Alternate object pool mechanism updates.
[git.git] / sha1_file.c
1 /*
2  * GIT - The information manager from hell
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  *
6  * This handles basic git sha1 object files - packing, unpacking,
7  * creation etc.
8  */
9 #include <sys/types.h>
10 #include <dirent.h>
11 #include "cache.h"
12 #include "delta.h"
13 #include "pack.h"
14
15 #ifndef O_NOATIME
16 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
17 #define O_NOATIME 01000000
18 #else
19 #define O_NOATIME 0
20 #endif
21 #endif
22
23 static unsigned int sha1_file_open_flag = O_NOATIME;
24
25 static unsigned hexval(char c)
26 {
27         if (c >= '0' && c <= '9')
28                 return c - '0';
29         if (c >= 'a' && c <= 'f')
30                 return c - 'a' + 10;
31         if (c >= 'A' && c <= 'F')
32                 return c - 'A' + 10;
33         return ~0;
34 }
35
36 int get_sha1_hex(const char *hex, unsigned char *sha1)
37 {
38         int i;
39         for (i = 0; i < 20; i++) {
40                 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
41                 if (val & ~0xff)
42                         return -1;
43                 *sha1++ = val;
44                 hex += 2;
45         }
46         return 0;
47 }
48
49 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir,
50         *git_graft_file;
51 static void setup_git_env(void)
52 {
53         git_dir = gitenv(GIT_DIR_ENVIRONMENT);
54         if (!git_dir)
55                 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
56         git_object_dir = gitenv(DB_ENVIRONMENT);
57         if (!git_object_dir) {
58                 git_object_dir = xmalloc(strlen(git_dir) + 9);
59                 sprintf(git_object_dir, "%s/objects", git_dir);
60         }
61         git_refs_dir = xmalloc(strlen(git_dir) + 6);
62         sprintf(git_refs_dir, "%s/refs", git_dir);
63         git_index_file = gitenv(INDEX_ENVIRONMENT);
64         if (!git_index_file) {
65                 git_index_file = xmalloc(strlen(git_dir) + 7);
66                 sprintf(git_index_file, "%s/index", git_dir);
67         }
68         git_graft_file = gitenv(GRAFT_ENVIRONMENT);
69         if (!git_graft_file)
70                 git_graft_file = strdup(git_path("info/grafts"));
71 }
72
73 char *get_object_directory(void)
74 {
75         if (!git_object_dir)
76                 setup_git_env();
77         return git_object_dir;
78 }
79
80 char *get_refs_directory(void)
81 {
82         if (!git_refs_dir)
83                 setup_git_env();
84         return git_refs_dir;
85 }
86
87 char *get_index_file(void)
88 {
89         if (!git_index_file)
90                 setup_git_env();
91         return git_index_file;
92 }
93
94 char *get_graft_file(void)
95 {
96         if (!git_graft_file)
97                 setup_git_env();
98         return git_graft_file;
99 }
100
101 int safe_create_leading_directories(char *path)
102 {
103         char *pos = path;
104
105         while (pos) {
106                 pos = strchr(pos, '/');
107                 if (!pos)
108                         break;
109                 *pos = 0;
110                 if (mkdir(path, 0777) < 0)
111                         if (errno != EEXIST) {
112                                 *pos = '/';
113                                 return -1;
114                         }
115                 *pos++ = '/';
116         }
117         return 0;
118 }
119
120 char * sha1_to_hex(const unsigned char *sha1)
121 {
122         static char buffer[50];
123         static const char hex[] = "0123456789abcdef";
124         char *buf = buffer;
125         int i;
126
127         for (i = 0; i < 20; i++) {
128                 unsigned int val = *sha1++;
129                 *buf++ = hex[val >> 4];
130                 *buf++ = hex[val & 0xf];
131         }
132         return buffer;
133 }
134
135 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
136 {
137         int i;
138         for (i = 0; i < 20; i++) {
139                 static char hex[] = "0123456789abcdef";
140                 unsigned int val = sha1[i];
141                 char *pos = pathbuf + i*2 + (i > 0);
142                 *pos++ = hex[val >> 4];
143                 *pos = hex[val & 0xf];
144         }
145 }
146
147 /*
148  * NOTE! This returns a statically allocated buffer, so you have to be
149  * careful about using it. Do a "strdup()" if you need to save the
150  * filename.
151  *
152  * Also note that this returns the location for creating.  Reading
153  * SHA1 file can happen from any alternate directory listed in the
154  * DB_ENVIRONMENT environment variable if it is not found in
155  * the primary object database.
156  */
157 char *sha1_file_name(const unsigned char *sha1)
158 {
159         static char *name, *base;
160
161         if (!base) {
162                 const char *sha1_file_directory = get_object_directory();
163                 int len = strlen(sha1_file_directory);
164                 base = xmalloc(len + 60);
165                 memcpy(base, sha1_file_directory, len);
166                 memset(base+len, 0, 60);
167                 base[len] = '/';
168                 base[len+3] = '/';
169                 name = base + len + 1;
170         }
171         fill_sha1_path(name, sha1);
172         return base;
173 }
174
175 char *sha1_pack_name(const unsigned char *sha1)
176 {
177         static const char hex[] = "0123456789abcdef";
178         static char *name, *base, *buf;
179         int i;
180
181         if (!base) {
182                 const char *sha1_file_directory = get_object_directory();
183                 int len = strlen(sha1_file_directory);
184                 base = xmalloc(len + 60);
185                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
186                 name = base + len + 11;
187         }
188
189         buf = name;
190
191         for (i = 0; i < 20; i++) {
192                 unsigned int val = *sha1++;
193                 *buf++ = hex[val >> 4];
194                 *buf++ = hex[val & 0xf];
195         }
196         
197         return base;
198 }
199
200 char *sha1_pack_index_name(const unsigned char *sha1)
201 {
202         static const char hex[] = "0123456789abcdef";
203         static char *name, *base, *buf;
204         int i;
205
206         if (!base) {
207                 const char *sha1_file_directory = get_object_directory();
208                 int len = strlen(sha1_file_directory);
209                 base = xmalloc(len + 60);
210                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
211                 name = base + len + 11;
212         }
213
214         buf = name;
215
216         for (i = 0; i < 20; i++) {
217                 unsigned int val = *sha1++;
218                 *buf++ = hex[val >> 4];
219                 *buf++ = hex[val & 0xf];
220         }
221         
222         return base;
223 }
224
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
227
228 /*
229  * Prepare alternate object database registry.
