[PATCH] git-merge-cache -q doesn't complain about failing merge program
[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 int get_sha1_file(const char *path, unsigned char *result)
50 {
51         char buffer[60];
52         int fd = open(path, O_RDONLY);
53         int len;
54
55         if (fd < 0)
56                 return -1;
57         len = read(fd, buffer, sizeof(buffer));
58         close(fd);
59         if (len < 40)
60                 return -1;
61         return get_sha1_hex(buffer, result);
62 }
63
64 static char *git_dir, *git_object_dir, *git_index_file, *git_refs_dir,
65         *git_graft_file;
66 static void setup_git_env(void)
67 {
68         git_dir = gitenv(GIT_DIR_ENVIRONMENT);
69         if (!git_dir)
70                 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
71         git_object_dir = gitenv(DB_ENVIRONMENT);
72         if (!git_object_dir) {
73                 git_object_dir = xmalloc(strlen(git_dir) + 9);
74                 sprintf(git_object_dir, "%s/objects", git_dir);
75         }
76         git_refs_dir = xmalloc(strlen(git_dir) + 6);
77         sprintf(git_refs_dir, "%s/refs", git_dir);
78         git_index_file = gitenv(INDEX_ENVIRONMENT);
79         if (!git_index_file) {
80                 git_index_file = xmalloc(strlen(git_dir) + 7);
81                 sprintf(git_index_file, "%s/index", git_dir);
82         }
83         git_graft_file = gitenv(GRAFT_ENVIRONMENT);
84         if (!git_graft_file)
85                 git_graft_file = strdup(git_path("info/grafts"));
86 }
87
88 char *get_object_directory(void)
89 {
90         if (!git_object_dir)
91                 setup_git_env();
92         return git_object_dir;
93 }
94
95 char *get_refs_directory(void)
96 {
97         if (!git_refs_dir)
98                 setup_git_env();
99         return git_refs_dir;
100 }
101
102 char *get_index_file(void)
103 {
104         if (!git_index_file)
105                 setup_git_env();
106         return git_index_file;
107 }
108
109 char *get_graft_file(void)
110 {
111         if (!git_graft_file)
112                 setup_git_env();
113         return git_graft_file;
114 }
115
116 int safe_create_leading_directories(char *path)
117 {
118         char *pos = path;
119
120         while (pos) {
121                 pos = strchr(pos, '/');
122                 if (!pos)
123                         break;
124                 *pos = 0;
125                 if (mkdir(path, 0777) < 0)
126                         if (errno != EEXIST) {
127                                 *pos = '/';
128                                 return -1;
129                         }
130                 *pos++ = '/';
131         }
132         return 0;
133 }
134
135 int get_sha1(const char *str, unsigned char *sha1)
136 {
137         static const char *prefix[] = {
138                 "",
139                 "refs",
140                 "refs/tags",
141                 "refs/heads",
142                 "refs/snap",
143                 NULL
144         };
145         const char **p;
146
147         if (!get_sha1_hex(str, sha1))
148                 return 0;
149
150         for (p = prefix; *p; p++) {
151                 char * pathname = git_path("%s/%s", *p, str);
152                 if (!get_sha1_file(pathname, sha1))
153                         return 0;
154         }
155
156         return -1;
157 }
158
159 char * sha1_to_hex(const unsigned char *sha1)
160 {
161         static char buffer[50];
162         static const char hex[] = "0123456789abcdef";
163         char *buf = buffer;
164         int i;
165
166         for (i = 0; i < 20; i++) {
167                 unsigned int val = *sha1++;
168                 *buf++ = hex[val >> 4];
169                 *buf++ = hex[val & 0xf];
170         }
171         return buffer;
172 }
173
174 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
175 {
176         int i;
177         for (i = 0; i < 20; i++) {
178                 static char hex[] = "0123456789abcdef";
179                 unsigned int val = sha1[i];
180                 char *pos = pathbuf + i*2 + (i > 0);
181                 *pos++ = hex[val >> 4];
182                 *pos = hex[val & 0xf];
183         }
184 }
185
186 /*
187  * NOTE! This returns a statically allocated buffer, so you have to be
188  * careful about using it. Do a "strdup()" if you need to save the
189  * filename.
190  *
191  * Also note that this returns the location for creating.  Reading
192  * SHA1 file can happen from any alternate directory listed in the
193  * DB_ENVIRONMENT environment variable if it is not found in
194  * the primary object database.
195  */
196 char *sha1_file_name(const unsigned char *sha1)
197 {
198         static char *name, *base;
199
200         if (!base) {
201                 const char *sha1_file_directory = get_object_directory();
202                 int len = strlen(sha1_file_directory);
203                 base = xmalloc(len + 60);
204                 memcpy(base, sha1_file_directory, len);
205                 memset(base+len, 0, 60);
206                 base[len] = '/';
207                 base[len+3] = '/';
208                 name = base + len + 1;
209         }
210         fill_sha1_path(name, sha1);
211         return base;
212 }
213
214 char *sha1_pack_name(const unsigned char *sha1)
215 {
216         static const char hex[] = "0123456789abcdef";
217         static char *name, *base, *buf;
218         int i;
219
220         if (!base) {
221                 const char *sha1_file_directory = get_object_directory();
222                 int len = strlen(sha1_file_directory);
223                 base = xmalloc(len + 60);
224                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
225                 name = base + len + 11;
226         }
227
228         buf = name;
229
230         for (i = 0; i < 20; i++) {
231                 unsigned int val = *sha1++;
232                 *buf++ = hex[val >> 4];
233                 *buf++ = hex[val & 0xf];
234         }
235         
236         return base;
237 }
238
239 char *sha1_pack_index_name(const unsigned char *sha1)
240 {
241         static const char hex[] = "0123456789abcdef";
242         static char *name, *base, *buf;
243         int i;
244
245         if (!base) {
246                 const char *sha1_file_directory = get_object_directory();
247                 int len = strlen(sha1_file_directory);
248                 base = xmalloc(len + 60);
249                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
250                 name = base + len + 11;
251         }
252
253         buf = name;
254
255         for (i = 0; i < 20; i++) {
256                 unsigned int val = *sha1++;
257                 *buf++ = hex[val >> 4];
258                 *buf++ = hex[val & 0xf];
259         }
260         
261         return base;
262 }
263
264 struct alternate_object_database *alt_odb;
265
266 /*
267  * Prepare alternate object database registry.
