Fix llist_sorted_difference_inplace in git-pack-redundant
[git/git-svn.git] / sha1_file.c
blob82a01887c261676eb39fe72b108bfa56041d71f5
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 <sys/types.h>
10 #include <dirent.h>
11 #include "cache.h"
12 #include "delta.h"
13 #include "pack.h"
15 #ifndef O_NOATIME
16 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
17 #define O_NOATIME 01000000
18 #else
19 #define O_NOATIME 0
20 #endif
21 #endif
23 const unsigned char null_sha1[20] = { 0, };
25 static unsigned int sha1_file_open_flag = O_NOATIME;
27 static unsigned hexval(char c)
29 if (c >= '0' && c <= '9')
30 return c - '0';
31 if (c >= 'a' && c <= 'f')
32 return c - 'a' + 10;
33 if (c >= 'A' && c <= 'F')
34 return c - 'A' + 10;
35 return ~0;
38 int get_sha1_hex(const char *hex, unsigned char *sha1)
40 int i;
41 for (i = 0; i < 20; i++) {
42 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
43 if (val & ~0xff)
44 return -1;
45 *sha1++ = val;
46 hex += 2;
48 return 0;
51 int safe_create_leading_directories(char *path)
53 char *pos = path;
54 if (*pos == '/')
55 pos++;
57 while (pos) {
58 pos = strchr(pos, '/');
59 if (!pos)
60 break;
61 *pos = 0;
62 if (mkdir(path, 0777) < 0)
63 if (errno != EEXIST) {
64 *pos = '/';
65 return -1;
67 *pos++ = '/';
69 return 0;
72 char * sha1_to_hex(const unsigned char *sha1)
74 static char buffer[50];
75 static const char hex[] = "0123456789abcdef";
76 char *buf = buffer;
77 int i;
79 for (i = 0; i < 20; i++) {
80 unsigned int val = *sha1++;
81 *buf++ = hex[val >> 4];
82 *buf++ = hex[val & 0xf];
84 return buffer;
87 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
89 int i;
90 for (i = 0; i < 20; i++) {
91 static char hex[] = "0123456789abcdef";
92 unsigned int val = sha1[i];
93 char *pos = pathbuf + i*2 + (i > 0);
94 *pos++ = hex[val >> 4];
95 *pos = hex[val & 0xf];
100 * NOTE! This returns a statically allocated buffer, so you have to be
101 * careful about using it. Do a "strdup()" if you need to save the
102 * filename.
104 * Also note that this returns the location for creating. Reading
105 * SHA1 file can happen from any alternate directory listed in the
106 * DB_ENVIRONMENT environment variable if it is not found in
107 * the primary object database.
109 char *sha1_file_name(const unsigned char *sha1)
111 static char *name, *base;
113 if (!base) {
114 const char *sha1_file_directory = get_object_directory();
115 int len = strlen(sha1_file_directory);
116 base = xmalloc(len + 60);
117 memcpy(base, sha1_file_directory, len);
118 memset(base+len, 0, 60);
119 base[len] = '/';
120 base[len+3] = '/';
121 name = base + len + 1;
123 fill_sha1_path(name, sha1);
124 return base;
127 char *sha1_pack_name(const unsigned char *sha1)
129 static const char hex[] = "0123456789abcdef";
130 static char *name, *base, *buf;
131 int i;
133 if (!base) {
134 const char *sha1_file_directory = get_object_directory();
135 int len = strlen(sha1_file_directory);
136 base = xmalloc(len + 60);
137 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
138 name = base + len + 11;
141 buf = name;
143 for (i = 0; i < 20; i++) {
144 unsigned int val = *sha1++;
145 *buf++ = hex[val >> 4];
146 *buf++ = hex[val & 0xf];
149 return base;
152 char *sha1_pack_index_name(const unsigned char *sha1)
154 static const char hex[] = "0123456789abcdef";
155 static char *name, *base, *buf;
156 int i;
158 if (!base) {
159 const char *sha1_file_directory = get_object_directory();
160 int len = strlen(sha1_file_directory);
161 base = xmalloc(len + 60);
162 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
163 name = base + len + 11;
166 buf = name;
168 for (i = 0; i < 20; i++) {
169 unsigned int val = *sha1++;
170 *buf++ = hex[val >> 4];
171 *buf++ = hex[val & 0xf];
174 return base;
177 struct alternate_object_database *alt_odb_list;
178 static struct alternate_object_database **alt_odb_tail;
181 * Prepare alternate object database registry.
183 * The variable alt_odb_list points at the list of struct
184 * alternate_object_database. The elements on this list come from
185 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
186 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
187 * whose contents is exactly in the same format as that environment
188 * variable. Its base points at a statically allocated buffer that
189 * contains "/the/directory/corresponding/to/.git/objects/...", while
190 * its name points just after the slash at the end of ".git/objects/"
191 * in the example above, and has enough space to hold 40-byte hex
192 * SHA1, an extra slash for the first level indirection, and the
193 * terminating NUL.
