[PATCH] Prevent git-rev-list without --merge-order producing duplicates in output
[git/dscho.git] / sha1_file.c
blobe648c068b558fbdbbdc866c8829af3641e2c5b60
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
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
20 static unsigned int sha1_file_open_flag = O_NOATIME;
22 static unsigned hexval(char c)
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;
33 int get_sha1_hex(const char *hex, unsigned char *sha1)
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;
43 return 0;
46 static int get_sha1_file(const char *path, unsigned char *result)
48 char buffer[60];
49 int fd = open(path, O_RDONLY);
50 int len;
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);
61 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir;
62 static void setup_git_env(void)
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);
72 git_refs_dir = xmalloc(strlen(git_dir) + 6);
73 sprintf(git_refs_dir, "%s/refs", git_dir);
74 git_index_file = gitenv(INDEX_ENVIRONMENT);
75 if (!git_index_file) {
76 git_index_file = xmalloc(strlen(git_dir) + 7);
77 sprintf(git_index_file, "%s/index", git_dir);
81 char *get_object_directory(void)
83 if (!git_object_dir)
84 setup_git_env();
85 return git_object_dir;
88 char *get_refs_directory(void)
90 if (!git_refs_dir)
91 setup_git_env();
92 return git_refs_dir;
95 char *get_index_file(void)
97 if (!git_index_file)
98 setup_git_env();
99 return git_index_file;
102 int get_sha1(const char *str, unsigned char *sha1)
104 static char pathname[PATH_MAX];
105 static const char *prefix[] = {
107 "refs",
108 "refs/tags",
109 "refs/heads",
110 "refs/snap",
111 NULL
113 const char **p;
115 if (!get_sha1_hex(str, sha1))
116 return 0;
118 if (!git_dir)
119 setup_git_env();
120 for (p = prefix; *p; p++) {
121 snprintf(pathname, sizeof(pathname), "%s/%s/%s",
122 git_dir, *p, str);
123 if (!get_sha1_file(pathname, sha1))
124 return 0;
127 return -1;
130 char * sha1_to_hex(const unsigned char *sha1)
132 static char buffer[50];
133 static const char hex[] = "0123456789abcdef";
134 char *buf = buffer;
135 int i;
137 for (i = 0; i < 20; i++) {
138 unsigned int val = *sha1++;
139 *buf++ = hex[val >> 4];
140 *buf++ = hex[val & 0xf];
142 return buffer;
145 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
147 int i;
148 for (i = 0; i < 20; i++) {
149 static char hex[] = "0123456789abcdef";
150 unsigned int val = sha1[i];
151 char *pos = pathbuf + i*2 + (i > 0);
152 *pos++ = hex[val >> 4];
153 *pos = hex[val & 0xf];
158 * NOTE! This returns a statically allocated buffer, so you have to be
159 * careful about using it. Do a "strdup()" if you need to save the
160 * filename.
162 * Also note that this returns the location for creating. Reading
163 * SHA1 file can happen from any alternate directory listed in the
164 * DB_ENVIRONMENT environment variable if it is not found in
165 * the primary object database.
167 char *sha1_file_name(const unsigned char *sha1)
169 static char *name, *base;
171 if (!base) {
172 const char *sha1_file_directory = get_object_directory();
173 int len = strlen(sha1_file_directory);
174 base = xmalloc(len + 60);
175 memcpy(base, sha1_file_directory, len);
176 memset(base+len, 0, 60);
177 base[len] = '/';
178 base[len+3] = '/';
179 name = base + len + 1;
181 fill_sha1_path(name, sha1);
182 return base;
185 static struct alternate_object_database {
186 char *base;
187 char *name;
188 } *alt_odb;
191 * Prepare alternate object database registry.
192 * alt_odb points at an array of struct alternate_object_database.
193 * This array is terminated with an element that has both its base
194 * and name set to NULL. alt_odb[n] comes from n'th non-empty
195 * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
196 * variable, and its base points at a statically allocated buffer
197 * that contains "/the/directory/corresponding/to/.git/objects/...",
198 * while its name points just after the slash at the end of
199 * ".git/objects/" in the example above, and has enough space to hold
200 * 40-byte hex SHA1, an extra slash for the first level indirection,
201 * and the terminating NUL.
202 * This function allocates the alt_odb array and all the strings
203 * pointed by base fields of the array elements with one xmalloc();
204 * the string pool immediately follows the array.
206 static void prepare_alt_odb(void)
208 int pass, totlen, i;
209 const char *cp, *last;
210 char *op = NULL;
211 const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
213 /* The first pass counts how large an area to allocate to
214 * hold the entire alt_odb structure, including array of
215 * structs and path buffers for them. The second pass fills
216 * the structure and prepares the path buffers for use by
217 * fill_sha1_path().
