Add bug isolation howto, scraped from Linus.
[git/jnareb-git.git] / sha1_file.c
blob946a35346bdaf21432b7f463712f5bf8881fee39
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;
428 if (check_packed_git_idx(path, &idx_size, &idx_map))
429 return NULL;
431 /* do we have a corresponding .pack file? */
432 strcpy(path + path_len - 4, ".pack");
433 if (stat(path, &st) || !S_ISREG(st.st_mode)) {
434 munmap(idx_map, idx_size);
435 return NULL;
437 /* ok, it looks sane as far as we can check without
438 * actually mapping the pack file.
440 p = xmalloc(sizeof(*p) + path_len + 2);
441 strcpy(p->pack_name, path);
442 p->index_size = idx_size;
443 p->pack_size = st.st_size;
444 p->index_base = idx_map;
445 p->next = NULL;
446 p->pack_base = NULL;
447 p->pack_last_used = 0;
448 p->pack_use_cnt = 0;
449 p->pack_local = local;
450 return p;
453 struct packed_git *parse_pack_index(unsigned char *sha1)
455 char *path = sha1_pack_index_name(sha1);
456 return parse_pack_index_file(sha1, path);
459 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
461 struct packed_git *p;
462 unsigned long idx_size;
463 void *idx_map;
464 char *path;
466 if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
467 return NULL;
469 path = sha1_pack_name(sha1);
471 p = xmalloc(sizeof(*p) + strlen(path) + 2);
472 strcpy(p->pack_name, path);
473 p->index_size = idx_size;
474 p->pack_size = 0;
475 p->index_base = idx_map;
476 p->next = NULL;
477 p->pack_base = NULL;
478 p->pack_last_used = 0;
479 p->pack_use_cnt = 0;
480 memcpy(p->sha1, sha1, 20);
481 return p;
484 void install_packed_git(struct packed_git *pack)
486 pack->next = packed_git;
487 packed_git = pack;
490 static void prepare_packed_git_one(char *objdir, int local)
492 char path[PATH_MAX];
493 int len;
494 DIR *dir;
495 struct dirent *de;
497 sprintf(path, "%s/pack", objdir);
498 len = strlen(path);
499 dir = opendir(path);
500 if (!dir)
501 return;
502 path[len++] = '/';
503 while ((de = readdir(dir)) != NULL) {
504 int namelen = strlen(de->d_name);
505 struct packed_git *p;
507 if (strcmp(de->d_name + namelen - 4, ".idx"))
508 continue;
510 /* we have .idx. Is it a file we can map? */
511 strcpy(path + len, de->d_name);
512 p = add_packed_git(path, len + namelen, local);
513 if (!p)
514 continue;
515 p->next = packed_git;
516 packed_git = p;
518 closedir(dir);
521 void prepare_packed_git(void)
523 static int run_once = 0;
524 struct alternate_object_database *alt;
526 if (run_once)
527 return;
528 prepare_packed_git_one(get_object_directory(), 1);
529 prepare_alt_odb();
530 for (alt = alt_odb_list; alt; alt = alt->next) {
531 alt->name[0] = 0;
532 prepare_packed_git_one(alt->base, 0);
534 run_once = 1;
537 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
539 char header[100];
540 unsigned char real_sha1[20];
541 SHA_CTX c;
543 SHA1_Init(&c);
544 SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
545 SHA1_Update(&c, map, size);
546 SHA1_Final(real_sha1, &c);
547 return memcmp(sha1, real_sha1, 20) ? -1 : 0;
550 static void *map_sha1_file_internal(const unsigned char *sha1,
551 unsigned long *size)
553 struct stat st;
554 void *map;
555 int fd;
556 char *filename = find_sha1_file(sha1, &st);
558 if (!filename) {
559 return NULL;
562 fd = open(filename, O_RDONLY | sha1_file_open_flag);
563 if (fd < 0) {
564 /* See if it works without O_NOATIME */
565 switch (sha1_file_open_flag) {
566 default:
567 fd = open(filename, O_RDONLY);
568 if (fd >= 0)
569 break;
570 /* Fallthrough */
571 case 0:
572 return NULL;
575 /* If it failed once, it will probably fail again.
