Merge branch 'nd/index-doc'
[git/git.git] / sha1_file.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "tag.h"
15 #include "tree.h"
16 #include "tree-walk.h"
17 #include "refs.h"
18 #include "pack-revindex.h"
19 #include "sha1-lookup.h"
20
21 #ifndef O_NOATIME
22 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
23 #define O_NOATIME 01000000
24 #else
25 #define O_NOATIME 0
26 #endif
27 #endif
28
29 #define SZ_FMT PRIuMAX
30 static inline uintmax_t sz_fmt(size_t s) { return s; }
31
32 const unsigned char null_sha1[20];
33
34 static int git_open_noatime(const char *name, struct packed_git *p);
35
36 /*
37 * This is meant to hold a *small* number of objects that you would
38 * want read_sha1_file() to be able to return, but yet you do not want
39 * to write them into the object store (e.g. a browse-only
40 * application).
41 */
42 static struct cached_object {
43 unsigned char sha1[20];
44 enum object_type type;
45 void *buf;
46 unsigned long size;
47 } *cached_objects;
48 static int cached_object_nr, cached_object_alloc;
49
50 static struct cached_object empty_tree = {
51 EMPTY_TREE_SHA1_BIN_LITERAL,
52 OBJ_TREE,
53 "",
54 0
55 };
56
57 static struct cached_object *find_cached_object(const unsigned char *sha1)
58 {
59 int i;
60 struct cached_object *co = cached_objects;
61
62 for (i = 0; i < cached_object_nr; i++, co++) {
63 if (!hashcmp(co->sha1, sha1))
64 return co;
65 }
66 if (!hashcmp(sha1, empty_tree.sha1))
67 return &empty_tree;
68 return NULL;
69 }
70
71 int safe_create_leading_directories(char *path)
72 {
73 char *pos = path + offset_1st_component(path);
74 struct stat st;
75
76 while (pos) {
77 pos = strchr(pos, '/');
78 if (!pos)
79 break;
80 while (*++pos == '/')
81 ;
82 if (!*pos)
83 break;
84 *--pos = '\0';
85 if (!stat(path, &st)) {
86 /* path exists */
87 if (!S_ISDIR(st.st_mode)) {
88 *pos = '/';
89 return -3;
90 }
91 }
92 else if (mkdir(path, 0777)) {
93 *pos = '/';
94 return -1;
95 }
96 else if (adjust_shared_perm(path)) {
97 *pos = '/';
98 return -2;
99 }
100 *pos++ = '/';
101 }
102 return 0;
103 }
104
105 int safe_create_leading_directories_const(const char *path)
106 {
107 /* path points to cache entries, so xstrdup before messing with it */
108 char *buf = xstrdup(path);
109 int result = safe_create_leading_directories(buf);
110 free(buf);
111 return result;
112 }
113
114 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
115 {
116 int i;
117 for (i = 0; i < 20; i++) {
118 static char hex[] = "0123456789abcdef";
119 unsigned int val = sha1[i];
120 char *pos = pathbuf + i*2 + (i > 0);
121 *pos++ = hex[val >> 4];
122 *pos = hex[val & 0xf];
123 }
124 }
125
126 /*
127 * NOTE! This returns a statically allocated buffer, so you have to be
128 * careful about using it. Do an "xstrdup()" if you need to save the
129 * filename.
130 *
131 * Also note that this returns the location for creating. Reading
132 * SHA1 file can happen from any alternate directory listed in the
133 * DB_ENVIRONMENT environment variable if it is not found in
134 * the primary object database.
135 */
136 char *sha1_file_name(const unsigned char *sha1)
137 {
138 static char buf[PATH_MAX];
139 const char *objdir;
140 int len;
141
142 objdir = get_object_directory();
143 len = strlen(objdir);
144
145 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
146 if (len + 43 > PATH_MAX)
147 die("insanely long object directory %s", objdir);
148 memcpy(buf, objdir, len);
149 buf[len] = '/';
150 buf[len+3] = '/';
151 buf[len+42] = '\0';
152 fill_sha1_path(buf + len + 1, sha1);
153 return buf;
154 }
155
156 static char *sha1_get_pack_name(const unsigned char *sha1,
157 char **name, char **base, const char *which)
158 {
159 static const char hex[] = "0123456789abcdef";
160 char *buf;
161 int i;
162
163 if (!*base) {
164 const char *sha1_file_directory = get_object_directory();
165 int len = strlen(sha1_file_directory);
166 *base = xmalloc(len + 60);
167 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
168 sha1_file_directory, which);
169 *name = *base + len + 11;
170 }
171
172 buf = *name;
173
174 for (i = 0; i < 20; i++) {
175 unsigned int val = *sha1++;
176 *buf++ = hex[val >> 4];
177 *buf++ = hex[val & 0xf];
178 }
179
180 return *base;
181 }
182
183 char *sha1_pack_name(const unsigned char *sha1)
184 {
185 static char *name, *base;
186
187 return sha1_get_pack_name(sha1, &name, &base, "pack");
188 }
189
190 char *sha1_pack_index_name(const unsigned char *sha1)
191 {
192 static char *name, *base;
193
194 return sha1_get_pack_name(sha1, &name, &base, "idx");
195 }
196
197 struct alternate_object_database *alt_odb_list;
198 static struct alternate_object_database **alt_odb_tail;
199
200 static void read_info_alternates(const char * alternates, int depth);
201
202 /*
203 * Prepare alternate object database registry.
204 *
205 * The variable alt_odb_list points at the list of struct
206 * alternate_object_database. The elements on this list come from
207 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
208 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
209 * whose contents is similar to that environment variable but can be
210 * LF separated. Its base points at a statically allocated buffer that
211 * contains "/the/directory/corresponding/to/.git/objects/...", while
212 * its name points just after the slash at the end of ".git/objects/"
213 * in the example above, and has enough space to hold 40-byte hex
214 * SHA1, an extra slash for the first level indirection, and the
215 * terminating NUL.
216 */
217 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
218 {
219 const char *objdir = get_object_directory();
220 struct alternate_object_database *ent;
221 struct alternate_object_database *alt;
222 /* 43 = 40-byte + 2 '/' + terminating NUL */
223 int pfxlen = len;
224 int entlen = pfxlen + 43;
225 int base_len = -1;
226
227 if (!is_absolute_path(entry) && relative_base) {
228 /* Relative alt-odb */
229 if (base_len < 0)
230 base_len = strlen(relative_base) + 1;
231 entlen += base_len;
232 pfxlen += base_len;
233 }
234 ent = xmalloc(sizeof(*ent) + entlen);
235
236 if (!is_absolute_path(entry) && relative_base) {
237 memcpy(ent->base, relative_base, base_len - 1);
238 ent->base[base_len - 1] = '/';
239 memcpy(ent->base + base_len, entry, len);
240 }
241 else
242 memcpy(ent->base, entry, pfxlen);
243
244 ent->name = ent->base + pfxlen + 1;
245 ent->base[pfxlen + 3] = '/';
246 ent->base[pfxlen] = ent->base[entlen-1] = 0;
247
248 /* Detect cases where alternate disappeared */
249 if (!is_directory(ent->base)) {
250 error("object directory %s does not exist; "
251 "check .git/objects/info/alternates.",
252 ent->base);
253 free(ent);
254 return -1;
255 }
256
257 /* Prevent the common mistake of listing the same
258 * thing twice, or object directory itself.
