Merge with master.kernel.org:/pub/scm/git/git.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 const unsigned char null_sha1[20] = { 0, };
24
25 static unsigned int sha1_file_open_flag = O_NOATIME;
26
27 static unsigned hexval(char c)
28 {
29         if (c >= '0' && c <= '9')
30                 return c - '0';
31         if (c >= 'a' && c <= 'f')
32                 return c - 'a' + 10;
33         if (c >= 'A' && c <= 'F')
34                 return c - 'A' + 10;
35         return ~0;
36 }
37
38 int get_sha1_hex(const char *hex, unsigned char *sha1)
39 {
40         int i;
41         for (i = 0; i < 20; i++) {
42                 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
43                 if (val & ~0xff)
44                         return -1;
45                 *sha1++ = val;
46                 hex += 2;
47         }
48         return 0;
49 }
50
51 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir,
52         *git_graft_file;
53 static void setup_git_env(void)
54 {
55         git_dir = getenv(GIT_DIR_ENVIRONMENT);
56         if (!git_dir)
57                 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
58         git_object_dir = getenv(DB_ENVIRONMENT);
59         if (!git_object_dir) {
60                 git_object_dir = xmalloc(strlen(git_dir) + 9);
61                 sprintf(git_object_dir, "%s/objects", git_dir);
62         }
63         git_refs_dir = xmalloc(strlen(git_dir) + 6);
64         sprintf(git_refs_dir, "%s/refs", git_dir);
65         git_index_file = getenv(INDEX_ENVIRONMENT);
66         if (!git_index_file) {
67                 git_index_file = xmalloc(strlen(git_dir) + 7);
68                 sprintf(git_index_file, "%s/index", git_dir);
69         }
70         git_graft_file = getenv(GRAFT_ENVIRONMENT);
71         if (!git_graft_file)
72                 git_graft_file = strdup(git_path("info/grafts"));
73 }
74
75 char *get_git_dir(void)
76 {
77         if (!git_dir)
78                 setup_git_env();
79         return git_dir;
80 }
81
82 char *get_object_directory(void)
83 {
84         if (!git_object_dir)
85                 setup_git_env();
86         return git_object_dir;
87 }
88
89 char *get_refs_directory(void)
90 {
91         if (!git_refs_dir)
92                 setup_git_env();
93         return git_refs_dir;
94 }
95
96 char *get_index_file(void)
97 {
98         if (!git_index_file)
99                 setup_git_env();
100         return git_index_file;
101 }
102
103 char *get_graft_file(void)
104 {
105         if (!git_graft_file)
106                 setup_git_env();
107         return git_graft_file;
108 }
109
110 int safe_create_leading_directories(char *path)
111 {
112         char *pos = path;
113
114         while (pos) {
115                 pos = strchr(pos, '/');
116                 if (!pos)
117                         break;
118                 *pos = 0;
119                 if (mkdir(path, 0777) < 0)
120                         if (errno != EEXIST) {
121                                 *pos = '/';
122                                 return -1;
123                         }
124                 *pos++ = '/';
125         }
126         return 0;
127 }
128
129 char * sha1_to_hex(const unsigned char *sha1)
130 {
131         static char buffer[50];
132         static const char hex[] = "0123456789abcdef";
133         char *buf = buffer;
134         int i;
135
136         for (i = 0; i < 20; i++) {
137                 unsigned int val = *sha1++;
138                 *buf++ = hex[val >> 4];
139                 *buf++ = hex[val & 0xf];
140         }
141         return buffer;
142 }
143
144 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
145 {
146         int i;
147         for (i = 0; i < 20; i++) {
148                 static char hex[] = "0123456789abcdef";
149                 unsigned int val = sha1[i];
150                 char *pos = pathbuf + i*2 + (i > 0);
151                 *pos++ = hex[val >> 4];
152                 *pos = hex[val & 0xf];
153         }
154 }
155
156 /*
157  * NOTE! This returns a statically allocated buffer, so you have to be
158  * careful about using it. Do a "strdup()" if you need to save the
159  * filename.
160  *
161  * Also note that this returns the location for creating.  Reading
162  * SHA1 file can happen from any alternate directory listed in the
163  * DB_ENVIRONMENT environment variable if it is not found in
164  * the primary object database.
165  */
166 char *sha1_file_name(const unsigned char *sha1)
167 {
168         static char *name, *base;
169
170         if (!base) {
171                 const char *sha1_file_directory = get_object_directory();
172                 int len = strlen(sha1_file_directory);
173                 base = xmalloc(len + 60);
174                 memcpy(base, sha1_file_directory, len);
175                 memset(base+len, 0, 60);
176                 base[len] = '/';
177                 base[len+3] = '/';
178                 name = base + len + 1;
179         }
180         fill_sha1_path(name, sha1);
181         return base;
182 }
183
184 char *sha1_pack_name(const unsigned char *sha1)
185 {
186         static const char hex[] = "0123456789abcdef";
187         static char *name, *base, *buf;
188         int i;
189
190         if (!base) {
191                 const char *sha1_file_directory = get_object_directory();
192                 int len = strlen(sha1_file_directory);
193                 base = xmalloc(len + 60);
194                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
195                 name = base + len + 11;
196         }
197
198         buf = name;
199
200         for (i = 0; i < 20; i++) {
201                 unsigned int val = *sha1++;
202                 *buf++ = hex[val >> 4];
203                 *buf++ = hex[val & 0xf];
204         }
205         
206         return base;
207 }
208
209 char *sha1_pack_index_name(const unsigned char *sha1)
210 {
211         static const char hex[] = "0123456789abcdef";
212         static char *name, *base, *buf;
213         int i;
214
215         if (!base) {
216                 const char *sha1_file_directory = get_object_directory();
217                 int len = strlen(sha1_file_directory);
218                 base = xmalloc(len + 60);
219                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
220                 name = base + len + 11;
221         }
222
223         buf = name;
224
225         for (i = 0; i < 20; i++) {
226                 unsigned int val = *sha1++;
227                 *buf++ = hex[val >> 4];
228                 *buf++ = hex[val & 0xf];
229         }
230         
231         return base;
232 }
233
234 struct alternate_object_database *alt_odb_list;
235 static struct alternate_object_database **alt_odb_tail;
236
237 /*
238  * Prepare alternate object database registry.
