Make git-send-pack exit with error when some refs couldn't be pushed out
[git/dscho.git] / sha1_file.c
blobfa22e9c71a0360ad5c3a865e0d6b3b49ba97e064
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 similar to that environment variable but can be
188 * LF separated. 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 const char *objdir = get_object_directory();
201 int base_len = -1;
203 last = alt;
204 while (last < ep) {
205 cp = last;
206 if (cp < ep && *cp == '#') {
207 while (cp < ep && *cp != sep)
208 cp++;
209 last = cp + 1;
210 continue;
212 for ( ; cp < ep && *cp != sep; cp++)
214 if (last != cp) {
215 struct alternate_object_database *alt;
216 /* 43 = 40-byte + 2 '/' + terminating NUL */
217 int pfxlen = cp - last;
218 int entlen = pfxlen + 43;
220 if (*last != '/' && relative_base) {
221 /* Relative alt-odb */
222 if (base_len < 0)
223 base_len = strlen(relative_base) + 1;
224 entlen += base_len;
225 pfxlen += base_len;
227 ent = xmalloc(sizeof(*ent) + entlen);
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 /* Prevent the common mistake of listing the same
242 * thing twice, or object directory itself.
244 for (alt = alt_odb_list; alt; alt = alt->next)
245 if (!memcmp(ent->base, alt->base, pfxlen))
246 goto bad;
247 if (!memcmp(ent->base, objdir, pfxlen)) {
248 bad:
249 free(ent);
251 else {
252 *alt_odb_tail = ent;
253 alt_odb_tail = &(ent->next);
254 ent->next = NULL;
257 while (cp < ep && *cp == sep)
258 cp++;
259 last = cp;
263 void prepare_alt_odb(void)
265 char path[PATH_MAX];
266 char *map;
267 int fd;
268 struct stat st;
269 char *alt;
271 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
272 if (!alt) alt = "";
274 if (alt_odb_tail)
275 return;
276 alt_odb_tail = &alt_odb_list;
277 link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL);
279 sprintf(path, "%s/info/alternates", get_object_directory());
280 fd = open(path, O_RDONLY);
281 if (fd < 0)
282 return;
283 if (fstat(fd, &st) || (st.st_size == 0)) {
284 close(fd);
285 return;
287 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
288 close(fd);
289 if (map == MAP_FAILED)
290 return;
292 link_alt_odb_entries(map, map + st.st_size, '\n',
293 get_object_directory());
294 munmap(map, st.st_size);
297 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
299 char *name = sha1_file_name(sha1);
300 struct alternate_object_database *alt;
302 if (!stat(name, st))
303 return name;
304 prepare_alt_odb();
305 for (alt = alt_odb_list; alt; alt = alt->next) {
306 name = alt->name;
307 fill_sha1_path(name, sha1);
308 if (!stat(alt->base, st))
309 return alt->base;
311 return NULL;
314 #define PACK_MAX_SZ (1<<26)
315 static int pack_used_ctr;
316 static unsigned long pack_mapped;
317 struct packed_git *packed_git;
319 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
320 void **idx_map_)
322 void *idx_map;
323 unsigned int *index;
324 unsigned long idx_size;
325 int nr, i;
326 int fd = open(path, O_RDONLY);
327 struct stat st;
328 if (fd < 0)
329 return -1;
330 if (fstat(fd, &st)) {
331 close(fd);
332 return -1;
334 idx_size = st.st_size;
335 idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
336 close(fd);
337 if (idx_map == MAP_FAILED)
338 return -1;
340 index = idx_map;
341 *idx_map_ = idx_map;
342 *idx_size_ = idx_size;
344 /* check index map */
345 if (idx_size < 4*256 + 20 + 20)
346 return error("index file too small");
347 nr = 0;
348 for (i = 0; i < 256; i++) {
349 unsigned int n = ntohl(index[i]);
350 if (n < nr)
351 return error("non-monotonic index");
352 nr = n;
356 * Total size:
357 * - 256 index entries 4 bytes each
358 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
359 * - 20-byte SHA1 of the packfile
360 * - 20-byte SHA1 file checksum
362 if (idx_size != 4*256 + nr * 24 + 20 + 20)
363 return error("wrong index file size");
365 return 0;
368 static int unuse_one_packed_git(void)
370 struct packed_git *p, *lru = NULL;
372 for (p = packed_git; p; p = p->next) {
373 if (p->pack_use_cnt || !p->pack_base)
374 continue;
375 if (!lru || p->pack_last_used < lru->pack_last_used)
376 lru = p;
378 if (!lru)
379 return 0;
380 munmap(lru->pack_base, lru->pack_size);
381 lru->pack_base = NULL;
382 return 1;
385 void unuse_packed_git(struct packed_git *p)
387 p->pack_use_cnt--;
390 int use_packed_git(struct packed_git *p)
392 if (!p->pack_size) {
393 struct stat st;
394 // We created the struct before we had the pack
395 stat(p->pack_name, &st);
396 if (!S_ISREG(st.st_mode))
397 die("packfile %s not a regular file", p->pack_name);
398 p->pack_size = st.st_size;
400 if (!p->pack_base) {
401 int fd;
402 struct stat st;
403 void *map;
405 pack_mapped += p->pack_size;
406 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
407 ; /* nothing */
408 fd = open(p->pack_name, O_RDONLY);
409 if (fd < 0)
410 die("packfile %s cannot be opened", p->pack_name);
411 if (fstat(fd, &st)) {
412 close(fd);
413 die("packfile %s cannot be opened", p->pack_name);
415 if (st.st_size != p->pack_size)
416 die("packfile %s size mismatch.", p->pack_name);
417 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
418 close(fd);
419 if (map == MAP_FAILED)
420 die("packfile %s cannot be mapped.", p->pack_name);
421 p->pack_base = map;
423 /* Check if the pack file matches with the index file.
