Refactor how we open pack files to prepare for multiple windows.
[git/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 close(p->pack_fd);
469 p->pack_fd = -1;
470 return 1;
471 }
472
473 void unuse_packed_git(struct packed_git *p)
474 {
475 if (p->windows)
476 p->windows->inuse_cnt--;
477 }
478
479 static void open_packed_git(struct packed_git *p)
480 {
481 struct stat st;
482 struct pack_header hdr;
483 unsigned char sha1[20];
484 unsigned char *idx_sha1;
485
486 p->pack_fd = open(p->pack_name, O_RDONLY);
487 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
488 die("packfile %s cannot be opened", p->pack_name);
489
490 /* If we created the struct before we had the pack we lack size. */
491 if (!p->pack_size) {
492 if (!S_ISREG(st.st_mode))
493 die("packfile %s not a regular file", p->pack_name);
494 p->pack_size = st.st_size;
495 } else if (p->pack_size != st.st_size)
496 die("packfile %s size changed", p->pack_name);
497
498 /* Verify we recognize this pack file format. */
499 read_or_die(p->pack_fd, &hdr, sizeof(hdr));
500 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
501 die("file %s is not a GIT packfile", p->pack_name);
502 if (!pack_version_ok(hdr.hdr_version))
503 die("packfile %s is version %u and not supported"
504 " (try upgrading GIT to a newer version)",
505 p->pack_name, ntohl(hdr.hdr_version));
506
507 /* Verify the pack matches its index. */
508 if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
509 die("packfile %s claims to have %u objects"
510 " while index size indicates %u objects",
511 p->pack_name, ntohl(hdr.hdr_entries),
512 num_packed_objects(p));
513 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
514 die("end of packfile %s is unavailable", p->pack_name);
515 read_or_die(p->pack_fd, sha1, sizeof(sha1));
516 idx_sha1 = ((unsigned char *)p->index_base) + p->index_size - 40;
517 if (hashcmp(sha1, idx_sha1))
518 die("packfile %s does not match index", p->pack_name);
519 }
520
521 int use_packed_git(struct packed_git *p)
522 {
523 if (p->pack_fd == -1)
524 open_packed_git(p);
525 if (!p->windows) {
526 struct pack_window *win;
527
528 pack_mapped += p->pack_size;
529 while (packed_git_limit < pack_mapped && unuse_one_packed_git())
530 ; /* nothing */
531 win = xcalloc(1, sizeof(*win));
532 win->len = p->pack_size;
533 win->base = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, p->pack_fd, 0);
534 if (win->base == MAP_FAILED)
535 die("packfile %s cannot be mapped.", p->pack_name);
536 p->windows = win;
537 }
538 p->windows->last_used = pack_used_ctr++;
539 p->windows->inuse_cnt++;
540 return 0;
541 }
542
543 struct packed_git *add_packed_git(char *path, int path_len, int local)
544 {
545 struct stat st;
546 struct packed_git *p;
547 unsigned long idx_size;
548 void *idx_map;
549 unsigned char sha1[20];
550
551 if (check_packed_git_idx(path, &idx_size, &idx_map))
552 return NULL;
553
554 /* do we have a corresponding .pack file? */
555 strcpy(path + path_len - 4, ".pack");
556 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
557 munmap(idx_map, idx_size);
558 return NULL;
559 }
560 /* ok, it looks sane as far as we can check without
561 * actually mapping the pack file.
562 */
563 p = xmalloc(sizeof(*p) + path_len + 2);
564 strcpy(p->pack_name, path);
565 p->index_size = idx_size;
566 p->pack_size = st.st_size;
567 p->index_base = idx_map;
568 p->next = NULL;
569 p->windows = NULL;
570 p->pack_fd = -1;
571 p->pack_local = local;
572 if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
573 hashcpy(p->sha1, sha1);
574 return p;
575 }
576
577 struct packed_git *parse_pack_index(unsigned char *sha1)
578 {
579 char *path = sha1_pack_index_name(sha1);
580 return parse_pack_index_file(sha1, path);
581 }
582
583 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
584 {
585 struct packed_git *p;
586 unsigned long idx_size;
587 void *idx_map;
588 char *path;
589
590 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
591 return NULL;
592
593 path = sha1_pack_name(sha1);
594
595 p = xmalloc(sizeof(*p) + strlen(path) + 2);
596 strcpy(p->pack_name, path);
597 p->index_size = idx_size;
598 p->pack_size = 0;
599 p->index_base = idx_map;
600 p->next = NULL;
601 p->windows = NULL;
602 p->pack_fd = -1;
603 hashcpy(p->sha1, sha1);
604 return p;
605 }
606
607 void install_packed_git(struct packed_git *pack)
608 {
609 pack->next = packed_git;
610 packed_git = pack;
611 }
612
613 static void prepare_packed_git_one(char *objdir, int local)
614 {
615 char path[PATH_MAX];
616 int len;
617 DIR *dir;
618 struct dirent *de;
619
620 sprintf(path, "%s/pack", objdir);
621 len = strlen(path);
622 dir = opendir(path);
623 if (!dir) {
624 if (errno != ENOENT)
625 error("unable to open object pack directory: %s: %s",
626 path, strerror(errno));
627 return;
628 }
