Reject hexstring longer than 40-bytes in get_short_sha1()
[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 "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
16
17 #ifndef O_NOATIME
18 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
19 #define O_NOATIME 01000000
20 #else
21 #define O_NOATIME 0
22 #endif
23 #endif
24
25 const unsigned char null_sha1[20];
26
27 static unsigned int sha1_file_open_flag = O_NOATIME;
28
29 signed char hexval_table[256] = {
30          -1, -1, -1, -1, -1, -1, -1, -1,                /* 00-07 */
31          -1, -1, -1, -1, -1, -1, -1, -1,                /* 08-0f */
32          -1, -1, -1, -1, -1, -1, -1, -1,                /* 10-17 */
33          -1, -1, -1, -1, -1, -1, -1, -1,                /* 18-1f */
34          -1, -1, -1, -1, -1, -1, -1, -1,                /* 20-27 */
35          -1, -1, -1, -1, -1, -1, -1, -1,                /* 28-2f */
36           0,  1,  2,  3,  4,  5,  6,  7,                /* 30-37 */
37           8,  9, -1, -1, -1, -1, -1, -1,                /* 38-3f */
38          -1, 10, 11, 12, 13, 14, 15, -1,                /* 40-47 */
39          -1, -1, -1, -1, -1, -1, -1, -1,                /* 48-4f */
40          -1, -1, -1, -1, -1, -1, -1, -1,                /* 50-57 */
41          -1, -1, -1, -1, -1, -1, -1, -1,                /* 58-5f */
42          -1, 10, 11, 12, 13, 14, 15, -1,                /* 60-67 */
43          -1, -1, -1, -1, -1, -1, -1, -1,                /* 68-67 */
44          -1, -1, -1, -1, -1, -1, -1, -1,                /* 70-77 */
45          -1, -1, -1, -1, -1, -1, -1, -1,                /* 78-7f */
46          -1, -1, -1, -1, -1, -1, -1, -1,                /* 80-87 */
47          -1, -1, -1, -1, -1, -1, -1, -1,                /* 88-8f */
48          -1, -1, -1, -1, -1, -1, -1, -1,                /* 90-97 */
49          -1, -1, -1, -1, -1, -1, -1, -1,                /* 98-9f */
50          -1, -1, -1, -1, -1, -1, -1, -1,                /* a0-a7 */
51          -1, -1, -1, -1, -1, -1, -1, -1,                /* a8-af */
52          -1, -1, -1, -1, -1, -1, -1, -1,                /* b0-b7 */
53          -1, -1, -1, -1, -1, -1, -1, -1,                /* b8-bf */
54          -1, -1, -1, -1, -1, -1, -1, -1,                /* c0-c7 */
55          -1, -1, -1, -1, -1, -1, -1, -1,                /* c8-cf */
56          -1, -1, -1, -1, -1, -1, -1, -1,                /* d0-d7 */
57          -1, -1, -1, -1, -1, -1, -1, -1,                /* d8-df */
58          -1, -1, -1, -1, -1, -1, -1, -1,                /* e0-e7 */
59          -1, -1, -1, -1, -1, -1, -1, -1,                /* e8-ef */
60          -1, -1, -1, -1, -1, -1, -1, -1,                /* f0-f7 */
61          -1, -1, -1, -1, -1, -1, -1, -1,                /* f8-ff */
62 };
63
64 int get_sha1_hex(const char *hex, unsigned char *sha1)
65 {
66         int i;
67         for (i = 0; i < 20; i++) {
68                 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
69                 if (val & ~0xff)
70                         return -1;
71                 *sha1++ = val;
72                 hex += 2;
73         }
74         return 0;
75 }
76
77 int safe_create_leading_directories(char *path)
78 {
79         char *pos = path;
80         struct stat st;
81
82         if (*pos == '/')
83                 pos++;
84
85         while (pos) {
86                 pos = strchr(pos, '/');
87                 if (!pos)
88                         break;
89                 *pos = 0;
90                 if (!stat(path, &st)) {
91                         /* path exists */
92                         if (!S_ISDIR(st.st_mode)) {
93                                 *pos = '/';
94                                 return -3;
95                         }
96                 }
97                 else if (mkdir(path, 0777)) {
98                         *pos = '/';
99                         return -1;
100                 }
101                 else if (adjust_shared_perm(path)) {
102                         *pos = '/';
103                         return -2;
104                 }
105                 *pos++ = '/';
106         }
107         return 0;
108 }
109
110 char * sha1_to_hex(const unsigned char *sha1)
111 {
112         static int bufno;
113         static char hexbuffer[4][50];
114         static const char hex[] = "0123456789abcdef";
115         char *buffer = hexbuffer[3 & ++bufno], *buf = buffer;
116         int i;
117
118         for (i = 0; i < 20; i++) {
119                 unsigned int val = *sha1++;
120                 *buf++ = hex[val >> 4];
121                 *buf++ = hex[val & 0xf];
122         }
123         *buf = '\0';
124
125         return buffer;
126 }
127
128 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
129 {
130         int i;
131         for (i = 0; i < 20; i++) {
132                 static char hex[] = "0123456789abcdef";
133                 unsigned int val = sha1[i];
134                 char *pos = pathbuf + i*2 + (i > 0);
135                 *pos++ = hex[val >> 4];
136                 *pos = hex[val & 0xf];
137         }
138 }
139
140 /*
141  * NOTE! This returns a statically allocated buffer, so you have to be
142  * careful about using it. Do a "xstrdup()" if you need to save the
143  * filename.
144  *
145  * Also note that this returns the location for creating.  Reading
146  * SHA1 file can happen from any alternate directory listed in the
147  * DB_ENVIRONMENT environment variable if it is not found in
148  * the primary object database.
149  */
150 char *sha1_file_name(const unsigned char *sha1)
151 {
152         static char *name, *base;
153
154         if (!base) {
155                 const char *sha1_file_directory = get_object_directory();
156                 int len = strlen(sha1_file_directory);
157                 base = xmalloc(len + 60);
158                 memcpy(base, sha1_file_directory, len);
159                 memset(base+len, 0, 60);
160                 base[len] = '/';
161                 base[len+3] = '/';
162                 name = base + len + 1;
163         }
164         fill_sha1_path(name, sha1);
165         return base;
166 }
167
168 char *sha1_pack_name(const unsigned char *sha1)
169 {
170         static const char hex[] = "0123456789abcdef";
171         static char *name, *base, *buf;
172         int i;
173
174         if (!base) {
175                 const char *sha1_file_directory = get_object_directory();
176                 int len = strlen(sha1_file_directory);
177                 base = xmalloc(len + 60);
178                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.pack", sha1_file_directory);
179                 name = base + len + 11;
180         }
181
182         buf = name;
183
184         for (i = 0; i < 20; i++) {
185                 unsigned int val = *sha1++;
186                 *buf++ = hex[val >> 4];
187                 *buf++ = hex[val & 0xf];
188         }
189         
190         return base;
191 }
192
193 char *sha1_pack_index_name(const unsigned char *sha1)
194 {
195         static const char hex[] = "0123456789abcdef";
196         static char *name, *base, *buf;
197         int i;
198
199         if (!base) {
200                 const char *sha1_file_directory = get_object_directory();
201                 int len = strlen(sha1_file_directory);
202                 base = xmalloc(len + 60);
203                 sprintf(base, "%s/pack/pack-1234567890123456789012345678901234567890.idx", sha1_file_directory);
204                 name = base + len + 11;
205         }
206
207         buf = name;
208
209         for (i = 0; i < 20; i++) {
210                 unsigned int val = *sha1++;
211                 *buf++ = hex[val >> 4];
212                 *buf++ = hex[val & 0xf];
213         }
214         
215         return base;
216 }
217
218 struct alternate_object_database *alt_odb_list;
219 static struct alternate_object_database **alt_odb_tail;
220
221 static void read_info_alternates(const char * alternates, int depth);
222
223 /*
224  * Prepare alternate object database registry.