424 * this is cheap.
426 if (memcmp((char*)(p->index_base) + p->index_size - 40,
427 p->pack_base + p->pack_size - 20, 20)) {
429 die("packfile %s does not match index.", p->pack_name);
432 p->pack_last_used = pack_used_ctr++;
433 p->pack_use_cnt++;
434 return 0;
437 struct packed_git *add_packed_git(char *path, int path_len, int local)
439 struct stat st;
440 struct packed_git *p;
441 unsigned long idx_size;
442 void *idx_map;
443 unsigned char sha1[20];
445 if (check_packed_git_idx(path, &idx_size, &idx_map))
446 return NULL;
448 /* do we have a corresponding .pack file? */
449 strcpy(path + path_len - 4, ".pack");
450 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
451 munmap(idx_map, idx_size);
452 return NULL;
454 /* ok, it looks sane as far as we can check without
455 * actually mapping the pack file.
457 p = xmalloc(sizeof(*p) + path_len + 2);
458 strcpy(p->pack_name, path);
459 p->index_size = idx_size;
460 p->pack_size = st.st_size;
461 p->index_base = idx_map;
462 p->next = NULL;
463 p->pack_base = NULL;
464 p->pack_last_used = 0;
465 p->pack_use_cnt = 0;
466 p->pack_local = local;
467 if (!get_sha1_hex(path + path_len - 40 - 4, sha1))
468 memcpy(p->sha1, sha1, 20);
469 return p;
472 struct packed_git *parse_pack_index(unsigned char *sha1)
474 char *path = sha1_pack_index_name(sha1);
475 return parse_pack_index_file(sha1, path);
478 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
480 struct packed_git *p;
481 unsigned long idx_size;
482 void *idx_map;
483 char *path;
485 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
486 return NULL;
488 path = sha1_pack_name(sha1);
490 p = xmalloc(sizeof(*p) + strlen(path) + 2);
491 strcpy(p->pack_name, path);
492 p->index_size = idx_size;
493 p->pack_size = 0;
494 p->index_base = idx_map;
495 p->next = NULL;
496 p->pack_base = NULL;
497 p->pack_last_used = 0;
498 p->pack_use_cnt = 0;
499 memcpy(p->sha1, sha1, 20);
500 return p;
503 void install_packed_git(struct packed_git *pack)
505 pack->next = packed_git;
506 packed_git = pack;
509 static void prepare_packed_git_one(char *objdir, int local)
511 char path[PATH_MAX];
512 int len;
513 DIR *dir;
514 struct dirent *de;
516 sprintf(path, "%s/pack", objdir);
517 len = strlen(path);
518 dir = opendir(path);
519 if (!dir)
520 return;
521 path[len++] = '/';
522 while ((de = readdir(dir)) != NULL) {
523 int namelen = strlen(de->d_name);
524 struct packed_git *p;
526 if (strcmp(de->d_name + namelen - 4, ".idx"))
527 continue;
529 /* we have .idx. Is it a file we can map? */
530 strcpy(path + len, de->d_name);
531 p = add_packed_git(path, len + namelen, local);
532 if (!p)
533 continue;
534 p->next = packed_git;
535 packed_git = p;
537 closedir(dir);
540 void prepare_packed_git(void)
542 static int run_once = 0;
543 struct alternate_object_database *alt;
545 if (run_once)
546 return;
547 prepare_packed_git_one(get_object_directory(), 1);
548 prepare_alt_odb();
549 for (alt = alt_odb_list; alt; alt = alt->next) {
550 alt->name[-1] = 0;
551 prepare_packed_git_one(alt->base, 0);
552 alt->name[-1] = '/';
554 run_once = 1;
557 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
559 char header[100];
560 unsigned char real_sha1[20];
561 SHA_CTX c;
563 SHA1_Init(&c);
564 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
565 SHA1_Update(&c, map, size);
566 SHA1_Final(real_sha1, &c);
567 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
570 static void *map_sha1_file_internal(const unsigned char *sha1,
571 unsigned long *size)
573 struct stat st;
574 void *map;
575 int fd;
576 char *filename = find_sha1_file(sha1, &st);
578 if (!filename) {
579 return NULL;
582 fd = open(filename, O_RDONLY | sha1_file_open_flag);
583 if (fd < 0) {
584 /* See if it works without O_NOATIME */
585 switch (sha1_file_open_flag) {
586 default:
587 fd = open(filename, O_RDONLY);
588 if (fd >= 0)
589 break;
590 /* Fallthrough */
591 case 0:
592 return NULL;
595 /* If it failed once, it will probably fail again.