268  * alt_odb points at an array of struct alternate_object_database.
269  * This array is terminated with an element that has both its base
270  * and name set to NULL.  alt_odb[n] comes from n'th non-empty
271  * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
272  * variable, and its base points at a statically allocated buffer
273  * that contains "/the/directory/corresponding/to/.git/objects/...",
274  * while its name points just after the slash at the end of
275  * ".git/objects/" in the example above, and has enough space to hold
276  * 40-byte hex SHA1, an extra slash for the first level indirection,
277  * and the terminating NUL.
278  * This function allocates the alt_odb array and all the strings
279  * pointed by base fields of the array elements with one xmalloc();
280  * the string pool immediately follows the array.
281  */
282 void prepare_alt_odb(void)
283 {
284         int pass, totlen, i;
285         const char *cp, *last;
286         char *op = NULL;
287         const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
288
289         if (alt_odb)
290                 return;
291         /* The first pass counts how large an area to allocate to
292          * hold the entire alt_odb structure, including array of
293          * structs and path buffers for them.  The second pass fills
294          * the structure and prepares the path buffers for use by
295          * fill_sha1_path().
296          */
297         for (totlen = pass = 0; pass < 2; pass++) {
298                 last = alt;
299                 i = 0;
300                 do {
301                         cp = strchr(last, ':') ? : last + strlen(last);
302                         if (last != cp) {
303                                 /* 43 = 40-byte + 2 '/' + terminating NUL */
304                                 int pfxlen = cp - last;
305                                 int entlen = pfxlen + 43;
306                                 if (pass == 0)
307                                         totlen += entlen;
308                                 else {
309                                         alt_odb[i].base = op;
310                                         alt_odb[i].name = op + pfxlen + 1;
311                                         memcpy(op, last, pfxlen);
312                                         op[pfxlen] = op[pfxlen + 3] = '/';
313                                         op[entlen-1] = 0;
314                                         op += entlen;
315                                 }
316                                 i++;
317                         }
318                         while (*cp && *cp == ':')
319                                 cp++;
320                         last = cp;
321                 } while (*cp);
322                 if (pass)
323                         break;
324                 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
325                 alt_odb[i].base = alt_odb[i].name = NULL;
326                 op = (char*)(&alt_odb[i+1]);
327         }
328 }
329
330 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
331 {
332         int i;
333         char *name = sha1_file_name(sha1);
334
335         if (!stat(name, st))
336                 return name;
337         prepare_alt_odb();
338         for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
339                 fill_sha1_path(name, sha1);
340                 if (!stat(alt_odb[i].base, st))
341                         return alt_odb[i].base;
342         }
343         return NULL;
344 }
345
346 #define PACK_MAX_SZ (1<<26)
347 static int pack_used_ctr;
348 static unsigned long pack_mapped;
349 struct packed_git *packed_git;
350
351 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
352                                 void **idx_map_)
353 {
354         void *idx_map;
355         unsigned int *index;
356         unsigned long idx_size;
357         int nr, i;
358         int fd = open(path, O_RDONLY);
359         struct stat st;
360         if (fd < 0)
361                 return -1;
362         if (fstat(fd, &st)) {
363                 close(fd);
364                 return -1;
365         }
366         idx_size = st.st_size;
367         idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
368         close(fd);
369         if (idx_map == MAP_FAILED)
370                 return -1;
371
372         index = idx_map;
373         *idx_map_ = idx_map;
374         *idx_size_ = idx_size;
375
376         /* check index map */
377         if (idx_size < 4*256 + 20 + 20)
378                 return error("index file too small");
379         nr = 0;
380         for (i = 0; i < 256; i++) {
381                 unsigned int n = ntohl(index[i]);
382                 if (n < nr)
383                         return error("non-monotonic index");
384                 nr = n;
385         }
386
387         /*
388          * Total size:
389          *  - 256 index entries 4 bytes each
390          *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
391          *  - 20-byte SHA1 of the packfile
392          *  - 20-byte SHA1 file checksum
393          */
394         if (idx_size != 4*256 + nr * 24 + 20 + 20)
395                 return error("wrong index file size");
396
397         return 0;
398 }
399
400 static int unuse_one_packed_git(void)
401 {
402         struct packed_git *p, *lru = NULL;
403
404         for (p = packed_git; p; p = p->next) {
405                 if (p->pack_use_cnt || !p->pack_base)
406                         continue;
407                 if (!lru || p->pack_last_used < lru->pack_last_used)
408                         lru = p;
409         }
410         if (!lru)
411                 return 0;
412         munmap(lru->pack_base, lru->pack_size);
413         lru->pack_base = NULL;
414         return 1;
415 }
416
417 void unuse_packed_git(struct packed_git *p)
418 {
419         p->pack_use_cnt--;
420 }
421
422 int use_packed_git(struct packed_git *p)
423 {
424         if (!p->pack_size) {
425                 struct stat st;
426                 // We created the struct before we had the pack
427                 stat(p->pack_name, &st);
428                 if (!S_ISREG(st.st_mode))
429                         die("packfile %s not a regular file", p->pack_name);
430                 p->pack_size = st.st_size;
431         }
432         if (!p->pack_base) {
433                 int fd;
434                 struct stat st;
435                 void *map;
436
437                 pack_mapped += p->pack_size;
438                 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
439                         ; /* nothing */
440                 fd = open(p->pack_name, O_RDONLY);
441                 if (fd < 0)
442                         die("packfile %s cannot be opened", p->pack_name);
443                 if (fstat(fd, &st)) {
444                         close(fd);
445                         die("packfile %s cannot be opened", p->pack_name);
446                 }
447                 if (st.st_size != p->pack_size)
448                         die("packfile %s size mismatch.", p->pack_name);
449                 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
450                 close(fd);
451                 if (map == MAP_FAILED)
452                         die("packfile %s cannot be mapped.", p->pack_name);
453                 p->pack_base = map;
454
455                 /* Check if the pack file matches with the index file.
