![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
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"
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
30 #define SZ_FMT PRIuMAX
31 static inline uintmax_t sz_fmt(size_t s) { return s; }
33 const unsigned char null_sha1[20];
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;
49 static struct cached_object empty_tree = {
50 EMPTY_TREE_SHA1_BIN_LITERAL,
51 OBJ_TREE,
52 "",
53 0
54 };
56 static struct cached_object *find_cached_object(const unsigned char *sha1)
57 {
58 int i;
59 struct cached_object *co = cached_objects;
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 }
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;
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 }
99 int safe_create_leading_directories(char *path)
100 {
101 char *pos = path + offset_1st_component(path);
102 struct stat st;
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 }
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 }
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 }
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;
170 objdir = get_object_directory();
171 len = strlen(objdir);
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 }
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;
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 }
200 buf = *name;
202 for (i = 0; i < 20; i++) {
203 unsigned int val = *sha1++;
204 *buf++ = hex[val >> 4];
205 *buf++ = hex[val & 0xf];
206 }
208 return *base;
209 }
211 char *sha1_pack_name(const unsigned char *sha1)
212 {
213 static char *name, *base;
215 return sha1_get_pack_name(sha1, &name, &base, "pack");
216 }
218 char *sha1_pack_index_name(const unsigned char *sha1)
219 {
220 static char *name, *base;
222 return sha1_get_pack_name(sha1, &name, &base, "idx");
223 }
225 struct alternate_object_database *alt_odb_list;
226 static struct alternate_object_database **alt_odb_tail;
228 static void read_info_alternates(const char * alternates, int depth);
229 static int git_open_noatime(const char *name);
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;
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);
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);
273 ent->name = ent->base + pfxlen + 1;
274 ent->base[pfxlen + 3] = '/';
275 ent->base[pfxlen] = ent->base[entlen-1] = 0;
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 }
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 }
300 /* add the alternate entry */
301 *alt_odb_tail = ent;
302 alt_odb_tail = &(ent->next);
303 ent->next = NULL;
305 /* recursively add alternates */
306 read_info_alternates(ent->base, depth + 1);
308 ent->base[pfxlen] = '/';
310 return 0;
311 }
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;
318 if (depth > 5) {
319 error("%s: ignoring alternate object stores, nesting too deep.",
320 relative_base);
321 return;
322 }
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 }
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;
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);
374 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
376 munmap(map, mapsz);
377 }
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 }
391 void foreach_alt_odb(alt_odb_fn fn, void *cb)
392 {
393 struct alternate_object_database *ent;
395 prepare_alt_odb();
396 for (ent = alt_odb_list; ent; ent = ent->next)
397 if (fn(ent, cb))
398 return;
399 }
401 void prepare_alt_odb(void)
402 {
403 const char *alt;
405 if (alt_odb_tail)
406 return;
408 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
409 if (!alt) alt = "";
411 alt_odb_tail = &alt_odb_list;
412 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
414 read_info_alternates(get_object_directory(), 0);
415 }
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 }
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 }
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 }
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;
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 }
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;
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);
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;
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 }
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 }
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 }
574 int open_pack_index(struct packed_git *p)
575 {
576 char *idx_name;
577 int ret;
579 if (p->index_data)
580 return 0;
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 }
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;
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 }
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;
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 }
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 }
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 }
660 void close_pack_windows(struct packed_git *p)
661 {
662 while (p->windows) {
663 struct pack_window *w = p->windows;
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 }
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 }
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 }
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;
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 }
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;
738 if (!p->index_data && open_pack_index(p))
739 return error("packfile %s index unavailable", p->pack_name);
741 if (!pack_max_fds) {
742 struct rlimit lim;
743 unsigned int max_fds;
745 if (getrlimit(RLIMIT_NOFILE, &lim))
746 die_errno("cannot get RLIMIT_NOFILE");
748 max_fds = lim.rlim_cur;
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 }
757 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
758 ; /* nothing */
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++;
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);
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");
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));
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 }
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 }
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 }
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;
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?)");
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;
862 if (p->pack_fd == -1 && open_packed_git(p))
863 die("packfile %s cannot be accessed", p->pack_name);
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 }
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 }
917 static void try_to_free_pack_memory(size_t size)
918 {
919 release_pack_memory(size, -1);
920 }
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);
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 }
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);
944 strcpy(p->pack_name + path_len, ".keep");
945 if (!access(p->pack_name, F_OK))
946 p->pack_keep = 1;
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 }
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 }
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);
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 }
977 return p;
978 }
980 void install_packed_git(struct packed_git *pack)
981 {
982 if (pack->pack_fd != -1)
983 pack_open_fds++;
985 pack->next = packed_git;
986 packed_git = pack;
987 }
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;
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;
1013 if (!has_extension(de->d_name, ".idx"))
1014 continue;
1016 if (len + namelen + 1 > sizeof(path))
1017 continue;
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 }
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;
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;
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 }
1066 static void rearrange_packed_git(void)
1067 {
1068 struct packed_git **ary, *p;
1069 int i, n;
1071 for (n = 0, p = packed_git; p; p = p->next)
1072 n++;
1073 if (n < 2)
1074 return;
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;
1081 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
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];
1089 free(ary);
1090 }
1092 static int prepare_packed_git_run_once = 0;
1093 void prepare_packed_git(void)
1094 {
1095 struct alternate_object_database *alt;
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 }
1110 void reprepare_packed_git(void)
1111 {
1112 discard_revindex();
1113 prepare_packed_git_run_once = 0;
1114 prepare_packed_git();
1115 }
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 }
1129 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1130 {
1131 struct packed_git *p;
1132 unsigned i;
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 }
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 }
1148 static int git_open_noatime(const char *name)
1149 {
1150 static int sha1_file_open_flag = O_NOATIME;
1152 for (;;) {
1153 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1154 if (fd >= 0)
1155 return fd;
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 }
1163 return -1;
1164 }
1165 }
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;
1173 fd = git_open_noatime(name);
1174 if (fd >= 0)
1175 return fd;
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 }
1189 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1190 {
1191 void *map;
1192 int fd;
1194 fd = open_sha1_file(sha1);
1195 map = NULL;
1196 if (fd >= 0) {
1197 struct stat st;
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 }
1208 static int legacy_loose_object(unsigned char *map)
1209 {
1210 unsigned int word;
1212 /*
1213 * Is it a zlib-compressed buffer? If so, the first byte
1214 * must be 0x78 (15-bit window size, deflated), and the
1215 * first 16-bit word is evenly divisible by 31
1216 */
1217 word = (map[0] << 8) + map[1];
1218 if (map[0] == 0x78 && !(word % 31))
1219 return 1;
1220 else
1221 return 0;
1222 }
1224 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1225 unsigned long len, enum object_type *type, unsigned long *sizep)
1226 {
1227 unsigned shift;
1228 unsigned long size, c;
1229 unsigned long used = 0;
1231 c = buf[used++];
1232 *type = (c >> 4) & 7;
1233 size = c & 15;
1234 shift = 4;
1235 while (c & 0x80) {
1236 if (len <= used || bitsizeof(long) <= shift) {
1237 error("bad object header");
1238 return 0;
1239 }
1240 c = buf[used++];
1241 size += (c & 0x7f) << shift;
1242 shift += 7;
1243 }
1244 *sizep = size;
1245 return used;
1246 }
1248 int unpack_sha1_header(z_stream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1249 {
1250 unsigned long size, used;
1251 static const char valid_loose_object_type[8] = {
1252 0, /* OBJ_EXT */
1253 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1254 0, /* "delta" and others are invalid in a loose object */
1255 };
1256 enum object_type type;
1258 /* Get the data stream */
1259 memset(stream, 0, sizeof(*stream));
1260 stream->next_in = map;
1261 stream->avail_in = mapsize;
1262 stream->next_out = buffer;
1263 stream->avail_out = bufsiz;
1265 if (legacy_loose_object(map)) {
1266 git_inflate_init(stream);
1267 return git_inflate(stream, 0);
1268 }
1271 /*
1272 * There used to be a second loose object header format which
1273 * was meant to mimic the in-pack format, allowing for direct
1274 * copy of the object data. This format turned up not to be
1275 * really worth it and we don't write it any longer. But we
1276 * can still read it.
