git-http-fetch needs $(X)
[git] / sha1_file.c
1 /*
2  * GIT - The information manager from hell
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  *
6  * This handles basic git sha1 object files - packing, unpacking,
7  * creation etc.
8  */
9 #include <sys/types.h>
10 #include <dirent.h>
11 #include "cache.h"
12 #include "delta.h"
13 #include "pack.h"
14
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
22
23 static unsigned int sha1_file_open_flag = O_NOATIME;
24
25 static unsigned hexval(char c)
26 {
27         if (c >= '0' && c <= '9')
28                 return c - '0';
29         if (c >= 'a' && c <= 'f')
30                 return c - 'a' + 10;
31         if (c >= 'A' && c <= 'F')
32                 return c - 'A' + 10;
33         return ~0;
34 }
35
36 int get_sha1_hex(const char *hex, unsigned char *sha1)
37 {
38         int i;
39         for (i = 0; i < 20; i++) {
40                 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
41                 if (val & ~0xff)
42                         return -1;
43                 *sha1++ = val;
44                 hex += 2;
45         }
46         return 0;
47 }
48
49 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir,
50         *git_graft_file;
51 static void setup_git_env(void)
52 {
53         git_dir = getenv(GIT_DIR_ENVIRONMENT);
54         if (!git_dir)
55                 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
56         git_object_dir = getenv(DB_ENVIRONMENT);
57         if (!git_object_dir) {
58                 git_object_dir = xmalloc(strlen(git_dir) + 9);
59                 sprintf(git_object_dir, "%s/objects", git_dir);
60         }
61         git_refs_dir = xmalloc(strlen(git_dir) + 6);
62         sprintf(git_refs_dir, "%s/refs", git_dir);
63         git_index_file = getenv(INDEX_ENVIRONMENT);
64         if (!git_index_file) {
65                 git_index_file = xmalloc(strlen(git_dir) + 7);
66                 sprintf(git_index_file, "%s/index", git_dir);
67         }
68         git_graft_file = getenv(GRAFT_ENVIRONMENT);
69         if (!git_graft_file)
70                 git_graft_file = strdup(git_path("info/grafts"));
71 }
72
73 char *get_git_dir(void)
74 {
75         if (!git_dir)
76                 setup_git_env();
77         return git_dir;
78 }
79
80 char *get_object_directory(void)
81 {
82         if (!git_object_dir)
83                 setup_git_env();
84         return git_object_dir;
85 }
86
87 char *get_refs_directory(void)
88 {
89         if (!git_refs_dir)
90                 setup_git_env();
91         return git_refs_dir;
92 }
93
94 char *get_index_file(void)
95 {
96         if (!git_index_file)
97                 setup_git_env();
98         return git_index_file;
99 }
100
101 char *get_graft_file(void)
102 {
103         if (!git_graft_file)
104                 setup_git_env();
105         return git_graft_file;
106 }
107
108 int safe_create_leading_directories(char *path)
109 {
110         char *pos = path;
111
112         while (pos) {
113                 pos = strchr(pos, '/');
114                 if (!pos)
115                         break;
116                 *pos = 0;
117                 if (mkdir(path, 0777) < 0)
118                         if (errno != EEXIST) {
119                                 *pos = '/';
120                                 return -1;
121                         }
122                 *pos++ = '/';
123         }
124         return 0;
125 }
126
127 char * sha1_to_hex(const unsigned char *sha1)
128 {
129         static char buffer[50];
130         static const char hex[] = "0123456789abcdef";
131         char *buf = buffer;
132         int i;
133
134         for (i = 0; i < 20; i++) {
135                 unsigned int val = *sha1++;
136                 *buf++ = hex[val >> 4];
137                 *buf++ = hex[val & 0xf];
138         }
139         return buffer;
140 }
141
142 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
143 {
144         int i;
145         for (i = 0; i < 20; i++) {
146                 static char hex[] = "0123456789abcdef";
147                 unsigned int val = sha1[i];
148                 char *pos = pathbuf + i*2 + (i > 0);
149                 *pos++ = hex[val >> 4];
150                 *pos = hex[val & 0xf];
151         }
152 }
153
154 /*
155  * NOTE! This returns a statically allocated buffer, so you have to be
156  * careful about using it. Do a "strdup()" if you need to save the
157  * filename.
158  *
159  * Also note that this returns the location for creating.  Reading
160  * SHA1 file can happen from any alternate directory listed in the
161  * DB_ENVIRONMENT environment variable if it is not found in
162  * the primary object database.
163  */
164 char *sha1_file_name(const unsigned char *sha1)
165 {
166         static char *name, *base;
167
168         if (!base) {
169                 const char *sha1_file_directory = get_object_directory();
170                 int len = strlen(sha1_file_directory);
171                 base = xmalloc(len + 60);
172                 memcpy(base, sha1_file_directory, len);
173                 memset(base+len, 0, 60);
174                 base[len] = '/';
175                 base[len+3] = '/';
176                 name = base + len + 1;
177         }
178         fill_sha1_path(name, sha1);
179         return base;
180 }
181
182 char *sha1_pack_name(const unsigned char *sha1)
183 {
184         static const char hex[] = "0123456789abcdef";
185         static char *name, *base, *buf;
186         int i;
187
188         if (!base) {
189                 const char *sha1_file_directory = get_object_directory();
190                 int len = strlen(sha1_file_directory);
191                 base = xmalloc(len + 60);
192                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
193                 name = base + len + 11;
194         }
195
196         buf = name;
197
198         for (i = 0; i < 20; i++) {
199                 unsigned int val = *sha1++;
200                 *buf++ = hex[val >> 4];
201                 *buf++ = hex[val & 0xf];
202         }
203         
204         return base;
205 }
206
207 char *sha1_pack_index_name(const unsigned char *sha1)
208 {
209         static const char hex[] = "0123456789abcdef";
210         static char *name, *base, *buf;
211         int i;
212
213         if (!base) {
214                 const char *sha1_file_directory = get_object_directory();
215                 int len = strlen(sha1_file_directory);
216                 base = xmalloc(len + 60);
217                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
218                 name = base + len + 11;
219         }
220
221         buf = name;
222
223         for (i = 0; i < 20; i++) {
224                 unsigned int val = *sha1++;
225                 *buf++ = hex[val >> 4];
226                 *buf++ = hex[val & 0xf];
227         }
228         
229         return base;
230 }
231
232 struct alternate_object_database *alt_odb_list;
233 static struct alternate_object_database **alt_odb_tail;
234
235 /*
236  * Prepare alternate object database registry.