259 */
260 for (alt = alt_odb_list; alt; alt = alt->next) {
261 if (!memcmp(ent->base, alt->base, pfxlen)) {
262 free(ent);
263 return -1;
264 }
265 }
266 if (!memcmp(ent->base, objdir, pfxlen)) {
267 free(ent);
268 return -1;
269 }
270
271 /* add the alternate entry */
272 *alt_odb_tail = ent;
273 alt_odb_tail = &(ent->next);
274 ent->next = NULL;
275
276 /* recursively add alternates */
277 read_info_alternates(ent->base, depth + 1);
278
279 ent->base[pfxlen] = '/';
280
281 return 0;
282 }
283
284 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
285 const char *relative_base, int depth)
286 {
287 const char *cp, *last;
288
289 if (depth > 5) {
290 error("%s: ignoring alternate object stores, nesting too deep.",
291 relative_base);
292 return;
293 }
294
295 last = alt;
296 while (last < ep) {
297 cp = last;
298 if (cp < ep && *cp == '#') {
299 while (cp < ep && *cp != sep)
300 cp++;
301 last = cp + 1;
302 continue;
303 }
304 while (cp < ep && *cp != sep)
305 cp++;
306 if (last != cp) {
307 if (!is_absolute_path(last) && depth) {
308 error("%s: ignoring relative alternate object store %s",
309 relative_base, last);
310 } else {
311 link_alt_odb_entry(last, cp - last,
312 relative_base, depth);
313 }
314 }
315 while (cp < ep && *cp == sep)
316 cp++;
317 last = cp;
318 }
319 }
320
321 static void read_info_alternates(const char * relative_base, int depth)
322 {
323 char *map;
324 size_t mapsz;
325 struct stat st;
326 const char alt_file_name[] = "info/alternates";
327 /* Given that relative_base is no longer than PATH_MAX,
328 ensure that "path" has enough space to append "/", the
329 file name, "info/alternates", and a trailing NUL. */
330 char path[PATH_MAX + 1 + sizeof alt_file_name];
331 int fd;
332
333 sprintf(path, "%s/%s", relative_base, alt_file_name);
334 fd = git_open_noatime(path, NULL);
335 if (fd < 0)
336 return;
337 if (fstat(fd, &st) || (st.st_size == 0)) {
338 close(fd);
339 return;
340 }
341 mapsz = xsize_t(st.st_size);
342 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
343 close(fd);
344
345 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
346
347 munmap(map, mapsz);
348 }
349
350 void add_to_alternates_file(const char *reference)
351 {
352 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
353 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
354 char *alt = mkpath("%s/objects\n", reference);
355 write_or_die(fd, alt, strlen(alt));
356 if (commit_lock_file(lock))
357 die("could not close alternates file");
358 if (alt_odb_tail)
359 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
360 }
361
362 void foreach_alt_odb(alt_odb_fn fn, void *cb)
363 {
364 struct alternate_object_database *ent;
365
366 prepare_alt_odb();
367 for (ent = alt_odb_list; ent; ent = ent->next)
368 if (fn(ent, cb))
369 return;
370 }
371
372 void prepare_alt_odb(void)
373 {
374 const char *alt;
375
376 if (alt_odb_tail)
377 return;
378
379 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
380 if (!alt) alt = "";
381
382 alt_odb_tail = &alt_odb_list;
383 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
384
385 read_info_alternates(get_object_directory(), 0);
386 }
387
388 static int has_loose_object_local(const unsigned char *sha1)
389 {
390 char *name = sha1_file_name(sha1);
391 return !access(name, F_OK);
392 }
393
394 int has_loose_object_nonlocal(const unsigned char *sha1)
395 {
396 struct alternate_object_database *alt;
397 prepare_alt_odb();
398 for (alt = alt_odb_list; alt; alt = alt->next) {
399 fill_sha1_path(alt->name, sha1);
400 if (!access(alt->base, F_OK))
401 return 1;
402 }
403 return 0;
404 }
405
406 static int has_loose_object(const unsigned char *sha1)
407 {
408 return has_loose_object_local(sha1) ||
409 has_loose_object_nonlocal(sha1);
410 }
411
412 static unsigned int pack_used_ctr;
413 static unsigned int pack_mmap_calls;
414 static unsigned int peak_pack_open_windows;
415 static unsigned int pack_open_windows;
416 static unsigned int pack_open_fds;
417 static unsigned int pack_max_fds;
418 static size_t peak_pack_mapped;
419 static size_t pack_mapped;
420 struct packed_git *packed_git;
421
422 void pack_report(void)
423 {
424 fprintf(stderr,
425 "pack_report: getpagesize() = %10" SZ_FMT "\n"
426 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
427 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
428 sz_fmt(getpagesize()),
429 sz_fmt(packed_git_window_size),
430 sz_fmt(packed_git_limit));
431 fprintf(stderr,
432 "pack_report: pack_used_ctr = %10u\n"
433 "pack_report: pack_mmap_calls = %10u\n"
434 "pack_report: pack_open_windows = %10u / %10u\n"
435 "pack_report: pack_mapped = "
436 "%10" SZ_FMT " / %10" SZ_FMT "\n",
437 pack_used_ctr,
438 pack_mmap_calls,
439 pack_open_windows, peak_pack_open_windows,
440 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
441 }
442
443 static int check_packed_git_idx(const char *path, struct packed_git *p)
444 {
445 void *idx_map;
446 struct pack_idx_header *hdr;
447 size_t idx_size;
448 uint32_t version, nr, i, *index;
449 int fd = git_open_noatime(path, p);
450 struct stat st;
451
452 if (fd < 0)
453 return -1;
454 if (fstat(fd, &st)) {
455 close(fd);
456 return -1;
457 }
458 idx_size = xsize_t(st.st_size);
459 if (idx_size < 4 * 256 + 20 + 20) {
460 close(fd);
461 return error("index file %s is too small", path);
462 }
463 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
464 close(fd);
465
466 hdr = idx_map;
467 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
468 version = ntohl(hdr->idx_version);
469 if (version < 2 || version > 2) {
470 munmap(idx_map, idx_size);
471 return error("index file %s is version %"PRIu32
472 " and is not supported by this binary"
473 " (try upgrading GIT to a newer version)",
474 path, version);
475 }
476 } else
477 version = 1;
478
479 nr = 0;
480 index = idx_map;
481 if (version > 1)
482 index += 2; /* skip index header */
483 for (i = 0; i < 256; i++) {
484 uint32_t n = ntohl(index[i]);
485 if (n < nr) {
486 munmap(idx_map, idx_size);
487 return error("non-monotonic index %s", path);
488 }
489 nr = n;
490 }
491
492 if (version == 1) {
493 /*
494 * Total size:
495 * - 256 index entries 4 bytes each
496 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
497 * - 20-byte SHA1 of the packfile
498 * - 20-byte SHA1 file checksum
499 */
500 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
501 munmap(idx_map, idx_size);
502 return error("wrong index v1 file size in %s", path);
503 }
504 } else if (version == 2) {
505 /*
506 * Minimum size:
507 * - 8 bytes of header
508 * - 256 index entries 4 bytes each
509 * - 20-byte sha1 entry * nr
510 * - 4-byte crc entry * nr
511 * - 4-byte offset entry * nr
512 * - 20-byte SHA1 of the packfile
513 * - 20-byte SHA1 file checksum
514 * And after the 4-byte offset table might be a
515 * variable sized table containing 8-byte entries
516 * for offsets larger than 2^31.
517 */
518 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
519 unsigned long max_size = min_size;
520 if (nr)
521 max_size += (nr - 1)*8;
522 if (idx_size < min_size || idx_size > max_size) {
523 munmap(idx_map, idx_size);
524 return error("wrong index v2 file size in %s", path);
525 }
526 if (idx_size != min_size &&
527 /*
528 * make sure we can deal with large pack offsets.
529 * 31-bit signed offset won't be enough, neither
530 * 32-bit unsigned one will be.
531 */
532 (sizeof(off_t) <= 4)) {
533 munmap(idx_map, idx_size);
534 return error("pack too large for current definition of off_t in %s", path);
535 }
536 }
537
538 p->index_version = version;
539 p->index_data = idx_map;
540 p->index_size = idx_size;
541 p->num_objects = nr;
542 return 0;
543 }
544
545 int open_pack_index(struct packed_git *p)
546 {
547 char *idx_name;
548 int ret;
549
550 if (p->index_data)
551 return 0;
552
553 idx_name = xstrdup(p->pack_name);
554 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
555 ret = check_packed_git_idx(idx_name, p);
556 free(idx_name);
557 return ret;
558 }
559
560 static void scan_windows(struct packed_git *p,
561 struct packed_git **lru_p,
562 struct pack_window **lru_w,
563 struct pack_window **lru_l)
564 {
565 struct pack_window *w, *w_l;
566
567 for (w_l = NULL, w = p->windows; w; w = w->next) {
568 if (!w->inuse_cnt) {
569 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
570 *lru_p = p;
571 *lru_w = w;
572 *lru_l = w_l;
573 }
574 }
575 w_l = w;
576 }
577 }
578
579 static int unuse_one_window(struct packed_git *current, int keep_fd)
580 {
581 struct packed_git *p, *lru_p = NULL;
582 struct pack_window *lru_w = NULL, *lru_l = NULL;
583
584 if (current)
585 scan_windows(current, &lru_p, &lru_w, &lru_l);
586 for (p = packed_git; p; p = p->next)
587 scan_windows(p, &lru_p, &lru_w, &lru_l);
588 if (lru_p) {
589 munmap(lru_w->base, lru_w->len);
590 pack_mapped -= lru_w->len;
591 if (lru_l)
592 lru_l->next = lru_w->next;
593 else {
594 lru_p->windows = lru_w->next;
595 if (!lru_p->windows && lru_p->pack_fd != -1
596 && lru_p->pack_fd != keep_fd) {
597 close(lru_p->pack_fd);
598 pack_open_fds--;
599 lru_p->pack_fd = -1;
600 }
601 }
602 free(lru_w);
603 pack_open_windows--;
604 return 1;
605 }
606 return 0;
607 }
608
609 void release_pack_memory(size_t need, int fd)
610 {
611 size_t cur = pack_mapped;
612 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
613 ; /* nothing */
614 }
615
616 void *xmmap(void *start, size_t length,
617 int prot, int flags, int fd, off_t offset)
618 {
619 void *ret = mmap(start, length, prot, flags, fd, offset);
620 if (ret == MAP_FAILED) {
621 if (!length)
622 return NULL;
623 release_pack_memory(length, fd);
624 ret = mmap(start, length, prot, flags, fd, offset);
625 if (ret == MAP_FAILED)
626 die_errno("Out of memory? mmap failed");
627 }
628 return ret;
629 }
630
631 void close_pack_windows(struct packed_git *p)
632 {
633 while (p->windows) {
634 struct pack_window *w = p->windows;
635
636 if (w->inuse_cnt)
637 die("pack '%s' still has open windows to it",
638 p->pack_name);
639 munmap(w->base, w->len);
640 pack_mapped -= w->len;
641 pack_open_windows--;
642 p->windows = w->next;
643 free(w);
644 }
645 }
646
647 void unuse_pack(struct pack_window **w_cursor)
648 {
649 struct pack_window *w = *w_cursor;
650 if (w) {
651 w->inuse_cnt--;
652 *w_cursor = NULL;
653 }
654 }
655
656 void close_pack_index(struct packed_git *p)
657 {
658 if (p->index_data) {
659 munmap((void *)p->index_data, p->index_size);
660 p->index_data = NULL;
661 }
662 }
663
664 /*
665 * This is used by git-repack in case a newly created pack happens to
666 * contain the same set of objects as an existing one. In that case
667 * the resulting file might be different even if its name would be the
668 * same. It is best to close any reference to the old pack before it is
669 * replaced on disk. Of course no index pointers nor windows for given pack
670 * must subsist at this point. If ever objects from this pack are requested
671 * again, the new version of the pack will be reinitialized through
672 * reprepare_packed_git().
673 */
674 void free_pack_by_name(const char *pack_name)
675 {
676 struct packed_git *p, **pp = &packed_git;
677
678 while (*pp) {
679 p = *pp;
680 if (strcmp(pack_name, p->pack_name) == 0) {
681 clear_delta_base_cache();
682 close_pack_windows(p);
683 if (p->pack_fd != -1) {
684 close(p->pack_fd);
685 pack_open_fds--;
686 }
687 close_pack_index(p);
688 free(p->bad_object_sha1);
689 *pp = p->next;
690 free(p);
691 return;
692 }
693 pp = &p->next;
694 }
695 }
696
697 /*
698 * Do not call this directly as this leaks p->pack_fd on error return;
699 * call open_packed_git() instead.
