Merge branch 'jl/maint-fetch-recursive-fix'
[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 "run-command.h"
15 #include "tag.h"
16 #include "tree.h"
17 #include "tree-walk.h"
18 #include "refs.h"
19 #include "pack-revindex.h"
20 #include "sha1-lookup.h"
21
22 #ifndef O_NOATIME
23 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
24 #define O_NOATIME 01000000
25 #else
26 #define O_NOATIME 0
27 #endif
28 #endif
29
30 #define SZ_FMT PRIuMAX
31 static inline uintmax_t sz_fmt(size_t s) { return s; }
32
33 const unsigned char null_sha1[20];
34
35 /*
36  * This is meant to hold a *small* number of objects that you would
37  * want read_sha1_file() to be able to return, but yet you do not want
38  * to write them into the object store (e.g. a browse-only
39  * application).
40  */
41 static struct cached_object {
42         unsigned char sha1[20];
43         enum object_type type;
44         void *buf;
45         unsigned long size;
46 } *cached_objects;
47 static int cached_object_nr, cached_object_alloc;
48
49 static struct cached_object empty_tree = {
50         EMPTY_TREE_SHA1_BIN_LITERAL,
51         OBJ_TREE,
52         "",
53         0
54 };
55
56 static struct cached_object *find_cached_object(const unsigned char *sha1)
57 {
58         int i;
59         struct cached_object *co = cached_objects;
60
61         for (i = 0; i < cached_object_nr; i++, co++) {
62                 if (!hashcmp(co->sha1, sha1))
63                         return co;
64         }
65         if (!hashcmp(sha1, empty_tree.sha1))
66                 return &empty_tree;
67         return NULL;
68 }
69
70 int mkdir_in_gitdir(const char *path)
71 {
72         if (mkdir(path, 0777)) {
73                 int saved_errno = errno;
74                 struct stat st;
75                 struct strbuf sb = STRBUF_INIT;
76
77                 if (errno != EEXIST)
78                         return -1;
79                 /*
80                  * Are we looking at a path in a symlinked worktree
81                  * whose original repository does not yet have it?
82                  * e.g. .git/rr-cache pointing at its original
83                  * repository in which the user hasn't performed any
84                  * conflict resolution yet?
85                  */
86                 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
87                     strbuf_readlink(&sb, path, st.st_size) ||
88                     !is_absolute_path(sb.buf) ||
89                     mkdir(sb.buf, 0777)) {
90                         strbuf_release(&sb);
91                         errno = saved_errno;
92                         return -1;
93                 }
94                 strbuf_release(&sb);
95         }
96         return adjust_shared_perm(path);
97 }
98
99 int safe_create_leading_directories(char *path)
100 {
101         char *pos = path + offset_1st_component(path);
102         struct stat st;
103
104         while (pos) {
105                 pos = strchr(pos, '/');
106                 if (!pos)
107                         break;
108                 while (*++pos == '/')
109                         ;
110                 if (!*pos)
111                         break;
112                 *--pos = '\0';
113                 if (!stat(path, &st)) {
114                         /* path exists */
115                         if (!S_ISDIR(st.st_mode)) {
116                                 *pos = '/';
117                                 return -3;
118                         }
119                 }
120                 else if (mkdir(path, 0777)) {
121                         *pos = '/';
122                         return -1;
123                 }
124                 else if (adjust_shared_perm(path)) {
125                         *pos = '/';
126                         return -2;
127                 }
128                 *pos++ = '/';
129         }
130         return 0;
131 }
132
133 int safe_create_leading_directories_const(const char *path)
134 {
135         /* path points to cache entries, so xstrdup before messing with it */
136         char *buf = xstrdup(path);
137         int result = safe_create_leading_directories(buf);
138         free(buf);
139         return result;
140 }
141
142 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
143 {
144         int i;
145         for (i = 0; i < 20; i++) {
146                 static char hex[] = "0123456789abcdef";
147                 unsigned int val = sha1[i];
148                 char *pos = pathbuf + i*2 + (i > 0);
149                 *pos++ = hex[val >> 4];
150                 *pos = hex[val & 0xf];
151         }
152 }
153
154 /*
155  * NOTE! This returns a statically allocated buffer, so you have to be
156  * careful about using it. Do an "xstrdup()" if you need to save the
157  * filename.
158  *
159  * Also note that this returns the location for creating.  Reading
160  * SHA1 file can happen from any alternate directory listed in the
161  * DB_ENVIRONMENT environment variable if it is not found in
162  * the primary object database.
163  */
164 char *sha1_file_name(const unsigned char *sha1)
165 {
166         static char buf[PATH_MAX];
167         const char *objdir;
168         int len;
169
170         objdir = get_object_directory();
171         len = strlen(objdir);
172
173         /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
174         if (len + 43 > PATH_MAX)
175                 die("insanely long object directory %s", objdir);
176         memcpy(buf, objdir, len);
177         buf[len] = '/';
178         buf[len+3] = '/';
179         buf[len+42] = '\0';
180         fill_sha1_path(buf + len + 1, sha1);
181         return buf;
182 }
183
184 static char *sha1_get_pack_name(const unsigned char *sha1,
185                                 char **name, char **base, const char *which)
186 {
187         static const char hex[] = "0123456789abcdef";
188         char *buf;
189         int i;
190
191         if (!*base) {
192                 const char *sha1_file_directory = get_object_directory();
193                 int len = strlen(sha1_file_directory);
194                 *base = xmalloc(len + 60);
195                 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
196                         sha1_file_directory, which);
197                 *name = *base + len + 11;
198         }
199
200         buf = *name;
201
202         for (i = 0; i < 20; i++) {
203                 unsigned int val = *sha1++;
204                 *buf++ = hex[val >> 4];
205                 *buf++ = hex[val & 0xf];
206         }
207
208         return *base;
209 }
210
211 char *sha1_pack_name(const unsigned char *sha1)
212 {
213         static char *name, *base;
214
215         return sha1_get_pack_name(sha1, &name, &base, "pack");
216 }
217
218 char *sha1_pack_index_name(const unsigned char *sha1)
219 {
220         static char *name, *base;
221
222         return sha1_get_pack_name(sha1, &name, &base, "idx");
223 }
224
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
227
228 static void read_info_alternates(const char * alternates, int depth);
229 static int git_open_noatime(const char *name);
230
231 /*
232  * Prepare alternate object database registry.
233  *
234  * The variable alt_odb_list points at the list of struct
235  * alternate_object_database.  The elements on this list come from
236  * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
237  * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
238  * whose contents is similar to that environment variable but can be
239  * LF separated.  Its base points at a statically allocated buffer that
240  * contains "/the/directory/corresponding/to/.git/objects/...", while
241  * its name points just after the slash at the end of ".git/objects/"
242  * in the example above, and has enough space to hold 40-byte hex
243  * SHA1, an extra slash for the first level indirection, and the
244  * terminating NUL.
245  */
246 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
247 {
248         const char *objdir = get_object_directory();
249         struct alternate_object_database *ent;
250         struct alternate_object_database *alt;
251         /* 43 = 40-byte + 2 '/' + terminating NUL */
252         int pfxlen = len;
253         int entlen = pfxlen + 43;
254         int base_len = -1;
255
256         if (!is_absolute_path(entry) && relative_base) {
257                 /* Relative alt-odb */
258                 if (base_len < 0)
259                         base_len = strlen(relative_base) + 1;
260                 entlen += base_len;
261                 pfxlen += base_len;
262         }
263         ent = xmalloc(sizeof(*ent) + entlen);
264
265         if (!is_absolute_path(entry) && relative_base) {
266                 memcpy(ent->base, relative_base, base_len - 1);
267                 ent->base[base_len - 1] = '/';
268                 memcpy(ent->base + base_len, entry, len);
269         }
270         else
271                 memcpy(ent->base, entry, pfxlen);
272
273         ent->name = ent->base + pfxlen + 1;
274         ent->base[pfxlen + 3] = '/';
275         ent->base[pfxlen] = ent->base[entlen-1] = 0;
276
277         /* Detect cases where alternate disappeared */
278         if (!is_directory(ent->base)) {
279                 error("object directory %s does not exist; "
280                       "check .git/objects/info/alternates.",
281                       ent->base);
282                 free(ent);
283                 return -1;
284         }
285
286         /* Prevent the common mistake of listing the same
287          * thing twice, or object directory itself.