576 * Stop using O_NOATIME
578 sha1_file_open_flag = 0;
580 map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
581 close(fd);
582 if (map == MAP_FAILED)
583 return NULL;
584 *size = st.st_size;
585 return map;
588 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
590 /* Get the data stream */
591 memset(stream, 0, sizeof(*stream));
592 stream->next_in = map;
593 stream->avail_in = mapsize;
594 stream->next_out = buffer;
595 stream->avail_out = size;
597 inflateInit(stream);
598 return inflate(stream, 0);
601 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
603 int bytes = strlen(buffer) + 1;
604 unsigned char *buf = xmalloc(1+size);
606 memcpy(buf, buffer + bytes, stream->total_out - bytes);
607 bytes = stream->total_out - bytes;
608 if (bytes < size) {
609 stream->next_out = buf + bytes;
610 stream->avail_out = size - bytes;
611 while (inflate(stream, Z_FINISH) == Z_OK)
612 /* nothing */;
614 buf[size] = 0;
615 inflateEnd(stream);
616 return buf;
620 * We used to just use "sscanf()", but that's actually way
621 * too permissive for what we want to check. So do an anal
622 * object header parse by hand.
624 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
626 int i;
627 unsigned long size;
630 * The type can be at most ten bytes (including the
631 * terminating '\0' that we add), and is followed by
632 * a space.
634 i = 10;
635 for (;;) {
636 char c = *hdr++;
637 if (c == ' ')
638 break;
639 if (!--i)
640 return -1;
641 *type++ = c;
643 *type = 0;
646 * The length must follow immediately, and be in canonical
647 * decimal format (ie "010" is not valid).
649 size = *hdr++ - '0';
650 if (size > 9)
651 return -1;
652 if (size) {
653 for (;;) {
654 unsigned long c = *hdr - '0';
655 if (c > 9)
656 break;
657 hdr++;
658 size = size * 10 + c;
661 *sizep = size;
664 * The length must be followed by a zero byte
666 return *hdr ? -1 : 0;
669 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
671 int ret;
672 z_stream stream;
673 char hdr[8192];
675 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
676 if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
677 return NULL;
679 return unpack_sha1_rest(&stream, hdr, *size);
682 /* forward declaration for a mutually recursive function */
683 static int packed_object_info(struct pack_entry *entry,
684 char *type, unsigned long *sizep);
686 static int packed_delta_info(unsigned char *base_sha1,
687 unsigned long delta_size,
688 unsigned long left,
689 char *type,
690 unsigned long *sizep,
691 struct packed_git *p)
693 struct pack_entry base_ent;
695 if (left < 20)
696 die("truncated pack file");
698 /* The base entry _must_ be in the same pack */
699 if (!find_pack_entry_one(base_sha1, &base_ent, p))
700 die("failed to find delta-pack base object %s",
701 sha1_to_hex(base_sha1));
703 /* We choose to only get the type of the base object and
704 * ignore potentially corrupt pack file that expects the delta
705 * based on a base with a wrong size. This saves tons of
706 * inflate() calls.
709 if (packed_object_info(&base_ent, type, NULL))
710 die("cannot get info for delta-pack base");
712 if (sizep) {
713 const unsigned char *data;
714 unsigned char delta_head[64];
715 unsigned long result_size;
716 z_stream stream;
717 int st;
719 memset(&stream, 0, sizeof(stream));
721 data = stream.next_in = base_sha1 + 20;
722 stream.avail_in = left - 20;
723 stream.next_out = delta_head;
724 stream.avail_out = sizeof(delta_head);
726 inflateInit(&stream);
727 st = inflate(&stream, Z_FINISH);
728 inflateEnd(&stream);
729 if ((st != Z_STREAM_END) &&
730 stream.total_out != sizeof(delta_head))
731 die("delta data unpack-initial failed");
733 /* Examine the initial part of the delta to figure out
734 * the result size.