700 */
701 static int open_packed_git_1(struct packed_git *p)
702 {
703 struct stat st;
704 struct pack_header hdr;
705 unsigned char sha1[20];
706 unsigned char *idx_sha1;
707 long fd_flag;
708
709 if (!p->index_data && open_pack_index(p))
710 return error("packfile %s index unavailable", p->pack_name);
711
712 if (!pack_max_fds) {
713 struct rlimit lim;
714 unsigned int max_fds;
715
716 if (getrlimit(RLIMIT_NOFILE, &lim))
717 die_errno("cannot get RLIMIT_NOFILE");
718
719 max_fds = lim.rlim_cur;
720
721 /* Save 3 for stdin/stdout/stderr, 22 for work */
722 if (25 < max_fds)
723 pack_max_fds = max_fds - 25;
724 else
725 pack_max_fds = 1;
726 }
727
728 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
729 ; /* nothing */
730
731 p->pack_fd = git_open_noatime(p->pack_name, p);
732 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
733 return -1;
734 pack_open_fds++;
735
736 /* If we created the struct before we had the pack we lack size. */
737 if (!p->pack_size) {
738 if (!S_ISREG(st.st_mode))
739 return error("packfile %s not a regular file", p->pack_name);
740 p->pack_size = st.st_size;
741 } else if (p->pack_size != st.st_size)
742 return error("packfile %s size changed", p->pack_name);
743
744 /* We leave these file descriptors open with sliding mmap;
745 * there is no point keeping them open across exec(), though.
746 */
747 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
748 if (fd_flag < 0)
749 return error("cannot determine file descriptor flags");
750 fd_flag |= FD_CLOEXEC;
751 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
752 return error("cannot set FD_CLOEXEC");
753
754 /* Verify we recognize this pack file format. */
755 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
756 return error("file %s is far too short to be a packfile", p->pack_name);
757 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
758 return error("file %s is not a GIT packfile", p->pack_name);
759 if (!pack_version_ok(hdr.hdr_version))
760 return error("packfile %s is version %"PRIu32" and not"
761 " supported (try upgrading GIT to a newer version)",
762 p->pack_name, ntohl(hdr.hdr_version));
763
764 /* Verify the pack matches its index. */
765 if (p->num_objects != ntohl(hdr.hdr_entries))
766 return error("packfile %s claims to have %"PRIu32" objects"
767 " while index indicates %"PRIu32" objects",
768 p->pack_name, ntohl(hdr.hdr_entries),
769 p->num_objects);
770 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
771 return error("end of packfile %s is unavailable", p->pack_name);
772 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
773 return error("packfile %s signature is unavailable", p->pack_name);
774 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
775 if (hashcmp(sha1, idx_sha1))
776 return error("packfile %s does not match index", p->pack_name);
777 return 0;
778 }
779
780 static int open_packed_git(struct packed_git *p)
781 {
782 if (!open_packed_git_1(p))
783 return 0;
784 if (p->pack_fd != -1) {
785 close(p->pack_fd);
786 pack_open_fds--;
787 p->pack_fd = -1;
788 }
789 return -1;
790 }
791
792 static int in_window(struct pack_window *win, off_t offset)
793 {
794 /* We must promise at least 20 bytes (one hash) after the
795 * offset is available from this window, otherwise the offset
796 * is not actually in this window and a different window (which
797 * has that one hash excess) must be used. This is to support
798 * the object header and delta base parsing routines below.
799 */
800 off_t win_off = win->offset;
801 return win_off <= offset
802 && (offset + 20) <= (win_off + win->len);
803 }
804
805 unsigned char *use_pack(struct packed_git *p,
806 struct pack_window **w_cursor,
807 off_t offset,
808 unsigned int *left)
809 {
810 struct pack_window *win = *w_cursor;
811
812 /* Since packfiles end in a hash of their content and it's
813 * pointless to ask for an offset into the middle of that
814 * hash, and the in_window function above wouldn't match
815 * don't allow an offset too close to the end of the file.
816 */
817 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
818 die("packfile %s cannot be accessed", p->pack_name);
819 if (offset > (p->pack_size - 20))
820 die("offset beyond end of packfile (truncated pack?)");
821
822 if (!win || !in_window(win, offset)) {
823 if (win)
824 win->inuse_cnt--;
825 for (win = p->windows; win; win = win->next) {
826 if (in_window(win, offset))
827 break;
828 }
829 if (!win) {
830 size_t window_align = packed_git_window_size / 2;
831 off_t len;
832
833 if (p->pack_fd == -1 && open_packed_git(p))
834 die("packfile %s cannot be accessed", p->pack_name);
835
836 win = xcalloc(1, sizeof(*win));
837 win->offset = (offset / window_align) * window_align;
838 len = p->pack_size - win->offset;
839 if (len > packed_git_window_size)
840 len = packed_git_window_size;
841 win->len = (size_t)len;
842 pack_mapped += win->len;
843 while (packed_git_limit < pack_mapped
844 && unuse_one_window(p, p->pack_fd))
845 ; /* nothing */
846 win->base = xmmap(NULL, win->len,
847 PROT_READ, MAP_PRIVATE,
848 p->pack_fd, win->offset);
849 if (win->base == MAP_FAILED)
850 die("packfile %s cannot be mapped: %s",
851 p->pack_name,
852 strerror(errno));
853 if (!win->offset && win->len == p->pack_size
854 && !p->do_not_close) {
855 close(p->pack_fd);
856 pack_open_fds--;
857 p->pack_fd = -1;
858 }
859 pack_mmap_calls++;
860 pack_open_windows++;
861 if (pack_mapped > peak_pack_mapped)
862 peak_pack_mapped = pack_mapped;
863 if (pack_open_windows > peak_pack_open_windows)
864 peak_pack_open_windows = pack_open_windows;
865 win->next = p->windows;
866 p->windows = win;
867 }
868 }
869 if (win != *w_cursor) {
870 win->last_used = pack_used_ctr++;
871 win->inuse_cnt++;
872 *w_cursor = win;
873 }
874 offset -= win->offset;
875 if (left)
876 *left = win->len - xsize_t(offset);
877 return win->base + offset;
878 }
879
880 static struct packed_git *alloc_packed_git(int extra)
881 {
882 struct packed_git *p = xmalloc(sizeof(*p) + extra);
883 memset(p, 0, sizeof(*p));
884 p->pack_fd = -1;
885 return p;
886 }
887
888 static void try_to_free_pack_memory(size_t size)
889 {
890 release_pack_memory(size, -1);
891 }
892
893 struct packed_git *add_packed_git(const char *path, int path_len, int local)
894 {
895 static int have_set_try_to_free_routine;
896 struct stat st;
897 struct packed_git *p = alloc_packed_git(path_len + 2);
898
899 if (!have_set_try_to_free_routine) {
900 have_set_try_to_free_routine = 1;
901 set_try_to_free_routine(try_to_free_pack_memory);
902 }
903
904 /*
905 * Make sure a corresponding .pack file exists and that
906 * the index looks sane.
907 */
908 path_len -= strlen(".idx");
909 if (path_len < 1) {
910 free(p);
911 return NULL;
912 }
913 memcpy(p->pack_name, path, path_len);
914
915 strcpy(p->pack_name + path_len, ".keep");
916 if (!access(p->pack_name, F_OK))
917 p->pack_keep = 1;
918
919 strcpy(p->pack_name + path_len, ".pack");
920 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
921 free(p);
922 return NULL;
923 }
924
925 /* ok, it looks sane as far as we can check without
926 * actually mapping the pack file.
927 */
928 p->pack_size = st.st_size;
929 p->pack_local = local;
930 p->mtime = st.st_mtime;
931 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
932 hashclr(p->sha1);
933 return p;
934 }
935
936 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
937 {
938 const char *path = sha1_pack_name(sha1);
939 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
940
941 strcpy(p->pack_name, path);
942 hashcpy(p->sha1, sha1);
943 if (check_packed_git_idx(idx_path, p)) {
944 free(p);
945 return NULL;
946 }
947
948 return p;
949 }
950
951 void install_packed_git(struct packed_git *pack)
952 {
953 if (pack->pack_fd != -1)
954 pack_open_fds++;
955
956 pack->next = packed_git;
957 packed_git = pack;
958 }
959
960 static void prepare_packed_git_one(char *objdir, int local)
961 {
962 /* Ensure that this buffer is large enough so that we can
963 append "/pack/" without clobbering the stack even if
964 strlen(objdir) were PATH_MAX. */
965 char path[PATH_MAX + 1 + 4 + 1 + 1];
966 int len;
967 DIR *dir;
968 struct dirent *de;
969
970 sprintf(path, "%s/pack", objdir);
971 len = strlen(path);
972 dir = opendir(path);
973 if (!dir) {
974 if (errno != ENOENT)
975 error("unable to open object pack directory: %s: %s",
976 path, strerror(errno));
977 return;
978 }
979 path[len++] = '/';
980 while ((de = readdir(dir)) != NULL) {
981 int namelen = strlen(de->d_name);
982 struct packed_git *p;
983
984 if (!has_extension(de->d_name, ".idx"))
985 continue;
986
987 if (len + namelen + 1 > sizeof(path))
988 continue;
989
990 /* Don't reopen a pack we already have. */
991 strcpy(path + len, de->d_name);
992 for (p = packed_git; p; p = p->next) {
993 if (!memcmp(path, p->pack_name, len + namelen - 4))
994 break;
995 }
996 if (p)
997 continue;
998 /* See if it really is a valid .idx file with corresponding
999 * .pack file that we can map.
1000 */
1001 p = add_packed_git(path, len + namelen, local);
1002 if (!p)
1003 continue;
1004 install_packed_git(p);
1005 }
1006 closedir(dir);
1007 }
1008
1009 static int sort_pack(const void *a_, const void *b_)
1010 {
1011 struct packed_git *a = *((struct packed_git **)a_);
1012 struct packed_git *b = *((struct packed_git **)b_);
1013 int st;
1014
1015 /*
1016 * Local packs tend to contain objects specific to our
1017 * variant of the project than remote ones. In addition,
1018 * remote ones could be on a network mounted filesystem.
1019 * Favor local ones for these reasons.