195 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
196 const char *relative_base)
198 const char *cp, *last;
199 struct alternate_object_database *ent;
200 int base_len = -1;
202 last = alt;
203 while (last < ep) {
204 cp = last;
205 if (cp < ep && *cp == '#') {
206 while (cp < ep && *cp != sep)
207 cp++;
208 last = cp + 1;
209 continue;
211 for ( ; cp < ep && *cp != sep; cp++)
213 if (last != cp) {
214 /* 43 = 40-byte + 2 '/' + terminating NUL */
215 int pfxlen = cp - last;
216 int entlen = pfxlen + 43;
218 if (*last != '/' && relative_base) {
219 /* Relative alt-odb */
220 if (base_len < 0)
221 base_len = strlen(relative_base) + 1;
222 entlen += base_len;
223 pfxlen += base_len;
225 ent = xmalloc(sizeof(*ent) + entlen);
226 *alt_odb_tail = ent;
227 alt_odb_tail = &(ent->next);
228 ent->next = NULL;
229 if (*last != '/' && relative_base) {
230 memcpy(ent->base, relative_base, base_len - 1);
231 ent->base[base_len - 1] = '/';
232 memcpy(ent->base + base_len,
233 last, cp - last);
235 else
236 memcpy(ent->base, last, pfxlen);
237 ent->name = ent->base + pfxlen + 1;
238 ent->base[pfxlen] = ent->base[pfxlen + 3] = '/';
239 ent->base[entlen-1] = 0;
241 while (cp < ep && *cp == sep)
242 cp++;
243 last = cp;
247 void prepare_alt_odb(void)
249 char path[PATH_MAX];
250 char *map;
251 int fd;
252 struct stat st;
253 char *alt;
255 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
256 if (!alt) alt = "";
258 if (alt_odb_tail)
259 return;
260 alt_odb_tail = &alt_odb_list;
261 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL);
263 sprintf(path, "%s/info/alternates", get_object_directory());
264 fd = open(path, O_RDONLY);
265 if (fd < 0)
266 return;
267 if (fstat(fd, &st) || (st.st_size == 0)) {
268 close(fd);
269 return;
271 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
272 close(fd);
273 if (map == MAP_FAILED)
274 return;
276 link_alt_odb_entries(map, map + st.st_size, '\n',
277 get_object_directory());
278 munmap(map, st.st_size);
281 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
283 char *name = sha1_file_name(sha1);
284 struct alternate_object_database *alt;
286 if (!stat(name, st))
287 return name;
288 prepare_alt_odb();
289 for (alt = alt_odb_list; alt; alt = alt->next) {
290 name = alt->name;
291 fill_sha1_path(name, sha1);
292 if (!stat(alt->base, st))
293 return alt->base;
295 return NULL;
298 #define PACK_MAX_SZ (1<<26)
299 static int pack_used_ctr;
300 static unsigned long pack_mapped;
301 struct packed_git *packed_git;
303 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
304 void **idx_map_)
306 void *idx_map;
307 unsigned int *index;
308 unsigned long idx_size;
309 int nr, i;
310 int fd = open(path, O_RDONLY);
311 struct stat st;
312 if (fd < 0)
313 return -1;
314 if (fstat(fd, &st)) {
315 close(fd);
316 return -1;
318 idx_size = st.st_size;
319 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
320 close(fd);
321 if (idx_map == MAP_FAILED)
322 return -1;
324 index = idx_map;
325 *idx_map_ = idx_map;
326 *idx_size_ = idx_size;
328 /* check index map */
329 if (idx_size < 4*256 + 20 + 20)
330 return error("index file too small");
331 nr = 0;
332 for (i = 0; i < 256; i++) {
333 unsigned int n = ntohl(index[i]);
334 if (n < nr)
335 return error("non-monotonic index");
336 nr = n;
340 * Total size:
341 * - 256 index entries 4 bytes each
342 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
343 * - 20-byte SHA1 of the packfile
344 * - 20-byte SHA1 file checksum
346 if (idx_size != 4*256 + nr * 24 + 20 + 20)
347 return error("wrong index file size");
349 return 0;
352 static int unuse_one_packed_git(void)
354 struct packed_git *p, *lru = NULL;
356 for (p = packed_git; p; p = p->next) {
357 if (p->pack_use_cnt || !p->pack_base)
358 continue;
359 if (!lru || p->pack_last_used < lru->pack_last_used)
360 lru = p;
362 if (!lru)
363 return 0;
364 munmap(lru->pack_base, lru->pack_size);
365 lru->pack_base = NULL;
366 return 1;
369 void unuse_packed_git(struct packed_git *p)
371 p->pack_use_cnt--;
374 int use_packed_git(struct packed_git *p)
376 if (!p->pack_size) {
377 struct stat st;
378 // We created the struct before we had the pack
379 stat(p->pack_name, &st);
380 if (!S_ISREG(st.st_mode))
381 die("packfile %s not a regular file", p->pack_name);
382 p->pack_size = st.st_size;
384 if (!p->pack_base) {
385 int fd;
386 struct stat st;
387 void *map;
389 pack_mapped += p->pack_size;
390 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
391 ; /* nothing */
392 fd = open(p->pack_name, O_RDONLY);
393 if (fd < 0)
394 die("packfile %s cannot be opened", p->pack_name);
395 if (fstat(fd, &st)) {
396 close(fd);
397 die("packfile %s cannot be opened", p->pack_name);
399 if (st.st_size != p->pack_size)
400 die("packfile %s size mismatch.", p->pack_name);
401 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
402 close(fd);
403 if (map == MAP_FAILED)
404 die("packfile %s cannot be mapped.", p->pack_name);
405 p->pack_base = map;
407 /* Check if the pack file matches with the index file.
