Move setup_diff_pager to libgit.a
[git/jnareb-git.git] / sha1_file.c
blob9152974642986de2307ec4ed6beda4efa49e8383
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
9 #include "cache.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "blob.h"
13 #include "commit.h"
14 #include "run-command.h"
15 #include "tag.h"
16 #include "tree.h"
17 #include "tree-walk.h"
18 #include "refs.h"
19 #include "pack-revindex.h"
20 #include "sha1-lookup.h"
21 #include "bulk-checkin.h"
22 #include "streaming.h"
24 #ifndef O_NOATIME
25 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
26 #define O_NOATIME 01000000
27 #else
28 #define O_NOATIME 0
29 #endif
30 #endif
32 #define SZ_FMT PRIuMAX
33 static inline uintmax_t sz_fmt(size_t s) { return s; }
35 const unsigned char null_sha1[20];
38 * This is meant to hold a *small* number of objects that you would
39 * want read_sha1_file() to be able to return, but yet you do not want
40 * to write them into the object store (e.g. a browse-only
41 * application).
43 static struct cached_object {
44 unsigned char sha1[20];
45 enum object_type type;
46 void *buf;
47 unsigned long size;
48 } *cached_objects;
49 static int cached_object_nr, cached_object_alloc;
51 static struct cached_object empty_tree = {
52 EMPTY_TREE_SHA1_BIN_LITERAL,
53 OBJ_TREE,
54 "",
58 static struct packed_git *last_found_pack;
60 static struct cached_object *find_cached_object(const unsigned char *sha1)
62 int i;
63 struct cached_object *co = cached_objects;
65 for (i = 0; i < cached_object_nr; i++, co++) {
66 if (!hashcmp(co->sha1, sha1))
67 return co;
69 if (!hashcmp(sha1, empty_tree.sha1))
70 return &empty_tree;
71 return NULL;
74 int mkdir_in_gitdir(const char *path)
76 if (mkdir(path, 0777)) {
77 int saved_errno = errno;
78 struct stat st;
79 struct strbuf sb = STRBUF_INIT;
81 if (errno != EEXIST)
82 return -1;
84 * Are we looking at a path in a symlinked worktree
85 * whose original repository does not yet have it?
86 * e.g. .git/rr-cache pointing at its original
87 * repository in which the user hasn't performed any
88 * conflict resolution yet?
90 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
91 strbuf_readlink(&sb, path, st.st_size) ||
92 !is_absolute_path(sb.buf) ||
93 mkdir(sb.buf, 0777)) {
94 strbuf_release(&sb);
95 errno = saved_errno;
96 return -1;
98 strbuf_release(&sb);
100 return adjust_shared_perm(path);
103 int safe_create_leading_directories(char *path)
105 char *pos = path + offset_1st_component(path);
106 struct stat st;
108 while (pos) {
109 pos = strchr(pos, '/');
110 if (!pos)
111 break;
112 while (*++pos == '/')
114 if (!*pos)
115 break;
116 *--pos = '\0';
117 if (!stat(path, &st)) {
118 /* path exists */
119 if (!S_ISDIR(st.st_mode)) {
120 *pos = '/';
121 return -3;
124 else if (mkdir(path, 0777)) {
125 *pos = '/';
126 return -1;
128 else if (adjust_shared_perm(path)) {
129 *pos = '/';
130 return -2;
132 *pos++ = '/';
134 return 0;
137 int safe_create_leading_directories_const(const char *path)
139 /* path points to cache entries, so xstrdup before messing with it */
140 char *buf = xstrdup(path);
141 int result = safe_create_leading_directories(buf);
142 free(buf);
143 return result;
146 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
148 int i;
149 for (i = 0; i < 20; i++) {
150 static char hex[] = "0123456789abcdef";
151 unsigned int val = sha1[i];
152 char *pos = pathbuf + i*2 + (i > 0);
153 *pos++ = hex[val >> 4];
154 *pos = hex[val & 0xf];
159 * NOTE! This returns a statically allocated buffer, so you have to be
160 * careful about using it. Do an "xstrdup()" if you need to save the
161 * filename.
163 * Also note that this returns the location for creating. Reading
164 * SHA1 file can happen from any alternate directory listed in the
165 * DB_ENVIRONMENT environment variable if it is not found in
166 * the primary object database.