1277 */
1278 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1279 if (!used || !valid_loose_object_type[type])
1280 return -1;
1281 map += used;
1282 mapsize -= used;
1284 /* Set up the stream for the rest.. */
1285 stream->next_in = map;
1286 stream->avail_in = mapsize;
1287 git_inflate_init(stream);
1289 /* And generate the fake traditional header */
1290 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1291 typename(type), size);
1292 return 0;
1293 }
1295 static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1296 {
1297 int bytes = strlen(buffer) + 1;
1298 unsigned char *buf = xmallocz(size);
1299 unsigned long n;
1300 int status = Z_OK;
1302 n = stream->total_out - bytes;
1303 if (n > size)
1304 n = size;
1305 memcpy(buf, (char *) buffer + bytes, n);
1306 bytes = n;
1307 if (bytes <= size) {
1308 /*
1309 * The above condition must be (bytes <= size), not
1310 * (bytes < size). In other words, even though we
1311 * expect no more output and set avail_out to zero,
1312 * the input zlib stream may have bytes that express
1313 * "this concludes the stream", and we *do* want to
1314 * eat that input.
1315 *
1316 * Otherwise we would not be able to test that we
1317 * consumed all the input to reach the expected size;
1318 * we also want to check that zlib tells us that all
1319 * went well with status == Z_STREAM_END at the end.
1320 */
1321 stream->next_out = buf + bytes;
1322 stream->avail_out = size - bytes;
1323 while (status == Z_OK)
1324 status = git_inflate(stream, Z_FINISH);
1325 }
1326 if (status == Z_STREAM_END && !stream->avail_in) {
1327 git_inflate_end(stream);
1328 return buf;
1329 }
1331 if (status < 0)
1332 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1333 else if (stream->avail_in)
1334 error("garbage at end of loose object '%s'",
1335 sha1_to_hex(sha1));
1336 free(buf);
1337 return NULL;
1338 }
1340 /*
1341 * We used to just use "sscanf()", but that's actually way
1342 * too permissive for what we want to check. So do an anal
1343 * object header parse by hand.
1344 */
1345 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1346 {
1347 char type[10];
1348 int i;
1349 unsigned long size;
1351 /*
1352 * The type can be at most ten bytes (including the
1353 * terminating '\0' that we add), and is followed by
1354 * a space.
1355 */
1356 i = 0;
1357 for (;;) {
1358 char c = *hdr++;
1359 if (c == ' ')
1360 break;
1361 type[i++] = c;
1362 if (i >= sizeof(type))
1363 return -1;
1364 }
1365 type[i] = 0;
1367 /*
1368 * The length must follow immediately, and be in canonical
1369 * decimal format (ie "010" is not valid).
1370 */
1371 size = *hdr++ - '0';
1372 if (size > 9)
1373 return -1;
1374 if (size) {
1375 for (;;) {
1376 unsigned long c = *hdr - '0';
1377 if (c > 9)
1378 break;
1379 hdr++;
1380 size = size * 10 + c;
1381 }
1382 }
1383 *sizep = size;
1385 /*
1386 * The length must be followed by a zero byte
1387 */
1388 return *hdr ? -1 : type_from_string(type);
1389 }
1391 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1392 {
1393 int ret;
1394 z_stream stream;
1395 char hdr[8192];
1397 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1398 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1399 return NULL;
1401 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1402 }
1404 unsigned long get_size_from_delta(struct packed_git *p,
1405 struct pack_window **w_curs,
1406 off_t curpos)
1407 {
1408 const unsigned char *data;
1409 unsigned char delta_head[20], *in;
1410 z_stream stream;
1411 int st;
1413 memset(&stream, 0, sizeof(stream));
1414 stream.next_out = delta_head;
1415 stream.avail_out = sizeof(delta_head);
1417 git_inflate_init(&stream);
1418 do {
1419 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1420 stream.next_in = in;
1421 st = git_inflate(&stream, Z_FINISH);
1422 curpos += stream.next_in - in;
1423 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1424 stream.total_out < sizeof(delta_head));
1425 git_inflate_end(&stream);
1426 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1427 error("delta data unpack-initial failed");
1428 return 0;
1429 }
1431 /* Examine the initial part of the delta to figure out
1432 * the result size.
1433 */
1434 data = delta_head;
1436 /* ignore base size */
1437 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1439 /* Read the result size */
1440 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1441 }
1443 static off_t get_delta_base(struct packed_git *p,
1444 struct pack_window **w_curs,
1445 off_t *curpos,
1446 enum object_type type,
1447 off_t delta_obj_offset)
1448 {
1449 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1450 off_t base_offset;
1452 /* use_pack() assured us we have [base_info, base_info + 20)
1453 * as a range that we can look at without walking off the
1454 * end of the mapped window. Its actually the hash size
1455 * that is assured. An OFS_DELTA longer than the hash size
1456 * is stupid, as then a REF_DELTA would be smaller to store.