225  *
226  * The variable alt_odb_list points at the list of struct
227  * alternate_object_database.  The elements on this list come from
228  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
229  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
230  * whose contents is similar to that environment variable but can be
231  * LF separated.  Its base points at a statically allocated buffer that
232  * contains "/the/directory/corresponding/to/.git/objects/...", while
233  * its name points just after the slash at the end of ".git/objects/"
234  * in the example above, and has enough space to hold 40-byte hex
235  * SHA1, an extra slash for the first level indirection, and the
236  * terminating NUL.
237  */
238 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
239 {
240         struct stat st;
241         const char *objdir = get_object_directory();
242         struct alternate_object_database *ent;
243         struct alternate_object_database *alt;
244         /* 43 = 40-byte + 2 '/' + terminating NUL */
245         int pfxlen = len;
246         int entlen = pfxlen + 43;
247         int base_len = -1;
248
249         if (*entry != '/' && relative_base) {
250                 /* Relative alt-odb */
251                 if (base_len < 0)
252                         base_len = strlen(relative_base) + 1;
253                 entlen += base_len;
254                 pfxlen += base_len;
255         }
256         ent = xmalloc(sizeof(*ent) + entlen);
257
258         if (*entry != '/' && relative_base) {
259                 memcpy(ent->base, relative_base, base_len - 1);
260                 ent->base[base_len - 1] = '/';
261                 memcpy(ent->base + base_len, entry, len);
262         }
263         else
264                 memcpy(ent->base, entry, pfxlen);
265
266         ent->name = ent->base + pfxlen + 1;
267         ent->base[pfxlen + 3] = '/';
268         ent->base[pfxlen] = ent->base[entlen-1] = 0;
269
270         /* Detect cases where alternate disappeared */
271         if (stat(ent->base, &st) || !S_ISDIR(st.st_mode)) {
272                 error("object directory %s does not exist; "
273                       "check .git/objects/info/alternates.",
274                       ent->base);
275                 free(ent);
276                 return -1;
277         }
278
279         /* Prevent the common mistake of listing the same
280          * thing twice, or object directory itself.
281          */
282         for (alt = alt_odb_list; alt; alt = alt->next) {
283                 if (!memcmp(ent->base, alt->base, pfxlen)) {
284                         free(ent);
285                         return -1;
286                 }
287         }
288         if (!memcmp(ent->base, objdir, pfxlen)) {
289                 free(ent);
290                 return -1;
291         }
292
293         /* add the alternate entry */
294         *alt_odb_tail = ent;
295         alt_odb_tail = &(ent->next);
296         ent->next = NULL;
297
298         /* recursively add alternates */
299         read_info_alternates(ent->base, depth + 1);
300
301         ent->base[pfxlen] = '/';
302
303         return 0;
304 }
305
306 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
307                                  const char *relative_base, int depth)
308 {
309         const char *cp, *last;
310
311         if (depth > 5) {
312                 error("%s: ignoring alternate object stores, nesting too deep.",
313                                 relative_base);
314                 return;
315         }
316
317         last = alt;
318         while (last < ep) {
319                 cp = last;
320                 if (cp < ep && *cp == '#') {
321                         while (cp < ep && *cp != sep)
322                                 cp++;
323                         last = cp + 1;
324                         continue;
325                 }
326                 while (cp < ep && *cp != sep)
327                         cp++;
328                 if (last != cp) {
329                         if ((*last != '/') && depth) {
330                                 error("%s: ignoring relative alternate object store %s",
331                                                 relative_base, last);
332                         } else {
333                                 link_alt_odb_entry(last, cp - last,
334                                                 relative_base, depth);
335                         }
336                 }
337                 while (cp < ep && *cp == sep)
338                         cp++;
339                 last = cp;
340         }
341 }
342
343 static void read_info_alternates(const char * relative_base, int depth)
344 {
345         char *map;
346         struct stat st;
347         char path[PATH_MAX];
348         int fd;
349
350         sprintf(path, "%s/info/alternates", relative_base);
351         fd = open(path, O_RDONLY);
352         if (fd < 0)
353                 return;
354         if (fstat(fd, &st) || (st.st_size == 0)) {
355                 close(fd);
356                 return;
357         }
358         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
359         close(fd);
360         if (map == MAP_FAILED)
361                 return;
362
363         link_alt_odb_entries(map, map + st.st_size, '\n', relative_base, depth);
364
365         munmap(map, st.st_size);
366 }
367
368 void prepare_alt_odb(void)
369 {
370         const char *alt;
371
372         alt = getenv(ALTERNATE_DB_ENVIRONMENT);
373         if (!alt) alt = "";
374
375         if (alt_odb_tail)
376                 return;
377         alt_odb_tail = &alt_odb_list;
378         link_alt_odb_entries(alt, alt + strlen(alt), ':', NULL, 0);
379
380         read_info_alternates(get_object_directory(), 0);
381 }
382
383 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
384 {
385         char *name = sha1_file_name(sha1);
386         struct alternate_object_database *alt;
387
388         if (!stat(name, st))
389                 return name;
390         prepare_alt_odb();
391         for (alt = alt_odb_list; alt; alt = alt->next) {
392                 name = alt->name;
393                 fill_sha1_path(name, sha1);
394                 if (!stat(alt->base, st))
395                         return alt->base;
396         }
397         return NULL;
398 }
399
400 #define PACK_MAX_SZ (1<<26)
401 static int pack_used_ctr;
402 static unsigned long pack_mapped;
403 struct packed_git *packed_git;
404
405 static int check_packed_git_idx(const char *path, unsigned long *idx_size_,
406                                 void **idx_map_)
407 {
408         void *idx_map;
409         unsigned int *index;
410         unsigned long idx_size;
411         int nr, i;
412         int fd = open(path, O_RDONLY);
413         struct stat st;
414         if (fd < 0)
415                 return -1;
416         if (fstat(fd, &st)) {
417                 close(fd);
418                 return -1;
419         }
420         idx_size = st.st_size;
421         idx_map = mmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
422         close(fd);
423         if (idx_map == MAP_FAILED)
424                 return -1;
425
426         index = idx_map;
427         *idx_map_ = idx_map;
428         *idx_size_ = idx_size;
429
430         /* check index map */
431         if (idx_size < 4*256 + 20 + 20)
432                 return error("index file too small");
433         nr = 0;
434         for (i = 0; i < 256; i++) {
435                 unsigned int n = ntohl(index[i]);
436                 if (n < nr)
437                         return error("non-monotonic index");
438                 nr = n;
439         }
440
441         /*
442          * Total size:
443          *  - 256 index entries 4 bytes each
444          *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
445          *  - 20-byte SHA1 of the packfile
446          *  - 20-byte SHA1 file checksum
447          */
448         if (idx_size != 4*256 + nr * 24 + 20 + 20)
449                 return error("wrong index file size");
450
451         return 0;
452 }
453
454 static int unuse_one_packed_git(void)
455 {
456         struct packed_git *p, *lru = NULL;
457
458         for (p = packed_git; p; p = p->next) {
459                 if (p->pack_use_cnt || !p->pack_base)
460                         continue;
461                 if (!lru || p->pack_last_used < lru->pack_last_used)
462                         lru = p;
463         }
464         if (!lru)
465                 return 0;
466         munmap(lru->pack_base, lru->pack_size);
467         lru->pack_base = NULL;
468         return 1;
469 }
470
471 void unuse_packed_git(struct packed_git *p)
472 {
473         p->pack_use_cnt--;
474 }
475
476 int use_packed_git(struct packed_git *p)
477 {
478         if (!p->pack_size) {
479                 struct stat st;
480                 /* We created the struct before we had the pack */
481                 stat(p->pack_name, &st);
482                 if (!S_ISREG(st.st_mode))
483                         die("packfile %s not a regular file", p->pack_name);
484                 p->pack_size = st.st_size;
485         }
486         if (!p->pack_base) {
487                 int fd;
488                 struct stat st;
489                 void *map;
490                 struct pack_header *hdr;
491
492                 pack_mapped += p->pack_size;
493                 while (PACK_MAX_SZ < pack_mapped && unuse_one_packed_git())
494                         ; /* nothing */
495                 fd = open(p->pack_name, O_RDONLY);
496                 if (fd < 0)
497                         die("packfile %s cannot be opened", p->pack_name);
498                 if (fstat(fd, &st)) {
499                         close(fd);
500                         die("packfile %s cannot be opened", p->pack_name);
501                 }
502                 if (st.st_size != p->pack_size)
503                         die("packfile %s size mismatch.", p->pack_name);
504                 map = mmap(NULL, p->pack_size, PROT_READ, MAP_PRIVATE, fd, 0);
505                 close(fd);
506                 if (map == MAP_FAILED)
507                         die("packfile %s cannot be mapped.", p->pack_name);
508                 p->pack_base = map;
509
510                 /* Check if we understand this pack file.  If we don't we're
511                  * likely too old to handle it.