288          */
289         for (alt = alt_odb_list; alt; alt = alt->next) {
290                 if (!memcmp(ent->base, alt->base, pfxlen)) {
291                         free(ent);
292                         return -1;
293                 }
294         }
295         if (!memcmp(ent->base, objdir, pfxlen)) {
296                 free(ent);
297                 return -1;
298         }
299
300         /* add the alternate entry */
301         *alt_odb_tail = ent;
302         alt_odb_tail = &(ent->next);
303         ent->next = NULL;
304
305         /* recursively add alternates */
306         read_info_alternates(ent->base, depth + 1);
307
308         ent->base[pfxlen] = '/';
309
310         return 0;
311 }
312
313 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
314                                  const char *relative_base, int depth)
315 {
316         const char *cp, *last;
317
318         if (depth > 5) {
319                 error("%s: ignoring alternate object stores, nesting too deep.",
320                                 relative_base);
321                 return;
322         }
323
324         last = alt;
325         while (last < ep) {
326                 cp = last;
327                 if (cp < ep && *cp == '#') {
328                         while (cp < ep && *cp != sep)
329                                 cp++;
330                         last = cp + 1;
331                         continue;
332                 }
333                 while (cp < ep && *cp != sep)
334                         cp++;
335                 if (last != cp) {
336                         if (!is_absolute_path(last) && depth) {
337                                 error("%s: ignoring relative alternate object store %s",
338                                                 relative_base, last);
339                         } else {
340                                 link_alt_odb_entry(last, cp - last,
341                                                 relative_base, depth);
342                         }
343                 }
344                 while (cp < ep && *cp == sep)
345                         cp++;
346                 last = cp;
347         }
348 }
349
350 static void read_info_alternates(const char * relative_base, int depth)
351 {
352         char *map;
353         size_t mapsz;
354         struct stat st;
355         const char alt_file_name[] = "info/alternates";
356         /* Given that relative_base is no longer than PATH_MAX,
357            ensure that "path" has enough space to append "/", the
358            file name, "info/alternates", and a trailing NUL.  */
359         char path[PATH_MAX + 1 + sizeof alt_file_name];
360         int fd;
361
362         sprintf(path, "%s/%s", relative_base, alt_file_name);
363         fd = git_open_noatime(path);
364         if (fd < 0)
365                 return;
366         if (fstat(fd, &st) || (st.st_size == 0)) {
367                 close(fd);
368                 return;
369         }
370         mapsz = xsize_t(st.st_size);
371         map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
372         close(fd);
373
374         link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
375
376         munmap(map, mapsz);
377 }
378
379 void add_to_alternates_file(const char *reference)
380 {
381         struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
382         int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
383         char *alt = mkpath("%s/objects\n", reference);
384         write_or_die(fd, alt, strlen(alt));
385         if (commit_lock_file(lock))
386                 die("could not close alternates file");
387         if (alt_odb_tail)
388                 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
389 }
390
391 void foreach_alt_odb(alt_odb_fn fn, void *cb)
392 {
393         struct alternate_object_database *ent;
394
395         prepare_alt_odb();
396         for (ent = alt_odb_list; ent; ent = ent->next)
397                 if (fn(ent, cb))
398                         return;
399 }
400
401 void prepare_alt_odb(void)
402 {
403         const char *alt;
404
405         if (alt_odb_tail)
406                 return;
407
408         alt = getenv(ALTERNATE_DB_ENVIRONMENT);
409         if (!alt) alt = "";
410
411         alt_odb_tail = &alt_odb_list;
412         link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
413
414         read_info_alternates(get_object_directory(), 0);
415 }
416
417 static int has_loose_object_local(const unsigned char *sha1)
418 {
419         char *name = sha1_file_name(sha1);
420         return !access(name, F_OK);
421 }
422
423 int has_loose_object_nonlocal(const unsigned char *sha1)
424 {
425         struct alternate_object_database *alt;
426         prepare_alt_odb();
427         for (alt = alt_odb_list; alt; alt = alt->next) {
428                 fill_sha1_path(alt->name, sha1);
429                 if (!access(alt->base, F_OK))
430                         return 1;
431         }
432         return 0;
433 }
434
435 static int has_loose_object(const unsigned char *sha1)
436 {
437         return has_loose_object_local(sha1) ||
438                has_loose_object_nonlocal(sha1);
439 }
440
441 static unsigned int pack_used_ctr;
442 static unsigned int pack_mmap_calls;
443 static unsigned int peak_pack_open_windows;
444 static unsigned int pack_open_windows;
445 static unsigned int pack_open_fds;
446 static unsigned int pack_max_fds;
447 static size_t peak_pack_mapped;
448 static size_t pack_mapped;
449 struct packed_git *packed_git;
450
451 void pack_report(void)
452 {
453         fprintf(stderr,
454                 "pack_report: getpagesize()            = %10" SZ_FMT "\n"
455                 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
456                 "pack_report: core.packedGitLimit      = %10" SZ_FMT "\n",
457                 sz_fmt(getpagesize()),
458                 sz_fmt(packed_git_window_size),
459                 sz_fmt(packed_git_limit));
460         fprintf(stderr,
461                 "pack_report: pack_used_ctr            = %10u\n"
462                 "pack_report: pack_mmap_calls          = %10u\n"
463                 "pack_report: pack_open_windows        = %10u / %10u\n"
464                 "pack_report: pack_mapped              = "
465                         "%10" SZ_FMT " / %10" SZ_FMT "\n",
466                 pack_used_ctr,
467                 pack_mmap_calls,
468                 pack_open_windows, peak_pack_open_windows,
469                 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
470 }
471
472 static int check_packed_git_idx(const char *path,  struct packed_git *p)
473 {
474         void *idx_map;
475         struct pack_idx_header *hdr;
476         size_t idx_size;
477         uint32_t version, nr, i, *index;
478         int fd = git_open_noatime(path);
479         struct stat st;
480
481         if (fd < 0)
482                 return -1;
483         if (fstat(fd, &st)) {
484                 close(fd);
485                 return -1;
486         }
487         idx_size = xsize_t(st.st_size);
488         if (idx_size < 4 * 256 + 20 + 20) {
489                 close(fd);
490                 return error("index file %s is too small", path);
491         }
492         idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
493         close(fd);
494
495         hdr = idx_map;
496         if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
497                 version = ntohl(hdr->idx_version);
498                 if (version < 2 || version > 2) {
499                         munmap(idx_map, idx_size);
500                         return error("index file %s is version %"PRIu32
501                                      " and is not supported by this binary"
502                                      " (try upgrading GIT to a newer version)",
503                                      path, version);
504                 }
505         } else
506                 version = 1;
507
508         nr = 0;
509         index = idx_map;
510         if (version > 1)
511                 index += 2;  /* skip index header */
512         for (i = 0; i < 256; i++) {
513                 uint32_t n = ntohl(index[i]);
514                 if (n < nr) {
515                         munmap(idx_map, idx_size);
516                         return error("non-monotonic index %s", path);
517                 }
518                 nr = n;
519         }
520
521         if (version == 1) {
522                 /*
523                  * Total size:
524                  *  - 256 index entries 4 bytes each
525                  *  - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
526                  *  - 20-byte SHA1 of the packfile
527                  *  - 20-byte SHA1 file checksum
528                  */
529                 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
530                         munmap(idx_map, idx_size);
531                         return error("wrong index v1 file size in %s", path);
532                 }
533         } else if (version == 2) {
534                 /*
535                  * Minimum size:
536                  *  - 8 bytes of header
537                  *  - 256 index entries 4 bytes each
538                  *  - 20-byte sha1 entry * nr
539                  *  - 4-byte crc entry * nr
540                  *  - 4-byte offset entry * nr
541                  *  - 20-byte SHA1 of the packfile
542                  *  - 20-byte SHA1 file checksum
543                  * And after the 4-byte offset table might be a
544                  * variable sized table containing 8-byte entries
545                  * for offsets larger than 2^31.
546                  */
547                 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
548                 unsigned long max_size = min_size;
549                 if (nr)
550                         max_size += (nr - 1)*8;
551                 if (idx_size < min_size || idx_size > max_size) {
552                         munmap(idx_map, idx_size);
553                         return error("wrong index v2 file size in %s", path);
554                 }
555                 if (idx_size != min_size &&
556                     /*
557                      * make sure we can deal with large pack offsets.
558                      * 31-bit signed offset won't be enough, neither
559                      * 32-bit unsigned one will be.
560                      */
561                     (sizeof(off_t) <= 4)) {
562                         munmap(idx_map, idx_size);
563                         return error("pack too large for current definition of off_t in %s", path);
564                 }
565         }
566
567         p->index_version = version;
568         p->index_data = idx_map;
569         p->index_size = idx_size;
570         p->num_objects = nr;
571         return 0;
572 }
573
574 int open_pack_index(struct packed_git *p)
575 {
576         char *idx_name;
577         int ret;
578
579         if (p->index_data)
580                 return 0;
581
582         idx_name = xstrdup(p->pack_name);
583         strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
584         ret = check_packed_git_idx(idx_name, p);
585         free(idx_name);
586         return ret;
587 }
588
589 static void scan_windows(struct packed_git *p,
590         struct packed_git **lru_p,
591         struct pack_window **lru_w,
592         struct pack_window **lru_l)
593 {
594         struct pack_window *w, *w_l;
595
596         for (w_l = NULL, w = p->windows; w; w = w->next) {
597                 if (!w->inuse_cnt) {
598                         if (!*lru_w || w->last_used < (*lru_w)->last_used) {
599                                 *lru_p = p;
600                                 *lru_w = w;
601                                 *lru_l = w_l;
602                         }
603                 }
604                 w_l = w;
605         }
606 }
607
608 static int unuse_one_window(struct packed_git *current, int keep_fd)
609 {
610         struct packed_git *p, *lru_p = NULL;
611         struct pack_window *lru_w = NULL, *lru_l = NULL;
612
613         if (current)
614                 scan_windows(current, &lru_p, &lru_w, &lru_l);
615         for (p = packed_git; p; p = p->next)
616                 scan_windows(p, &lru_p, &lru_w, &lru_l);
617         if (lru_p) {
618                 munmap(lru_w->base, lru_w->len);
619                 pack_mapped -= lru_w->len;
620                 if (lru_l)
621                         lru_l->next = lru_w->next;
622                 else {
623                         lru_p->windows = lru_w->next;
624                         if (!lru_p->windows && lru_p->pack_fd != -1
625                                 && lru_p->pack_fd != keep_fd) {
626                                 close(lru_p->pack_fd);
627                                 pack_open_fds--;
628                                 lru_p->pack_fd = -1;
629                         }
630                 }
631                 free(lru_w);
632                 pack_open_windows--;
633                 return 1;
634         }
635         return 0;
636 }
637
638 void release_pack_memory(size_t need, int fd)
639 {
640         size_t cur = pack_mapped;
641         while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
642                 ; /* nothing */
643 }
644
645 void *xmmap(void *start, size_t length,
646         int prot, int flags, int fd, off_t offset)
647 {
648         void *ret = mmap(start, length, prot, flags, fd, offset);
649         if (ret == MAP_FAILED) {
650                 if (!length)
651                         return NULL;
652                 release_pack_memory(length, fd);
653                 ret = mmap(start, length, prot, flags, fd, offset);
654                 if (ret == MAP_FAILED)
655                         die_errno("Out of memory? mmap failed");
656         }
657         return ret;
658 }
659
660 void close_pack_windows(struct packed_git *p)
661 {
662         while (p->windows) {
663                 struct pack_window *w = p->windows;
664
665                 if (w->inuse_cnt)
666                         die("pack '%s' still has open windows to it",
667                             p->pack_name);
668                 munmap(w->base, w->len);
669                 pack_mapped -= w->len;
670                 pack_open_windows--;
671                 p->windows = w->next;
672                 free(w);
673         }
674 }
675
676 void unuse_pack(struct pack_window **w_cursor)
677 {
678         struct pack_window *w = *w_cursor;
679         if (w) {
680                 w->inuse_cnt--;
681                 *w_cursor = NULL;
682         }
683 }
684
685 void close_pack_index(struct packed_git *p)
686 {
687         if (p->index_data) {
688                 munmap((void *)p->index_data, p->index_size);
689                 p->index_data = NULL;
690         }
691 }
692
693 /*
694  * This is used by git-repack in case a newly created pack happens to
695  * contain the same set of objects as an existing one.  In that case
696  * the resulting file might be different even if its name would be the
697  * same.  It is best to close any reference to the old pack before it is
698  * replaced on disk.  Of course no index pointers nor windows for given pack
699  * must subsist at this point.  If ever objects from this pack are requested
700  * again, the new version of the pack will be reinitialized through
701  * reprepare_packed_git().
702  */
703 void free_pack_by_name(const char *pack_name)
704 {
705         struct packed_git *p, **pp = &packed_git;
706
707         while (*pp) {
708                 p = *pp;
709                 if (strcmp(pack_name, p->pack_name) == 0) {
710                         clear_delta_base_cache();
711                         close_pack_windows(p);
712                         if (p->pack_fd != -1) {
713                                 close(p->pack_fd);
714                                 pack_open_fds--;
715                         }
716                         close_pack_index(p);
717                         free(p->bad_object_sha1);
718                         *pp = p->next;
719                         free(p);
720                         return;
721                 }
722                 pp = &p->next;
723         }
724 }
725
726 /*
727  * Do not call this directly as this leaks p->pack_fd on error return;
728  * call open_packed_git() instead.