1457 */
1458 if (type == OBJ_OFS_DELTA) {
1459 unsigned used = 0;
1460 unsigned char c = base_info[used++];
1461 base_offset = c & 127;
1462 while (c & 128) {
1463 base_offset += 1;
1464 if (!base_offset || MSB(base_offset, 7))
1465 return 0; /* overflow */
1466 c = base_info[used++];
1467 base_offset = (base_offset << 7) + (c & 127);
1468 }
1469 base_offset = delta_obj_offset - base_offset;
1470 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1471 return 0; /* out of bound */
1472 *curpos += used;
1473 } else if (type == OBJ_REF_DELTA) {
1474 /* The base entry _must_ be in the same pack */
1475 base_offset = find_pack_entry_one(base_info, p);
1476 *curpos += 20;
1477 } else
1478 die("I am totally screwed");
1479 return base_offset;
1480 }
1482 /* forward declaration for a mutually recursive function */
1483 static int packed_object_info(struct packed_git *p, off_t offset,
1484 unsigned long *sizep, int *rtype);
1486 static int packed_delta_info(struct packed_git *p,
1487 struct pack_window **w_curs,
1488 off_t curpos,
1489 enum object_type type,
1490 off_t obj_offset,
1491 unsigned long *sizep)
1492 {
1493 off_t base_offset;
1495 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1496 if (!base_offset)
1497 return OBJ_BAD;
1498 type = packed_object_info(p, base_offset, NULL, NULL);
1499 if (type <= OBJ_NONE) {
1500 struct revindex_entry *revidx;
1501 const unsigned char *base_sha1;
1502 revidx = find_pack_revindex(p, base_offset);
1503 if (!revidx)
1504 return OBJ_BAD;
1505 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1506 mark_bad_packed_object(p, base_sha1);
1507 type = sha1_object_info(base_sha1, NULL);
1508 if (type <= OBJ_NONE)
1509 return OBJ_BAD;
1510 }
1512 /* We choose to only get the type of the base object and
1513 * ignore potentially corrupt pack file that expects the delta
1514 * based on a base with a wrong size. This saves tons of
1515 * inflate() calls.
1516 */
1517 if (sizep) {
1518 *sizep = get_size_from_delta(p, w_curs, curpos);
1519 if (*sizep == 0)
1520 type = OBJ_BAD;
1521 }
1523 return type;
1524 }
1526 int unpack_object_header(struct packed_git *p,
1527 struct pack_window **w_curs,
1528 off_t *curpos,
1529 unsigned long *sizep)
1530 {
1531 unsigned char *base;
1532 unsigned int left;
1533 unsigned long used;
1534 enum object_type type;
1536 /* use_pack() assures us we have [base, base + 20) available
1537 * as a range that we can look at. (Its actually the hash
1538 * size that is assured.) With our object header encoding
1539 * the maximum deflated object size is 2^137, which is just
1540 * insane, so we know won't exceed what we have been given.
1541 */
1542 base = use_pack(p, w_curs, *curpos, &left);
1543 used = unpack_object_header_buffer(base, left, &type, sizep);
1544 if (!used) {
1545 type = OBJ_BAD;
1546 } else
1547 *curpos += used;
1549 return type;
1550 }
1552 int packed_object_info_detail(struct packed_git *p,
1553 off_t obj_offset,
1554 unsigned long *size,
1555 unsigned long *store_size,
1556 unsigned int *delta_chain_length,
1557 unsigned char *base_sha1)
1558 {
1559 struct pack_window *w_curs = NULL;
1560 off_t curpos;
1561 unsigned long dummy;
1562 unsigned char *next_sha1;
1563 enum object_type type;
1564 struct revindex_entry *revidx;
1566 *delta_chain_length = 0;
1567 curpos = obj_offset;
1568 type = unpack_object_header(p, &w_curs, &curpos, size);
1570 revidx = find_pack_revindex(p, obj_offset);
1571 *store_size = revidx[1].offset - obj_offset;
1573 for (;;) {
1574 switch (type) {
1575 default:
1576 die("pack %s contains unknown object type %d",
1577 p->pack_name, type);
1578 case OBJ_COMMIT:
1579 case OBJ_TREE:
1580 case OBJ_BLOB:
1581 case OBJ_TAG:
1582 unuse_pack(&w_curs);
1583 return type;
1584 case OBJ_OFS_DELTA:
1585 obj_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
1586 if (!obj_offset)
1587 die("pack %s contains bad delta base reference of type %s",
1588 p->pack_name, typename(type));
1589 if (*delta_chain_length == 0) {
1590 revidx = find_pack_revindex(p, obj_offset);
1591 hashcpy(base_sha1, nth_packed_object_sha1(p, revidx->nr));
1592 }
1593 break;
1594 case OBJ_REF_DELTA:
1595 next_sha1 = use_pack(p, &w_curs, curpos, NULL);
1596 if (*delta_chain_length == 0)
1597 hashcpy(base_sha1, next_sha1);
1598 obj_offset = find_pack_entry_one(next_sha1, p);
1599 break;
1600 }
1601 (*delta_chain_length)++;
1602 curpos = obj_offset;
1603 type = unpack_object_header(p, &w_curs, &curpos, &dummy);
1604 }
1605 }
1607 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1608 unsigned long *sizep, int *rtype)
1609 {
1610 struct pack_window *w_curs = NULL;
1611 unsigned long size;
1612 off_t curpos = obj_offset;
1613 enum object_type type;
1615 type = unpack_object_header(p, &w_curs, &curpos, &size);
1616 if (rtype)
1617 *rtype = type; /* representation type */
1619 switch (type) {
1620 case OBJ_OFS_DELTA:
1621 case OBJ_REF_DELTA:
1622 type = packed_delta_info(p, &w_curs, curpos,
1623 type, obj_offset, sizep);
1624 break;
1625 case OBJ_COMMIT:
1626 case OBJ_TREE:
1627 case OBJ_BLOB:
1628 case OBJ_TAG:
1629 if (sizep)
1630 *sizep = size;
1631 break;
1632 default:
1633 error("unknown object type %i at offset %"PRIuMAX" in %s",
1634 type, (uintmax_t)obj_offset, p->pack_name);
1635 type = OBJ_BAD;
1636 }
1637 unuse_pack(&w_curs);
1638 return type;
1639 }
1641 static void *unpack_compressed_entry(struct packed_git *p,
1642 struct pack_window **w_curs,
1643 off_t curpos,
1644 unsigned long size)
1645 {
1646 int st;
1647 z_stream stream;
1648 unsigned char *buffer, *in;
1650 buffer = xmallocz(size);
1651 memset(&stream, 0, sizeof(stream));
1652 stream.next_out = buffer;
1653 stream.avail_out = size + 1;
1655 git_inflate_init(&stream);
1656 do {
1657 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1658 stream.next_in = in;
1659 st = git_inflate(&stream, Z_FINISH);
1660 if (!stream.avail_out)
1661 break; /* the payload is larger than it should be */
1662 curpos += stream.next_in - in;
1663 } while (st == Z_OK || st == Z_BUF_ERROR);
1664 git_inflate_end(&stream);
1665 if ((st != Z_STREAM_END) || stream.total_out != size) {
1666 free(buffer);
1667 return NULL;
1668 }
1670 return buffer;
1671 }
1673 #define MAX_DELTA_CACHE (256)
1675 static size_t delta_base_cached;
1677 static struct delta_base_cache_lru_list {
1678 struct delta_base_cache_lru_list *prev;
1679 struct delta_base_cache_lru_list *next;
1680 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1682 static struct delta_base_cache_entry {
1683 struct delta_base_cache_lru_list lru;
1684 void *data;
1685 struct packed_git *p;
1686 off_t base_offset;
1687 unsigned long size;
1688 enum object_type type;
1689 } delta_base_cache[MAX_DELTA_CACHE];