512                  */
513                 hdr = map;
514                 if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
515                         die("packfile %s isn't actually a pack.", p->pack_name);
516                 if (!pack_version_ok(hdr->hdr_version))
517                         die("packfile %s is version %i and not supported"
518                                 " (try upgrading GIT to a newer version)",
519                                 p->pack_name, ntohl(hdr->hdr_version));
520
521                 /* Check if the pack file matches with the index file.
522                  * this is cheap.
523                  */
524                 if (hashcmp((unsigned char *)(p->index_base) +
525                             p->index_size - 40,
526                             (unsigned char *)p->pack_base +
527                             p->pack_size - 20)) {
528                         die("packfile %s does not match index.", p->pack_name);
529                 }
530         }
531         p->pack_last_used = pack_used_ctr++;
532         p->pack_use_cnt++;
533         return 0;
534 }
535
536 struct packed_git *add_packed_git(char *path, int path_len, int local)
537 {
538         struct stat st;
539         struct packed_git *p;
540         unsigned long idx_size;
541         void *idx_map;
542         unsigned char sha1[20];
543
544         if (check_packed_git_idx(path, &idx_size, &idx_map))
545                 return NULL;
546
547         /* do we have a corresponding .pack file? */
548         strcpy(path + path_len - 4, ".pack");
549         if (stat(path, &st) || !S_ISREG(st.st_mode)) {
550                 munmap(idx_map, idx_size);
551                 return NULL;
552         }
553         /* ok, it looks sane as far as we can check without
554          * actually mapping the pack file.
555          */
556         p = xmalloc(sizeof(*p) + path_len + 2);
557         strcpy(p->pack_name, path);
558         p->index_size = idx_size;
559         p->pack_size = st.st_size;
560         p->index_base = idx_map;
561         p->next = NULL;
562         p->pack_base = NULL;
563         p->pack_last_used = 0;
564         p->pack_use_cnt = 0;
565         p->pack_local = local;
566         if ((path_len > 44) && !get_sha1_hex(path + path_len - 44, sha1))
567                 hashcpy(p->sha1, sha1);
568         return p;
569 }
570
571 struct packed_git *parse_pack_index(unsigned char *sha1)
572 {
573         char *path = sha1_pack_index_name(sha1);
574         return parse_pack_index_file(sha1, path);
575 }
576
577 struct packed_git *parse_pack_index_file(const unsigned char *sha1, char *idx_path)
578 {
579         struct packed_git *p;
580         unsigned long idx_size;
581         void *idx_map;
582         char *path;
583
584         if (check_packed_git_idx(idx_path, &idx_size, &idx_map))
585                 return NULL;
586
587         path = sha1_pack_name(sha1);
588
589         p = xmalloc(sizeof(*p) + strlen(path) + 2);
590         strcpy(p->pack_name, path);
591         p->index_size = idx_size;
592         p->pack_size = 0;
593         p->index_base = idx_map;
594         p->next = NULL;
595         p->pack_base = NULL;
596         p->pack_last_used = 0;
597         p->pack_use_cnt = 0;
598         hashcpy(p->sha1, sha1);
599         return p;
600 }
601
602 void install_packed_git(struct packed_git *pack)
603 {
604         pack->next = packed_git;
605         packed_git = pack;
606 }
607
608 static void prepare_packed_git_one(char *objdir, int local)
609 {
610         char path[PATH_MAX];
611         int len;
612         DIR *dir;
613         struct dirent *de;
614
615         sprintf(path, "%s/pack", objdir);
616         len = strlen(path);
617         dir = opendir(path);
618         if (!dir) {
619                 if (errno != ENOENT)
620                         error("unable to open object pack directory: %s: %s",
621                               path, strerror(errno));
622                 return;
623         }
624         path[len++] = '/';
625         while ((de = readdir(dir)) != NULL) {
626                 int namelen = strlen(de->d_name);
627                 struct packed_git *p;
628
629                 if (!has_extension(de->d_name, ".idx"))
630                         continue;
631
632                 /* we have .idx.  Is it a file we can map? */
633                 strcpy(path + len, de->d_name);
634                 for (p = packed_git; p; p = p->next) {
635                         if (!memcmp(path, p->pack_name, len + namelen - 4))
636                                 break;
637                 }
638                 if (p)
639                         continue;
640                 p = add_packed_git(path, len + namelen, local);
641                 if (!p)
642                         continue;
643                 p->next = packed_git;
644                 packed_git = p;
645         }
646         closedir(dir);
647 }
648
649 static int prepare_packed_git_run_once = 0;
650 void prepare_packed_git(void)
651 {
652         struct alternate_object_database *alt;
653
654         if (prepare_packed_git_run_once)
655                 return;
656         prepare_packed_git_one(get_object_directory(), 1);
657         prepare_alt_odb();
658         for (alt = alt_odb_list; alt; alt = alt->next) {
659                 alt->name[-1] = 0;
660                 prepare_packed_git_one(alt->base, 0);
661                 alt->name[-1] = '/';
662         }
663         prepare_packed_git_run_once = 1;
664 }
665
666 static void reprepare_packed_git(void)
667 {
668         prepare_packed_git_run_once = 0;
669         prepare_packed_git();
670 }
671
672 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
673 {
674         unsigned char real_sha1[20];
675         hash_sha1_file(map, size, type, real_sha1);
676         return hashcmp(sha1, real_sha1) ? -1 : 0;
677 }
678
679 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
680 {
681         struct stat st;
682         void *map;
683         int fd;
684         char *filename = find_sha1_file(sha1, &st);
685
686         if (!filename) {
687                 return NULL;
688         }
689
690         fd = open(filename, O_RDONLY | sha1_file_open_flag);
691         if (fd < 0) {
692                 /* See if it works without O_NOATIME */
693                 switch (sha1_file_open_flag) {
694                 default:
695                         fd = open(filename, O_RDONLY);
696                         if (fd >= 0)
697                                 break;
698                 /* Fallthrough */
699                 case 0:
700                         return NULL;
701                 }
702
703                 /* If it failed once, it will probably fail again.