168 char *sha1_file_name(const unsigned char *sha1)
170 static char buf[PATH_MAX];
171 const char *objdir;
172 int len;
174 objdir = get_object_directory();
175 len = strlen(objdir);
177 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
178 if (len + 43 > PATH_MAX)
179 die("insanely long object directory %s", objdir);
180 memcpy(buf, objdir, len);
181 buf[len] = '/';
182 buf[len+3] = '/';
183 buf[len+42] = '\0';
184 fill_sha1_path(buf + len + 1, sha1);
185 return buf;
188 static char *sha1_get_pack_name(const unsigned char *sha1,
189 char **name, char **base, const char *which)
191 static const char hex[] = "0123456789abcdef";
192 char *buf;
193 int i;
195 if (!*base) {
196 const char *sha1_file_directory = get_object_directory();
197 int len = strlen(sha1_file_directory);
198 *base = xmalloc(len + 60);
199 sprintf(*base, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
200 sha1_file_directory, which);
201 *name = *base + len + 11;
204 buf = *name;
206 for (i = 0; i < 20; i++) {
207 unsigned int val = *sha1++;
208 *buf++ = hex[val >> 4];
209 *buf++ = hex[val & 0xf];
212 return *base;
215 char *sha1_pack_name(const unsigned char *sha1)
217 static char *name, *base;
219 return sha1_get_pack_name(sha1, &name, &base, "pack");
222 char *sha1_pack_index_name(const unsigned char *sha1)
224 static char *name, *base;
226 return sha1_get_pack_name(sha1, &name, &base, "idx");
229 struct alternate_object_database *alt_odb_list;
230 static struct alternate_object_database **alt_odb_tail;
232 static int git_open_noatime(const char *name);
235 * Prepare alternate object database registry.
237 * The variable alt_odb_list points at the list of struct
238 * alternate_object_database. The elements on this list come from
239 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
240 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
241 * whose contents is similar to that environment variable but can be
242 * LF separated. Its base points at a statically allocated buffer that
243 * contains "/the/directory/corresponding/to/.git/objects/...", while
244 * its name points just after the slash at the end of ".git/objects/"
245 * in the example above, and has enough space to hold 40-byte hex
246 * SHA1, an extra slash for the first level indirection, and the
247 * terminating NUL.
249 static int link_alt_odb_entry(const char * entry, int len, const char * relative_base, int depth)
251 const char *objdir = get_object_directory();
252 struct alternate_object_database *ent;
253 struct alternate_object_database *alt;
254 int pfxlen, entlen;
255 struct strbuf pathbuf = STRBUF_INIT;
257 if (!is_absolute_path(entry) && relative_base) {
258 strbuf_addstr(&pathbuf, real_path(relative_base));
259 strbuf_addch(&pathbuf, '/');
261 strbuf_add(&pathbuf, entry, len);
263 normalize_path_copy(pathbuf.buf, pathbuf.buf);
265 pfxlen = strlen(pathbuf.buf);
268 * The trailing slash after the directory name is given by
269 * this function at the end. Remove duplicates.
271 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
272 pfxlen -= 1;
274 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
275 ent = xmalloc(sizeof(*ent) + entlen);
276 memcpy(ent->base, pathbuf.buf, pfxlen);
277 strbuf_release(&pathbuf);
279 ent->name = ent->base + pfxlen + 1;
280 ent->base[pfxlen + 3] = '/';
281 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283 /* Detect cases where alternate disappeared */
284 if (!is_directory(ent->base)) {
285 error("object directory %s does not exist; "
286 "check .git/objects/info/alternates.",
287 ent->base);
288 free(ent);
289 return -1;
292 /* Prevent the common mistake of listing the same
293 * thing twice, or object directory itself.
295 for (alt = alt_odb_list; alt; alt = alt->next) {
296 if (!memcmp(ent->base, alt->base, pfxlen)) {
297 free(ent);
298 return -1;
301 if (!strcmp(ent->base, objdir)) {
302 free(ent);
303 return -1;
306 /* add the alternate entry */
307 *alt_odb_tail = ent;
308 alt_odb_tail = &(ent->next);
309 ent->next = NULL;
311 /* recursively add alternates */
312 read_info_alternates(ent->base, depth + 1);
314 ent->base[pfxlen] = '/';
316 return 0;
319 static void link_alt_odb_entries(const char *alt, const char *ep, int sep,
320 const char *relative_base, int depth)
322 const char *cp, *last;
324 if (depth > 5) {
325 error("%s: ignoring alternate object stores, nesting too deep.",
326 relative_base);
327 return;
330 last = alt;
331 while (last < ep) {
332 cp = last;
333 if (cp < ep && *cp == '#') {
334 while (cp < ep && *cp != sep)
335 cp++;
336 last = cp + 1;
337 continue;
339 while (cp < ep && *cp != sep)
340 cp++;
341 if (last != cp) {
342 if (!is_absolute_path(last) && depth) {
343 error("%s: ignoring relative alternate object store %s",
344 relative_base, last);
345 } else {
346 link_alt_odb_entry(last, cp - last,
347 relative_base, depth);
350 while (cp < ep && *cp == sep)
351 cp++;
352 last = cp;
356 void read_info_alternates(const char * relative_base, int depth)
358 char *map;
359 size_t mapsz;
360 struct stat st;
361 const char alt_file_name[] = "info/alternates";
362 /* Given that relative_base is no longer than PATH_MAX,