596 * Stop using O_NOATIME
598 sha1_file_open_flag = 0;
600 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
601 close(fd);
602 if (map == MAP_FAILED)
603 return NULL;
604 *size = st.st_size;
605 return map;
608 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
610 /* Get the data stream */
611 memset(stream, 0, sizeof(*stream));
612 stream->next_in = map;
613 stream->avail_in = mapsize;
614 stream->next_out = buffer;
615 stream->avail_out = size;
617 inflateInit(stream);
618 return inflate(stream, 0);
621 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
623 int bytes = strlen(buffer) + 1;
624 unsigned char *buf = xmalloc(1+size);
626 memcpy(buf, buffer + bytes, stream->total_out - bytes);
627 bytes = stream->total_out - bytes;
628 if (bytes < size) {
629 stream->next_out = buf + bytes;
630 stream->avail_out = size - bytes;
631 while (inflate(stream, Z_FINISH) == Z_OK)
632 /* nothing */;
634 buf[size] = 0;
635 inflateEnd(stream);
636 return buf;
640 * We used to just use "sscanf()", but that's actually way
641 * too permissive for what we want to check. So do an anal
642 * object header parse by hand.
644 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
646 int i;
647 unsigned long size;
650 * The type can be at most ten bytes (including the
651 * terminating '\0' that we add), and is followed by
652 * a space.
654 i = 10;
655 for (;;) {
656 char c = *hdr++;
657 if (c == ' ')
658 break;
659 if (!--i)
660 return -1;
661 *type++ = c;
663 *type = 0;
666 * The length must follow immediately, and be in canonical
667 * decimal format (ie "010" is not valid).
669 size = *hdr++ - '0';
670 if (size > 9)
671 return -1;
672 if (size) {
673 for (;;) {
674 unsigned long c = *hdr - '0';
675 if (c > 9)
676 break;
677 hdr++;
678 size = size * 10 + c;
681 *sizep = size;
684 * The length must be followed by a zero byte
686 return *hdr ? -1 : 0;
689 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
691 int ret;
692 z_stream stream;
693 char hdr[8192];
695 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
696 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
697 return NULL;
699 return unpack_sha1_rest(&stream, hdr, *size);
702 /* forward declaration for a mutually recursive function */
703 static int packed_object_info(struct pack_entry *entry,
704 char *type, unsigned long *sizep);
706 static int packed_delta_info(unsigned char *base_sha1,
707 unsigned long delta_size,
708 unsigned long left,
709 char *type,
710 unsigned long *sizep,
711 struct packed_git *p)
713 struct pack_entry base_ent;
715 if (left < 20)
716 die("truncated pack file");
718 /* The base entry _must_ be in the same pack */
719 if (!find_pack_entry_one(base_sha1, &base_ent, p))
720 die("failed to find delta-pack base object %s",
721 sha1_to_hex(base_sha1));
723 /* We choose to only get the type of the base object and
724 * ignore potentially corrupt pack file that expects the delta
725 * based on a base with a wrong size. This saves tons of
726 * inflate() calls.
729 if (packed_object_info(&base_ent, type, NULL))
730 die("cannot get info for delta-pack base");
732 if (sizep) {
733 const unsigned char *data;
734 unsigned char delta_head[64];
735 unsigned long result_size;
736 z_stream stream;
737 int st;
739 memset(&stream, 0, sizeof(stream));
741 data = stream.next_in = base_sha1 + 20;
742 stream.avail_in = left - 20;
743 stream.next_out = delta_head;
744 stream.avail_out = sizeof(delta_head);
746 inflateInit(&stream);
747 st = inflate(&stream, Z_FINISH);
748 inflateEnd(&stream);
749 if ((st != Z_STREAM_END) &&
750 stream.total_out != sizeof(delta_head))
751 die("delta data unpack-initial failed");
753 /* Examine the initial part of the delta to figure out
754 * the result size.