704                  * Stop using O_NOATIME
705                  */
706                 sha1_file_open_flag = 0;
707         }
708         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
709         close(fd);
710         if (map == MAP_FAILED)
711                 return NULL;
712         *size = st.st_size;
713         return map;
714 }
715
716 int legacy_loose_object(unsigned char *map)
717 {
718         unsigned int word;
719
720         /*
721          * Is it a zlib-compressed buffer? If so, the first byte
722          * must be 0x78 (15-bit window size, deflated), and the
723          * first 16-bit word is evenly divisible by 31
724          */
725         word = (map[0] << 8) + map[1];
726         if (map[0] == 0x78 && !(word % 31))
727                 return 1;
728         else
729                 return 0;
730 }
731
732 unsigned long unpack_object_header_gently(const unsigned char *buf, unsigned long len, enum object_type *type, unsigned long *sizep)
733 {
734         unsigned shift;
735         unsigned char c;
736         unsigned long size;
737         unsigned long used = 0;
738
739         c = buf[used++];
740         *type = (c >> 4) & 7;
741         size = c & 15;
742         shift = 4;
743         while (c & 0x80) {
744                 if (len <= used)
745                         return 0;
746                 if (sizeof(long) * 8 <= shift)
747                         return 0;
748                 c = buf[used++];
749                 size += (c & 0x7f) << shift;
750                 shift += 7;
751         }
752         *sizep = size;
753         return used;
754 }
755
756 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
757 {
758         unsigned long size, used;
759         static const char valid_loose_object_type[8] = {
760                 0, /* OBJ_EXT */
761                 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
762                 0, /* "delta" and others are invalid in a loose object */
763         };
764         enum object_type type;
765
766         /* Get the data stream */
767         memset(stream, 0, sizeof(*stream));
768         stream->next_in = map;
769         stream->avail_in = mapsize;
770         stream->next_out = buffer;
771         stream->avail_out = bufsiz;
772
773         if (legacy_loose_object(map)) {
774                 inflateInit(stream);
775                 return inflate(stream, 0);
776         }
777
778         used = unpack_object_header_gently(map, mapsize, &type, &size);
779         if (!used || !valid_loose_object_type[type])
780                 return -1;
781         map += used;
782         mapsize -= used;
783
784         /* Set up the stream for the rest.. */
785         stream->next_in = map;
786         stream->avail_in = mapsize;
787         inflateInit(stream);
788
789         /* And generate the fake traditional header */
790         stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
791                                          type_names[type], size);
792         return 0;
793 }
794
795 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
796 {
797         int bytes = strlen(buffer) + 1;
798         unsigned char *buf = xmalloc(1+size);
799         unsigned long n;
800
801         n = stream->total_out - bytes;
802         if (n > size)
803                 n = size;
804         memcpy(buf, (char *) buffer + bytes, n);
805         bytes = n;
806         if (bytes < size) {
807                 stream->next_out = buf + bytes;
808                 stream->avail_out = size - bytes;
809                 while (inflate(stream, Z_FINISH) == Z_OK)
810                         /* nothing */;
811         }
812         buf[size] = 0;
813         inflateEnd(stream);
814         return buf;
815 }
816
817 /*
818  * We used to just use "sscanf()", but that's actually way
819  * too permissive for what we want to check. So do an anal
820  * object header parse by hand.
821  */
822 static int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
823 {
824         int i;
825         unsigned long size;
826
827         /*
828          * The type can be at most ten bytes (including the 
829          * terminating '\0' that we add), and is followed by
830          * a space. 
831          */
832         i = 10;
833         for (;;) {
834                 char c = *hdr++;
835                 if (c == ' ')
836                         break;
837                 if (!--i)
838                         return -1;
839                 *type++ = c;
840         }
841         *type = 0;
842
843         /*
844          * The length must follow immediately, and be in canonical
845          * decimal format (ie "010" is not valid).
846          */
847         size = *hdr++ - '0';
848         if (size > 9)
849                 return -1;
850         if (size) {
851                 for (;;) {
852                         unsigned long c = *hdr - '0';
853                         if (c > 9)
854                                 break;
855                         hdr++;
856                         size = size * 10 + c;
857                 }
858         }
859         *sizep = size;
860
861         /*
862          * The length must be followed by a zero byte
863          */
864         return *hdr ? -1 : 0;
865 }
866
867 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
868 {
869         int ret;
870         z_stream stream;
871         char hdr[8192];
872
873         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
874         if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
875                 return NULL;
876
877         return unpack_sha1_rest(&stream, hdr, *size);
878 }
879
880 /* forward declaration for a mutually recursive function */
881 static int packed_object_info(struct packed_git *p, unsigned long offset,
882                               char *type, unsigned long *sizep);
883
884 static int packed_delta_info(struct packed_git *p,
885                              unsigned long offset,
886                              char *type,
887                              unsigned long *sizep)
888 {
889         unsigned long base_offset;
890         unsigned char *base_sha1 = (unsigned char *) p->pack_base + offset;
891
892         if (p->pack_size < offset + 20)
893                 die("truncated pack file");
894         /* The base entry _must_ be in the same pack */
895         base_offset = find_pack_entry_one(base_sha1, p);
896         if (!base_offset)
897                 die("failed to find delta-pack base object %s",
898                     sha1_to_hex(base_sha1));
899         offset += 20;
900
901         /* We choose to only get the type of the base object and
902          * ignore potentially corrupt pack file that expects the delta
903          * based on a base with a wrong size.  This saves tons of
904          * inflate() calls.
905          */
906         if (packed_object_info(p, base_offset, type, NULL))
907                 die("cannot get info for delta-pack base");
908
909         if (sizep) {
910                 const unsigned char *data;
911                 unsigned char delta_head[20];
912                 unsigned long result_size;
913                 z_stream stream;
914                 int st;
915
916                 memset(&stream, 0, sizeof(stream));
917
918                 stream.next_in = (unsigned char *) p->pack_base + offset;
919                 stream.avail_in = p->pack_size - offset;
920                 stream.next_out = delta_head;
921                 stream.avail_out = sizeof(delta_head);
922
923                 inflateInit(&stream);
924                 st = inflate(&stream, Z_FINISH);
925                 inflateEnd(&stream);
926                 if ((st != Z_STREAM_END) &&
927                     stream.total_out != sizeof(delta_head))
928                         die("delta data unpack-initial failed");
929
930                 /* Examine the initial part of the delta to figure out
931                  * the result size.