408 * this is cheap.
410 if (memcmp((char*)(p->index_base) + p->index_size - 40,
411 p->pack_base + p->pack_size - 20, 20)) {
413 die("packfile %s does not match index.", p->pack_name);
416 p->pack_last_used = pack_used_ctr++;
417 p->pack_use_cnt++;
418 return 0;
421 struct packed_git *add_packed_git(char *path, int path_len, int local)
423 struct stat st;
424 struct packed_git *p;
425 unsigned long idx_size;
426 void *idx_map;
427 unsigned char sha1[20];
429 if (check_packed_git_idx(path, &idx_size, &idx_map))
430 return NULL;
432 /* do we have a corresponding .pack file? */
433 strcpy(path + path_len - 4, ".pack");
434 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
435 munmap(idx_map, idx_size);
436 return NULL;
438 /* ok, it looks sane as far as we can check without
439 * actually mapping the pack file.
441 p = xmalloc(sizeof(*p) + path_len + 2);
442 strcpy(p->pack_name, path);
443 p->index_size = idx_size;
444 p->pack_size = st.st_size;
445 p->index_base = idx_map;
446 p->next = NULL;
447 p->pack_base = NULL;
448 p->pack_last_used = 0;
449 p->pack_use_cnt = 0;
450 p->pack_local = local;
451 if (!get_sha1_hex(path + path_len - 40 - 4, sha1))
452 memcpy(p->sha1, sha1, 20);
453 return p;
456 struct packed_git *parse_pack_index(unsigned char *sha1)
458 char *path = sha1_pack_index_name(sha1);
459 return parse_pack_index_file(sha1, path);
462 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
464 struct packed_git *p;
465 unsigned long idx_size;
466 void *idx_map;
467 char *path;
469 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
470 return NULL;
472 path = sha1_pack_name(sha1);
474 p = xmalloc(sizeof(*p) + strlen(path) + 2);
475 strcpy(p->pack_name, path);
476 p->index_size = idx_size;
477 p->pack_size = 0;
478 p->index_base = idx_map;
479 p->next = NULL;
480 p->pack_base = NULL;
481 p->pack_last_used = 0;
482 p->pack_use_cnt = 0;
483 memcpy(p->sha1, sha1, 20);
484 return p;
487 void install_packed_git(struct packed_git *pack)
489 pack->next = packed_git;
490 packed_git = pack;
493 static void prepare_packed_git_one(char *objdir, int local)
495 char path[PATH_MAX];
496 int len;
497 DIR *dir;
498 struct dirent *de;
500 sprintf(path, "%s/pack", objdir);
501 len = strlen(path);
502 dir = opendir(path);
503 if (!dir)
504 return;
505 path[len++] = '/';
506 while ((de = readdir(dir)) != NULL) {
507 int namelen = strlen(de->d_name);
508 struct packed_git *p;
510 if (strcmp(de->d_name + namelen - 4, ".idx"))
511 continue;
513 /* we have .idx. Is it a file we can map? */
514 strcpy(path + len, de->d_name);
515 p = add_packed_git(path, len + namelen, local);
516 if (!p)
517 continue;
518 p->next = packed_git;
519 packed_git = p;
521 closedir(dir);
524 void prepare_packed_git(void)
526 static int run_once = 0;
527 struct alternate_object_database *alt;
529 if (run_once)
530 return;
531 prepare_packed_git_one(get_object_directory(), 1);
532 prepare_alt_odb();
533 for (alt = alt_odb_list; alt; alt = alt->next) {
534 alt->name[0] = 0;
535 prepare_packed_git_one(alt->base, 0);
537 run_once = 1;
540 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
542 char header[100];
543 unsigned char real_sha1[20];
544 SHA_CTX c;
546 SHA1_Init(&c);
547 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
548 SHA1_Update(&c, map, size);
549 SHA1_Final(real_sha1, &c);
550 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
553 static void *map_sha1_file_internal(const unsigned char *sha1,
554 unsigned long *size)
556 struct stat st;
557 void *map;
558 int fd;
559 char *filename = find_sha1_file(sha1, &st);
561 if (!filename) {
562 return NULL;
565 fd = open(filename, O_RDONLY | sha1_file_open_flag);
566 if (fd < 0) {
567 /* See if it works without O_NOATIME */
568 switch (sha1_file_open_flag) {
569 default:
570 fd = open(filename, O_RDONLY);
571 if (fd >= 0)
572 break;
573 /* Fallthrough */
574 case 0:
575 return NULL;
578 /* If it failed once, it will probably fail again.