1020 */
1021 st = a->pack_local - b->pack_local;
1022 if (st)
1023 return -st;
1024
1025 /*
1026 * Younger packs tend to contain more recent objects,
1027 * and more recent objects tend to get accessed more
1028 * often.
1029 */
1030 if (a->mtime < b->mtime)
1031 return 1;
1032 else if (a->mtime == b->mtime)
1033 return 0;
1034 return -1;
1035 }
1036
1037 static void rearrange_packed_git(void)
1038 {
1039 struct packed_git **ary, *p;
1040 int i, n;
1041
1042 for (n = 0, p = packed_git; p; p = p->next)
1043 n++;
1044 if (n < 2)
1045 return;
1046
1047 /* prepare an array of packed_git for easier sorting */
1048 ary = xcalloc(n, sizeof(struct packed_git *));
1049 for (n = 0, p = packed_git; p; p = p->next)
1050 ary[n++] = p;
1051
1052 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1053
1054 /* link them back again */
1055 for (i = 0; i < n - 1; i++)
1056 ary[i]->next = ary[i + 1];
1057 ary[n - 1]->next = NULL;
1058 packed_git = ary[0];
1059
1060 free(ary);
1061 }
1062
1063 static int prepare_packed_git_run_once = 0;
1064 void prepare_packed_git(void)
1065 {
1066 struct alternate_object_database *alt;
1067
1068 if (prepare_packed_git_run_once)
1069 return;
1070 prepare_packed_git_one(get_object_directory(), 1);
1071 prepare_alt_odb();
1072 for (alt = alt_odb_list; alt; alt = alt->next) {
1073 alt->name[-1] = 0;
1074 prepare_packed_git_one(alt->base, 0);
1075 alt->name[-1] = '/';
1076 }
1077 rearrange_packed_git();
1078 prepare_packed_git_run_once = 1;
1079 }
1080
1081 void reprepare_packed_git(void)
1082 {
1083 discard_revindex();
1084 prepare_packed_git_run_once = 0;
1085 prepare_packed_git();
1086 }
1087
1088 static void mark_bad_packed_object(struct packed_git *p,
1089 const unsigned char *sha1)
1090 {
1091 unsigned i;
1092 for (i = 0; i < p->num_bad_objects; i++)
1093 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1094 return;
1095 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1096 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1097 p->num_bad_objects++;
1098 }
1099
1100 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1101 {
1102 struct packed_git *p;
1103 unsigned i;
1104
1105 for (p = packed_git; p; p = p->next)
1106 for (i = 0; i < p->num_bad_objects; i++)
1107 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1108 return p;
1109 return NULL;
1110 }
1111
1112 int check_sha1_signature(const unsigned char *sha1, void *map, unsigned long size, const char *type)
1113 {
1114 unsigned char real_sha1[20];
1115 hash_sha1_file(map, size, type, real_sha1);
1116 return hashcmp(sha1, real_sha1) ? -1 : 0;
1117 }
1118
1119 static int git_open_noatime(const char *name, struct packed_git *p)
1120 {
1121 static int sha1_file_open_flag = O_NOATIME;
1122
1123 for (;;) {
1124 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1125 if (fd >= 0)
1126 return fd;
1127
1128 /* Might the failure be due to O_NOATIME? */
1129 if (errno != ENOENT && sha1_file_open_flag) {
1130 sha1_file_open_flag = 0;
1131 continue;
1132 }
1133
1134 return -1;
1135 }
1136 }
1137
1138 static int open_sha1_file(const unsigned char *sha1)
1139 {
1140 int fd;
1141 char *name = sha1_file_name(sha1);
1142 struct alternate_object_database *alt;
1143
1144 fd = git_open_noatime(name, NULL);
1145 if (fd >= 0)
1146 return fd;
1147
1148 prepare_alt_odb();
1149 errno = ENOENT;
1150 for (alt = alt_odb_list; alt; alt = alt->next) {
1151 name = alt->name;
1152 fill_sha1_path(name, sha1);
1153 fd = git_open_noatime(alt->base, NULL);
1154 if (fd >= 0)
1155 return fd;
1156 }
1157 return -1;
1158 }
1159
1160 static void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1161 {
1162 void *map;
1163 int fd;
1164
1165 fd = open_sha1_file(sha1);
1166 map = NULL;
1167 if (fd >= 0) {
1168 struct stat st;
1169
1170 if (!fstat(fd, &st)) {
1171 *size = xsize_t(st.st_size);
1172 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1173 }
1174 close(fd);
1175 }
1176 return map;
1177 }
1178
1179 static int legacy_loose_object(unsigned char *map)
1180 {
1181 unsigned int word;
1182
1183 /*
1184 * Is it a zlib-compressed buffer? If so, the first byte
1185 * must be 0x78 (15-bit window size, deflated), and the
1186 * first 16-bit word is evenly divisible by 31
1187 */
1188 word = (map[0] << 8) + map[1];
1189 if (map[0] == 0x78 && !(word % 31))
1190 return 1;
1191 else
1192 return 0;
1193 }
1194
1195 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1196 unsigned long len, enum object_type *type, unsigned long *sizep)
1197 {
1198 unsigned shift;
1199 unsigned long size, c;
1200 unsigned long used = 0;
1201
1202 c = buf[used++];
1203 *type = (c >> 4) & 7;
1204 size = c & 15;
1205 shift = 4;
1206 while (c & 0x80) {
1207 if (len <= used || bitsizeof(long) <= shift) {
1208 error("bad object header");
1209 return 0;
1210 }
1211 c = buf[used++];
1212 size += (c & 0x7f) << shift;
1213 shift += 7;
1214 }
1215 *sizep = size;
1216 return used;
1217 }
1218
1219 static int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1220 {
1221 unsigned long size, used;
1222 static const char valid_loose_object_type[8] = {
1223 0, /* OBJ_EXT */
1224 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1225 0, /* "delta" and others are invalid in a loose object */
1226 };
1227 enum object_type type;
1228
1229 /* Get the data stream */
1230 memset(stream, 0, sizeof(*stream));
1231 stream->next_in = map;
1232 stream->avail_in = mapsize;
1233 stream->next_out = buffer;
1234 stream->avail_out = bufsiz;
1235
1236 if (legacy_loose_object(map)) {
1237 git_inflate_init(stream);
1238 return git_inflate(stream, 0);
1239 }
1240
1241
1242 /*
1243 * There used to be a second loose object header format which
1244 * was meant to mimic the in-pack format, allowing for direct
1245 * copy of the object data. This format turned up not to be
1246 * really worth it and we don't write it any longer. But we
1247 * can still read it.
1248 */
1249 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1250 if (!used || !valid_loose_object_type[type])
1251 return -1;
1252 map += used;
1253 mapsize -= used;
1254
1255 /* Set up the stream for the rest.. */
1256 stream->next_in = map;
1257 stream->avail_in = mapsize;
1258 git_inflate_init(stream);
1259
1260 /* And generate the fake traditional header */
1261 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1262 typename(type), size);
1263 return 0;
1264 }
1265
1266 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1267 {
1268 int bytes = strlen(buffer) + 1;
1269 unsigned char *buf = xmallocz(size);
1270 unsigned long n;
1271 int status = Z_OK;
1272
1273 n = stream->total_out - bytes;
1274 if (n > size)
1275 n = size;
1276 memcpy(buf, (char *) buffer + bytes, n);
1277 bytes = n;
1278 if (bytes <= size) {
1279 /*
1280 * The above condition must be (bytes <= size), not
1281 * (bytes < size). In other words, even though we
1282 * expect no more output and set avail_out to zer0,
1283 * the input zlib stream may have bytes that express
1284 * "this concludes the stream", and we *do* want to
1285 * eat that input.
1286 *
1287 * Otherwise we would not be able to test that we
1288 * consumed all the input to reach the expected size;
1289 * we also want to check that zlib tells us that all
1290 * went well with status == Z_STREAM_END at the end.
1291 */
1292 stream->next_out = buf + bytes;
1293 stream->avail_out = size - bytes;
1294 while (status == Z_OK)
1295 status = git_inflate(stream, Z_FINISH);
1296 }
1297 if (status == Z_STREAM_END && !stream->avail_in) {
1298 git_inflate_end(stream);
1299 return buf;
1300 }
1301
1302 if (status < 0)
1303 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1304 else if (stream->avail_in)
1305 error("garbage at end of loose object '%s'",
1306 sha1_to_hex(sha1));
1307 free(buf);
1308 return NULL;
1309 }
1310
1311 /*
1312 * We used to just use "sscanf()", but that's actually way
1313 * too permissive for what we want to check. So do an anal
1314 * object header parse by hand.
1315 */
1316 static int parse_sha1_header(const char *hdr, unsigned long *sizep)
1317 {
1318 char type[10];
1319 int i;
1320 unsigned long size;
1321
1322 /*
1323 * The type can be at most ten bytes (including the
1324 * terminating '\0' that we add), and is followed by
1325 * a space.
1326 */
1327 i = 0;
1328 for (;;) {
1329 char c = *hdr++;
1330 if (c == ' ')
1331 break;
1332 type[i++] = c;
1333 if (i >= sizeof(type))
1334 return -1;
1335 }
1336 type[i] = 0;
1337
1338 /*
1339 * The length must follow immediately, and be in canonical
1340 * decimal format (ie "010" is not valid).
1341 */
1342 size = *hdr++ - '0';
1343 if (size > 9)
1344 return -1;
1345 if (size) {
1346 for (;;) {
1347 unsigned long c = *hdr - '0';
1348 if (c > 9)
1349 break;
1350 hdr++;
1351 size = size * 10 + c;
1352 }
1353 }
1354 *sizep = size;
1355
1356 /*
1357 * The length must be followed by a zero byte
1358 */
1359 return *hdr ? -1 : type_from_string(type);
1360 }
1361
1362 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1363 {
1364 int ret;
1365 z_stream stream;
1366 char hdr[8192];
1367
1368 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1369 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1370 return NULL;
1371
1372 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1373 }
1374
1375 unsigned long get_size_from_delta(struct packed_git *p,
1376 struct pack_window **w_curs,
1377 off_t curpos)
1378 {
1379 const unsigned char *data;
1380 unsigned char delta_head[20], *in;
1381 z_stream stream;
1382 int st;
1383
1384 memset(&stream, 0, sizeof(stream));
1385 stream.next_out = delta_head;
1386 stream.avail_out = sizeof(delta_head);
1387
1388 git_inflate_init(&stream);
1389 do {
1390 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1391 stream.next_in = in;
1392 st = git_inflate(&stream, Z_FINISH);
1393 curpos += stream.next_in - in;
1394 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1395 stream.total_out < sizeof(delta_head));
1396 git_inflate_end(&stream);
1397 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1398 error("delta data unpack-initial failed");
1399 return 0;
1400 }
1401
1402 /* Examine the initial part of the delta to figure out
1403 * the result size.