932                  */
933                 data = delta_head;
934
935                 /* ignore base size */
936                 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
937
938                 /* Read the result size */
939                 result_size = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
940                 *sizep = result_size;
941         }
942         return 0;
943 }
944
945 static unsigned long unpack_object_header(struct packed_git *p, unsigned long offset,
946         enum object_type *type, unsigned long *sizep)
947 {
948         unsigned long used;
949
950         if (p->pack_size <= offset)
951                 die("object offset outside of pack file");
952
953         used = unpack_object_header_gently((unsigned char *)p->pack_base +
954                                            offset,
955                                            p->pack_size - offset, type, sizep);
956         if (!used)
957                 die("object offset outside of pack file");
958
959         return offset + used;
960 }
961
962 int check_reuse_pack_delta(struct packed_git *p, unsigned long offset,
963                            unsigned char *base, unsigned long *sizep,
964                            enum object_type *kindp)
965 {
966         unsigned long ptr;
967         int status = -1;
968
969         use_packed_git(p);
970         ptr = offset;
971         ptr = unpack_object_header(p, ptr, kindp, sizep);
972         if (*kindp != OBJ_DELTA)
973                 goto done;
974         hashcpy(base, (unsigned char *) p->pack_base + ptr);
975         status = 0;
976  done:
977         unuse_packed_git(p);
978         return status;
979 }
980
981 void packed_object_info_detail(struct packed_git *p,
982                                unsigned long offset,
983                                char *type,
984                                unsigned long *size,
985                                unsigned long *store_size,
986                                unsigned int *delta_chain_length,
987                                unsigned char *base_sha1)
988 {
989         unsigned long val;
990         unsigned char *next_sha1;
991         enum object_type kind;
992
993         *delta_chain_length = 0;
994         offset = unpack_object_header(p, offset, &kind, size);
995
996         for (;;) {
997                 switch (kind) {
998                 default:
999                         die("corrupted pack file %s containing object of kind %d",
1000                             p->pack_name, kind);
1001                 case OBJ_COMMIT:
1002                 case OBJ_TREE:
1003                 case OBJ_BLOB:
1004                 case OBJ_TAG:
1005                         strcpy(type, type_names[kind]);
1006                         *store_size = 0; /* notyet */
1007                         return;
1008                 case OBJ_DELTA:
1009                         if (p->pack_size <= offset + 20)
1010                                 die("pack file %s records an incomplete delta base",
1011                                     p->pack_name);
1012                         next_sha1 = (unsigned char *) p->pack_base + offset;
1013                         if (*delta_chain_length == 0)
1014                                 hashcpy(base_sha1, next_sha1);
1015                         offset = find_pack_entry_one(next_sha1, p);
1016                         break;
1017                 }
1018                 offset = unpack_object_header(p, offset, &kind, &val);
1019                 (*delta_chain_length)++;
1020         }
1021 }
1022
1023 static int packed_object_info(struct packed_git *p, unsigned long offset,
1024                               char *type, unsigned long *sizep)
1025 {
1026         unsigned long size;
1027         enum object_type kind;
1028
1029         offset = unpack_object_header(p, offset, &kind, &size);
1030
1031         if (kind == OBJ_DELTA)
1032                 return packed_delta_info(p, offset, type, sizep);
1033
1034         switch (kind) {
1035         case OBJ_COMMIT:
1036         case OBJ_TREE:
1037         case OBJ_BLOB:
1038         case OBJ_TAG:
1039                 strcpy(type, type_names[kind]);
1040                 break;
1041         default:
1042                 die("corrupted pack file %s containing object of kind %d",
1043                     p->pack_name, kind);
1044         }
1045         if (sizep)
1046                 *sizep = size;
1047         return 0;
1048 }
1049
1050 static void *unpack_compressed_entry(struct packed_git *p,
1051                                     unsigned long offset,
1052                                     unsigned long size)
1053 {
1054         int st;
1055         z_stream stream;
1056         unsigned char *buffer;
1057
1058         buffer = xmalloc(size + 1);
1059         buffer[size] = 0;
1060         memset(&stream, 0, sizeof(stream));
1061         stream.next_in = (unsigned char*)p->pack_base + offset;
1062         stream.avail_in = p->pack_size - offset;
1063         stream.next_out = buffer;
1064         stream.avail_out = size;
1065
1066         inflateInit(&stream);
1067         st = inflate(&stream, Z_FINISH);
1068         inflateEnd(&stream);
1069         if ((st != Z_STREAM_END) || stream.total_out != size) {
1070                 free(buffer);
1071                 return NULL;
1072         }
1073
1074         return buffer;
1075 }
1076
1077 static void *unpack_delta_entry(struct packed_git *p,
1078                                 unsigned long offset,
1079                                 unsigned long delta_size,
1080                                 char *type,
1081                                 unsigned long *sizep)
1082 {
1083         void *delta_data, *result, *base;
1084         unsigned long result_size, base_size, base_offset;
1085         unsigned char *base_sha1;
1086
1087         if (p->pack_size < offset + 20)
1088                 die("truncated pack file");
1089         /* The base entry _must_ be in the same pack */
1090         base_sha1 = (unsigned char*)p->pack_base + offset;
1091         base_offset = find_pack_entry_one(base_sha1, p);
1092         if (!base_offset)
1093                 die("failed to find delta-pack base object %s",
1094                     sha1_to_hex(base_sha1));
1095         offset += 20;
1096
1097         base = unpack_entry_gently(p, base_offset, type, &base_size);
1098         if (!base)
1099                 die("failed to read delta base object at %lu from %s",
1100                     base_offset, p->pack_name);
1101
1102         delta_data = unpack_compressed_entry(p, offset, delta_size);
1103         result = patch_delta(base, base_size,
1104                              delta_data, delta_size,
1105                              &result_size);
1106         if (!result)
1107                 die("failed to apply delta");
1108         free(delta_data);
1109         free(base);
1110         *sizep = result_size;
1111         return result;
1112 }
1113
1114 static void *unpack_entry(struct pack_entry *entry,
1115                           char *type, unsigned long *sizep)
1116 {
1117         struct packed_git *p = entry->p;
1118         void *retval;
1119
1120         if (use_packed_git(p))
1121                 die("cannot map packed file");
1122         retval = unpack_entry_gently(p, entry->offset, type, sizep);
1123         unuse_packed_git(p);
1124         if (!retval)
1125                 die("corrupted pack file %s", p->pack_name);
1126         return retval;
1127 }
1128
1129 /* The caller is responsible for use_packed_git()/unuse_packed_git() pair */
1130 void *unpack_entry_gently(struct packed_git *p, unsigned long offset,
1131                           char *type, unsigned long *sizep)
1132 {
1133         unsigned long size;
1134         enum object_type kind;
1135
1136         offset = unpack_object_header(p, offset, &kind, &size);
1137         switch (kind) {
1138         case OBJ_DELTA:
1139                 return unpack_delta_entry(p, offset, size, type, sizep);
1140         case OBJ_COMMIT:
1141         case OBJ_TREE:
1142         case OBJ_BLOB:
1143         case OBJ_TAG:
1144                 strcpy(type, type_names[kind]);
1145                 *sizep = size;
1146                 return unpack_compressed_entry(p, offset, size);
1147         default:
1148                 return NULL;
1149         }
1150 }
1151
1152 int num_packed_objects(const struct packed_git *p)
1153 {
