[PATCH] Remove "delta" object representation.
[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
11 #ifndef O_NOATIME
12 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
13 #define O_NOATIME 01000000
14 #else
15 #define O_NOATIME 0
16 #endif
17 #endif
18
19 static unsigned int sha1_file_open_flag = O_NOATIME;
20
21 static unsigned hexval(char c)
22 {
23 if (c >= '0' && c <= '9')
24 return c - '0';
25 if (c >= 'a' && c <= 'f')
26 return c - 'a' + 10;
27 if (c >= 'A' && c <= 'F')
28 return c - 'A' + 10;
29 return ~0;
30 }
31
32 int get_sha1_hex(const char *hex, unsigned char *sha1)
33 {
34 int i;
35 for (i = 0; i < 20; i++) {
36 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
37 if (val & ~0xff)
38 return -1;
39 *sha1++ = val;
40 hex += 2;
41 }
42 return 0;
43 }
44
45 static int get_sha1_file(const char *path, unsigned char *result)
46 {
47 char buffer[60];
48 int fd = open(path, O_RDONLY);
49 int len;
50
51 if (fd < 0)
52 return -1;
53 len = read(fd, buffer, sizeof(buffer));
54 close(fd);
55 if (len < 40)
56 return -1;
57 return get_sha1_hex(buffer, result);
58 }
59
60 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir;
61 static void setup_git_env(void)
62 {
63 git_dir = gitenv(GIT_DIR_ENVIRONMENT);
64 if (!git_dir)
65 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
66 git_object_dir = gitenv(DB_ENVIRONMENT);
67 if (!git_object_dir) {
68 git_object_dir = xmalloc(strlen(git_dir) + 9);
69 sprintf(git_object_dir, "%s/objects", git_dir);
70 }
71 git_refs_dir = xmalloc(strlen(git_dir) + 6);
72 sprintf(git_refs_dir, "%s/refs", git_dir);
73 git_index_file = gitenv(INDEX_ENVIRONMENT);
74 if (!git_index_file) {
75 git_index_file = xmalloc(strlen(git_dir) + 7);
76 sprintf(git_index_file, "%s/index", git_dir);
77 }
78 }
79
80 char *get_object_directory(void)
81 {
82 if (!git_object_dir)
83 setup_git_env();
84 return git_object_dir;
85 }
86
87 char *get_refs_directory(void)
88 {
89 if (!git_refs_dir)
90 setup_git_env();
91 return git_refs_dir;
92 }
93
94 char *get_index_file(void)
95 {
96 if (!git_index_file)
97 setup_git_env();
98 return git_index_file;
99 }
100
101 int get_sha1(const char *str, unsigned char *sha1)
102 {
103 static char pathname[PATH_MAX];
104 static const char *prefix[] = {
105 "",
106 "refs",
107 "refs/tags",
108 "refs/heads",
109 "refs/snap",
110 NULL
111 };
112 const char **p;
113
114 if (!get_sha1_hex(str, sha1))
115 return 0;
116
117 if (!git_dir)
118 setup_git_env();
119 for (p = prefix; *p; p++) {
120 snprintf(pathname, sizeof(pathname), "%s/%s/%s",
121 git_dir, *p, str);
122 if (!get_sha1_file(pathname, sha1))
123 return 0;
124 }
125
126 return -1;
127 }
128
129 char * sha1_to_hex(const unsigned char *sha1)
130 {
131 static char buffer[50];
132 static const char hex[] = "0123456789abcdef";
133 char *buf = buffer;
134 int i;
135
136 for (i = 0; i < 20; i++) {
137 unsigned int val = *sha1++;
138 *buf++ = hex[val >> 4];
139 *buf++ = hex[val & 0xf];
140 }
141 return buffer;
142 }
143
144 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
145 {
146 int i;
147 for (i = 0; i < 20; i++) {
148 static char hex[] = "0123456789abcdef";
149 unsigned int val = sha1[i];
150 char *pos = pathbuf + i*2 + (i > 0);
151 *pos++ = hex[val >> 4];
152 *pos = hex[val & 0xf];
153 }
154 }
155
156 /*
157 * NOTE! This returns a statically allocated buffer, so you have to be
158 * careful about using it. Do a "strdup()" if you need to save the
159 * filename.
160 *
161 * Also note that this returns the location for creating. Reading
162 * SHA1 file can happen from any alternate directory listed in the
163 * DB_ENVIRONMENT environment variable if it is not found in
164 * the primary object database.
