[PATCH] Find size of SHA1 object without inflating everything.
[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
12 #ifndef O_NOATIME
13 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
14 #define O_NOATIME 01000000
15 #else
16 #define O_NOATIME 0
17 #endif
18 #endif
19
20 static unsigned int sha1_file_open_flag = O_NOATIME;
21
22 static unsigned hexval(char c)
23 {
24 if (c >= '0' && c <= '9')
25 return c - '0';
26 if (c >= 'a' && c <= 'f')
27 return c - 'a' + 10;
28 if (c >= 'A' && c <= 'F')
29 return c - 'A' + 10;
30 return ~0;
31 }
32
33 int get_sha1_hex(const char *hex, unsigned char *sha1)
34 {
35 int i;
36 for (i = 0; i < 20; i++) {
37 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
38 if (val & ~0xff)
39 return -1;
40 *sha1++ = val;
41 hex += 2;
42 }
43 return 0;
44 }
45
46 static int get_sha1_file(const char *path, unsigned char *result)
47 {
48 char buffer[60];
49 int fd = open(path, O_RDONLY);
50 int len;
51
52 if (fd < 0)
53 return -1;
54 len = read(fd, buffer, sizeof(buffer));
55 close(fd);
56 if (len < 40)
57 return -1;
58 return get_sha1_hex(buffer, result);
59 }
60
61 static char *git_dir, *git_object_dir, *git_index_file;
62 static void setup_git_env(void)
63 {
64 git_dir = gitenv(GIT_DIR_ENVIRONMENT);
65 if (!git_dir)
66 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
67 git_object_dir = gitenv(DB_ENVIRONMENT);
68 if (!git_object_dir) {
69 git_object_dir = xmalloc(strlen(git_dir) + 9);
70 sprintf(git_object_dir, "%s/objects", git_dir);
71 }
72 git_index_file = gitenv(INDEX_ENVIRONMENT);
73 if (!git_index_file) {
74 git_index_file = xmalloc(strlen(git_dir) + 7);
75 sprintf(git_index_file, "%s/index", git_dir);
76 }
77 }
78
79 char *get_object_directory(void)
80 {
81 if (!git_object_dir)
82 setup_git_env();
83 return git_object_dir;
84 }
85
86 char *get_index_file(void)
87 {
88 if (!git_index_file)
89 setup_git_env();
90 return git_index_file;
91 }
92
93 int get_sha1(const char *str, unsigned char *sha1)
94 {
95 static char pathname[PATH_MAX];
96 static const char *prefix[] = {
97 "",
98 "refs",
99 "refs/tags",
100 "refs/heads",
101 "refs/snap",
102 NULL
103 };
104 const char **p;
105
106 if (!get_sha1_hex(str, sha1))
107 return 0;
108
109 if (!git_dir)
110 setup_git_env();
111 for (p = prefix; *p; p++) {
112 snprintf(pathname, sizeof(pathname), "%s/%s/%s",
113 git_dir, *p, str);
114 if (!get_sha1_file(pathname, sha1))
115 return 0;
116 }
117
118 return -1;
119 }
120
121 char * sha1_to_hex(const unsigned char *sha1)
122 {
123 static char buffer[50];
124 static const char hex[] = "0123456789abcdef";
125 char *buf = buffer;
126 int i;
127
128 for (i = 0; i < 20; i++) {
129 unsigned int val = *sha1++;
130 *buf++ = hex[val >> 4];
131 *buf++ = hex[val & 0xf];
132 }
133 return buffer;
134 }
135
136 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 {
138 int i;
139 for (i = 0; i < 20; i++) {
140 static char hex[] = "0123456789abcdef";
141 unsigned int val = sha1[i];
142 char *pos = pathbuf + i*2 + (i > 0);
143 *pos++ = hex[val >> 4];
144 *pos = hex[val & 0xf];
145 }
146 }
147
148 /*
149 * NOTE! This returns a statically allocated buffer, so you have to be
150 * careful about using it. Do a "strdup()" if you need to save the
151 * filename.
152 *
153 * Also note that this returns the location for creating. Reading
154 * SHA1 file can happen from any alternate directory listed in the
155 * DB_ENVIRONMENT environment variable if it is not found in
156 * the primary object database.