239  *
240  * The variable alt_odb_list points at the list of struct
241  * alternate_object_database.  The elements on this list come from
242  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
243  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
244  * whose contents is exactly in the same format as that environment
245  * variable.  Its base points at a statically allocated buffer that
246  * contains "/the/directory/corresponding/to/.git/objects/...", while
247  * its name points just after the slash at the end of ".git/objects/"
248  * in the example above, and has enough space to hold 40-byte hex
249  * SHA1, an extra slash for the first level indirection, and the
250  * terminating NUL.
251  */
252 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
253                                  const char *relative_base)
254 {
255         const char *cp, *last;
256         struct alternate_object_database *ent;
257         int base_len = -1;
258
259         last = alt;
260         while (last < ep) {
261                 cp = last;
262                 if (cp < ep && *cp == '#') {
263                         while (cp < ep && *cp != sep)
264                                 cp++;
265                         last = cp + 1;
266                         continue;
267                 }
268                 for ( ; cp < ep && *cp != sep; cp++)
269                         ;
270                 if (last != cp) {
271                         /* 43 = 40-byte + 2 '/' + terminating NUL */
272                         int pfxlen = cp - last;
273                         int entlen = pfxlen + 43;
274
275                         if (*last != '/' && relative_base) {
276                                 /* Relative alt-odb */
277                                 if (base_len < 0)
278                                         base_len = strlen(relative_base) + 1;
279                                 entlen += base_len;
280                                 pfxlen += base_len;
281                         }
282                         ent = xmalloc(sizeof(*ent) + entlen);
283                         *alt_odb_tail = ent;
284                         alt_odb_tail = &(ent->next);
285                         ent->next = NULL;
286                         if (*last != '/' && relative_base) {
287                                 memcpy(ent->base, relative_base, base_len - 1);
288                                 ent->base[base_len - 1] = '/';
289                                 memcpy(ent->base + base_len,
290                                        last, cp - last);
291                         }
292                         else
293                                 memcpy(ent->base, last, pfxlen);
294                         ent->name = ent->base + pfxlen + 1;
295                         ent->base[pfxlen] = ent->base[pfxlen + 3] = '/';
296                         ent->base[entlen-1] = 0;
297                 }
298                 while (cp < ep && *cp == sep)
299                         cp++;
300                 last = cp;
301         }
302 }
303
304 void prepare_alt_odb(void)
305 {
306         char path[PATH_MAX];
307         char *map;
308         int fd;
309         struct stat st;
310         char *alt;
311
312         alt = getenv(ALTERNATE_DB_ENVIRONMENT);
313         if (!alt) alt = "";
314
315         if (alt_odb_tail)
316                 return;
317         alt_odb_tail = &alt_odb_list;
318         link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL);
319
320         sprintf(path, "%s/info/alternates", get_object_directory());
321         fd = open(path, O_RDONLY);
322         if (fd < 0)
323                 return;
324         if (fstat(fd, &st) || (st.st_size == 0)) {
325                 close(fd);
326                 return;
327         }
328         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
329         close(fd);
330         if (map == MAP_FAILED)
331                 return;
332
333         link_alt_odb_entries(map, map + st.st_size, '\n',
334                              get_object_directory());
335         munmap(map, st.st_size);
336 }
337
338 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
339 {
340         char *name = sha1_file_name(sha1);
341         struct alternate_object_database *alt;
342
343         if (!stat(name, st))
344                 return name;
345         prepare_alt_odb();
346         for (alt = alt_odb_list; alt; alt = alt->next) {
347                 name = alt->name;
348                 fill_sha1_path(name, sha1);
349                 if (!stat(alt->base, st))
350                         return alt->base;
351         }
352         return NULL;
353 }
354
355 #define PACK_MAX_SZ (1<<26)
356 static int pack_used_ctr;
357 static unsigned long pack_mapped;
358 struct packed_git *packed_git;
359
360 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
361                                 void **idx_map_)
362 {
363         void *idx_map;
364         unsigned int *index;
365         unsigned long idx_size;
366         int nr, i;
367         int fd = open(path, O_RDONLY);
368         struct stat st;
369         if (fd < 0)
370                 return -1;
371         if (fstat(fd, &st)) {
372                 close(fd);
373                 return -1;
374         }
375         idx_size = st.st_size;
376         idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
377         close(fd);
378         if (idx_map == MAP_FAILED)
379                 return -1;
380
381         index = idx_map;
382         *idx_map_ = idx_map;
383         *idx_size_ = idx_size;
384
385         /* check index map */
386         if (idx_size < 4*256 + 20 + 20)
387                 return error("index file too small");
388         nr = 0;
389         for (i = 0; i < 256; i++) {
390                 unsigned int n = ntohl(index[i]);
391                 if (n < nr)
392                         return error("non-monotonic index");
393                 nr = n;
394         }
395
396         /*
397          * Total size:
398          *  - 256 index entries 4 bytes each
399          *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
400          *  - 20-byte SHA1 of the packfile
401          *  - 20-byte SHA1 file checksum
402          */
403         if (idx_size != 4*256 + nr * 24 + 20 + 20)
404                 return error("wrong index file size");
405
406         return 0;
407 }
408
409 static int unuse_one_packed_git(void)
410 {
411         struct packed_git *p, *lru = NULL;
412
413         for (p = packed_git; p; p = p->next) {
414                 if (p->pack_use_cnt || !p->pack_base)
415                         continue;
416                 if (!lru || p->pack_last_used < lru->pack_last_used)
417                         lru = p;
418         }
419         if (!lru)
420                 return 0;
421         munmap(lru->pack_base, lru->pack_size);
422         lru->pack_base = NULL;
423         return 1;
424 }
425
426 void unuse_packed_git(struct packed_git *p)
427 {