629 path[len++] = '/';
630 while ((de = readdir(dir)) != NULL) {
631 int namelen = strlen(de->d_name);
632 struct packed_git *p;
633
634 if (!has_extension(de->d_name, ".idx"))
635 continue;
636
637 /* we have .idx. Is it a file we can map? */
638 strcpy(path + len, de->d_name);
639 for (p = packed_git; p; p = p->next) {
640 if (!memcmp(path, p->pack_name, len + namelen - 4))
641 break;
642 }
643 if (p)
644 continue;
645 p = add_packed_git(path, len + namelen, local);
646 if (!p)
647 continue;
648 p->next = packed_git;
649 packed_git = p;
650 }
651 closedir(dir);
652 }
653
654 static int prepare_packed_git_run_once = 0;
655 void prepare_packed_git(void)
656 {
657 struct alternate_object_database *alt;
658
659 if (prepare_packed_git_run_once)
660 return;
661 prepare_packed_git_one(get_object_directory(), 1);
662 prepare_alt_odb();
663 for (alt = alt_odb_list; alt; alt = alt->next) {
664 alt->name[-1] = 0;
665 prepare_packed_git_one(alt->base, 0);
666 alt->name[-1] = '/';
667 }
668 prepare_packed_git_run_once = 1;
669 }
670
671 void reprepare_packed_git(void)
672 {
673 prepare_packed_git_run_once = 0;
674 prepare_packed_git();
675 }
676
677 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
678 {
679 unsigned char real_sha1[20];
680 hash_sha1_file(map, size, type, real_sha1);
681 return hashcmp(sha1, real_sha1) ? -1 : 0;
682 }
683
684 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
685 {
686 struct stat st;
687 void *map;
688 int fd;
689 char *filename = find_sha1_file(sha1, &st);
690
691 if (!filename) {
692 return NULL;
693 }
694
695 fd = open(filename, O_RDONLY | sha1_file_open_flag);
696 if (fd < 0) {
697 /* See if it works without O_NOATIME */
698 switch (sha1_file_open_flag) {
699 default:
700 fd = open(filename, O_RDONLY);
701 if (fd >= 0)
702 break;
703 /* Fallthrough */
704 case 0:
705 return NULL;
706 }
707
708 /* If it failed once, it will probably fail again.
709 * Stop using O_NOATIME
710 */
711 sha1_file_open_flag = 0;
712 }
713 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
714 close(fd);
715 if (map == MAP_FAILED)
716 return NULL;
717 *size = st.st_size;
718 return map;
719 }
720
721 int legacy_loose_object(unsigned char *map)
722 {
723 unsigned int word;
724
725 /*
726 * Is it a zlib-compressed buffer? If so, the first byte
727 * must be 0x78 (15-bit window size, deflated), and the
728 * first 16-bit word is evenly divisible by 31
729 */
730 word = (map[0] << 8) + map[1];
731 if (map[0] == 0x78 && !(word % 31))
732 return 1;
733 else
734 return 0;
735 }
736
737 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
738 {
739 unsigned shift;
740 unsigned char c;
741 unsigned long size;
742 unsigned long used = 0;
743
744 c = buf[used++];
745 *type = (c >> 4) & 7;
746 size = c & 15;
747 shift = 4;
748 while (c & 0x80) {
749 if (len <= used)
750 return 0;
751 if (sizeof(long) * 8 <= shift)
752 return 0;
753 c = buf[used++];
754 size += (c & 0x7f) << shift;
755 shift += 7;
756 }
757 *sizep = size;
758 return used;
759 }
760
761 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
762 {
763 unsigned long size, used;
764 static const char valid_loose_object_type[8] = {
765 0, /* OBJ_EXT */
766 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
767 0, /* "delta" and others are invalid in a loose object */
768 };
769 enum object_type type;
770
771 /* Get the data stream */
772 memset(stream, 0, sizeof(*stream));
773 stream->next_in = map;
774 stream->avail_in = mapsize;
775 stream->next_out = buffer;
776 stream->avail_out = bufsiz;
777
778 if (legacy_loose_object(map)) {
779 inflateInit(stream);
780 return inflate(stream, 0);
781 }
782
783 used = unpack_object_header_gently(map, mapsize, &type, &size);
784 if (!used || !valid_loose_object_type[type])
785 return -1;
786 map += used;
787 mapsize -= used;
788
789 /* Set up the stream for the rest.. */
790 stream->next_in = map;
791 stream->avail_in = mapsize;
792 inflateInit(stream);
793
794 /* And generate the fake traditional header */
795 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
796 type_names[type], size);
797 return 0;
798 }
799
800 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
801 {
802 int bytes = strlen(buffer) + 1;
803 unsigned char *buf = xmalloc(1+size);
804 unsigned long n;
805
806 n = stream->total_out - bytes;
807 if (n > size)
808 n = size;
809 memcpy(buf, (char *) buffer + bytes, n);
810 bytes = n;
811 if (bytes < size) {
812 stream->next_out = buf + bytes;
813 stream->avail_out = size - bytes;
814 while (inflate(stream, Z_FINISH) == Z_OK)
815 /* nothing */;
816 }
817 buf[size] = 0;
818 inflateEnd(stream);
819 return buf;
820 }
821
822 /*
823 * We used to just use "sscanf()", but that's actually way
824 * too permissive for what we want to check. So do an anal
825 * object header parse by hand.