230  *
231  * The variable alt_odb_list points at the list of struct
232  * alternate_object_database.  The elements on this list come from
233  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
234  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
235  * whose contents is exactly in the same format as that environment
236  * variable.  Its base points at a statically allocated buffer that
237  * contains "/the/directory/corresponding/to/.git/objects/...", while
238  * its name points just after the slash at the end of ".git/objects/"
239  * in the example above, and has enough space to hold 40-byte hex
240  * SHA1, an extra slash for the first level indirection, and the
241  * terminating NUL.
242  */
243 static void link_alt_odb_entries(const char *alt, const char *ep)
244 {
245         const char *cp, *last;
246         struct alternate_object_database *ent;
247
248         last = alt;
249         do {
250                 for (cp = last; cp < ep && *cp != ':'; cp++)
251                         ;
252                 if (last != cp) {
253                         /* 43 = 40-byte + 2 '/' + terminating NUL */
254                         int pfxlen = cp - last;
255                         int entlen = pfxlen + 43;
256
257                         ent = xmalloc(sizeof(*ent) + entlen);
258                         *alt_odb_tail = ent;
259                         alt_odb_tail = &(ent->next);
260                         ent->next = NULL;
261
262                         memcpy(ent->base, last, pfxlen);
263                         ent->name = ent->base + pfxlen + 1;
264                         ent->base[pfxlen] = ent->base[pfxlen + 3] = '/';
265                         ent->base[entlen-1] = 0;
266                 }
267                 while (cp < ep && *cp == ':')
268                         cp++;
269                 last = cp;
270         } while (cp < ep);
271 }
272
273 void prepare_alt_odb(void)
274 {
275         char path[PATH_MAX];
276         char *map, *ep;
277         int fd;
278         struct stat st;
279         char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
280
281         sprintf(path, "%s/info/alternates", get_object_directory());
282         if (alt_odb_tail)
283                 return;
284         alt_odb_tail = &alt_odb_list;
285         link_alt_odb_entries(alt, alt + strlen(alt));
286
287         fd = open(path, O_RDONLY);
288         if (fd < 0)
289                 return;
290         if (fstat(fd, &st) || (st.st_size == 0)) {
291                 close(fd);
292                 return;
293         }
294         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
295         close(fd);
296         if (map == MAP_FAILED)
297                 return;
298
299         /* Remove the trailing newline */
300         for (ep = map + st.st_size - 1; map < ep && ep[-1] == '\n'; ep--)
301                 ;
302         link_alt_odb_entries(map, ep);
303         munmap(map, st.st_size);
304 }
305
306 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
307 {
308         char *name = sha1_file_name(sha1);
309         struct alternate_object_database *alt;
310
311         if (!stat(name, st))
312                 return name;
313         prepare_alt_odb();
314         for (alt = alt_odb_list; alt; alt = alt->next) {
315                 name = alt->name;
316                 fill_sha1_path(name, sha1);
317                 if (!stat(alt->base, st))
318                         return alt->base;
319         }
320         return NULL;
321 }
322
323 #define PACK_MAX_SZ (1<<26)
324 static int pack_used_ctr;
325 static unsigned long pack_mapped;
326 struct packed_git *packed_git;
327
328 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
329                                 void **idx_map_)
330 {
331         void *idx_map;
332         unsigned int *index;
333         unsigned long idx_size;
334         int nr, i;
335         int fd = open(path, O_RDONLY);
336         struct stat st;
337         if (fd < 0)
338                 return -1;
339         if (fstat(fd, &st)) {
340                 close(fd);
341                 return -1;
342         }
343         idx_size = st.st_size;
344         idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
345         close(fd);
346         if (idx_map == MAP_FAILED)
347                 return -1;
348
349         index = idx_map;
350         *idx_map_ = idx_map;
351         *idx_size_ = idx_size;
352
353         /* check index map */
354         if (idx_size < 4*256 + 20 + 20)
355                 return error("index file too small");
356         nr = 0;
357         for (i = 0; i < 256; i++) {
358                 unsigned int n = ntohl(index[i]);
359                 if (n < nr)
360                         return error("non-monotonic index");
361                 nr = n;
362         }
363
364         /*
365          * Total size:
366          *  - 256 index entries 4 bytes each
367          *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
368          *  - 20-byte SHA1 of the packfile
369          *  - 20-byte SHA1 file checksum
370          */
371         if (idx_size != 4*256 + nr * 24 + 20 + 20)
372                 return error("wrong index file size");
373
374         return 0;
375 }
376
377 static int unuse_one_packed_git(void)
378 {
379         struct packed_git *p, *lru = NULL;
380
381         for (p = packed_git; p; p = p->next) {
382                 if (p->pack_use_cnt || !p->pack_base)
383                         continue;
384                 if (!lru || p->pack_last_used < lru->pack_last_used)
385                         lru = p;
386         }
387         if (!lru)
388                 return 0;
389         munmap(lru->pack_base, lru->pack_size);
390         lru->pack_base = NULL;
391         return 1;
392 }
393
394 void unuse_packed_git(struct packed_git *p)
395 {
396         p->pack_use_cnt--;
397 }
398
399 int use_packed_git(struct packed_git *p)
400 {
401         if (!p->pack_size) {
402                 struct stat st;
403                 // We created the struct before we had the pack
404                 stat(p->pack_name, &st);
405                 if (!S_ISREG(st.st_mode))
406                         die("packfile %s not a regular file", p->pack_name);
407                 p->pack_size = st.st_size;
408         }
409         if (!p->pack_base) {
410                 int fd;
411                 struct stat st;
412                 void *map;
413
414                 pack_mapped += p->pack_size;
415                 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
416                         ; /* nothing */
417                 fd = open(p->pack_name, O_RDONLY);
418                 if (fd < 0)
419                         die("packfile %s cannot be opened", p->pack_name);
420                 if (fstat(fd, &st)) {
421                         close(fd);
422                         die("packfile %s cannot be opened", p->pack_name);
423                 }
424                 if (st.st_size != p->pack_size)
425                         die("packfile %s size mismatch.", p->pack_name);
426                 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
427                 close(fd);
428                 if (map == MAP_FAILED)
429                         die("packfile %s cannot be mapped.", p->pack_name);
430                 p->pack_base = map;
431
432                 /* Check if the pack file matches with the index file.