736 data = delta_head;
737 get_delta_hdr_size(&data); /* ignore base size */
739 /* Read the result size */
740 result_size = get_delta_hdr_size(&data);
741 *sizep = result_size;
743 return 0;
746 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
747 enum object_type *type, unsigned long *sizep)
749 unsigned shift;
750 unsigned char *pack, c;
751 unsigned long size;
753 if (offset >= p->pack_size)
754 die("object offset outside of pack file");
756 pack = p->pack_base + offset;
757 c = *pack++;
758 offset++;
759 *type = (c >> 4) & 7;
760 size = c & 15;
761 shift = 4;
762 while (c & 0x80) {
763 if (offset >= p->pack_size)
764 die("object offset outside of pack file");
765 c = *pack++;
766 offset++;
767 size += (c & 0x7f) << shift;
768 shift += 7;
770 *sizep = size;
771 return offset;
774 void packed_object_info_detail(struct pack_entry *e,
775 char *type,
776 unsigned long *size,
777 unsigned long *store_size,
778 int *delta_chain_length,
779 unsigned char *base_sha1)
781 struct packed_git *p = e->p;
782 unsigned long offset, left;
783 unsigned char *pack;
784 enum object_type kind;
786 offset = unpack_object_header(p, e->offset, &kind, size);
787 pack = p->pack_base + offset;
788 left = p->pack_size - offset;
789 if (kind != OBJ_DELTA)
790 *delta_chain_length = 0;
791 else {
792 int chain_length = 0;
793 memcpy(base_sha1, pack, 20);
794 do {
795 struct pack_entry base_ent;
796 unsigned long junk;
798 find_pack_entry_one(pack, &base_ent, p);
799 offset = unpack_object_header(p, base_ent.offset,
800 &kind, &junk);
801 pack = p->pack_base + offset;
802 chain_length++;
803 } while (kind == OBJ_DELTA);
804 *delta_chain_length = chain_length;
806 switch (kind) {
807 case OBJ_COMMIT:
808 strcpy(type, "commit");
809 break;
810 case OBJ_TREE:
811 strcpy(type, "tree");
812 break;
813 case OBJ_BLOB:
814 strcpy(type, "blob");
815 break;
816 case OBJ_TAG:
817 strcpy(type, "tag");
818 break;
819 default:
820 die("corrupted pack file %s containing object of kind %d",
821 p->pack_name, kind);
823 *store_size = 0; /* notyet */
826 static int packed_object_info(struct pack_entry *entry,
827 char *type, unsigned long *sizep)
829 struct packed_git *p = entry->p;
830 unsigned long offset, size, left;
831 unsigned char *pack;
832 enum object_type kind;
833 int retval;
835 if (use_packed_git(p))
836 die("cannot map packed file");
838 offset = unpack_object_header(p, entry->offset, &kind, &size);
839 pack = p->pack_base + offset;
840 left = p->pack_size - offset;
842 switch (kind) {
843 case OBJ_DELTA:
844 retval = packed_delta_info(pack, size, left, type, sizep, p);
845 unuse_packed_git(p);
846 return retval;
847 case OBJ_COMMIT:
848 strcpy(type, "commit");
849 break;
850 case OBJ_TREE:
851 strcpy(type, "tree");
852 break;
853 case OBJ_BLOB:
854 strcpy(type, "blob");
855 break;
856 case OBJ_TAG:
857 strcpy(type, "tag");
858 break;
859 default:
860 die("corrupted pack file %s containing object of kind %d",
861 p->pack_name, kind);
863 if (sizep)
864 *sizep = size;
865 unuse_packed_git(p);
866 return 0;
869 /* forward declaration for a mutually recursive function */
870 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
872 static void *unpack_delta_entry(unsigned char *base_sha1,
873 unsigned long delta_size,
874 unsigned long left,
875 char *type,
876 unsigned long *sizep,