1404 */
1405 data = delta_head;
1406
1407 /* ignore base size */
1408 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1409
1410 /* Read the result size */
1411 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1412 }
1413
1414 static off_t get_delta_base(struct packed_git *p,
1415 struct pack_window **w_curs,
1416 off_t *curpos,
1417 enum object_type type,
1418 off_t delta_obj_offset)
1419 {
1420 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1421 off_t base_offset;
1422
1423 /* use_pack() assured us we have [base_info, base_info + 20)
1424 * as a range that we can look at without walking off the
1425 * end of the mapped window. Its actually the hash size
1426 * that is assured. An OFS_DELTA longer than the hash size
1427 * is stupid, as then a REF_DELTA would be smaller to store.
1428 */
1429 if (type == OBJ_OFS_DELTA) {
1430 unsigned used = 0;
1431 unsigned char c = base_info[used++];
1432 base_offset = c & 127;
1433 while (c & 128) {
1434 base_offset += 1;
1435 if (!base_offset || MSB(base_offset, 7))
1436 return 0; /* overflow */
1437 c = base_info[used++];
1438 base_offset = (base_offset << 7) + (c & 127);
1439 }
1440 base_offset = delta_obj_offset - base_offset;
1441 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1442 return 0; /* out of bound */
1443 *curpos += used;
1444 } else if (type == OBJ_REF_DELTA) {
1445 /* The base entry _must_ be in the same pack */
1446 base_offset = find_pack_entry_one(base_info, p);
1447 *curpos += 20;
1448 } else
1449 die("I am totally screwed");
1450 return base_offset;
1451 }
1452
1453 /* forward declaration for a mutually recursive function */
1454 static int packed_object_info(struct packed_git *p, off_t offset,
1455 unsigned long *sizep);
1456
1457 static int packed_delta_info(struct packed_git *p,
1458 struct pack_window **w_curs,
1459 off_t curpos,
1460 enum object_type type,
1461 off_t obj_offset,
1462 unsigned long *sizep)
1463 {
1464 off_t base_offset;
1465
1466 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1467 if (!base_offset)
1468 return OBJ_BAD;
1469 type = packed_object_info(p, base_offset, NULL);
1470 if (type <= OBJ_NONE) {
1471 struct revindex_entry *revidx;
1472 const unsigned char *base_sha1;
1473 revidx = find_pack_revindex(p, base_offset);
1474 if (!revidx)
1475 return OBJ_BAD;
1476 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1477 mark_bad_packed_object(p, base_sha1);
1478 type = sha1_object_info(base_sha1, NULL);
1479 if (type <= OBJ_NONE)
1480 return OBJ_BAD;
1481 }
1482
1483 /* We choose to only get the type of the base object and
1484 * ignore potentially corrupt pack file that expects the delta
1485 * based on a base with a wrong size. This saves tons of
1486 * inflate() calls.
1487 */
1488 if (sizep) {
1489 *sizep = get_size_from_delta(p, w_curs, curpos);
1490 if (*sizep == 0)
1491 type = OBJ_BAD;
1492 }
1493
1494 return type;
1495 }
1496
1497 static int unpack_object_header(struct packed_git *p,
1498 struct pack_window **w_curs,
1499 off_t *curpos,
1500 unsigned long *sizep)
1501 {
1502 unsigned char *base;
1503 unsigned int left;
1504 unsigned long used;
1505 enum object_type type;
1506
1507 /* use_pack() assures us we have [base, base + 20) available
1508 * as a range that we can look at at. (Its actually the hash
1509 * size that is assured.) With our object header encoding
1510 * the maximum deflated object size is 2^137, which is just
1511 * insane, so we know won't exceed what we have been given.
1512 */
1513 base = use_pack(p, w_curs, *curpos, &left);
1514 used = unpack_object_header_buffer(base, left, &type, sizep);
1515 if (!used) {
1516 type = OBJ_BAD;
1517 } else
1518 *curpos += used;
1519
1520 return type;
1521 }
1522
1523 const char *packed_object_info_detail(struct packed_git *p,
1524 off_t obj_offset,
1525 unsigned long *size,
1526 unsigned long *store_size,
1527 unsigned int *delta_chain_length,
1528 unsigned char *base_sha1)
1529 {
1530 struct pack_window *w_curs = NULL;
1531 off_t curpos;
1532 unsigned long dummy;
1533 unsigned char *next_sha1;
1534 enum object_type type;
1535 struct revindex_entry *revidx;
1536
1537 *delta_chain_length = 0;
1538 curpos = obj_offset;
1539 type = unpack_object_header(p, &w_curs, &curpos, size);
1540
1541 revidx = find_pack_revindex(p, obj_offset);
1542 *store_size = revidx[1].offset - obj_offset;
1543
1544 for (;;) {
1545 switch (type) {
1546 default:
1547 die("pack %s contains unknown object type %d",
1548 p->pack_name, type);
1549 case OBJ_COMMIT:
1550 case OBJ_TREE:
1551 case OBJ_BLOB:
1552 case OBJ_TAG:
1553 unuse_pack(&w_curs);
1554 return typename(type);
1555 case OBJ_OFS_DELTA:
1556 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1557 if (!obj_offset)
1558 die("pack %s contains bad delta base reference of type %s",
1559 p->pack_name, typename(type));
1560 if (*delta_chain_length == 0) {
1561 revidx = find_pack_revindex(p, obj_offset);
1562 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1563 }
1564 break;
1565 case OBJ_REF_DELTA:
1566 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1567 if (*delta_chain_length == 0)
1568 hashcpy(base_sha1, next_sha1);
1569 obj_offset = find_pack_entry_one(next_sha1, p);
1570 break;
1571 }
1572 (*delta_chain_length)++;
1573 curpos = obj_offset;
1574 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1575 }
1576 }
1577
1578 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1579 unsigned long *sizep)
1580 {
1581 struct pack_window *w_curs = NULL;
1582 unsigned long size;
1583 off_t curpos = obj_offset;
1584 enum object_type type;
1585
1586 type = unpack_object_header(p, &w_curs, &curpos, &size);
1587
1588 switch (type) {
1589 case OBJ_OFS_DELTA:
1590 case OBJ_REF_DELTA:
1591 type = packed_delta_info(p, &w_curs, curpos,
1592 type, obj_offset, sizep);
1593 break;
1594 case OBJ_COMMIT:
1595 case OBJ_TREE:
1596 case OBJ_BLOB:
1597 case OBJ_TAG:
1598 if (sizep)
1599 *sizep = size;
1600 break;
1601 default:
1602 error("unknown object type %i at offset %"PRIuMAX" in %s",
1603 type, (uintmax_t)obj_offset, p->pack_name);
1604 type = OBJ_BAD;
1605 }
1606 unuse_pack(&w_curs);
1607 return type;
1608 }
1609
1610 static void *unpack_compressed_entry(struct packed_git *p,
1611 struct pack_window **w_curs,
1612 off_t curpos,
1613 unsigned long size)
1614 {
1615 int st;
1616 z_stream stream;
1617 unsigned char *buffer, *in;
1618
1619 buffer = xmallocz(size);
1620 memset(&stream, 0, sizeof(stream));
1621 stream.next_out = buffer;
1622 stream.avail_out = size + 1;
1623
1624 git_inflate_init(&stream);
1625 do {
1626 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1627 stream.next_in = in;
1628 st = git_inflate(&stream, Z_FINISH);
1629 if (!stream.avail_out)
1630 break; /* the payload is larger than it should be */
1631 curpos += stream.next_in - in;
1632 } while (st == Z_OK || st == Z_BUF_ERROR);
1633 git_inflate_end(&stream);
1634 if ((st != Z_STREAM_END) || stream.total_out != size) {
1635 free(buffer);
1636 return NULL;
1637 }
1638
1639 return buffer;
1640 }
1641
1642 #define MAX_DELTA_CACHE (256)
1643
1644 static size_t delta_base_cached;
1645
1646 static struct delta_base_cache_lru_list {
1647 struct delta_base_cache_lru_list *prev;
1648 struct delta_base_cache_lru_list *next;
1649 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1650
1651 static struct delta_base_cache_entry {
1652 struct delta_base_cache_lru_list lru;
1653 void *data;
1654 struct packed_git *p;
1655 off_t base_offset;
1656 unsigned long size;
1657 enum object_type type;
1658 } delta_base_cache[MAX_DELTA_CACHE];
1659
1660 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1661 {
1662 unsigned long hash;
1663
1664 hash = (unsigned long)p + (unsigned long)base_offset;
1665 hash += (hash >> 8) + (hash >> 16);
1666 return hash % MAX_DELTA_CACHE;
1667 }
1668
1669 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1670 unsigned long *base_size, enum object_type *type, int keep_cache)
1671 {
1672 void *ret;
1673 unsigned long hash = pack_entry_hash(p, base_offset);
1674 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1675
1676 ret = ent->data;
1677 if (!ret || ent->p != p || ent->base_offset != base_offset)
1678 return unpack_entry(p, base_offset, type, base_size);
1679
1680 if (!keep_cache) {
1681 ent->data = NULL;
1682 ent->lru.next->prev = ent->lru.prev;
1683 ent->lru.prev->next = ent->lru.next;
1684 delta_base_cached -= ent->size;
1685 } else {
1686 ret = xmemdupz(ent->data, ent->size);
1687 }
1688 *type = ent->type;
1689 *base_size = ent->size;
1690 return ret;
1691 }
1692
1693 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1694 {
1695 if (ent->data) {
1696 free(ent->data);
1697 ent->data = NULL;
1698 ent->lru.next->prev = ent->lru.prev;
1699 ent->lru.prev->next = ent->lru.next;
1700 delta_base_cached -= ent->size;
1701 }
1702 }
1703
1704 void clear_delta_base_cache(void)
1705 {
1706 unsigned long p;
1707 for (p = 0; p < MAX_DELTA_CACHE; p++)
1708 release_delta_base_cache(&delta_base_cache[p]);
1709 }
1710
1711 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1712 void *base, unsigned long base_size, enum object_type type)
1713 {
1714 unsigned long hash = pack_entry_hash(p, base_offset);
1715 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1716 struct delta_base_cache_lru_list *lru;
1717
1718 release_delta_base_cache(ent);
1719 delta_base_cached += base_size;
1720
1721 for (lru = delta_base_cache_lru.next;
1722 delta_base_cached > delta_base_cache_limit
1723 && lru != &delta_base_cache_lru;
1724 lru = lru->next) {
1725 struct delta_base_cache_entry *f = (void *)lru;
1726 if (f->type == OBJ_BLOB)
1727 release_delta_base_cache(f);
1728 }
1729 for (lru = delta_base_cache_lru.next;
1730 delta_base_cached > delta_base_cache_limit
1731 && lru != &delta_base_cache_lru;
1732 lru = lru->next) {
1733 struct delta_base_cache_entry *f = (void *)lru;
1734 release_delta_base_cache(f);
1735 }
1736
1737 ent->p = p;
1738 ent->base_offset = base_offset;
1739 ent->type = type;
1740 ent->data = base;
1741 ent->size = base_size;
1742 ent->lru.next = &delta_base_cache_lru;
1743 ent->lru.prev = delta_base_cache_lru.prev;
1744 delta_base_cache_lru.prev->next = &ent->lru;
1745 delta_base_cache_lru.prev = &ent->lru;
1746 }
1747
1748 static void *read_object(const unsigned char *sha1, enum object_type *type,
1749 unsigned long *size);
1750
1751 static void *unpack_delta_entry(struct packed_git *p,
1752 struct pack_window **w_curs,
1753 off_t curpos,
1754 unsigned long delta_size,
1755 off_t obj_offset,
1756 enum object_type *type,
1757 unsigned long *sizep)
1758 {
1759 void *delta_data, *result, *base;
1760 unsigned long base_size;
1761 off_t base_offset;
1762
1763 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1764 if (!base_offset) {
1765 error("failed to validate delta base reference "
1766 "at offset %"PRIuMAX" from %s",
1767 (uintmax_t)curpos, p->pack_name);
1768 return NULL;
1769 }
1770 unuse_pack(w_curs);
1771 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1772 if (!base) {
1773 /*
1774 * We're probably in deep shit, but let's try to fetch
1775 * the required base anyway from another pack or loose.