826 */
827 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
828 {
829 int i;
830 unsigned long size;
831
832 /*
833 * The type can be at most ten bytes (including the
834 * terminating '\0' that we add), and is followed by
835 * a space.
836 */
837 i = 10;
838 for (;;) {
839 char c = *hdr++;
840 if (c == ' ')
841 break;
842 if (!--i)
843 return -1;
844 *type++ = c;
845 }
846 *type = 0;
847
848 /*
849 * The length must follow immediately, and be in canonical
850 * decimal format (ie "010" is not valid).
851 */
852 size = *hdr++ - '0';
853 if (size > 9)
854 return -1;
855 if (size) {
856 for (;;) {
857 unsigned long c = *hdr - '0';
858 if (c > 9)
859 break;
860 hdr++;
861 size = size * 10 + c;
862 }
863 }
864 *sizep = size;
865
866 /*
867 * The length must be followed by a zero byte
868 */
869 return *hdr ? -1 : 0;
870 }
871
872 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
873 {
874 int ret;
875 z_stream stream;
876 char hdr[8192];
877
878 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
879 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
880 return NULL;
881
882 return unpack_sha1_rest(&stream, hdr, *size);
883 }
884
885 static unsigned long get_delta_base(struct packed_git *p,
886 unsigned long offset,
887 enum object_type kind,
888 unsigned long delta_obj_offset,
889 unsigned long *base_obj_offset)
890 {
891 unsigned char *base_info = p->windows->base + offset;
892 unsigned long base_offset;
893
894 /* there must be at least 20 bytes left regardless of delta type */
895 if (p->pack_size <= offset + 20)
896 die("truncated pack file");
897
898 if (kind == OBJ_OFS_DELTA) {
899 unsigned used = 0;
900 unsigned char c = base_info[used++];
901 base_offset = c & 127;
902 while (c & 128) {
903 base_offset += 1;
904 if (!base_offset || base_offset & ~(~0UL >> 7))
905 die("offset value overflow for delta base object");
906 c = base_info[used++];
907 base_offset = (base_offset << 7) + (c & 127);
908 }
909 base_offset = delta_obj_offset - base_offset;
910 if (base_offset >= delta_obj_offset)
911 die("delta base offset out of bound");
912 offset += used;
913 } else if (kind == OBJ_REF_DELTA) {
914 /* The base entry _must_ be in the same pack */
915 base_offset = find_pack_entry_one(base_info, p);
916 if (!base_offset)
917 die("failed to find delta-pack base object %s",
918 sha1_to_hex(base_info));
919 offset += 20;
920 } else
921 die("I am totally screwed");
922 *base_obj_offset = base_offset;
923 return offset;
924 }
925
926 /* forward declaration for a mutually recursive function */
927 static int packed_object_info(struct packed_git *p, unsigned long offset,
928 char *type, unsigned long *sizep);
929
930 static int packed_delta_info(struct packed_git *p,
931 unsigned long offset,
932 enum object_type kind,
933 unsigned long obj_offset,
934 char *type,
935 unsigned long *sizep)
936 {
937 unsigned long base_offset;
938
939 offset = get_delta_base(p, offset, kind, obj_offset, &base_offset);
940
941 /* We choose to only get the type of the base object and
942 * ignore potentially corrupt pack file that expects the delta
943 * based on a base with a wrong size. This saves tons of
944 * inflate() calls.
945 */
946 if (packed_object_info(p, base_offset, type, NULL))
947 die("cannot get info for delta-pack base");
948
949 if (sizep) {
950 const unsigned char *data;
951 unsigned char delta_head[20];
952 unsigned long result_size;
953 z_stream stream;
954 int st;
955
956 memset(&stream, 0, sizeof(stream));
957
958 stream.next_in = p->windows->base + offset;
959 stream.avail_in = p->pack_size - offset;
960 stream.next_out = delta_head;
961 stream.avail_out = sizeof(delta_head);
962
963 inflateInit(&stream);
964 st = inflate(&stream, Z_FINISH);
965 inflateEnd(&stream);
966 if ((st != Z_STREAM_END) &&
967 stream.total_out != sizeof(delta_head))
968 die("delta data unpack-initial failed");
969
970 /* Examine the initial part of the delta to figure out
971 * the result size.