433                  * this is cheap.
434                  */
435                 if (memcmp((char*)(p->index_base) + p->index_size - 40,
436                            p->pack_base + p->pack_size - 20, 20)) {
437                               
438                         die("packfile %s does not match index.", p->pack_name);
439                 }
440         }
441         p->pack_last_used = pack_used_ctr++;
442         p->pack_use_cnt++;
443         return 0;
444 }
445
446 struct packed_git *add_packed_git(char *path, int path_len)
447 {
448         struct stat st;
449         struct packed_git *p;
450         unsigned long idx_size;
451         void *idx_map;
452
453         if (check_packed_git_idx(path, &idx_size, &idx_map))
454                 return NULL;
455
456         /* do we have a corresponding .pack file? */
457         strcpy(path + path_len - 4, ".pack");
458         if (stat(path, &st) || !S_ISREG(st.st_mode)) {
459                 munmap(idx_map, idx_size);
460                 return NULL;
461         }
462         /* ok, it looks sane as far as we can check without
463          * actually mapping the pack file.
464          */
465         p = xmalloc(sizeof(*p) + path_len + 2);
466         strcpy(p->pack_name, path);
467         p->index_size = idx_size;
468         p->pack_size = st.st_size;
469         p->index_base = idx_map;
470         p->next = NULL;
471         p->pack_base = NULL;
472         p->pack_last_used = 0;
473         p->pack_use_cnt = 0;
474         return p;
475 }
476
477 struct packed_git *parse_pack_index(unsigned char *sha1)
478 {
479         struct packed_git *p;
480         unsigned long idx_size;
481         void *idx_map;
482         char *path = sha1_pack_index_name(sha1);
483
484         if (check_packed_git_idx(path, &idx_size, &idx_map))
485                 return NULL;
486
487         path = sha1_pack_name(sha1);
488
489         p = xmalloc(sizeof(*p) + strlen(path) + 2);
490         strcpy(p->pack_name, path);
491         p->index_size = idx_size;
492         p->pack_size = 0;
493         p->index_base = idx_map;
494         p->next = NULL;
495         p->pack_base = NULL;
496         p->pack_last_used = 0;
497         p->pack_use_cnt = 0;
498         memcpy(p->sha1, sha1, 20);
499         return p;
500 }
501
502 void install_packed_git(struct packed_git *pack)
503 {
504         pack->next = packed_git;
505         packed_git = pack;
506 }
507
508 static void prepare_packed_git_one(char *objdir)
509 {
510         char path[PATH_MAX];
511         int len;
512         DIR *dir;
513         struct dirent *de;
514
515         sprintf(path, "%s/pack", objdir);
516         len = strlen(path);
517         dir = opendir(path);
518         if (!dir)
519                 return;
520         path[len++] = '/';
521         while ((de = readdir(dir)) != NULL) {
522                 int namelen = strlen(de->d_name);
523                 struct packed_git *p;
524
525                 if (strcmp(de->d_name + namelen - 4, ".idx"))
526                         continue;
527
528                 /* we have .idx.  Is it a file we can map? */
529                 strcpy(path + len, de->d_name);
530                 p = add_packed_git(path, len + namelen);
531                 if (!p)
532                         continue;
533                 p->next = packed_git;
534                 packed_git = p;
535         }
536         closedir(dir);
537 }
538
539 void prepare_packed_git(void)
540 {
541         static int run_once = 0;
542         struct alternate_object_database *alt;
543
544         if (run_once)
545                 return;
546         prepare_packed_git_one(get_object_directory());
547         prepare_alt_odb();
548         for (alt = alt_odb_list; alt; alt = alt->next) {
549                 alt->name[0] = 0;
550                 prepare_packed_git_one(alt->base);
551         }
552         run_once = 1;
553 }
554
555 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
556 {
557         char header[100];
558         unsigned char real_sha1[20];
559         SHA_CTX c;
560
561         SHA1_Init(&c);
562         SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
563         SHA1_Update(&c, map, size);
564         SHA1_Final(real_sha1, &c);
565         return memcmp(sha1, real_sha1, 20) ? -1 : 0;
566 }
567
568 static void *map_sha1_file_internal(const unsigned char *sha1,
569                                     unsigned long *size)
570 {
571         struct stat st;
572         void *map;
573         int fd;
574         char *filename = find_sha1_file(sha1, &st);
575
576         if (!filename) {
577                 return NULL;
578         }
579
580         fd = open(filename, O_RDONLY | sha1_file_open_flag);
581         if (fd < 0) {
582                 /* See if it works without O_NOATIME */
583                 switch (sha1_file_open_flag) {
584                 default:
585                         fd = open(filename, O_RDONLY);
586                         if (fd >= 0)
587                                 break;
588                 /* Fallthrough */
589                 case 0:
590                         return NULL;
591                 }
592
593                 /* If it failed once, it will probably fail again.