219 for (totlen = pass = 0; pass < 2; pass++) {
220 last = alt;
221 i = 0;
222 do {
223 cp = strchr(last, ':') ? : last + strlen(last);
224 if (last != cp) {
225 /* 43 = 40-byte + 2 '/' + terminating NUL */
226 int pfxlen = cp - last;
227 int entlen = pfxlen + 43;
228 if (pass == 0)
229 totlen += entlen;
230 else {
231 alt_odb[i].base = op;
232 alt_odb[i].name = op + pfxlen + 1;
233 memcpy(op, last, pfxlen);
234 op[pfxlen] = op[pfxlen + 3] = '/';
235 op[entlen-1] = 0;
236 op += entlen;
238 i++;
240 while (*cp && *cp == ':')
241 cp++;
242 last = cp;
243 } while (*cp);
244 if (pass)
245 break;
246 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
247 alt_odb[i].base = alt_odb[i].name = NULL;
248 op = (char*)(&alt_odb[i+1]);
252 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
254 int i;
255 char *name = sha1_file_name(sha1);
257 if (!stat(name, st))
258 return name;
259 if (!alt_odb)
260 prepare_alt_odb();
261 for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
262 fill_sha1_path(name, sha1);
263 if (!stat(alt_odb[i].base, st))
264 return alt_odb[i].base;
266 return NULL;
269 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
271 char header[100];
272 unsigned char real_sha1[20];
273 SHA_CTX c;
275 SHA1_Init(&c);
276 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
277 SHA1_Update(&c, map, size);
278 SHA1_Final(real_sha1, &c);
279 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
282 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
284 struct stat st;
285 void *map;
286 int fd;
287 char *filename = find_sha1_file(sha1, &st);
289 if (!filename) {
290 error("cannot map sha1 file %s", sha1_to_hex(sha1));
291 return NULL;
294 fd = open(filename, O_RDONLY | sha1_file_open_flag);
295 if (fd < 0) {
296 /* See if it works without O_NOATIME */
297 switch (sha1_file_open_flag) {
298 default:
299 fd = open(filename, O_RDONLY);
300 if (fd >= 0)
301 break;
302 /* Fallthrough */
303 case 0:
304 perror(filename);
305 return NULL;
308 /* If it failed once, it will probably fail again. Stop using O_NOATIME */
309 sha1_file_open_flag = 0;
311 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
312 close(fd);
313 if (-1 == (int)(long)map)
314 return NULL;
315 *size = st.st_size;
316 return map;
319 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
321 /* Get the data stream */
322 memset(stream, 0, sizeof(*stream));
323 stream->next_in = map;
324 stream->avail_in = mapsize;
325 stream->next_out = buffer;
326 stream->avail_out = size;
328 inflateInit(stream);
329 return inflate(stream, 0);
332 void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
334 int bytes = strlen(buffer) + 1;
335 char *buf = xmalloc(1+size);
337 memcpy(buf, buffer + bytes, stream->total_out - bytes);
338 bytes = stream->total_out - bytes;
339 if (bytes < size) {
340 stream->next_out = buf + bytes;
341 stream->avail_out = size - bytes;
342 while (inflate(stream, Z_FINISH) == Z_OK)
343 /* nothing */;
345 buf[size] = 0;
346 inflateEnd(stream);
347 return buf;
351 * We used to just use "sscanf()", but that's actually way
352 * too permissive for what we want to check. So do an anal
353 * object header parse by hand.
355 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
357 int i;
358 unsigned long size;
361 * The type can be at most ten bytes (including the
362 * terminating '\0' that we add), and is followed by
363 * a space.
365 i = 10;
366 for (;;) {
367 char c = *hdr++;
368 if (c == ' ')
369 break;
370 if (!--i)
371 return -1;
372 *type++ = c;
374 *type = 0;
377 * The length must follow immediately, and be in canonical
378 * decimal format (ie "010" is not valid).