972 */
973 data = delta_head;
974
975 /* ignore base size */
976 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
977
978 /* Read the result size */
979 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
980 *sizep = result_size;
981 }
982 return 0;
983 }
984
985 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
986 enum object_type *type, unsigned long *sizep)
987 {
988 unsigned long used;
989
990 if (p->pack_size <= offset)
991 die("object offset outside of pack file");
992
993 used = unpack_object_header_gently(p->windows->base + offset,
994 p->pack_size - offset, type, sizep);
995 if (!used)
996 die("object offset outside of pack file");
997
998 return offset + used;
999 }
1000
1001 void packed_object_info_detail(struct packed_git *p,
1002 unsigned long offset,
1003 char *type,
1004 unsigned long *size,
1005 unsigned long *store_size,
1006 unsigned int *delta_chain_length,
1007 unsigned char *base_sha1)
1008 {
1009 unsigned long obj_offset, val;
1010 unsigned char *next_sha1;
1011 enum object_type kind;
1012
1013 *delta_chain_length = 0;
1014 obj_offset = offset;
1015 offset = unpack_object_header(p, offset, &kind, size);
1016
1017 for (;;) {
1018 switch (kind) {
1019 default:
1020 die("pack %s contains unknown object type %d",
1021 p->pack_name, kind);
1022 case OBJ_COMMIT:
1023 case OBJ_TREE:
1024 case OBJ_BLOB:
1025 case OBJ_TAG:
1026 strcpy(type, type_names[kind]);
1027 *store_size = 0; /* notyet */
1028 return;
1029 case OBJ_OFS_DELTA:
1030 get_delta_base(p, offset, kind, obj_offset, &offset);
1031 if (*delta_chain_length == 0) {
1032 /* TODO: find base_sha1 as pointed by offset */
1033 }
1034 break;
1035 case OBJ_REF_DELTA:
1036 if (p->pack_size <= offset + 20)
1037 die("pack file %s records an incomplete delta base",
1038 p->pack_name);
1039 next_sha1 = p->windows->base + offset;
1040 if (*delta_chain_length == 0)
1041 hashcpy(base_sha1, next_sha1);
1042 offset = find_pack_entry_one(next_sha1, p);
1043 break;
1044 }
1045 obj_offset = offset;
1046 offset = unpack_object_header(p, offset, &kind, &val);
1047 (*delta_chain_length)++;
1048 }
1049 }
1050
1051 static int packed_object_info(struct packed_git *p, unsigned long offset,
1052 char *type, unsigned long *sizep)
1053 {
1054 unsigned long size, obj_offset = offset;
1055 enum object_type kind;
1056
1057 offset = unpack_object_header(p, offset, &kind, &size);
1058
1059 switch (kind) {
1060 case OBJ_OFS_DELTA:
1061 case OBJ_REF_DELTA:
1062 return packed_delta_info(p, offset, kind, obj_offset, type, sizep);
1063 case OBJ_COMMIT:
1064 case OBJ_TREE:
1065 case OBJ_BLOB:
1066 case OBJ_TAG:
1067 strcpy(type, type_names[kind]);
1068 break;
1069 default:
1070 die("pack %s contains unknown object type %d",
1071 p->pack_name, kind);
1072 }
1073 if (sizep)
1074 *sizep = size;
1075 return 0;
1076 }
1077
1078 static void *unpack_compressed_entry(struct packed_git *p,
1079 unsigned long offset,
1080 unsigned long size)
1081 {
1082 int st;
1083 z_stream stream;
1084 unsigned char *buffer;
1085
1086 buffer = xmalloc(size + 1);
1087 buffer[size] = 0;
1088 memset(&stream, 0, sizeof(stream));
1089 stream.next_in = p->windows->base + offset;
1090 stream.avail_in = p->pack_size - offset;
1091 stream.next_out = buffer;
1092 stream.avail_out = size;
1093
1094 inflateInit(&stream);
1095 st = inflate(&stream, Z_FINISH);
1096 inflateEnd(&stream);
1097 if ((st != Z_STREAM_END) || stream.total_out != size) {
1098 free(buffer);
1099 return NULL;
1100 }
1101
1102 return buffer;
1103 }
1104
1105 static void *unpack_delta_entry(struct packed_git *p,
1106 unsigned long offset,
1107 unsigned long delta_size,
1108 enum object_type kind,
1109 unsigned long obj_offset,
1110 char *type,
1111 unsigned long *sizep)
1112 {
1113 void *delta_data, *result, *base;
1114 unsigned long result_size, base_size, base_offset;
1115
1116 offset = get_delta_base(p, offset, kind, obj_offset, &base_offset);
1117 base = unpack_entry(p, base_offset, type, &base_size);
1118 if (!base)
1119 die("failed to read delta base object at %lu from %s",
1120 base_offset, p->pack_name);
1121
1122 delta_data = unpack_compressed_entry(p, offset, delta_size);
1123 result = patch_delta(base, base_size,
1124 delta_data, delta_size,
1125 &result_size);
1126 if (!result)
1127 die("failed to apply delta");
1128 free(delta_data);
1129 free(base);
1130 *sizep = result_size;
1131 return result;
1132 }
1133
1134 void *unpack_entry(struct packed_git *p, unsigned long offset,
1135 char *type, unsigned long *sizep)
1136 {
1137 unsigned long size, obj_offset = offset;
1138 enum object_type kind;
1139 void *retval;
1140
1141 if (use_packed_git(p))
1142 die("cannot map packed file");
1143 offset = unpack_object_header(p, offset, &kind, &size);
1144 switch (kind) {
1145 case OBJ_OFS_DELTA:
1146 case OBJ_REF_DELTA:
1147 retval = unpack_delta_entry(p, offset, size, kind, obj_offset, type, sizep);
1148 break;
1149 case OBJ_COMMIT:
1150 case OBJ_TREE:
1151 case OBJ_BLOB:
1152 case OBJ_TAG:
1153 strcpy(type, type_names[kind]);
1154 *sizep = size;
1155 retval = unpack_compressed_entry(p, offset, size);
1156 break;
1157 default:
1158 die("unknown object type %i in %s", kind, p->pack_name);
1159 }
1160 unuse_packed_git(p);
1161 return retval;
1162 }
1163
1164 int num_packed_objects(const struct packed_git *p)
1165 {
1166 /* See check_packed_git_idx() */
1167 return (p->index_size - 20 - 20 - 4*256) / 24;
1168 }
1169
1170 int nth_packed_object_sha1(const struct packed_git *p, int n,
1171 unsigned char* sha1)
1172 {
1173 void *index = p->index_base + 256;
1174 if (n < 0 || num_packed_objects(p) <= n)
1175 return -1;