594                  * Stop using O_NOATIME
595                  */
596                 sha1_file_open_flag = 0;
597         }
598         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
599         close(fd);
600         if (map == MAP_FAILED)
601                 return NULL;
602         *size = st.st_size;
603         return map;
604 }
605
606 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
607 {
608         /* Get the data stream */
609         memset(stream, 0, sizeof(*stream));
610         stream->next_in = map;
611         stream->avail_in = mapsize;
612         stream->next_out = buffer;
613         stream->avail_out = size;
614
615         inflateInit(stream);
616         return inflate(stream, 0);
617 }
618
619 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
620 {
621         int bytes = strlen(buffer) + 1;
622         unsigned char *buf = xmalloc(1+size);
623
624         memcpy(buf, buffer + bytes, stream->total_out - bytes);
625         bytes = stream->total_out - bytes;
626         if (bytes < size) {
627                 stream->next_out = buf + bytes;
628                 stream->avail_out = size - bytes;
629                 while (inflate(stream, Z_FINISH) == Z_OK)
630                         /* nothing */;
631         }
632         buf[size] = 0;
633         inflateEnd(stream);
634         return buf;
635 }
636
637 /*
638  * We used to just use "sscanf()", but that's actually way
639  * too permissive for what we want to check. So do an anal
640  * object header parse by hand.
641  */
642 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
643 {
644         int i;
645         unsigned long size;
646
647         /*
648          * The type can be at most ten bytes (including the 
649          * terminating '\0' that we add), and is followed by
650          * a space. 
651          */
652         i = 10;
653         for (;;) {
654                 char c = *hdr++;
655                 if (c == ' ')
656                         break;
657                 if (!--i)
658                         return -1;
659                 *type++ = c;
660         }
661         *type = 0;
662
663         /*
664          * The length must follow immediately, and be in canonical
665          * decimal format (ie "010" is not valid).
666          */
667         size = *hdr++ - '0';
668         if (size > 9)
669                 return -1;
670         if (size) {
671                 for (;;) {
672                         unsigned long c = *hdr - '0';
673                         if (c > 9)
674                                 break;
675                         hdr++;
676                         size = size * 10 + c;
677                 }
678         }
679         *sizep = size;
680
681         /*
682          * The length must be followed by a zero byte
683          */
684         return *hdr ? -1 : 0;
685 }
686
687 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
688 {
689         int ret;
690         z_stream stream;
691         char hdr[8192];
692
693         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
694         if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
695                 return NULL;
696
697         return unpack_sha1_rest(&stream, hdr, *size);
698 }
699
700 /* forward declaration for a mutually recursive function */
701 static int packed_object_info(struct pack_entry *entry,
702                               char *type, unsigned long *sizep);
703
704 static int packed_delta_info(unsigned char *base_sha1,
705                              unsigned long delta_size,
706                              unsigned long left,
707                              char *type,
708                              unsigned long *sizep,
709                              struct packed_git *p)
710 {
711         struct pack_entry base_ent;
712
713         if (left < 20)
714                 die("truncated pack file");
715
716         /* The base entry _must_ be in the same pack */
717         if (!find_pack_entry_one(base_sha1, &base_ent, p))
718                 die("failed to find delta-pack base object %s",
719                     sha1_to_hex(base_sha1));
720
721         /* We choose to only get the type of the base object and
722          * ignore potentially corrupt pack file that expects the delta
723          * based on a base with a wrong size.  This saves tons of
724          * inflate() calls.
725          */
726
727         if (packed_object_info(&base_ent, type, NULL))
728                 die("cannot get info for delta-pack base");
729
730         if (sizep) {
731                 const unsigned char *data;
732                 unsigned char delta_head[64];
733                 unsigned long result_size;
734                 z_stream stream;
735                 int st;
736
737                 memset(&stream, 0, sizeof(stream));
738
739                 data = stream.next_in = base_sha1 + 20;
740                 stream.avail_in = left - 20;
741                 stream.next_out = delta_head;
742                 stream.avail_out = sizeof(delta_head);
743
744                 inflateInit(&stream);
745                 st = inflate(&stream, Z_FINISH);
746                 inflateEnd(&stream);
747                 if ((st != Z_STREAM_END) &&
748                     stream.total_out != sizeof(delta_head))
749                         die("delta data unpack-initial failed");
750
751                 /* Examine the initial part of the delta to figure out
752                  * the result size.