428         p->pack_use_cnt--;
429 }
430
431 int use_packed_git(struct packed_git *p)
432 {
433         if (!p->pack_size) {
434                 struct stat st;
435                 // We created the struct before we had the pack
436                 stat(p->pack_name, &st);
437                 if (!S_ISREG(st.st_mode))
438                         die("packfile %s not a regular file", p->pack_name);
439                 p->pack_size = st.st_size;
440         }
441         if (!p->pack_base) {
442                 int fd;
443                 struct stat st;
444                 void *map;
445
446                 pack_mapped += p->pack_size;
447                 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
448                         ; /* nothing */
449                 fd = open(p->pack_name, O_RDONLY);
450                 if (fd < 0)
451                         die("packfile %s cannot be opened", p->pack_name);
452                 if (fstat(fd, &st)) {
453                         close(fd);
454                         die("packfile %s cannot be opened", p->pack_name);
455                 }
456                 if (st.st_size != p->pack_size)
457                         die("packfile %s size mismatch.", p->pack_name);
458                 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
459                 close(fd);
460                 if (map == MAP_FAILED)
461                         die("packfile %s cannot be mapped.", p->pack_name);
462                 p->pack_base = map;
463
464                 /* Check if the pack file matches with the index file.
465                  * this is cheap.
466                  */
467                 if (memcmp((char*)(p->index_base) + p->index_size - 40,
468                            p->pack_base + p->pack_size - 20, 20)) {
469                               
470                         die("packfile %s does not match index.", p->pack_name);
471                 }
472         }
473         p->pack_last_used = pack_used_ctr++;
474         p->pack_use_cnt++;
475         return 0;
476 }
477
478 struct packed_git *add_packed_git(char *path, int path_len)
479 {
480         struct stat st;
481         struct packed_git *p;
482         unsigned long idx_size;
483         void *idx_map;
484
485         if (check_packed_git_idx(path, &idx_size, &idx_map))
486                 return NULL;
487
488         /* do we have a corresponding .pack file? */
489         strcpy(path + path_len - 4, ".pack");
490         if (stat(path, &st) || !S_ISREG(st.st_mode)) {
491                 munmap(idx_map, idx_size);
492                 return NULL;
493         }
494         /* ok, it looks sane as far as we can check without
495          * actually mapping the pack file.
496          */
497         p = xmalloc(sizeof(*p) + path_len + 2);
498         strcpy(p->pack_name, path);
499         p->index_size = idx_size;
500         p->pack_size = st.st_size;
501         p->index_base = idx_map;
502         p->next = NULL;
503         p->pack_base = NULL;
504         p->pack_last_used = 0;
505         p->pack_use_cnt = 0;
506         return p;
507 }
508
509 struct packed_git *parse_pack_index(unsigned char *sha1)
510 {
511         char *path = sha1_pack_index_name(sha1);
512         return parse_pack_index_file(sha1, path);
513 }
514
515 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
516 {
517         struct packed_git *p;
518         unsigned long idx_size;
519         void *idx_map;
520         char *path;
521
522         if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
523                 return NULL;
524
525         path = sha1_pack_name(sha1);
526
527         p = xmalloc(sizeof(*p) + strlen(path) + 2);
528         strcpy(p->pack_name, path);
529         p->index_size = idx_size;
530         p->pack_size = 0;
531         p->index_base = idx_map;
532         p->next = NULL;
533         p->pack_base = NULL;
534         p->pack_last_used = 0;
535         p->pack_use_cnt = 0;
536         memcpy(p->sha1, sha1, 20);
537         return p;
538 }
539
540 void install_packed_git(struct packed_git *pack)
541 {
542         pack->next = packed_git;
543         packed_git = pack;
544 }
545
546 static void prepare_packed_git_one(char *objdir)
547 {
548         char path[PATH_MAX];
549         int len;
550         DIR *dir;
551         struct dirent *de;
552
553         sprintf(path, "%s/pack", objdir);
554         len = strlen(path);
555         dir = opendir(path);
556         if (!dir)
557                 return;
558         path[len++] = '/';
559         while ((de = readdir(dir)) != NULL) {
560                 int namelen = strlen(de->d_name);
561                 struct packed_git *p;
562
563                 if (strcmp(de->d_name + namelen - 4, ".idx"))
564                         continue;
565
566                 /* we have .idx.  Is it a file we can map? */
567                 strcpy(path + len, de->d_name);
568                 p = add_packed_git(path, len + namelen);
569                 if (!p)
570                         continue;
571                 p->next = packed_git;
572                 packed_git = p;
573         }
574         closedir(dir);
575 }
576
577 void prepare_packed_git(void)
578 {
579         static int run_once = 0;
580         struct alternate_object_database *alt;
581
582         if (run_once)
583                 return;
584         prepare_packed_git_one(get_object_directory());
585         prepare_alt_odb();
586         for (alt = alt_odb_list; alt; alt = alt->next) {
587                 alt->name[0] = 0;
588                 prepare_packed_git_one(alt->base);
589         }
590         run_once = 1;
591 }
592
593 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
594 {
595         char header[100];
596         unsigned char real_sha1[20];
597         SHA_CTX c;
598
599         SHA1_Init(&c);
600         SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
601         SHA1_Update(&c, map, size);
602         SHA1_Final(real_sha1, &c);
603         return memcmp(sha1, real_sha1, 20) ? -1 : 0;
604 }
605
606 static void *map_sha1_file_internal(const unsigned char *sha1,
607                                     unsigned long *size)
608 {
609         struct stat st;
610         void *map;
611         int fd;
612         char *filename = find_sha1_file(sha1, &st);
613
614         if (!filename) {
615                 return NULL;
616         }
617
618         fd = open(filename, O_RDONLY | sha1_file_open_flag);
619         if (fd < 0) {
620                 /* See if it works without O_NOATIME */
621                 switch (sha1_file_open_flag) {
622                 default:
623                         fd = open(filename, O_RDONLY);
624                         if (fd >= 0)
625                                 break;
626                 /* Fallthrough */
627                 case 0:
628                         return NULL;
629                 }
630
631                 /* If it failed once, it will probably fail again.