1154         /* See check_packed_git_idx() */
1155         return (p->index_size - 20 - 20 - 4*256) / 24;
1156 }
1157
1158 int nth_packed_object_sha1(const struct packed_git *p, int n,
1159                            unsigned char* sha1)
1160 {
1161         void *index = p->index_base + 256;
1162         if (n < 0 || num_packed_objects(p) <= n)
1163                 return -1;
1164         hashcpy(sha1, (unsigned char *) index + (24 * n) + 4);
1165         return 0;
1166 }
1167
1168 unsigned long find_pack_entry_one(const unsigned char *sha1,
1169                                   struct packed_git *p)
1170 {
1171         unsigned int *level1_ofs = p->index_base;
1172         int hi = ntohl(level1_ofs[*sha1]);
1173         int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1174         void *index = p->index_base + 256;
1175
1176         do {
1177                 int mi = (lo + hi) / 2;
1178                 int cmp = hashcmp((unsigned char *)index + (24 * mi) + 4, sha1);
1179                 if (!cmp)
1180                         return ntohl(*((unsigned int *) ((char *) index + (24 * mi))));
1181                 if (cmp > 0)
1182                         hi = mi;
1183                 else
1184                         lo = mi+1;
1185         } while (lo < hi);
1186         return 0;
1187 }
1188
1189 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e, const char **ignore_packed)
1190 {
1191         struct packed_git *p;
1192         unsigned long offset;
1193
1194         prepare_packed_git();
1195
1196         for (p = packed_git; p; p = p->next) {
1197                 if (ignore_packed) {
1198                         const char **ig;
1199                         for (ig = ignore_packed; *ig; ig++)
1200                                 if (!strcmp(p->pack_name, *ig))
1201                                         break;
1202                         if (*ig)
1203                                 continue;
1204                 }
1205                 offset = find_pack_entry_one(sha1, p);
1206                 if (offset) {
1207                         e->offset = offset;
1208                         e->p = p;
1209                         hashcpy(e->sha1, sha1);
1210                         return 1;
1211                 }
1212         }
1213         return 0;
1214 }
1215
1216 struct packed_git *find_sha1_pack(const unsigned char *sha1, 
1217                                   struct packed_git *packs)
1218 {
1219         struct packed_git *p;
1220
1221         for (p = packs; p; p = p->next) {
1222                 if (find_pack_entry_one(sha1, p))
1223                         return p;
1224         }
1225         return NULL;
1226         
1227 }
1228
1229 int sha1_object_info(const unsigned char *sha1, char *type, unsigned long *sizep)
1230 {
1231         int status;
1232         unsigned long mapsize, size;
1233         void *map;
1234         z_stream stream;
1235         char hdr[128];
1236
1237         map = map_sha1_file(sha1, &mapsize);
1238         if (!map) {
1239                 struct pack_entry e;
1240
1241                 if (!find_pack_entry(sha1, &e, NULL)) {
1242                         reprepare_packed_git();
1243                         if (!find_pack_entry(sha1, &e, NULL))
1244                                 return error("unable to find %s", sha1_to_hex(sha1));
1245                 }
1246                 if (use_packed_git(e.p))
1247                         die("cannot map packed file");
1248                 status = packed_object_info(e.p, e.offset, type, sizep);
1249                 unuse_packed_git(e.p);
1250                 return status;
1251         }
1252         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
1253                 status = error("unable to unpack %s header",
1254                                sha1_to_hex(sha1));
1255         if (parse_sha1_header(hdr, type, &size) < 0)
1256                 status = error("unable to parse %s header", sha1_to_hex(sha1));
1257         else {
1258                 status = 0;
1259                 if (sizep)
1260                         *sizep = size;
1261         }
1262         inflateEnd(&stream);
1263         munmap(map, mapsize);
1264         return status;
1265 }
1266
1267 static void *read_packed_sha1(const unsigned char *sha1, char *type, unsigned long *size)
1268 {
1269         struct pack_entry e;
1270
1271         if (!find_pack_entry(sha1, &e, NULL)) {
1272                 error("cannot read sha1_file for %s", sha1_to_hex(sha1));
1273                 return NULL;
1274         }
1275         return unpack_entry(&e, type, size);
1276 }
1277
1278 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
1279 {
1280         unsigned long mapsize;
1281         void *map, *buf;
1282         struct pack_entry e;
1283
1284         if (find_pack_entry(sha1, &e, NULL))
1285                 return read_packed_sha1(sha1, type, size);
1286         map = map_sha1_file(sha1, &mapsize);
1287         if (map) {
1288                 buf = unpack_sha1_file(map, mapsize, type, size);
1289                 munmap(map, mapsize);
1290                 return buf;
1291         }
1292         reprepare_packed_git();
1293         if (find_pack_entry(sha1, &e, NULL))
1294                 return read_packed_sha1(sha1, type, size);
1295         return NULL;
1296 }
1297
1298 void *read_object_with_reference(const unsigned char *sha1,
1299                                  const char *required_type,
1300                                  unsigned long *size,
1301                                  unsigned char *actual_sha1_return)
1302 {
1303         char type[20];
1304         void *buffer;
1305         unsigned long isize;
1306         unsigned char actual_sha1[20];
1307
1308         hashcpy(actual_sha1, sha1);
1309         while (1) {
1310                 int ref_length = -1;
1311                 const char *ref_type = NULL;
1312
1313                 buffer = read_sha1_file(actual_sha1, type, &isize);
1314                 if (!buffer)
1315                         return NULL;
1316                 if (!strcmp(type, required_type)) {
1317                         *size = isize;
1318                         if (actual_sha1_return)
1319                                 hashcpy(actual_sha1_return, actual_sha1);
1320                         return buffer;
1321                 }
1322                 /* Handle references */
1323                 else if (!strcmp(type, commit_type))
1324                         ref_type = "tree ";
1325                 else if (!strcmp(type, tag_type))
1326                         ref_type = "object ";
1327                 else {
1328                         free(buffer);
1329                         return NULL;
1330                 }
1331                 ref_length = strlen(ref_type);
1332
1333                 if (memcmp(buffer, ref_type, ref_length) ||
1334                     get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
1335                         free(buffer);
1336                         return NULL;
1337                 }
1338                 free(buffer);
1339                 /* Now we have the ID of the referred-to object in
1340                  * actual_sha1.  Check again. */
1341         }
1342 }
1343
1344 static void write_sha1_file_prepare(void *buf, unsigned long len,
1345                                     const char *type, unsigned char *sha1,
1346                                     unsigned char *hdr, int *hdrlen)
1347 {
1348         SHA_CTX c;
1349
1350         /* Generate the header */
1351         *hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
1352
1353         /* Sha1.. */
1354         SHA1_Init(&c);
1355         SHA1_Update(&c, hdr, *hdrlen);
1356         SHA1_Update(&c, buf, len);
1357         SHA1_Final(sha1, &c);
1358 }
1359
1360 /*
1361  * Link the tempfile to the final place, possibly creating the
1362  * last directory level as you do so.
1363  *
1364  * Returns the errno on failure, 0 on success.
1365  */
1366 static int link_temp_to_file(const char *tmpfile, const char *filename)
1367 {
1368         int ret;
1369         char *dir;
1370
1371         if (!link(tmpfile, filename))
1372                 return 0;
1373
1374         /*
1375          * Try to mkdir the last path component if that failed.
1376          *
1377          * Re-try the "link()" regardless of whether the mkdir
1378          * succeeds, since a race might mean that somebody
1379          * else succeeded.