753                  */
754                 data = delta_head;
755                 get_delta_hdr_size(&data); /* ignore base size */
756
757                 /* Read the result size */
758                 result_size = get_delta_hdr_size(&data);
759                 *sizep = result_size;
760         }
761         return 0;
762 }
763
764 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
765         enum object_type *type, unsigned long *sizep)
766 {
767         unsigned shift;
768         unsigned char *pack, c;
769         unsigned long size;
770
771         if (offset >= p->pack_size)
772                 die("object offset outside of pack file");
773
774         pack =  p->pack_base + offset;
775         c = *pack++;
776         offset++;
777         *type = (c >> 4) & 7;
778         size = c & 15;
779         shift = 4;
780         while (c & 0x80) {
781                 if (offset >= p->pack_size)
782                         die("object offset outside of pack file");
783                 c = *pack++;
784                 offset++;
785                 size += (c & 0x7f) << shift;
786                 shift += 7;
787         }
788         *sizep = size;
789         return offset;
790 }
791
792 void packed_object_info_detail(struct pack_entry *e,
793                                char *type,
794                                unsigned long *size,
795                                unsigned long *store_size,
796                                int *delta_chain_length,
797                                unsigned char *base_sha1)
798 {
799         struct packed_git *p = e->p;
800         unsigned long offset, left;
801         unsigned char *pack;
802         enum object_type kind;
803
804         offset = unpack_object_header(p, e->offset, &kind, size);
805         pack = p->pack_base + offset;
806         left = p->pack_size - offset;
807         if (kind != OBJ_DELTA)
808                 *delta_chain_length = 0;
809         else {
810                 int chain_length = 0;
811                 memcpy(base_sha1, pack, 20);
812                 do {
813                         struct pack_entry base_ent;
814                         unsigned long junk;
815
816                         find_pack_entry_one(pack, &base_ent, p);
817                         offset = unpack_object_header(p, base_ent.offset,
818                                                       &kind, &junk);
819                         pack = p->pack_base + offset;
820                         chain_length++;
821                 } while (kind == OBJ_DELTA);
822                 *delta_chain_length = chain_length;
823         }
824         switch (kind) {
825         case OBJ_COMMIT:
826                 strcpy(type, "commit");
827                 break;
828         case OBJ_TREE:
829                 strcpy(type, "tree");
830                 break;
831         case OBJ_BLOB:
832                 strcpy(type, "blob");
833                 break;
834         case OBJ_TAG:
835                 strcpy(type, "tag");
836                 break;
837         default:
838                 die("corrupted pack file");
839         }
840         *store_size = 0; /* notyet */
841 }
842
843 static int packed_object_info(struct pack_entry *entry,
844                               char *type, unsigned long *sizep)
845 {
846         struct packed_git *p = entry->p;
847         unsigned long offset, size, left;
848         unsigned char *pack;
849         enum object_type kind;
850         int retval;
851
852         if (use_packed_git(p))
853                 die("cannot map packed file");
854
855         offset = unpack_object_header(p, entry->offset, &kind, &size);
856         pack = p->pack_base + offset;
857         left = p->pack_size - offset;
858
859         switch (kind) {
860         case OBJ_DELTA:
861                 retval = packed_delta_info(pack, size, left, type, sizep, p);
862                 unuse_packed_git(p);
863                 return retval;
864         case OBJ_COMMIT:
865                 strcpy(type, "commit");
866                 break;
867         case OBJ_TREE:
868                 strcpy(type, "tree");
869                 break;
870         case OBJ_BLOB:
871                 strcpy(type, "blob");
872                 break;
873         case OBJ_TAG:
874                 strcpy(type, "tag");
875                 break;
876         default:
877                 die("corrupted pack file");
878         }
879         if (sizep)
880                 *sizep = size;
881         unuse_packed_git(p);
882         return 0;
883 }
884
885 /* forward declaration for a mutually recursive function */
886 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
887
888 static void *unpack_delta_entry(unsigned char *base_sha1,
889                                 unsigned long delta_size,
890                                 unsigned long left,
891                                 char *type,
892                                 unsigned long *sizep,
893                                 struct packed_git *p)
894 {
895         struct pack_entry base_ent;
896         void *data, *delta_data, *result, *base;
897         unsigned long data_size, result_size, base_size;
898         z_stream stream;
899         int st;
900
901         if (left < 20)
902                 die("truncated pack file");
903         data = base_sha1 + 20;
904         data_size = left - 20;
905         delta_data = xmalloc(delta_size);
906
907         memset(&stream, 0, sizeof(stream));
908
909         stream.next_in = data;
910         stream.avail_in = data_size;
911         stream.next_out = delta_data;
912         stream.avail_out = delta_size;
913
914         inflateInit(&stream);
915         st = inflate(&stream, Z_FINISH);
916         inflateEnd(&stream);
917         if ((st != Z_STREAM_END) || stream.total_out != delta_size)
918                 die("delta data unpack failed");
919
920         /* The base entry _must_ be in the same pack */
921         if (!find_pack_entry_one(base_sha1, &base_ent, p))
922                 die("failed to find delta-pack base object %s",
923                     sha1_to_hex(base_sha1));
924         base = unpack_entry_gently(&base_ent, type, &base_size);
925         if (!base)
926                 die("failed to read delta-pack base object %s",
927                     sha1_to_hex(base_sha1));
928         result = patch_delta(base, base_size,
929                              delta_data, delta_size,