363 ensure that "path" has enough space to append "/", the
364 file name, "info/alternates", and a trailing NUL. */
365 char path[PATH_MAX + 1 + sizeof alt_file_name];
366 int fd;
368 sprintf(path, "%s/%s", relative_base, alt_file_name);
369 fd = git_open_noatime(path);
370 if (fd < 0)
371 return;
372 if (fstat(fd, &st) || (st.st_size == 0)) {
373 close(fd);
374 return;
376 mapsz = xsize_t(st.st_size);
377 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
378 close(fd);
380 link_alt_odb_entries(map, map + mapsz, '\n', relative_base, depth);
382 munmap(map, mapsz);
385 void add_to_alternates_file(const char *reference)
387 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
388 int fd = hold_lock_file_for_append(lock, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR);
389 char *alt = mkpath("%s\n", reference);
390 write_or_die(fd, alt, strlen(alt));
391 if (commit_lock_file(lock))
392 die("could not close alternates file");
393 if (alt_odb_tail)
394 link_alt_odb_entries(alt, alt + strlen(alt), '\n', NULL, 0);
397 void foreach_alt_odb(alt_odb_fn fn, void *cb)
399 struct alternate_object_database *ent;
401 prepare_alt_odb();
402 for (ent = alt_odb_list; ent; ent = ent->next)
403 if (fn(ent, cb))
404 return;
407 void prepare_alt_odb(void)
409 const char *alt;
411 if (alt_odb_tail)
412 return;
414 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
415 if (!alt) alt = "";
417 alt_odb_tail = &alt_odb_list;
418 link_alt_odb_entries(alt, alt + strlen(alt), PATH_SEP, NULL, 0);
420 read_info_alternates(get_object_directory(), 0);
423 static int has_loose_object_local(const unsigned char *sha1)
425 char *name = sha1_file_name(sha1);
426 return !access(name, F_OK);
429 int has_loose_object_nonlocal(const unsigned char *sha1)
431 struct alternate_object_database *alt;
432 prepare_alt_odb();
433 for (alt = alt_odb_list; alt; alt = alt->next) {
434 fill_sha1_path(alt->name, sha1);
435 if (!access(alt->base, F_OK))
436 return 1;
438 return 0;
441 static int has_loose_object(const unsigned char *sha1)
443 return has_loose_object_local(sha1) ||
444 has_loose_object_nonlocal(sha1);
447 static unsigned int pack_used_ctr;
448 static unsigned int pack_mmap_calls;
449 static unsigned int peak_pack_open_windows;
450 static unsigned int pack_open_windows;
451 static unsigned int pack_open_fds;
452 static unsigned int pack_max_fds;
453 static size_t peak_pack_mapped;
454 static size_t pack_mapped;
455 struct packed_git *packed_git;
457 void pack_report(void)
459 fprintf(stderr,
460 "pack_report: getpagesize() = %10" SZ_FMT "\n"
461 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
462 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
463 sz_fmt(getpagesize()),
464 sz_fmt(packed_git_window_size),
465 sz_fmt(packed_git_limit));
466 fprintf(stderr,
467 "pack_report: pack_used_ctr = %10u\n"
468 "pack_report: pack_mmap_calls = %10u\n"
469 "pack_report: pack_open_windows = %10u / %10u\n"
470 "pack_report: pack_mapped = "
471 "%10" SZ_FMT " / %10" SZ_FMT "\n",
472 pack_used_ctr,
473 pack_mmap_calls,
474 pack_open_windows, peak_pack_open_windows,
475 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
478 static int check_packed_git_idx(const char *path, struct packed_git *p)
480 void *idx_map;
481 struct pack_idx_header *hdr;
482 size_t idx_size;
483 uint32_t version, nr, i, *index;
484 int fd = git_open_noatime(path);
485 struct stat st;
487 if (fd < 0)
488 return -1;
489 if (fstat(fd, &st)) {
490 close(fd);
491 return -1;
493 idx_size = xsize_t(st.st_size);
494 if (idx_size < 4 * 256 + 20 + 20) {
495 close(fd);
496 return error("index file %s is too small", path);
498 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
499 close(fd);
501 hdr = idx_map;
502 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
503 version = ntohl(hdr->idx_version);
504 if (version < 2 || version > 2) {
505 munmap(idx_map, idx_size);
506 return error("index file %s is version %"PRIu32
507 " and is not supported by this binary"
508 " (try upgrading GIT to a newer version)",
509 path, version);
511 } else
512 version = 1;
514 nr = 0;
515 index = idx_map;
516 if (version > 1)
517 index += 2; /* skip index header */
518 for (i = 0; i < 256; i++) {
519 uint32_t n = ntohl(index[i]);
520 if (n < nr) {
521 munmap(idx_map, idx_size);
522 return error("non-monotonic index %s", path);
524 nr = n;
527 if (version == 1) {
529 * Total size:
530 * - 256 index entries 4 bytes each
531 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
532 * - 20-byte SHA1 of the packfile
533 * - 20-byte SHA1 file checksum
535 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
536 munmap(idx_map, idx_size);
537 return error("wrong index v1 file size in %s", path);
539 } else if (version == 2) {
541 * Minimum size:
542 * - 8 bytes of header
543 * - 256 index entries 4 bytes each
544 * - 20-byte sha1 entry * nr
545 * - 4-byte crc entry * nr
546 * - 4-byte offset entry * nr
547 * - 20-byte SHA1 of the packfile
548 * - 20-byte SHA1 file checksum
549 * And after the 4-byte offset table might be a
550 * variable sized table containing 8-byte entries
551 * for offsets larger than 2^31.