632                  * Stop using O_NOATIME
633                  */
634                 sha1_file_open_flag = 0;
635         }
636         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
637         close(fd);
638         if (map == MAP_FAILED)
639                 return NULL;
640         *size = st.st_size;
641         return map;
642 }
643
644 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
645 {
646         /* Get the data stream */
647         memset(stream, 0, sizeof(*stream));
648         stream->next_in = map;
649         stream->avail_in = mapsize;
650         stream->next_out = buffer;
651         stream->avail_out = size;
652
653         inflateInit(stream);
654         return inflate(stream, 0);
655 }
656
657 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
658 {
659         int bytes = strlen(buffer) + 1;
660         unsigned char *buf = xmalloc(1+size);
661
662         memcpy(buf, buffer + bytes, stream->total_out - bytes);
663         bytes = stream->total_out - bytes;
664         if (bytes < size) {
665                 stream->next_out = buf + bytes;
666                 stream->avail_out = size - bytes;
667                 while (inflate(stream, Z_FINISH) == Z_OK)
668                         /* nothing */;
669         }
670         buf[size] = 0;
671         inflateEnd(stream);
672         return buf;
673 }
674
675 /*
676  * We used to just use "sscanf()", but that's actually way
677  * too permissive for what we want to check. So do an anal
678  * object header parse by hand.
679  */
680 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
681 {
682         int i;
683         unsigned long size;
684
685         /*
686          * The type can be at most ten bytes (including the 
687          * terminating '\0' that we add), and is followed by
688          * a space. 
689          */
690         i = 10;
691         for (;;) {
692                 char c = *hdr++;
693                 if (c == ' ')
694                         break;
695                 if (!--i)
696                         return -1;
697                 *type++ = c;
698         }
699         *type = 0;
700
701         /*
702          * The length must follow immediately, and be in canonical
703          * decimal format (ie "010" is not valid).
704          */
705         size = *hdr++ - '0';
706         if (size > 9)
707                 return -1;
708         if (size) {
709                 for (;;) {
710                         unsigned long c = *hdr - '0';
711                         if (c > 9)
712                                 break;
713                         hdr++;
714                         size = size * 10 + c;
715                 }
716         }
717         *sizep = size;
718
719         /*
720          * The length must be followed by a zero byte
721          */
722         return *hdr ? -1 : 0;
723 }
724
725 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
726 {
727         int ret;
728         z_stream stream;
729         char hdr[8192];
730
731         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
732         if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
733                 return NULL;
734
735         return unpack_sha1_rest(&stream, hdr, *size);
736 }
737
738 /* forward declaration for a mutually recursive function */
739 static int packed_object_info(struct pack_entry *entry,
740                               char *type, unsigned long *sizep);
741
742 static int packed_delta_info(unsigned char *base_sha1,
743                              unsigned long delta_size,
744                              unsigned long left,
745                              char *type,
746                              unsigned long *sizep,
747                              struct packed_git *p)
748 {
749         struct pack_entry base_ent;
750
751         if (left < 20)
752                 die("truncated pack file");
753
754         /* The base entry _must_ be in the same pack */
755         if (!find_pack_entry_one(base_sha1, &base_ent, p))
756                 die("failed to find delta-pack base object %s",
757                     sha1_to_hex(base_sha1));
758
759         /* We choose to only get the type of the base object and
760          * ignore potentially corrupt pack file that expects the delta
761          * based on a base with a wrong size.  This saves tons of
762          * inflate() calls.
763          */
764
765         if (packed_object_info(&base_ent, type, NULL))
766                 die("cannot get info for delta-pack base");
767
768         if (sizep) {
769                 const unsigned char *data;
770                 unsigned char delta_head[64];
771                 unsigned long result_size;
772                 z_stream stream;
773                 int st;
774
775                 memset(&stream, 0, sizeof(stream));
776
777                 data = stream.next_in = base_sha1 + 20;
778                 stream.avail_in = left - 20;
779                 stream.next_out = delta_head;
780                 stream.avail_out = sizeof(delta_head);
781
782                 inflateInit(&stream);
783                 st = inflate(&stream, Z_FINISH);
784                 inflateEnd(&stream);
785                 if ((st != Z_STREAM_END) &&
786                     stream.total_out != sizeof(delta_head))
787                         die("delta data unpack-initial failed");
788
789                 /* Examine the initial part of the delta to figure out
790                  * the result size.