456                  * this is cheap.
457                  */
458                 if (memcmp((char*)(p->index_base) + p->index_size - 40,
459                            p->pack_base + p->pack_size - 20, 20)) {
460                               
461                         die("packfile %s does not match index.", p->pack_name);
462                 }
463         }
464         p->pack_last_used = pack_used_ctr++;
465         p->pack_use_cnt++;
466         return 0;
467 }
468
469 struct packed_git *add_packed_git(char *path, int path_len)
470 {
471         struct stat st;
472         struct packed_git *p;
473         unsigned long idx_size;
474         void *idx_map;
475
476         if (check_packed_git_idx(path, &idx_size, &idx_map))
477                 return NULL;
478
479         /* do we have a corresponding .pack file? */
480         strcpy(path + path_len - 4, ".pack");
481         if (stat(path, &st) || !S_ISREG(st.st_mode)) {
482                 munmap(idx_map, idx_size);
483                 return NULL;
484         }
485         /* ok, it looks sane as far as we can check without
486          * actually mapping the pack file.
487          */
488         p = xmalloc(sizeof(*p) + path_len + 2);
489         strcpy(p->pack_name, path);
490         p->index_size = idx_size;
491         p->pack_size = st.st_size;
492         p->index_base = idx_map;
493         p->next = NULL;
494         p->pack_base = NULL;
495         p->pack_last_used = 0;
496         p->pack_use_cnt = 0;
497         return p;
498 }
499
500 struct packed_git *parse_pack_index(unsigned char *sha1)
501 {
502         struct packed_git *p;
503         unsigned long idx_size;
504         void *idx_map;
505         char *path = sha1_pack_index_name(sha1);
506
507         if (check_packed_git_idx(path, &idx_size, &idx_map))
508                 return NULL;
509
510         path = sha1_pack_name(sha1);
511
512         p = xmalloc(sizeof(*p) + strlen(path) + 2);
513         strcpy(p->pack_name, path);
514         p->index_size = idx_size;
515         p->pack_size = 0;
516         p->index_base = idx_map;
517         p->next = NULL;
518         p->pack_base = NULL;
519         p->pack_last_used = 0;
520         p->pack_use_cnt = 0;
521         memcpy(p->sha1, sha1, 20);
522         return p;
523 }
524
525 void install_packed_git(struct packed_git *pack)
526 {
527         pack->next = packed_git;
528         packed_git = pack;
529 }
530
531 static void prepare_packed_git_one(char *objdir)
532 {
533         char path[PATH_MAX];
534         int len;
535         DIR *dir;
536         struct dirent *de;
537
538         sprintf(path, "%s/pack", objdir);
539         len = strlen(path);
540         dir = opendir(path);
541         if (!dir)
542                 return;
543         path[len++] = '/';
544         while ((de = readdir(dir)) != NULL) {
545                 int namelen = strlen(de->d_name);
546                 struct packed_git *p;
547
548                 if (strcmp(de->d_name + namelen - 4, ".idx"))
549                         continue;
550
551                 /* we have .idx.  Is it a file we can map? */
552                 strcpy(path + len, de->d_name);
553                 p = add_packed_git(path, len + namelen);
554                 if (!p)
555                         continue;
556                 p->next = packed_git;
557                 packed_git = p;
558         }
559         closedir(dir);
560 }
561
562 void prepare_packed_git(void)
563 {
564         int i;
565         static int run_once = 0;
566
567         if (run_once++)
568                 return;
569
570         prepare_packed_git_one(get_object_directory());
571         prepare_alt_odb();
572         for (i = 0; alt_odb[i].base != NULL; i++) {
573                 alt_odb[i].name[0] = 0;
574                 prepare_packed_git_one(alt_odb[i].base);
575         }
576 }
577
578 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
579 {
580         char header[100];
581         unsigned char real_sha1[20];
582         SHA_CTX c;
583
584         SHA1_Init(&c);
585         SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
586         SHA1_Update(&c, map, size);
587         SHA1_Final(real_sha1, &c);
588         return memcmp(sha1, real_sha1, 20) ? -1 : 0;
589 }
590
591 static void *map_sha1_file_internal(const unsigned char *sha1,
592                                     unsigned long *size)
593 {
594         struct stat st;
595         void *map;
596         int fd;
597         char *filename = find_sha1_file(sha1, &st);
598
599         if (!filename) {
600                 return NULL;
601         }
602
603         fd = open(filename, O_RDONLY | sha1_file_open_flag);
604         if (fd < 0) {
605                 /* See if it works without O_NOATIME */
606                 switch (sha1_file_open_flag) {
607                 default:
608                         fd = open(filename, O_RDONLY);
609                         if (fd >= 0)
610                                 break;
611                 /* Fallthrough */
612                 case 0:
613                         return NULL;
614                 }
615
616                 /* If it failed once, it will probably fail again.