1776 * This is costly but should happen only in the presence
1777 * of a corrupted pack, and is better than failing outright.
1778 */
1779 struct revindex_entry *revidx;
1780 const unsigned char *base_sha1;
1781 revidx = find_pack_revindex(p, base_offset);
1782 if (!revidx)
1783 return NULL;
1784 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1785 error("failed to read delta base object %s"
1786 " at offset %"PRIuMAX" from %s",
1787 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1788 p->pack_name);
1789 mark_bad_packed_object(p, base_sha1);
1790 base = read_object(base_sha1, type, &base_size);
1791 if (!base)
1792 return NULL;
1793 }
1794
1795 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1796 if (!delta_data) {
1797 error("failed to unpack compressed delta "
1798 "at offset %"PRIuMAX" from %s",
1799 (uintmax_t)curpos, p->pack_name);
1800 free(base);
1801 return NULL;
1802 }
1803 result = patch_delta(base, base_size,
1804 delta_data, delta_size,
1805 sizep);
1806 if (!result)
1807 die("failed to apply delta");
1808 free(delta_data);
1809 add_delta_base_cache(p, base_offset, base, base_size, *type);
1810 return result;
1811 }
1812
1813 int do_check_packed_object_crc;
1814
1815 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1816 enum object_type *type, unsigned long *sizep)
1817 {
1818 struct pack_window *w_curs = NULL;
1819 off_t curpos = obj_offset;
1820 void *data;
1821
1822 if (do_check_packed_object_crc && p->index_version > 1) {
1823 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1824 unsigned long len = revidx[1].offset - obj_offset;
1825 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1826 const unsigned char *sha1 =
1827 nth_packed_object_sha1(p, revidx->nr);
1828 error("bad packed object CRC for %s",
1829 sha1_to_hex(sha1));
1830 mark_bad_packed_object(p, sha1);
1831 unuse_pack(&w_curs);
1832 return NULL;
1833 }
1834 }
1835
1836 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1837 switch (*type) {
1838 case OBJ_OFS_DELTA:
1839 case OBJ_REF_DELTA:
1840 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1841 obj_offset, type, sizep);
1842 break;
1843 case OBJ_COMMIT:
1844 case OBJ_TREE:
1845 case OBJ_BLOB:
1846 case OBJ_TAG:
1847 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1848 break;
1849 default:
1850 data = NULL;
1851 error("unknown object type %i at offset %"PRIuMAX" in %s",
1852 *type, (uintmax_t)obj_offset, p->pack_name);
1853 }
1854 unuse_pack(&w_curs);
1855 return data;
1856 }
1857
1858 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1859 uint32_t n)
1860 {
1861 const unsigned char *index = p->index_data;
1862 if (!index) {
1863 if (open_pack_index(p))
1864 return NULL;
1865 index = p->index_data;
1866 }
1867 if (n >= p->num_objects)
1868 return NULL;
1869 index += 4 * 256;
1870 if (p->index_version == 1) {
1871 return index + 24 * n + 4;
1872 } else {
1873 index += 8;
1874 return index + 20 * n;
1875 }
1876 }
1877
1878 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1879 {
1880 const unsigned char *index = p->index_data;
1881 index += 4 * 256;
1882 if (p->index_version == 1) {
1883 return ntohl(*((uint32_t *)(index + 24 * n)));
1884 } else {
1885 uint32_t off;
1886 index += 8 + p->num_objects * (20 + 4);
1887 off = ntohl(*((uint32_t *)(index + 4 * n)));
1888 if (!(off & 0x80000000))
1889 return off;
1890 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1891 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1892 ntohl(*((uint32_t *)(index + 4)));
1893 }
1894 }
1895
1896 off_t find_pack_entry_one(const unsigned char *sha1,
1897 struct packed_git *p)
1898 {
1899 const uint32_t *level1_ofs = p->index_data;
1900 const unsigned char *index = p->index_data;
1901 unsigned hi, lo, stride;
1902 static int use_lookup = -1;
1903 static int debug_lookup = -1;
1904
1905 if (debug_lookup < 0)
1906 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1907
1908 if (!index) {
1909 if (open_pack_index(p))
1910 return 0;
1911 level1_ofs = p->index_data;
1912 index = p->index_data;
1913 }
1914 if (p->index_version > 1) {
1915 level1_ofs += 2;
1916 index += 8;
1917 }
1918 index += 4 * 256;
1919 hi = ntohl(level1_ofs[*sha1]);
1920 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1921 if (p->index_version > 1) {
1922 stride = 20;
1923 } else {
1924 stride = 24;
1925 index += 4;
1926 }
1927
1928 if (debug_lookup)
1929 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1930 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1931
1932 if (use_lookup < 0)
1933 use_lookup = !!getenv("GIT_USE_LOOKUP");
1934 if (use_lookup) {
1935 int pos = sha1_entry_pos(index, stride, 0,
1936 lo, hi, p->num_objects, sha1);
1937 if (pos < 0)
1938 return 0;
1939 return nth_packed_object_offset(p, pos);
1940 }
1941
1942 do {
1943 unsigned mi = (lo + hi) / 2;
1944 int cmp = hashcmp(index + mi * stride, sha1);
1945
1946 if (debug_lookup)
1947 printf("lo %u hi %u rg %u mi %u\n",
1948 lo, hi, hi - lo, mi);
1949 if (!cmp)
1950 return nth_packed_object_offset(p, mi);
1951 if (cmp > 0)
1952 hi = mi;
1953 else
1954 lo = mi+1;
1955 } while (lo < hi);
1956 return 0;
1957 }
1958
1959 static int is_pack_valid(struct packed_git *p)
1960 {
1961 /* An already open pack is known to be valid. */
1962 if (p->pack_fd != -1)
1963 return 1;
1964
1965 /* If the pack has one window completely covering the
1966 * file size, the pack is known to be valid even if
1967 * the descriptor is not currently open.
1968 */
1969 if (p->windows) {
1970 struct pack_window *w = p->windows;
1971
1972 if (!w->offset && w->len == p->pack_size)
1973 return 1;
1974 }
1975
1976 /* Force the pack to open to prove its valid. */
1977 return !open_packed_git(p);
1978 }
1979
1980 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1981 {
1982 static struct packed_git *last_found = (void *)1;
1983 struct packed_git *p;
1984 off_t offset;
1985
1986 prepare_packed_git();
1987 if (!packed_git)
1988 return 0;
1989 p = (last_found == (void *)1) ? packed_git : last_found;
1990
1991 do {
1992 if (p->num_bad_objects) {
1993 unsigned i;
1994 for (i = 0; i < p->num_bad_objects; i++)
1995 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1996 goto next;
1997 }
1998
1999 offset = find_pack_entry_one(sha1, p);
2000 if (offset) {
2001 /*
2002 * We are about to tell the caller where they can
2003 * locate the requested object. We better make
2004 * sure the packfile is still here and can be
2005 * accessed before supplying that answer, as
2006 * it may have been deleted since the index
2007 * was loaded!