791                  */
792                 data = delta_head;
793                 get_delta_hdr_size(&data); /* ignore base size */
794
795                 /* Read the result size */
796                 result_size = get_delta_hdr_size(&data);
797                 *sizep = result_size;
798         }
799         return 0;
800 }
801
802 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
803         enum object_type *type, unsigned long *sizep)
804 {
805         unsigned shift;
806         unsigned char *pack, c;
807         unsigned long size;
808
809         if (offset >= p->pack_size)
810                 die("object offset outside of pack file");
811
812         pack =  p->pack_base + offset;
813         c = *pack++;
814         offset++;
815         *type = (c >> 4) & 7;
816         size = c & 15;
817         shift = 4;
818         while (c & 0x80) {
819                 if (offset >= p->pack_size)
820                         die("object offset outside of pack file");
821                 c = *pack++;
822                 offset++;
823                 size += (c & 0x7f) << shift;
824                 shift += 7;
825         }
826         *sizep = size;
827         return offset;
828 }
829
830 void packed_object_info_detail(struct pack_entry *e,
831                                char *type,
832                                unsigned long *size,
833                                unsigned long *store_size,
834                                int *delta_chain_length,
835                                unsigned char *base_sha1)
836 {
837         struct packed_git *p = e->p;
838         unsigned long offset, left;
839         unsigned char *pack;
840         enum object_type kind;
841
842         offset = unpack_object_header(p, e->offset, &kind, size);
843         pack = p->pack_base + offset;
844         left = p->pack_size - offset;
845         if (kind != OBJ_DELTA)
846                 *delta_chain_length = 0;
847         else {
848                 int chain_length = 0;
849                 memcpy(base_sha1, pack, 20);
850                 do {
851                         struct pack_entry base_ent;
852                         unsigned long junk;
853
854                         find_pack_entry_one(pack, &base_ent, p);
855                         offset = unpack_object_header(p, base_ent.offset,
856                                                       &kind, &junk);
857                         pack = p->pack_base + offset;
858                         chain_length++;
859                 } while (kind == OBJ_DELTA);
860                 *delta_chain_length = chain_length;
861         }
862         switch (kind) {
863         case OBJ_COMMIT:
864                 strcpy(type, "commit");
865                 break;
866         case OBJ_TREE:
867                 strcpy(type, "tree");
868                 break;
869         case OBJ_BLOB:
870                 strcpy(type, "blob");
871                 break;
872         case OBJ_TAG:
873                 strcpy(type, "tag");
874                 break;
875         default:
876                 die("corrupted pack file %s containing object of kind %d",
877                     p->pack_name, kind);
878         }
879         *store_size = 0; /* notyet */
880 }
881
882 static int packed_object_info(struct pack_entry *entry,
883                               char *type, unsigned long *sizep)
884 {
885         struct packed_git *p = entry->p;
886         unsigned long offset, size, left;
887         unsigned char *pack;
888         enum object_type kind;
889         int retval;
890
891         if (use_packed_git(p))
892                 die("cannot map packed file");
893
894         offset = unpack_object_header(p, entry->offset, &kind, &size);
895         pack = p->pack_base + offset;
896         left = p->pack_size - offset;
897
898         switch (kind) {
899         case OBJ_DELTA:
900                 retval = packed_delta_info(pack, size, left, type, sizep, p);
901                 unuse_packed_git(p);
902                 return retval;
903         case OBJ_COMMIT:
904                 strcpy(type, "commit");
905                 break;
906         case OBJ_TREE:
907                 strcpy(type, "tree");
908                 break;
909         case OBJ_BLOB:
910                 strcpy(type, "blob");
911                 break;
912         case OBJ_TAG:
913                 strcpy(type, "tag");
914                 break;
915         default:
916                 die("corrupted pack file %s containing object of kind %d",
917                     p->pack_name, kind);
918         }
919         if (sizep)
920                 *sizep = size;
921         unuse_packed_git(p);
922         return 0;
923 }
924
925 /* forward declaration for a mutually recursive function */
926 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
927
928 static void *unpack_delta_entry(unsigned char *base_sha1,
929                                 unsigned long delta_size,
930                                 unsigned long left,
931                                 char *type,
932                                 unsigned long *sizep,
933                                 struct packed_git *p)
934 {
935         struct pack_entry base_ent;
936         void *data, *delta_data, *result, *base;
937         unsigned long data_size, result_size, base_size;
938         z_stream stream;
939         int st;
940
941         if (left < 20)
942                 die("truncated pack file");
943         data = base_sha1 + 20;
944         data_size = left - 20;
945         delta_data = xmalloc(delta_size);
946
947         memset(&stream, 0, sizeof(stream));
948
949         stream.next_in = data;
950         stream.avail_in = data_size;
951         stream.next_out = delta_data;
952         stream.avail_out = delta_size;
953
954         inflateInit(&stream);
955         st = inflate(&stream, Z_FINISH);
956         inflateEnd(&stream);
957         if ((st != Z_STREAM_END) || stream.total_out != delta_size)
958                 die("delta data unpack failed");
959
960         /* The base entry _must_ be in the same pack */
961         if (!find_pack_entry_one(base_sha1, &base_ent, p))
962                 die("failed to find delta-pack base object %s",
963                     sha1_to_hex(base_sha1));
964         base = unpack_entry_gently(&base_ent, type, &base_size);
965         if (!base)
966                 die("failed to read delta-pack base object %s",
967                     sha1_to_hex(base_sha1));