617                  * Stop using O_NOATIME
618                  */
619                 sha1_file_open_flag = 0;
620         }
621         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
622         close(fd);
623         if (map == MAP_FAILED)
624                 return NULL;
625         *size = st.st_size;
626         return map;
627 }
628
629 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
630 {
631         /* Get the data stream */
632         memset(stream, 0, sizeof(*stream));
633         stream->next_in = map;
634         stream->avail_in = mapsize;
635         stream->next_out = buffer;
636         stream->avail_out = size;
637
638         inflateInit(stream);
639         return inflate(stream, 0);
640 }
641
642 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
643 {
644         int bytes = strlen(buffer) + 1;
645         unsigned char *buf = xmalloc(1+size);
646
647         memcpy(buf, buffer + bytes, stream->total_out - bytes);
648         bytes = stream->total_out - bytes;
649         if (bytes < size) {
650                 stream->next_out = buf + bytes;
651                 stream->avail_out = size - bytes;
652                 while (inflate(stream, Z_FINISH) == Z_OK)
653                         /* nothing */;
654         }
655         buf[size] = 0;
656         inflateEnd(stream);
657         return buf;
658 }
659
660 /*
661  * We used to just use "sscanf()", but that's actually way
662  * too permissive for what we want to check. So do an anal
663  * object header parse by hand.
664  */
665 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
666 {
667         int i;
668         unsigned long size;
669
670         /*
671          * The type can be at most ten bytes (including the 
672          * terminating '\0' that we add), and is followed by
673          * a space. 
674          */
675         i = 10;
676         for (;;) {
677                 char c = *hdr++;
678                 if (c == ' ')
679                         break;
680                 if (!--i)
681                         return -1;
682                 *type++ = c;
683         }
684         *type = 0;
685
686         /*
687          * The length must follow immediately, and be in canonical
688          * decimal format (ie "010" is not valid).
689          */
690         size = *hdr++ - '0';
691         if (size > 9)
692                 return -1;
693         if (size) {
694                 for (;;) {
695                         unsigned long c = *hdr - '0';
696                         if (c > 9)
697                                 break;
698                         hdr++;
699                         size = size * 10 + c;
700                 }
701         }
702         *sizep = size;
703
704         /*
705          * The length must be followed by a zero byte
706          */
707         return *hdr ? -1 : 0;
708 }
709
710 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
711 {
712         int ret;
713         z_stream stream;
714         char hdr[8192];
715
716         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
717         if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
718                 return NULL;
719
720         return unpack_sha1_rest(&stream, hdr, *size);
721 }
722
723 /* forward declaration for a mutually recursive function */
724 static int packed_object_info(struct pack_entry *entry,
725                               char *type, unsigned long *sizep);
726
727 static int packed_delta_info(unsigned char *base_sha1,
728                              unsigned long delta_size,
729                              unsigned long left,
730                              char *type,
731                              unsigned long *sizep,
732                              struct packed_git *p)
733 {
734         struct pack_entry base_ent;
735
736         if (left < 20)
737                 die("truncated pack file");
738
739         /* The base entry _must_ be in the same pack */
740         if (!find_pack_entry_one(base_sha1, &base_ent, p))
741                 die("failed to find delta-pack base object %s",
742                     sha1_to_hex(base_sha1));
743
744         /* We choose to only get the type of the base object and
745          * ignore potentially corrupt pack file that expects the delta
746          * based on a base with a wrong size.  This saves tons of
747          * inflate() calls.
748          */
749
750         if (packed_object_info(&base_ent, type, NULL))
751                 die("cannot get info for delta-pack base");
752
753         if (sizep) {
754                 const unsigned char *data;
755                 unsigned char delta_head[64];
756                 unsigned long result_size;
757                 z_stream stream;
758                 int st;
759
760                 memset(&stream, 0, sizeof(stream));
761
762                 data = stream.next_in = base_sha1 + 20;
763                 stream.avail_in = left - 20;
764                 stream.next_out = delta_head;
765                 stream.avail_out = sizeof(delta_head);
766
767                 inflateInit(&stream);
768                 st = inflate(&stream, Z_FINISH);
769                 inflateEnd(&stream);
770                 if ((st != Z_STREAM_END) &&
771                     stream.total_out != sizeof(delta_head))
772                         die("delta data unpack-initial failed");
773
774                 /* Examine the initial part of the delta to figure out
775                  * the result size.