553 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
554 unsigned long max_size = min_size;
555 if (nr)
556 max_size += (nr - 1)*8;
557 if (idx_size < min_size || idx_size > max_size) {
558 munmap(idx_map, idx_size);
559 return error("wrong index v2 file size in %s", path);
561 if (idx_size != min_size &&
563 * make sure we can deal with large pack offsets.
564 * 31-bit signed offset won't be enough, neither
565 * 32-bit unsigned one will be.
567 (sizeof(off_t) <= 4)) {
568 munmap(idx_map, idx_size);
569 return error("pack too large for current definition of off_t in %s", path);
573 p->index_version = version;
574 p->index_data = idx_map;
575 p->index_size = idx_size;
576 p->num_objects = nr;
577 return 0;
580 int open_pack_index(struct packed_git *p)
582 char *idx_name;
583 int ret;
585 if (p->index_data)
586 return 0;
588 idx_name = xstrdup(p->pack_name);
589 strcpy(idx_name + strlen(idx_name) - strlen(".pack"), ".idx");
590 ret = check_packed_git_idx(idx_name, p);
591 free(idx_name);
592 return ret;
595 static void scan_windows(struct packed_git *p,
596 struct packed_git **lru_p,
597 struct pack_window **lru_w,
598 struct pack_window **lru_l)
600 struct pack_window *w, *w_l;
602 for (w_l = NULL, w = p->windows; w; w = w->next) {
603 if (!w->inuse_cnt) {
604 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
605 *lru_p = p;
606 *lru_w = w;
607 *lru_l = w_l;
610 w_l = w;
614 static int unuse_one_window(struct packed_git *current, int keep_fd)
616 struct packed_git *p, *lru_p = NULL;
617 struct pack_window *lru_w = NULL, *lru_l = NULL;
619 if (current)
620 scan_windows(current, &lru_p, &lru_w, &lru_l);
621 for (p = packed_git; p; p = p->next)
622 scan_windows(p, &lru_p, &lru_w, &lru_l);
623 if (lru_p) {
624 munmap(lru_w->base, lru_w->len);
625 pack_mapped -= lru_w->len;
626 if (lru_l)
627 lru_l->next = lru_w->next;
628 else {
629 lru_p->windows = lru_w->next;
630 if (!lru_p->windows && lru_p->pack_fd != -1
631 && lru_p->pack_fd != keep_fd) {
632 close(lru_p->pack_fd);
633 pack_open_fds--;
634 lru_p->pack_fd = -1;
637 free(lru_w);
638 pack_open_windows--;
639 return 1;
641 return 0;
644 void release_pack_memory(size_t need, int fd)
646 size_t cur = pack_mapped;
647 while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
648 ; /* nothing */
651 void *xmmap(void *start, size_t length,
652 int prot, int flags, int fd, off_t offset)
654 void *ret = mmap(start, length, prot, flags, fd, offset);
655 if (ret == MAP_FAILED) {
656 if (!length)
657 return NULL;
658 release_pack_memory(length, fd);
659 ret = mmap(start, length, prot, flags, fd, offset);
660 if (ret == MAP_FAILED)
661 die_errno("Out of memory? mmap failed");
663 return ret;
666 void close_pack_windows(struct packed_git *p)
668 while (p->windows) {
669 struct pack_window *w = p->windows;
671 if (w->inuse_cnt)
672 die("pack '%s' still has open windows to it",
673 p->pack_name);
674 munmap(w->base, w->len);
675 pack_mapped -= w->len;
676 pack_open_windows--;
677 p->windows = w->next;
678 free(w);
682 void unuse_pack(struct pack_window **w_cursor)
684 struct pack_window *w = *w_cursor;
685 if (w) {
686 w->inuse_cnt--;
687 *w_cursor = NULL;
691 void close_pack_index(struct packed_git *p)
693 if (p->index_data) {
694 munmap((void *)p->index_data, p->index_size);
695 p->index_data = NULL;
700 * This is used by git-repack in case a newly created pack happens to
701 * contain the same set of objects as an existing one. In that case
702 * the resulting file might be different even if its name would be the
703 * same. It is best to close any reference to the old pack before it is
704 * replaced on disk. Of course no index pointers nor windows for given pack
705 * must subsist at this point. If ever objects from this pack are requested
706 * again, the new version of the pack will be reinitialized through
707 * reprepare_packed_git().