1176 hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1177 return 0;
1178 }
1179
1180 unsigned long find_pack_entry_one(const unsigned char *sha1,
1181 struct packed_git *p)
1182 {
1183 unsigned int *level1_ofs = p->index_base;
1184 int hi = ntohl(level1_ofs[*sha1]);
1185 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1186 void *index = p->index_base + 256;
1187
1188 do {
1189 int mi = (lo + hi) / 2;
1190 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1191 if (!cmp)
1192 return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1193 if (cmp > 0)
1194 hi = mi;
1195 else
1196 lo = mi+1;
1197 } while (lo < hi);
1198 return 0;
1199 }
1200
1201 static int matches_pack_name(struct packed_git *p, const char *ig)
1202 {
1203 const char *last_c, *c;
1204
1205 if (!strcmp(p->pack_name, ig))
1206 return 0;
1207
1208 for (c = p->pack_name, last_c = c; *c;)
1209 if (*c == '/')
1210 last_c = ++c;
1211 else
1212 ++c;
1213 if (!strcmp(last_c, ig))
1214 return 0;
1215
1216 return 1;
1217 }
1218
1219 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1220 {
1221 struct packed_git *p;
1222 unsigned long offset;
1223
1224 prepare_packed_git();
1225
1226 for (p = packed_git; p; p = p->next) {
1227 if (ignore_packed) {
1228 const char **ig;
1229 for (ig = ignore_packed; *ig; ig++)
1230 if (!matches_pack_name(p, *ig))
1231 break;
1232 if (*ig)
1233 continue;
1234 }
1235 offset = find_pack_entry_one(sha1, p);
1236 if (offset) {
1237 e->offset = offset;
1238 e->p = p;
1239 hashcpy(e->sha1, sha1);
1240 return 1;
1241 }
1242 }
1243 return 0;
1244 }
1245
1246 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1247 struct packed_git *packs)
1248 {
1249 struct packed_git *p;
1250
1251 for (p = packs; p; p = p->next) {
1252 if (find_pack_entry_one(sha1, p))
1253 return p;
1254 }
1255 return NULL;
1256
1257 }
1258
1259 static int sha1_loose_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1260 {
1261 int status;
1262 unsigned long mapsize, size;
1263 void *map;
1264 z_stream stream;
1265 char hdr[128];
1266
1267 map = map_sha1_file(sha1, &mapsize);
1268 if (!map)
1269 return error("unable to find %s", sha1_to_hex(sha1));
1270 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1271 status = error("unable to unpack %s header",
1272 sha1_to_hex(sha1));
1273 if (parse_sha1_header(hdr, type, &size) < 0)
1274 status = error("unable to parse %s header", sha1_to_hex(sha1));
1275 else {
1276 status = 0;
1277 if (sizep)
1278 *sizep = size;
1279 }
1280 inflateEnd(&stream);
1281 munmap(map, mapsize);
1282 return status;
1283 }
1284
1285 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1286 {
1287 int status;
1288 struct pack_entry e;
1289
1290 if (!find_pack_entry(sha1, &e, NULL)) {
1291 reprepare_packed_git();
1292 if (!find_pack_entry(sha1, &e, NULL))
1293 return sha1_loose_object_info(sha1, type, sizep);
1294 }
1295 if (use_packed_git(e.p))
1296 die("cannot map packed file");
1297 status = packed_object_info(e.p, e.offset, type, sizep);
1298 unuse_packed_git(e.p);
1299 return status;
1300 }
1301
1302 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1303 {
1304 struct pack_entry e;
1305
1306 if (!find_pack_entry(sha1, &e, NULL)) {
1307 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1308 return NULL;
1309 }
1310 return unpack_entry(e.p, e.offset, type, size);
1311 }
1312
1313 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1314 {
1315 unsigned long mapsize;
1316 void *map, *buf;
1317 struct pack_entry e;
1318
1319 if (find_pack_entry(sha1, &e, NULL))
1320 return read_packed_sha1(sha1, type, size);
1321 map = map_sha1_file(sha1, &mapsize);
1322 if (map) {
1323 buf = unpack_sha1_file(map, mapsize, type, size);
1324 munmap(map, mapsize);
1325 return buf;
1326 }
1327 reprepare_packed_git();
1328 if (find_pack_entry(sha1, &e, NULL))
1329 return read_packed_sha1(sha1, type, size);
1330 return NULL;
1331 }
1332
1333 void *read_object_with_reference(const unsigned char *sha1,
1334 const char *required_type,
1335 unsigned long *size,
1336 unsigned char *actual_sha1_return)
1337 {
1338 char type[20];
1339 void *buffer;
1340 unsigned long isize;
1341 unsigned char actual_sha1[20];
1342
1343 hashcpy(actual_sha1, sha1);
1344 while (1) {
1345 int ref_length = -1;
1346 const char *ref_type = NULL;
1347
1348 buffer = read_sha1_file(actual_sha1, type, &isize);
1349 if (!buffer)
1350 return NULL;
1351 if (!strcmp(type, required_type)) {
1352 *size = isize;
1353 if (actual_sha1_return)
1354 hashcpy(actual_sha1_return, actual_sha1);
1355 return buffer;
1356 }
1357 /* Handle references */
1358 else if (!strcmp(type, commit_type))
1359 ref_type = "tree ";
1360 else if (!strcmp(type, tag_type))
1361 ref_type = "object ";
1362 else {
1363 free(buffer);
1364 return NULL;
1365 }
1366 ref_length = strlen(ref_type);
1367
1368 if (memcmp(buffer, ref_type, ref_length) ||
1369 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1370 free(buffer);
1371 return NULL;
1372 }
1373 free(buffer);
1374 /* Now we have the ID of the referred-to object in
1375 * actual_sha1. Check again. */
1376 }
1377 }
1378
1379 static void write_sha1_file_prepare(void *buf, unsigned long len,
1380 const char *type, unsigned char *sha1,
1381 unsigned char *hdr, int *hdrlen)
1382 {
1383 SHA_CTX c;
1384
1385 /* Generate the header */
1386 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1387
1388 /* Sha1.. */
1389 SHA1_Init(&c);
1390 SHA1_Update(&c, hdr, *hdrlen);
1391 SHA1_Update(&c, buf, len);
1392 SHA1_Final(sha1, &c);
1393 }
1394
1395 /*
1396 * Link the tempfile to the final place, possibly creating the
1397 * last directory level as you do so.
