unpack_sha1_file: zero-pad the unpacked object.
[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 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
311 {
312 int ret, bytes;
313 z_stream stream;
314 char buffer[8192];
315 unsigned char *buf;
316
317 /* Get the data stream */
318 memset(&stream, 0, sizeof(stream));
319 stream.next_in = map;
320 stream.avail_in = mapsize;
321 stream.next_out = (unsigned char *)buffer;
322 stream.avail_out = sizeof(buffer);
323
324 inflateInit(&stream);
325 ret = inflate(&stream, 0);
326 if (ret < Z_OK)
327 return NULL;
328 if (sscanf(buffer, "%10s %lu", type, size) != 2)
329 return NULL;
330
331 bytes = strlen(buffer) + 1;
332 buf = xmalloc(1+*size);
333
334 memcpy(buf, buffer + bytes, stream.total_out - bytes);
335 bytes = stream.total_out - bytes;
336 if (bytes < *size && ret == Z_OK) {
337 stream.next_out = buf + bytes;
338 stream.avail_out = *size - bytes;
339 while (inflate(&stream, Z_FINISH) == Z_OK)
340 /* nothing */;
341 }
342 buf[*size] = 0;
343 inflateEnd(&stream);
344 return buf;
345 }
346
347 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
348 {
349 unsigned long mapsize;
350 void *map, *buf;
351
352 map = map_sha1_file(sha1, &mapsize);
353 if (map) {
354 buf = unpack_sha1_file(map, mapsize, type, size);
355 munmap(map, mapsize);
356 if (buf && !strcmp(type, "delta")) {
357 void *ref = NULL, *delta = buf;
358 unsigned long ref_size, delta_size = *size;
359 buf = NULL;
360 if (delta_size > 20)
361 ref = read_sha1_file(delta, type, &ref_size);
362 if (ref)
363 buf = patch_delta(ref, ref_size,
364 delta+20, delta_size-20,
365 size);
366 free(delta);
367 free(ref);
368 }
369 return buf;
370 }
371 return NULL;
372 }
373
374 void *read_object_with_reference(const unsigned char *sha1,
375 const char *required_type,
376 unsigned long *size,
377 unsigned char *actual_sha1_return)
378 {
379 char type[20];
380 void *buffer;
381 unsigned long isize;
382 unsigned char actual_sha1[20];
383
384 memcpy(actual_sha1, sha1, 20);
385 while (1) {
386 int ref_length = -1;
387 const char *ref_type = NULL;
388
389 buffer = read_sha1_file(actual_sha1, type, &isize);
390 if (!buffer)
391 return NULL;
392 if (!strcmp(type, required_type)) {
393 *size = isize;
394 if (actual_sha1_return)
395 memcpy(actual_sha1_return, actual_sha1, 20);
396 return buffer;
397 }
398 /* Handle references */
399 else if (!strcmp(type, "commit"))
400 ref_type = "tree ";
401 else if (!strcmp(type, "tag"))
402 ref_type = "object ";
403 else {
404 free(buffer);
405 return NULL;
406 }
407 ref_length = strlen(ref_type);
408
409 if (memcmp(buffer, ref_type, ref_length) ||
410 get_sha1_hex(buffer + ref_length, actual_sha1)) {
411 free(buffer);
412 return NULL;
413 }
414 /* Now we have the ID of the referred-to object in
415 * actual_sha1. Check again. */
416 }
417 }
418
419 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
420 {
421 int size;
422 unsigned char *compressed;
423 z_stream stream;
424 unsigned char sha1[20];
425 SHA_CTX c;
426 char *filename;
427 static char tmpfile[PATH_MAX];
428 unsigned char hdr[50];
429 int fd, hdrlen, ret;
430
431 /* Generate the header */
432 hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
433
434 /* Sha1.. */
435 SHA1_Init(&c);
436 SHA1_Update(&c, hdr, hdrlen);
437 SHA1_Update(&c, buf, len);
438 SHA1_Final(sha1, &c);
439
440 if (returnsha1)
441 memcpy(returnsha1, sha1, 20);
442
443 filename = sha1_file_name(sha1);
444 fd = open(filename, O_RDONLY);
445 if (fd >= 0) {
446 /*
447 * FIXME!!! We might do collision checking here, but we'd
448 * need to uncompress the old file and check it. Later.