729  */
730 static int open_packed_git_1(struct packed_git *p)
731 {
732         struct stat st;
733         struct pack_header hdr;
734         unsigned char sha1[20];
735         unsigned char *idx_sha1;
736         long fd_flag;
737
738         if (!p->index_data && open_pack_index(p))
739                 return error("packfile %s index unavailable", p->pack_name);
740
741         if (!pack_max_fds) {
742                 struct rlimit lim;
743                 unsigned int max_fds;
744
745                 if (getrlimit(RLIMIT_NOFILE, &lim))
746                         die_errno("cannot get RLIMIT_NOFILE");
747
748                 max_fds = lim.rlim_cur;
749
750                 /* Save 3 for stdin/stdout/stderr, 22 for work */
751                 if (25 < max_fds)
752                         pack_max_fds = max_fds - 25;
753                 else
754                         pack_max_fds = 1;
755         }
756
757         while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
758                 ; /* nothing */
759
760         p->pack_fd = git_open_noatime(p->pack_name);
761         if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
762                 return -1;
763         pack_open_fds++;
764
765         /* If we created the struct before we had the pack we lack size. */
766         if (!p->pack_size) {
767                 if (!S_ISREG(st.st_mode))
768                         return error("packfile %s not a regular file", p->pack_name);
769                 p->pack_size = st.st_size;
770         } else if (p->pack_size != st.st_size)
771                 return error("packfile %s size changed", p->pack_name);
772
773         /* We leave these file descriptors open with sliding mmap;
774          * there is no point keeping them open across exec(), though.
775          */
776         fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
777         if (fd_flag < 0)
778                 return error("cannot determine file descriptor flags");
779         fd_flag |= FD_CLOEXEC;
780         if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
781                 return error("cannot set FD_CLOEXEC");
782
783         /* Verify we recognize this pack file format. */
784         if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
785                 return error("file %s is far too short to be a packfile", p->pack_name);
786         if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
787                 return error("file %s is not a GIT packfile", p->pack_name);
788         if (!pack_version_ok(hdr.hdr_version))
789                 return error("packfile %s is version %"PRIu32" and not"
790                         " supported (try upgrading GIT to a newer version)",
791                         p->pack_name, ntohl(hdr.hdr_version));
792
793         /* Verify the pack matches its index. */
794         if (p->num_objects != ntohl(hdr.hdr_entries))
795                 return error("packfile %s claims to have %"PRIu32" objects"
796                              " while index indicates %"PRIu32" objects",
797                              p->pack_name, ntohl(hdr.hdr_entries),
798                              p->num_objects);
799         if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
800                 return error("end of packfile %s is unavailable", p->pack_name);
801         if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
802                 return error("packfile %s signature is unavailable", p->pack_name);
803         idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
804         if (hashcmp(sha1, idx_sha1))
805                 return error("packfile %s does not match index", p->pack_name);
806         return 0;
807 }
808
809 static int open_packed_git(struct packed_git *p)
810 {
811         if (!open_packed_git_1(p))
812                 return 0;
813         if (p->pack_fd != -1) {
814                 close(p->pack_fd);
815                 pack_open_fds--;
816                 p->pack_fd = -1;
817         }
818         return -1;
819 }
820
821 static int in_window(struct pack_window *win, off_t offset)
822 {
823         /* We must promise at least 20 bytes (one hash) after the
824          * offset is available from this window, otherwise the offset
825          * is not actually in this window and a different window (which
826          * has that one hash excess) must be used.  This is to support
827          * the object header and delta base parsing routines below.
828          */
829         off_t win_off = win->offset;
830         return win_off <= offset
831                 && (offset + 20) <= (win_off + win->len);
832 }
833
834 unsigned char *use_pack(struct packed_git *p,
835                 struct pack_window **w_cursor,
836                 off_t offset,
837                 unsigned int *left)
838 {
839         struct pack_window *win = *w_cursor;
840
841         /* Since packfiles end in a hash of their content and it's
842          * pointless to ask for an offset into the middle of that
843          * hash, and the in_window function above wouldn't match
844          * don't allow an offset too close to the end of the file.
845          */
846         if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
847                 die("packfile %s cannot be accessed", p->pack_name);
848         if (offset > (p->pack_size - 20))
849                 die("offset beyond end of packfile (truncated pack?)");
850
851         if (!win || !in_window(win, offset)) {
852                 if (win)
853                         win->inuse_cnt--;
854                 for (win = p->windows; win; win = win->next) {
855                         if (in_window(win, offset))
856                                 break;
857                 }
858                 if (!win) {
859                         size_t window_align = packed_git_window_size / 2;
860                         off_t len;
861
862                         if (p->pack_fd == -1 && open_packed_git(p))
863                                 die("packfile %s cannot be accessed", p->pack_name);
864
865                         win = xcalloc(1, sizeof(*win));
866                         win->offset = (offset / window_align) * window_align;
867                         len = p->pack_size - win->offset;
868                         if (len > packed_git_window_size)
869                                 len = packed_git_window_size;
870                         win->len = (size_t)len;
871                         pack_mapped += win->len;
872                         while (packed_git_limit < pack_mapped
873                                 && unuse_one_window(p, p->pack_fd))
874                                 ; /* nothing */
875                         win->base = xmmap(NULL, win->len,
876                                 PROT_READ, MAP_PRIVATE,
877                                 p->pack_fd, win->offset);
878                         if (win->base == MAP_FAILED)
879                                 die("packfile %s cannot be mapped: %s",
880                                         p->pack_name,
881                                         strerror(errno));
882                         if (!win->offset && win->len == p->pack_size
883                                 && !p->do_not_close) {
884                                 close(p->pack_fd);
885                                 pack_open_fds--;
886                                 p->pack_fd = -1;
887                         }
888                         pack_mmap_calls++;
889                         pack_open_windows++;
890                         if (pack_mapped > peak_pack_mapped)
891                                 peak_pack_mapped = pack_mapped;
892                         if (pack_open_windows > peak_pack_open_windows)
893                                 peak_pack_open_windows = pack_open_windows;
894                         win->next = p->windows;
895                         p->windows = win;
896                 }
897         }
898         if (win != *w_cursor) {
899                 win->last_used = pack_used_ctr++;
900                 win->inuse_cnt++;
901                 *w_cursor = win;
902         }
903         offset -= win->offset;
904         if (left)
905                 *left = win->len - xsize_t(offset);
906         return win->base + offset;
907 }
908
909 static struct packed_git *alloc_packed_git(int extra)
910 {
911         struct packed_git *p = xmalloc(sizeof(*p) + extra);
912         memset(p, 0, sizeof(*p));
913         p->pack_fd = -1;
914         return p;
915 }
916
917 static void try_to_free_pack_memory(size_t size)
918 {
919         release_pack_memory(size, -1);
920 }
921
922 struct packed_git *add_packed_git(const char *path, int path_len, int local)
923 {
924         static int have_set_try_to_free_routine;
925         struct stat st;
926         struct packed_git *p = alloc_packed_git(path_len + 2);
927
928         if (!have_set_try_to_free_routine) {
929                 have_set_try_to_free_routine = 1;
930                 set_try_to_free_routine(try_to_free_pack_memory);
931         }
932
933         /*
934          * Make sure a corresponding .pack file exists and that
935          * the index looks sane.
936          */
937         path_len -= strlen(".idx");
938         if (path_len < 1) {
939                 free(p);
940                 return NULL;
941         }
942         memcpy(p->pack_name, path, path_len);
943
944         strcpy(p->pack_name + path_len, ".keep");
945         if (!access(p->pack_name, F_OK))
946                 p->pack_keep = 1;
947
948         strcpy(p->pack_name + path_len, ".pack");
949         if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
950                 free(p);
951                 return NULL;
952         }
953
954         /* ok, it looks sane as far as we can check without
955          * actually mapping the pack file.
956          */
957         p->pack_size = st.st_size;
958         p->pack_local = local;
959         p->mtime = st.st_mtime;
960         if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
961                 hashclr(p->sha1);
962         return p;
963 }
964
965 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
966 {
967         const char *path = sha1_pack_name(sha1);
968         struct packed_git *p = alloc_packed_git(strlen(path) + 1);
969
970         strcpy(p->pack_name, path);
971         hashcpy(p->sha1, sha1);
972         if (check_packed_git_idx(idx_path, p)) {
973                 free(p);
974                 return NULL;
975         }
976
977         return p;
978 }
979
980 void install_packed_git(struct packed_git *pack)
981 {
982         if (pack->pack_fd != -1)
983                 pack_open_fds++;
984
985         pack->next = packed_git;
986         packed_git = pack;
987 }
988
989 static void prepare_packed_git_one(char *objdir, int local)
990 {
991         /* Ensure that this buffer is large enough so that we can
992            append "/pack/" without clobbering the stack even if
993            strlen(objdir) were PATH_MAX.  */
994         char path[PATH_MAX + 1 + 4 + 1 + 1];
995         int len;
996         DIR *dir;
997         struct dirent *de;
998
999         sprintf(path, "%s/pack", objdir);
1000         len = strlen(path);
1001         dir = opendir(path);
1002         if (!dir) {
1003                 if (errno != ENOENT)
1004                         error("unable to open object pack directory: %s: %s",
1005                               path, strerror(errno));
1006                 return;
1007         }
1008         path[len++] = '/';
1009         while ((de = readdir(dir)) != NULL) {
1010                 int namelen = strlen(de->d_name);
1011                 struct packed_git *p;
1012
1013                 if (!has_extension(de->d_name, ".idx"))
1014                         continue;
1015
1016                 if (len + namelen + 1 > sizeof(path))
1017                         continue;
1018
1019                 /* Don't reopen a pack we already have. */
1020                 strcpy(path + len, de->d_name);
1021                 for (p = packed_git; p; p = p->next) {
1022                         if (!memcmp(path, p->pack_name, len + namelen - 4))
1023                                 break;
1024                 }
1025                 if (p)
1026                         continue;
1027                 /* See if it really is a valid .idx file with corresponding
1028                  * .pack file that we can map.
1029                  */
1030                 p = add_packed_git(path, len + namelen, local);
1031                 if (!p)
1032                         continue;
1033                 install_packed_git(p);
1034         }
1035         closedir(dir);
1036 }
1037
1038 static int sort_pack(const void *a_, const void *b_)
1039 {
1040         struct packed_git *a = *((struct packed_git **)a_);
1041         struct packed_git *b = *((struct packed_git **)b_);
1042         int st;
1043
1044         /*
1045          * Local packs tend to contain objects specific to our
1046          * variant of the project than remote ones.  In addition,
1047          * remote ones could be on a network mounted filesystem.
1048          * Favor local ones for these reasons.
1049          */
1050         st = a->pack_local - b->pack_local;
1051         if (st)
1052                 return -st;
1053
1054         /*
1055          * Younger packs tend to contain more recent objects,
1056          * and more recent objects tend to get accessed more
1057          * often.
