744 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1745 }
1746
1747 int has_pack_index(const unsigned char *sha1)
1748 {
1749 struct stat st;
1750 if (stat(sha1_pack_index_name(sha1), &st))
1751 return 0;
1752 return 1;
1753 }
1754
1755 int has_pack_file(const unsigned char *sha1)
1756 {
1757 struct stat st;
1758 if (stat(sha1_pack_name(sha1), &st))
1759 return 0;
1760 return 1;
1761 }
1762
1763 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1764 {
1765 struct pack_entry e;
1766 return find_pack_entry(sha1, &e, ignore_packed);
1767 }
1768
1769 int has_sha1_file(const unsigned char *sha1)
1770 {
1771 struct stat st;
1772 struct pack_entry e;
1773
1774 if (find_pack_entry(sha1, &e, NULL))
1775 return 1;
1776 return find_sha1_file(sha1, &st) ? 1 : 0;
1777 }
1778
1779 /*
1780 * reads from fd as long as possible into a supplied buffer of size bytes.
1781 * If necessary the buffer's size is increased using realloc()
1782 *
1783 * returns 0 if anything went fine and -1 otherwise
1784 *
1785 * NOTE: both buf and size may change, but even when -1 is returned
1786 * you still have to free() it yourself.
1787 */
1788 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1789 {
1790 char* buf = *return_buf;
1791 unsigned long size = *return_size;
1792 int iret;
1793 unsigned long off = 0;
1794
1795 do {
1796 iret = xread(fd, buf + off, size - off);
1797 if (iret > 0) {
1798 off += iret;
1799 if (off == size) {
1800 size *= 2;
1801 buf = xrealloc(buf, size);
1802 }
1803 }
1804 } while (iret > 0);
1805
1806 *return_buf = buf;
1807 *return_size = off;
1808
1809 if (iret < 0)
1810 return -1;
1811 return 0;
1812 }
1813
1814 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1815 {
1816 unsigned long size = 4096;
1817 char *buf = xmalloc(size);
1818 int ret;
1819
1820 if (read_pipe(fd, &buf, &size)) {
1821 free(buf);
1822 return -1;
1823 }
1824
1825 if (!type)
1826 type = blob_type;
1827 if (write_object)
1828 ret = write_sha1_file(buf, size, type, sha1);
1829 else
1830 ret = hash_sha1_file(buf, size, type, sha1);
1831 free(buf);
1832 return ret;
1833 }
1834
1835 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1836 {
1837 unsigned long size = st->st_size;
1838 void *buf;
1839 int ret;
1840
1841 buf = "";
1842 if (size)
1843 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1844 close(fd);
1845 if (buf == MAP_FAILED)
1846 return -1;
1847
1848 if (!type)
1849 type = blob_type;
1850 if (write_object)
1851 ret = write_sha1_file(buf, size, type, sha1);
1852 else
1853 ret = hash_sha1_file(buf, size, type, sha1);
1854 if (size)
1855 munmap(buf, size);
1856 return ret;
1857 }
1858
1859 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1860 {
1861 int fd;
1862 char *target;
1863
1864 switch (st->st_mode & S_IFMT) {
1865 case S_IFREG:
1866 fd = open(path, O_RDONLY);
1867 if (fd < 0)
1868 return error("open(\"%s\"): %s", path,
1869 strerror(errno));
1870 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1871 return error("%s: failed to insert into database",
1872 path);
1873 break;
1874 case S_IFLNK:
1875 target = xmalloc(st->st_size+1);
1876 if (readlink(path, target, st->st_size+1) != st->st_size) {
1877 char *errstr = strerror(errno);
1878 free(target);
1879 return error("readlink(\"%s\"): %s", path,
1880 errstr);
1881 }
1882 if (!write_object)
1883 hash_sha1_file(target, st->st_size, blob_type, sha1);
1884 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1885 return error("%s: failed to insert into database",
1886 path);
1887 free(target);
1888 break;
1889 default:
1890 return error("%s: unsupported file type", path);
1891 }
1892 return 0;
1893 }