165 */
166 char *sha1_file_name(const unsigned char *sha1)
167 {
168 static char *name, *base;
169
170 if (!base) {
171 const char *sha1_file_directory = get_object_directory();
172 int len = strlen(sha1_file_directory);
173 base = xmalloc(len + 60);
174 memcpy(base, sha1_file_directory, len);
175 memset(base+len, 0, 60);
176 base[len] = '/';
177 base[len+3] = '/';
178 name = base + len + 1;
179 }
180 fill_sha1_path(name, sha1);
181 return base;
182 }
183
184 static struct alternate_object_database {
185 char *base;
186 char *name;
187 } *alt_odb;
188
189 /*
190 * Prepare alternate object database registry.
191 * alt_odb points at an array of struct alternate_object_database.
192 * This array is terminated with an element that has both its base
193 * and name set to NULL. alt_odb[n] comes from n'th non-empty
194 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
195 * variable, and its base points at a statically allocated buffer
196 * that contains "/the/directory/corresponding/to/.git/objects/...",
197 * while its name points just after the slash at the end of
198 * ".git/objects/" in the example above, and has enough space to hold
199 * 40-byte hex SHA1, an extra slash for the first level indirection,
200 * and the terminating NUL.
201 * This function allocates the alt_odb array and all the strings
202 * pointed by base fields of the array elements with one xmalloc();
203 * the string pool immediately follows the array.
204 */
205 static void prepare_alt_odb(void)
206 {
207 int pass, totlen, i;
208 const char *cp, *last;
209 char *op = NULL;
210 const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
211
212 /* The first pass counts how large an area to allocate to
213 * hold the entire alt_odb structure, including array of
214 * structs and path buffers for them. The second pass fills
215 * the structure and prepares the path buffers for use by
216 * fill_sha1_path().
217 */
218 for (totlen = pass = 0; pass < 2; pass++) {
219 last = alt;
220 i = 0;
221 do {
222 cp = strchr(last, ':') ? : last + strlen(last);
223 if (last != cp) {
224 /* 43 = 40-byte + 2 '/' + terminating NUL */
225 int pfxlen = cp - last;
226 int entlen = pfxlen + 43;
227 if (pass == 0)
228 totlen += entlen;
229 else {
230 alt_odb[i].base = op;
231 alt_odb[i].name = op + pfxlen + 1;
232 memcpy(op, last, pfxlen);
233 op[pfxlen] = op[pfxlen + 3] = '/';
234 op[entlen-1] = 0;
235 op += entlen;
236 }
237 i++;
238 }
239 while (*cp && *cp == ':')
240 cp++;
241 last = cp;
242 } while (*cp);
243 if (pass)
244 break;
245 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
246 alt_odb[i].base = alt_odb[i].name = NULL;
247 op = (char*)(&alt_odb[i+1]);
248 }
249 }
250
251 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
252 {
253 int i;
254 char *name = sha1_file_name(sha1);
255
256 if (!stat(name, st))
257 return name;
258 if (!alt_odb)
259 prepare_alt_odb();
260 for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
261 fill_sha1_path(name, sha1);
262 if (!stat(alt_odb[i].base, st))
263 return alt_odb[i].base;
264 }
265 return NULL;
266 }
267
268 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
269 {
270 char header[100];
271 unsigned char real_sha1[20];
272 SHA_CTX c;
273
274 SHA1_Init(&c);
275 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
276 SHA1_Update(&c, map, size);
277 SHA1_Final(real_sha1, &c);
278 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
279 }
280
281 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
282 {
283 struct stat st;
284 void *map;
285 int fd;
286 char *filename = find_sha1_file(sha1, &st);
287
288 if (!filename) {
289 error("cannot map sha1 file %s", sha1_to_hex(sha1));
290 return NULL;
291 }
292
293 fd = open(filename, O_RDONLY | sha1_file_open_flag);
294 if (fd < 0) {
295 /* See if it works without O_NOATIME */
296 switch (sha1_file_open_flag) {
297 default:
298 fd = open(filename, O_RDONLY);
299 if (fd >= 0)
300 break;
301 /* Fallthrough */
302 case 0:
303 perror(filename);
304 return NULL;
305 }
306
307 /* If it failed once, it will probably fail again. Stop using O_NOATIME */
308 sha1_file_open_flag = 0;
309 }
310 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
311 close(fd);
312 if (-1 == (int)(long)map)
313 return NULL;
314 *size = st.st_size;
315 return map;
316 }
317
318 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
319 {
320 /* Get the data stream */
321 memset(stream, 0, sizeof(*stream));
322 stream->next_in = map;
323 stream->avail_in = mapsize;
324 stream->next_out = buffer;
325 stream->avail_out = size;
326
327 inflateInit(stream);
328 return inflate(stream, 0);
329 }
330
331 void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
332 {
333 int bytes = strlen(buffer) + 1;
334 unsigned char *buf = xmalloc(1+size);
335
336 memcpy(buf, buffer + bytes, stream->total_out - bytes);
337 bytes = stream->total_out - bytes;
338 if (bytes < size) {
339 stream->next_out = buf + bytes;
340 stream->avail_out = size - bytes;
341 while (inflate(stream, Z_FINISH) == Z_OK)
342 /* nothing */;
343 }
344 buf[size] = 0;
345 inflateEnd(stream);
346 return buf;
347 }
348
349 /*
350 * We used to just use "sscanf()", but that's actually way
351 * too permissive for what we want to check. So do an anal
352 * object header parse by hand.