1058          */
1059         if (a->mtime < b->mtime)
1060                 return 1;
1061         else if (a->mtime == b->mtime)
1062                 return 0;
1063         return -1;
1064 }
1065
1066 static void rearrange_packed_git(void)
1067 {
1068         struct packed_git **ary, *p;
1069         int i, n;
1070
1071         for (n = 0, p = packed_git; p; p = p->next)
1072                 n++;
1073         if (n < 2)
1074                 return;
1075
1076         /* prepare an array of packed_git for easier sorting */
1077         ary = xcalloc(n, sizeof(struct packed_git *));
1078         for (n = 0, p = packed_git; p; p = p->next)
1079                 ary[n++] = p;
1080
1081         qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1082
1083         /* link them back again */
1084         for (i = 0; i < n - 1; i++)
1085                 ary[i]->next = ary[i + 1];
1086         ary[n - 1]->next = NULL;
1087         packed_git = ary[0];
1088
1089         free(ary);
1090 }
1091
1092 static int prepare_packed_git_run_once = 0;
1093 void prepare_packed_git(void)
1094 {
1095         struct alternate_object_database *alt;
1096
1097         if (prepare_packed_git_run_once)
1098                 return;
1099         prepare_packed_git_one(get_object_directory(), 1);
1100         prepare_alt_odb();
1101         for (alt = alt_odb_list; alt; alt = alt->next) {
1102                 alt->name[-1] = 0;
1103                 prepare_packed_git_one(alt->base, 0);
1104                 alt->name[-1] = '/';
1105         }
1106         rearrange_packed_git();
1107         prepare_packed_git_run_once = 1;
1108 }
1109
1110 void reprepare_packed_git(void)
1111 {
1112         discard_revindex();
1113         prepare_packed_git_run_once = 0;
1114         prepare_packed_git();
1115 }
1116
1117 static void mark_bad_packed_object(struct packed_git *p,
1118                                    const unsigned char *sha1)
1119 {
1120         unsigned i;
1121         for (i = 0; i < p->num_bad_objects; i++)
1122                 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1123                         return;
1124         p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1125         hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1126         p->num_bad_objects++;
1127 }
1128
1129 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1130 {
1131         struct packed_git *p;
1132         unsigned i;
1133
1134         for (p = packed_git; p; p = p->next)
1135                 for (i = 0; i < p->num_bad_objects; i++)
1136                         if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1137                                 return p;
1138         return NULL;
1139 }
1140
1141 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1142 {
1143         unsigned char real_sha1[20];
1144         hash_sha1_file(map, size, type, real_sha1);
1145         return hashcmp(sha1, real_sha1) ? -1 : 0;
1146 }
1147
1148 static int git_open_noatime(const char *name)
1149 {
1150         static int sha1_file_open_flag = O_NOATIME;
1151
1152         for (;;) {
1153                 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1154                 if (fd >= 0)
1155                         return fd;
1156
1157                 /* Might the failure be due to O_NOATIME? */
1158                 if (errno != ENOENT && sha1_file_open_flag) {
1159                         sha1_file_open_flag = 0;
1160                         continue;
1161                 }
1162
1163                 return -1;
1164         }
1165 }
1166
1167 static int open_sha1_file(const unsigned char *sha1)
1168 {
1169         int fd;
1170         char *name = sha1_file_name(sha1);
1171         struct alternate_object_database *alt;
1172
1173         fd = git_open_noatime(name);
1174         if (fd >= 0)
1175                 return fd;
1176
1177         prepare_alt_odb();
1178         errno = ENOENT;
1179         for (alt = alt_odb_list; alt; alt = alt->next) {
1180                 name = alt->name;
1181                 fill_sha1_path(name, sha1);
1182                 fd = git_open_noatime(alt->base);
1183                 if (fd >= 0)
1184                         return fd;
1185         }
1186         return -1;
1187 }
1188
1189 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1190 {
1191         void *map;
1192         int fd;
1193
1194         fd = open_sha1_file(sha1);
1195         map = NULL;
1196         if (fd >= 0) {
1197                 struct stat st;
1198
1199                 if (!fstat(fd, &st)) {
1200                         *size = xsize_t(st.st_size);
1201                         map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1202                 }
1203                 close(fd);
1204         }
1205         return map;
1206 }
1207
1208 /*
1209  * There used to be a second loose object header format which
1210  * was meant to mimic the in-pack format, allowing for direct
1211  * copy of the object data.  This format turned up not to be
1212  * really worth it and we no longer write loose objects in that
1213  * format.
1214  */
1215 static int experimental_loose_object(unsigned char *map)
1216 {
1217         unsigned int word;
1218
1219         /*
1220          * Is it a zlib-compressed buffer? If so, the first byte
1221          * must be 0x78 (15-bit window size, deflated), and the
1222          * first 16-bit word is evenly divisible by 31. If so,
1223          * we are looking at the official format, not the experimental
1224          * one.
1225          */
1226         word = (map[0] << 8) + map[1];
1227         if (map[0] == 0x78 && !(word % 31))
1228                 return 0;
1229         else
1230                 return 1;
1231 }
1232
1233 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1234                 unsigned long len, enum object_type *type, unsigned long *sizep)
1235 {
1236         unsigned shift;
1237         unsigned long size, c;
1238         unsigned long used = 0;
1239
1240         c = buf[used++];
1241         *type = (c >> 4) & 7;
1242         size = c & 15;
1243         shift = 4;
1244         while (c & 0x80) {
1245                 if (len <= used || bitsizeof(long) <= shift) {
1246                         error("bad object header");
1247                         return 0;
1248                 }
1249                 c = buf[used++];
1250                 size += (c & 0x7f) << shift;
1251                 shift += 7;
1252         }
1253         *sizep = size;
1254         return used;
1255 }
1256
1257 int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1258 {
1259         unsigned long size, used;
1260         static const char valid_loose_object_type[8] = {
1261                 0, /* OBJ_EXT */
1262                 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1263                 0, /* "delta" and others are invalid in a loose object */
1264         };
1265         enum object_type type;
1266
1267         /* Get the data stream */
1268         memset(stream, 0, sizeof(*stream));
1269         stream->next_in = map;
1270         stream->avail_in = mapsize;
1271         stream->next_out = buffer;
1272         stream->avail_out = bufsiz;
1273
1274         if (experimental_loose_object(map)) {
1275                 /*
1276                  * The old experimental format we no longer produce;
1277                  * we can still read it.
1278                  */
1279                 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1280                 if (!used || !valid_loose_object_type[type])
1281                         return -1;
1282                 map += used;
1283                 mapsize -= used;
1284
1285                 /* Set up the stream for the rest.. */
1286                 stream->next_in = map;
1287                 stream->avail_in = mapsize;
1288                 git_inflate_init(stream);
1289
1290                 /* And generate the fake traditional header */
1291                 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1292                                                  typename(type), size);
1293                 return 0;
1294         }
1295         git_inflate_init(stream);
1296         return git_inflate(stream, 0);
1297 }
1298
1299 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1300 {
1301         int bytes = strlen(buffer) + 1;
1302         unsigned char *buf = xmallocz(size);
1303         unsigned long n;
1304         int status = Z_OK;
1305
1306         n = stream->total_out - bytes;
1307         if (n > size)
1308                 n = size;
1309         memcpy(buf, (char *) buffer + bytes, n);
1310         bytes = n;
1311         if (bytes <= size) {
1312                 /*
1313                  * The above condition must be (bytes <= size), not
1314                  * (bytes < size).  In other words, even though we
1315                  * expect no more output and set avail_out to zero,
1316                  * the input zlib stream may have bytes that express
1317                  * "this concludes the stream", and we *do* want to
1318                  * eat that input.
1319                  *
1320                  * Otherwise we would not be able to test that we
1321                  * consumed all the input to reach the expected size;
1322                  * we also want to check that zlib tells us that all
1323                  * went well with status == Z_STREAM_END at the end.
1324                  */
1325                 stream->next_out = buf + bytes;
1326                 stream->avail_out = size - bytes;
1327                 while (status == Z_OK)
1328                         status = git_inflate(stream, Z_FINISH);
1329         }
1330         if (status == Z_STREAM_END && !stream->avail_in) {
1331                 git_inflate_end(stream);
1332                 return buf;
1333         }
1334
1335         if (status < 0)
1336                 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1337         else if (stream->avail_in)
1338                 error("garbage at end of loose object '%s'",
1339                       sha1_to_hex(sha1));
1340         free(buf);
1341         return NULL;
1342 }
1343
1344 /*
1345  * We used to just use "sscanf()", but that's actually way
1346  * too permissive for what we want to check. So do an anal
1347  * object header parse by hand.
1348  */
1349 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1350 {
1351         char type[10];
1352         int i;
1353         unsigned long size;
1354
1355         /*
1356          * The type can be at most ten bytes (including the
1357          * terminating '\0' that we add), and is followed by
1358          * a space.
1359          */
1360         i = 0;
1361         for (;;) {
1362                 char c = *hdr++;
1363                 if (c == ' ')
1364                         break;
1365                 type[i++] = c;
1366                 if (i >= sizeof(type))
1367                         return -1;
1368         }
1369         type[i] = 0;
1370
1371         /*
1372          * The length must follow immediately, and be in canonical
1373          * decimal format (ie "010" is not valid).
1374          */
1375         size = *hdr++ - '0';
1376         if (size > 9)
1377                 return -1;
1378         if (size) {
1379                 for (;;) {
1380                         unsigned long c = *hdr - '0';
1381                         if (c > 9)
1382                                 break;
1383                         hdr++;
1384                         size = size * 10 + c;
1385                 }
1386         }
1387         *sizep = size;
1388
1389         /*
1390          * The length must be followed by a zero byte
1391          */
1392         return *hdr ? -1 : type_from_string(type);
1393 }
1394
1395 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1396 {
1397         int ret;
1398         z_stream stream;
1399         char hdr[8192];
1400
1401         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1402         if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1403                 return NULL;
1404
1405         return unpack_sha1_rest(&stream, hdr, *size, sha1);
1406 }
1407
1408 unsigned long get_size_from_delta(struct packed_git *p,
1409                                   struct pack_window **w_curs,
1410                                   off_t curpos)
1411 {
1412         const unsigned char *data;
1413         unsigned char delta_head[20], *in;
1414         z_stream stream;
1415         int st;
1416
1417         memset(&stream, 0, sizeof(stream));
1418         stream.next_out = delta_head;
1419         stream.avail_out = sizeof(delta_head);
1420
1421         git_inflate_init(&stream);
1422         do {
1423                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1424                 stream.next_in = in;
1425                 st = git_inflate(&stream, Z_FINISH);
1426                 curpos += stream.next_in - in;
1427         } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1428                  stream.total_out < sizeof(delta_head));
1429         git_inflate_end(&stream);
1430         if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1431                 error("delta data unpack-initial failed");
1432                 return 0;
1433         }
1434
1435         /* Examine the initial part of the delta to figure out
1436          * the result size.
1437          */
1438         data = delta_head;
1439
1440         /* ignore base size */
1441         get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1442
1443         /* Read the result size */
1444         return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1445 }
1446
1447 static off_t get_delta_base(struct packed_git *p,
1448                                     struct pack_window **w_curs,
1449                                     off_t *curpos,
1450                                     enum object_type type,
1451                                     off_t delta_obj_offset)
1452 {
1453         unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1454         off_t base_offset;
1455
1456         /* use_pack() assured us we have [base_info, base_info + 20)
1457          * as a range that we can look at without walking off the
1458          * end of the mapped window.  Its actually the hash size
1459          * that is assured.  An OFS_DELTA longer than the hash size
1460          * is stupid, as then a REF_DELTA would be smaller to store.