237  *
238  * The variable alt_odb_list points at the list of struct
239  * alternate_object_database.  The elements on this list come from
240  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
242  * whose contents is exactly in the same format as that environment
243  * variable.  Its base points at a statically allocated buffer that
244  * contains "/the/directory/corresponding/to/.git/objects/...", while
245  * its name points just after the slash at the end of ".git/objects/"
246  * in the example above, and has enough space to hold 40-byte hex
247  * SHA1, an extra slash for the first level indirection, and the
248  * terminating NUL.
249  */
250 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
251                                  const char *relative_base)
252 {
253         const char *cp, *last;
254         struct alternate_object_database *ent;
255         int base_len = -1;
256
257         last = alt;
258         while (last < ep) {
259                 cp = last;
260                 if (cp < ep && *cp == '#') {
261                         while (cp < ep && *cp != sep)
262                                 cp++;
263                         last = cp + 1;
264                         continue;
265                 }
266                 for ( ; cp < ep && *cp != sep; cp++)
267                         ;
268                 if (last != cp) {
269                         /* 43 = 40-byte + 2 '/' + terminating NUL */
270                         int pfxlen = cp - last;
271                         int entlen = pfxlen + 43;
272
273                         if (*last != '/' && relative_base) {
274                                 /* Relative alt-odb */
275                                 if (base_len < 0)
276                                         base_len = strlen(relative_base) + 1;
277                                 entlen += base_len;
278                                 pfxlen += base_len;
279                         }
280                         ent = xmalloc(sizeof(*ent) + entlen);
281                         *alt_odb_tail = ent;
282                         alt_odb_tail = &(ent->next);
283                         ent->next = NULL;
284                         if (*last != '/' && relative_base) {
285                                 memcpy(ent->base, relative_base, base_len - 1);
286                                 ent->base[base_len - 1] = '/';
287                                 memcpy(ent->base + base_len,
288                                        last, cp - last);
289                         }
290                         else
291                                 memcpy(ent->base, last, pfxlen);
292                         ent->name = ent->base + pfxlen + 1;
293                         ent->base[pfxlen] = ent->base[pfxlen + 3] = '/';
294                         ent->base[entlen-1] = 0;
295                 }
296                 while (cp < ep && *cp == sep)
297                         cp++;
298                 last = cp;
299         }
300 }
301
302 void prepare_alt_odb(void)
303 {
304         char path[PATH_MAX];
305         char *map;
306         int fd;
307         struct stat st;
308         char *alt;
309
310         alt = getenv(ALTERNATE_DB_ENVIRONMENT);
311         if (!alt) alt = "";
312
313         if (alt_odb_tail)
314                 return;
315         alt_odb_tail = &alt_odb_list;
316         link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL);
317
318         sprintf(path, "%s/info/alternates", get_object_directory());
319         fd = open(path, O_RDONLY);
320         if (fd < 0)
321                 return;
322         if (fstat(fd, &st) || (st.st_size == 0)) {
323                 close(fd);
324                 return;
325         }
326         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
327         close(fd);
328         if (map == MAP_FAILED)
329                 return;
330
331         link_alt_odb_entries(map, map + st.st_size, '\n',
332                              get_object_directory());
333         munmap(map, st.st_size);
334 }
335
336 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
337 {
338         char *name = sha1_file_name(sha1);
339         struct alternate_object_database *alt;
340
341         if (!stat(name, st))
342                 return name;
343         prepare_alt_odb();
344         for (alt = alt_odb_list; alt; alt = alt->next) {
345                 name = alt->name;
346                 fill_sha1_path(name, sha1);
347                 if (!stat(alt->base, st))
348                         return alt->base;
349         }
350         return NULL;
351 }
352
353 #define PACK_MAX_SZ (1<<26)
354 static int pack_used_ctr;
355 static unsigned long pack_mapped;
356 struct packed_git *packed_git;
357
358 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
359                                 void **idx_map_)
360 {
361         void *idx_map;
362         unsigned int *index;
363         unsigned long idx_size;
364         int nr, i;
365         int fd = open(path, O_RDONLY);
366         struct stat st;
367         if (fd < 0)
368                 return -1;
369         if (fstat(fd, &st)) {
370                 close(fd);
371                 return -1;
372         }
373         idx_size = st.st_size;
374         idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
375         close(fd);
376         if (idx_map == MAP_FAILED)
377                 return -1;
378
379         index = idx_map;
380         *idx_map_ = idx_map;
381         *idx_size_ = idx_size;
382
383         /* check index map */
384         if (idx_size < 4*256 + 20 + 20)
385                 return error("index file too small");
386         nr = 0;
387         for (i = 0; i < 256; i++) {
388                 unsigned int n = ntohl(index[i]);
389                 if (n < nr)
390                         return error("non-monotonic index");
391                 nr = n;
392         }
393
394         /*
395          * Total size:
396          *  - 256 index entries 4 bytes each
397          *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
398          *  - 20-byte SHA1 of the packfile
399          *  - 20-byte SHA1 file checksum
400          */
401         if (idx_size != 4*256 + nr * 24 + 20 + 20)
402                 return error("wrong index file size");
403
404         return 0;
405 }
406
407 static int unuse_one_packed_git(void)
408 {
409         struct packed_git *p, *lru = NULL;
410
411         for (p = packed_git; p; p = p->next) {
412                 if (p->pack_use_cnt || !p->pack_base)
413                         continue;
414                 if (!lru || p->pack_last_used < lru->pack_last_used)
415                         lru = p;
416         }
417         if (!lru)
418                 return 0;
419         munmap(lru->pack_base, lru->pack_size);
420         lru->pack_base = NULL;
421         return 1;
422 }
423
424 void unuse_packed_git(struct packed_git *p)
425 {
426         p->pack_use_cnt--;