1398 *
1399 * Returns the errno on failure, 0 on success.
1400 */
1401 static int link_temp_to_file(const char *tmpfile, const char *filename)
1402 {
1403 int ret;
1404 char *dir;
1405
1406 if (!link(tmpfile, filename))
1407 return 0;
1408
1409 /*
1410 * Try to mkdir the last path component if that failed.
1411 *
1412 * Re-try the "link()" regardless of whether the mkdir
1413 * succeeds, since a race might mean that somebody
1414 * else succeeded.
1415 */
1416 ret = errno;
1417 dir = strrchr(filename, '/');
1418 if (dir) {
1419 *dir = 0;
1420 if (!mkdir(filename, 0777) && adjust_shared_perm(filename)) {
1421 *dir = '/';
1422 return -2;
1423 }
1424 *dir = '/';
1425 if (!link(tmpfile, filename))
1426 return 0;
1427 ret = errno;
1428 }
1429 return ret;
1430 }
1431
1432 /*
1433 * Move the just written object into its final resting place
1434 */
1435 int move_temp_to_file(const char *tmpfile, const char *filename)
1436 {
1437 int ret = link_temp_to_file(tmpfile, filename);
1438
1439 /*
1440 * Coda hack - coda doesn't like cross-directory links,
1441 * so we fall back to a rename, which will mean that it
1442 * won't be able to check collisions, but that's not a
1443 * big deal.
1444 *
1445 * The same holds for FAT formatted media.
1446 *
1447 * When this succeeds, we just return 0. We have nothing
1448 * left to unlink.
1449 */
1450 if (ret && ret != EEXIST) {
1451 if (!rename(tmpfile, filename))
1452 return 0;
1453 ret = errno;
1454 }
1455 unlink(tmpfile);
1456 if (ret) {
1457 if (ret != EEXIST) {
1458 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1459 }
1460 /* FIXME!!! Collision check here ? */
1461 }
1462
1463 return 0;
1464 }
1465
1466 static int write_buffer(int fd, const void *buf, size_t len)
1467 {
1468 while (len) {
1469 ssize_t size;
1470
1471 size = write(fd, buf, len);
1472 if (!size)
1473 return error("file write: disk full");
1474 if (size < 0) {
1475 if (errno == EINTR || errno == EAGAIN)
1476 continue;
1477 return error("file write error (%s)", strerror(errno));
1478 }
1479 len -= size;
1480 buf = (char *) buf + size;
1481 }
1482 return 0;
1483 }
1484
1485 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1486 {
1487 int hdr_len;
1488 unsigned char c;
1489
1490 c = (type << 4) | (len & 15);
1491 len >>= 4;
1492 hdr_len = 1;
1493 while (len) {
1494 *hdr++ = c | 0x80;
1495 hdr_len++;
1496 c = (len & 0x7f);
1497 len >>= 7;
1498 }
1499 *hdr = c;
1500 return hdr_len;
1501 }
1502
1503 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1504 {
1505 int obj_type, hdr;
1506
1507 if (use_legacy_headers) {
1508 while (deflate(stream, 0) == Z_OK)
1509 /* nothing */;
1510 return;
1511 }
1512 if (!strcmp(type, blob_type))
1513 obj_type = OBJ_BLOB;
1514 else if (!strcmp(type, tree_type))
1515 obj_type = OBJ_TREE;
1516 else if (!strcmp(type, commit_type))
1517 obj_type = OBJ_COMMIT;
1518 else if (!strcmp(type, tag_type))
1519 obj_type = OBJ_TAG;
1520 else
1521 die("trying to generate bogus object of type '%s'", type);
1522 hdr = write_binary_header(stream->next_out, obj_type, len);
1523 stream->total_out = hdr;
1524 stream->next_out += hdr;
1525 stream->avail_out -= hdr;
1526 }
1527
1528 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1529 unsigned char *sha1)
1530 {
1531 unsigned char hdr[50];
1532 int hdrlen;
1533 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1534 return 0;
1535 }
1536
1537 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1538 {
1539 int size;
1540 unsigned char *compressed;
1541 z_stream stream;
1542 unsigned char sha1[20];
1543 char *filename;
1544 static char tmpfile[PATH_MAX];
1545 unsigned char hdr[50];
1546 int fd, hdrlen;
1547
1548 /* Normally if we have it in the pack then we do not bother writing
1549 * it out into .git/objects/??/?{38} file.