1461          */
1462         if (type == OBJ_OFS_DELTA) {
1463                 unsigned used = 0;
1464                 unsigned char c = base_info[used++];
1465                 base_offset = c & 127;
1466                 while (c & 128) {
1467                         base_offset += 1;
1468                         if (!base_offset || MSB(base_offset, 7))
1469                                 return 0;  /* overflow */
1470                         c = base_info[used++];
1471                         base_offset = (base_offset << 7) + (c & 127);
1472                 }
1473                 base_offset = delta_obj_offset - base_offset;
1474                 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1475                         return 0;  /* out of bound */
1476                 *curpos += used;
1477         } else if (type == OBJ_REF_DELTA) {
1478                 /* The base entry _must_ be in the same pack */
1479                 base_offset = find_pack_entry_one(base_info, p);
1480                 *curpos += 20;
1481         } else
1482                 die("I am totally screwed");
1483         return base_offset;
1484 }
1485
1486 /* forward declaration for a mutually recursive function */
1487 static int packed_object_info(struct packed_git *p, off_t offset,
1488                               unsigned long *sizep, int *rtype);
1489
1490 static int packed_delta_info(struct packed_git *p,
1491                              struct pack_window **w_curs,
1492                              off_t curpos,
1493                              enum object_type type,
1494                              off_t obj_offset,
1495                              unsigned long *sizep)
1496 {
1497         off_t base_offset;
1498
1499         base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1500         if (!base_offset)
1501                 return OBJ_BAD;
1502         type = packed_object_info(p, base_offset, NULL, NULL);
1503         if (type <= OBJ_NONE) {
1504                 struct revindex_entry *revidx;
1505                 const unsigned char *base_sha1;
1506                 revidx = find_pack_revindex(p, base_offset);
1507                 if (!revidx)
1508                         return OBJ_BAD;
1509                 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1510                 mark_bad_packed_object(p, base_sha1);
1511                 type = sha1_object_info(base_sha1, NULL);
1512                 if (type <= OBJ_NONE)
1513                         return OBJ_BAD;
1514         }
1515
1516         /* We choose to only get the type of the base object and
1517          * ignore potentially corrupt pack file that expects the delta
1518          * based on a base with a wrong size.  This saves tons of
1519          * inflate() calls.
1520          */
1521         if (sizep) {
1522                 *sizep = get_size_from_delta(p, w_curs, curpos);
1523                 if (*sizep == 0)
1524                         type = OBJ_BAD;
1525         }
1526
1527         return type;
1528 }
1529
1530 int unpack_object_header(struct packed_git *p,
1531                          struct pack_window **w_curs,
1532                          off_t *curpos,
1533                          unsigned long *sizep)
1534 {
1535         unsigned char *base;
1536         unsigned int left;
1537         unsigned long used;
1538         enum object_type type;
1539
1540         /* use_pack() assures us we have [base, base + 20) available
1541          * as a range that we can look at.  (Its actually the hash
1542          * size that is assured.)  With our object header encoding
1543          * the maximum deflated object size is 2^137, which is just
1544          * insane, so we know won't exceed what we have been given.
1545          */
1546         base = use_pack(p, w_curs, *curpos, &left);
1547         used = unpack_object_header_buffer(base, left, &type, sizep);
1548         if (!used) {
1549                 type = OBJ_BAD;
1550         } else
1551                 *curpos += used;
1552
1553         return type;
1554 }
1555
1556 int packed_object_info_detail(struct packed_git *p,
1557                                       off_t obj_offset,
1558                                       unsigned long *size,
1559                                       unsigned long *store_size,
1560                                       unsigned int *delta_chain_length,
1561                                       unsigned char *base_sha1)
1562 {
1563         struct pack_window *w_curs = NULL;
1564         off_t curpos;
1565         unsigned long dummy;
1566         unsigned char *next_sha1;
1567         enum object_type type;
1568         struct revindex_entry *revidx;
1569
1570         *delta_chain_length = 0;
1571         curpos = obj_offset;
1572         type = unpack_object_header(p, &w_curs, &curpos, size);
1573
1574         revidx = find_pack_revindex(p, obj_offset);
1575         *store_size = revidx[1].offset - obj_offset;
1576
1577         for (;;) {
1578                 switch (type) {
1579                 default:
1580                         die("pack %s contains unknown object type %d",
1581                             p->pack_name, type);
1582                 case OBJ_COMMIT:
1583                 case OBJ_TREE:
1584                 case OBJ_BLOB:
1585                 case OBJ_TAG:
1586                         unuse_pack(&w_curs);
1587                         return type;
1588                 case OBJ_OFS_DELTA:
1589                         obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1590                         if (!obj_offset)
1591                                 die("pack %s contains bad delta base reference of type %s",
1592                                     p->pack_name, typename(type));
1593                         if (*delta_chain_length == 0) {
1594                                 revidx = find_pack_revindex(p, obj_offset);
1595                                 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1596                         }
1597                         break;
1598                 case OBJ_REF_DELTA:
1599                         next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1600                         if (*delta_chain_length == 0)
1601                                 hashcpy(base_sha1, next_sha1);
1602                         obj_offset = find_pack_entry_one(next_sha1, p);
1603                         break;
1604                 }
1605                 (*delta_chain_length)++;
1606                 curpos = obj_offset;
1607                 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1608         }
1609 }
1610
1611 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1612                               unsigned long *sizep, int *rtype)
1613 {
1614         struct pack_window *w_curs = NULL;
1615         unsigned long size;
1616         off_t curpos = obj_offset;
1617         enum object_type type;
1618
1619         type = unpack_object_header(p, &w_curs, &curpos, &size);
1620         if (rtype)
1621                 *rtype = type; /* representation type */
1622
1623         switch (type) {
1624         case OBJ_OFS_DELTA:
1625         case OBJ_REF_DELTA:
1626                 type = packed_delta_info(p, &w_curs, curpos,
1627                                          type, obj_offset, sizep);
1628                 break;
1629         case OBJ_COMMIT:
1630         case OBJ_TREE:
1631         case OBJ_BLOB:
1632         case OBJ_TAG:
1633                 if (sizep)
1634                         *sizep = size;
1635                 break;
1636         default:
1637                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1638                       type, (uintmax_t)obj_offset, p->pack_name);
1639                 type = OBJ_BAD;
1640         }
1641         unuse_pack(&w_curs);
1642         return type;
1643 }
1644
1645 static void *unpack_compressed_entry(struct packed_git *p,
1646                                     struct pack_window **w_curs,
1647                                     off_t curpos,
1648                                     unsigned long size)
1649 {
1650         int st;
1651         z_stream stream;
1652         unsigned char *buffer, *in;
1653
1654         buffer = xmallocz(size);
1655         memset(&stream, 0, sizeof(stream));
1656         stream.next_out = buffer;
1657         stream.avail_out = size + 1;
1658
1659         git_inflate_init(&stream);
1660         do {
1661                 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1662                 stream.next_in = in;
1663                 st = git_inflate(&stream, Z_FINISH);
1664                 if (!stream.avail_out)
1665                         break; /* the payload is larger than it should be */
1666                 curpos += stream.next_in - in;
1667         } while (st == Z_OK || st == Z_BUF_ERROR);
1668         git_inflate_end(&stream);
1669         if ((st != Z_STREAM_END) || stream.total_out != size) {
1670                 free(buffer);
1671                 return NULL;
1672         }
1673
1674         return buffer;
1675 }
1676
1677 #define MAX_DELTA_CACHE (256)
1678
1679 static size_t delta_base_cached;
1680
1681 static struct delta_base_cache_lru_list {
1682         struct delta_base_cache_lru_list *prev;
1683         struct delta_base_cache_lru_list *next;
1684 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1685
1686 static struct delta_base_cache_entry {
1687         struct delta_base_cache_lru_list lru;
1688         void *data;
1689         struct packed_git *p;
1690         off_t base_offset;
1691         unsigned long size;
1692         enum object_type type;
1693 } delta_base_cache[MAX_DELTA_CACHE];
1694
1695 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1696 {
1697         unsigned long hash;
1698
1699         hash = (unsigned long)p + (unsigned long)base_offset;
1700         hash += (hash >> 8) + (hash >> 16);
1701         return hash % MAX_DELTA_CACHE;
1702 }
1703
1704 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1705 {
1706         unsigned long hash = pack_entry_hash(p, base_offset);
1707         struct delta_base_cache_entry *ent = delta_base_cache + hash;
1708         return (ent->data && ent->p == p && ent->base_offset == base_offset);
1709 }
1710
1711 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1712         unsigned long *base_size, enum object_type *type, int keep_cache)
1713 {
1714         void *ret;
1715         unsigned long hash = pack_entry_hash(p, base_offset);
1716         struct delta_base_cache_entry *ent = delta_base_cache + hash;
1717
1718         ret = ent->data;
1719         if (!ret || ent->p != p || ent->base_offset != base_offset)
1720                 return unpack_entry(p, base_offset, type, base_size);
1721
1722         if (!keep_cache) {
1723                 ent->data = NULL;
1724                 ent->lru.next->prev = ent->lru.prev;
1725                 ent->lru.prev->next = ent->lru.next;
1726                 delta_base_cached -= ent->size;
1727         } else {
1728                 ret = xmemdupz(ent->data, ent->size);
1729         }
1730         *type = ent->type;
1731         *base_size = ent->size;
1732         return ret;
1733 }
1734
1735 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1736 {
1737         if (ent->data) {
1738                 free(ent->data);
1739                 ent->data = NULL;
1740                 ent->lru.next->prev = ent->lru.prev;
1741                 ent->lru.prev->next = ent->lru.next;
1742                 delta_base_cached -= ent->size;
1743         }
1744 }
1745
1746 void clear_delta_base_cache(void)
1747 {
1748         unsigned long p;
1749         for (p = 0; p < MAX_DELTA_CACHE; p++)
1750                 release_delta_base_cache(&delta_base_cache[p]);
1751 }
1752
1753 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1754         void *base, unsigned long base_size, enum object_type type)
1755 {
1756         unsigned long hash = pack_entry_hash(p, base_offset);
1757         struct delta_base_cache_entry *ent = delta_base_cache + hash;
1758         struct delta_base_cache_lru_list *lru;
1759
1760         release_delta_base_cache(ent);
1761         delta_base_cached += base_size;
1762
1763         for (lru = delta_base_cache_lru.next;
1764              delta_base_cached > delta_base_cache_limit
1765              && lru != &delta_base_cache_lru;
1766              lru = lru->next) {
1767                 struct delta_base_cache_entry *f = (void *)lru;
1768                 if (f->type == OBJ_BLOB)
1769                         release_delta_base_cache(f);
1770         }
1771         for (lru = delta_base_cache_lru.next;
1772              delta_base_cached > delta_base_cache_limit
1773              && lru != &delta_base_cache_lru;
1774              lru = lru->next) {
1775                 struct delta_base_cache_entry *f = (void *)lru;
1776                 release_delta_base_cache(f);
1777         }
1778
1779         ent->p = p;
1780         ent->base_offset = base_offset;
1781         ent->type = type;
1782         ent->data = base;
1783         ent->size = base_size;
1784         ent->lru.next = &delta_base_cache_lru;
1785         ent->lru.prev = delta_base_cache_lru.prev;
1786         delta_base_cache_lru.prev->next = &ent->lru;
1787         delta_base_cache_lru.prev = &ent->lru;
1788 }
1789
1790 static void *read_object(const unsigned char *sha1, enum object_type *type,
1791                          unsigned long *size);
1792
1793 static void *unpack_delta_entry(struct packed_git *p,
1794                                 struct pack_window **w_curs,
1795                                 off_t curpos,
1796                                 unsigned long delta_size,
1797                                 off_t obj_offset,
1798                                 enum object_type *type,
1799                                 unsigned long *sizep)
1800 {
1801         void *delta_data, *result, *base;
1802         unsigned long base_size;
1803         off_t base_offset;
1804
1805         base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1806         if (!base_offset) {
1807                 error("failed to validate delta base reference "
1808                       "at offset %"PRIuMAX" from %s",
1809                       (uintmax_t)curpos, p->pack_name);
1810                 return NULL;
1811         }
1812         unuse_pack(w_curs);
1813         base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1814         if (!base) {
1815                 /*
1816                  * We're probably in deep shit, but let's try to fetch
1817                  * the required base anyway from another pack or loose.