427 }
428
429 int use_packed_git(struct packed_git *p)
430 {
431         if (!p->pack_size) {
432                 struct stat st;
433                 // We created the struct before we had the pack
434                 stat(p->pack_name, &st);
435                 if (!S_ISREG(st.st_mode))
436                         die("packfile %s not a regular file", p->pack_name);
437                 p->pack_size = st.st_size;
438         }
439         if (!p->pack_base) {
440                 int fd;
441                 struct stat st;
442                 void *map;
443
444                 pack_mapped += p->pack_size;
445                 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
446                         ; /* nothing */
447                 fd = open(p->pack_name, O_RDONLY);
448                 if (fd < 0)
449                         die("packfile %s cannot be opened", p->pack_name);
450                 if (fstat(fd, &st)) {
451                         close(fd);
452                         die("packfile %s cannot be opened", p->pack_name);
453                 }
454                 if (st.st_size != p->pack_size)
455                         die("packfile %s size mismatch.", p->pack_name);
456                 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
457                 close(fd);
458                 if (map == MAP_FAILED)
459                         die("packfile %s cannot be mapped.", p->pack_name);
460                 p->pack_base = map;
461
462                 /* Check if the pack file matches with the index file.
463                  * this is cheap.
464                  */
465                 if (memcmp((char*)(p->index_base) + p->index_size - 40,
466                            p->pack_base + p->pack_size - 20, 20)) {
467                               
468                         die("packfile %s does not match index.", p->pack_name);
469                 }
470         }
471         p->pack_last_used = pack_used_ctr++;
472         p->pack_use_cnt++;
473         return 0;
474 }
475
476 struct packed_git *add_packed_git(char *path, int path_len)
477 {
478         struct stat st;
479         struct packed_git *p;
480         unsigned long idx_size;
481         void *idx_map;
482
483         if (check_packed_git_idx(path, &idx_size, &idx_map))
484                 return NULL;
485
486         /* do we have a corresponding .pack file? */
487         strcpy(path + path_len - 4, ".pack");
488         if (stat(path, &st) || !S_ISREG(st.st_mode)) {
489                 munmap(idx_map, idx_size);
490                 return NULL;
491         }
492         /* ok, it looks sane as far as we can check without
493          * actually mapping the pack file.
494          */
495         p = xmalloc(sizeof(*p) + path_len + 2);
496         strcpy(p->pack_name, path);
497         p->index_size = idx_size;
498         p->pack_size = st.st_size;
499         p->index_base = idx_map;
500         p->next = NULL;
501         p->pack_base = NULL;
502         p->pack_last_used = 0;
503         p->pack_use_cnt = 0;
504         return p;
505 }
506
507 struct packed_git *parse_pack_index(unsigned char *sha1)
508 {
509         char *path = sha1_pack_index_name(sha1);
510         return parse_pack_index_file(sha1, path);
511 }
512
513 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
514 {
515         struct packed_git *p;
516         unsigned long idx_size;
517         void *idx_map;
518         char *path;
519
520         if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
521                 return NULL;
522
523         path = sha1_pack_name(sha1);
524
525         p = xmalloc(sizeof(*p) + strlen(path) + 2);
526         strcpy(p->pack_name, path);
527         p->index_size = idx_size;
528         p->pack_size = 0;
529         p->index_base = idx_map;
530         p->next = NULL;
531         p->pack_base = NULL;
532         p->pack_last_used = 0;
533         p->pack_use_cnt = 0;
534         memcpy(p->sha1, sha1, 20);
535         return p;
536 }
537
538 void install_packed_git(struct packed_git *pack)
539 {
540         pack->next = packed_git;
541         packed_git = pack;
542 }
543
544 static void prepare_packed_git_one(char *objdir)
545 {
546         char path[PATH_MAX];
547         int len;
548         DIR *dir;
549         struct dirent *de;
550
551         sprintf(path, "%s/pack", objdir);
552         len = strlen(path);
553         dir = opendir(path);
554         if (!dir)
555                 return;
556         path[len++] = '/';
557         while ((de = readdir(dir)) != NULL) {
558                 int namelen = strlen(de->d_name);
559                 struct packed_git *p;
560
561                 if (strcmp(de->d_name + namelen - 4, ".idx"))
562                         continue;
563
564                 /* we have .idx.  Is it a file we can map? */
565                 strcpy(path + len, de->d_name);
566                 p = add_packed_git(path, len + namelen);
567                 if (!p)
568                         continue;
569                 p->next = packed_git;
570                 packed_git = p;
571         }
572         closedir(dir);
573 }
574
575 void prepare_packed_git(void)
576 {
577         static int run_once = 0;
578         struct alternate_object_database *alt;
579
580         if (run_once)
581                 return;
582         prepare_packed_git_one(get_object_directory());
583         prepare_alt_odb();
584         for (alt = alt_odb_list; alt; alt = alt->next) {
585                 alt->name[0] = 0;
586                 prepare_packed_git_one(alt->base);
587         }
588         run_once = 1;
589 }
590
591 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
592 {
593         char header[100];
594         unsigned char real_sha1[20];
595         SHA_CTX c;
596
597         SHA1_Init(&c);
598         SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
599         SHA1_Update(&c, map, size);
600         SHA1_Final(real_sha1, &c);
601         return memcmp(sha1, real_sha1, 20) ? -1 : 0;
602 }
603
604 static void *map_sha1_file_internal(const unsigned char *sha1,
605                                     unsigned long *size)
606 {
607         struct stat st;
608         void *map;
609         int fd;
610         char *filename = find_sha1_file(sha1, &st);
611
612         if (!filename) {
613                 return NULL;
614         }
615
616         fd = open(filename, O_RDONLY | sha1_file_open_flag);
617         if (fd < 0) {
618                 /* See if it works without O_NOATIME */
619                 switch (sha1_file_open_flag) {
620                 default:
621                         fd = open(filename, O_RDONLY);
622                         if (fd >= 0)
623                                 break;
624                 /* Fallthrough */
625                 case 0:
626                         return NULL;
627                 }
628
629                 /* If it failed once, it will probably fail again.