756 data = delta_head;
757 get_delta_hdr_size(&data); /* ignore base size */
759 /* Read the result size */
760 result_size = get_delta_hdr_size(&data);
761 *sizep = result_size;
763 return 0;
766 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
767 enum object_type *type, unsigned long *sizep)
769 unsigned shift;
770 unsigned char *pack, c;
771 unsigned long size;
773 if (offset >= p->pack_size)
774 die("object offset outside of pack file");
776 pack = p->pack_base + offset;
777 c = *pack++;
778 offset++;
779 *type = (c >> 4) & 7;
780 size = c & 15;
781 shift = 4;
782 while (c & 0x80) {
783 if (offset >= p->pack_size)
784 die("object offset outside of pack file");
785 c = *pack++;
786 offset++;
787 size += (c & 0x7f) << shift;
788 shift += 7;
790 *sizep = size;
791 return offset;
794 void packed_object_info_detail(struct pack_entry *e,
795 char *type,
796 unsigned long *size,
797 unsigned long *store_size,
798 int *delta_chain_length,
799 unsigned char *base_sha1)
801 struct packed_git *p = e->p;
802 unsigned long offset, left;
803 unsigned char *pack;
804 enum object_type kind;
806 offset = unpack_object_header(p, e->offset, &kind, size);
807 pack = p->pack_base + offset;
808 left = p->pack_size - offset;
809 if (kind != OBJ_DELTA)
810 *delta_chain_length = 0;
811 else {
812 int chain_length = 0;
813 memcpy(base_sha1, pack, 20);
814 do {
815 struct pack_entry base_ent;
816 unsigned long junk;
818 find_pack_entry_one(pack, &base_ent, p);
819 offset = unpack_object_header(p, base_ent.offset,
820 &kind, &junk);
821 pack = p->pack_base + offset;
822 chain_length++;
823 } while (kind == OBJ_DELTA);
824 *delta_chain_length = chain_length;
826 switch (kind) {
827 case OBJ_COMMIT:
828 strcpy(type, "commit");
829 break;
830 case OBJ_TREE:
831 strcpy(type, "tree");
832 break;
833 case OBJ_BLOB:
834 strcpy(type, "blob");
835 break;
836 case OBJ_TAG:
837 strcpy(type, "tag");
838 break;
839 default:
840 die("corrupted pack file %s containing object of kind %d",
841 p->pack_name, kind);
843 *store_size = 0; /* notyet */
846 static int packed_object_info(struct pack_entry *entry,
847 char *type, unsigned long *sizep)
849 struct packed_git *p = entry->p;
850 unsigned long offset, size, left;
851 unsigned char *pack;
852 enum object_type kind;
853 int retval;
855 if (use_packed_git(p))
856 die("cannot map packed file");
858 offset = unpack_object_header(p, entry->offset, &kind, &size);
859 pack = p->pack_base + offset;
860 left = p->pack_size - offset;
862 switch (kind) {
863 case OBJ_DELTA:
864 retval = packed_delta_info(pack, size, left, type, sizep, p);
865 unuse_packed_git(p);
866 return retval;
867 case OBJ_COMMIT:
868 strcpy(type, "commit");
869 break;
870 case OBJ_TREE:
871 strcpy(type, "tree");
872 break;
873 case OBJ_BLOB:
874 strcpy(type, "blob");
875 break;
876 case OBJ_TAG:
877 strcpy(type, "tag");
878 break;
879 default:
880 die("corrupted pack file %s containing object of kind %d",
881 p->pack_name, kind);
883 if (sizep)
884 *sizep = size;
885 unuse_packed_git(p);
886 return 0;
889 /* forward declaration for a mutually recursive function */
890 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
892 static void *unpack_delta_entry(unsigned char *base_sha1,
893 unsigned long delta_size,
894 unsigned long left,
895 char *type,
896 unsigned long *sizep,
897 struct packed_git *p)
899 struct pack_entry base_ent;
900 void *data, *delta_data, *result, *base;
901 unsigned long data_size, result_size, base_size;
902 z_stream stream;
903 int st;
905 if (left < 20)
906 die("truncated pack file");
907 data = base_sha1 + 20;
908 data_size = left - 20;
909 delta_data = xmalloc(delta_size);
911 memset(&stream, 0, sizeof(stream));
913 stream.next_in = data;
914 stream.avail_in = data_size;
915 stream.next_out = delta_data;
916 stream.avail_out = delta_size;
918 inflateInit(&stream);
919 st = inflate(&stream, Z_FINISH);
920 inflateEnd(&stream);
921 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
922 die("delta data unpack failed");
924 /* The base entry _must_ be in the same pack */