1691 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1692 {
1693 unsigned long hash;
1695 hash = (unsigned long)p + (unsigned long)base_offset;
1696 hash += (hash >> 8) + (hash >> 16);
1697 return hash % MAX_DELTA_CACHE;
1698 }
1700 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1701 {
1702 unsigned long hash = pack_entry_hash(p, base_offset);
1703 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1704 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1705 }
1707 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1708 unsigned long *base_size, enum object_type *type, int keep_cache)
1709 {
1710 void *ret;
1711 unsigned long hash = pack_entry_hash(p, base_offset);
1712 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1714 ret = ent->data;
1715 if (!ret || ent->p != p || ent->base_offset != base_offset)
1716 return unpack_entry(p, base_offset, type, base_size);
1718 if (!keep_cache) {
1719 ent->data = NULL;
1720 ent->lru.next->prev = ent->lru.prev;
1721 ent->lru.prev->next = ent->lru.next;
1722 delta_base_cached -= ent->size;
1723 } else {
1724 ret = xmemdupz(ent->data, ent->size);
1725 }
1726 *type = ent->type;
1727 *base_size = ent->size;
1728 return ret;
1729 }
1731 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1732 {
1733 if (ent->data) {
1734 free(ent->data);
1735 ent->data = NULL;
1736 ent->lru.next->prev = ent->lru.prev;
1737 ent->lru.prev->next = ent->lru.next;
1738 delta_base_cached -= ent->size;
1739 }
1740 }
1742 void clear_delta_base_cache(void)
1743 {
1744 unsigned long p;
1745 for (p = 0; p < MAX_DELTA_CACHE; p++)
1746 release_delta_base_cache(&delta_base_cache[p]);
1747 }
1749 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1750 void *base, unsigned long base_size, enum object_type type)
1751 {
1752 unsigned long hash = pack_entry_hash(p, base_offset);
1753 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1754 struct delta_base_cache_lru_list *lru;
1756 release_delta_base_cache(ent);
1757 delta_base_cached += base_size;
1759 for (lru = delta_base_cache_lru.next;
1760 delta_base_cached > delta_base_cache_limit
1761 && lru != &delta_base_cache_lru;
1762 lru = lru->next) {
1763 struct delta_base_cache_entry *f = (void *)lru;
1764 if (f->type == OBJ_BLOB)
1765 release_delta_base_cache(f);
1766 }
1767 for (lru = delta_base_cache_lru.next;
1768 delta_base_cached > delta_base_cache_limit
1769 && lru != &delta_base_cache_lru;
1770 lru = lru->next) {
1771 struct delta_base_cache_entry *f = (void *)lru;
1772 release_delta_base_cache(f);
1773 }
1775 ent->p = p;
1776 ent->base_offset = base_offset;
1777 ent->type = type;
1778 ent->data = base;
1779 ent->size = base_size;
1780 ent->lru.next = &delta_base_cache_lru;
1781 ent->lru.prev = delta_base_cache_lru.prev;
1782 delta_base_cache_lru.prev->next = &ent->lru;
1783 delta_base_cache_lru.prev = &ent->lru;
1784 }
1786 static void *read_object(const unsigned char *sha1, enum object_type *type,
1787 unsigned long *size);
1789 static void *unpack_delta_entry(struct packed_git *p,
1790 struct pack_window **w_curs,
1791 off_t curpos,
1792 unsigned long delta_size,
1793 off_t obj_offset,
1794 enum object_type *type,
1795 unsigned long *sizep)
1796 {
1797 void *delta_data, *result, *base;
1798 unsigned long base_size;
1799 off_t base_offset;
1801 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1802 if (!base_offset) {
1803 error("failed to validate delta base reference "
1804 "at offset %"PRIuMAX" from %s",
1805 (uintmax_t)curpos, p->pack_name);
1806 return NULL;
1807 }
1808 unuse_pack(w_curs);
1809 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1810 if (!base) {
1811 /*
1812 * We're probably in deep shit, but let's try to fetch
1813 * the required base anyway from another pack or loose.
1814 * This is costly but should happen only in the presence
1815 * of a corrupted pack, and is better than failing outright.
1816 */
1817 struct revindex_entry *revidx;
1818 const unsigned char *base_sha1;
1819 revidx = find_pack_revindex(p, base_offset);
1820 if (!revidx)
1821 return NULL;
1822 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1823 error("failed to read delta base object %s"
1824 " at offset %"PRIuMAX" from %s",
1825 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1826 p->pack_name);
1827 mark_bad_packed_object(p, base_sha1);
1828 base = read_object(base_sha1, type, &base_size);
1829 if (!base)
1830 return NULL;
1831 }
1833 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1834 if (!delta_data) {
1835 error("failed to unpack compressed delta "
1836 "at offset %"PRIuMAX" from %s",
1837 (uintmax_t)curpos, p->pack_name);
1838 free(base);
1839 return NULL;
1840 }
1841 result = patch_delta(base, base_size,
1842 delta_data, delta_size,
1843 sizep);
1844 if (!result)
1845 die("failed to apply delta");
1846 free(delta_data);
1847 add_delta_base_cache(p, base_offset, base, base_size, *type);
1848 return result;
1849 }
1851 int do_check_packed_object_crc;
1853 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1854 enum object_type *type, unsigned long *sizep)
1855 {
1856 struct pack_window *w_curs = NULL;
1857 off_t curpos = obj_offset;
1858 void *data;
1860 if (do_check_packed_object_crc && p->index_version > 1) {
1861 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1862 unsigned long len = revidx[1].offset - obj_offset;
1863 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1864 const unsigned char *sha1 =
1865 nth_packed_object_sha1(p, revidx->nr);
1866 error("bad packed object CRC for %s",
1867 sha1_to_hex(sha1));
1868 mark_bad_packed_object(p, sha1);
1869 unuse_pack(&w_curs);
1870 return NULL;
1871 }
1872 }
1874 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1875 switch (*type) {
1876 case OBJ_OFS_DELTA:
1877 case OBJ_REF_DELTA:
1878 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1879 obj_offset, type, sizep);
1880 break;
1881 case OBJ_COMMIT:
1882 case OBJ_TREE:
1883 case OBJ_BLOB:
1884 case OBJ_TAG:
1885 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1886 break;
1887 default:
1888 data = NULL;
1889 error("unknown object type %i at offset %"PRIuMAX" in %s",
1890 *type, (uintmax_t)obj_offset, p->pack_name);
1891 }
1892 unuse_pack(&w_curs);
1893 return data;
1894 }
1896 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1897 uint32_t n)
1898 {
1899 const unsigned char *index = p->index_data;
1900 if (!index) {
1901 if (open_pack_index(p))
1902 return NULL;
1903 index = p->index_data;
1904 }
1905 if (n >= p->num_objects)
1906 return NULL;
1907 index += 4 * 256;
1908 if (p->index_version == 1) {
1909 return index + 24 * n + 4;
1910 } else {
1911 index += 8;
1912 return index + 20 * n;
1913 }
1914 }
1916 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1917 {