380 size = *hdr++ - '0';
381 if (size > 9)
382 return -1;
383 if (size) {
384 for (;;) {
385 unsigned long c = *hdr - '0';
386 if (c > 9)
387 break;
388 hdr++;
389 size = size * 10 + c;
392 *sizep = size;
395 * The length must be followed by a zero byte
397 return *hdr ? -1 : 0;
400 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
402 int ret;
403 z_stream stream;
404 char hdr[8192];
406 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
407 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
408 return NULL;
410 return unpack_sha1_rest(&stream, hdr, *size);
413 int sha1_delta_base(const unsigned char *sha1, unsigned char *base_sha1)
415 int ret;
416 unsigned long mapsize, size;
417 void *map;
418 z_stream stream;
419 char hdr[64], type[20];
420 void *delta_data_head;
422 map = map_sha1_file(sha1, &mapsize);
423 if (!map)
424 return -1;
425 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
426 if (ret < Z_OK || parse_sha1_header(hdr, type, &size) < 0) {
427 ret = -1;
428 goto out;
430 if (strcmp(type, "delta")) {
431 ret = 0;
432 goto out;
435 delta_data_head = hdr + strlen(hdr) + 1;
436 ret = 1;
437 memcpy(base_sha1, delta_data_head, 20);
438 out:
439 inflateEnd(&stream);
440 munmap(map, mapsize);
441 return ret;
444 int sha1_file_size(const unsigned char *sha1, unsigned long *sizep)
446 int ret, status;
447 unsigned long mapsize, size;
448 void *map;
449 z_stream stream;
450 char hdr[64], type[20];
451 const unsigned char *data;
452 unsigned char cmd;
453 int i;
455 map = map_sha1_file(sha1, &mapsize);
456 if (!map)
457 return -1;
458 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
459 status = -1;
460 if (ret < Z_OK || parse_sha1_header(hdr, type, &size) < 0)
461 goto out;
462 if (strcmp(type, "delta")) {
463 *sizep = size;
464 status = 0;
465 goto out;
468 /* We are dealing with a delta object. Inflated, the first
469 * 20 bytes hold the base object SHA1, and delta data follows
470 * immediately after it.
472 * The initial part of the delta starts at delta_data_head +
473 * 20. Borrow code from patch-delta to read the result size.
475 data = hdr + strlen(hdr) + 1 + 20;
477 /* Skip over the source size; we are not interested in
478 * it and we cannot verify it because we do not want
479 * to read the base object.
481 cmd = *data++;
482 while (cmd) {
483 if (cmd & 1)
484 data++;
485 cmd >>= 1;
487 /* Read the result size */
488 size = i = 0;
489 cmd = *data++;
490 while (cmd) {
491 if (cmd & 1)
492 size |= *data++ << i;
493 i += 8;
494 cmd >>= 1;
496 *sizep = size;
497 status = 0;
498 out:
499 inflateEnd(&stream);
500 munmap(map, mapsize);
501 return status;
504 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
506 unsigned long mapsize;
507 void *map, *buf;
509 map = map_sha1_file(sha1, &mapsize);
510 if (map) {
511 buf = unpack_sha1_file(map, mapsize, type, size);
512 munmap(map, mapsize);
513 if (buf && !strcmp(type, "delta")) {
514 void *ref = NULL, *delta = buf;
515 unsigned long ref_size, delta_size = *size;
516 buf = NULL;
517 if (delta_size > 20)
518 ref = read_sha1_file(delta, type, &ref_size);
519 if (ref)
520 buf = patch_delta(ref, ref_size,
521 delta+20, delta_size-20,
522 size);
523 free(delta);
524 free(ref);
526 return buf;
528 return NULL;
531 void *read_object_with_reference(const unsigned char *sha1,
532 const char *required_type,
533 unsigned long *size,
534 unsigned char *actual_sha1_return)
536 char type[20];
537 void *buffer;
538 unsigned long isize;
539 unsigned char actual_sha1[20];
541 memcpy(actual_sha1, sha1, 20);
542 while (1) {
543 int ref_length = -1;
544 const char *ref_type = NULL;
546 buffer = read_sha1_file(actual_sha1, type, &isize);
547 if (!buffer)
548 return NULL;
549 if (!strcmp(type, required_type)) {
550 *size = isize;
551 if (actual_sha1_return)
552 memcpy(actual_sha1_return, actual_sha1, 20);
553 return buffer;
555 /* Handle references */
556 else if (!strcmp(type, "commit"))
557 ref_type = "tree ";
558 else if (!strcmp(type, "tag"))
559 ref_type = "object ";
560 else {
561 free(buffer);
562 return NULL;
564 ref_length = strlen(ref_type);
566 if (memcmp(buffer, ref_type, ref_length) ||
567 get_sha1_hex(buffer + ref_length, actual_sha1)) {
568 free(buffer);
569 return NULL;
571 /* Now we have the ID of the referred-to object in
572 * actual_sha1. Check again. */
576 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
578 int size;
579 unsigned char *compressed;
580 z_stream stream;
581 unsigned char sha1[20];
582 SHA_CTX c;
583 char *filename;
584 static char tmpfile[PATH_MAX];
585 unsigned char hdr[50];
586 int fd, hdrlen, ret;
588 /* Generate the header */
589 hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
591 /* Sha1.. */
592 SHA1_Init(&c);
593 SHA1_Update(&c, hdr, hdrlen);
594 SHA1_Update(&c, buf, len);
595 SHA1_Final(sha1, &c);
597 if (returnsha1)
598 memcpy(returnsha1, sha1, 20);
600 filename = sha1_file_name(sha1);
601 fd = open(filename, O_RDONLY);
602 if (fd >= 0) {
604 * FIXME!!! We might do collision checking here, but we'd
605 * need to uncompress the old file and check it. Later.