1818                  * This is costly but should happen only in the presence
1819                  * of a corrupted pack, and is better than failing outright.
1820                  */
1821                 struct revindex_entry *revidx;
1822                 const unsigned char *base_sha1;
1823                 revidx = find_pack_revindex(p, base_offset);
1824                 if (!revidx)
1825                         return NULL;
1826                 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1827                 error("failed to read delta base object %s"
1828                       " at offset %"PRIuMAX" from %s",
1829                       sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1830                       p->pack_name);
1831                 mark_bad_packed_object(p, base_sha1);
1832                 base = read_object(base_sha1, type, &base_size);
1833                 if (!base)
1834                         return NULL;
1835         }
1836
1837         delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1838         if (!delta_data) {
1839                 error("failed to unpack compressed delta "
1840                       "at offset %"PRIuMAX" from %s",
1841                       (uintmax_t)curpos, p->pack_name);
1842                 free(base);
1843                 return NULL;
1844         }
1845         result = patch_delta(base, base_size,
1846                              delta_data, delta_size,
1847                              sizep);
1848         if (!result)
1849                 die("failed to apply delta");
1850         free(delta_data);
1851         add_delta_base_cache(p, base_offset, base, base_size, *type);
1852         return result;
1853 }
1854
1855 int do_check_packed_object_crc;
1856
1857 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1858                    enum object_type *type, unsigned long *sizep)
1859 {
1860         struct pack_window *w_curs = NULL;
1861         off_t curpos = obj_offset;
1862         void *data;
1863
1864         if (do_check_packed_object_crc && p->index_version > 1) {
1865                 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1866                 unsigned long len = revidx[1].offset - obj_offset;
1867                 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1868                         const unsigned char *sha1 =
1869                                 nth_packed_object_sha1(p, revidx->nr);
1870                         error("bad packed object CRC for %s",
1871                               sha1_to_hex(sha1));
1872                         mark_bad_packed_object(p, sha1);
1873                         unuse_pack(&w_curs);
1874                         return NULL;
1875                 }
1876         }
1877
1878         *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1879         switch (*type) {
1880         case OBJ_OFS_DELTA:
1881         case OBJ_REF_DELTA:
1882                 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1883                                           obj_offset, type, sizep);
1884                 break;
1885         case OBJ_COMMIT:
1886         case OBJ_TREE:
1887         case OBJ_BLOB:
1888         case OBJ_TAG:
1889                 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1890                 break;
1891         default:
1892                 data = NULL;
1893                 error("unknown object type %i at offset %"PRIuMAX" in %s",
1894                       *type, (uintmax_t)obj_offset, p->pack_name);
1895         }
1896         unuse_pack(&w_curs);
1897         return data;
1898 }
1899
1900 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1901                                             uint32_t n)
1902 {
1903         const unsigned char *index = p->index_data;
1904         if (!index) {
1905                 if (open_pack_index(p))
1906                         return NULL;
1907                 index = p->index_data;
1908         }
1909         if (n >= p->num_objects)
1910                 return NULL;
1911         index += 4 * 256;
1912         if (p->index_version == 1) {
1913                 return index + 24 * n + 4;
1914         } else {
1915                 index += 8;
1916                 return index + 20 * n;
1917         }
1918 }
1919
1920 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1921 {
1922         const unsigned char *index = p->index_data;
1923         index += 4 * 256;
1924         if (p->index_version == 1) {
1925                 return ntohl(*((uint32_t *)(index + 24 * n)));
1926         } else {
1927                 uint32_t off;
1928                 index += 8 + p->num_objects * (20 + 4);
1929                 off = ntohl(*((uint32_t *)(index + 4 * n)));
1930                 if (!(off & 0x80000000))
1931                         return off;
1932                 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1933                 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1934                                    ntohl(*((uint32_t *)(index + 4)));
1935         }
1936 }
1937
1938 off_t find_pack_entry_one(const unsigned char *sha1,
1939                                   struct packed_git *p)
1940 {
1941         const uint32_t *level1_ofs = p->index_data;
1942         const unsigned char *index = p->index_data;
1943         unsigned hi, lo, stride;
1944         static int use_lookup = -1;
1945         static int debug_lookup = -1;
1946
1947         if (debug_lookup < 0)
1948                 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1949
1950         if (!index) {
1951                 if (open_pack_index(p))
1952                         return 0;
1953                 level1_ofs = p->index_data;
1954                 index = p->index_data;
1955         }
1956         if (p->index_version > 1) {
1957                 level1_ofs += 2;
1958                 index += 8;
1959         }
1960         index += 4 * 256;
1961         hi = ntohl(level1_ofs[*sha1]);
1962         lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1963         if (p->index_version > 1) {
1964                 stride = 20;
1965         } else {
1966                 stride = 24;
1967                 index += 4;
1968         }
1969
1970         if (debug_lookup)
1971                 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1972                        sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1973
1974         if (use_lookup < 0)
1975                 use_lookup = !!getenv("GIT_USE_LOOKUP");
1976         if (use_lookup) {
1977                 int pos = sha1_entry_pos(index, stride, 0,
1978                                          lo, hi, p->num_objects, sha1);
1979                 if (pos < 0)
1980                         return 0;
1981                 return nth_packed_object_offset(p, pos);
1982         }
1983
1984         do {
1985                 unsigned mi = (lo + hi) / 2;
1986                 int cmp = hashcmp(index + mi * stride, sha1);
1987
1988                 if (debug_lookup)
1989                         printf("lo %u hi %u rg %u mi %u\n",
1990                                lo, hi, hi - lo, mi);
1991                 if (!cmp)
1992                         return nth_packed_object_offset(p, mi);
1993                 if (cmp > 0)
1994                         hi = mi;
1995                 else
1996                         lo = mi+1;
1997         } while (lo < hi);
1998         return 0;
1999 }
2000
2001 static int is_pack_valid(struct packed_git *p)
2002 {
2003         /* An already open pack is known to be valid. */
2004         if (p->pack_fd != -1)
2005                 return 1;
2006
2007         /* If the pack has one window completely covering the
2008          * file size, the pack is known to be valid even if
2009          * the descriptor is not currently open.
2010          */
2011         if (p->windows) {
2012                 struct pack_window *w = p->windows;
2013
2014                 if (!w->offset && w->len == p->pack_size)
2015                         return 1;
2016         }
2017
2018         /* Force the pack to open to prove its valid. */
2019         return !open_packed_git(p);
2020 }
2021
2022 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2023 {
2024         static struct packed_git *last_found = (void *)1;
2025         struct packed_git *p;
2026         off_t offset;
2027
2028         prepare_packed_git();
2029         if (!packed_git)
2030                 return 0;
2031         p = (last_found == (void *)1) ? packed_git : last_found;
2032
2033         do {
2034                 if (p->num_bad_objects) {
2035                         unsigned i;
2036                         for (i = 0; i < p->num_bad_objects; i++)
2037                                 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2038                                         goto next;
2039                 }
2040
2041                 offset = find_pack_entry_one(sha1, p);
2042                 if (offset) {
2043                         /*
2044                          * We are about to tell the caller where they can
2045                          * locate the requested object.  We better make
2046                          * sure the packfile is still here and can be
2047                          * accessed before supplying that answer, as
2048                          * it may have been deleted since the index
2049                          * was loaded!
2050                          */
2051                         if (!is_pack_valid(p)) {
2052                                 error("packfile %s cannot be accessed", p->pack_name);
2053                                 goto next;
2054                         }
2055                         e->offset = offset;
2056                         e->p = p;
2057                         hashcpy(e->sha1, sha1);
2058                         last_found = p;
2059                         return 1;
2060                 }
2061
2062                 next:
2063                 if (p == last_found)
2064                         p = packed_git;
2065                 else
2066                         p = p->next;
2067                 if (p == last_found)
2068                         p = p->next;
2069         } while (p);
2070         return 0;
2071 }
2072
2073 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2074                                   struct packed_git *packs)
2075 {
2076         struct packed_git *p;
2077
2078         for (p = packs; p; p = p->next) {
2079                 if (find_pack_entry_one(sha1, p))
2080                         return p;
2081         }
2082         return NULL;
2083
2084 }
2085
2086 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2087 {
2088         int status;
2089         unsigned long mapsize, size;
2090         void *map;
2091         z_stream stream;
2092         char hdr[32];
2093
2094         map = map_sha1_file(sha1, &mapsize);
2095         if (!map)
2096                 return error("unable to find %s", sha1_to_hex(sha1));
2097         if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2098                 status = error("unable to unpack %s header",
2099                                sha1_to_hex(sha1));
2100         else if ((status = parse_sha1_header(hdr, &size)) < 0)
2101                 status = error("unable to parse %s header", sha1_to_hex(sha1));
2102         else if (sizep)
2103                 *sizep = size;
2104         git_inflate_end(&stream);
2105         munmap(map, mapsize);
2106         return status;
2107 }
2108
2109 /* returns enum object_type or negative */
2110 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2111 {
2112         struct cached_object *co;
2113         struct pack_entry e;
2114         int status, rtype;
2115
2116         co = find_cached_object(sha1);
2117         if (co) {
2118                 if (oi->sizep)
2119                         *(oi->sizep) = co->size;
2120                 oi->whence = OI_CACHED;
2121                 return co->type;
2122         }
2123
2124         if (!find_pack_entry(sha1, &e)) {
2125                 /* Most likely it's a loose object. */
2126                 status = sha1_loose_object_info(sha1, oi->sizep);
2127                 if (status >= 0) {
2128                         oi->whence = OI_LOOSE;
2129                         return status;
2130                 }
2131
2132                 /* Not a loose object; someone else may have just packed it. */
2133                 reprepare_packed_git();
2134                 if (!find_pack_entry(sha1, &e))
2135                         return status;
2136         }
2137
2138         status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2139         if (status < 0) {
2140                 mark_bad_packed_object(e.p, sha1);
2141                 status = sha1_object_info_extended(sha1, oi);
2142         } else if (in_delta_base_cache(e.p, e.offset)) {
2143                 oi->whence = OI_DBCACHED;
2144         } else {
2145                 oi->whence = OI_PACKED;
2146                 oi->u.packed.offset = e.offset;
2147                 oi->u.packed.pack = e.p;
2148                 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2149                                          rtype == OBJ_OFS_DELTA);
2150         }
2151
2152         return status;
2153 }
2154
2155 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2156 {
2157         struct object_info oi;
2158
2159         oi.sizep = sizep;
2160         return sha1_object_info_extended(sha1, &oi);
2161 }
2162
2163 static void *read_packed_sha1(const unsigned char *sha1,
2164                               enum object_type *type, unsigned long *size)
2165 {
2166         struct pack_entry e;
2167         void *data;
2168
2169         if (!find_pack_entry(sha1, &e))
2170                 return NULL;
2171         data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2172         if (!data) {
2173                 /*
2174                  * We're probably in deep shit, but let's try to fetch
2175                  * the required object anyway from another pack or loose.