157 */
158 char *sha1_file_name(const unsigned char *sha1)
159 {
160 static char *name, *base;
161
162 if (!base) {
163 const char *sha1_file_directory = get_object_directory();
164 int len = strlen(sha1_file_directory);
165 base = xmalloc(len + 60);
166 memcpy(base, sha1_file_directory, len);
167 memset(base+len, 0, 60);
168 base[len] = '/';
169 base[len+3] = '/';
170 name = base + len + 1;
171 }
172 fill_sha1_path(name, sha1);
173 return base;
174 }
175
176 static struct alternate_object_database {
177 char *base;
178 char *name;
179 } *alt_odb;
180
181 /*
182 * Prepare alternate object database registry.
183 * alt_odb points at an array of struct alternate_object_database.
184 * This array is terminated with an element that has both its base
185 * and name set to NULL. alt_odb[n] comes from n'th non-empty
186 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
187 * variable, and its base points at a statically allocated buffer
188 * that contains "/the/directory/corresponding/to/.git/objects/...",
189 * while its name points just after the slash at the end of
190 * ".git/objects/" in the example above, and has enough space to hold
191 * 40-byte hex SHA1, an extra slash for the first level indirection,
192 * and the terminating NUL.
193 * This function allocates the alt_odb array and all the strings
194 * pointed by base fields of the array elements with one xmalloc();
195 * the string pool immediately follows the array.
196 */
197 static void prepare_alt_odb(void)
198 {
199 int pass, totlen, i;
200 const char *cp, *last;
201 char *op = NULL;
202 const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
203
204 /* The first pass counts how large an area to allocate to
205 * hold the entire alt_odb structure, including array of
206 * structs and path buffers for them. The second pass fills
207 * the structure and prepares the path buffers for use by
208 * fill_sha1_path().
209 */
210 for (totlen = pass = 0; pass < 2; pass++) {
211 last = alt;
212 i = 0;
213 do {
214 cp = strchr(last, ':') ? : last + strlen(last);
215 if (last != cp) {
216 /* 43 = 40-byte + 2 '/' + terminating NUL */
217 int pfxlen = cp - last;
218 int entlen = pfxlen + 43;
219 if (pass == 0)
220 totlen += entlen;
221 else {
222 alt_odb[i].base = op;
223 alt_odb[i].name = op + pfxlen + 1;
224 memcpy(op, last, pfxlen);
225 op[pfxlen] = op[pfxlen + 3] = '/';
226 op[entlen-1] = 0;
227 op += entlen;
228 }
229 i++;
230 }
231 while (*cp && *cp == ':')
232 cp++;
233 last = cp;
234 } while (*cp);
235 if (pass)
236 break;
237 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
238 alt_odb[i].base = alt_odb[i].name = NULL;
239 op = (char*)(&alt_odb[i+1]);
240 }
241 }
242
243 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
244 {
245 int i;
246 char *name = sha1_file_name(sha1);
247
248 if (!stat(name, st))
249 return name;
250 if (!alt_odb)
251 prepare_alt_odb();
252 for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
253 fill_sha1_path(name, sha1);
254 if (!stat(alt_odb[i].base, st))
255 return alt_odb[i].base;
256 }
257 return NULL;
258 }
259
260 int check_sha1_signature(unsigned char *sha1, void *map, unsigned long size, const char *type)
261 {
262 char header[100];
263 unsigned char real_sha1[20];
264 SHA_CTX c;
265
266 SHA1_Init(&c);
267 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
268 SHA1_Update(&c, map, size);
269 SHA1_Final(real_sha1, &c);
270 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
271 }
272
273 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
274 {
275 struct stat st;
276 void *map;
277 int fd;
278 char *filename = find_sha1_file(sha1, &st);
279
280 if (!filename) {
281 error("cannot map sha1 file %s", sha1_to_hex(sha1));
282 return NULL;
283 }
284
285 fd = open(filename, O_RDONLY | sha1_file_open_flag);
286 if (fd < 0) {
287 /* See if it works without O_NOATIME */
288 switch (sha1_file_open_flag) {
289 default:
290 fd = open(filename, O_RDONLY);
291 if (fd >= 0)