925 if (!find_pack_entry_one(base_sha1, &base_ent, p))
926 die("failed to find delta-pack base object %s",
927 sha1_to_hex(base_sha1));
928 base = unpack_entry_gently(&base_ent, type, &base_size);
929 if (!base)
930 die("failed to read delta-pack base object %s",
931 sha1_to_hex(base_sha1));
932 result = patch_delta(base, base_size,
933 delta_data, delta_size,
934 &result_size);
935 if (!result)
936 die("failed to apply delta");
937 free(delta_data);
938 free(base);
939 *sizep = result_size;
940 return result;
943 static void *unpack_non_delta_entry(unsigned char *data,
944 unsigned long size,
945 unsigned long left)
947 int st;
948 z_stream stream;
949 unsigned char *buffer;
951 buffer = xmalloc(size + 1);
952 buffer[size] = 0;
953 memset(&stream, 0, sizeof(stream));
954 stream.next_in = data;
955 stream.avail_in = left;
956 stream.next_out = buffer;
957 stream.avail_out = size;
959 inflateInit(&stream);
960 st = inflate(&stream, Z_FINISH);
961 inflateEnd(&stream);
962 if ((st != Z_STREAM_END) || stream.total_out != size) {
963 free(buffer);
964 return NULL;
967 return buffer;
970 static void *unpack_entry(struct pack_entry *entry,
971 char *type, unsigned long *sizep)
973 struct packed_git *p = entry->p;
974 void *retval;
976 if (use_packed_git(p))
977 die("cannot map packed file");
978 retval = unpack_entry_gently(entry, type, sizep);
979 unuse_packed_git(p);
980 if (!retval)
981 die("corrupted pack file %s", p->pack_name);
982 return retval;
985 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
986 void *unpack_entry_gently(struct pack_entry *entry,
987 char *type, unsigned long *sizep)
989 struct packed_git *p = entry->p;
990 unsigned long offset, size, left;
991 unsigned char *pack;
992 enum object_type kind;
993 void *retval;
995 offset = unpack_object_header(p, entry->offset, &kind, &size);
996 pack = p->pack_base + offset;
997 left = p->pack_size - offset;
998 switch (kind) {
999 case OBJ_DELTA:
1000 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1001 return retval;
1002 case OBJ_COMMIT:
1003 strcpy(type, "commit");
1004 break;
1005 case OBJ_TREE:
1006 strcpy(type, "tree");
1007 break;
1008 case OBJ_BLOB:
1009 strcpy(type, "blob");
1010 break;
1011 case OBJ_TAG:
1012 strcpy(type, "tag");
1013 break;
1014 default:
1015 return NULL;
1017 *sizep = size;
1018 retval = unpack_non_delta_entry(pack, size, left);
1019 return retval;
1022 int num_packed_objects(const struct packed_git *p)
1024 /* See check_packed_git_idx() */
1025 return (p->index_size - 20 - 20 - 4*256) / 24;
1028 int nth_packed_object_sha1(const struct packed_git *p, int n,
1029 unsigned char* sha1)
1031 void *index = p->index_base + 256;
1032 if (n < 0 || num_packed_objects(p) <= n)
1033 return -1;
1034 memcpy(sha1, (index + 24 * n + 4), 20);
1035 return 0;
1038 int find_pack_entry_one(const unsigned char *sha1,
1039 struct pack_entry *e, struct packed_git *p)
1041 unsigned int *level1_ofs = p->index_base;
1042 int hi = ntohl(level1_ofs[*sha1]);
1043 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1044 void *index = p->index_base + 256;
1046 do {
1047 int mi = (lo + hi) / 2;
1048 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1049 if (!cmp) {
1050 e->offset = ntohl(*((int*)(index + 24 * mi)));
1051 memcpy(e->sha1, sha1, 20);
1052 e->p = p;
1053 return 1;
1055 if (cmp > 0)
1056 hi = mi;
1057 else
1058 lo = mi+1;
1059 } while (lo < hi);
1060 return 0;
1063 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1065 struct packed_git *p;
1066 prepare_packed_git();
1068 for (p = packed_git; p; p = p->next) {
1069 if (find_pack_entry_one(sha1, e, p))
1070 return 1;
1072 return 0;
1075 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1076 struct packed_git *packs)
1078 struct packed_git *p;
1079 struct pack_entry e;
1081 for (p = packs; p; p = p->next) {
1082 if (find_pack_entry_one(sha1, &e, p))
1083 return p;
1085 return NULL;
1089 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1091 int status;
1092 unsigned long mapsize, size;
1093 void *map;
1094 z_stream stream;
1095 char hdr[128];
1097 map = map_sha1_file_internal(sha1, &mapsize);
