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