877 struct packed_git *p)
879 struct pack_entry base_ent;
880 void *data, *delta_data, *result, *base;
881 unsigned long data_size, result_size, base_size;
882 z_stream stream;
883 int st;
885 if (left < 20)
886 die("truncated pack file");
887 data = base_sha1 + 20;
888 data_size = left - 20;
889 delta_data = xmalloc(delta_size);
891 memset(&stream, 0, sizeof(stream));
893 stream.next_in = data;
894 stream.avail_in = data_size;
895 stream.next_out = delta_data;
896 stream.avail_out = delta_size;
898 inflateInit(&stream);
899 st = inflate(&stream, Z_FINISH);
900 inflateEnd(&stream);
901 if ((st != Z_STREAM_END) || stream.total_out != delta_size)
902 die("delta data unpack failed");
904 /* The base entry _must_ be in the same pack */
905 if (!find_pack_entry_one(base_sha1, &base_ent, p))
906 die("failed to find delta-pack base object %s",
907 sha1_to_hex(base_sha1));
908 base = unpack_entry_gently(&base_ent, type, &base_size);
909 if (!base)
910 die("failed to read delta-pack base object %s",
911 sha1_to_hex(base_sha1));
912 result = patch_delta(base, base_size,
913 delta_data, delta_size,
914 &result_size);
915 if (!result)
916 die("failed to apply delta");
917 free(delta_data);
918 free(base);
919 *sizep = result_size;
920 return result;
923 static void *unpack_non_delta_entry(unsigned char *data,
924 unsigned long size,
925 unsigned long left)
927 int st;
928 z_stream stream;
929 unsigned char *buffer;
931 buffer = xmalloc(size + 1);
932 buffer[size] = 0;
933 memset(&stream, 0, sizeof(stream));
934 stream.next_in = data;
935 stream.avail_in = left;
936 stream.next_out = buffer;
937 stream.avail_out = size;
939 inflateInit(&stream);
940 st = inflate(&stream, Z_FINISH);
941 inflateEnd(&stream);
942 if ((st != Z_STREAM_END) || stream.total_out != size) {
943 free(buffer);
944 return NULL;
947 return buffer;
950 static void *unpack_entry(struct pack_entry *entry,
951 char *type, unsigned long *sizep)
953 struct packed_git *p = entry->p;
954 void *retval;
956 if (use_packed_git(p))
957 die("cannot map packed file");
958 retval = unpack_entry_gently(entry, type, sizep);
959 unuse_packed_git(p);
960 if (!retval)
961 die("corrupted pack file %s", p->pack_name);
962 return retval;
965 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
966 void *unpack_entry_gently(struct pack_entry *entry,
967 char *type, unsigned long *sizep)
969 struct packed_git *p = entry->p;
970 unsigned long offset, size, left;
971 unsigned char *pack;
972 enum object_type kind;
973 void *retval;
975 offset = unpack_object_header(p, entry->offset, &kind, &size);
976 pack = p->pack_base + offset;
977 left = p->pack_size - offset;
978 switch (kind) {
979 case OBJ_DELTA:
980 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
981 return retval;
982 case OBJ_COMMIT:
983 strcpy(type, "commit");
984 break;
985 case OBJ_TREE:
986 strcpy(type, "tree");
987 break;
988 case OBJ_BLOB:
989 strcpy(type, "blob");
990 break;
991 case OBJ_TAG:
992 strcpy(type, "tag");
993 break;
994 default:
995 return NULL;
997 *sizep = size;
998 retval = unpack_non_delta_entry(pack, size, left);
999 return retval;
1002 int num_packed_objects(const struct packed_git *p)
1004 /* See check_packed_git_idx() */
1005 return (p->index_size - 20 - 20 - 4*256) / 24;
1008 int nth_packed_object_sha1(const struct packed_git *p, int n,