292 break;
293 /* Fallthrough */
294 case 0:
295 perror(filename);
296 return NULL;
297 }
298
299 /* If it failed once, it will probably fail again. Stop using O_NOATIME */
300 sha1_file_open_flag = 0;
301 }
302 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
303 close(fd);
304 if (-1 == (int)(long)map)
305 return NULL;
306 *size = st.st_size;
307 return map;
308 }
309
310 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
311 {
312 /* Get the data stream */
313 memset(stream, 0, sizeof(*stream));
314 stream->next_in = map;
315 stream->avail_in = mapsize;
316 stream->next_out = buffer;
317 stream->avail_out = size;
318
319 inflateInit(stream);
320 return inflate(stream, 0);
321 }
322
323 void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
324 {
325 int bytes = strlen(buffer) + 1;
326 char *buf = xmalloc(1+size);
327
328 memcpy(buf, buffer + bytes, stream->total_out - bytes);
329 bytes = stream->total_out - bytes;
330 if (bytes < size) {
331 stream->next_out = buf + bytes;
332 stream->avail_out = size - bytes;
333 while (inflate(stream, Z_FINISH) == Z_OK)
334 /* nothing */;
335 }
336 buf[size] = 0;
337 inflateEnd(stream);
338 return buf;
339 }
340
341 /*
342 * We used to just use "sscanf()", but that's actually way
343 * too permissive for what we want to check. So do an anal
344 * object header parse by hand.
345 */
346 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
347 {
348 int i;
349 unsigned long size;
350
351 /*
352 * The type can be at most ten bytes (including the
353 * terminating '\0' that we add), and is followed by
354 * a space.
355 */
356 i = 10;
357 for (;;) {
358 char c = *hdr++;
359 if (c == ' ')
360 break;
361 if (!--i)
362 return -1;
363 *type++ = c;
364 }
365 *type = 0;
366
367 /*
368 * The length must follow immediately, and be in canonical
369 * decimal format (ie "010" is not valid).
370 */
371 size = *hdr++ - '0';
372 if (size > 9)
373 return -1;
374 if (size) {
375 for (;;) {
376 unsigned long c = *hdr - '0';
377 if (c > 9)
378 break;
379 hdr++;
380 size = size * 10 + c;
381 }
382 }
383 *sizep = size;
384
385 /*
386 * The length must be followed by a zero byte
387 */
388 return *hdr ? -1 : 0;
389 }
390
391 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
392 {
393 int ret;
394 z_stream stream;
395 char hdr[8192];
396
397 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
398 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
399 return NULL;
400
401 return unpack_sha1_rest(&stream, hdr, *size);
402 }
403
404 int sha1_delta_base(const unsigned char *sha1, unsigned char *base_sha1)
405 {
406 int ret;
407 unsigned long mapsize, size;
408 void *map;
409 z_stream stream;
410 char hdr[64], type[20];
411 void *delta_data_head;
412
413 map = map_sha1_file(sha1, &mapsize);
414 if (!map)
415 return -1;
416 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
417 if (ret < Z_OK || parse_sha1_header(hdr, type, &size) < 0) {
418 ret = -1;
419 goto out;
420 }
421 if (strcmp(type, "delta")) {
422 ret = 0;
423 goto out;
424 }
425
426 delta_data_head = hdr + strlen(hdr) + 1;
427 ret = 1;
428 memcpy(base_sha1, delta_data_head, 20);
429 out:
430 inflateEnd(&stream);
431 munmap(map, mapsize);
432 return ret;
433 }
434
435 int sha1_file_size(const unsigned char *sha1, unsigned long *sizep)
436 {
437 int ret, status;
438 unsigned long mapsize, size;
439 void *map;
440 z_stream stream;
441 char hdr[64], type[20];
442 const unsigned char *data;
443 unsigned char cmd;
444 int i;
445
446 map = map_sha1_file(sha1, &mapsize);
447 if (!map)
448 return -1;
449 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
450 status = -1;
451 if (ret < Z_OK || parse_sha1_header(hdr, type, &size) < 0)
452 goto out;
453 if (strcmp(type, "delta")) {
454 *sizep = size;
455 status = 0;
456 goto out;
457 }
458
459 /* We are dealing with a delta object. Inflated, the first
460 * 20 bytes hold the base object SHA1, and delta data follows
461 * immediately after it.