630                  * Stop using O_NOATIME
631                  */
632                 sha1_file_open_flag = 0;
633         }
634         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
635         close(fd);
636         if (map == MAP_FAILED)
637                 return NULL;
638         *size = st.st_size;
639         return map;
640 }
641
642 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
643 {
644         /* Get the data stream */
645         memset(stream, 0, sizeof(*stream));
646         stream->next_in = map;
647         stream->avail_in = mapsize;
648         stream->next_out = buffer;
649         stream->avail_out = size;
650
651         inflateInit(stream);
652         return inflate(stream, 0);
653 }
654
655 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
656 {
657         int bytes = strlen(buffer) + 1;
658         unsigned char *buf = xmalloc(1+size);
659
660         memcpy(buf, buffer + bytes, stream->total_out - bytes);
661         bytes = stream->total_out - bytes;
662         if (bytes < size) {
663                 stream->next_out = buf + bytes;
664                 stream->avail_out = size - bytes;
665                 while (inflate(stream, Z_FINISH) == Z_OK)
666                         /* nothing */;
667         }
668         buf[size] = 0;
669         inflateEnd(stream);
670         return buf;
671 }
672
673 /*
674  * We used to just use "sscanf()", but that's actually way
675  * too permissive for what we want to check. So do an anal
676  * object header parse by hand.
677  */
678 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
679 {
680         int i;
681         unsigned long size;
682
683         /*
684          * The type can be at most ten bytes (including the 
685          * terminating '\0' that we add), and is followed by
686          * a space. 
687          */
688         i = 10;
689         for (;;) {
690                 char c = *hdr++;
691                 if (c == ' ')
692                         break;
693                 if (!--i)
694                         return -1;
695                 *type++ = c;
696         }
697         *type = 0;
698
699         /*
700          * The length must follow immediately, and be in canonical
701          * decimal format (ie "010" is not valid).
702          */
703         size = *hdr++ - '0';
704         if (size > 9)
705                 return -1;
706         if (size) {
707                 for (;;) {
708                         unsigned long c = *hdr - '0';
709                         if (c > 9)
710                                 break;
711                         hdr++;
712                         size = size * 10 + c;
713                 }
714         }
715         *sizep = size;
716
717         /*
718          * The length must be followed by a zero byte
719          */
720         return *hdr ? -1 : 0;
721 }
722
723 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
724 {
725         int ret;
726         z_stream stream;
727         char hdr[8192];
728
729         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
730         if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
731                 return NULL;
732
733         return unpack_sha1_rest(&stream, hdr, *size);
734 }
735
736 /* forward declaration for a mutually recursive function */
737 static int packed_object_info(struct pack_entry *entry,
738                               char *type, unsigned long *sizep);
739
740 static int packed_delta_info(unsigned char *base_sha1,
741                              unsigned long delta_size,
742                              unsigned long left,
743                              char *type,
744                              unsigned long *sizep,
745                              struct packed_git *p)
746 {
747         struct pack_entry base_ent;
748
749         if (left < 20)
750                 die("truncated pack file");
751
752         /* The base entry _must_ be in the same pack */
753         if (!find_pack_entry_one(base_sha1, &base_ent, p))
754                 die("failed to find delta-pack base object %s",
755                     sha1_to_hex(base_sha1));
756
757         /* We choose to only get the type of the base object and
758          * ignore potentially corrupt pack file that expects the delta
759          * based on a base with a wrong size.  This saves tons of
760          * inflate() calls.
761          */
762
763         if (packed_object_info(&base_ent, type, NULL))
764                 die("cannot get info for delta-pack base");
765
766         if (sizep) {
767                 const unsigned char *data;
768                 unsigned char delta_head[64];
769                 unsigned long result_size;
770                 z_stream stream;
771                 int st;
772
773                 memset(&stream, 0, sizeof(stream));
774
775                 data = stream.next_in = base_sha1 + 20;
776                 stream.avail_in = left - 20;
777                 stream.next_out = delta_head;
778                 stream.avail_out = sizeof(delta_head);
779
780                 inflateInit(&stream);
781                 st = inflate(&stream, Z_FINISH);
782                 inflateEnd(&stream);
783                 if ((st != Z_STREAM_END) &&
784                     stream.total_out != sizeof(delta_head))
785                         die("delta data unpack-initial failed");
786
787                 /* Examine the initial part of the delta to figure out
788                  * the result size.