579 * Stop using O_NOATIME
581 sha1_file_open_flag = 0;
583 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
584 close(fd);
585 if (map == MAP_FAILED)
586 return NULL;
587 *size = st.st_size;
588 return map;
591 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
593 /* Get the data stream */
594 memset(stream, 0, sizeof(*stream));
595 stream->next_in = map;
596 stream->avail_in = mapsize;
597 stream->next_out = buffer;
598 stream->avail_out = size;
600 inflateInit(stream);
601 return inflate(stream, 0);
604 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
606 int bytes = strlen(buffer) + 1;
607 unsigned char *buf = xmalloc(1+size);
609 memcpy(buf, buffer + bytes, stream->total_out - bytes);
610 bytes = stream->total_out - bytes;
611 if (bytes < size) {
612 stream->next_out = buf + bytes;
613 stream->avail_out = size - bytes;
614 while (inflate(stream, Z_FINISH) == Z_OK)
615 /* nothing */;
617 buf[size] = 0;
618 inflateEnd(stream);
619 return buf;
623 * We used to just use "sscanf()", but that's actually way
624 * too permissive for what we want to check. So do an anal
625 * object header parse by hand.
627 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
629 int i;
630 unsigned long size;
633 * The type can be at most ten bytes (including the
634 * terminating '\0' that we add), and is followed by
635 * a space.
637 i = 10;
638 for (;;) {
639 char c = *hdr++;
640 if (c == ' ')
641 break;
642 if (!--i)
643 return -1;
644 *type++ = c;
646 *type = 0;
649 * The length must follow immediately, and be in canonical
650 * decimal format (ie "010" is not valid).
652 size = *hdr++ - '0';
653 if (size > 9)
654 return -1;
655 if (size) {
656 for (;;) {
657 unsigned long c = *hdr - '0';
658 if (c > 9)
659 break;
660 hdr++;
661 size = size * 10 + c;
664 *sizep = size;
667 * The length must be followed by a zero byte
669 return *hdr ? -1 : 0;
672 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
674 int ret;
675 z_stream stream;
676 char hdr[8192];
678 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
679 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
680 return NULL;
682 return unpack_sha1_rest(&stream, hdr, *size);
685 /* forward declaration for a mutually recursive function */
686 static int packed_object_info(struct pack_entry *entry,
687 char *type, unsigned long *sizep);
689 static int packed_delta_info(unsigned char *base_sha1,
690 unsigned long delta_size,
691 unsigned long left,
692 char *type,
693 unsigned long *sizep,
694 struct packed_git *p)
696 struct pack_entry base_ent;
698 if (left < 20)
699 die("truncated pack file");
701 /* The base entry _must_ be in the same pack */
702 if (!find_pack_entry_one(base_sha1, &base_ent, p))
703 die("failed to find delta-pack base object %s",
704 sha1_to_hex(base_sha1));
706 /* We choose to only get the type of the base object and
707 * ignore potentially corrupt pack file that expects the delta
708 * based on a base with a wrong size. This saves tons of
709 * inflate() calls.
712 if (packed_object_info(&base_ent, type, NULL))
713 die("cannot get info for delta-pack base");
715 if (sizep) {
716 const unsigned char *data;
717 unsigned char delta_head[64];
718 unsigned long result_size;
719 z_stream stream;
720 int st;
722 memset(&stream, 0, sizeof(stream));
724 data = stream.next_in = base_sha1 + 20;
725 stream.avail_in = left - 20;
726 stream.next_out = delta_head;
727 stream.avail_out = sizeof(delta_head);
729 inflateInit(&stream);
730 st = inflate(&stream, Z_FINISH);
731 inflateEnd(&stream);
732 if ((st != Z_STREAM_END) &&
733 stream.total_out != sizeof(delta_head))
734 die("delta data unpack-initial failed");
736 /* Examine the initial part of the delta to figure out
737 * the result size.
739 data = delta_head;
740 get_delta_hdr_size(&data); /* ignore base size */
742 /* Read the result size */
743 result_size = get_delta_hdr_size(&data);
744 *sizep = result_size;
746 return 0;
749 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
750 enum object_type *type, unsigned long *sizep)
752 unsigned shift;
753 unsigned char *pack, c;
754 unsigned long size;
756 if (offset >= p->pack_size)
757 die("object offset outside of pack file");
759 pack = p->pack_base + offset;
760 c = *pack++;
761 offset++;
762 *type = (c >> 4) & 7;
763 size = c & 15;
764 shift = 4;
765 while (c & 0x80) {
766 if (offset >= p->pack_size)
767 die("object offset outside of pack file");
768 c = *pack++;
769 offset++;
770 size += (c & 0x7f) << shift;
771 shift += 7;
773 *sizep = size;
774 return offset;
777 void packed_object_info_detail(struct pack_entry *e,
778 char *type,
779 unsigned long *size,
780 unsigned long *store_size,
781 int *delta_chain_length,
782 unsigned char *base_sha1)
784 struct packed_git *p = e->p;
785 unsigned long offset, left;
786 unsigned char *pack;
787 enum object_type kind;
789 offset = unpack_object_header(p, e->offset, &kind, size);
790 pack = p->pack_base + offset;