2008 */
2009 if (!is_pack_valid(p)) {
2010 error("packfile %s cannot be accessed", p->pack_name);
2011 goto next;
2012 }
2013 e->offset = offset;
2014 e->p = p;
2015 hashcpy(e->sha1, sha1);
2016 last_found = p;
2017 return 1;
2018 }
2019
2020 next:
2021 if (p == last_found)
2022 p = packed_git;
2023 else
2024 p = p->next;
2025 if (p == last_found)
2026 p = p->next;
2027 } while (p);
2028 return 0;
2029 }
2030
2031 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2032 struct packed_git *packs)
2033 {
2034 struct packed_git *p;
2035
2036 for (p = packs; p; p = p->next) {
2037 if (find_pack_entry_one(sha1, p))
2038 return p;
2039 }
2040 return NULL;
2041
2042 }
2043
2044 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2045 {
2046 int status;
2047 unsigned long mapsize, size;
2048 void *map;
2049 z_stream stream;
2050 char hdr[32];
2051
2052 map = map_sha1_file(sha1, &mapsize);
2053 if (!map)
2054 return error("unable to find %s", sha1_to_hex(sha1));
2055 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2056 status = error("unable to unpack %s header",
2057 sha1_to_hex(sha1));
2058 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2059 status = error("unable to parse %s header", sha1_to_hex(sha1));
2060 else if (sizep)
2061 *sizep = size;
2062 git_inflate_end(&stream);
2063 munmap(map, mapsize);
2064 return status;
2065 }
2066
2067 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2068 {
2069 struct cached_object *co;
2070 struct pack_entry e;
2071 int status;
2072
2073 co = find_cached_object(sha1);
2074 if (co) {
2075 if (sizep)
2076 *sizep = co->size;
2077 return co->type;
2078 }
2079
2080 if (!find_pack_entry(sha1, &e)) {
2081 /* Most likely it's a loose object. */
2082 status = sha1_loose_object_info(sha1, sizep);
2083 if (status >= 0)
2084 return status;
2085
2086 /* Not a loose object; someone else may have just packed it. */
2087 reprepare_packed_git();
2088 if (!find_pack_entry(sha1, &e))
2089 return status;
2090 }
2091
2092 status = packed_object_info(e.p, e.offset, sizep);
2093 if (status < 0) {
2094 mark_bad_packed_object(e.p, sha1);
2095 status = sha1_object_info(sha1, sizep);
2096 }
2097
2098 return status;
2099 }
2100
2101 static void *read_packed_sha1(const unsigned char *sha1,
2102 enum object_type *type, unsigned long *size)
2103 {
2104 struct pack_entry e;
2105 void *data;
2106
2107 if (!find_pack_entry(sha1, &e))
2108 return NULL;
2109 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2110 if (!data) {
2111 /*
2112 * We're probably in deep shit, but let's try to fetch
2113 * the required object anyway from another pack or loose.
2114 * This should happen only in the presence of a corrupted
2115 * pack, and is better than failing outright.
2116 */
2117 error("failed to read object %s at offset %"PRIuMAX" from %s",
2118 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2119 mark_bad_packed_object(e.p, sha1);
2120 data = read_object(sha1, type, size);
2121 }
2122 return data;
2123 }
2124
2125 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2126 unsigned char *sha1)
2127 {
2128 struct cached_object *co;
2129
2130 hash_sha1_file(buf, len, typename(type), sha1);
2131 if (has_sha1_file(sha1) || find_cached_object(sha1))
2132 return 0;
2133 if (cached_object_alloc <= cached_object_nr) {
2134 cached_object_alloc = alloc_nr(cached_object_alloc);
2135 cached_objects = xrealloc(cached_objects,
2136 sizeof(*cached_objects) *
2137 cached_object_alloc);
2138 }
2139 co = &cached_objects[cached_object_nr++];
2140 co->size = len;
2141 co->type = type;
2142 co->buf = xmalloc(len);
2143 memcpy(co->buf, buf, len);
2144 hashcpy(co->sha1, sha1);
2145 return 0;
2146 }
2147
2148 static void *read_object(const unsigned char *sha1, enum object_type *type,
2149 unsigned long *size)
2150 {
2151 unsigned long mapsize;
2152 void *map, *buf;
2153 struct cached_object *co;
2154
2155 co = find_cached_object(sha1);
2156 if (co) {
2157 *type = co->type;
2158 *size = co->size;
2159 return xmemdupz(co->buf, co->size);
2160 }
2161
2162 buf = read_packed_sha1(sha1, type, size);
2163 if (buf)
2164 return buf;
2165 map = map_sha1_file(sha1, &mapsize);
2166 if (map) {
2167 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2168 munmap(map, mapsize);
2169 return buf;
2170 }
2171 reprepare_packed_git();
2172 return read_packed_sha1(sha1, type, size);
2173 }
2174
2175 /*
2176 * This function dies on corrupt objects; the callers who want to
2177 * deal with them should arrange to call read_object() and give error
2178 * messages themselves.
2179 */
2180 void *read_sha1_file_repl(const unsigned char *sha1,
2181 enum object_type *type,
2182 unsigned long *size,
2183 const unsigned char **replacement)
2184 {
2185 const unsigned char *repl = lookup_replace_object(sha1);
2186 void *data;
2187 char *path;
2188 const struct packed_git *p;
2189
2190 errno = 0;
2191 data = read_object(repl, type, size);
2192 if (data) {
2193 if (replacement)
2194 *replacement = repl;
2195 return data;
2196 }
2197
2198 if (errno && errno != ENOENT)
2199 die_errno("failed to read object %s", sha1_to_hex(sha1));
2200
2201 /* die if we replaced an object with one that does not exist */
2202 if (repl != sha1)
2203 die("replacement %s not found for %s",
2204 sha1_to_hex(repl), sha1_to_hex(sha1));
2205
2206 if (has_loose_object(repl)) {
2207 path = sha1_file_name(sha1);
2208 die("loose object %s (stored in %s) is corrupt",
2209 sha1_to_hex(repl), path);
2210 }
2211
2212 if ((p = has_packed_and_bad(repl)) != NULL)
2213 die("packed object %s (stored in %s) is corrupt",
2214 sha1_to_hex(repl), p->pack_name);
2215
2216 return NULL;
2217 }
2218
2219 void *read_object_with_reference(const unsigned char *sha1,
2220 const char *required_type_name,
2221 unsigned long *size,
2222 unsigned char *actual_sha1_return)
2223 {
2224 enum object_type type, required_type;
2225 void *buffer;
2226 unsigned long isize;
2227 unsigned char actual_sha1[20];
2228
2229 required_type = type_from_string(required_type_name);
2230 hashcpy(actual_sha1, sha1);
2231 while (1) {
2232 int ref_length = -1;
2233 const char *ref_type = NULL;
2234
2235 buffer = read_sha1_file(actual_sha1, &type, &isize);
2236 if (!buffer)
2237 return NULL;
2238 if (type == required_type) {
2239 *size = isize;
2240 if (actual_sha1_return)
2241 hashcpy(actual_sha1_return, actual_sha1);
2242 return buffer;
2243 }
2244 /* Handle references */
2245 else if (type == OBJ_COMMIT)
2246 ref_type = "tree ";
2247 else if (type == OBJ_TAG)
2248 ref_type = "object ";
2249 else {
2250 free(buffer);
2251 return NULL;
2252 }
2253 ref_length = strlen(ref_type);
2254
2255 if (ref_length + 40 > isize ||
2256 memcmp(buffer, ref_type, ref_length) ||
2257 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2258 free(buffer);
2259 return NULL;
2260 }
2261 free(buffer);
2262 /* Now we have the ID of the referred-to object in
2263 * actual_sha1. Check again. */
2264 }
2265 }
2266
2267 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2268 const char *type, unsigned char *sha1,
2269 char *hdr, int *hdrlen)
2270 {
2271 git_SHA_CTX c;
2272
2273 /* Generate the header */
2274 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2275
2276 /* Sha1.. */
2277 git_SHA1_Init(&c);
2278 git_SHA1_Update(&c, hdr, *hdrlen);
2279 git_SHA1_Update(&c, buf, len);
2280 git_SHA1_Final(sha1, &c);
2281 }
2282
2283 /*
2284 * Move the just written object into its final resting place.
2285 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2286 * "moving" is only a part of what it does, when no patch between
2287 * master to pu changes the call sites of this function.
2288 */
2289 int move_temp_to_file(const char *tmpfile, const char *filename)
2290 {
2291 int ret = 0;
2292
2293 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2294 goto try_rename;
2295 else if (link(tmpfile, filename))
2296 ret = errno;
2297
2298 /*
2299 * Coda hack - coda doesn't like cross-directory links,
2300 * so we fall back to a rename, which will mean that it
2301 * won't be able to check collisions, but that's not a
2302 * big deal.
2303 *
2304 * The same holds for FAT formatted media.
2305 *
2306 * When this succeeds, we just return. We have nothing
2307 * left to unlink.
2308 */
2309 if (ret && ret != EEXIST) {
2310 try_rename:
2311 if (!rename(tmpfile, filename))
2312 goto out;
2313 ret = errno;
2314 }
2315 unlink_or_warn(tmpfile);
2316 if (ret) {
2317 if (ret != EEXIST) {
2318 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2319 }
2320 /* FIXME!!! Collision check here ? */
2321 }
2322
2323 out:
2324 if (adjust_shared_perm(filename))
2325 return error("unable to set permission to '%s'", filename);
2326 return 0;
2327 }
2328
2329 static int write_buffer(int fd, const void *buf, size_t len)
2330 {
2331 if (write_in_full(fd, buf, len) < 0)
2332 return error("file write error (%s)", strerror(errno));
2333 return 0;
2334 }
2335
2336 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2337 unsigned char *sha1)
2338 {
2339 char hdr[32];
2340 int hdrlen;
2341 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2342 return 0;
2343 }
2344
2345 /* Finalize a file on disk, and close it. */
2346 static void close_sha1_file(int fd)
2347 {
2348 if (fsync_object_files)
2349 fsync_or_die(fd, "sha1 file");
2350 if (close(fd) != 0)
2351 die_errno("error when closing sha1 file");
2352 }
2353
2354 /* Size of directory component, including the ending '/' */
2355 static inline int directory_size(const char *filename)
2356 {
2357 const char *s = strrchr(filename, '/');
2358 if (!s)
2359 return 0;
2360 return s - filename + 1;
2361 }
2362
2363 /*
2364 * This creates a temporary file in the same directory as the final
2365 * 'filename'
2366 *
2367 * We want to avoid cross-directory filename renames, because those
2368 * can have problems on various filesystems (FAT, NFS, Coda).