776                  */
777                 data = delta_head;
778                 get_delta_hdr_size(&data); /* ignore base size */
779
780                 /* Read the result size */
781                 result_size = get_delta_hdr_size(&data);
782                 *sizep = result_size;
783         }
784         return 0;
785 }
786
787 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
788         enum object_type *type, unsigned long *sizep)
789 {
790         unsigned shift;
791         unsigned char *pack, c;
792         unsigned long size;
793
794         if (offset >= p->pack_size)
795                 die("object offset outside of pack file");
796
797         pack =  p->pack_base + offset;
798         c = *pack++;
799         offset++;
800         *type = (c >> 4) & 7;
801         size = c & 15;
802         shift = 4;
803         while (c & 0x80) {
804                 if (offset >= p->pack_size)
805                         die("object offset outside of pack file");
806                 c = *pack++;
807                 offset++;
808                 size += (c & 0x7f) << shift;
809                 shift += 7;
810         }
811         *sizep = size;
812         return offset;
813 }
814
815 void packed_object_info_detail(struct pack_entry *e,
816                                char *type,
817                                unsigned long *size,
818                                unsigned long *store_size,
819                                int *delta_chain_length,
820                                unsigned char *base_sha1)
821 {
822         struct packed_git *p = e->p;
823         unsigned long offset, left;
824         unsigned char *pack;
825         enum object_type kind;
826
827         offset = unpack_object_header(p, e->offset, &kind, size);
828         pack = p->pack_base + offset;
829         left = p->pack_size - offset;
830         if (kind != OBJ_DELTA)
831                 *delta_chain_length = 0;
832         else {
833                 int chain_length = 0;
834                 memcpy(base_sha1, pack, 20);
835                 do {
836                         struct pack_entry base_ent;
837                         unsigned long junk;
838
839                         find_pack_entry_one(pack, &base_ent, p);
840                         offset = unpack_object_header(p, base_ent.offset,
841                                                       &kind, &junk);
842                         pack = p->pack_base + offset;
843                         chain_length++;
844                 } while (kind == OBJ_DELTA);
845                 *delta_chain_length = chain_length;
846         }
847         switch (kind) {
848         case OBJ_COMMIT:
849                 strcpy(type, "commit");
850                 break;
851         case OBJ_TREE:
852                 strcpy(type, "tree");
853                 break;
854         case OBJ_BLOB:
855                 strcpy(type, "blob");
856                 break;
857         case OBJ_TAG:
858                 strcpy(type, "tag");
859                 break;
860         default:
861                 die("corrupted pack file");
862         }
863         *store_size = 0; /* notyet */
864 }
865
866 static int packed_object_info(struct pack_entry *entry,
867                               char *type, unsigned long *sizep)
868 {
869         struct packed_git *p = entry->p;
870         unsigned long offset, size, left;
871         unsigned char *pack;
872         enum object_type kind;
873         int retval;
874
875         if (use_packed_git(p))
876                 die("cannot map packed file");
877
878         offset = unpack_object_header(p, entry->offset, &kind, &size);
879         pack = p->pack_base + offset;
880         left = p->pack_size - offset;
881
882         switch (kind) {
883         case OBJ_DELTA:
884                 retval = packed_delta_info(pack, size, left, type, sizep, p);
885                 unuse_packed_git(p);
886                 return retval;
887         case OBJ_COMMIT:
888                 strcpy(type, "commit");
889                 break;
890         case OBJ_TREE:
891                 strcpy(type, "tree");
892                 break;
893         case OBJ_BLOB:
894                 strcpy(type, "blob");
895                 break;
896         case OBJ_TAG:
897                 strcpy(type, "tag");
898                 break;
899         default:
900                 die("corrupted pack file");
901         }
902         if (sizep)
903                 *sizep = size;
904         unuse_packed_git(p);
905         return 0;
906 }
907
908 /* forward declaration for a mutually recursive function */
909 static void *unpack_entry(struct pack_entry *, char *, unsigned long *);
910
911 static void *unpack_delta_entry(unsigned char *base_sha1,
912                                 unsigned long delta_size,
913                                 unsigned long left,
914                                 char *type,
915                                 unsigned long *sizep,
916                                 struct packed_git *p)
917 {
918         struct pack_entry base_ent;
919         void *data, *delta_data, *result, *base;
920         unsigned long data_size, result_size, base_size;
921         z_stream stream;
922         int st;
923
924         if (left < 20)
925                 die("truncated pack file");
926         data = base_sha1 + 20;
927         data_size = left - 20;
928         delta_data = xmalloc(delta_size);
929
930         memset(&stream, 0, sizeof(stream));
931
932         stream.next_in = data;
933         stream.avail_in = data_size;
934         stream.next_out = delta_data;
935         stream.avail_out = delta_size;
936
937         inflateInit(&stream);
938         st = inflate(&stream, Z_FINISH);
939         inflateEnd(&stream);
940         if ((st != Z_STREAM_END) || stream.total_out != delta_size)
941                 die("delta data unpack failed");
942
943         /* The base entry _must_ be in the same pack */
944         if (!find_pack_entry_one(base_sha1, &base_ent, p))
945                 die("failed to find delta-pack base object %s",
946                     sha1_to_hex(base_sha1));
947         base = unpack_entry_gently(&base_ent, type, &base_size);
948         if (!base)
949                 die("failed to read delta-pack base object %s",
950                     sha1_to_hex(base_sha1));
951         result = patch_delta(base, base_size,