462 *
463 * The initial part of the delta starts at delta_data_head +
464 * 20. Borrow code from patch-delta to read the result size.
465 */
466 data = hdr + strlen(hdr) + 1 + 20;
467
468 /* Skip over the source size; we are not interested in
469 * it and we cannot verify it because we do not want
470 * to read the base object.
471 */
472 cmd = *data++;
473 while (cmd) {
474 if (cmd & 1)
475 data++;
476 cmd >>= 1;
477 }
478 /* Read the result size */
479 size = i = 0;
480 cmd = *data++;
481 while (cmd) {
482 if (cmd & 1)
483 size |= *data++ << i;
484 i += 8;
485 cmd >>= 1;
486 }
487 *sizep = size;
488 status = 0;
489 out:
490 inflateEnd(&stream);
491 munmap(map, mapsize);
492 return status;
493 }
494
495 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
496 {
497 unsigned long mapsize;
498 void *map, *buf;
499
500 map = map_sha1_file(sha1, &mapsize);
501 if (map) {
502 buf = unpack_sha1_file(map, mapsize, type, size);
503 munmap(map, mapsize);
504 if (buf && !strcmp(type, "delta")) {
505 void *ref = NULL, *delta = buf;
506 unsigned long ref_size, delta_size = *size;
507 buf = NULL;
508 if (delta_size > 20)
509 ref = read_sha1_file(delta, type, &ref_size);
510 if (ref)
511 buf = patch_delta(ref, ref_size,
512 delta+20, delta_size-20,
513 size);
514 free(delta);
515 free(ref);
516 }
517 return buf;
518 }
519 return NULL;
520 }
521
522 void *read_object_with_reference(const unsigned char *sha1,
523 const char *required_type,
524 unsigned long *size,
525 unsigned char *actual_sha1_return)
526 {
527 char type[20];
528 void *buffer;
529 unsigned long isize;
530 unsigned char actual_sha1[20];
531
532 memcpy(actual_sha1, sha1, 20);
533 while (1) {
534 int ref_length = -1;
535 const char *ref_type = NULL;
536
537 buffer = read_sha1_file(actual_sha1, type, &isize);
538 if (!buffer)
539 return NULL;
540 if (!strcmp(type, required_type)) {
541 *size = isize;
542 if (actual_sha1_return)
543 memcpy(actual_sha1_return, actual_sha1, 20);
544 return buffer;
545 }
546 /* Handle references */
547 else if (!strcmp(type, "commit"))
548 ref_type = "tree ";
549 else if (!strcmp(type, "tag"))
550 ref_type = "object ";
551 else {
552 free(buffer);
553 return NULL;
554 }
555 ref_length = strlen(ref_type);
556
557 if (memcmp(buffer, ref_type, ref_length) ||
558 get_sha1_hex(buffer + ref_length, actual_sha1)) {
559 free(buffer);
560 return NULL;
561 }
562 /* Now we have the ID of the referred-to object in
563 * actual_sha1. Check again. */
564 }
565 }
566
567 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
568 {
569 int size;
570 unsigned char *compressed;
571 z_stream stream;
572 unsigned char sha1[20];
573 SHA_CTX c;
574 char *filename;
575 static char tmpfile[PATH_MAX];
576 unsigned char hdr[50];
577 int fd, hdrlen, ret;
578
579 /* Generate the header */
580 hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
581
582 /* Sha1.. */
583 SHA1_Init(&c);
584 SHA1_Update(&c, hdr, hdrlen);
585 SHA1_Update(&c, buf, len);
586 SHA1_Final(sha1, &c);
587
588 if (returnsha1)
589 memcpy(returnsha1, sha1, 20);
590
591 filename = sha1_file_name(sha1);
592 fd = open(filename, O_RDONLY);
593 if (fd >= 0) {
594 /*
595 * FIXME!!! We might do collision checking here, but we'd
596 * need to uncompress the old file and check it. Later.