1550 */
1551 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1552 filename = sha1_file_name(sha1);
1553 if (returnsha1)
1554 hashcpy(returnsha1, sha1);
1555 if (has_sha1_file(sha1))
1556 return 0;
1557 fd = open(filename, O_RDONLY);
1558 if (fd >= 0) {
1559 /*
1560 * FIXME!!! We might do collision checking here, but we'd
1561 * need to uncompress the old file and check it. Later.
1562 */
1563 close(fd);
1564 return 0;
1565 }
1566
1567 if (errno != ENOENT) {
1568 return error("sha1 file %s: %s\n", filename, strerror(errno));
1569 }
1570
1571 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1572
1573 fd = mkstemp(tmpfile);
1574 if (fd < 0) {
1575 if (errno == EPERM)
1576 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1577 else
1578 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1579 }
1580
1581 /* Set it up */
1582 memset(&stream, 0, sizeof(stream));
1583 deflateInit(&stream, zlib_compression_level);
1584 size = 8 + deflateBound(&stream, len+hdrlen);
1585 compressed = xmalloc(size);
1586
1587 /* Compress it */
1588 stream.next_out = compressed;
1589 stream.avail_out = size;
1590
1591 /* First header.. */
1592 stream.next_in = hdr;
1593 stream.avail_in = hdrlen;
1594 setup_object_header(&stream, type, len);
1595
1596 /* Then the data itself.. */
1597 stream.next_in = buf;
1598 stream.avail_in = len;
1599 while (deflate(&stream, Z_FINISH) == Z_OK)
1600 /* nothing */;
1601 deflateEnd(&stream);
1602 size = stream.total_out;
1603
1604 if (write_buffer(fd, compressed, size) < 0)
1605 die("unable to write sha1 file");
1606 fchmod(fd, 0444);
1607 close(fd);
1608 free(compressed);
1609
1610 return move_temp_to_file(tmpfile, filename);
1611 }
1612
1613 /*
1614 * We need to unpack and recompress the object for writing
1615 * it out to a different file.
1616 */
1617 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1618 {
1619 size_t size;
1620 z_stream stream;
1621 unsigned char *unpacked;
1622 unsigned long len;
1623 char type[20];
1624 char hdr[50];
1625 int hdrlen;
1626 void *buf;
1627
1628 /* need to unpack and recompress it by itself */
1629 unpacked = read_packed_sha1(sha1, type, &len);
1630
1631 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1632
1633 /* Set it up */
1634 memset(&stream, 0, sizeof(stream));
1635 deflateInit(&stream, zlib_compression_level);
1636 size = deflateBound(&stream, len + hdrlen);
1637 buf = xmalloc(size);
1638
1639 /* Compress it */
1640 stream.next_out = buf;
1641 stream.avail_out = size;
1642
1643 /* First header.. */
1644 stream.next_in = (void *)hdr;
1645 stream.avail_in = hdrlen;
1646 while (deflate(&stream, 0) == Z_OK)
1647 /* nothing */;
1648
1649 /* Then the data itself.. */
1650 stream.next_in = unpacked;
1651 stream.avail_in = len;
1652 while (deflate(&stream, Z_FINISH) == Z_OK)
1653 /* nothing */;
1654 deflateEnd(&stream);
1655 free(unpacked);
1656
1657 *objsize = stream.total_out;
1658 return buf;
1659 }
1660
1661 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1662 {
1663 int retval;
1664 unsigned long objsize;
1665 void *buf = map_sha1_file(sha1, &objsize);
1666
1667 if (buf) {
1668 retval = write_buffer(fd, buf, objsize);
1669 munmap(buf, objsize);
1670 return retval;
1671 }
1672
1673 buf = repack_object(sha1, &objsize);
1674 retval = write_buffer(fd, buf, objsize);
1675 free(buf);
1676 return retval;
1677 }
1678
1679 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1680 size_t bufsize, size_t *bufposn)
1681 {
1682 char tmpfile[PATH_MAX];
1683 int local;
1684 z_stream stream;
1685 unsigned char real_sha1[20];
1686 unsigned char discard[4096];
1687 int ret;
1688 SHA_CTX c;
1689
1690 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1691
1692 local = mkstemp(tmpfile);
1693 if (local < 0) {
1694 if (errno == EPERM)
1695 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
1696 else
1697 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1698 }
1699
1700 memset(&stream, 0, sizeof(stream));
1701
1702 inflateInit(&stream);
1703
1704 SHA1_Init(&c);
1705
1706 do {
1707 ssize_t size;
1708 if (*bufposn) {
1709 stream.avail_in = *bufposn;
1710 stream.next_in = (unsigned char *) buffer;
1711 do {
1712 stream.next_out = discard;
1713 stream.avail_out = sizeof(discard);
1714 ret = inflate(&stream, Z_SYNC_FLUSH);
1715 SHA1_Update(&c, discard, sizeof(discard) -
1716 stream.avail_out);
1717 } while (stream.avail_in && ret == Z_OK);
1718 if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1719 die("unable to write sha1 file");