952                              delta_data, delta_size,
953                              &result_size);
954         if (!result)
955                 die("failed to apply delta");
956         free(delta_data);
957         free(base);
958         *sizep = result_size;
959         return result;
960 }
961
962 static void *unpack_non_delta_entry(unsigned char *data,
963                                     unsigned long size,
964                                     unsigned long left)
965 {
966         int st;
967         z_stream stream;
968         unsigned char *buffer;
969
970         buffer = xmalloc(size + 1);
971         buffer[size] = 0;
972         memset(&stream, 0, sizeof(stream));
973         stream.next_in = data;
974         stream.avail_in = left;
975         stream.next_out = buffer;
976         stream.avail_out = size;
977
978         inflateInit(&stream);
979         st = inflate(&stream, Z_FINISH);
980         inflateEnd(&stream);
981         if ((st != Z_STREAM_END) || stream.total_out != size) {
982                 free(buffer);
983                 return NULL;
984         }
985
986         return buffer;
987 }
988
989 static void *unpack_entry(struct pack_entry *entry,
990                           char *type, unsigned long *sizep)
991 {
992         struct packed_git *p = entry->p;
993         void *retval;
994
995         if (use_packed_git(p))
996                 die("cannot map packed file");
997         retval = unpack_entry_gently(entry, type, sizep);
998         unuse_packed_git(p);
999         if (!retval)
1000                 die("corrupted pack file");
1001         return retval;
1002 }
1003
1004 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1005 void *unpack_entry_gently(struct pack_entry *entry,
1006                           char *type, unsigned long *sizep)
1007 {
1008         struct packed_git *p = entry->p;
1009         unsigned long offset, size, left;
1010         unsigned char *pack;
1011         enum object_type kind;
1012         void *retval;
1013
1014         offset = unpack_object_header(p, entry->offset, &kind, &size);
1015         pack = p->pack_base + offset;
1016         left = p->pack_size - offset;
1017         switch (kind) {
1018         case OBJ_DELTA:
1019                 retval = unpack_delta_entry(pack, size, left, type, sizep, p);
1020                 return retval;
1021         case OBJ_COMMIT:
1022                 strcpy(type, "commit");
1023                 break;
1024         case OBJ_TREE:
1025                 strcpy(type, "tree");
1026                 break;
1027         case OBJ_BLOB:
1028                 strcpy(type, "blob");
1029                 break;
1030         case OBJ_TAG:
1031                 strcpy(type, "tag");
1032                 break;
1033         default:
1034                 return NULL;
1035         }
1036         *sizep = size;
1037         retval = unpack_non_delta_entry(pack, size, left);
1038         return retval;
1039 }
1040
1041 int num_packed_objects(const struct packed_git *p)
1042 {
1043         /* See check_packed_git_idx() */
1044         return (p->index_size - 20 - 20 - 4*256) / 24;
1045 }
1046
1047 int nth_packed_object_sha1(const struct packed_git *p, int n,
1048                            unsigned char* sha1)
1049 {
1050         void *index = p->index_base + 256;
1051         if (n < 0 || num_packed_objects(p) <= n)
1052                 return -1;
1053         memcpy(sha1, (index + 24 * n + 4), 20);
1054         return 0;
1055 }
1056
1057 int find_pack_entry_one(const unsigned char *sha1,
1058                         struct pack_entry *e, struct packed_git *p)
1059 {
1060         unsigned int *level1_ofs = p->index_base;
1061         int hi = ntohl(level1_ofs[*sha1]);
1062         int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1063         void *index = p->index_base + 256;
1064
1065         do {
1066                 int mi = (lo + hi) / 2;
1067                 int cmp = memcmp(index + 24 * mi + 4, sha1, 20);
1068                 if (!cmp) {
1069                         e->offset = ntohl(*((int*)(index + 24 * mi)));
1070                         memcpy(e->sha1, sha1, 20);
1071                         e->p = p;
1072                         return 1;
1073                 }
1074                 if (cmp > 0)
1075                         hi = mi;
1076                 else
1077                         lo = mi+1;
1078         } while (lo < hi);
1079         return 0;
1080 }
1081
1082 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1083 {
1084         struct packed_git *p;
1085         prepare_packed_git();
1086
1087         for (p = packed_git; p; p = p->next) {
1088                 if (find_pack_entry_one(sha1, e, p))
1089                         return 1;
1090         }
1091         return 0;
1092 }
1093
1094 struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1095                                   struct packed_git *packs)
1096 {
1097         struct packed_git *p;
1098         struct pack_entry e;
1099
1100         for (p = packs; p; p = p->next) {
1101                 if (find_pack_entry_one(sha1, &e, p))
1102                         return p;
1103         }
1104         return NULL;
1105         
1106 }
1107
1108 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1109 {
1110         int status;
1111         unsigned long mapsize, size;
1112         void *map;
1113         z_stream stream;
1114         char hdr[128];
1115
1116         map = map_sha1_file_internal(sha1, &mapsize);
1117         if (!map) {
1118                 struct pack_entry e;
1119
1120                 if (!find_pack_entry(sha1, &e))
1121                         return error("unable to find %s", sha1_to_hex(sha1));
1122                 return packed_object_info(&e, type, sizep);
1123         }
1124         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1125                 status = error("unable to unpack %s header",
1126                                sha1_to_hex(sha1));
1127         if (parse_sha1_header(hdr, type, &size) < 0)
1128                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1129         else {
1130                 status = 0;
1131                 if (sizep)
1132                         *sizep = size;
1133         }
1134         inflateEnd(&stream);
1135         munmap(map, mapsize);
1136         return status;
1137 }
1138
1139 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1140 {
1141         struct pack_entry e;