1098 if (!map) {
1099 struct pack_entry e;
1101 if (!find_pack_entry(sha1, &e))
1102 return error("unable to find %s", sha1_to_hex(sha1));
1103 return packed_object_info(&e, type, sizep);
1105 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1106 status = error("unable to unpack %s header",
1107 sha1_to_hex(sha1));
1108 if (parse_sha1_header(hdr, type, &size) < 0)
1109 status = error("unable to parse %s header", sha1_to_hex(sha1));
1110 else {
1111 status = 0;
1112 if (sizep)
1113 *sizep = size;
1115 inflateEnd(&stream);
1116 munmap(map, mapsize);
1117 return status;
1120 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1122 struct pack_entry e;
1124 if (!find_pack_entry(sha1, &e)) {
1125 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1126 return NULL;
1128 return unpack_entry(&e, type, size);
1131 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1133 unsigned long mapsize;
1134 void *map, *buf;
1135 struct pack_entry e;
1137 if (find_pack_entry(sha1, &e))
1138 return read_packed_sha1(sha1, type, size);
1139 map = map_sha1_file_internal(sha1, &mapsize);
1140 if (map) {
1141 buf = unpack_sha1_file(map, mapsize, type, size);
1142 munmap(map, mapsize);
1143 return buf;
1145 return NULL;
1148 void *read_object_with_reference(const unsigned char *sha1,
1149 const char *required_type,
1150 unsigned long *size,
1151 unsigned char *actual_sha1_return)
1153 char type[20];
1154 void *buffer;
1155 unsigned long isize;
1156 unsigned char actual_sha1[20];
1158 memcpy(actual_sha1, sha1, 20);
1159 while (1) {
1160 int ref_length = -1;
1161 const char *ref_type = NULL;
1163 buffer = read_sha1_file(actual_sha1, type, &isize);
1164 if (!buffer)
1165 return NULL;
1166 if (!strcmp(type, required_type)) {
1167 *size = isize;
1168 if (actual_sha1_return)
1169 memcpy(actual_sha1_return, actual_sha1, 20);
1170 return buffer;
1172 /* Handle references */
1173 else if (!strcmp(type, "commit"))
1174 ref_type = "tree ";
1175 else if (!strcmp(type, "tag"))
1176 ref_type = "object ";
1177 else {
1178 free(buffer);
1179 return NULL;
1181 ref_length = strlen(ref_type);
1183 if (memcmp(buffer, ref_type, ref_length) ||
1184 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1185 free(buffer);
1186 return NULL;
1188 free(buffer);
1189 /* Now we have the ID of the referred-to object in
1190 * actual_sha1. Check again. */
1194 char *write_sha1_file_prepare(void *buf,
1195 unsigned long len,
1196 const char *type,
1197 unsigned char *sha1,
1198 unsigned char *hdr,
1199 int *hdrlen)
1201 SHA_CTX c;
1203 /* Generate the header */
1204 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1206 /* Sha1.. */
1207 SHA1_Init(&c);
1208 SHA1_Update(&c, hdr, *hdrlen);
1209 SHA1_Update(&c, buf, len);
1210 SHA1_Final(sha1, &c);
1212 return sha1_file_name(sha1);
1216 * Link the tempfile to the final place, possibly creating the
1217 * last directory level as you do so.
1219 * Returns the errno on failure, 0 on success.
1221 static int link_temp_to_file(const char *tmpfile, char *filename)
1223 int ret;
1225 if (!link(tmpfile, filename))
1226 return 0;
1229 * Try to mkdir the last path component if that failed
1230 * with an ENOENT.
1232 * Re-try the "link()" regardless of whether the mkdir
1233 * succeeds, since a race might mean that somebody
1234 * else succeeded.
1236 ret = errno;
1237 if (ret == ENOENT) {
1238 char *dir = strrchr(filename, '/');
1239 if (dir) {
1240 *dir = 0;
1241 mkdir(filename, 0777);
1242 *dir = '/';
1243 if (!link(tmpfile, filename))
1244 return 0;
1245 ret = errno;
1248 return ret;
1252 * Move the just written object into its final resting place
1254 int move_temp_to_file(const char *tmpfile, char *filename)
1256 int ret = link_temp_to_file(tmpfile, filename);
1259 * Coda hack - coda doesn't like cross-directory links,
1260 * so we fall back to a rename, which will mean that it
1261 * won't be able to check collisions, but that's not a
1262 * big deal.
1264 * The same holds for FAT formatted media.
1266 * When this succeeds, we just return 0. We have nothing
1267 * left to unlink.