1918 const unsigned char *index = p->index_data;
1919 index += 4 * 256;
1920 if (p->index_version == 1) {
1921 return ntohl(*((uint32_t *)(index + 24 * n)));
1922 } else {
1923 uint32_t off;
1924 index += 8 + p->num_objects * (20 + 4);
1925 off = ntohl(*((uint32_t *)(index + 4 * n)));
1926 if (!(off & 0x80000000))
1927 return off;
1928 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1929 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1930 ntohl(*((uint32_t *)(index + 4)));
1931 }
1932 }
1934 off_t find_pack_entry_one(const unsigned char *sha1,
1935 struct packed_git *p)
1936 {
1937 const uint32_t *level1_ofs = p->index_data;
1938 const unsigned char *index = p->index_data;
1939 unsigned hi, lo, stride;
1940 static int use_lookup = -1;
1941 static int debug_lookup = -1;
1943 if (debug_lookup < 0)
1944 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1946 if (!index) {
1947 if (open_pack_index(p))
1948 return 0;
1949 level1_ofs = p->index_data;
1950 index = p->index_data;
1951 }
1952 if (p->index_version > 1) {
1953 level1_ofs += 2;
1954 index += 8;
1955 }
1956 index += 4 * 256;
1957 hi = ntohl(level1_ofs[*sha1]);
1958 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
1959 if (p->index_version > 1) {
1960 stride = 20;
1961 } else {
1962 stride = 24;
1963 index += 4;
1964 }
1966 if (debug_lookup)
1967 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
1968 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
1970 if (use_lookup < 0)
1971 use_lookup = !!getenv("GIT_USE_LOOKUP");
1972 if (use_lookup) {
1973 int pos = sha1_entry_pos(index, stride, 0,
1974 lo, hi, p->num_objects, sha1);
1975 if (pos < 0)
1976 return 0;
1977 return nth_packed_object_offset(p, pos);
1978 }
1980 do {
1981 unsigned mi = (lo + hi) / 2;
1982 int cmp = hashcmp(index + mi * stride, sha1);
1984 if (debug_lookup)
1985 printf("lo %u hi %u rg %u mi %u\n",
1986 lo, hi, hi - lo, mi);
1987 if (!cmp)
1988 return nth_packed_object_offset(p, mi);
1989 if (cmp > 0)
1990 hi = mi;
1991 else
1992 lo = mi+1;
1993 } while (lo < hi);
1994 return 0;
1995 }
1997 static int is_pack_valid(struct packed_git *p)
1998 {
1999 /* An already open pack is known to be valid. */
2000 if (p->pack_fd != -1)
2001 return 1;
2003 /* If the pack has one window completely covering the
2004 * file size, the pack is known to be valid even if
2005 * the descriptor is not currently open.
2006 */
2007 if (p->windows) {
2008 struct pack_window *w = p->windows;
2010 if (!w->offset && w->len == p->pack_size)
2011 return 1;
2012 }
2014 /* Force the pack to open to prove its valid. */
2015 return !open_packed_git(p);
2016 }
2018 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2019 {
2020 static struct packed_git *last_found = (void *)1;
2021 struct packed_git *p;
2022 off_t offset;
2024 prepare_packed_git();
2025 if (!packed_git)
2026 return 0;
2027 p = (last_found == (void *)1) ? packed_git : last_found;
2029 do {
2030 if (p->num_bad_objects) {
2031 unsigned i;
2032 for (i = 0; i < p->num_bad_objects; i++)
2033 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2034 goto next;
2035 }
2037 offset = find_pack_entry_one(sha1, p);
2038 if (offset) {
2039 /*
2040 * We are about to tell the caller where they can
2041 * locate the requested object. We better make
2042 * sure the packfile is still here and can be
2043 * accessed before supplying that answer, as
2044 * it may have been deleted since the index
2045 * was loaded!
2046 */
2047 if (!is_pack_valid(p)) {
2048 error("packfile %s cannot be accessed", p->pack_name);
2049 goto next;
2050 }
2051 e->offset = offset;
2052 e->p = p;
2053 hashcpy(e->sha1, sha1);
2054 last_found = p;
2055 return 1;
2056 }
2058 next:
2059 if (p == last_found)
2060 p = packed_git;
2061 else
2062 p = p->next;
2063 if (p == last_found)
2064 p = p->next;
2065 } while (p);
2066 return 0;
2067 }
2069 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2070 struct packed_git *packs)
2071 {
2072 struct packed_git *p;
2074 for (p = packs; p; p = p->next) {
2075 if (find_pack_entry_one(sha1, p))
2076 return p;
2077 }
2078 return NULL;
2080 }
2082 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2083 {
2084 int status;
2085 unsigned long mapsize, size;
2086 void *map;
2087 z_stream stream;
2088 char hdr[32];
2090 map = map_sha1_file(sha1, &mapsize);
2091 if (!map)
2092 return error("unable to find %s", sha1_to_hex(sha1));
2093 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2094 status = error("unable to unpack %s header",
2095 sha1_to_hex(sha1));
2096 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2097 status = error("unable to parse %s header", sha1_to_hex(sha1));
2098 else if (sizep)
2099 *sizep = size;
2100 git_inflate_end(&stream);
2101 munmap(map, mapsize);
2102 return status;
2103 }
2105 /* returns enum object_type or negative */
2106 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2107 {
2108 struct cached_object *co;
2109 struct pack_entry e;
2110 int status, rtype;
2112 co = find_cached_object(sha1);
2113 if (co) {
2114 if (oi->sizep)
2115 *(oi->sizep) = co->size;
2116 oi->whence = OI_CACHED;
2117 return co->type;
2118 }
2120 if (!find_pack_entry(sha1, &e)) {
2121 /* Most likely it's a loose object. */
2122 status = sha1_loose_object_info(sha1, oi->sizep);
2123 if (status >= 0) {
2124 oi->whence = OI_LOOSE;
2125 return status;
2126 }
2128 /* Not a loose object; someone else may have just packed it. */
2129 reprepare_packed_git();
2130 if (!find_pack_entry(sha1, &e))
2131 return status;
2132 }
2134 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2135 if (status < 0) {
2136 mark_bad_packed_object(e.p, sha1);
2137 status = sha1_object_info_extended(sha1, oi);
2138 } else if (in_delta_base_cache(e.p, e.offset)) {
2139 oi->whence = OI_DBCACHED;
2140 } else {
2141 oi->whence = OI_PACKED;
2142 oi->u.packed.offset = e.offset;
2143 oi->u.packed.pack = e.p;
2144 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2145 rtype == OBJ_OFS_DELTA);
2146 }
2148 return status;
2149 }
2151 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2152 {
2153 struct object_info oi;
2155 oi.sizep = sizep;
2156 return sha1_object_info_extended(sha1, &oi);
2157 }
2159 static void *read_packed_sha1(const unsigned char *sha1,
2160 enum object_type *type, unsigned long *size)
2161 {
2162 struct pack_entry e;
2163 void *data;
2165 if (!find_pack_entry(sha1, &e))
2166 return NULL;
2167 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2168 if (!data) {
2169 /*
2170 * We're probably in deep shit, but let's try to fetch
2171 * the required object anyway from another pack or loose.
2172 * This should happen only in the presence of a corrupted
2173 * pack, and is better than failing outright.