709 void free_pack_by_name(const char *pack_name)
711 struct packed_git *p, **pp = &packed_git;
713 while (*pp) {
714 p = *pp;
715 if (strcmp(pack_name, p->pack_name) == 0) {
716 clear_delta_base_cache();
717 close_pack_windows(p);
718 if (p->pack_fd != -1) {
719 close(p->pack_fd);
720 pack_open_fds--;
722 close_pack_index(p);
723 free(p->bad_object_sha1);
724 *pp = p->next;
725 if (last_found_pack == p)
726 last_found_pack = NULL;
727 free(p);
728 return;
730 pp = &p->next;
734 static unsigned int get_max_fd_limit(void)
736 #ifdef RLIMIT_NOFILE
737 struct rlimit lim;
739 if (getrlimit(RLIMIT_NOFILE, &lim))
740 die_errno("cannot get RLIMIT_NOFILE");
742 return lim.rlim_cur;
743 #elif defined(_SC_OPEN_MAX)
744 return sysconf(_SC_OPEN_MAX);
745 #elif defined(OPEN_MAX)
746 return OPEN_MAX;
747 #else
748 return 1; /* see the caller ;-) */
749 #endif
753 * Do not call this directly as this leaks p->pack_fd on error return;
754 * call open_packed_git() instead.
756 static int open_packed_git_1(struct packed_git *p)
758 struct stat st;
759 struct pack_header hdr;
760 unsigned char sha1[20];
761 unsigned char *idx_sha1;
762 long fd_flag;
764 if (!p->index_data && open_pack_index(p))
765 return error("packfile %s index unavailable", p->pack_name);
767 if (!pack_max_fds) {
768 unsigned int max_fds = get_max_fd_limit();
770 /* Save 3 for stdin/stdout/stderr, 22 for work */
771 if (25 < max_fds)
772 pack_max_fds = max_fds - 25;
773 else
774 pack_max_fds = 1;
777 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
778 ; /* nothing */
780 p->pack_fd = git_open_noatime(p->pack_name);
781 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
782 return -1;
783 pack_open_fds++;
785 /* If we created the struct before we had the pack we lack size. */
786 if (!p->pack_size) {
787 if (!S_ISREG(st.st_mode))
788 return error("packfile %s not a regular file", p->pack_name);
789 p->pack_size = st.st_size;
790 } else if (p->pack_size != st.st_size)
791 return error("packfile %s size changed", p->pack_name);
793 /* We leave these file descriptors open with sliding mmap;
794 * there is no point keeping them open across exec(), though.
796 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
797 if (fd_flag < 0)
798 return error("cannot determine file descriptor flags");
799 fd_flag |= FD_CLOEXEC;
800 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
801 return error("cannot set FD_CLOEXEC");
803 /* Verify we recognize this pack file format. */
804 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
805 return error("file %s is far too short to be a packfile", p->pack_name);
806 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
807 return error("file %s is not a GIT packfile", p->pack_name);
808 if (!pack_version_ok(hdr.hdr_version))
809 return error("packfile %s is version %"PRIu32" and not"
810 " supported (try upgrading GIT to a newer version)",
811 p->pack_name, ntohl(hdr.hdr_version));
813 /* Verify the pack matches its index. */
814 if (p->num_objects != ntohl(hdr.hdr_entries))
815 return error("packfile %s claims to have %"PRIu32" objects"
816 " while index indicates %"PRIu32" objects",
817 p->pack_name, ntohl(hdr.hdr_entries),
818 p->num_objects);
819 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
820 return error("end of packfile %s is unavailable", p->pack_name);
821 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
822 return error("packfile %s signature is unavailable", p->pack_name);
823 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
824 if (hashcmp(sha1, idx_sha1))
825 return error("packfile %s does not match index", p->pack_name);
826 return 0;
829 static int open_packed_git(struct packed_git *p)
831 if (!open_packed_git_1(p))
832 return 0;
833 if (p->pack_fd != -1) {
834 close(p->pack_fd);
835 pack_open_fds--;
836 p->pack_fd = -1;
838 return -1;
841 static int in_window(struct pack_window *win, off_t offset)
843 /* We must promise at least 20 bytes (one hash) after the
844 * offset is available from this window, otherwise the offset
845 * is not actually in this window and a different window (which
846 * has that one hash excess) must be used. This is to support
847 * the object header and delta base parsing routines below.
849 off_t win_off = win->offset;
850 return win_off <= offset
851 && (offset + 20) <= (win_off + win->len);
854 unsigned char *use_pack(struct packed_git *p,
855 struct pack_window **w_cursor,
856 off_t offset,
857 unsigned long *left)
859 struct pack_window *win = *w_cursor;
861 /* Since packfiles end in a hash of their content and it's
862 * pointless to ask for an offset into the middle of that
863 * hash, and the in_window function above wouldn't match
864 * don't allow an offset too close to the end of the file.