1142
1143         if (!find_pack_entry(sha1, &e)) {
1144                 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1145                 return NULL;
1146         }
1147         return unpack_entry(&e, type, size);
1148 }
1149
1150 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1151 {
1152         unsigned long mapsize;
1153         void *map, *buf;
1154         struct pack_entry e;
1155
1156         if (find_pack_entry(sha1, &e))
1157                 return read_packed_sha1(sha1, type, size);
1158         map = map_sha1_file_internal(sha1, &mapsize);
1159         if (map) {
1160                 buf = unpack_sha1_file(map, mapsize, type, size);
1161                 munmap(map, mapsize);
1162                 return buf;
1163         }
1164         return NULL;
1165 }
1166
1167 void *read_object_with_reference(const unsigned char *sha1,
1168                                  const char *required_type,
1169                                  unsigned long *size,
1170                                  unsigned char *actual_sha1_return)
1171 {
1172         char type[20];
1173         void *buffer;
1174         unsigned long isize;
1175         unsigned char actual_sha1[20];
1176
1177         memcpy(actual_sha1, sha1, 20);
1178         while (1) {
1179                 int ref_length = -1;
1180                 const char *ref_type = NULL;
1181
1182                 buffer = read_sha1_file(actual_sha1, type, &isize);
1183                 if (!buffer)
1184                         return NULL;
1185                 if (!strcmp(type, required_type)) {
1186                         *size = isize;
1187                         if (actual_sha1_return)
1188                                 memcpy(actual_sha1_return, actual_sha1, 20);
1189                         return buffer;
1190                 }
1191                 /* Handle references */
1192                 else if (!strcmp(type, "commit"))
1193                         ref_type = "tree ";
1194                 else if (!strcmp(type, "tag"))
1195                         ref_type = "object ";
1196                 else {
1197                         free(buffer);
1198                         return NULL;
1199                 }
1200                 ref_length = strlen(ref_type);
1201
1202                 if (memcmp(buffer, ref_type, ref_length) ||
1203                     get_sha1_hex(buffer + ref_length, actual_sha1)) {
1204                         free(buffer);
1205                         return NULL;
1206                 }
1207                 /* Now we have the ID of the referred-to object in
1208                  * actual_sha1.  Check again. */
1209         }
1210 }
1211
1212 char *write_sha1_file_prepare(void *buf,
1213                               unsigned long len,
1214                               const char *type,
1215                               unsigned char *sha1,
1216                               unsigned char *hdr,
1217                               int *hdrlen)
1218 {
1219         SHA_CTX c;
1220
1221         /* Generate the header */
1222         *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1223
1224         /* Sha1.. */
1225         SHA1_Init(&c);
1226         SHA1_Update(&c, hdr, *hdrlen);
1227         SHA1_Update(&c, buf, len);
1228         SHA1_Final(sha1, &c);
1229
1230         return sha1_file_name(sha1);
1231 }
1232
1233 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1234 {
1235         int size;
1236         unsigned char *compressed;
1237         z_stream stream;
1238         unsigned char sha1[20];
1239         char *filename;
1240         static char tmpfile[PATH_MAX];
1241         unsigned char hdr[50];
1242         int fd, hdrlen, ret;
1243
1244         /* Normally if we have it in the pack then we do not bother writing
1245          * it out into .git/objects/??/?{38} file.
1246          */
1247         filename = write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1248         if (returnsha1)
1249                 memcpy(returnsha1, sha1, 20);
1250         if (has_sha1_file(sha1))
1251                 return 0;
1252         fd = open(filename, O_RDONLY);
1253         if (fd >= 0) {
1254                 /*
1255                  * FIXME!!! We might do collision checking here, but we'd
1256                  * need to uncompress the old file and check it. Later.
1257                  */
1258                 close(fd);
1259                 return 0;
1260         }
1261
1262         if (errno != ENOENT) {
1263                 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
1264                 return -1;
1265         }
1266
1267         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1268
1269         fd = mkstemp(tmpfile);
1270         if (fd < 0) {
1271                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
1272                 return -1;
1273         }
1274
1275         /* Set it up */
1276         memset(&stream, 0, sizeof(stream));
1277         deflateInit(&stream, Z_BEST_COMPRESSION);
1278         size = deflateBound(&stream, len+hdrlen);
1279         compressed = xmalloc(size);
1280
1281         /* Compress it */
1282         stream.next_out = compressed;
1283         stream.avail_out = size;
1284
1285         /* First header.. */
1286         stream.next_in = hdr;
1287         stream.avail_in = hdrlen;
1288         while (deflate(&stream, 0) == Z_OK)
1289                 /* nothing */;
1290
1291         /* Then the data itself.. */
1292         stream.next_in = buf;
1293         stream.avail_in = len;
1294         while (deflate(&stream, Z_FINISH) == Z_OK)
1295                 /* nothing */;
1296         deflateEnd(&stream);
1297         size = stream.total_out;
1298
1299         if (write(fd, compressed, size) != size)
1300                 die("unable to write file");
1301         fchmod(fd, 0444);
1302         close(fd);
1303         free(compressed);
1304
1305         ret = link(tmpfile, filename);
1306         if (ret < 0) {
1307                 ret = errno;
1308
1309                 /*
1310                  * Coda hack - coda doesn't like cross-directory links,
1311                  * so we fall back to a rename, which will mean that it
1312                  * won't be able to check collisions, but that's not a
1313                  * big deal.
1314                  *
1315                  * When this succeeds, we just return 0. We have nothing
1316                  * left to unlink.