2174 */
2175 error("failed to read object %s at offset %"PRIuMAX" from %s",
2176 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2177 mark_bad_packed_object(e.p, sha1);
2178 data = read_object(sha1, type, size);
2179 }
2180 return data;
2181 }
2183 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2184 unsigned char *sha1)
2185 {
2186 struct cached_object *co;
2188 hash_sha1_file(buf, len, typename(type), sha1);
2189 if (has_sha1_file(sha1) || find_cached_object(sha1))
2190 return 0;
2191 if (cached_object_alloc <= cached_object_nr) {
2192 cached_object_alloc = alloc_nr(cached_object_alloc);
2193 cached_objects = xrealloc(cached_objects,
2194 sizeof(*cached_objects) *
2195 cached_object_alloc);
2196 }
2197 co = &cached_objects[cached_object_nr++];
2198 co->size = len;
2199 co->type = type;
2200 co->buf = xmalloc(len);
2201 memcpy(co->buf, buf, len);
2202 hashcpy(co->sha1, sha1);
2203 return 0;
2204 }
2206 static void *read_object(const unsigned char *sha1, enum object_type *type,
2207 unsigned long *size)
2208 {
2209 unsigned long mapsize;
2210 void *map, *buf;
2211 struct cached_object *co;
2213 co = find_cached_object(sha1);
2214 if (co) {
2215 *type = co->type;
2216 *size = co->size;
2217 return xmemdupz(co->buf, co->size);
2218 }
2220 buf = read_packed_sha1(sha1, type, size);
2221 if (buf)
2222 return buf;
2223 map = map_sha1_file(sha1, &mapsize);
2224 if (map) {
2225 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2226 munmap(map, mapsize);
2227 return buf;
2228 }
2229 reprepare_packed_git();
2230 return read_packed_sha1(sha1, type, size);
2231 }
2233 /*
2234 * This function dies on corrupt objects; the callers who want to
2235 * deal with them should arrange to call read_object() and give error
2236 * messages themselves.
2237 */
2238 void *read_sha1_file_extended(const unsigned char *sha1,
2239 enum object_type *type,
2240 unsigned long *size,
2241 unsigned flag)
2242 {
2243 void *data;
2244 char *path;
2245 const struct packed_git *p;
2246 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2247 ? lookup_replace_object(sha1) : sha1;
2249 errno = 0;
2250 data = read_object(repl, type, size);
2251 if (data)
2252 return data;
2254 if (errno && errno != ENOENT)
2255 die_errno("failed to read object %s", sha1_to_hex(sha1));
2257 /* die if we replaced an object with one that does not exist */
2258 if (repl != sha1)
2259 die("replacement %s not found for %s",
2260 sha1_to_hex(repl), sha1_to_hex(sha1));
2262 if (has_loose_object(repl)) {
2263 path = sha1_file_name(sha1);
2264 die("loose object %s (stored in %s) is corrupt",
2265 sha1_to_hex(repl), path);
2266 }
2268 if ((p = has_packed_and_bad(repl)) != NULL)
2269 die("packed object %s (stored in %s) is corrupt",
2270 sha1_to_hex(repl), p->pack_name);
2272 return NULL;
2273 }
2275 void *read_object_with_reference(const unsigned char *sha1,
2276 const char *required_type_name,
2277 unsigned long *size,
2278 unsigned char *actual_sha1_return)
2279 {
2280 enum object_type type, required_type;
2281 void *buffer;
2282 unsigned long isize;
2283 unsigned char actual_sha1[20];
2285 required_type = type_from_string(required_type_name);
2286 hashcpy(actual_sha1, sha1);
2287 while (1) {
2288 int ref_length = -1;
2289 const char *ref_type = NULL;
2291 buffer = read_sha1_file(actual_sha1, &type, &isize);
2292 if (!buffer)
2293 return NULL;
2294 if (type == required_type) {
2295 *size = isize;
2296 if (actual_sha1_return)
2297 hashcpy(actual_sha1_return, actual_sha1);
2298 return buffer;
2299 }
2300 /* Handle references */
2301 else if (type == OBJ_COMMIT)
2302 ref_type = "tree ";
2303 else if (type == OBJ_TAG)
2304 ref_type = "object ";
2305 else {
2306 free(buffer);
2307 return NULL;
2308 }
2309 ref_length = strlen(ref_type);
2311 if (ref_length + 40 > isize ||
2312 memcmp(buffer, ref_type, ref_length) ||
2313 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2314 free(buffer);
2315 return NULL;
2316 }
2317 free(buffer);
2318 /* Now we have the ID of the referred-to object in
2319 * actual_sha1. Check again. */
2320 }
2321 }
2323 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2324 const char *type, unsigned char *sha1,
2325 char *hdr, int *hdrlen)
2326 {
2327 git_SHA_CTX c;
2329 /* Generate the header */
2330 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2332 /* Sha1.. */
2333 git_SHA1_Init(&c);
2334 git_SHA1_Update(&c, hdr, *hdrlen);
2335 git_SHA1_Update(&c, buf, len);
2336 git_SHA1_Final(sha1, &c);
2337 }
2339 /*
2340 * Move the just written object into its final resting place.
2341 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2342 * "moving" is only a part of what it does, when no patch between
2343 * master to pu changes the call sites of this function.
2344 */
2345 int move_temp_to_file(const char *tmpfile, const char *filename)
2346 {
2347 int ret = 0;
2349 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2350 goto try_rename;
2351 else if (link(tmpfile, filename))
2352 ret = errno;
2354 /*
2355 * Coda hack - coda doesn't like cross-directory links,
2356 * so we fall back to a rename, which will mean that it
2357 * won't be able to check collisions, but that's not a
2358 * big deal.
2359 *
2360 * The same holds for FAT formatted media.
2361 *
2362 * When this succeeds, we just return. We have nothing
2363 * left to unlink.
2364 */
2365 if (ret && ret != EEXIST) {
2366 try_rename:
2367 if (!rename(tmpfile, filename))
2368 goto out;
2369 ret = errno;
2370 }
2371 unlink_or_warn(tmpfile);
2372 if (ret) {
2373 if (ret != EEXIST) {
2374 return error("unable to write sha1 filename %s: %s\n", filename, strerror(ret));
2375 }
2376 /* FIXME!!! Collision check here ? */
2377 }
2379 out:
2380 if (adjust_shared_perm(filename))
2381 return error("unable to set permission to '%s'", filename);
2382 return 0;
2383 }
2385 static int write_buffer(int fd, const void *buf, size_t len)
2386 {
2387 if (write_in_full(fd, buf, len) < 0)
2388 return error("file write error (%s)", strerror(errno));
2389 return 0;
2390 }
2392 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2393 unsigned char *sha1)
2394 {
2395 char hdr[32];
2396 int hdrlen;
2397 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2398 return 0;
2399 }
2401 /* Finalize a file on disk, and close it. */
2402 static void close_sha1_file(int fd)
2403 {
2404 if (fsync_object_files)
2405 fsync_or_die(fd, "sha1 file");
2406 if (close(fd) != 0)
2407 die_errno("error when closing sha1 file");
2408 }
2410 /* Size of directory component, including the ending '/' */
2411 static inline int directory_size(const char *filename)
2412 {
2413 const char *s = strrchr(filename, '/');
2414 if (!s)
2415 return 0;
2416 return s - filename + 1;
2417 }
2419 /*
2420 * This creates a temporary file in the same directory as the final
2421 * 'filename'
2422 *
2423 * We want to avoid cross-directory filename renames, because those
2424 * can have problems on various filesystems (FAT, NFS, Coda).