866 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
867 die("packfile %s cannot be accessed", p->pack_name);
868 if (offset > (p->pack_size - 20))
869 die("offset beyond end of packfile (truncated pack?)");
871 if (!win || !in_window(win, offset)) {
872 if (win)
873 win->inuse_cnt--;
874 for (win = p->windows; win; win = win->next) {
875 if (in_window(win, offset))
876 break;
878 if (!win) {
879 size_t window_align = packed_git_window_size / 2;
880 off_t len;
882 if (p->pack_fd == -1 && open_packed_git(p))
883 die("packfile %s cannot be accessed", p->pack_name);
885 win = xcalloc(1, sizeof(*win));
886 win->offset = (offset / window_align) * window_align;
887 len = p->pack_size - win->offset;
888 if (len > packed_git_window_size)
889 len = packed_git_window_size;
890 win->len = (size_t)len;
891 pack_mapped += win->len;
892 while (packed_git_limit < pack_mapped
893 && unuse_one_window(p, p->pack_fd))
894 ; /* nothing */
895 win->base = xmmap(NULL, win->len,
896 PROT_READ, MAP_PRIVATE,
897 p->pack_fd, win->offset);
898 if (win->base == MAP_FAILED)
899 die("packfile %s cannot be mapped: %s",
900 p->pack_name,
901 strerror(errno));
902 if (!win->offset && win->len == p->pack_size
903 && !p->do_not_close) {
904 close(p->pack_fd);
905 pack_open_fds--;
906 p->pack_fd = -1;
908 pack_mmap_calls++;
909 pack_open_windows++;
910 if (pack_mapped > peak_pack_mapped)
911 peak_pack_mapped = pack_mapped;
912 if (pack_open_windows > peak_pack_open_windows)
913 peak_pack_open_windows = pack_open_windows;
914 win->next = p->windows;
915 p->windows = win;
918 if (win != *w_cursor) {
919 win->last_used = pack_used_ctr++;
920 win->inuse_cnt++;
921 *w_cursor = win;
923 offset -= win->offset;
924 if (left)
925 *left = win->len - xsize_t(offset);
926 return win->base + offset;
929 static struct packed_git *alloc_packed_git(int extra)
931 struct packed_git *p = xmalloc(sizeof(*p) + extra);
932 memset(p, 0, sizeof(*p));
933 p->pack_fd = -1;
934 return p;
937 static void try_to_free_pack_memory(size_t size)
939 release_pack_memory(size, -1);
942 struct packed_git *add_packed_git(const char *path, int path_len, int local)
944 static int have_set_try_to_free_routine;
945 struct stat st;
946 struct packed_git *p = alloc_packed_git(path_len + 2);
948 if (!have_set_try_to_free_routine) {
949 have_set_try_to_free_routine = 1;
950 set_try_to_free_routine(try_to_free_pack_memory);
954 * Make sure a corresponding .pack file exists and that
955 * the index looks sane.
957 path_len -= strlen(".idx");
958 if (path_len < 1) {
959 free(p);
960 return NULL;
962 memcpy(p->pack_name, path, path_len);
964 strcpy(p->pack_name + path_len, ".keep");
965 if (!access(p->pack_name, F_OK))
966 p->pack_keep = 1;
968 strcpy(p->pack_name + path_len, ".pack");
969 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
970 free(p);
971 return NULL;
974 /* ok, it looks sane as far as we can check without
975 * actually mapping the pack file.
977 p->pack_size = st.st_size;
978 p->pack_local = local;
979 p->mtime = st.st_mtime;
980 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
981 hashclr(p->sha1);
982 return p;
985 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
987 const char *path = sha1_pack_name(sha1);
988 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
990 strcpy(p->pack_name, path);
991 hashcpy(p->sha1, sha1);
992 if (check_packed_git_idx(idx_path, p)) {
993 free(p);
994 return NULL;
997 return p;
1000 void install_packed_git(struct packed_git *pack)
1002 if (pack->pack_fd != -1)
1003 pack_open_fds++;
1005 pack->next = packed_git;
1006 packed_git = pack;
1009 static void prepare_packed_git_one(char *objdir, int local)
1011 /* Ensure that this buffer is large enough so that we can
1012 append "/pack/" without clobbering the stack even if
1013 strlen(objdir) were PATH_MAX. */
1014 char path[PATH_MAX + 1 + 4 + 1 + 1];
1015 int len;
1016 DIR *dir;
1017 struct dirent *de;
1019 sprintf(path, "%s/pack", objdir);
1020 len = strlen(path);
1021 dir = opendir(path);
1022 if (!dir) {
1023 if (errno != ENOENT)
1024 error("unable to open object pack directory: %s: %s",
1025 path, strerror(errno));
1026 return;
1028 path[len++] = '/';
1029 while ((de = readdir(dir)) != NULL) {
1030 int namelen = strlen(de->d_name);
1031 struct packed_git *p;
1033 if (!has_extension(de->d_name, ".idx"))
1034 continue;
1036 if (len + namelen + 1 > sizeof(path))
1037 continue;
1039 /* Don't reopen a pack we already have. */
1040 strcpy(path + len, de->d_name);
1041 for (p = packed_git; p; p = p->next) {
1042 if (!memcmp(path, p->pack_name, len + namelen - 4))
1043 break;
1045 if (p)
1046 continue;
1047 /* See if it really is a valid .idx file with corresponding
1048 * .pack file that we can map.