1009 unsigned char* sha1)
1011 void *index = p->index_base + 256;
1012 if (n < 0 || num_packed_objects(p) <= n)
1013 return -1;
1014 memcpy(sha1, (index + 24 * n + 4), 20);
1015 return 0;
1018 int find_pack_entry_one(const unsigned char *sha1,
1019 struct pack_entry *e, struct packed_git *p)
1021 unsigned int *level1_ofs = p->index_base;
1022 int hi = ntohl(level1_ofs[*sha1]);
1023 int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1024 void *index = p->index_base + 256;
1026 do {
1027 int mi = (lo + hi) / 2;
1028 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1029 if (!cmp) {
1030 e->offset = ntohl(*((int*)(index + 24 * mi)));
1031 memcpy(e->sha1, sha1, 20);
1032 e->p = p;
1033 return 1;
1035 if (cmp > 0)
1036 hi = mi;
1037 else
1038 lo = mi+1;
1039 } while (lo < hi);
1040 return 0;
1043 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1045 struct packed_git *p;
1046 prepare_packed_git();
1048 for (p = packed_git; p; p = p->next) {
1049 if (find_pack_entry_one(sha1, e, p))
1050 return 1;
1052 return 0;
1055 struct packed_git *find_sha1_pack(const unsigned char *sha1,
1056 struct packed_git *packs)
1058 struct packed_git *p;
1059 struct pack_entry e;
1061 for (p = packs; p; p = p->next) {
1062 if (find_pack_entry_one(sha1, &e, p))
1063 return p;
1065 return NULL;
1069 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1071 int status;
1072 unsigned long mapsize, size;
1073 void *map;
1074 z_stream stream;
1075 char hdr[128];
1077 map = map_sha1_file_internal(sha1, &mapsize);
1078 if (!map) {
1079 struct pack_entry e;
1081 if (!find_pack_entry(sha1, &e))
1082 return error("unable to find %s", sha1_to_hex(sha1));
1083 return packed_object_info(&e, type, sizep);
1085 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1086 status = error("unable to unpack %s header",
1087 sha1_to_hex(sha1));
1088 if (parse_sha1_header(hdr, type, &size) < 0)
1089 status = error("unable to parse %s header", sha1_to_hex(sha1));
1090 else {
1091 status = 0;
1092 if (sizep)
1093 *sizep = size;
1095 inflateEnd(&stream);
1096 munmap(map, mapsize);
1097 return status;
1100 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1102 struct pack_entry e;
1104 if (!find_pack_entry(sha1, &e)) {
1105 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1106 return NULL;
1108 return unpack_entry(&e, type, size);
1111 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1113 unsigned long mapsize;
1114 void *map, *buf;
1115 struct pack_entry e;
1117 if (find_pack_entry(sha1, &e))
1118 return read_packed_sha1(sha1, type, size);
1119 map = map_sha1_file_internal(sha1, &mapsize);
1120 if (map) {
1121 buf = unpack_sha1_file(map, mapsize, type, size);
1122 munmap(map, mapsize);
1123 return buf;
1125 return NULL;
1128 void *read_object_with_reference(const unsigned char *sha1,
1129 const char *required_type,
1130 unsigned long *size,
1131 unsigned char *actual_sha1_return)
1133 char type[20];
1134 void *buffer;
1135 unsigned long isize;
1136 unsigned char actual_sha1[20];
1138 memcpy(actual_sha1, sha1, 20);
1139 while (1) {
1140 int ref_length = -1;
1141 const char *ref_type = NULL;
1143 buffer = read_sha1_file(actual_sha1, type, &isize);
1144 if (!buffer)
1145 return NULL;
1146 if (!strcmp(type, required_type)) {
1147 *size = isize;