791 left = p->pack_size - offset;
792 if (kind != OBJ_DELTA)
793 *delta_chain_length = 0;
794 else {
795 int chain_length = 0;
796 memcpy(base_sha1, pack, 20);
797 do {
798 struct pack_entry base_ent;
799 unsigned long junk;
801 find_pack_entry_one(pack, &base_ent, p);
802 offset = unpack_object_header(p, base_ent.offset,
803 &kind, &junk);
804 pack = p->pack_base + offset;
805 chain_length++;
806 } while (kind == OBJ_DELTA);
807 *delta_chain_length = chain_length;
809 switch (kind) {
810 case OBJ_COMMIT:
811 strcpy(type, "commit");
812 break;
813 case OBJ_TREE:
814 strcpy(type, "tree");
815 break;
816 case OBJ_BLOB:
817 strcpy(type, "blob");
818 break;
819 case OBJ_TAG:
820 strcpy(type, "tag");
821 break;
822 default:
823 die("corrupted pack file %s containing object of kind %d",
824 p->pack_name, kind);
826 *store_size = 0; /* notyet */
829 static int packed_object_info(struct pack_entry *entry,
830 char *type, unsigned long *sizep)
832 struct packed_git *p = entry->p;
833 unsigned long offset, size, left;
834 unsigned char *pack;
835 enum object_type kind;
836 int retval;
838 if (use_packed_git(p))
839 die("cannot map packed file");
841 offset = unpack_object_header(p, entry->offset, &kind, &size);
842 pack = p->pack_base + offset;
843 left = p->pack_size - offset;
845 switch (kind) {
846 case OBJ_DELTA:
847 retval = packed_delta_info(pack, size, left, type, sizep, p);
848 unuse_packed_git(p);
849 return retval;
850 case OBJ_COMMIT:
851 strcpy(type, "commit");
852 break;
853 case OBJ_TREE:
854 strcpy(type, "tree");
855 break;
856 case OBJ_BLOB:
857 strcpy(type, "blob");
858 break;
859 case OBJ_TAG:
860 strcpy(type, "tag");
861 break;
862 default:
863 die("corrupted pack file %s containing object of kind %d",
864 p->pack_name, kind);
866 if (sizep)
867 *sizep = size;
868 unuse_packed_git(p);
869 return 0;
872 /* forward declaration for a mutually recursive function */
873 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
875 static void *unpack_delta_entry(unsigned char *base_sha1,
876 unsigned long delta_size,
877 unsigned long left,
878 char *type,
879 unsigned long *sizep,
880 struct packed_git *p)
882 struct pack_entry base_ent;
883 void *data, *delta_data, *result, *base;
884 unsigned long data_size, result_size, base_size;
885 z_stream stream;
886 int st;
888 if (left < 20)
889 die("truncated pack file");
890 data = base_sha1 + 20;
891 data_size = left - 20;
892 delta_data = xmalloc(delta_size);
894 memset(&stream, 0, sizeof(stream));
896 stream.next_in = data;
897 stream.avail_in = data_size;
898 stream.next_out = delta_data;
899 stream.avail_out = delta_size;
901 inflateInit(&stream);
902 st = inflate(&stream, Z_FINISH);
903 inflateEnd(&stream);
904 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
905 die("delta data unpack failed");
907 /* The base entry _must_ be in the same pack */
908 if (!find_pack_entry_one(base_sha1, &base_ent, p))
909 die("failed to find delta-pack base object %s",
910 sha1_to_hex(base_sha1));
911 base = unpack_entry_gently(&base_ent, type, &base_size);
912 if (!base)
913 die("failed to read delta-pack base object %s",
914 sha1_to_hex(base_sha1));
915 result = patch_delta(base, base_size,
916 delta_data, delta_size,
917 &result_size);
918 if (!result)
919 die("failed to apply delta");
920 free(delta_data);
921 free(base);
922 *sizep = result_size;
923 return result;
926 static void *unpack_non_delta_entry(unsigned char *data,
927 unsigned long size,
928 unsigned long left)
930 int st;
931 z_stream stream;
932 unsigned char *buffer;
934 buffer = xmalloc(size + 1);
935 buffer[size] = 0;
936 memset(&stream, 0, sizeof(stream));
937 stream.next_in = data;
938 stream.avail_in = left;
939 stream.next_out = buffer;
940 stream.avail_out = size;
942 inflateInit(&stream);
943 st = inflate(&stream, Z_FINISH);
944 inflateEnd(&stream);
945 if ((st != Z_STREAM_END) || stream.total_out != size) {
946 free(buffer);
947 return NULL;
950 return buffer;
953 static void *unpack_entry(struct pack_entry *entry,
954 char *type, unsigned long *sizep)
956 struct packed_git *p = entry->p;
957 void *retval;
959 if (use_packed_git(p))
960 die("cannot map packed file");
961 retval = unpack_entry_gently(entry, type, sizep);
962 unuse_packed_git(p);
963 if (!retval)
964 die("corrupted pack file %s", p->pack_name);
965 return retval;
968 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
969 void *unpack_entry_gently(struct pack_entry *entry,
970 char *type, unsigned long *sizep)
972 struct packed_git *p = entry->p;
973 unsigned long offset, size, left;
974 unsigned char *pack;
975 enum object_type kind;
976 void *retval;
978 offset = unpack_object_header(p, entry->offset, &kind, &size);
979 pack = p->pack_base + offset;
980 left = p->pack_size - offset;
981 switch (kind) {
982 case OBJ_DELTA:
983 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
984 return retval;
985 case OBJ_COMMIT:
986 strcpy(type, "commit");