1050 p = add_packed_git(path, len + namelen, local);
1051 if (!p)
1052 continue;
1053 install_packed_git(p);
1055 closedir(dir);
1058 static int sort_pack(const void *a_, const void *b_)
1060 struct packed_git *a = *((struct packed_git **)a_);
1061 struct packed_git *b = *((struct packed_git **)b_);
1062 int st;
1065 * Local packs tend to contain objects specific to our
1066 * variant of the project than remote ones. In addition,
1067 * remote ones could be on a network mounted filesystem.
1068 * Favor local ones for these reasons.
1070 st = a->pack_local - b->pack_local;
1071 if (st)
1072 return -st;
1075 * Younger packs tend to contain more recent objects,
1076 * and more recent objects tend to get accessed more
1077 * often.
1079 if (a->mtime < b->mtime)
1080 return 1;
1081 else if (a->mtime == b->mtime)
1082 return 0;
1083 return -1;
1086 static void rearrange_packed_git(void)
1088 struct packed_git **ary, *p;
1089 int i, n;
1091 for (n = 0, p = packed_git; p; p = p->next)
1092 n++;
1093 if (n < 2)
1094 return;
1096 /* prepare an array of packed_git for easier sorting */
1097 ary = xcalloc(n, sizeof(struct packed_git *));
1098 for (n = 0, p = packed_git; p; p = p->next)
1099 ary[n++] = p;
1101 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1103 /* link them back again */
1104 for (i = 0; i < n - 1; i++)
1105 ary[i]->next = ary[i + 1];
1106 ary[n - 1]->next = NULL;
1107 packed_git = ary[0];
1109 free(ary);
1112 static int prepare_packed_git_run_once = 0;
1113 void prepare_packed_git(void)
1115 struct alternate_object_database *alt;
1117 if (prepare_packed_git_run_once)
1118 return;
1119 prepare_packed_git_one(get_object_directory(), 1);
1120 prepare_alt_odb();
1121 for (alt = alt_odb_list; alt; alt = alt->next) {
1122 alt->name[-1] = 0;
1123 prepare_packed_git_one(alt->base, 0);
1124 alt->name[-1] = '/';
1126 rearrange_packed_git();
1127 prepare_packed_git_run_once = 1;
1130 void reprepare_packed_git(void)
1132 discard_revindex();
1133 prepare_packed_git_run_once = 0;
1134 prepare_packed_git();
1137 static void mark_bad_packed_object(struct packed_git *p,
1138 const unsigned char *sha1)
1140 unsigned i;
1141 for (i = 0; i < p->num_bad_objects; i++)
1142 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1143 return;
1144 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1145 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1146 p->num_bad_objects++;
1149 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1151 struct packed_git *p;
1152 unsigned i;
1154 for (p = packed_git; p; p = p->next)
1155 for (i = 0; i < p->num_bad_objects; i++)
1156 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1157 return p;
1158 return NULL;
1162 * With an in-core object data in "map", rehash it to make sure the
1163 * object name actually matches "sha1" to detect object corruption.
1164 * With "map" == NULL, try reading the object named with "sha1" using
1165 * the streaming interface and rehash it to do the same.
1167 int check_sha1_signature(const unsigned char *sha1, void *map,
1168 unsigned long size, const char *type)
1170 unsigned char real_sha1[20];
1171 enum object_type obj_type;
1172 struct git_istream *st;
1173 git_SHA_CTX c;
1174 char hdr[32];
1175 int hdrlen;
1177 if (map) {
1178 hash_sha1_file(map, size, type, real_sha1);
1179 return hashcmp(sha1, real_sha1) ? -1 : 0;
1182 st = open_istream(sha1, &obj_type, &size, NULL);
1183 if (!st)
1184 return -1;
1186 /* Generate the header */
1187 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1189 /* Sha1.. */
1190 git_SHA1_Init(&c);
1191 git_SHA1_Update(&c, hdr, hdrlen);
1192 for (;;) {
1193 char buf[1024 * 16];
1194 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1196 if (!readlen)
1197 break;
1198 git_SHA1_Update(&c, buf, readlen);
1200 git_SHA1_Final(real_sha1, &c);
1201 close_istream(st);
1202 return hashcmp(sha1, real_sha1) ? -1 : 0;
1205 static int git_open_noatime(const char *name)
1207 static int sha1_file_open_flag = O_NOATIME;
1209 for (;;) {
1210 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1211 if (fd >= 0)
1212 return fd;
1214 /* Might the failure be due to O_NOATIME? */
1215 if (errno != ENOENT && sha1_file_open_flag) {
1216 sha1_file_open_flag = 0;
1217 continue;
1220 return -1;
1224 static int open_sha1_file(const unsigned char *sha1)
1226 int fd;
1227 char *name = sha1_file_name(sha1);
1228 struct alternate_object_database *alt;
1230 fd = git_open_noatime(name);
1231 if (fd >= 0)
1232 return fd;
1234 prepare_alt_odb();
1235 errno = ENOENT;
1236 for (alt = alt_odb_list; alt; alt = alt->next) {
1237 name = alt->name;
1238 fill_sha1_path(name, sha1);
1239 fd = git_open_noatime(alt->base);
1240 if (fd >= 0)
1241 return fd;
1243 return -1;
1246 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1248 void *map;
1249 int fd;
1251 fd = open_sha1_file(sha1);
1252 map = NULL;
1253 if (fd >= 0) {
1254 struct stat st;
1256 if (!fstat(fd, &st)) {
1257 *size = xsize_t(st.st_size);
1258 if (!*size) {
1259 /* mmap() is forbidden on empty files */
1260 error("object file %s is empty", sha1_file_name(sha1));
1261 return NULL;
1263 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1265 close(fd);
1267 return map;
1271 * There used to be a second loose object header format which
1272 * was meant to mimic the in-pack format, allowing for direct
1273 * copy of the object data. This format turned up not to be
1274 * really worth it and we no longer write loose objects in that
1275 * format.