1720 memmove(buffer, buffer + *bufposn - stream.avail_in,
1721 stream.avail_in);
1722 *bufposn = stream.avail_in;
1723 if (ret != Z_OK)
1724 break;
1725 }
1726 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1727 if (size <= 0) {
1728 close(local);
1729 unlink(tmpfile);
1730 if (!size)
1731 return error("Connection closed?");
1732 perror("Reading from connection");
1733 return -1;
1734 }
1735 *bufposn += size;
1736 } while (1);
1737 inflateEnd(&stream);
1738
1739 close(local);
1740 SHA1_Final(real_sha1, &c);
1741 if (ret != Z_STREAM_END) {
1742 unlink(tmpfile);
1743 return error("File %s corrupted", sha1_to_hex(sha1));
1744 }
1745 if (hashcmp(sha1, real_sha1)) {
1746 unlink(tmpfile);
1747 return error("File %s has bad hash", sha1_to_hex(sha1));
1748 }
1749
1750 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1751 }
1752
1753 int has_pack_index(const unsigned char *sha1)
1754 {
1755 struct stat st;
1756 if (stat(sha1_pack_index_name(sha1), &st))
1757 return 0;
1758 return 1;
1759 }
1760
1761 int has_pack_file(const unsigned char *sha1)
1762 {
1763 struct stat st;
1764 if (stat(sha1_pack_name(sha1), &st))
1765 return 0;
1766 return 1;
1767 }
1768
1769 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1770 {
1771 struct pack_entry e;
1772 return find_pack_entry(sha1, &e, ignore_packed);
1773 }
1774
1775 int has_sha1_file(const unsigned char *sha1)
1776 {
1777 struct stat st;
1778 struct pack_entry e;
1779
1780 if (find_pack_entry(sha1, &e, NULL))
1781 return 1;
1782 return find_sha1_file(sha1, &st) ? 1 : 0;
1783 }
1784
1785 /*
1786 * reads from fd as long as possible into a supplied buffer of size bytes.
1787 * If necessary the buffer's size is increased using realloc()
1788 *
1789 * returns 0 if anything went fine and -1 otherwise
1790 *
1791 * NOTE: both buf and size may change, but even when -1 is returned
1792 * you still have to free() it yourself.
1793 */
1794 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1795 {
1796 char* buf = *return_buf;
1797 unsigned long size = *return_size;
1798 int iret;
1799 unsigned long off = 0;
1800
1801 do {
1802 iret = xread(fd, buf + off, size - off);
1803 if (iret > 0) {
1804 off += iret;
1805 if (off == size) {
1806 size *= 2;
1807 buf = xrealloc(buf, size);
1808 }
1809 }
1810 } while (iret > 0);
1811
1812 *return_buf = buf;
1813 *return_size = off;
1814
1815 if (iret < 0)
1816 return -1;
1817 return 0;
1818 }
1819
1820 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1821 {
1822 unsigned long size = 4096;
1823 char *buf = xmalloc(size);
1824 int ret;
1825
1826 if (read_pipe(fd, &buf, &size)) {
1827 free(buf);
1828 return -1;
1829 }
1830
1831 if (!type)
1832 type = blob_type;
1833 if (write_object)
1834 ret = write_sha1_file(buf, size, type, sha1);
1835 else
1836 ret = hash_sha1_file(buf, size, type, sha1);
1837 free(buf);
1838 return ret;
1839 }
1840
1841 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1842 {
1843 unsigned long size = st->st_size;
1844 void *buf;
1845 int ret;
1846
1847 buf = "";
1848 if (size)
1849 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1850 close(fd);
1851 if (buf == MAP_FAILED)
1852 return -1;
1853
1854 if (!type)
1855 type = blob_type;
1856 if (write_object)
1857 ret = write_sha1_file(buf, size, type, sha1);
1858 else
1859 ret = hash_sha1_file(buf, size, type, sha1);
1860 if (size)
1861 munmap(buf, size);
1862 return ret;
1863 }
1864
1865 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1866 {
1867 int fd;
1868 char *target;
1869
1870 switch (st->st_mode & S_IFMT) {
1871 case S_IFREG:
1872 fd = open(path, O_RDONLY);
1873 if (fd < 0)
1874 return error("open(\"%s\"): %s", path,
1875 strerror(errno));
1876 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1877 return error("%s: failed to insert into database",
1878 path);
1879 break;
1880 case S_IFLNK:
1881 target = xmalloc(st->st_size+1);
1882 if (readlink(path, target, st->st_size+1) != st->st_size) {
1883 char *errstr = strerror(errno);
1884 free(target);
1885 return error("readlink(\"%s\"): %s", path,
1886 errstr);
1887 }
1888 if (!write_object)
1889 hash_sha1_file(target, st->st_size, blob_type, sha1);
1890 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1891 return error("%s: failed to insert into database",
1892 path);
1893 free(target);
1894 break;
1895 default:
1896 return error("%s: unsupported file type", path);
1897 }
1898 return 0;
1899 }