607 close(fd);
608 return 0;
611 if (errno != ENOENT) {
612 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
613 return -1;
616 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
618 fd = mkstemp(tmpfile);
619 if (fd < 0) {
620 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
621 return -1;
624 /* Set it up */
625 memset(&stream, 0, sizeof(stream));
626 deflateInit(&stream, Z_BEST_COMPRESSION);
627 size = deflateBound(&stream, len+hdrlen);
628 compressed = xmalloc(size);
630 /* Compress it */
631 stream.next_out = compressed;
632 stream.avail_out = size;
634 /* First header.. */
635 stream.next_in = hdr;
636 stream.avail_in = hdrlen;
637 while (deflate(&stream, 0) == Z_OK)
638 /* nothing */;
640 /* Then the data itself.. */
641 stream.next_in = buf;
642 stream.avail_in = len;
643 while (deflate(&stream, Z_FINISH) == Z_OK)
644 /* nothing */;
645 deflateEnd(&stream);
646 size = stream.total_out;
648 if (write(fd, compressed, size) != size)
649 die("unable to write file");
650 fchmod(fd, 0444);
651 close(fd);
652 free(compressed);
654 ret = link(tmpfile, filename);
655 if (ret < 0) {
656 ret = errno;
659 * Coda hack - coda doesn't like cross-directory links,
660 * so we fall back to a rename, which will mean that it
661 * won't be able to check collisions, but that's not a
662 * big deal.
664 * When this succeeds, we just return 0. We have nothing
665 * left to unlink.
667 if (ret == EXDEV && !rename(tmpfile, filename))
668 return 0;
670 unlink(tmpfile);
671 if (ret) {
672 if (ret != EEXIST) {
673 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
674 return -1;
676 /* FIXME!!! Collision check here ? */
679 return 0;
682 int write_sha1_from_fd(const unsigned char *sha1, int fd)
684 char *filename = sha1_file_name(sha1);
686 int local;
687 z_stream stream;
688 unsigned char real_sha1[20];
689 unsigned char buf[4096];
690 unsigned char discard[4096];
691 int ret;
692 SHA_CTX c;
694 local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
696 if (local < 0)
697 return error("Couldn't open %s\n", filename);
699 memset(&stream, 0, sizeof(stream));
701 inflateInit(&stream);
703 SHA1_Init(&c);
705 do {
706 ssize_t size;
707 size = read(fd, buf, 4096);
708 if (size <= 0) {
709 close(local);
710 unlink(filename);
711 if (!size)
712 return error("Connection closed?");
713 perror("Reading from connection");
714 return -1;
716 write(local, buf, size);
717 stream.avail_in = size;
718 stream.next_in = buf;
719 do {
720 stream.next_out = discard;
721 stream.avail_out = sizeof(discard);
722 ret = inflate(&stream, Z_SYNC_FLUSH);
723 SHA1_Update(&c, discard, sizeof(discard) -
724 stream.avail_out);
725 } while (stream.avail_in && ret == Z_OK);
727 } while (ret == Z_OK);
728 inflateEnd(&stream);
730 close(local);
731 SHA1_Final(real_sha1, &c);
732 if (ret != Z_STREAM_END) {
733 unlink(filename);
734 return error("File %s corrupted", sha1_to_hex(sha1));
736 if (memcmp(sha1, real_sha1, 20)) {
737 unlink(filename);
738 return error("File %s has bad hash\n", sha1_to_hex(sha1));
741 return 0;
744 int has_sha1_file(const unsigned char *sha1)
746 struct stat st;
747 return !!find_sha1_file(sha1, &st);
750 int index_fd(unsigned char *sha1, int fd, struct stat *st)
752 unsigned long size = st->st_size;
753 void *buf;
754 int ret;
756 buf = "";
757 if (size)
758 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
759 close(fd);
760 if ((int)(long)buf == -1)
761 return -1;
763 ret = write_sha1_file(buf, size, "blob", sha1);
764 if (size)
765 munmap(buf, size);
766 return ret;