2176                  * This should happen only in the presence of a corrupted
2177                  * pack, and is better than failing outright.
2178                  */
2179                 error("failed to read object %s at offset %"PRIuMAX" from %s",
2180                       sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2181                 mark_bad_packed_object(e.p, sha1);
2182                 data = read_object(sha1, type, size);
2183         }
2184         return data;
2185 }
2186
2187 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2188                       unsigned char *sha1)
2189 {
2190         struct cached_object *co;
2191
2192         hash_sha1_file(buf, len, typename(type), sha1);
2193         if (has_sha1_file(sha1) || find_cached_object(sha1))
2194                 return 0;
2195         if (cached_object_alloc <= cached_object_nr) {
2196                 cached_object_alloc = alloc_nr(cached_object_alloc);
2197                 cached_objects = xrealloc(cached_objects,
2198                                           sizeof(*cached_objects) *
2199                                           cached_object_alloc);
2200         }
2201         co = &cached_objects[cached_object_nr++];
2202         co->size = len;
2203         co->type = type;
2204         co->buf = xmalloc(len);
2205         memcpy(co->buf, buf, len);
2206         hashcpy(co->sha1, sha1);
2207         return 0;
2208 }
2209
2210 static void *read_object(const unsigned char *sha1, enum object_type *type,
2211                          unsigned long *size)
2212 {
2213         unsigned long mapsize;
2214         void *map, *buf;
2215         struct cached_object *co;
2216
2217         co = find_cached_object(sha1);
2218         if (co) {
2219                 *type = co->type;
2220                 *size = co->size;
2221                 return xmemdupz(co->buf, co->size);
2222         }
2223
2224         buf = read_packed_sha1(sha1, type, size);
2225         if (buf)
2226                 return buf;
2227         map = map_sha1_file(sha1, &mapsize);
2228         if (map) {
2229                 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2230                 munmap(map, mapsize);
2231                 return buf;
2232         }
2233         reprepare_packed_git();
2234         return read_packed_sha1(sha1, type, size);
2235 }
2236
2237 /*
2238  * This function dies on corrupt objects; the callers who want to
2239  * deal with them should arrange to call read_object() and give error
2240  * messages themselves.
2241  */
2242 void *read_sha1_file_extended(const unsigned char *sha1,
2243                               enum object_type *type,
2244                               unsigned long *size,
2245                               unsigned flag)
2246 {
2247         void *data;
2248         char *path;
2249         const struct packed_git *p;
2250         const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2251                 ? lookup_replace_object(sha1) : sha1;
2252
2253         errno = 0;
2254         data = read_object(repl, type, size);
2255         if (data)
2256                 return data;
2257
2258         if (errno && errno != ENOENT)
2259                 die_errno("failed to read object %s", sha1_to_hex(sha1));
2260
2261         /* die if we replaced an object with one that does not exist */
2262         if (repl != sha1)
2263                 die("replacement %s not found for %s",
2264                     sha1_to_hex(repl), sha1_to_hex(sha1));
2265
2266         if (has_loose_object(repl)) {
2267                 path = sha1_file_name(sha1);
2268                 die("loose object %s (stored in %s) is corrupt",
2269                     sha1_to_hex(repl), path);
2270         }
2271
2272         if ((p = has_packed_and_bad(repl)) != NULL)
2273                 die("packed object %s (stored in %s) is corrupt",
2274                     sha1_to_hex(repl), p->pack_name);
2275
2276         return NULL;
2277 }
2278
2279 void *read_object_with_reference(const unsigned char *sha1,
2280                                  const char *required_type_name,
2281                                  unsigned long *size,
2282                                  unsigned char *actual_sha1_return)
2283 {
2284         enum object_type type, required_type;
2285         void *buffer;
2286         unsigned long isize;
2287         unsigned char actual_sha1[20];
2288
2289         required_type = type_from_string(required_type_name);
2290         hashcpy(actual_sha1, sha1);
2291         while (1) {
2292                 int ref_length = -1;
2293                 const char *ref_type = NULL;
2294
2295                 buffer = read_sha1_file(actual_sha1, &type, &isize);
2296                 if (!buffer)
2297                         return NULL;
2298                 if (type == required_type) {
2299                         *size = isize;
2300                         if (actual_sha1_return)
2301                                 hashcpy(actual_sha1_return, actual_sha1);
2302                         return buffer;
2303                 }
2304                 /* Handle references */
2305                 else if (type == OBJ_COMMIT)
2306                         ref_type = "tree ";
2307                 else if (type == OBJ_TAG)
2308                         ref_type = "object ";
2309                 else {
2310                         free(buffer);
2311                         return NULL;
2312                 }
2313                 ref_length = strlen(ref_type);
2314
2315                 if (ref_length + 40 > isize ||
2316                     memcmp(buffer, ref_type, ref_length) ||
2317                     get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2318                         free(buffer);
2319                         return NULL;
2320                 }
2321                 free(buffer);
2322                 /* Now we have the ID of the referred-to object in
2323                  * actual_sha1.  Check again. */
2324         }
2325 }
2326
2327 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2328                                     const char *type, unsigned char *sha1,
2329                                     char *hdr, int *hdrlen)
2330 {
2331         git_SHA_CTX c;
2332
2333         /* Generate the header */
2334         *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2335
2336         /* Sha1.. */
2337         git_SHA1_Init(&c);
2338         git_SHA1_Update(&c, hdr, *hdrlen);
2339         git_SHA1_Update(&c, buf, len);
2340         git_SHA1_Final(sha1, &c);
2341 }
2342
2343 /*
2344  * Move the just written object into its final resting place.
2345  * NEEDSWORK: this should be renamed to finalize_temp_file() as
2346  * "moving" is only a part of what it does, when no patch between
2347  * master to pu changes the call sites of this function.
2348  */
2349 int move_temp_to_file(const char *tmpfile, const char *filename)
2350 {
2351         int ret = 0;
2352
2353         if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2354                 goto try_rename;
2355         else if (link(tmpfile, filename))
2356                 ret = errno;
2357
2358         /*
2359          * Coda hack - coda doesn't like cross-directory links,
2360          * so we fall back to a rename, which will mean that it
2361          * won't be able to check collisions, but that's not a
2362          * big deal.
2363          *
2364          * The same holds for FAT formatted media.
2365          *
2366          * When this succeeds, we just return.  We have nothing
2367          * left to unlink.
2368          */
2369         if (ret && ret != EEXIST) {
2370         try_rename:
2371                 if (!rename(tmpfile, filename))
2372                         goto out;
2373                 ret = errno;
2374         }
2375         unlink_or_warn(tmpfile);
2376         if (ret) {
2377                 if (ret != EEXIST) {
2378                         return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2379                 }
2380                 /* FIXME!!! Collision check here ? */
2381         }
2382
2383 out:
2384         if (adjust_shared_perm(filename))
2385                 return error("unable to set permission to '%s'", filename);
2386         return 0;
2387 }
2388
2389 static int write_buffer(int fd, const void *buf, size_t len)
2390 {
2391         if (write_in_full(fd, buf, len) < 0)
2392                 return error("file write error (%s)", strerror(errno));
2393         return 0;
2394 }
2395
2396 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2397                    unsigned char *sha1)
2398 {
2399         char hdr[32];
2400         int hdrlen;
2401         write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2402         return 0;
2403 }
2404
2405 /* Finalize a file on disk, and close it. */
2406 static void close_sha1_file(int fd)
2407 {
2408         if (fsync_object_files)
2409                 fsync_or_die(fd, "sha1 file");
2410         if (close(fd) != 0)
2411                 die_errno("error when closing sha1 file");
2412 }
2413
2414 /* Size of directory component, including the ending '/' */
2415 static inline int directory_size(const char *filename)
2416 {
2417         const char *s = strrchr(filename, '/');
2418         if (!s)
2419                 return 0;
2420         return s - filename + 1;
2421 }
2422
2423 /*
2424  * This creates a temporary file in the same directory as the final
2425  * 'filename'
2426  *
2427  * We want to avoid cross-directory filename renames, because those
2428  * can have problems on various filesystems (FAT, NFS, Coda).
2429  */
2430 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2431 {
2432         int fd, dirlen = directory_size(filename);
2433
2434         if (dirlen + 20 > bufsiz) {
2435                 errno = ENAMETOOLONG;
2436                 return -1;
2437         }
2438         memcpy(buffer, filename, dirlen);
2439         strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2440         fd = git_mkstemp_mode(buffer, 0444);
2441         if (fd < 0 && dirlen && errno == ENOENT) {
2442                 /* Make sure the directory exists */
2443                 memcpy(buffer, filename, dirlen);
2444                 buffer[dirlen-1] = 0;
2445                 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2446                         return -1;
2447
2448                 /* Try again */
2449                 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2450                 fd = git_mkstemp_mode(buffer, 0444);
2451         }
2452         return fd;
2453 }
2454
2455 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2456                               const void *buf, unsigned long len, time_t mtime)
2457 {
2458         int fd, ret;
2459         unsigned char compressed[4096];
2460         z_stream stream;
2461         git_SHA_CTX c;
2462         unsigned char parano_sha1[20];
2463         char *filename;
2464         static char tmpfile[PATH_MAX];
2465
2466         filename = sha1_file_name(sha1);
2467         fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2468         if (fd < 0) {
2469                 if (errno == EACCES)
2470                         return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2471                 else
2472                         return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2473         }
2474
2475         /* Set it up */
2476         memset(&stream, 0, sizeof(stream));
2477         deflateInit(&stream, zlib_compression_level);
2478         stream.next_out = compressed;
2479         stream.avail_out = sizeof(compressed);
2480         git_SHA1_Init(&c);
2481
2482         /* First header.. */
2483         stream.next_in = (unsigned char *)hdr;
2484         stream.avail_in = hdrlen;
2485         while (deflate(&stream, 0) == Z_OK)
2486                 /* nothing */;
2487         git_SHA1_Update(&c, hdr, hdrlen);
2488
2489         /* Then the data itself.. */
2490         stream.next_in = (void *)buf;
2491         stream.avail_in = len;
2492         do {
2493                 unsigned char *in0 = stream.next_in;
2494                 ret = deflate(&stream, Z_FINISH);
2495                 git_SHA1_Update(&c, in0, stream.next_in - in0);
2496                 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2497                         die("unable to write sha1 file");
2498                 stream.next_out = compressed;
2499                 stream.avail_out = sizeof(compressed);
2500         } while (ret == Z_OK);
2501
2502         if (ret != Z_STREAM_END)
2503                 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2504         ret = deflateEnd(&stream);
2505         if (ret != Z_OK)
2506                 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2507         git_SHA1_Final(parano_sha1, &c);
2508         if (hashcmp(sha1, parano_sha1) != 0)
2509                 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2510
2511         close_sha1_file(fd);
2512
2513         if (mtime) {
2514                 struct utimbuf utb;
2515                 utb.actime = mtime;
2516                 utb.modtime = mtime;
2517                 if (utime(tmpfile, &utb) < 0)
2518                         warning("failed utime() on %s: %s",
2519                                 tmpfile, strerror(errno));
2520         }
2521
2522         return move_temp_to_file(tmpfile, filename);
2523 }
2524
2525 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2526 {
2527         unsigned char sha1[20];
2528         char hdr[32];
2529         int hdrlen;
2530
2531         /* Normally if we have it in the pack then we do not bother writing
2532          * it out into .git/objects/??/?{38} file.