930                              &result_size);
931         if (!result)
932                 die("failed to apply delta");
933         free(delta_data);
934         free(base);
935         *sizep = result_size;
936         return result;
937 }
938
939 static void *unpack_non_delta_entry(unsigned char *data,
940                                     unsigned long size,
941                                     unsigned long left)
942 {
943         int st;
944         z_stream stream;
945         unsigned char *buffer;
946
947         buffer = xmalloc(size + 1);
948         buffer[size] = 0;
949         memset(&stream, 0, sizeof(stream));
950         stream.next_in = data;
951         stream.avail_in = left;
952         stream.next_out = buffer;
953         stream.avail_out = size;
954
955         inflateInit(&stream);
956         st = inflate(&stream, Z_FINISH);
957         inflateEnd(&stream);
958         if ((st != Z_STREAM_END) || stream.total_out != size) {
959                 free(buffer);
960                 return NULL;
961         }
962
963         return buffer;
964 }
965
966 static void *unpack_entry(struct pack_entry *entry,
967                           char *type, unsigned long *sizep)
968 {
969         struct packed_git *p = entry->p;
970         void *retval;
971
972         if (use_packed_git(p))
973                 die("cannot map packed file");
974         retval = unpack_entry_gently(entry, type, sizep);
975         unuse_packed_git(p);
976         if (!retval)
977                 die("corrupted pack file");
978         return retval;
979 }
980
981 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
982 void *unpack_entry_gently(struct pack_entry *entry,
983                           char *type, unsigned long *sizep)
984 {
985         struct packed_git *p = entry->p;
986         unsigned long offset, size, left;
987         unsigned char *pack;
988         enum object_type kind;
989         void *retval;
990
991         offset = unpack_object_header(p, entry->offset, &kind, &size);
992         pack = p->pack_base + offset;
993         left = p->pack_size - offset;
994         switch (kind) {
995         case OBJ_DELTA:
996                 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
997                 return retval;
998         case OBJ_COMMIT:
999                 strcpy(type, "commit");
1000                 break;
1001         case OBJ_TREE:
1002                 strcpy(type, "tree");
1003                 break;
1004         case OBJ_BLOB:
1005                 strcpy(type, "blob");
1006                 break;
1007         case OBJ_TAG:
1008                 strcpy(type, "tag");
1009                 break;
1010         default:
1011                 return NULL;
1012         }
1013         *sizep = size;
1014         retval = unpack_non_delta_entry(pack, size, left);
1015         return retval;
1016 }
1017
1018 int num_packed_objects(const struct packed_git *p)
1019 {
1020         /* See check_packed_git_idx() */
1021         return (p->index_size - 20 - 20 - 4*256) / 24;
1022 }
1023
1024 int nth_packed_object_sha1(const struct packed_git *p, int n,
1025                            unsigned char* sha1)
1026 {
1027         void *index = p->index_base + 256;
1028         if (n < 0 || num_packed_objects(p) <= n)
1029                 return -1;
1030         memcpy(sha1, (index + 24 * n + 4), 20);
1031         return 0;
1032 }
1033
1034 int find_pack_entry_one(const unsigned char *sha1,
1035                         struct pack_entry *e, struct packed_git *p)
1036 {
1037         unsigned int *level1_ofs = p->index_base;
1038         int hi = ntohl(level1_ofs[*sha1]);
1039         int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1040         void *index = p->index_base + 256;
1041
1042         do {
1043                 int mi = (lo + hi) / 2;
1044                 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1045                 if (!cmp) {
1046                         e->offset = ntohl(*((int*)(index + 24 * mi)));
1047                         memcpy(e->sha1, sha1, 20);
1048                         e->p = p;
1049                         return 1;
1050                 }
1051                 if (cmp > 0)
1052                         hi = mi;
1053                 else
1054                         lo = mi+1;
1055         } while (lo < hi);
1056         return 0;
1057 }
1058
1059 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1060 {
1061         struct packed_git *p;
1062         prepare_packed_git();
1063
1064         for (p = packed_git; p; p = p->next) {
1065                 if (find_pack_entry_one(sha1, e, p))
1066                         return 1;
1067         }
1068         return 0;
1069 }
1070
1071 struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1072                                   struct packed_git *packs)
1073 {
1074         struct packed_git *p;
1075         struct pack_entry e;
1076
1077         for (p = packs; p; p = p->next) {
1078                 if (find_pack_entry_one(sha1, &e, p))
1079                         return p;
1080         }
1081         return NULL;
1082         
1083 }
1084
1085 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1086 {
1087         int status;
1088         unsigned long mapsize, size;
1089         void *map;
1090         z_stream stream;
1091         char hdr[128];
1092
1093         map = map_sha1_file_internal(sha1, &mapsize);
1094         if (!map) {
1095                 struct pack_entry e;
1096
1097                 if (!find_pack_entry(sha1, &e))
1098                         return error("unable to find %s", sha1_to_hex(sha1));
1099                 return packed_object_info(&e, type, sizep);
1100         }
1101         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1102                 status = error("unable to unpack %s header",
1103                                sha1_to_hex(sha1));
1104         if (parse_sha1_header(hdr, type, &size) < 0)
1105                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1106         else {
1107                 status = 0;
1108                 if (sizep)
1109                         *sizep = size;
1110         }
1111         inflateEnd(&stream);
1112         munmap(map, mapsize);
1113         return status;
1114 }
1115
1116 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1117 {
1118         struct pack_entry e;
1119
1120         if (!find_pack_entry(sha1, &e)) {