789                  */
790                 data = delta_head;
791                 get_delta_hdr_size(&data); /* ignore base size */
792
793                 /* Read the result size */
794                 result_size = get_delta_hdr_size(&data);
795                 *sizep = result_size;
796         }
797         return 0;
798 }
799
800 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
801         enum object_type *type, unsigned long *sizep)
802 {
803         unsigned shift;
804         unsigned char *pack, c;
805         unsigned long size;
806
807         if (offset >= p->pack_size)
808                 die("object offset outside of pack file");
809
810         pack =  p->pack_base + offset;
811         c = *pack++;
812         offset++;
813         *type = (c >> 4) & 7;
814         size = c & 15;
815         shift = 4;
816         while (c & 0x80) {
817                 if (offset >= p->pack_size)
818                         die("object offset outside of pack file");
819                 c = *pack++;
820                 offset++;
821                 size += (c & 0x7f) << shift;
822                 shift += 7;
823         }
824         *sizep = size;
825         return offset;
826 }
827
828 void packed_object_info_detail(struct pack_entry *e,
829                                char *type,
830                                unsigned long *size,
831                                unsigned long *store_size,
832                                int *delta_chain_length,
833                                unsigned char *base_sha1)
834 {
835         struct packed_git *p = e->p;
836         unsigned long offset, left;
837         unsigned char *pack;
838         enum object_type kind;
839
840         offset = unpack_object_header(p, e->offset, &kind, size);
841         pack = p->pack_base + offset;
842         left = p->pack_size - offset;
843         if (kind != OBJ_DELTA)
844                 *delta_chain_length = 0;
845         else {
846                 int chain_length = 0;
847                 memcpy(base_sha1, pack, 20);
848                 do {
849                         struct pack_entry base_ent;
850                         unsigned long junk;
851
852                         find_pack_entry_one(pack, &base_ent, p);
853                         offset = unpack_object_header(p, base_ent.offset,
854                                                       &kind, &junk);
855                         pack = p->pack_base + offset;
856                         chain_length++;
857                 } while (kind == OBJ_DELTA);
858                 *delta_chain_length = chain_length;
859         }
860         switch (kind) {
861         case OBJ_COMMIT:
862                 strcpy(type, "commit");
863                 break;
864         case OBJ_TREE:
865                 strcpy(type, "tree");
866                 break;
867         case OBJ_BLOB:
868                 strcpy(type, "blob");
869                 break;
870         case OBJ_TAG:
871                 strcpy(type, "tag");
872                 break;
873         default:
874                 die("corrupted pack file");
875         }
876         *store_size = 0; /* notyet */
877 }
878
879 static int packed_object_info(struct pack_entry *entry,
880                               char *type, unsigned long *sizep)
881 {
882         struct packed_git *p = entry->p;
883         unsigned long offset, size, left;
884         unsigned char *pack;
885         enum object_type kind;
886         int retval;
887
888         if (use_packed_git(p))
889                 die("cannot map packed file");
890
891         offset = unpack_object_header(p, entry->offset, &kind, &size);
892         pack = p->pack_base + offset;
893         left = p->pack_size - offset;
894
895         switch (kind) {
896         case OBJ_DELTA:
897                 retval = packed_delta_info(pack, size, left, type, sizep, p);
898                 unuse_packed_git(p);
899                 return retval;
900         case OBJ_COMMIT:
901                 strcpy(type, "commit");
902                 break;
903         case OBJ_TREE:
904                 strcpy(type, "tree");
905                 break;
906         case OBJ_BLOB:
907                 strcpy(type, "blob");
908                 break;
909         case OBJ_TAG:
910                 strcpy(type, "tag");
911                 break;
912         default:
913                 die("corrupted pack file");
914         }
915         if (sizep)
916                 *sizep = size;
917         unuse_packed_git(p);
918         return 0;
919 }
920
921 /* forward declaration for a mutually recursive function */
922 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
923
924 static void *unpack_delta_entry(unsigned char *base_sha1,
925                                 unsigned long delta_size,
926                                 unsigned long left,
927                                 char *type,
928                                 unsigned long *sizep,
929                                 struct packed_git *p)
930 {
931         struct pack_entry base_ent;
932         void *data, *delta_data, *result, *base;
933         unsigned long data_size, result_size, base_size;
934         z_stream stream;
935         int st;
936
937         if (left < 20)
938                 die("truncated pack file");
939         data = base_sha1 + 20;
940         data_size = left - 20;
941         delta_data = xmalloc(delta_size);
942
943         memset(&stream, 0, sizeof(stream));
944
945         stream.next_in = data;
946         stream.avail_in = data_size;
947         stream.next_out = delta_data;
948         stream.avail_out = delta_size;
949
950         inflateInit(&stream);
951         st = inflate(&stream, Z_FINISH);
952         inflateEnd(&stream);
953         if ((st != Z_STREAM_END) || stream.total_out != delta_size)
954                 die("delta data unpack failed");
955
956         /* The base entry _must_ be in the same pack */
957         if (!find_pack_entry_one(base_sha1, &base_ent, p))
958                 die("failed to find delta-pack base object %s",
959                     sha1_to_hex(base_sha1));
960         base = unpack_entry_gently(&base_ent, type, &base_size);
961         if (!base)
962                 die("failed to read delta-pack base object %s",
963                     sha1_to_hex(base_sha1));
964         result = patch_delta(base, base_size,
965                              delta_data, delta_size,