2533          */
2534         write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2535         if (returnsha1)
2536                 hashcpy(returnsha1, sha1);
2537         if (has_sha1_file(sha1))
2538                 return 0;
2539         return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2540 }
2541
2542 int force_object_loose(const unsigned char *sha1, time_t mtime)
2543 {
2544         void *buf;
2545         unsigned long len;
2546         enum object_type type;
2547         char hdr[32];
2548         int hdrlen;
2549         int ret;
2550
2551         if (has_loose_object(sha1))
2552                 return 0;
2553         buf = read_packed_sha1(sha1, &type, &len);
2554         if (!buf)
2555                 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2556         hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2557         ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2558         free(buf);
2559
2560         return ret;
2561 }
2562
2563 int has_pack_index(const unsigned char *sha1)
2564 {
2565         struct stat st;
2566         if (stat(sha1_pack_index_name(sha1), &st))
2567                 return 0;
2568         return 1;
2569 }
2570
2571 int has_sha1_pack(const unsigned char *sha1)
2572 {
2573         struct pack_entry e;
2574         return find_pack_entry(sha1, &e);
2575 }
2576
2577 int has_sha1_file(const unsigned char *sha1)
2578 {
2579         struct pack_entry e;
2580
2581         if (find_pack_entry(sha1, &e))
2582                 return 1;
2583         return has_loose_object(sha1);
2584 }
2585
2586 static void check_tree(const void *buf, size_t size)
2587 {
2588         struct tree_desc desc;
2589         struct name_entry entry;
2590
2591         init_tree_desc(&desc, buf, size);
2592         while (tree_entry(&desc, &entry))
2593                 /* do nothing
2594                  * tree_entry() will die() on malformed entries */
2595                 ;
2596 }
2597
2598 static void check_commit(const void *buf, size_t size)
2599 {
2600         struct commit c;
2601         memset(&c, 0, sizeof(c));
2602         if (parse_commit_buffer(&c, buf, size))
2603                 die("corrupt commit");
2604 }
2605
2606 static void check_tag(const void *buf, size_t size)
2607 {
2608         struct tag t;
2609         memset(&t, 0, sizeof(t));
2610         if (parse_tag_buffer(&t, buf, size))
2611                 die("corrupt tag");
2612 }
2613
2614 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2615                      enum object_type type,
2616                      const char *path, unsigned flags)
2617 {
2618         int ret, re_allocated = 0;
2619         int write_object = flags & HASH_WRITE_OBJECT;
2620
2621         if (!type)
2622                 type = OBJ_BLOB;
2623
2624         /*
2625          * Convert blobs to git internal format
2626          */
2627         if ((type == OBJ_BLOB) && path) {
2628                 struct strbuf nbuf = STRBUF_INIT;
2629                 if (convert_to_git(path, buf, size, &nbuf,
2630                                    write_object ? safe_crlf : 0)) {
2631                         buf = strbuf_detach(&nbuf, &size);
2632                         re_allocated = 1;
2633                 }
2634         }
2635         if (flags & HASH_FORMAT_CHECK) {
2636                 if (type == OBJ_TREE)
2637                         check_tree(buf, size);
2638                 if (type == OBJ_COMMIT)
2639                         check_commit(buf, size);
2640                 if (type == OBJ_TAG)
2641                         check_tag(buf, size);
2642         }
2643
2644         if (write_object)
2645                 ret = write_sha1_file(buf, size, typename(type), sha1);
2646         else
2647                 ret = hash_sha1_file(buf, size, typename(type), sha1);
2648         if (re_allocated)
2649                 free(buf);
2650         return ret;
2651 }
2652
2653 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2654                       const char *path, unsigned flags)
2655 {
2656         struct strbuf sbuf = STRBUF_INIT;
2657         int ret;
2658
2659         if (strbuf_read(&sbuf, fd, 4096) >= 0)
2660                 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2661         else
2662                 ret = -1;
2663         strbuf_release(&sbuf);
2664         return ret;
2665 }
2666
2667 #define SMALL_FILE_SIZE (32*1024)
2668
2669 static int index_core(unsigned char *sha1, int fd, size_t size,
2670                       enum object_type type, const char *path,
2671                       unsigned flags)
2672 {
2673         int ret;
2674
2675         if (!size) {
2676                 ret = index_mem(sha1, NULL, size, type, path, flags);
2677         } else if (size <= SMALL_FILE_SIZE) {
2678                 char *buf = xmalloc(size);
2679                 if (size == read_in_full(fd, buf, size))
2680                         ret = index_mem(sha1, buf, size, type, path, flags);
2681                 else
2682                         ret = error("short read %s", strerror(errno));
2683                 free(buf);
2684         } else {
2685                 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2686                 ret = index_mem(sha1, buf, size, type, path, flags);
2687                 munmap(buf, size);
2688         }
2689         return ret;
2690 }
2691
2692 /*
2693  * This creates one packfile per large blob, because the caller
2694  * immediately wants the result sha1, and fast-import can report the
2695  * object name via marks mechanism only by closing the created
2696  * packfile.
2697  *
2698  * This also bypasses the usual "convert-to-git" dance, and that is on
2699  * purpose. We could write a streaming version of the converting
2700  * functions and insert that before feeding the data to fast-import
2701  * (or equivalent in-core API described above), but the primary
2702  * motivation for trying to stream from the working tree file and to
2703  * avoid mmaping it in core is to deal with large binary blobs, and
2704  * by definition they do _not_ want to get any conversion.
2705  */
2706 static int index_stream(unsigned char *sha1, int fd, size_t size,
2707                         enum object_type type, const char *path,
2708                         unsigned flags)
2709 {
2710         struct child_process fast_import;
2711         char export_marks[512];
2712         const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2713         char tmpfile[512];
2714         char fast_import_cmd[512];
2715         char buf[512];
2716         int len, tmpfd;
2717
2718         strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2719         tmpfd = git_mkstemp_mode(tmpfile, 0600);
2720         if (tmpfd < 0)
2721                 die_errno("cannot create tempfile: %s", tmpfile);
2722         if (close(tmpfd))
2723                 die_errno("cannot close tempfile: %s", tmpfile);
2724         sprintf(export_marks, "--export-marks=%s", tmpfile);
2725
2726         memset(&fast_import, 0, sizeof(fast_import));
2727         fast_import.in = -1;
2728         fast_import.argv = argv;
2729         fast_import.git_cmd = 1;
2730         if (start_command(&fast_import))
2731                 die_errno("index-stream: git fast-import failed");
2732
2733         len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2734                       (unsigned long) size);
2735         write_or_whine(fast_import.in, fast_import_cmd, len,
2736                        "index-stream: feeding fast-import");
2737         while (size) {
2738                 char buf[10240];
2739                 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
2740                 ssize_t actual;
2741
2742                 actual = read_in_full(fd, buf, sz);
2743                 if (actual < 0)
2744                         die_errno("index-stream: reading input");
2745                 if (write_in_full(fast_import.in, buf, actual) != actual)
2746                         die_errno("index-stream: feeding fast-import");
2747                 size -= actual;
2748         }
2749         if (close(fast_import.in))
2750                 die_errno("index-stream: closing fast-import");
2751         if (finish_command(&fast_import))
2752                 die_errno("index-stream: finishing fast-import");
2753
2754         tmpfd = open(tmpfile, O_RDONLY);
2755         if (tmpfd < 0)
2756                 die_errno("index-stream: cannot open fast-import mark");
2757         len = read(tmpfd, buf, sizeof(buf));
2758         if (len < 0)
2759                 die_errno("index-stream: reading fast-import mark");
2760         if (close(tmpfd) < 0)
2761                 die_errno("index-stream: closing fast-import mark");
2762         if (unlink(tmpfile))
2763                 die_errno("index-stream: unlinking fast-import mark");
2764         if (len != 44 ||
2765             memcmp(":1 ", buf, 3) ||
2766             get_sha1_hex(buf + 3, sha1))
2767                 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2768         return 0;
2769 }
2770
2771 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2772              enum object_type type, const char *path, unsigned flags)
2773 {
2774         int ret;
2775         size_t size = xsize_t(st->st_size);
2776
2777         if (!S_ISREG(st->st_mode))
2778                 ret = index_pipe(sha1, fd, type, path, flags);
2779         else if (size <= big_file_threshold || type != OBJ_BLOB)
2780                 ret = index_core(sha1, fd, size, type, path, flags);
2781         else
2782                 ret = index_stream(sha1, fd, size, type, path, flags);
2783         close(fd);
2784         return ret;
2785 }
2786
2787 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2788 {
2789         int fd;
2790         struct strbuf sb = STRBUF_INIT;
2791
2792         switch (st->st_mode & S_IFMT) {
2793         case S_IFREG:
2794                 fd = open(path, O_RDONLY);
2795                 if (fd < 0)
2796                         return error("open(\"%s\"): %s", path,
2797                                      strerror(errno));
2798                 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2799                         return error("%s: failed to insert into database",
2800                                      path);
2801                 break;
2802         case S_IFLNK:
2803                 if (strbuf_readlink(&sb, path, st->st_size)) {
2804                         char *errstr = strerror(errno);
2805                         return error("readlink(\"%s\"): %s", path,
2806                                      errstr);
2807                 }
2808                 if (!(flags & HASH_WRITE_OBJECT))
2809                         hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2810                 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2811                         return error("%s: failed to insert into database",
2812                                      path);
2813                 strbuf_release(&sb);
2814                 break;
2815         case S_IFDIR:
2816                 return resolve_gitlink_ref(path, "HEAD", sha1);
2817         default:
2818                 return error("%s: unsupported file type", path);
2819         }
2820         return 0;
2821 }
2822
2823 int read_pack_header(int fd, struct pack_header *header)
2824 {
2825         if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2826                 /* "eof before pack header was fully read" */
2827                 return PH_ERROR_EOF;
2828
2829         if (header->hdr_signature != htonl(PACK_SIGNATURE))
2830                 /* "protocol error (pack signature mismatch detected)" */
2831                 return PH_ERROR_PACK_SIGNATURE;
2832         if (!pack_version_ok(header->hdr_version))
2833                 /* "protocol error (pack version unsupported)" */
2834                 return PH_ERROR_PROTOCOL;
2835         return 0;
2836 }
2837
2838 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2839 {
2840         enum object_type type = sha1_object_info(sha1, NULL);
2841         if (type < 0)
2842                 die("%s is not a valid object", sha1_to_hex(sha1));
2843         if (type != expect)
2844                 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2845                     typename(expect));
2846 }