2425 */
2426 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2427 {
2428 int fd, dirlen = directory_size(filename);
2430 if (dirlen + 20 > bufsiz) {
2431 errno = ENAMETOOLONG;
2432 return -1;
2433 }
2434 memcpy(buffer, filename, dirlen);
2435 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2436 fd = git_mkstemp_mode(buffer, 0444);
2437 if (fd < 0 && dirlen && errno == ENOENT) {
2438 /* Make sure the directory exists */
2439 memcpy(buffer, filename, dirlen);
2440 buffer[dirlen-1] = 0;
2441 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2442 return -1;
2444 /* Try again */
2445 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2446 fd = git_mkstemp_mode(buffer, 0444);
2447 }
2448 return fd;
2449 }
2451 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2452 const void *buf, unsigned long len, time_t mtime)
2453 {
2454 int fd, ret;
2455 unsigned char compressed[4096];
2456 z_stream stream;
2457 git_SHA_CTX c;
2458 unsigned char parano_sha1[20];
2459 char *filename;
2460 static char tmpfile[PATH_MAX];
2462 filename = sha1_file_name(sha1);
2463 fd = create_tmpfile(tmpfile, sizeof(tmpfile), filename);
2464 if (fd < 0) {
2465 if (errno == EACCES)
2466 return error("insufficient permission for adding an object to repository database %s\n", get_object_directory());
2467 else
2468 return error("unable to create temporary sha1 filename %s: %s\n", tmpfile, strerror(errno));
2469 }
2471 /* Set it up */
2472 memset(&stream, 0, sizeof(stream));
2473 deflateInit(&stream, zlib_compression_level);
2474 stream.next_out = compressed;
2475 stream.avail_out = sizeof(compressed);
2476 git_SHA1_Init(&c);
2478 /* First header.. */
2479 stream.next_in = (unsigned char *)hdr;
2480 stream.avail_in = hdrlen;
2481 while (deflate(&stream, 0) == Z_OK)
2482 /* nothing */;
2483 git_SHA1_Update(&c, hdr, hdrlen);
2485 /* Then the data itself.. */
2486 stream.next_in = (void *)buf;
2487 stream.avail_in = len;
2488 do {
2489 unsigned char *in0 = stream.next_in;
2490 ret = deflate(&stream, Z_FINISH);
2491 git_SHA1_Update(&c, in0, stream.next_in - in0);
2492 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2493 die("unable to write sha1 file");
2494 stream.next_out = compressed;
2495 stream.avail_out = sizeof(compressed);
2496 } while (ret == Z_OK);
2498 if (ret != Z_STREAM_END)
2499 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2500 ret = deflateEnd(&stream);
2501 if (ret != Z_OK)
2502 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2503 git_SHA1_Final(parano_sha1, &c);
2504 if (hashcmp(sha1, parano_sha1) != 0)
2505 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2507 close_sha1_file(fd);
2509 if (mtime) {
2510 struct utimbuf utb;
2511 utb.actime = mtime;
2512 utb.modtime = mtime;
2513 if (utime(tmpfile, &utb) < 0)
2514 warning("failed utime() on %s: %s",
2515 tmpfile, strerror(errno));
2516 }
2518 return move_temp_to_file(tmpfile, filename);
2519 }
2521 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2522 {
2523 unsigned char sha1[20];
2524 char hdr[32];
2525 int hdrlen;
2527 /* Normally if we have it in the pack then we do not bother writing
2528 * it out into .git/objects/??/?{38} file.
2529 */
2530 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2531 if (returnsha1)
2532 hashcpy(returnsha1, sha1);
2533 if (has_sha1_file(sha1))
2534 return 0;
2535 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2536 }
2538 int force_object_loose(const unsigned char *sha1, time_t mtime)
2539 {
2540 void *buf;
2541 unsigned long len;
2542 enum object_type type;
2543 char hdr[32];
2544 int hdrlen;
2545 int ret;
2547 if (has_loose_object(sha1))
2548 return 0;
2549 buf = read_packed_sha1(sha1, &type, &len);
2550 if (!buf)
2551 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2552 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2553 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2554 free(buf);
2556 return ret;
2557 }
2559 int has_pack_index(const unsigned char *sha1)
2560 {
2561 struct stat st;
2562 if (stat(sha1_pack_index_name(sha1), &st))
2563 return 0;
2564 return 1;
2565 }
2567 int has_sha1_pack(const unsigned char *sha1)
2568 {
2569 struct pack_entry e;
2570 return find_pack_entry(sha1, &e);
2571 }
2573 int has_sha1_file(const unsigned char *sha1)
2574 {
2575 struct pack_entry e;
2577 if (find_pack_entry(sha1, &e))
2578 return 1;
2579 return has_loose_object(sha1);
2580 }
2582 static void check_tree(const void *buf, size_t size)
2583 {
2584 struct tree_desc desc;
2585 struct name_entry entry;
2587 init_tree_desc(&desc, buf, size);
2588 while (tree_entry(&desc, &entry))
2589 /* do nothing
2590 * tree_entry() will die() on malformed entries */
2591 ;
2592 }
2594 static void check_commit(const void *buf, size_t size)
2595 {
2596 struct commit c;
2597 memset(&c, 0, sizeof(c));
2598 if (parse_commit_buffer(&c, buf, size))
2599 die("corrupt commit");
2600 }
2602 static void check_tag(const void *buf, size_t size)
2603 {
2604 struct tag t;
2605 memset(&t, 0, sizeof(t));
2606 if (parse_tag_buffer(&t, buf, size))
2607 die("corrupt tag");
2608 }
2610 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2611 enum object_type type,
2612 const char *path, unsigned flags)
2613 {
2614 int ret, re_allocated = 0;
2615 int write_object = flags & HASH_WRITE_OBJECT;
2617 if (!type)
2618 type = OBJ_BLOB;
2620 /*
2621 * Convert blobs to git internal format
2622 */
2623 if ((type == OBJ_BLOB) && path) {
2624 struct strbuf nbuf = STRBUF_INIT;
2625 if (convert_to_git(path, buf, size, &nbuf,
2626 write_object ? safe_crlf : 0)) {
2627 buf = strbuf_detach(&nbuf, &size);
2628 re_allocated = 1;
2629 }
2630 }
2631 if (flags & HASH_FORMAT_CHECK) {
2632 if (type == OBJ_TREE)
2633 check_tree(buf, size);
2634 if (type == OBJ_COMMIT)
2635 check_commit(buf, size);
2636 if (type == OBJ_TAG)
2637 check_tag(buf, size);
2638 }
2640 if (write_object)
2641 ret = write_sha1_file(buf, size, typename(type), sha1);
2642 else
2643 ret = hash_sha1_file(buf, size, typename(type), sha1);
2644 if (re_allocated)
2645 free(buf);
2646 return ret;
2647 }
2649 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2650 const char *path, unsigned flags)
2651 {
2652 struct strbuf sbuf = STRBUF_INIT;
2653 int ret;
2655 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2656 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2657 else
2658 ret = -1;
2659 strbuf_release(&sbuf);
2660 return ret;
2661 }
2663 #define SMALL_FILE_SIZE (32*1024)
2665 static int index_core(unsigned char *sha1, int fd, size_t size,
2666 enum object_type type, const char *path,
2667 unsigned flags)
2668 {
2669 int ret;
2671 if (!size) {
2672 ret = index_mem(sha1, NULL, size, type, path, flags);
2673 } else if (size <= SMALL_FILE_SIZE) {
2674 char *buf = xmalloc(size);
2675 if (size == read_in_full(fd, buf, size))
2676 ret = index_mem(sha1, buf, size, type, path, flags);
2677 else
2678 ret = error("short read %s", strerror(errno));
2679 free(buf);
2680 } else {
2681 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2682 ret = index_mem(sha1, buf, size, type, path, flags);
2683 munmap(buf, size);
2684 }
2685 return ret;
2686 }
2688 /*
2689 * This creates one packfile per large blob, because the caller
2690 * immediately wants the result sha1, and fast-import can report the
2691 * object name via marks mechanism only by closing the created
2692 * packfile.