1317                  */
1318                 if (ret == EXDEV && !rename(tmpfile, filename))
1319                         return 0;
1320         }
1321         unlink(tmpfile);
1322         if (ret) {
1323                 if (ret != EEXIST) {
1324                         fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
1325                         return -1;
1326                 }
1327                 /* FIXME!!! Collision check here ? */
1328         }
1329
1330         return 0;
1331 }
1332
1333 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1334 {
1335         ssize_t size;
1336         unsigned long objsize;
1337         int posn = 0;
1338         void *buf = map_sha1_file_internal(sha1, &objsize);
1339         z_stream stream;
1340         if (!buf) {
1341                 unsigned char *unpacked;
1342                 unsigned long len;
1343                 char type[20];
1344                 char hdr[50];
1345                 int hdrlen;
1346                 // need to unpack and recompress it by itself
1347                 unpacked = read_packed_sha1(sha1, type, &len);
1348
1349                 hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1350
1351                 /* Set it up */
1352                 memset(&stream, 0, sizeof(stream));
1353                 deflateInit(&stream, Z_BEST_COMPRESSION);
1354                 size = deflateBound(&stream, len + hdrlen);
1355                 buf = xmalloc(size);
1356
1357                 /* Compress it */
1358                 stream.next_out = buf;
1359                 stream.avail_out = size;
1360                 
1361                 /* First header.. */
1362                 stream.next_in = (void *)hdr;
1363                 stream.avail_in = hdrlen;
1364                 while (deflate(&stream, 0) == Z_OK)
1365                         /* nothing */;
1366
1367                 /* Then the data itself.. */
1368                 stream.next_in = unpacked;
1369                 stream.avail_in = len;
1370                 while (deflate(&stream, Z_FINISH) == Z_OK)
1371                         /* nothing */;
1372                 deflateEnd(&stream);
1373                 
1374                 objsize = stream.total_out;
1375         }
1376
1377         do {
1378                 size = write(fd, buf + posn, objsize - posn);
1379                 if (size <= 0) {
1380                         if (!size) {
1381                                 fprintf(stderr, "write closed");
1382                         } else {
1383                                 perror("write ");
1384                         }
1385                         return -1;
1386                 }
1387                 posn += size;
1388         } while (posn < objsize);
1389         return 0;
1390 }
1391
1392 int write_sha1_from_fd(const unsigned char *sha1, int fd)
1393 {
1394         char *filename = sha1_file_name(sha1);
1395
1396         int local;
1397         z_stream stream;
1398         unsigned char real_sha1[20];
1399         unsigned char buf[4096];
1400         unsigned char discard[4096];
1401         int ret;
1402         SHA_CTX c;
1403
1404         local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
1405
1406         if (local < 0)
1407                 return error("Couldn't open %s\n", filename);
1408
1409         memset(&stream, 0, sizeof(stream));
1410
1411         inflateInit(&stream);
1412
1413         SHA1_Init(&c);
1414
1415         do {
1416                 ssize_t size;
1417                 size = read(fd, buf, 4096);
1418                 if (size <= 0) {
1419                         close(local);
1420                         unlink(filename);
1421                         if (!size)
1422                                 return error("Connection closed?");
1423                         perror("Reading from connection");
1424                         return -1;
1425                 }
1426                 write(local, buf, size);
1427                 stream.avail_in = size;
1428                 stream.next_in = buf;
1429                 do {
1430                         stream.next_out = discard;
1431                         stream.avail_out = sizeof(discard);
1432                         ret = inflate(&stream, Z_SYNC_FLUSH);
1433                         SHA1_Update(&c, discard, sizeof(discard) -
1434                                     stream.avail_out);
1435                 } while (stream.avail_in && ret == Z_OK);
1436                 
1437         } while (ret == Z_OK);
1438         inflateEnd(&stream);
1439
1440         close(local);
1441         SHA1_Final(real_sha1, &c);
1442         if (ret != Z_STREAM_END) {
1443                 unlink(filename);
1444                 return error("File %s corrupted", sha1_to_hex(sha1));
1445         }
1446         if (memcmp(sha1, real_sha1, 20)) {
1447                 unlink(filename);
1448                 return error("File %s has bad hash\n", sha1_to_hex(sha1));
1449         }
1450         
1451         return 0;
1452 }
1453
1454 int has_pack_index(const unsigned char *sha1)
1455 {
1456         struct stat st;
1457         if (stat(sha1_pack_index_name(sha1), &st))
1458                 return 0;
1459         return 1;
1460 }
1461
1462 int has_pack_file(const unsigned char *sha1)
1463 {
1464         struct stat st;
1465         if (stat(sha1_pack_name(sha1), &st))
1466                 return 0;
1467         return 1;
1468 }
1469
1470 int has_sha1_pack(const unsigned char *sha1)
1471 {
1472         struct pack_entry e;
1473         return find_pack_entry(sha1, &e);
1474 }
1475
1476 int has_sha1_file(const unsigned char *sha1)
1477 {
1478         struct stat st;
1479         struct pack_entry e;
1480
1481         if (find_pack_entry(sha1, &e))
1482                 return 1;
1483         return find_sha1_file(sha1, &st) ? 1 : 0;
1484 }
1485
1486 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1487 {
1488         unsigned long size = st->st_size;
1489         void *buf;
1490         int ret;
1491         unsigned char hdr[50];
1492         int hdrlen;
1493
1494         buf = "";
1495         if (size)
1496                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1497         close(fd);
1498         if (buf == MAP_FAILED)
1499                 return -1;
1500
1501         if (!type)
1502                 type = "blob";
1503         if (write_object)
1504                 ret = write_sha1_file(buf, size, type, sha1);
1505         else {
1506                 write_sha1_file_prepare(buf, size, type, sha1, hdr, &hdrlen);
1507                 ret = 0;
1508         }
1509         if (size)
1510                 munmap(buf, size);
1511         return ret;
1512 }