353 */
354 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
355 {
356 int i;
357 unsigned long size;
358
359 /*
360 * The type can be at most ten bytes (including the
361 * terminating '\0' that we add), and is followed by
362 * a space.
363 */
364 i = 10;
365 for (;;) {
366 char c = *hdr++;
367 if (c == ' ')
368 break;
369 if (!--i)
370 return -1;
371 *type++ = c;
372 }
373 *type = 0;
374
375 /*
376 * The length must follow immediately, and be in canonical
377 * decimal format (ie "010" is not valid).
378 */
379 size = *hdr++ - '0';
380 if (size > 9)
381 return -1;
382 if (size) {
383 for (;;) {
384 unsigned long c = *hdr - '0';
385 if (c > 9)
386 break;
387 hdr++;
388 size = size * 10 + c;
389 }
390 }
391 *sizep = size;
392
393 /*
394 * The length must be followed by a zero byte
395 */
396 return *hdr ? -1 : 0;
397 }
398
399 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
400 {
401 int ret;
402 z_stream stream;
403 char hdr[8192];
404
405 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
406 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
407 return NULL;
408
409 return unpack_sha1_rest(&stream, hdr, *size);
410 }
411
412 int sha1_file_size(const unsigned char *sha1, unsigned long *sizep)
413 {
414 int ret, status;
415 unsigned long mapsize, size;
416 void *map;
417 z_stream stream;
418 char hdr[64], type[20];
419
420 map = map_sha1_file(sha1, &mapsize);
421 if (!map)
422 return -1;
423 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
424 if (ret < Z_OK || parse_sha1_header(hdr, type, &size) < 0)
425 status = -1;
426 else {
427 status = 0;
428 *sizep = size;
429 }
430 inflateEnd(&stream);
431 munmap(map, mapsize);
432 return status;
433 }
434
435 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
436 {
437 unsigned long mapsize;
438 void *map, *buf;
439
440 map = map_sha1_file(sha1, &mapsize);
441 if (map) {
442 buf = unpack_sha1_file(map, mapsize, type, size);
443 munmap(map, mapsize);
444 return buf;
445 }
446 return NULL;
447 }
448
449 void *read_object_with_reference(const unsigned char *sha1,
450 const char *required_type,
451 unsigned long *size,
452 unsigned char *actual_sha1_return)
453 {
454 char type[20];
455 void *buffer;
456 unsigned long isize;
457 unsigned char actual_sha1[20];
458
459 memcpy(actual_sha1, sha1, 20);
460 while (1) {
461 int ref_length = -1;
462 const char *ref_type = NULL;
463
464 buffer = read_sha1_file(actual_sha1, type, &isize);
465 if (!buffer)
466 return NULL;
467 if (!strcmp(type, required_type)) {
468 *size = isize;
469 if (actual_sha1_return)
470 memcpy(actual_sha1_return, actual_sha1, 20);
471 return buffer;
472 }
473 /* Handle references */
474 else if (!strcmp(type, "commit"))
475 ref_type = "tree ";
476 else if (!strcmp(type, "tag"))
477 ref_type = "object ";
478 else {
479 free(buffer);
480 return NULL;
481 }
482 ref_length = strlen(ref_type);
483
484 if (memcmp(buffer, ref_type, ref_length) ||
485 get_sha1_hex(buffer + ref_length, actual_sha1)) {
486 free(buffer);
487 return NULL;
488 }
489 /* Now we have the ID of the referred-to object in
490 * actual_sha1. Check again. */
491 }
492 }
493
494 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
495 {
496 int size;
497 unsigned char *compressed;
498 z_stream stream;
499 unsigned char sha1[20];
500 SHA_CTX c;
501 char *filename;
502 static char tmpfile[PATH_MAX];
503 unsigned char hdr[50];
504 int fd, hdrlen, ret;
505
506 /* Generate the header */
507 hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
508
509 /* Sha1.. */
510 SHA1_Init(&c);
511 SHA1_Update(&c, hdr, hdrlen);
512 SHA1_Update(&c, buf, len);
513 SHA1_Final(sha1, &c);
514
515 if (returnsha1)
516 memcpy(returnsha1, sha1, 20);
517
518 filename = sha1_file_name(sha1);
519 fd = open(filename, O_RDONLY);
520 if (fd >= 0) {
521 /*
522 * FIXME!!! We might do collision checking here, but we'd
523 * need to uncompress the old file and check it. Later.