1380          */
1381         ret = errno;
1382         dir = strrchr(filename, '/');
1383         if (dir) {
1384                 *dir = 0;
1385                 mkdir(filename, 0777);
1386                 if (adjust_shared_perm(filename))
1387                         return -2;
1388                 *dir = '/';
1389                 if (!link(tmpfile, filename))
1390                         return 0;
1391                 ret = errno;
1392         }
1393         return ret;
1394 }
1395
1396 /*
1397  * Move the just written object into its final resting place
1398  */
1399 int move_temp_to_file(const char *tmpfile, const char *filename)
1400 {
1401         int ret = link_temp_to_file(tmpfile, filename);
1402
1403         /*
1404          * Coda hack - coda doesn't like cross-directory links,
1405          * so we fall back to a rename, which will mean that it
1406          * won't be able to check collisions, but that's not a
1407          * big deal.
1408          *
1409          * The same holds for FAT formatted media.
1410          *
1411          * When this succeeds, we just return 0. We have nothing
1412          * left to unlink.
1413          */
1414         if (ret && ret != EEXIST) {
1415                 if (!rename(tmpfile, filename))
1416                         return 0;
1417                 ret = errno;
1418         }
1419         unlink(tmpfile);
1420         if (ret) {
1421                 if (ret != EEXIST) {
1422                         fprintf(stderr, "unable to write sha1 filename %s: %s\n", filename, strerror(ret));
1423                         return -1;
1424                 }
1425                 /* FIXME!!! Collision check here ? */
1426         }
1427
1428         return 0;
1429 }
1430
1431 static int write_buffer(int fd, const void *buf, size_t len)
1432 {
1433         while (len) {
1434                 ssize_t size;
1435
1436                 size = write(fd, buf, len);
1437                 if (!size)
1438                         return error("file write: disk full");
1439                 if (size < 0) {
1440                         if (errno == EINTR || errno == EAGAIN)
1441                                 continue;
1442                         return error("file write error (%s)", strerror(errno));
1443                 }
1444                 len -= size;
1445                 buf = (char *) buf + size;
1446         }
1447         return 0;
1448 }
1449
1450 static int write_binary_header(unsigned char *hdr, enum object_type type, unsigned long len)
1451 {
1452         int hdr_len;
1453         unsigned char c;
1454
1455         c = (type << 4) | (len & 15);
1456         len >>= 4;
1457         hdr_len = 1;
1458         while (len) {
1459                 *hdr++ = c | 0x80;
1460                 hdr_len++;
1461                 c = (len & 0x7f);
1462                 len >>= 7;
1463         }
1464         *hdr = c;
1465         return hdr_len;
1466 }
1467
1468 static void setup_object_header(z_stream *stream, const char *type, unsigned long len)
1469 {
1470         int obj_type, hdr;
1471
1472         if (use_legacy_headers) {
1473                 while (deflate(stream, 0) == Z_OK)
1474                         /* nothing */;
1475                 return;
1476         }
1477         if (!strcmp(type, blob_type))
1478                 obj_type = OBJ_BLOB;
1479         else if (!strcmp(type, tree_type))
1480                 obj_type = OBJ_TREE;
1481         else if (!strcmp(type, commit_type))
1482                 obj_type = OBJ_COMMIT;
1483         else if (!strcmp(type, tag_type))
1484                 obj_type = OBJ_TAG;
1485         else
1486                 die("trying to generate bogus object of type '%s'", type);
1487         hdr = write_binary_header(stream->next_out, obj_type, len);
1488         stream->total_out = hdr;
1489         stream->next_out += hdr;
1490         stream->avail_out -= hdr;
1491 }
1492
1493 int hash_sha1_file(void *buf, unsigned long len, const char *type,
1494                    unsigned char *sha1)
1495 {
1496         unsigned char hdr[50];
1497         int hdrlen;
1498         write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1499         return 0;
1500 }
1501
1502 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
1503 {
1504         int size;
1505         unsigned char *compressed;
1506         z_stream stream;
1507         unsigned char sha1[20];
1508         char *filename;
1509         static char tmpfile[PATH_MAX];
1510         unsigned char hdr[50];
1511         int fd, hdrlen;
1512
1513         /* Normally if we have it in the pack then we do not bother writing
1514          * it out into .git/objects/??/?{38} file.
1515          */
1516         write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
1517         filename = sha1_file_name(sha1);
1518         if (returnsha1)
1519                 hashcpy(returnsha1, sha1);
1520         if (has_sha1_file(sha1))
1521                 return 0;
1522         fd = open(filename, O_RDONLY);
1523         if (fd >= 0) {
1524                 /*
1525                  * FIXME!!! We might do collision checking here, but we'd
1526                  * need to uncompress the old file and check it. Later.
1527                  */
1528                 close(fd);
1529                 return 0;
1530         }
1531
1532         if (errno != ENOENT) {
1533                 fprintf(stderr, "sha1 file %s: %s\n", filename, strerror(errno));
1534                 return -1;
1535         }
1536
1537         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1538
1539         fd = mkstemp(tmpfile);
1540         if (fd < 0) {
1541                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
1542                 return -1;
1543         }
1544
1545         /* Set it up */
1546         memset(&stream, 0, sizeof(stream));
1547         deflateInit(&stream, zlib_compression_level);
1548         size = 8 + deflateBound(&stream, len+hdrlen);
1549         compressed = xmalloc(size);
1550
1551         /* Compress it */
1552         stream.next_out = compressed;
1553         stream.avail_out = size;
1554
1555         /* First header.. */
1556         stream.next_in = hdr;
1557         stream.avail_in = hdrlen;
1558         setup_object_header(&stream, type, len);
1559
1560         /* Then the data itself.. */
1561         stream.next_in = buf;
1562         stream.avail_in = len;
1563         while (deflate(&stream, Z_FINISH) == Z_OK)
1564                 /* nothing */;
1565         deflateEnd(&stream);
1566         size = stream.total_out;
1567
1568         if (write_buffer(fd, compressed, size) < 0)
1569                 die("unable to write sha1 file");
1570         fchmod(fd, 0444);
1571         close(fd);
1572         free(compressed);
1573
1574         return move_temp_to_file(tmpfile, filename);
1575 }
1576
1577 /*
1578  * We need to unpack and recompress the object for writing
1579  * it out to a different file.