987 break;
988 case OBJ_TREE:
989 strcpy(type, "tree");
990 break;
991 case OBJ_BLOB:
992 strcpy(type, "blob");
993 break;
994 case OBJ_TAG:
995 strcpy(type, "tag");
996 break;
997 default:
998 return NULL;
1000 *sizep = size;
1001 retval = unpack_non_delta_entry(pack, size, left);
1002 return retval;
1005 int num_packed_objects(const struct packed_git *p)
1007 /* See check_packed_git_idx() */
1008 return (p->index_size - 20 - 20 - 4*256) / 24;
1011 int nth_packed_object_sha1(const struct packed_git *p, int n,
1012 unsigned char* sha1)
1014 void *index = p->index_base + 256;
1015 if (n < 0 || num_packed_objects(p) <= n)
1016 return -1;
1017 memcpy(sha1, (index + 24 * n + 4), 20);
1018 return 0;
1021 int find_pack_entry_one(const unsigned char *sha1,
1022 struct pack_entry *e, struct packed_git *p)
1024 unsigned int *level1_ofs = p->index_base;
1025 int hi = ntohl(level1_ofs[*sha1]);
1026 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1027 void *index = p->index_base + 256;
1029 do {
1030 int mi = (lo + hi) / 2;
1031 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1032 if (!cmp) {
1033 e->offset = ntohl(*((int*)(index + 24 * mi)));
1034 memcpy(e->sha1, sha1, 20);
1035 e->p = p;
1036 return 1;
1038 if (cmp > 0)
1039 hi = mi;
1040 else
1041 lo = mi+1;
1042 } while (lo < hi);
1043 return 0;
1046 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1048 struct packed_git *p;
1049 prepare_packed_git();
1051 for (p = packed_git; p; p = p->next) {
1052 if (find_pack_entry_one(sha1, e, p))
1053 return 1;
1055 return 0;
1058 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1059 struct packed_git *packs)
1061 struct packed_git *p;
1062 struct pack_entry e;
1064 for (p = packs; p; p = p->next) {
1065 if (find_pack_entry_one(sha1, &e, p))
1066 return p;
1068 return NULL;
1072 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1074 int status;
1075 unsigned long mapsize, size;
1076 void *map;
1077 z_stream stream;
1078 char hdr[128];
1080 map = map_sha1_file_internal(sha1, &mapsize);
1081 if (!map) {
1082 struct pack_entry e;
1084 if (!find_pack_entry(sha1, &e))
1085 return error("unable to find %s", sha1_to_hex(sha1));
1086 return packed_object_info(&e, type, sizep);
1088 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1089 status = error("unable to unpack %s header",
1090 sha1_to_hex(sha1));
1091 if (parse_sha1_header(hdr, type, &size) < 0)
1092 status = error("unable to parse %s header", sha1_to_hex(sha1));
1093 else {
1094 status = 0;
1095 if (sizep)
1096 *sizep = size;
1098 inflateEnd(&stream);
1099 munmap(map, mapsize);
1100 return status;
1103 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1105 struct pack_entry e;
1107 if (!find_pack_entry(sha1, &e)) {
1108 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1109 return NULL;
1111 return unpack_entry(&e, type, size);
1114 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1116 unsigned long mapsize;
1117 void *map, *buf;
1118 struct pack_entry e;
1120 if (find_pack_entry(sha1, &e))
1121 return read_packed_sha1(sha1, type, size);
1122 map = map_sha1_file_internal(sha1, &mapsize);
1123 if (map) {
1124 buf = unpack_sha1_file(map, mapsize, type, size);
1125 munmap(map, mapsize);
1126 return buf;
1128 return NULL;
1131 void *read_object_with_reference(const unsigned char *sha1,
1132 const char *required_type,
1133 unsigned long *size,
1134 unsigned char *actual_sha1_return)
1136 char type[20];
1137 void *buffer;
1138 unsigned long isize;
1139 unsigned char actual_sha1[20];
1141 memcpy(actual_sha1, sha1, 20);
1142 while (1) {
1143 int ref_length = -1;
1144 const char *ref_type = NULL;
1146 buffer = read_sha1_file(actual_sha1, type, &isize);
1147 if (!buffer)
1148 return NULL;
1149 if (!strcmp(type, required_type)) {
1150 *size = isize;
1151 if (actual_sha1_return)
1152 memcpy(actual_sha1_return, actual_sha1, 20);
1153 return buffer;
1155 /* Handle references */
1156 else if (!strcmp(type, "commit"))
1157 ref_type = "tree ";
1158 else if (!strcmp(type, "tag"))
1159 ref_type = "object ";
1160 else {
1161 free(buffer);
1162 return NULL;
1164 ref_length = strlen(ref_type);
1166 if (memcmp(buffer, ref_type, ref_length) ||
1167 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1168 free(buffer);
1169 return NULL;
1171 free(buffer);
1172 /* Now we have the ID of the referred-to object in
1173 * actual_sha1. Check again. */
1177 char *write_sha1_file_prepare(void *buf,
1178 unsigned long len,
1179 const char *type,
1180 unsigned char *sha1,
1181 unsigned char *hdr,
1182 int *hdrlen)
1184 SHA_CTX c;
1186 /* Generate the header */
1187 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1189 /* Sha1.. */
1190 SHA1_Init(&c);
1191 SHA1_Update(&c, hdr, *hdrlen);
1192 SHA1_Update(&c, buf, len);
1193 SHA1_Final(sha1, &c);
1195 return sha1_file_name(sha1);
1199 * Link the tempfile to the final place, possibly creating the
1200 * last directory level as you do so.