1121                 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1122                 return NULL;
1123         }
1124         return unpack_entry(&e, type, size);
1125 }
1126
1127 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1128 {
1129         unsigned long mapsize;
1130         void *map, *buf;
1131         struct pack_entry e;
1132
1133         if (find_pack_entry(sha1, &e))
1134                 return read_packed_sha1(sha1, type, size);
1135         map = map_sha1_file_internal(sha1, &mapsize);
1136         if (map) {
1137                 buf = unpack_sha1_file(map, mapsize, type, size);
1138                 munmap(map, mapsize);
1139                 return buf;
1140         }
1141         return NULL;
1142 }
1143
1144 void *read_object_with_reference(const unsigned char *sha1,
1145                                  const char *required_type,
1146                                  unsigned long *size,
1147                                  unsigned char *actual_sha1_return)
1148 {
1149         char type[20];
1150         void *buffer;
1151         unsigned long isize;
1152         unsigned char actual_sha1[20];
1153
1154         memcpy(actual_sha1, sha1, 20);
1155         while (1) {
1156                 int ref_length = -1;
1157                 const char *ref_type = NULL;
1158
1159                 buffer = read_sha1_file(actual_sha1, type, &isize);
1160                 if (!buffer)
1161                         return NULL;
1162                 if (!strcmp(type, required_type)) {
1163                         *size = isize;
1164                         if (actual_sha1_return)
1165                                 memcpy(actual_sha1_return, actual_sha1, 20);
1166                         return buffer;
1167                 }
1168                 /* Handle references */
1169                 else if (!strcmp(type, "commit"))
1170                         ref_type = "tree ";
1171                 else if (!strcmp(type, "tag"))
1172                         ref_type = "object ";
1173                 else {
1174                         free(buffer);
1175                         return NULL;
1176                 }
1177                 ref_length = strlen(ref_type);
1178
1179                 if (memcmp(buffer, ref_type, ref_length) ||
1180                     get_sha1_hex(buffer + ref_length, actual_sha1)) {
1181                         free(buffer);
1182                         return NULL;
1183                 }
1184                 free(buffer);
1185                 /* Now we have the ID of the referred-to object in
1186                  * actual_sha1.  Check again. */
1187         }
1188 }
1189
1190 char *write_sha1_file_prepare(void *buf,
1191                               unsigned long len,
1192                               const char *type,
1193                               unsigned char *sha1,
1194                               unsigned char *hdr,
1195                               int *hdrlen)
1196 {
1197         SHA_CTX c;
1198
1199         /* Generate the header */
1200         *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1201
1202         /* Sha1.. */
1203         SHA1_Init(&c);
1204         SHA1_Update(&c, hdr, *hdrlen);
1205         SHA1_Update(&c, buf, len);
1206         SHA1_Final(sha1, &c);
1207
1208         return sha1_file_name(sha1);
1209 }
1210
1211 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1212 {
1213         int size;
1214         unsigned char *compressed;
1215         z_stream stream;
1216         unsigned char sha1[20];
1217         char *filename;
1218         static char tmpfile[PATH_MAX];
1219         unsigned char hdr[50];
1220         int fd, hdrlen, ret;
1221
1222         /* Normally if we have it in the pack then we do not bother writing
1223          * it out into .git/objects/??/?{38} file.
1224          */
1225         filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1226         if (returnsha1)
1227                 memcpy(returnsha1, sha1, 20);
1228         if (has_sha1_file(sha1))
1229                 return 0;
1230         fd = open(filename, O_RDONLY);
1231         if (fd >= 0) {
1232                 /*
1233                  * FIXME!!! We might do collision checking here, but we'd
1234                  * need to uncompress the old file and check it. Later.
1235                  */
1236                 close(fd);
1237                 return 0;
1238         }
1239
1240         if (errno != ENOENT) {
1241                 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1242                 return -1;
1243         }
1244
1245         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1246
1247         fd = mkstemp(tmpfile);
1248         if (fd < 0) {
1249                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1250                 return -1;
1251         }
1252
1253         /* Set it up */
1254         memset(&stream, 0, sizeof(stream));
1255         deflateInit(&stream, Z_BEST_COMPRESSION);
1256         size = deflateBound(&stream, len+hdrlen);
1257         compressed = xmalloc(size);
1258
1259         /* Compress it */
1260         stream.next_out = compressed;
1261         stream.avail_out = size;
1262
1263         /* First header.. */
1264         stream.next_in = hdr;
1265         stream.avail_in = hdrlen;
1266         while (deflate(&stream, 0) == Z_OK)
1267                 /* nothing */;
1268
1269         /* Then the data itself.. */
1270         stream.next_in = buf;
1271         stream.avail_in = len;
1272         while (deflate(&stream, Z_FINISH) == Z_OK)
1273                 /* nothing */;
1274         deflateEnd(&stream);
1275         size = stream.total_out;
1276
1277         if (write(fd, compressed, size) != size)
1278                 die("unable to write file");
1279         fchmod(fd, 0444);
1280         close(fd);
1281         free(compressed);
1282
1283         ret = link(tmpfile, filename);
1284         if (ret < 0) {
1285                 ret = errno;
1286
1287                 /*
1288                  * Coda hack - coda doesn't like cross-directory links,
1289                  * so we fall back to a rename, which will mean that it
1290                  * won't be able to check collisions, but that's not a
1291                  * big deal.
1292                  *
1293                  * When this succeeds, we just return 0. We have nothing
1294                  * left to unlink.