597 */
598 close(fd);
599 return 0;
600 }
601
602 if (errno != ENOENT) {
603 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
604 return -1;
605 }
606
607 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
608
609 fd = mkstemp(tmpfile);
610 if (fd < 0) {
611 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
612 return -1;
613 }
614
615 /* Set it up */
616 memset(&stream, 0, sizeof(stream));
617 deflateInit(&stream, Z_BEST_COMPRESSION);
618 size = deflateBound(&stream, len+hdrlen);
619 compressed = xmalloc(size);
620
621 /* Compress it */
622 stream.next_out = compressed;
623 stream.avail_out = size;
624
625 /* First header.. */
626 stream.next_in = hdr;
627 stream.avail_in = hdrlen;
628 while (deflate(&stream, 0) == Z_OK)
629 /* nothing */;
630
631 /* Then the data itself.. */
632 stream.next_in = buf;
633 stream.avail_in = len;
634 while (deflate(&stream, Z_FINISH) == Z_OK)
635 /* nothing */;
636 deflateEnd(&stream);
637 size = stream.total_out;
638
639 if (write(fd, compressed, size) != size)
640 die("unable to write file");
641 fchmod(fd, 0444);
642 close(fd);
643 free(compressed);
644
645 ret = link(tmpfile, filename);
646 if (ret < 0) {
647 ret = errno;
648
649 /*
650 * Coda hack - coda doesn't like cross-directory links,
651 * so we fall back to a rename, which will mean that it
652 * won't be able to check collisions, but that's not a
653 * big deal.
654 *
655 * When this succeeds, we just return 0. We have nothing
656 * left to unlink.
657 */
658 if (ret == EXDEV && !rename(tmpfile, filename))
659 return 0;
660 }
661 unlink(tmpfile);
662 if (ret) {
663 if (ret != EEXIST) {
664 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
665 return -1;
666 }
667 /* FIXME!!! Collision check here ? */
668 }
669
670 return 0;
671 }
672
673 int write_sha1_from_fd(const unsigned char *sha1, int fd)
674 {
675 char *filename = sha1_file_name(sha1);
676
677 int local;
678 z_stream stream;
679 unsigned char real_sha1[20];
680 unsigned char buf[4096];
681 unsigned char discard[4096];
682 int ret;
683 SHA_CTX c;
684
685 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
686
687 if (local < 0)
688 return error("Couldn't open %s\n", filename);
689
690 memset(&stream, 0, sizeof(stream));
691
692 inflateInit(&stream);
693
694 SHA1_Init(&c);
695
696 do {
697 ssize_t size;
698 size = read(fd, buf, 4096);
699 if (size <= 0) {
700 close(local);
701 unlink(filename);
702 if (!size)
703 return error("Connection closed?");
704 perror("Reading from connection");
705 return -1;
706 }
707 write(local, buf, size);
708 stream.avail_in = size;
709 stream.next_in = buf;
710 do {
711 stream.next_out = discard;
712 stream.avail_out = sizeof(discard);
713 ret = inflate(&stream, Z_SYNC_FLUSH);
714 SHA1_Update(&c, discard, sizeof(discard) -
715 stream.avail_out);
716 } while (stream.avail_in && ret == Z_OK);
717
718 } while (ret == Z_OK);
719 inflateEnd(&stream);
720
721 close(local);
722 SHA1_Final(real_sha1, &c);
723 if (ret != Z_STREAM_END) {
724 unlink(filename);
725 return error("File %s corrupted", sha1_to_hex(sha1));
726 }
727 if (memcmp(sha1, real_sha1, 20)) {
728 unlink(filename);
729 return error("File %s has bad hash\n", sha1_to_hex(sha1));
730 }
731
732 return 0;
733 }
734
735 int has_sha1_file(const unsigned char *sha1)
736 {
737 struct stat st;
738 return !!find_sha1_file(sha1, &st);
739 }
740
741 int index_fd(unsigned char *sha1, int fd, struct stat *st)
742 {
743 unsigned long size = st->st_size;
744 void *buf;
745 int ret;
746
747 buf = "";
748 if (size)
749 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
750 close(fd);
751 if ((int)(long)buf == -1)
752 return -1;
753
754 ret = write_sha1_file(buf, size, "blob", sha1);
755 if (size)
756 munmap(buf, size);
757 return ret;
758 }