524 */
525 close(fd);
526 return 0;
527 }
528
529 if (errno != ENOENT) {
530 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
531 return -1;
532 }
533
534 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
535
536 fd = mkstemp(tmpfile);
537 if (fd < 0) {
538 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
539 return -1;
540 }
541
542 /* Set it up */
543 memset(&stream, 0, sizeof(stream));
544 deflateInit(&stream, Z_BEST_COMPRESSION);
545 size = deflateBound(&stream, len+hdrlen);
546 compressed = xmalloc(size);
547
548 /* Compress it */
549 stream.next_out = compressed;
550 stream.avail_out = size;
551
552 /* First header.. */
553 stream.next_in = hdr;
554 stream.avail_in = hdrlen;
555 while (deflate(&stream, 0) == Z_OK)
556 /* nothing */;
557
558 /* Then the data itself.. */
559 stream.next_in = buf;
560 stream.avail_in = len;
561 while (deflate(&stream, Z_FINISH) == Z_OK)
562 /* nothing */;
563 deflateEnd(&stream);
564 size = stream.total_out;
565
566 if (write(fd, compressed, size) != size)
567 die("unable to write file");
568 fchmod(fd, 0444);
569 close(fd);
570 free(compressed);
571
572 ret = link(tmpfile, filename);
573 if (ret < 0) {
574 ret = errno;
575
576 /*
577 * Coda hack - coda doesn't like cross-directory links,
578 * so we fall back to a rename, which will mean that it
579 * won't be able to check collisions, but that's not a
580 * big deal.
581 *
582 * When this succeeds, we just return 0. We have nothing
583 * left to unlink.
584 */
585 if (ret == EXDEV && !rename(tmpfile, filename))
586 return 0;
587 }
588 unlink(tmpfile);
589 if (ret) {
590 if (ret != EEXIST) {
591 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
592 return -1;
593 }
594 /* FIXME!!! Collision check here ? */
595 }
596
597 return 0;
598 }
599
600 int write_sha1_from_fd(const unsigned char *sha1, int fd)
601 {
602 char *filename = sha1_file_name(sha1);
603
604 int local;
605 z_stream stream;
606 unsigned char real_sha1[20];
607 unsigned char buf[4096];
608 unsigned char discard[4096];
609 int ret;
610 SHA_CTX c;
611
612 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
613
614 if (local < 0)
615 return error("Couldn't open %s\n", filename);
616
617 memset(&stream, 0, sizeof(stream));
618
619 inflateInit(&stream);
620
621 SHA1_Init(&c);
622
623 do {
624 ssize_t size;
625 size = read(fd, buf, 4096);
626 if (size <= 0) {
627 close(local);
628 unlink(filename);
629 if (!size)
630 return error("Connection closed?");
631 perror("Reading from connection");
632 return -1;
633 }
634 write(local, buf, size);
635 stream.avail_in = size;
636 stream.next_in = buf;
637 do {
638 stream.next_out = discard;
639 stream.avail_out = sizeof(discard);
640 ret = inflate(&stream, Z_SYNC_FLUSH);
641 SHA1_Update(&c, discard, sizeof(discard) -
642 stream.avail_out);
643 } while (stream.avail_in && ret == Z_OK);
644
645 } while (ret == Z_OK);
646 inflateEnd(&stream);
647
648 close(local);
649 SHA1_Final(real_sha1, &c);
650 if (ret != Z_STREAM_END) {
651 unlink(filename);
652 return error("File %s corrupted", sha1_to_hex(sha1));
653 }
654 if (memcmp(sha1, real_sha1, 20)) {
655 unlink(filename);
656 return error("File %s has bad hash\n", sha1_to_hex(sha1));
657 }
658
659 return 0;
660 }
661
662 int has_sha1_file(const unsigned char *sha1)
663 {
664 struct stat st;
665 return !!find_sha1_file(sha1, &st);
666 }
667
668 int index_fd(unsigned char *sha1, int fd, struct stat *st)
669 {
670 unsigned long size = st->st_size;
671 void *buf;
672 int ret;
673
674 buf = "";
675 if (size)
676 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
677 close(fd);
678 if ((int)(long)buf == -1)
679 return -1;
680
681 ret = write_sha1_file(buf, size, "blob", sha1);
682 if (size)
683 munmap(buf, size);
684 return ret;
685 }