1202 * Returns the errno on failure, 0 on success.
1204 static int link_temp_to_file(const char *tmpfile, char *filename)
1206 int ret;
1208 if (!link(tmpfile, filename))
1209 return 0;
1212 * Try to mkdir the last path component if that failed
1213 * with an ENOENT.
1215 * Re-try the "link()" regardless of whether the mkdir
1216 * succeeds, since a race might mean that somebody
1217 * else succeeded.
1219 ret = errno;
1220 if (ret == ENOENT) {
1221 char *dir = strrchr(filename, '/');
1222 if (dir) {
1223 *dir = 0;
1224 mkdir(filename, 0777);
1225 *dir = '/';
1226 if (!link(tmpfile, filename))
1227 return 0;
1228 ret = errno;
1231 return ret;
1235 * Move the just written object into its final resting place
1237 int move_temp_to_file(const char *tmpfile, char *filename)
1239 int ret = link_temp_to_file(tmpfile, filename);
1242 * Coda hack - coda doesn't like cross-directory links,
1243 * so we fall back to a rename, which will mean that it
1244 * won't be able to check collisions, but that's not a
1245 * big deal.
1247 * The same holds for FAT formatted media.
1249 * When this succeeds, we just return 0. We have nothing
1250 * left to unlink.
1252 if (ret && ret != EEXIST) {
1253 if (!rename(tmpfile, filename))
1254 return 0;
1255 ret = errno;
1257 unlink(tmpfile);
1258 if (ret) {
1259 if (ret != EEXIST) {
1260 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1261 return -1;
1263 /* FIXME!!! Collision check here ? */
1266 return 0;
1269 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1271 int size;
1272 unsigned char *compressed;
1273 z_stream stream;
1274 unsigned char sha1[20];
1275 char *filename;
1276 static char tmpfile[PATH_MAX];
1277 unsigned char hdr[50];
1278 int fd, hdrlen;
1280 /* Normally if we have it in the pack then we do not bother writing
1281 * it out into .git/objects/??/?{38} file.
1283 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1284 if (returnsha1)
1285 memcpy(returnsha1, sha1, 20);
1286 if (has_sha1_file(sha1))
1287 return 0;
1288 fd = open(filename, O_RDONLY);
1289 if (fd >= 0) {
1291 * FIXME!!! We might do collision checking here, but we'd
1292 * need to uncompress the old file and check it. Later.
1294 close(fd);
1295 return 0;
1298 if (errno != ENOENT) {
1299 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1300 return -1;
1303 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1305 fd = mkstemp(tmpfile);
1306 if (fd < 0) {
1307 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1308 return -1;
1311 /* Set it up */
1312 memset(&stream, 0, sizeof(stream));
1313 deflateInit(&stream, Z_BEST_COMPRESSION);
1314 size = deflateBound(&stream, len+hdrlen);
1315 compressed = xmalloc(size);
1317 /* Compress it */
1318 stream.next_out = compressed;
1319 stream.avail_out = size;
1321 /* First header.. */
1322 stream.next_in = hdr;
1323 stream.avail_in = hdrlen;
1324 while (deflate(&stream, 0) == Z_OK)
1325 /* nothing */;
1327 /* Then the data itself.. */
1328 stream.next_in = buf;
1329 stream.avail_in = len;
1330 while (deflate(&stream, Z_FINISH) == Z_OK)
1331 /* nothing */;
1332 deflateEnd(&stream);
1333 size = stream.total_out;
1335 if (write(fd, compressed, size) != size)
1336 die("unable to write file");
1337 fchmod(fd, 0444);
1338 close(fd);
1339 free(compressed);
1341 return move_temp_to_file(tmpfile, filename);
1344 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1346 ssize_t size;
1347 unsigned long objsize;
1348 int posn = 0;
1349 void *map = map_sha1_file_internal(sha1, &objsize);
1350 void *buf = map;
1351 void *temp_obj = NULL;
1352 z_stream stream;
1354 if (!buf) {
1355 unsigned char *unpacked;
1356 unsigned long len;
1357 char type[20];
1358 char hdr[50];
1359 int hdrlen;
1360 // need to unpack and recompress it by itself
1361 unpacked = read_packed_sha1(sha1, type, &len);
1363 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1365 /* Set it up */
1366 memset(&stream, 0, sizeof(stream));
1367 deflateInit(&stream, Z_BEST_COMPRESSION);
1368 size = deflateBound(&stream, len + hdrlen);
1369 temp_obj = buf = xmalloc(size);
1371 /* Compress it */
1372 stream.next_out = buf;
1373 stream.avail_out = size;
1375 /* First header.. */
1376 stream.next_in = (void *)hdr;
1377 stream.avail_in = hdrlen;
1378 while (deflate(&stream, 0) == Z_OK)
1379 /* nothing */;
1381 /* Then the data itself.. */
1382 stream.next_in = unpacked;
1383 stream.avail_in = len;
1384 while (deflate(&stream, Z_FINISH) == Z_OK)