1580  */
1581 static void *repack_object(const unsigned char *sha1, unsigned long *objsize)
1582 {
1583         size_t size;
1584         z_stream stream;
1585         unsigned char *unpacked;
1586         unsigned long len;
1587         char type[20];
1588         char hdr[50];
1589         int hdrlen;
1590         void *buf;
1591
1592         /* need to unpack and recompress it by itself */
1593         unpacked = read_packed_sha1(sha1, type, &len);
1594
1595         hdrlen = sprintf(hdr, "%s %lu", type, len) + 1;
1596
1597         /* Set it up */
1598         memset(&stream, 0, sizeof(stream));
1599         deflateInit(&stream, zlib_compression_level);
1600         size = deflateBound(&stream, len + hdrlen);
1601         buf = xmalloc(size);
1602
1603         /* Compress it */
1604         stream.next_out = buf;
1605         stream.avail_out = size;
1606
1607         /* First header.. */
1608         stream.next_in = (void *)hdr;
1609         stream.avail_in = hdrlen;
1610         while (deflate(&stream, 0) == Z_OK)
1611                 /* nothing */;
1612
1613         /* Then the data itself.. */
1614         stream.next_in = unpacked;
1615         stream.avail_in = len;
1616         while (deflate(&stream, Z_FINISH) == Z_OK)
1617                 /* nothing */;
1618         deflateEnd(&stream);
1619         free(unpacked);
1620
1621         *objsize = stream.total_out;
1622         return buf;
1623 }
1624
1625 int write_sha1_to_fd(int fd, const unsigned char *sha1)
1626 {
1627         int retval;
1628         unsigned long objsize;
1629         void *buf = map_sha1_file(sha1, &objsize);
1630
1631         if (buf) {
1632                 retval = write_buffer(fd, buf, objsize);
1633                 munmap(buf, objsize);
1634                 return retval;
1635         }
1636
1637         buf = repack_object(sha1, &objsize);
1638         retval = write_buffer(fd, buf, objsize);
1639         free(buf);
1640         return retval;
1641 }
1642
1643 int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
1644                        size_t bufsize, size_t *bufposn)
1645 {
1646         char tmpfile[PATH_MAX];
1647         int local;
1648         z_stream stream;
1649         unsigned char real_sha1[20];
1650         unsigned char discard[4096];
1651         int ret;
1652         SHA_CTX c;
1653
1654         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
1655
1656         local = mkstemp(tmpfile);
1657         if (local < 0)
1658                 return error("Couldn't open %s for %s",
1659                              tmpfile, sha1_to_hex(sha1));
1660
1661         memset(&stream, 0, sizeof(stream));
1662
1663         inflateInit(&stream);
1664
1665         SHA1_Init(&c);
1666
1667         do {
1668                 ssize_t size;
1669                 if (*bufposn) {
1670                         stream.avail_in = *bufposn;
1671                         stream.next_in = (unsigned char *) buffer;
1672                         do {
1673                                 stream.next_out = discard;
1674                                 stream.avail_out = sizeof(discard);
1675                                 ret = inflate(&stream, Z_SYNC_FLUSH);
1676                                 SHA1_Update(&c, discard, sizeof(discard) -
1677                                             stream.avail_out);
1678                         } while (stream.avail_in && ret == Z_OK);
1679                         if (write_buffer(local, buffer, *bufposn - stream.avail_in) < 0)
1680                                 die("unable to write sha1 file");
1681                         memmove(buffer, buffer + *bufposn - stream.avail_in,
1682                                 stream.avail_in);
1683                         *bufposn = stream.avail_in;
1684                         if (ret != Z_OK)
1685                                 break;
1686                 }
1687                 size = read(fd, buffer + *bufposn, bufsize - *bufposn);
1688                 if (size <= 0) {
1689                         close(local);
1690                         unlink(tmpfile);
1691                         if (!size)
1692                                 return error("Connection closed?");
1693                         perror("Reading from connection");
1694                         return -1;
1695                 }
1696                 *bufposn += size;
1697         } while (1);
1698         inflateEnd(&stream);
1699
1700         close(local);
1701         SHA1_Final(real_sha1, &c);
1702         if (ret != Z_STREAM_END) {
1703                 unlink(tmpfile);
1704                 return error("File %s corrupted", sha1_to_hex(sha1));
1705         }
1706         if (hashcmp(sha1, real_sha1)) {
1707                 unlink(tmpfile);
1708                 return error("File %s has bad hash", sha1_to_hex(sha1));
1709         }
1710
1711         return move_temp_to_file(tmpfile, sha1_file_name(sha1));
1712 }
1713
1714 int has_pack_index(const unsigned char *sha1)
1715 {
1716         struct stat st;
1717         if (stat(sha1_pack_index_name(sha1), &st))
1718                 return 0;
1719         return 1;
1720 }
1721
1722 int has_pack_file(const unsigned char *sha1)
1723 {
1724         struct stat st;
1725         if (stat(sha1_pack_name(sha1), &st))
1726                 return 0;
1727         return 1;
1728 }
1729
1730 int has_sha1_pack(const unsigned char *sha1, const char **ignore_packed)
1731 {
1732         struct pack_entry e;
1733         return find_pack_entry(sha1, &e, ignore_packed);
1734 }
1735
1736 int has_sha1_file(const unsigned char *sha1)
1737 {
1738         struct stat st;
1739         struct pack_entry e;
1740
1741         if (find_pack_entry(sha1, &e, NULL))
1742                 return 1;
1743         return find_sha1_file(sha1, &st) ? 1 : 0;
1744 }
1745
1746 /*
1747  * reads from fd as long as possible into a supplied buffer of size bytes.
1748  * If necessary the buffer's size is increased using realloc()
1749  *
1750  * returns 0 if anything went fine and -1 otherwise
1751  *
1752  * NOTE: both buf and size may change, but even when -1 is returned
1753  * you still have to free() it yourself.
1754  */
1755 int read_pipe(int fd, char** return_buf, unsigned long* return_size)
1756 {
1757         char* buf = *return_buf;
1758         unsigned long size = *return_size;
1759         int iret;
1760         unsigned long off = 0;
1761
1762         do {
1763                 iret = xread(fd, buf + off, size - off);
1764                 if (iret > 0) {
1765                         off += iret;
1766                         if (off == size) {
1767                                 size *= 2;
1768                                 buf = xrealloc(buf, size);
1769                         }
1770                 }
1771         } while (iret > 0);
1772
1773         *return_buf = buf;
1774         *return_size = off;
1775
1776         if (iret < 0)
1777                 return -1;
1778         return 0;
1779 }
1780
1781 int index_pipe(unsigned char *sha1, int fd, const char *type, int write_object)
1782 {
1783         unsigned long size = 4096;
1784         char *buf = xmalloc(size);
1785         int ret;
1786
1787         if (read_pipe(fd, &buf, &size)) {
1788                 free(buf);
1789                 return -1;
1790         }
1791
1792         if (!type)
1793                 type = blob_type;
1794         if (write_object)
1795                 ret = write_sha1_file(buf, size, type, sha1);
1796         else
1797                 ret = hash_sha1_file(buf, size, type, sha1);
1798         free(buf);
1799         return ret;
1800 }
1801
1802 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object, const char *type)
1803 {
1804         unsigned long size = st->st_size;
1805         void *buf;
1806         int ret;
1807
1808         buf = "";
1809         if (size)
1810                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
1811         close(fd);
1812         if (buf == MAP_FAILED)
1813                 return -1;
1814
1815         if (!type)
1816                 type = blob_type;
1817         if (write_object)
1818                 ret = write_sha1_file(buf, size, type, sha1);
1819         else
1820                 ret = hash_sha1_file(buf, size, type, sha1);
1821         if (size)
1822                 munmap(buf, size);
1823         return ret;
1824 }
1825
1826 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
1827 {
1828         int fd;
1829         char *target;
1830
1831         switch (st->st_mode & S_IFMT) {
1832         case S_IFREG:
1833                 fd = open(path, O_RDONLY);
1834                 if (fd < 0)
1835                         return error("open(\"%s\"): %s", path,
1836                                      strerror(errno));
1837                 if (index_fd(sha1, fd, st, write_object, NULL) < 0)
1838                         return error("%s: failed to insert into database",
1839                                      path);
1840                 break;
1841         case S_IFLNK:
1842                 target = xmalloc(st->st_size+1);
1843                 if (readlink(path, target, st->st_size+1) != st->st_size) {
1844                         char *errstr = strerror(errno);
1845                         free(target);
1846                         return error("readlink(\"%s\"): %s", path,
1847                                      errstr);
1848                 }
1849                 if (!write_object)
1850                         hash_sha1_file(target, st->st_size, blob_type, sha1);
1851                 else if (write_sha1_file(target, st->st_size, blob_type, sha1))
1852                         return error("%s: failed to insert into database",
1853                                      path);
1854                 free(target);
1855                 break;
1856         default:
1857                 return error("%s: unsupported file type", path);
1858         }
1859         return 0;
1860 }