1269 if (ret && ret != EEXIST) {
1270 if (!rename(tmpfile, filename))
1271 return 0;
1272 ret = errno;
1274 unlink(tmpfile);
1275 if (ret) {
1276 if (ret != EEXIST) {
1277 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1278 return -1;
1280 /* FIXME!!! Collision check here ? */
1283 return 0;
1286 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1288 int size;
1289 unsigned char *compressed;
1290 z_stream stream;
1291 unsigned char sha1[20];
1292 char *filename;
1293 static char tmpfile[PATH_MAX];
1294 unsigned char hdr[50];
1295 int fd, hdrlen;
1297 /* Normally if we have it in the pack then we do not bother writing
1298 * it out into .git/objects/??/?{38} file.
1300 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1301 if (returnsha1)
1302 memcpy(returnsha1, sha1, 20);
1303 if (has_sha1_file(sha1))
1304 return 0;
1305 fd = open(filename, O_RDONLY);
1306 if (fd >= 0) {
1308 * FIXME!!! We might do collision checking here, but we'd
1309 * need to uncompress the old file and check it. Later.
1311 close(fd);
1312 return 0;
1315 if (errno != ENOENT) {
1316 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1317 return -1;
1320 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1322 fd = mkstemp(tmpfile);
1323 if (fd < 0) {
1324 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1325 return -1;
1328 /* Set it up */
1329 memset(&stream, 0, sizeof(stream));
1330 deflateInit(&stream, Z_BEST_COMPRESSION);
1331 size = deflateBound(&stream, len+hdrlen);
1332 compressed = xmalloc(size);
1334 /* Compress it */
1335 stream.next_out = compressed;
1336 stream.avail_out = size;
1338 /* First header.. */
1339 stream.next_in = hdr;
1340 stream.avail_in = hdrlen;
1341 while (deflate(&stream, 0) == Z_OK)
1342 /* nothing */;
1344 /* Then the data itself.. */
1345 stream.next_in = buf;
1346 stream.avail_in = len;
1347 while (deflate(&stream, Z_FINISH) == Z_OK)
1348 /* nothing */;
1349 deflateEnd(&stream);
1350 size = stream.total_out;
1352 if (write(fd, compressed, size) != size)
1353 die("unable to write file");
1354 fchmod(fd, 0444);
1355 close(fd);
1356 free(compressed);
1358 return move_temp_to_file(tmpfile, filename);
1361 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1363 ssize_t size;
1364 unsigned long objsize;
1365 int posn = 0;
1366 void *map = map_sha1_file_internal(sha1, &objsize);
1367 void *buf = map;
1368 void *temp_obj = NULL;
1369 z_stream stream;
1371 if (!buf) {
1372 unsigned char *unpacked;
1373 unsigned long len;
1374 char type[20];
1375 char hdr[50];
1376 int hdrlen;
1377 // need to unpack and recompress it by itself
1378 unpacked = read_packed_sha1(sha1, type, &len);
1380 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1382 /* Set it up */
1383 memset(&stream, 0, sizeof(stream));
1384 deflateInit(&stream, Z_BEST_COMPRESSION);
1385 size = deflateBound(&stream, len + hdrlen);
1386 temp_obj = buf = xmalloc(size);
1388 /* Compress it */
1389 stream.next_out = buf;
1390 stream.avail_out = size;
1392 /* First header.. */
1393 stream.next_in = (void *)hdr;
1394 stream.avail_in = hdrlen;
1395 while (deflate(&stream, 0) == Z_OK)
1396 /* nothing */;
1398 /* Then the data itself.. */
1399 stream.next_in = unpacked;
1400 stream.avail_in = len;
1401 while (deflate(&stream, Z_FINISH) == Z_OK)
1402 /* nothing */;
1403 deflateEnd(&stream);
1404 free(unpacked);
1406 objsize = stream.total_out;
1409 do {
1410 size = write(fd, buf + posn, objsize - posn);
1411 if (size <= 0) {
1412 if (!size) {
1413 fprintf(stderr, "write closed");
1414 } else {
1415 perror("write ");
1417 return -1;
1419 posn += size;
1420 } while (posn < objsize);
1422 if (map)
1423 munmap(map, objsize);
1424 if (temp_obj)
1425 free(temp_obj);
1427 return 0;
1430 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1431 size_t bufsize, size_t *bufposn)
1433 char tmpfile[PATH_MAX];
1434 int local;
1435 z_stream stream;
1436 unsigned char real_sha1[20];
1437 unsigned char discard[4096];
1438 int ret;
1439 SHA_CTX c;
1441 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1443 local = mkstemp(tmpfile);
1444 if (local < 0)
1445 return error("Couldn't open %s for %s\n", tmpfile, sha1_to_hex(sha1));
1447 memset(&stream, 0, sizeof(stream));