449 */
450 close(fd);
451 return 0;
452 }
453
454 if (errno != ENOENT) {
455 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
456 return -1;
457 }
458
459 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
460
461 fd = mkstemp(tmpfile);
462 if (fd < 0) {
463 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
464 return -1;
465 }
466
467 /* Set it up */
468 memset(&stream, 0, sizeof(stream));
469 deflateInit(&stream, Z_BEST_COMPRESSION);
470 size = deflateBound(&stream, len+hdrlen);
471 compressed = xmalloc(size);
472
473 /* Compress it */
474 stream.next_out = compressed;
475 stream.avail_out = size;
476
477 /* First header.. */
478 stream.next_in = hdr;
479 stream.avail_in = hdrlen;
480 while (deflate(&stream, 0) == Z_OK)
481 /* nothing */;
482
483 /* Then the data itself.. */
484 stream.next_in = buf;
485 stream.avail_in = len;
486 while (deflate(&stream, Z_FINISH) == Z_OK)
487 /* nothing */;
488 deflateEnd(&stream);
489 size = stream.total_out;
490
491 if (write(fd, compressed, size) != size)
492 die("unable to write file");
493 fchmod(fd, 0444);
494 close(fd);
495 free(compressed);
496
497 ret = link(tmpfile, filename);
498 if (ret < 0) {
499 ret = errno;
500
501 /*
502 * Coda hack - coda doesn't like cross-directory links,
503 * so we fall back to a rename, which will mean that it
504 * won't be able to check collisions, but that's not a
505 * big deal.
506 *
507 * When this succeeds, we just return 0. We have nothing
508 * left to unlink.
509 */
510 if (ret == EXDEV && !rename(tmpfile, filename))
511 return 0;
512 }
513 unlink(tmpfile);
514 if (ret) {
515 if (ret != EEXIST) {
516 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
517 return -1;
518 }
519 /* FIXME!!! Collision check here ? */
520 }
521
522 return 0;
523 }
524
525 int write_sha1_from_fd(const unsigned char *sha1, int fd)
526 {
527 char *filename = sha1_file_name(sha1);
528
529 int local;
530 z_stream stream;
531 unsigned char real_sha1[20];
532 unsigned char buf[4096];
533 unsigned char discard[4096];
534 int ret;
535 SHA_CTX c;
536
537 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
538
539 if (local < 0)
540 return error("Couldn't open %s\n", filename);
541
542 memset(&stream, 0, sizeof(stream));
543
544 inflateInit(&stream);
545
546 SHA1_Init(&c);
547
548 do {
549 ssize_t size;
550 size = read(fd, buf, 4096);
551 if (size <= 0) {
552 close(local);
553 unlink(filename);
554 if (!size)
555 return error("Connection closed?");
556 perror("Reading from connection");
557 return -1;
558 }
559 write(local, buf, size);
560 stream.avail_in = size;
561 stream.next_in = buf;
562 do {
563 stream.next_out = discard;
564 stream.avail_out = sizeof(discard);
565 ret = inflate(&stream, Z_SYNC_FLUSH);
566 SHA1_Update(&c, discard, sizeof(discard) -
567 stream.avail_out);
568 } while (stream.avail_in && ret == Z_OK);
569
570 } while (ret == Z_OK);
571 inflateEnd(&stream);
572
573 close(local);
574 SHA1_Final(real_sha1, &c);
575 if (ret != Z_STREAM_END) {
576 unlink(filename);
577 return error("File %s corrupted", sha1_to_hex(sha1));
578 }
579 if (memcmp(sha1, real_sha1, 20)) {
580 unlink(filename);
581 return error("File %s has bad hash\n", sha1_to_hex(sha1));
582 }
583
584 return 0;
585 }
586
587 int has_sha1_file(const unsigned char *sha1)
588 {
589 struct stat st;
590 return !!find_sha1_file(sha1, &st);
591 }
592
593 int index_fd(unsigned char *sha1, int fd, struct stat *st)
594 {
595 unsigned long size = st->st_size;
596 void *buf;
597 int ret;
598
599 buf = "";
600 if (size)
601 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
602 close(fd);
603 if ((int)(long)buf == -1)
604 return -1;
605
606 ret = write_sha1_file(buf, size, "blob", sha1);
607 if (size)
608 munmap(buf, size);
609 return ret;
610 }