2369 */
2370 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2371 {
2372 int fd, dirlen = directory_size(filename);
2373
2374 if (dirlen + 20 > bufsiz) {
2375 errno = ENAMETOOLONG;
2376 return -1;
2377 }
2378 memcpy(buffer, filename, dirlen);
2379 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2380 fd = git_mkstemp_mode(buffer, 0444);
2381 if (fd < 0 && dirlen && errno == ENOENT) {
2382 /* Make sure the directory exists */
2383 memcpy(buffer, filename, dirlen);
2384 buffer[dirlen-1] = 0;
2385 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2386 return -1;
2387
2388 /* Try again */
2389 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2390 fd = git_mkstemp_mode(buffer, 0444);
2391 }
2392 return fd;
2393 }
2394
2395 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2396 const void *buf, unsigned long len, time_t mtime)
2397 {
2398 int fd, ret;
2399 unsigned char compressed[4096];
2400 z_stream stream;
2401 git_SHA_CTX c;
2402 unsigned char parano_sha1[20];
2403 char *filename;
2404 static char tmpfile[PATH_MAX];
2405
2406 filename = sha1_file_name(sha1);
2407 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2408 if (fd < 0) {
2409 if (errno == EACCES)
2410 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2411 else
2412 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2413 }
2414
2415 /* Set it up */
2416 memset(&stream, 0, sizeof(stream));
2417 deflateInit(&stream, zlib_compression_level);
2418 stream.next_out = compressed;
2419 stream.avail_out = sizeof(compressed);
2420 git_SHA1_Init(&c);
2421
2422 /* First header.. */
2423 stream.next_in = (unsigned char *)hdr;
2424 stream.avail_in = hdrlen;
2425 while (deflate(&stream, 0) == Z_OK)
2426 /* nothing */;
2427 git_SHA1_Update(&c, hdr, hdrlen);
2428
2429 /* Then the data itself.. */
2430 stream.next_in = (void *)buf;
2431 stream.avail_in = len;
2432 do {
2433 unsigned char *in0 = stream.next_in;
2434 ret = deflate(&stream, Z_FINISH);
2435 git_SHA1_Update(&c, in0, stream.next_in - in0);
2436 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2437 die("unable to write sha1 file");
2438 stream.next_out = compressed;
2439 stream.avail_out = sizeof(compressed);
2440 } while (ret == Z_OK);
2441
2442 if (ret != Z_STREAM_END)
2443 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2444 ret = deflateEnd(&stream);
2445 if (ret != Z_OK)
2446 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2447 git_SHA1_Final(parano_sha1, &c);
2448 if (hashcmp(sha1, parano_sha1) != 0)
2449 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2450
2451 close_sha1_file(fd);
2452
2453 if (mtime) {
2454 struct utimbuf utb;
2455 utb.actime = mtime;
2456 utb.modtime = mtime;
2457 if (utime(tmpfile, &utb) < 0)
2458 warning("failed utime() on %s: %s",
2459 tmpfile, strerror(errno));
2460 }
2461
2462 return move_temp_to_file(tmpfile, filename);
2463 }
2464
2465 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2466 {
2467 unsigned char sha1[20];
2468 char hdr[32];
2469 int hdrlen;
2470
2471 /* Normally if we have it in the pack then we do not bother writing
2472 * it out into .git/objects/??/?{38} file.
2473 */
2474 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2475 if (returnsha1)
2476 hashcpy(returnsha1, sha1);
2477 if (has_sha1_file(sha1))
2478 return 0;
2479 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2480 }
2481
2482 int force_object_loose(const unsigned char *sha1, time_t mtime)
2483 {
2484 void *buf;
2485 unsigned long len;
2486 enum object_type type;
2487 char hdr[32];
2488 int hdrlen;
2489 int ret;
2490
2491 if (has_loose_object(sha1))
2492 return 0;
2493 buf = read_packed_sha1(sha1, &type, &len);
2494 if (!buf)
2495 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2496 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2497 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2498 free(buf);
2499
2500 return ret;
2501 }
2502
2503 int has_pack_index(const unsigned char *sha1)
2504 {
2505 struct stat st;
2506 if (stat(sha1_pack_index_name(sha1), &st))
2507 return 0;
2508 return 1;
2509 }
2510
2511 int has_sha1_pack(const unsigned char *sha1)
2512 {
2513 struct pack_entry e;
2514 return find_pack_entry(sha1, &e);
2515 }
2516
2517 int has_sha1_file(const unsigned char *sha1)
2518 {
2519 struct pack_entry e;
2520
2521 if (find_pack_entry(sha1, &e))
2522 return 1;
2523 return has_loose_object(sha1);
2524 }
2525
2526 static void check_tree(const void *buf, size_t size)
2527 {
2528 struct tree_desc desc;
2529 struct name_entry entry;
2530
2531 init_tree_desc(&desc, buf, size);
2532 while (tree_entry(&desc, &entry))
2533 /* do nothing
2534 * tree_entry() will die() on malformed entries */
2535 ;
2536 }
2537
2538 static void check_commit(const void *buf, size_t size)
2539 {
2540 struct commit c;
2541 memset(&c, 0, sizeof(c));
2542 if (parse_commit_buffer(&c, buf, size))
2543 die("corrupt commit");
2544 }
2545
2546 static void check_tag(const void *buf, size_t size)
2547 {
2548 struct tag t;
2549 memset(&t, 0, sizeof(t));
2550 if (parse_tag_buffer(&t, buf, size))
2551 die("corrupt tag");
2552 }
2553
2554 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2555 int write_object, enum object_type type,
2556 const char *path, int format_check)
2557 {
2558 int ret, re_allocated = 0;
2559
2560 if (!type)
2561 type = OBJ_BLOB;
2562
2563 /*
2564 * Convert blobs to git internal format
2565 */
2566 if ((type == OBJ_BLOB) && path) {
2567 struct strbuf nbuf = STRBUF_INIT;
2568 if (convert_to_git(path, buf, size, &nbuf,
2569 write_object ? safe_crlf : 0)) {
2570 buf = strbuf_detach(&nbuf, &size);
2571 re_allocated = 1;
2572 }
2573 }
2574 if (format_check) {
2575 if (type == OBJ_TREE)
2576 check_tree(buf, size);
2577 if (type == OBJ_COMMIT)
2578 check_commit(buf, size);
2579 if (type == OBJ_TAG)
2580 check_tag(buf, size);
2581 }
2582
2583 if (write_object)
2584 ret = write_sha1_file(buf, size, typename(type), sha1);
2585 else
2586 ret = hash_sha1_file(buf, size, typename(type), sha1);
2587 if (re_allocated)
2588 free(buf);
2589 return ret;
2590 }
2591
2592 #define SMALL_FILE_SIZE (32*1024)
2593
2594 int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
2595 enum object_type type, const char *path, int format_check)
2596 {
2597 int ret;
2598 size_t size = xsize_t(st->st_size);
2599
2600 if (!S_ISREG(st->st_mode)) {
2601 struct strbuf sbuf = STRBUF_INIT;
2602 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2603 ret = index_mem(sha1, sbuf.buf, sbuf.len, write_object,
2604 type, path, format_check);
2605 else
2606 ret = -1;
2607 strbuf_release(&sbuf);
2608 } else if (!size) {
2609 ret = index_mem(sha1, NULL, size, write_object, type, path,
2610 format_check);
2611 } else if (size <= SMALL_FILE_SIZE) {
2612 char *buf = xmalloc(size);
2613 if (size == read_in_full(fd, buf, size))
2614 ret = index_mem(sha1, buf, size, write_object, type,
2615 path, format_check);
2616 else
2617 ret = error("short read %s", strerror(errno));
2618 free(buf);
2619 } else {
2620 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2621 ret = index_mem(sha1, buf, size, write_object, type, path,
2622 format_check);
2623 munmap(buf, size);
2624 }
2625 close(fd);
2626 return ret;
2627 }
2628
2629 int index_path(unsigned char *sha1, const char *path, struct stat *st, int write_object)
2630 {
2631 int fd;
2632 struct strbuf sb = STRBUF_INIT;
2633
2634 switch (st->st_mode & S_IFMT) {
2635 case S_IFREG:
2636 fd = open(path, O_RDONLY);
2637 if (fd < 0)
2638 return error("open(\"%s\"): %s", path,
2639 strerror(errno));
2640 if (index_fd(sha1, fd, st, write_object, OBJ_BLOB, path, 0) < 0)
2641 return error("%s: failed to insert into database",
2642 path);
2643 break;
2644 case S_IFLNK:
2645 if (strbuf_readlink(&sb, path, st->st_size)) {
2646 char *errstr = strerror(errno);
2647 return error("readlink(\"%s\"): %s", path,
2648 errstr);
2649 }
2650 if (!write_object)
2651 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2652 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2653 return error("%s: failed to insert into database",
2654 path);
2655 strbuf_release(&sb);
2656 break;
2657 case S_IFDIR:
2658 return resolve_gitlink_ref(path, "HEAD", sha1);
2659 default:
2660 return error("%s: unsupported file type", path);
2661 }
2662 return 0;
2663 }
2664
2665 int read_pack_header(int fd, struct pack_header *header)
2666 {
2667 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2668 /* "eof before pack header was fully read" */
2669 return PH_ERROR_EOF;
2670
2671 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2672 /* "protocol error (pack signature mismatch detected)" */
2673 return PH_ERROR_PACK_SIGNATURE;
2674 if (!pack_version_ok(header->hdr_version))
2675 /* "protocol error (pack version unsupported)" */
2676 return PH_ERROR_PROTOCOL;
2677 return 0;
2678 }
2679
2680 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2681 {
2682 enum object_type type = sha1_object_info(sha1, NULL);
2683 if (type < 0)
2684 die("%s is not a valid object", sha1_to_hex(sha1));
2685 if (type != expect)
2686 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2687 typename(expect));
2688 }