966                              &result_size);
967         if (!result)
968                 die("failed to apply delta");
969         free(delta_data);
970         free(base);
971         *sizep = result_size;
972         return result;
973 }
974
975 static void *unpack_non_delta_entry(unsigned char *data,
976                                     unsigned long size,
977                                     unsigned long left)
978 {
979         int st;
980         z_stream stream;
981         unsigned char *buffer;
982
983         buffer = xmalloc(size + 1);
984         buffer[size] = 0;
985         memset(&stream, 0, sizeof(stream));
986         stream.next_in = data;
987         stream.avail_in = left;
988         stream.next_out = buffer;
989         stream.avail_out = size;
990
991         inflateInit(&stream);
992         st = inflate(&stream, Z_FINISH);
993         inflateEnd(&stream);
994         if ((st != Z_STREAM_END) || stream.total_out != size) {
995                 free(buffer);
996                 return NULL;
997         }
998
999         return buffer;
1000 }
1001
1002 static void *unpack_entry(struct pack_entry *entry,
1003                           char *type, unsigned long *sizep)
1004 {
1005         struct packed_git *p = entry->p;
1006         void *retval;
1007
1008         if (use_packed_git(p))
1009                 die("cannot map packed file");
1010         retval = unpack_entry_gently(entry, type, sizep);
1011         unuse_packed_git(p);
1012         if (!retval)
1013                 die("corrupted pack file");
1014         return retval;
1015 }
1016
1017 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1018 void *unpack_entry_gently(struct pack_entry *entry,
1019                           char *type, unsigned long *sizep)
1020 {
1021         struct packed_git *p = entry->p;
1022         unsigned long offset, size, left;
1023         unsigned char *pack;
1024         enum object_type kind;
1025         void *retval;
1026
1027         offset = unpack_object_header(p, entry->offset, &kind, &size);
1028         pack = p->pack_base + offset;
1029         left = p->pack_size - offset;
1030         switch (kind) {
1031         case OBJ_DELTA:
1032                 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1033                 return retval;
1034         case OBJ_COMMIT:
1035                 strcpy(type, "commit");
1036                 break;
1037         case OBJ_TREE:
1038                 strcpy(type, "tree");
1039                 break;
1040         case OBJ_BLOB:
1041                 strcpy(type, "blob");
1042                 break;
1043         case OBJ_TAG:
1044                 strcpy(type, "tag");
1045                 break;
1046         default:
1047                 return NULL;
1048         }
1049         *sizep = size;
1050         retval = unpack_non_delta_entry(pack, size, left);
1051         return retval;
1052 }
1053
1054 int num_packed_objects(const struct packed_git *p)
1055 {
1056         /* See check_packed_git_idx() */
1057         return (p->index_size - 20 - 20 - 4*256) / 24;
1058 }
1059
1060 int nth_packed_object_sha1(const struct packed_git *p, int n,
1061                            unsigned char* sha1)
1062 {
1063         void *index = p->index_base + 256;
1064         if (n < 0 || num_packed_objects(p) <= n)
1065                 return -1;
1066         memcpy(sha1, (index + 24 * n + 4), 20);
1067         return 0;
1068 }
1069
1070 int find_pack_entry_one(const unsigned char *sha1,
1071                         struct pack_entry *e, struct packed_git *p)
1072 {
1073         unsigned int *level1_ofs = p->index_base;
1074         int hi = ntohl(level1_ofs[*sha1]);
1075         int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1076         void *index = p->index_base + 256;
1077
1078         do {
1079                 int mi = (lo + hi) / 2;
1080                 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1081                 if (!cmp) {
1082                         e->offset = ntohl(*((int*)(index + 24 * mi)));
1083                         memcpy(e->sha1, sha1, 20);
1084                         e->p = p;
1085                         return 1;
1086                 }
1087                 if (cmp > 0)
1088                         hi = mi;
1089                 else
1090                         lo = mi+1;
1091         } while (lo < hi);
1092         return 0;
1093 }
1094
1095 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1096 {
1097         struct packed_git *p;
1098         prepare_packed_git();
1099
1100         for (p = packed_git; p; p = p->next) {
1101                 if (find_pack_entry_one(sha1, e, p))
1102                         return 1;
1103         }
1104         return 0;
1105 }
1106
1107 struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1108                                   struct packed_git *packs)
1109 {
1110         struct packed_git *p;
1111         struct pack_entry e;
1112
1113         for (p = packs; p; p = p->next) {
1114                 if (find_pack_entry_one(sha1, &e, p))
1115                         return p;
1116         }
1117         return NULL;
1118         
1119 }
1120
1121 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1122 {
1123         int status;
1124         unsigned long mapsize, size;
1125         void *map;
1126         z_stream stream;
1127         char hdr[128];
1128
1129         map = map_sha1_file_internal(sha1, &mapsize);
1130         if (!map) {
1131                 struct pack_entry e;
1132
1133                 if (!find_pack_entry(sha1, &e))
1134                         return error("unable to find %s", sha1_to_hex(sha1));
1135                 return packed_object_info(&e, type, sizep);
1136         }
1137         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1138                 status = error("unable to unpack %s header",
1139                                sha1_to_hex(sha1));
1140         if (parse_sha1_header(hdr, type, &size) < 0)
1141                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1142         else {
1143                 status = 0;
1144                 if (sizep)
1145                         *sizep = size;
1146         }
1147         inflateEnd(&stream);
1148         munmap(map, mapsize);
1149         return status;
1150 }
1151
1152 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1153 {
1154         struct pack_entry e;
1155