1295                  */
1296                 if (ret == EXDEV && !rename(tmpfile, filename))
1297                         return 0;
1298         }
1299         unlink(tmpfile);
1300         if (ret) {
1301                 if (ret != EEXIST) {
1302                         fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1303                         return -1;
1304                 }
1305                 /* FIXME!!! Collision check here ? */
1306         }
1307
1308         return 0;
1309 }
1310
1311 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1312 {
1313         ssize_t size;
1314         unsigned long objsize;
1315         int posn = 0;
1316         void *map = map_sha1_file_internal(sha1, &objsize);
1317         void *buf = map;
1318         void *temp_obj = NULL;
1319         z_stream stream;
1320
1321         if (!buf) {
1322                 unsigned char *unpacked;
1323                 unsigned long len;
1324                 char type[20];
1325                 char hdr[50];
1326                 int hdrlen;
1327                 // need to unpack and recompress it by itself
1328                 unpacked = read_packed_sha1(sha1, type, &len);
1329
1330                 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1331
1332                 /* Set it up */
1333                 memset(&stream, 0, sizeof(stream));
1334                 deflateInit(&stream, Z_BEST_COMPRESSION);
1335                 size = deflateBound(&stream, len + hdrlen);
1336                 temp_obj = buf = xmalloc(size);
1337
1338                 /* Compress it */
1339                 stream.next_out = buf;
1340                 stream.avail_out = size;
1341                 
1342                 /* First header.. */
1343                 stream.next_in = (void *)hdr;
1344                 stream.avail_in = hdrlen;
1345                 while (deflate(&stream, 0) == Z_OK)
1346                         /* nothing */;
1347
1348                 /* Then the data itself.. */
1349                 stream.next_in = unpacked;
1350                 stream.avail_in = len;
1351                 while (deflate(&stream, Z_FINISH) == Z_OK)
1352                         /* nothing */;
1353                 deflateEnd(&stream);
1354                 free(unpacked);
1355                 
1356                 objsize = stream.total_out;
1357         }
1358
1359         do {
1360                 size = write(fd, buf + posn, objsize - posn);
1361                 if (size <= 0) {
1362                         if (!size) {
1363                                 fprintf(stderr, "write closed");
1364                         } else {
1365                                 perror("write ");
1366                         }
1367                         return -1;
1368                 }
1369                 posn += size;
1370         } while (posn < objsize);
1371
1372         if (map)
1373                 munmap(map, objsize);
1374         if (temp_obj)
1375                 free(temp_obj);
1376
1377         return 0;
1378 }
1379
1380 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1381                        size_t bufsize, size_t *bufposn)
1382 {
1383         char *filename = sha1_file_name(sha1);
1384
1385         int local;
1386         z_stream stream;
1387         unsigned char real_sha1[20];
1388         unsigned char discard[4096];
1389         int ret;
1390         SHA_CTX c;
1391
1392         local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1393
1394         if (local < 0)
1395                 return error("Couldn't open %s\n", filename);
1396
1397         memset(&stream, 0, sizeof(stream));
1398
1399         inflateInit(&stream);
1400
1401         SHA1_Init(&c);
1402
1403         do {
1404                 ssize_t size;
1405                 if (*bufposn) {
1406                         stream.avail_in = *bufposn;
1407                         stream.next_in = (unsigned char *) buffer;
1408                         do {
1409                                 stream.next_out = discard;
1410                                 stream.avail_out = sizeof(discard);
1411                                 ret = inflate(&stream, Z_SYNC_FLUSH);
1412                                 SHA1_Update(&c, discard, sizeof(discard) -
1413                                             stream.avail_out);
1414                         } while (stream.avail_in && ret == Z_OK);
1415                         write(local, buffer, *bufposn - stream.avail_in);
1416                         memmove(buffer, buffer + *bufposn - stream.avail_in,
1417                                 stream.avail_in);
1418                         *bufposn = stream.avail_in;
1419                         if (ret != Z_OK)
1420                                 break;
1421                 }
1422                 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1423                 if (size <= 0) {
1424                         close(local);
1425                         unlink(filename);
1426                         if (!size)
1427                                 return error("Connection closed?");
1428                         perror("Reading from connection");
1429                         return -1;
1430                 }
1431                 *bufposn += size;
1432         } while (1);
1433         inflateEnd(&stream);
1434
1435         close(local);
1436         SHA1_Final(real_sha1, &c);
1437         if (ret != Z_STREAM_END) {
1438                 unlink(filename);
1439                 return error("File %s corrupted", sha1_to_hex(sha1));
1440         }
1441         if (memcmp(sha1, real_sha1, 20)) {
1442                 unlink(filename);
1443                 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1444         }
1445         
1446         return 0;
1447 }
1448
1449 int has_pack_index(const unsigned char *sha1)
1450 {
1451         struct stat st;
1452         if (stat(sha1_pack_index_name(sha1), &st))
1453                 return 0;
1454         return 1;
1455 }
1456
1457 int has_pack_file(const unsigned char *sha1)
1458 {
1459         struct stat st;
1460         if (stat(sha1_pack_name(sha1), &st))
1461                 return 0;
1462         return 1;
1463 }
1464
1465 int has_sha1_pack(const unsigned char *sha1)
1466 {
1467         struct pack_entry e;
1468         return find_pack_entry(sha1, &e);
1469 }
1470
1471 int has_sha1_file(const unsigned char *sha1)
1472 {
1473         struct stat st;
1474         struct pack_entry e;
1475
1476         if (find_pack_entry(sha1, &e))
1477                 return 1;
1478         return find_sha1_file(sha1, &st) ? 1 : 0;
1479 }
1480
1481 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1482 {
1483         unsigned long size = st->st_size;
1484         void *buf;
1485         int ret;
1486         unsigned char hdr[50];
1487         int hdrlen;
1488
1489         buf = "";
1490         if (size)
1491                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1492         close(fd);
1493         if (buf == MAP_FAILED)
1494                 return -1;
1495
1496         if (!type)
1497                 type = "blob";
1498         if (write_object)
1499                 ret = write_sha1_file(buf, size, type, sha1);
1500         else {
1501                 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1502                 ret = 0;
1503         }
1504         if (size)
1505                 munmap(buf, size);
1506         return ret;
1507 }