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