1156         if (!find_pack_entry(sha1, &e)) {
1157                 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1158                 return NULL;
1159         }
1160         return unpack_entry(&e, type, size);
1161 }
1162
1163 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1164 {
1165         unsigned long mapsize;
1166         void *map, *buf;
1167         struct pack_entry e;
1168
1169         if (find_pack_entry(sha1, &e))
1170                 return read_packed_sha1(sha1, type, size);
1171         map = map_sha1_file_internal(sha1, &mapsize);
1172         if (map) {
1173                 buf = unpack_sha1_file(map, mapsize, type, size);
1174                 munmap(map, mapsize);
1175                 return buf;
1176         }
1177         return NULL;
1178 }
1179
1180 void *read_object_with_reference(const unsigned char *sha1,
1181                                  const char *required_type,
1182                                  unsigned long *size,
1183                                  unsigned char *actual_sha1_return)
1184 {
1185         char type[20];
1186         void *buffer;
1187         unsigned long isize;
1188         unsigned char actual_sha1[20];
1189
1190         memcpy(actual_sha1, sha1, 20);
1191         while (1) {
1192                 int ref_length = -1;
1193                 const char *ref_type = NULL;
1194
1195                 buffer = read_sha1_file(actual_sha1, type, &isize);
1196                 if (!buffer)
1197                         return NULL;
1198                 if (!strcmp(type, required_type)) {
1199                         *size = isize;
1200                         if (actual_sha1_return)
1201                                 memcpy(actual_sha1_return, actual_sha1, 20);
1202                         return buffer;
1203                 }
1204                 /* Handle references */
1205                 else if (!strcmp(type, "commit"))
1206                         ref_type = "tree ";
1207                 else if (!strcmp(type, "tag"))
1208                         ref_type = "object ";
1209                 else {
1210                         free(buffer);
1211                         return NULL;
1212                 }
1213                 ref_length = strlen(ref_type);
1214
1215                 if (memcmp(buffer, ref_type, ref_length) ||
1216                     get_sha1_hex(buffer + ref_length, actual_sha1)) {
1217                         free(buffer);
1218                         return NULL;
1219                 }
1220                 free(buffer);
1221                 /* Now we have the ID of the referred-to object in
1222                  * actual_sha1.  Check again. */
1223         }
1224 }
1225
1226 char *write_sha1_file_prepare(void *buf,
1227                               unsigned long len,
1228                               const char *type,
1229                               unsigned char *sha1,
1230                               unsigned char *hdr,
1231                               int *hdrlen)
1232 {
1233         SHA_CTX c;
1234
1235         /* Generate the header */
1236         *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1237
1238         /* Sha1.. */
1239         SHA1_Init(&c);
1240         SHA1_Update(&c, hdr, *hdrlen);
1241         SHA1_Update(&c, buf, len);
1242         SHA1_Final(sha1, &c);
1243
1244         return sha1_file_name(sha1);
1245 }
1246
1247 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1248 {
1249         int size;
1250         unsigned char *compressed;
1251         z_stream stream;
1252         unsigned char sha1[20];
1253         char *filename;
1254         static char tmpfile[PATH_MAX];
1255         unsigned char hdr[50];
1256         int fd, hdrlen, ret;
1257
1258         /* Normally if we have it in the pack then we do not bother writing
1259          * it out into .git/objects/??/?{38} file.
1260          */
1261         filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1262         if (returnsha1)
1263                 memcpy(returnsha1, sha1, 20);
1264         if (has_sha1_file(sha1))
1265                 return 0;
1266         fd = open(filename, O_RDONLY);
1267         if (fd >= 0) {
1268                 /*
1269                  * FIXME!!! We might do collision checking here, but we'd
1270                  * need to uncompress the old file and check it. Later.
1271                  */
1272                 close(fd);
1273                 return 0;
1274         }
1275
1276         if (errno != ENOENT) {
1277                 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1278                 return -1;
1279         }
1280
1281         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1282
1283         fd = mkstemp(tmpfile);
1284         if (fd < 0) {
1285                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1286                 return -1;
1287         }
1288
1289         /* Set it up */
1290         memset(&stream, 0, sizeof(stream));
1291         deflateInit(&stream, Z_BEST_COMPRESSION);
1292         size = deflateBound(&stream, len+hdrlen);
1293         compressed = xmalloc(size);
1294
1295         /* Compress it */
1296         stream.next_out = compressed;
1297         stream.avail_out = size;
1298
1299         /* First header.. */
1300         stream.next_in = hdr;
1301         stream.avail_in = hdrlen;
1302         while (deflate(&stream, 0) == Z_OK)
1303                 /* nothing */;
1304
1305         /* Then the data itself.. */
1306         stream.next_in = buf;
1307         stream.avail_in = len;
1308         while (deflate(&stream, Z_FINISH) == Z_OK)
1309                 /* nothing */;
1310         deflateEnd(&stream);
1311         size = stream.total_out;
1312
1313         if (write(fd, compressed, size) != size)
1314                 die("unable to write file");
1315         fchmod(fd, 0444);
1316         close(fd);
1317         free(compressed);
1318
1319         ret = link(tmpfile, filename);
1320         if (ret < 0) {
1321                 ret = errno;
1322
1323                 /*
1324                  * Coda hack - coda doesn't like cross-directory links,
1325                  * so we fall back to a rename, which will mean that it
1326                  * won't be able to check collisions, but that's not a
1327                  * big deal.
1328                  *
1329                  * When this succeeds, we just return 0. We have nothing
1330                  * left to unlink.