1148 if (actual_sha1_return)
1149 memcpy(actual_sha1_return, actual_sha1, 20);
1150 return buffer;
1152 /* Handle references */
1153 else if (!strcmp(type, "commit"))
1154 ref_type = "tree ";
1155 else if (!strcmp(type, "tag"))
1156 ref_type = "object ";
1157 else {
1158 free(buffer);
1159 return NULL;
1161 ref_length = strlen(ref_type);
1163 if (memcmp(buffer, ref_type, ref_length) ||
1164 get_sha1_hex(buffer + ref_length, actual_sha1)) {
1165 free(buffer);
1166 return NULL;
1168 free(buffer);
1169 /* Now we have the ID of the referred-to object in
1170 * actual_sha1. Check again. */
1174 char *write_sha1_file_prepare(void *buf,
1175 unsigned long len,
1176 const char *type,
1177 unsigned char *sha1,
1178 unsigned char *hdr,
1179 int *hdrlen)
1181 SHA_CTX c;
1183 /* Generate the header */
1184 *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1186 /* Sha1.. */
1187 SHA1_Init(&c);
1188 SHA1_Update(&c, hdr, *hdrlen);
1189 SHA1_Update(&c, buf, len);
1190 SHA1_Final(sha1, &c);
1192 return sha1_file_name(sha1);
1196 * Link the tempfile to the final place, possibly creating the
1197 * last directory level as you do so.
1199 * Returns the errno on failure, 0 on success.
1201 static int link_temp_to_file(const char *tmpfile, char *filename)
1203 int ret;
1205 if (!link(tmpfile, filename))
1206 return 0;
1209 * Try to mkdir the last path component if that failed
1210 * with an ENOENT.
1212 * Re-try the "link()" regardless of whether the mkdir
1213 * succeeds, since a race might mean that somebody
1214 * else succeeded.
1216 ret = errno;
1217 if (ret == ENOENT) {
1218 char *dir = strrchr(filename, '/');
1219 if (dir) {
1220 *dir = 0;
1221 mkdir(filename, 0777);
1222 *dir = '/';
1223 if (!link(tmpfile, filename))
1224 return 0;
1225 ret = errno;
1228 return ret;
1232 * Move the just written object into its final resting place
1234 int move_temp_to_file(const char *tmpfile, char *filename)
1236 int ret = link_temp_to_file(tmpfile, filename);
1239 * Coda hack - coda doesn't like cross-directory links,
1240 * so we fall back to a rename, which will mean that it
1241 * won't be able to check collisions, but that's not a
1242 * big deal.
1244 * The same holds for FAT formatted media.
1246 * When this succeeds, we just return 0. We have nothing
1247 * left to unlink.
1249 if (ret && ret != EEXIST) {
1250 if (!rename(tmpfile, filename))
1251 return 0;
1252 ret = errno;
1254 unlink(tmpfile);
1255 if (ret) {
1256 if (ret != EEXIST) {
1257 fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1258 return -1;
1260 /* FIXME!!! Collision check here ? */
1263 return 0;
1266 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1268 int size;
1269 unsigned char *compressed;
1270 z_stream stream;
1271 unsigned char sha1[20];
1272 char *filename;
1273 static char tmpfile[PATH_MAX];
1274 unsigned char hdr[50];
1275 int fd, hdrlen;
1277 /* Normally if we have it in the pack then we do not bother writing
1278 * it out into .git/objects/??/?{38} file.
1280 filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1281 if (returnsha1)
1282 memcpy(returnsha1, sha1, 20);
1283 if (has_sha1_file(sha1))
1284 return 0;
1285 fd = open(filename, O_RDONLY);
1286 if (fd >= 0) {
1288 * FIXME!!! We might do collision checking here, but we'd
1289 * need to uncompress the old file and check it. Later.