968         result = patch_delta(base, base_size,
969                              delta_data, delta_size,
970                              &result_size);
971         if (!result)
972                 die("failed to apply delta");
973         free(delta_data);
974         free(base);
975         *sizep = result_size;
976         return result;
977 }
978
979 static void *unpack_non_delta_entry(unsigned char *data,
980                                     unsigned long size,
981                                     unsigned long left)
982 {
983         int st;
984         z_stream stream;
985         unsigned char *buffer;
986
987         buffer = xmalloc(size + 1);
988         buffer[size] = 0;
989         memset(&stream, 0, sizeof(stream));
990         stream.next_in = data;
991         stream.avail_in = left;
992         stream.next_out = buffer;
993         stream.avail_out = size;
994
995         inflateInit(&stream);
996         st = inflate(&stream, Z_FINISH);
997         inflateEnd(&stream);
998         if ((st != Z_STREAM_END) || stream.total_out != size) {
999                 free(buffer);
1000                 return NULL;
1001         }
1002
1003         return buffer;
1004 }
1005
1006 static void *unpack_entry(struct pack_entry *entry,
1007                           char *type, unsigned long *sizep)
1008 {
1009         struct packed_git *p = entry->p;
1010         void *retval;
1011
1012         if (use_packed_git(p))
1013                 die("cannot map packed file");
1014         retval = unpack_entry_gently(entry, type, sizep);
1015         unuse_packed_git(p);
1016         if (!retval)
1017                 die("corrupted pack file %s", p->pack_name);
1018         return retval;
1019 }
1020
1021 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1022 void *unpack_entry_gently(struct pack_entry *entry,
1023                           char *type, unsigned long *sizep)
1024 {
1025         struct packed_git *p = entry->p;
1026         unsigned long offset, size, left;
1027         unsigned char *pack;
1028         enum object_type kind;
1029         void *retval;
1030
1031         offset = unpack_object_header(p, entry->offset, &kind, &size);
1032         pack = p->pack_base + offset;
1033         left = p->pack_size - offset;
1034         switch (kind) {
1035         case OBJ_DELTA:
1036                 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1037                 return retval;
1038         case OBJ_COMMIT:
1039                 strcpy(type, "commit");
1040                 break;
1041         case OBJ_TREE:
1042                 strcpy(type, "tree");
1043                 break;
1044         case OBJ_BLOB:
1045                 strcpy(type, "blob");
1046                 break;
1047         case OBJ_TAG:
1048                 strcpy(type, "tag");
1049                 break;
1050         default:
1051                 return NULL;
1052         }
1053         *sizep = size;
1054         retval = unpack_non_delta_entry(pack, size, left);
1055         return retval;
1056 }
1057
1058 int num_packed_objects(const struct packed_git *p)
1059 {
1060         /* See check_packed_git_idx() */
1061         return (p->index_size - 20 - 20 - 4*256) / 24;
1062 }
1063
1064 int nth_packed_object_sha1(const struct packed_git *p, int n,
1065                            unsigned char* sha1)
1066 {
1067         void *index = p->index_base + 256;
1068         if (n < 0 || num_packed_objects(p) <= n)
1069                 return -1;
1070         memcpy(sha1, (index + 24 * n + 4), 20);
1071         return 0;
1072 }
1073
1074 int find_pack_entry_one(const unsigned char *sha1,
1075                         struct pack_entry *e, struct packed_git *p)
1076 {
1077         unsigned int *level1_ofs = p->index_base;
1078         int hi = ntohl(level1_ofs[*sha1]);
1079         int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1080         void *index = p->index_base + 256;
1081
1082         do {
1083                 int mi = (lo + hi) / 2;
1084                 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1085                 if (!cmp) {
1086                         e->offset = ntohl(*((int*)(index + 24 * mi)));
1087                         memcpy(e->sha1, sha1, 20);
1088                         e->p = p;
1089                         return 1;
1090                 }
1091                 if (cmp > 0)
1092                         hi = mi;
1093                 else
1094                         lo = mi+1;
1095         } while (lo < hi);
1096         return 0;
1097 }
1098
1099 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1100 {
1101         struct packed_git *p;
1102         prepare_packed_git();
1103
1104         for (p = packed_git; p; p = p->next) {
1105                 if (find_pack_entry_one(sha1, e, p))
1106                         return 1;
1107         }
1108         return 0;
1109 }
1110
1111 struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1112                                   struct packed_git *packs)
1113 {
1114         struct packed_git *p;
1115         struct pack_entry e;
1116
1117         for (p = packs; p; p = p->next) {
1118                 if (find_pack_entry_one(sha1, &e, p))
1119                         return p;
1120         }
1121         return NULL;
1122         
1123 }
1124
1125 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1126 {
1127         int status;
1128         unsigned long mapsize, size;
1129         void *map;
1130         z_stream stream;
1131         char hdr[128];
1132
1133         map = map_sha1_file_internal(sha1, &mapsize);
1134         if (!map) {
1135                 struct pack_entry e;
1136
1137                 if (!find_pack_entry(sha1, &e))
1138                         return error("unable to find %s", sha1_to_hex(sha1));
1139                 return packed_object_info(&e, type, sizep);
1140         }
1141         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1142                 status = error("unable to unpack %s header",
1143                                sha1_to_hex(sha1));
1144         if (parse_sha1_header(hdr, type, &size) < 0)
1145                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1146         else {
1147                 status = 0;
1148                 if (sizep)
1149                         *sizep = size;
1150         }
1151         inflateEnd(&stream);
1152         munmap(map, mapsize);
1153         return status;
1154 }
1155
1156 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1157 {