1331                  */
1332                 if (ret == EXDEV && !rename(tmpfile, filename))
1333                         return 0;
1334         }
1335         unlink(tmpfile);
1336         if (ret) {
1337                 if (ret != EEXIST) {
1338                         fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1339                         return -1;
1340                 }
1341                 /* FIXME!!! Collision check here ? */
1342         }
1343
1344         return 0;
1345 }
1346
1347 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1348 {
1349         ssize_t size;
1350         unsigned long objsize;
1351         int posn = 0;
1352         void *map = map_sha1_file_internal(sha1, &objsize);
1353         void *buf = map;
1354         void *temp_obj = NULL;
1355         z_stream stream;
1356
1357         if (!buf) {
1358                 unsigned char *unpacked;
1359                 unsigned long len;
1360                 char type[20];
1361                 char hdr[50];
1362                 int hdrlen;
1363                 // need to unpack and recompress it by itself
1364                 unpacked = read_packed_sha1(sha1, type, &len);
1365
1366                 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1367
1368                 /* Set it up */
1369                 memset(&stream, 0, sizeof(stream));
1370                 deflateInit(&stream, Z_BEST_COMPRESSION);
1371                 size = deflateBound(&stream, len + hdrlen);
1372                 temp_obj = buf = xmalloc(size);
1373
1374                 /* Compress it */
1375                 stream.next_out = buf;
1376                 stream.avail_out = size;
1377                 
1378                 /* First header.. */
1379                 stream.next_in = (void *)hdr;
1380                 stream.avail_in = hdrlen;
1381                 while (deflate(&stream, 0) == Z_OK)
1382                         /* nothing */;
1383
1384                 /* Then the data itself.. */
1385                 stream.next_in = unpacked;
1386                 stream.avail_in = len;
1387                 while (deflate(&stream, Z_FINISH) == Z_OK)
1388                         /* nothing */;
1389                 deflateEnd(&stream);
1390                 free(unpacked);
1391                 
1392                 objsize = stream.total_out;
1393         }
1394
1395         do {
1396                 size = write(fd, buf + posn, objsize - posn);
1397                 if (size <= 0) {
1398                         if (!size) {
1399                                 fprintf(stderr, "write closed");
1400                         } else {
1401                                 perror("write ");
1402                         }
1403                         return -1;
1404                 }
1405                 posn += size;
1406         } while (posn < objsize);
1407
1408         if (map)
1409                 munmap(map, objsize);
1410         if (temp_obj)
1411                 free(temp_obj);
1412
1413         return 0;
1414 }
1415
1416 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1417                        size_t bufsize, size_t *bufposn)
1418 {
1419         char *filename = sha1_file_name(sha1);
1420
1421         int local;
1422         z_stream stream;
1423         unsigned char real_sha1[20];
1424         unsigned char discard[4096];
1425         int ret;
1426         SHA_CTX c;
1427
1428         local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1429
1430         if (local < 0)
1431                 return error("Couldn't open %s\n", filename);
1432
1433         memset(&stream, 0, sizeof(stream));
1434
1435         inflateInit(&stream);
1436
1437         SHA1_Init(&c);
1438
1439         do {
1440                 ssize_t size;
1441                 if (*bufposn) {
1442                         stream.avail_in = *bufposn;
1443                         stream.next_in = (unsigned char *) buffer;
1444                         do {
1445                                 stream.next_out = discard;
1446                                 stream.avail_out = sizeof(discard);
1447                                 ret = inflate(&stream, Z_SYNC_FLUSH);
1448                                 SHA1_Update(&c, discard, sizeof(discard) -
1449                                             stream.avail_out);
1450                         } while (stream.avail_in && ret == Z_OK);
1451                         write(local, buffer, *bufposn - stream.avail_in);
1452                         memmove(buffer, buffer + *bufposn - stream.avail_in,
1453                                 stream.avail_in);
1454                         *bufposn = stream.avail_in;
1455                         if (ret != Z_OK)
1456                                 break;
1457                 }
1458                 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1459                 if (size <= 0) {
1460                         close(local);
1461                         unlink(filename);
1462                         if (!size)
1463                                 return error("Connection closed?");
1464                         perror("Reading from connection");
1465                         return -1;
1466                 }
1467                 *bufposn += size;
1468         } while (1);
1469         inflateEnd(&stream);
1470
1471         close(local);
1472         SHA1_Final(real_sha1, &c);
1473         if (ret != Z_STREAM_END) {
1474                 unlink(filename);
1475                 return error("File %s corrupted", sha1_to_hex(sha1));
1476         }
1477         if (memcmp(sha1, real_sha1, 20)) {
1478                 unlink(filename);
1479                 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1480         }
1481         
1482         return 0;
1483 }
1484
1485 int has_pack_index(const unsigned char *sha1)
1486 {
1487         struct stat st;
1488         if (stat(sha1_pack_index_name(sha1), &st))
1489                 return 0;
1490         return 1;
1491 }
1492
1493 int has_pack_file(const unsigned char *sha1)
1494 {
1495         struct stat st;
1496         if (stat(sha1_pack_name(sha1), &st))
1497                 return 0;
1498         return 1;
1499 }
1500
1501 int has_sha1_pack(const unsigned char *sha1)
1502 {
1503         struct pack_entry e;
1504         return find_pack_entry(sha1, &e);
1505 }
1506
1507 int has_sha1_file(const unsigned char *sha1)
1508 {
1509         struct stat st;
1510         struct pack_entry e;
1511
1512         if (find_pack_entry(sha1, &e))
1513                 return 1;
1514         return find_sha1_file(sha1, &st) ? 1 : 0;
1515 }
1516
1517 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1518 {
1519         unsigned long size = st->st_size;
1520         void *buf;
1521         int ret;
1522         unsigned char hdr[50];
1523         int hdrlen;
1524
1525         buf = "";
1526         if (size)
1527                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1528         close(fd);
1529         if (buf == MAP_FAILED)
1530                 return -1;
1531
1532         if (!type)
1533                 type = "blob";
1534         if (write_object)
1535                 ret = write_sha1_file(buf, size, type, sha1);
1536         else {
1537                 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1538                 ret = 0;
1539         }
1540         if (size)
1541                 munmap(buf, size);
1542         return ret;
1543 }