1291 close(fd);
1292 return 0;
1295 if (errno != ENOENT) {
1296 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1297 return -1;
1300 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1302 fd = mkstemp(tmpfile);
1303 if (fd < 0) {
1304 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1305 return -1;
1308 /* Set it up */
1309 memset(&stream, 0, sizeof(stream));
1310 deflateInit(&stream, Z_BEST_COMPRESSION);
1311 size = deflateBound(&stream, len+hdrlen);
1312 compressed = xmalloc(size);
1314 /* Compress it */
1315 stream.next_out = compressed;
1316 stream.avail_out = size;
1318 /* First header.. */
1319 stream.next_in = hdr;
1320 stream.avail_in = hdrlen;
1321 while (deflate(&stream, 0) == Z_OK)
1322 /* nothing */;
1324 /* Then the data itself.. */
1325 stream.next_in = buf;
1326 stream.avail_in = len;
1327 while (deflate(&stream, Z_FINISH) == Z_OK)
1328 /* nothing */;
1329 deflateEnd(&stream);
1330 size = stream.total_out;
1332 if (write(fd, compressed, size) != size)
1333 die("unable to write file");
1334 fchmod(fd, 0444);
1335 close(fd);
1336 free(compressed);
1338 return move_temp_to_file(tmpfile, filename);
1341 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1343 ssize_t size;
1344 unsigned long objsize;
1345 int posn = 0;
1346 void *map = map_sha1_file_internal(sha1, &objsize);
1347 void *buf = map;
1348 void *temp_obj = NULL;
1349 z_stream stream;
1351 if (!buf) {
1352 unsigned char *unpacked;
1353 unsigned long len;
1354 char type[20];
1355 char hdr[50];
1356 int hdrlen;
1357 // need to unpack and recompress it by itself
1358 unpacked = read_packed_sha1(sha1, type, &len);
1360 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1362 /* Set it up */
1363 memset(&stream, 0, sizeof(stream));
1364 deflateInit(&stream, Z_BEST_COMPRESSION);
1365 size = deflateBound(&stream, len + hdrlen);
1366 temp_obj = buf = xmalloc(size);
1368 /* Compress it */
1369 stream.next_out = buf;
1370 stream.avail_out = size;
1372 /* First header.. */
1373 stream.next_in = (void *)hdr;
1374 stream.avail_in = hdrlen;
1375 while (deflate(&stream, 0) == Z_OK)
1376 /* nothing */;
1378 /* Then the data itself.. */
1379 stream.next_in = unpacked;
1380 stream.avail_in = len;
1381 while (deflate(&stream, Z_FINISH) == Z_OK)
1382 /* nothing */;
1383 deflateEnd(&stream);
1384 free(unpacked);
1386 objsize = stream.total_out;
1389 do {
1390 size = write(fd, buf + posn, objsize - posn);
1391 if (size <= 0) {
1392 if (!size) {
1393 fprintf(stderr, "write closed");
1394 } else {
1395 perror("write ");
1397 return -1;
1399 posn += size;
1400 } while (posn < objsize);
1402 if (map)
1403 munmap(map, objsize);
1404 if (temp_obj)
1405 free(temp_obj);
1407 return 0;
1410 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1411 size_t bufsize, size_t *bufposn)
1413 char tmpfile[PATH_MAX];
1414 int local;
1415 z_stream stream;
1416 unsigned char real_sha1[20];
1417 unsigned char discard[4096];
1418 int ret;
1419 SHA_CTX c;
1421 snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1423 local = mkstemp(tmpfile);
1424 if (local < 0)
1425 return error("Couldn't open %s for %s\n", tmpfile, sha1_to_hex(sha1));
1427 memset(&stream, 0, sizeof(stream));
1429 inflateInit(&stream);
1431 SHA1_Init(&c);
1433 do {
1434 ssize_t size;
1435 if (*bufposn) {
1436 stream.avail_in = *bufposn;
1437 stream.next_in = (unsigned char *) buffer;