1158         struct pack_entry e;
1159
1160         if (!find_pack_entry(sha1, &e)) {
1161                 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1162                 return NULL;
1163         }
1164         return unpack_entry(&e, type, size);
1165 }
1166
1167 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1168 {
1169         unsigned long mapsize;
1170         void *map, *buf;
1171         struct pack_entry e;
1172
1173         if (find_pack_entry(sha1, &e))
1174                 return read_packed_sha1(sha1, type, size);
1175         map = map_sha1_file_internal(sha1, &mapsize);
1176         if (map) {
1177                 buf = unpack_sha1_file(map, mapsize, type, size);
1178                 munmap(map, mapsize);
1179                 return buf;
1180         }
1181         return NULL;
1182 }
1183
1184 void *read_object_with_reference(const unsigned char *sha1,
1185                                  const char *required_type,
1186                                  unsigned long *size,
1187                                  unsigned char *actual_sha1_return)
1188 {
1189         char type[20];
1190         void *buffer;
1191         unsigned long isize;
1192         unsigned char actual_sha1[20];
1193
1194         memcpy(actual_sha1, sha1, 20);
1195         while (1) {
1196                 int ref_length = -1;
1197                 const char *ref_type = NULL;
1198
1199                 buffer = read_sha1_file(actual_sha1, type, &isize);
1200                 if (!buffer)
1201                         return NULL;
1202                 if (!strcmp(type, required_type)) {
1203                         *size = isize;
1204                         if (actual_sha1_return)
1205                                 memcpy(actual_sha1_return, actual_sha1, 20);
1206                         return buffer;
1207                 }
1208                 /* Handle references */
1209                 else if (!strcmp(type, "commit"))
1210                         ref_type = "tree ";
1211                 else if (!strcmp(type, "tag"))
1212                         ref_type = "object ";
1213                 else {
1214                         free(buffer);
1215                         return NULL;
1216                 }
1217                 ref_length = strlen(ref_type);
1218
1219                 if (memcmp(buffer, ref_type, ref_length) ||
1220                     get_sha1_hex(buffer + ref_length, actual_sha1)) {
1221                         free(buffer);
1222                         return NULL;
1223                 }
1224                 free(buffer);
1225                 /* Now we have the ID of the referred-to object in
1226                  * actual_sha1.  Check again. */
1227         }
1228 }
1229
1230 char *write_sha1_file_prepare(void *buf,
1231                               unsigned long len,
1232                               const char *type,
1233                               unsigned char *sha1,
1234                               unsigned char *hdr,
1235                               int *hdrlen)
1236 {
1237         SHA_CTX c;
1238
1239         /* Generate the header */
1240         *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1241
1242         /* Sha1.. */
1243         SHA1_Init(&c);
1244         SHA1_Update(&c, hdr, *hdrlen);
1245         SHA1_Update(&c, buf, len);
1246         SHA1_Final(sha1, &c);
1247
1248         return sha1_file_name(sha1);
1249 }
1250
1251 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1252 {
1253         int size;
1254         unsigned char *compressed;
1255         z_stream stream;
1256         unsigned char sha1[20];
1257         char *filename;
1258         static char tmpfile[PATH_MAX];
1259         unsigned char hdr[50];
1260         int fd, hdrlen, ret;
1261
1262         /* Normally if we have it in the pack then we do not bother writing
1263          * it out into .git/objects/??/?{38} file.
1264          */
1265         filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1266         if (returnsha1)
1267                 memcpy(returnsha1, sha1, 20);
1268         if (has_sha1_file(sha1))
1269                 return 0;
1270         fd = open(filename, O_RDONLY);
1271         if (fd >= 0) {
1272                 /*
1273                  * FIXME!!! We might do collision checking here, but we'd
1274                  * need to uncompress the old file and check it. Later.
1275                  */
1276                 close(fd);
1277                 return 0;
1278         }
1279
1280         if (errno != ENOENT) {
1281                 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1282                 return -1;
1283         }
1284
1285         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1286
1287         fd = mkstemp(tmpfile);
1288         if (fd < 0) {
1289                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1290                 return -1;
1291         }
1292
1293         /* Set it up */
1294         memset(&stream, 0, sizeof(stream));
1295         deflateInit(&stream, Z_BEST_COMPRESSION);
1296         size = deflateBound(&stream, len+hdrlen);
1297         compressed = xmalloc(size);
1298
1299         /* Compress it */
1300         stream.next_out = compressed;
1301         stream.avail_out = size;
1302
1303         /* First header.. */
1304         stream.next_in = hdr;
1305         stream.avail_in = hdrlen;
1306         while (deflate(&stream, 0) == Z_OK)
1307                 /* nothing */;
1308
1309         /* Then the data itself.. */
1310         stream.next_in = buf;
1311         stream.avail_in = len;
1312         while (deflate(&stream, Z_FINISH) == Z_OK)
1313                 /* nothing */;
1314         deflateEnd(&stream);
1315         size = stream.total_out;
1316
1317         if (write(fd, compressed, size) != size)
1318                 die("unable to write file");
1319         fchmod(fd, 0444);
1320         close(fd);
1321         free(compressed);
1322
1323         ret = link(tmpfile, filename);
1324         if (ret < 0) {
1325                 ret = errno;
1326
1327                 /*
1328                  * Coda hack - coda doesn't like cross-directory links,
1329                  * so we fall back to a rename, which will mean that it
1330                  * won't be able to check collisions, but that's not a
1331                  * big deal.
1332                  *
1333                  * When this succeeds, we just return 0. We have nothing
1334                  * left to unlink.