1449 inflateInit(&stream);
1451 SHA1_Init(&c);
1453 do {
1454 ssize_t size;
1455 if (*bufposn) {
1456 stream.avail_in = *bufposn;
1457 stream.next_in = (unsigned char *) buffer;
1458 do {
1459 stream.next_out = discard;
1460 stream.avail_out = sizeof(discard);
1461 ret = inflate(&stream, Z_SYNC_FLUSH);
1462 SHA1_Update(&c, discard, sizeof(discard) -
1463 stream.avail_out);
1464 } while (stream.avail_in && ret == Z_OK);
1465 write(local, buffer, *bufposn - stream.avail_in);
1466 memmove(buffer, buffer + *bufposn - stream.avail_in,
1467 stream.avail_in);
1468 *bufposn = stream.avail_in;
1469 if (ret != Z_OK)
1470 break;
1472 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1473 if (size <= 0) {
1474 close(local);
1475 unlink(tmpfile);
1476 if (!size)
1477 return error("Connection closed?");
1478 perror("Reading from connection");
1479 return -1;
1481 *bufposn += size;
1482 } while (1);
1483 inflateEnd(&stream);
1485 close(local);
1486 SHA1_Final(real_sha1, &c);
1487 if (ret != Z_STREAM_END) {
1488 unlink(tmpfile);
1489 return error("File %s corrupted", sha1_to_hex(sha1));
1491 if (memcmp(sha1, real_sha1, 20)) {
1492 unlink(tmpfile);
1493 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1496 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1499 int has_pack_index(const unsigned char *sha1)
1501 struct stat st;
1502 if (stat(sha1_pack_index_name(sha1), &st))
1503 return 0;
1504 return 1;
1507 int has_pack_file(const unsigned char *sha1)
1509 struct stat st;
1510 if (stat(sha1_pack_name(sha1), &st))
1511 return 0;
1512 return 1;
1515 int has_sha1_pack(const unsigned char *sha1)
1517 struct pack_entry e;
1518 return find_pack_entry(sha1, &e);
1521 int has_sha1_file(const unsigned char *sha1)
1523 struct stat st;
1524 struct pack_entry e;
1526 if (find_pack_entry(sha1, &e))
1527 return 1;
1528 return find_sha1_file(sha1, &st) ? 1 : 0;
1531 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1533 unsigned long size = 4096;
1534 char *buf = malloc(size);
1535 int iret, ret;
1536 unsigned long off = 0;
1537 unsigned char hdr[50];
1538 int hdrlen;
1539 do {
1540 iret = read(fd, buf + off, size - off);
1541 if (iret > 0) {
1542 off += iret;
1543 if (off == size) {
1544 size *= 2;
1545 buf = realloc(buf, size);
1548 } while (iret > 0);
1549 if (iret < 0) {
1550 free(buf);
1551 return -1;
1553 if (!type)
1554 type = "blob";
1555 if (write_object)
1556 ret = write_sha1_file(buf, off, type, sha1);
1557 else {
1558 write_sha1_file_prepare(buf, off, type, sha1, hdr, &hdrlen);
1559 ret = 0;
1561 free(buf);
1562 return ret;
1565 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1567 unsigned long size = st->st_size;
1568 void *buf;
1569 int ret;
1570 unsigned char hdr[50];
1571 int hdrlen;
1573 buf = "";
1574 if (size)
1575 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1576 close(fd);
1577 if (buf == MAP_FAILED)
1578 return -1;
1580 if (!type)
1581 type = "blob";
1582 if (write_object)
1583 ret = write_sha1_file(buf, size, type, sha1);
1584 else {
1585 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1586 ret = 0;
1588 if (size)
1589 munmap(buf, size);
1590 return ret;
1593 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1595 int fd;
1596 char *target;
1598 switch (st->st_mode & S_IFMT) {
1599 case S_IFREG:
1600 fd = open(path, O_RDONLY);
1601 if (fd < 0)
1602 return error("open(\"%s\"): %s", path,
1603 strerror(errno));
1604 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1605 return error("%s: failed to insert into database",
1606 path);
1607 break;
1608 case S_IFLNK:
1609 target = xmalloc(st->st_size+1);
1610 if (readlink(path, target, st->st_size+1) != st->st_size) {
1611 char *errstr = strerror(errno);
1612 free(target);
1613 return error("readlink(\"%s\"): %s", path,
1614 errstr);
1616 if (!write_object) {
1617 unsigned char hdr[50];
1618 int hdrlen;
1619 write_sha1_file_prepare(target, st->st_size, "blob",
1620 sha1, hdr, &hdrlen);
1621 } else if (write_sha1_file(target, st->st_size, "blob", sha1))
1622 return error("%s: failed to insert into database",
1623 path);
1624 free(target);
1625 break;
1626 default:
1627 return error("%s: unsupported file type", path);
1629 return 0;