1438 do {
1439 stream.next_out = discard;
1440 stream.avail_out = sizeof(discard);
1441 ret = inflate(&stream, Z_SYNC_FLUSH);
1442 SHA1_Update(&c, discard, sizeof(discard) -
1443 stream.avail_out);
1444 } while (stream.avail_in && ret == Z_OK);
1445 write(local, buffer, *bufposn - stream.avail_in);
1446 memmove(buffer, buffer + *bufposn - stream.avail_in,
1447 stream.avail_in);
1448 *bufposn = stream.avail_in;
1449 if (ret != Z_OK)
1450 break;
1452 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1453 if (size <= 0) {
1454 close(local);
1455 unlink(tmpfile);
1456 if (!size)
1457 return error("Connection closed?");
1458 perror("Reading from connection");
1459 return -1;
1461 *bufposn += size;
1462 } while (1);
1463 inflateEnd(&stream);
1465 close(local);
1466 SHA1_Final(real_sha1, &c);
1467 if (ret != Z_STREAM_END) {
1468 unlink(tmpfile);
1469 return error("File %s corrupted", sha1_to_hex(sha1));
1471 if (memcmp(sha1, real_sha1, 20)) {
1472 unlink(tmpfile);
1473 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1476 return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1479 int has_pack_index(const unsigned char *sha1)
1481 struct stat st;
1482 if (stat(sha1_pack_index_name(sha1), &st))
1483 return 0;
1484 return 1;
1487 int has_pack_file(const unsigned char *sha1)
1489 struct stat st;
1490 if (stat(sha1_pack_name(sha1), &st))
1491 return 0;
1492 return 1;
1495 int has_sha1_pack(const unsigned char *sha1)
1497 struct pack_entry e;
1498 return find_pack_entry(sha1, &e);
1501 int has_sha1_file(const unsigned char *sha1)
1503 struct stat st;
1504 struct pack_entry e;
1506 if (find_pack_entry(sha1, &e))
1507 return 1;
1508 return find_sha1_file(sha1, &st) ? 1 : 0;
1511 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1513 unsigned long size = st->st_size;
1514 void *buf;
1515 int ret;
1516 unsigned char hdr[50];
1517 int hdrlen;
1519 buf = "";
1520 if (size)
1521 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1522 close(fd);
1523 if (buf == MAP_FAILED)
1524 return -1;
1526 if (!type)
1527 type = "blob";
1528 if (write_object)
1529 ret = write_sha1_file(buf, size, type, sha1);
1530 else {
1531 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1532 ret = 0;
1534 if (size)
1535 munmap(buf, size);
1536 return ret;
1539 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1541 int fd;
1542 char *target;
1544 switch (st->st_mode & S_IFMT) {
1545 case S_IFREG:
1546 fd = open(path, O_RDONLY);
1547 if (fd < 0)
1548 return error("open(\"%s\"): %s", path,
1549 strerror(errno));
1550 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1551 return error("%s: failed to insert into database",
1552 path);
1553 break;
1554 case S_IFLNK:
1555 target = xmalloc(st->st_size+1);
1556 if (readlink(path, target, st->st_size+1) != st->st_size) {
1557 char *errstr = strerror(errno);
1558 free(target);
1559 return error("readlink(\"%s\"): %s", path,
1560 errstr);
1562 if (!write_object) {
1563 unsigned char hdr[50];
1564 int hdrlen;
1565 write_sha1_file_prepare(target, st->st_size, "blob",
1566 sha1, hdr, &hdrlen);
1567 } else if (write_sha1_file(target, st->st_size, "blob", sha1))
1568 return error("%s: failed to insert into database",
1569 path);
1570 free(target);
1571 break;
1572 default:
1573 return error("%s: unsupported file type", path);
1575 return 0;