2693 *
2694 * This also bypasses the usual "convert-to-git" dance, and that is on
2695 * purpose. We could write a streaming version of the converting
2696 * functions and insert that before feeding the data to fast-import
2697 * (or equivalent in-core API described above), but the primary
2698 * motivation for trying to stream from the working tree file and to
2699 * avoid mmaping it in core is to deal with large binary blobs, and
2700 * by definition they do _not_ want to get any conversion.
2701 */
2702 static int index_stream(unsigned char *sha1, int fd, size_t size,
2703 enum object_type type, const char *path,
2704 unsigned flags)
2705 {
2706 struct child_process fast_import;
2707 char export_marks[512];
2708 const char *argv[] = { "fast-import", "--quiet", export_marks, NULL };
2709 char tmpfile[512];
2710 char fast_import_cmd[512];
2711 char buf[512];
2712 int len, tmpfd;
2714 strcpy(tmpfile, git_path("hashstream_XXXXXX"));
2715 tmpfd = git_mkstemp_mode(tmpfile, 0600);
2716 if (tmpfd < 0)
2717 die_errno("cannot create tempfile: %s", tmpfile);
2718 if (close(tmpfd))
2719 die_errno("cannot close tempfile: %s", tmpfile);
2720 sprintf(export_marks, "--export-marks=%s", tmpfile);
2722 memset(&fast_import, 0, sizeof(fast_import));
2723 fast_import.in = -1;
2724 fast_import.argv = argv;
2725 fast_import.git_cmd = 1;
2726 if (start_command(&fast_import))
2727 die_errno("index-stream: git fast-import failed");
2729 len = sprintf(fast_import_cmd, "blob\nmark :1\ndata %lu\n",
2730 (unsigned long) size);
2731 write_or_whine(fast_import.in, fast_import_cmd, len,
2732 "index-stream: feeding fast-import");
2733 while (size) {
2734 char buf[10240];
2735 size_t sz = size < sizeof(buf) ? size : sizeof(buf);
2736 ssize_t actual;
2738 actual = read_in_full(fd, buf, sz);
2739 if (actual < 0)
2740 die_errno("index-stream: reading input");
2741 if (write_in_full(fast_import.in, buf, actual) != actual)
2742 die_errno("index-stream: feeding fast-import");
2743 size -= actual;
2744 }
2745 if (close(fast_import.in))
2746 die_errno("index-stream: closing fast-import");
2747 if (finish_command(&fast_import))
2748 die_errno("index-stream: finishing fast-import");
2750 tmpfd = open(tmpfile, O_RDONLY);
2751 if (tmpfd < 0)
2752 die_errno("index-stream: cannot open fast-import mark");
2753 len = read(tmpfd, buf, sizeof(buf));
2754 if (len < 0)
2755 die_errno("index-stream: reading fast-import mark");
2756 if (close(tmpfd) < 0)
2757 die_errno("index-stream: closing fast-import mark");
2758 if (unlink(tmpfile))
2759 die_errno("index-stream: unlinking fast-import mark");
2760 if (len != 44 ||
2761 memcmp(":1 ", buf, 3) ||
2762 get_sha1_hex(buf + 3, sha1))
2763 die_errno("index-stream: unexpected fast-import mark: <%s>", buf);
2764 return 0;
2765 }
2767 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2768 enum object_type type, const char *path, unsigned flags)
2769 {
2770 int ret;
2771 size_t size = xsize_t(st->st_size);
2773 if (!S_ISREG(st->st_mode))
2774 ret = index_pipe(sha1, fd, type, path, flags);
2775 else if (size <= big_file_threshold || type != OBJ_BLOB)
2776 ret = index_core(sha1, fd, size, type, path, flags);
2777 else
2778 ret = index_stream(sha1, fd, size, type, path, flags);
2779 close(fd);
2780 return ret;
2781 }
2783 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2784 {
2785 int fd;
2786 struct strbuf sb = STRBUF_INIT;
2788 switch (st->st_mode & S_IFMT) {
2789 case S_IFREG:
2790 fd = open(path, O_RDONLY);
2791 if (fd < 0)
2792 return error("open(\"%s\"): %s", path,
2793 strerror(errno));
2794 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2795 return error("%s: failed to insert into database",
2796 path);
2797 break;
2798 case S_IFLNK:
2799 if (strbuf_readlink(&sb, path, st->st_size)) {
2800 char *errstr = strerror(errno);
2801 return error("readlink(\"%s\"): %s", path,
2802 errstr);
2803 }
2804 if (!(flags & HASH_WRITE_OBJECT))
2805 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2806 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2807 return error("%s: failed to insert into database",
2808 path);
2809 strbuf_release(&sb);
2810 break;
2811 case S_IFDIR:
2812 return resolve_gitlink_ref(path, "HEAD", sha1);
2813 default:
2814 return error("%s: unsupported file type", path);
2815 }
2816 return 0;
2817 }
2819 int read_pack_header(int fd, struct pack_header *header)
2820 {
2821 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2822 /* "eof before pack header was fully read" */
2823 return PH_ERROR_EOF;
2825 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2826 /* "protocol error (pack signature mismatch detected)" */
2827 return PH_ERROR_PACK_SIGNATURE;
2828 if (!pack_version_ok(header->hdr_version))
2829 /* "protocol error (pack version unsupported)" */
2830 return PH_ERROR_PROTOCOL;
2831 return 0;
2832 }
2834 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2835 {
2836 enum object_type type = sha1_object_info(sha1, NULL);
2837 if (type < 0)
2838 die("%s is not a valid object", sha1_to_hex(sha1));
2839 if (type != expect)
2840 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2841 typename(expect));
2842 }