1335                  */
1336                 if (ret == EXDEV && !rename(tmpfile, filename))
1337                         return 0;
1338         }
1339         unlink(tmpfile);
1340         if (ret) {
1341                 if (ret != EEXIST) {
1342                         fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1343                         return -1;
1344                 }
1345                 /* FIXME!!! Collision check here ? */
1346         }
1347
1348         return 0;
1349 }
1350
1351 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1352 {
1353         ssize_t size;
1354         unsigned long objsize;
1355         int posn = 0;
1356         void *map = map_sha1_file_internal(sha1, &objsize);
1357         void *buf = map;
1358         void *temp_obj = NULL;
1359         z_stream stream;
1360
1361         if (!buf) {
1362                 unsigned char *unpacked;
1363                 unsigned long len;
1364                 char type[20];
1365                 char hdr[50];
1366                 int hdrlen;
1367                 // need to unpack and recompress it by itself
1368                 unpacked = read_packed_sha1(sha1, type, &len);
1369
1370                 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1371
1372                 /* Set it up */
1373                 memset(&stream, 0, sizeof(stream));
1374                 deflateInit(&stream, Z_BEST_COMPRESSION);
1375                 size = deflateBound(&stream, len + hdrlen);
1376                 temp_obj = buf = xmalloc(size);
1377
1378                 /* Compress it */
1379                 stream.next_out = buf;
1380                 stream.avail_out = size;
1381                 
1382                 /* First header.. */
1383                 stream.next_in = (void *)hdr;
1384                 stream.avail_in = hdrlen;
1385                 while (deflate(&stream, 0) == Z_OK)
1386                         /* nothing */;
1387
1388                 /* Then the data itself.. */
1389                 stream.next_in = unpacked;
1390                 stream.avail_in = len;
1391                 while (deflate(&stream, Z_FINISH) == Z_OK)
1392                         /* nothing */;
1393                 deflateEnd(&stream);
1394                 free(unpacked);
1395                 
1396                 objsize = stream.total_out;
1397         }
1398
1399         do {
1400                 size = write(fd, buf + posn, objsize - posn);
1401                 if (size <= 0) {
1402                         if (!size) {
1403                                 fprintf(stderr, "write closed");
1404                         } else {
1405                                 perror("write ");
1406                         }
1407                         return -1;
1408                 }
1409                 posn += size;
1410         } while (posn < objsize);
1411
1412         if (map)
1413                 munmap(map, objsize);
1414         if (temp_obj)
1415                 free(temp_obj);
1416
1417         return 0;
1418 }
1419
1420 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1421                        size_t bufsize, size_t *bufposn)
1422 {
1423         char *filename = sha1_file_name(sha1);
1424
1425         int local;
1426         z_stream stream;
1427         unsigned char real_sha1[20];
1428         unsigned char discard[4096];
1429         int ret;
1430         SHA_CTX c;
1431
1432         local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1433
1434         if (local < 0)
1435                 return error("Couldn't open %s\n", filename);
1436
1437         memset(&stream, 0, sizeof(stream));
1438
1439         inflateInit(&stream);
1440
1441         SHA1_Init(&c);
1442
1443         do {
1444                 ssize_t size;
1445                 if (*bufposn) {
1446                         stream.avail_in = *bufposn;
1447                         stream.next_in = (unsigned char *) buffer;
1448                         do {
1449                                 stream.next_out = discard;
1450                                 stream.avail_out = sizeof(discard);
1451                                 ret = inflate(&stream, Z_SYNC_FLUSH);
1452                                 SHA1_Update(&c, discard, sizeof(discard) -
1453                                             stream.avail_out);
1454                         } while (stream.avail_in && ret == Z_OK);
1455                         write(local, buffer, *bufposn - stream.avail_in);
1456                         memmove(buffer, buffer + *bufposn - stream.avail_in,
1457                                 stream.avail_in);
1458                         *bufposn = stream.avail_in;
1459                         if (ret != Z_OK)
1460                                 break;
1461                 }
1462                 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1463                 if (size <= 0) {
1464                         close(local);
1465                         unlink(filename);
1466                         if (!size)
1467                                 return error("Connection closed?");
1468                         perror("Reading from connection");
1469                         return -1;
1470                 }
1471                 *bufposn += size;
1472         } while (1);
1473         inflateEnd(&stream);
1474
1475         close(local);
1476         SHA1_Final(real_sha1, &c);
1477         if (ret != Z_STREAM_END) {
1478                 unlink(filename);
1479                 return error("File %s corrupted", sha1_to_hex(sha1));
1480         }
1481         if (memcmp(sha1, real_sha1, 20)) {
1482                 unlink(filename);
1483                 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1484         }
1485         
1486         return 0;
1487 }
1488
1489 int has_pack_index(const unsigned char *sha1)
1490 {
1491         struct stat st;
1492         if (stat(sha1_pack_index_name(sha1), &st))
1493                 return 0;
1494         return 1;
1495 }
1496
1497 int has_pack_file(const unsigned char *sha1)
1498 {
1499         struct stat st;
1500         if (stat(sha1_pack_name(sha1), &st))
1501                 return 0;
1502         return 1;
1503 }
1504
1505 int has_sha1_pack(const unsigned char *sha1)
1506 {
1507         struct pack_entry e;
1508         return find_pack_entry(sha1, &e);
1509 }
1510
1511 int has_sha1_file(const unsigned char *sha1)
1512 {
1513         struct stat st;
1514         struct pack_entry e;
1515
1516         if (find_pack_entry(sha1, &e))
1517                 return 1;
1518         return find_sha1_file(sha1, &st) ? 1 : 0;
1519 }
1520
1521 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1522 {
1523         unsigned long size = st->st_size;
1524         void *buf;
1525         int ret;
1526         unsigned char hdr[50];
1527         int hdrlen;
1528
1529         buf = "";
1530         if (size)
1531                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1532         close(fd);
1533         if (buf == MAP_FAILED)
1534                 return -1;
1535
1536         if (!type)
1537                 type = "blob";
1538         if (write_object)
1539                 ret = write_sha1_file(buf, size, type, sha1);
1540         else {
1541                 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1542                 ret = 0;
1543         }
1544         if (size)
1545                 munmap(buf, size);
1546         return ret;
1547 }