1385 /* nothing */;
1386 deflateEnd(&stream);
1387 free(unpacked);
1389 objsize = stream.total_out;
1392 do {
1393 size = write(fd, buf + posn, objsize - posn);
1394 if (size <= 0) {
1395 if (!size) {
1396 fprintf(stderr, "write closed");
1397 } else {
1398 perror("write ");
1400 return -1;
1402 posn += size;
1403 } while (posn < objsize);
1405 if (map)
1406 munmap(map, objsize);
1407 if (temp_obj)
1408 free(temp_obj);
1410 return 0;
1413 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1414 size_t bufsize, size_t *bufposn)
1416 char tmpfile[PATH_MAX];
1417 int local;
1418 z_stream stream;
1419 unsigned char real_sha1[20];
1420 unsigned char discard[4096];
1421 int ret;
1422 SHA_CTX c;
1424 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1426 local = mkstemp(tmpfile);
1427 if (local < 0)
1428 return error("Couldn't open %s for %s\n", tmpfile, sha1_to_hex(sha1));
1430 memset(&stream, 0, sizeof(stream));
1432 inflateInit(&stream);
1434 SHA1_Init(&c);
1436 do {
1437 ssize_t size;
1438 if (*bufposn) {
1439 stream.avail_in = *bufposn;
1440 stream.next_in = (unsigned char *) buffer;
1441 do {
1442 stream.next_out = discard;
1443 stream.avail_out = sizeof(discard);
1444 ret = inflate(&stream, Z_SYNC_FLUSH);
1445 SHA1_Update(&c, discard, sizeof(discard) -
1446 stream.avail_out);
1447 } while (stream.avail_in && ret == Z_OK);
1448 write(local, buffer, *bufposn - stream.avail_in);
1449 memmove(buffer, buffer + *bufposn - stream.avail_in,
1450 stream.avail_in);
1451 *bufposn = stream.avail_in;
1452 if (ret != Z_OK)
1453 break;
1455 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1456 if (size <= 0) {
1457 close(local);
1458 unlink(tmpfile);
1459 if (!size)
1460 return error("Connection closed?");
1461 perror("Reading from connection");
1462 return -1;
1464 *bufposn += size;
1465 } while (1);
1466 inflateEnd(&stream);
1468 close(local);
1469 SHA1_Final(real_sha1, &c);
1470 if (ret != Z_STREAM_END) {
1471 unlink(tmpfile);
1472 return error("File %s corrupted", sha1_to_hex(sha1));
1474 if (memcmp(sha1, real_sha1, 20)) {
1475 unlink(tmpfile);
1476 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1479 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1482 int has_pack_index(const unsigned char *sha1)
1484 struct stat st;
1485 if (stat(sha1_pack_index_name(sha1), &st))
1486 return 0;
1487 return 1;
1490 int has_pack_file(const unsigned char *sha1)
1492 struct stat st;
1493 if (stat(sha1_pack_name(sha1), &st))
1494 return 0;
1495 return 1;
1498 int has_sha1_pack(const unsigned char *sha1)
1500 struct pack_entry e;
1501 return find_pack_entry(sha1, &e);
1504 int has_sha1_file(const unsigned char *sha1)
1506 struct stat st;
1507 struct pack_entry e;
1509 if (find_pack_entry(sha1, &e))
1510 return 1;
1511 return find_sha1_file(sha1, &st) ? 1 : 0;
1514 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1516 unsigned long size = st->st_size;
1517 void *buf;
1518 int ret;
1519 unsigned char hdr[50];
1520 int hdrlen;
1522 buf = "";
1523 if (size)
1524 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1525 close(fd);
1526 if (buf == MAP_FAILED)
1527 return -1;
1529 if (!type)
1530 type = "blob";
1531 if (write_object)
1532 ret = write_sha1_file(buf, size, type, sha1);
1533 else {
1534 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1535 ret = 0;
1537 if (size)
1538 munmap(buf, size);
1539 return ret;
1542 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1544 int fd;
1545 char *target;
1547 switch (st->st_mode & S_IFMT) {
1548 case S_IFREG:
1549 fd = open(path, O_RDONLY);
1550 if (fd < 0)
1551 return error("open(\"%s\"): %s", path,
1552 strerror(errno));
1553 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1554 return error("%s: failed to insert into database",
1555 path);
1556 break;
1557 case S_IFLNK:
1558 target = xmalloc(st->st_size+1);
1559 if (readlink(path, target, st->st_size+1) != st->st_size) {
1560 char *errstr = strerror(errno);
1561 free(target);
1562 return error("readlink(\"%s\"): %s", path,
1563 errstr);
1565 if (!write_object) {
1566 unsigned char hdr[50];
1567 int hdrlen;
1568 write_sha1_file_prepare(target, st->st_size, "blob",
1569 sha1, hdr, &hdrlen);
1570 } else if (write_sha1_file(target, st->st_size, "blob", sha1))
1571 return error("%s: failed to insert into database",
1572 path);
1573 free(target);
1574 break;
1575 default:
1576 return error("%s: unsupported file type", path);
1578 return 0;