1277 static int experimental_loose_object(unsigned char *map)
1279 unsigned int word;
1282 * We must determine if the buffer contains the standard
1283 * zlib-deflated stream or the experimental format based
1284 * on the in-pack object format. Compare the header byte
1285 * for each format:
1287 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1288 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1290 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1291 * in standard loose-object format, UNLESS it is a Git-pack
1292 * format object *exactly* 8 bytes in size when inflated.
1294 * However, RFC1950 also specifies that the 1st 16-bit word
1295 * must be divisible by 31 - this checksum tells us our buffer
1296 * is in the standard format, giving a false positive only if
1297 * the 1st word of the Git-pack format object happens to be
1298 * divisible by 31, ie:
1299 * ((byte0 * 256) + byte1) % 31 = 0
1300 * => 0ttt10000www1000 % 31 = 0
1302 * As it happens, this case can only arise for www=3 & ttt=1
1303 * - ie, a Commit object, which would have to be 8 bytes in
1304 * size. As no Commit can be that small, we find that the
1305 * combination of these two criteria (bitmask & checksum)
1306 * can always correctly determine the buffer format.
1308 word = (map[0] << 8) + map[1];
1309 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1310 return 0;
1311 else
1312 return 1;
1315 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1316 unsigned long len, enum object_type *type, unsigned long *sizep)
1318 unsigned shift;
1319 unsigned long size, c;
1320 unsigned long used = 0;
1322 c = buf[used++];
1323 *type = (c >> 4) & 7;
1324 size = c & 15;
1325 shift = 4;
1326 while (c & 0x80) {
1327 if (len <= used || bitsizeof(long) <= shift) {
1328 error("bad object header");
1329 size = used = 0;
1330 break;
1332 c = buf[used++];
1333 size += (c & 0x7f) << shift;
1334 shift += 7;
1336 *sizep = size;
1337 return used;
1340 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1342 unsigned long size, used;
1343 static const char valid_loose_object_type[8] = {
1344 0, /* OBJ_EXT */
1345 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1346 0, /* "delta" and others are invalid in a loose object */
1348 enum object_type type;
1350 /* Get the data stream */
1351 memset(stream, 0, sizeof(*stream));
1352 stream->next_in = map;
1353 stream->avail_in = mapsize;
1354 stream->next_out = buffer;
1355 stream->avail_out = bufsiz;
1357 if (experimental_loose_object(map)) {
1359 * The old experimental format we no longer produce;
1360 * we can still read it.
1362 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1363 if (!used || !valid_loose_object_type[type])
1364 return -1;
1365 map += used;
1366 mapsize -= used;
1368 /* Set up the stream for the rest.. */
1369 stream->next_in = map;
1370 stream->avail_in = mapsize;
1371 git_inflate_init(stream);
1373 /* And generate the fake traditional header */
1374 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1375 typename(type), size);
1376 return 0;
1378 git_inflate_init(stream);
1379 return git_inflate(stream, 0);
1382 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1384 int bytes = strlen(buffer) + 1;
1385 unsigned char *buf = xmallocz(size);
1386 unsigned long n;
1387 int status = Z_OK;
1389 n = stream->total_out - bytes;
1390 if (n > size)
1391 n = size;
1392 memcpy(buf, (char *) buffer + bytes, n);
1393 bytes = n;
1394 if (bytes <= size) {
1396 * The above condition must be (bytes <= size), not
1397 * (bytes < size). In other words, even though we
1398 * expect no more output and set avail_out to zero,
1399 * the input zlib stream may have bytes that express
1400 * "this concludes the stream", and we *do* want to
1401 * eat that input.
1403 * Otherwise we would not be able to test that we
1404 * consumed all the input to reach the expected size;
1405 * we also want to check that zlib tells us that all
1406 * went well with status == Z_STREAM_END at the end.
1408 stream->next_out = buf + bytes;
1409 stream->avail_out = size - bytes;
1410 while (status == Z_OK)
1411 status = git_inflate(stream, Z_FINISH);
1413 if (status == Z_STREAM_END && !stream->avail_in) {
1414 git_inflate_end(stream);
1415 return buf;
1418 if (status < 0)
1419 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1420 else if (stream->avail_in)
1421 error("garbage at end of loose object '%s'",
1422 sha1_to_hex(sha1));
1423 free(buf);
1424 return NULL;
1428 * We used to just use "sscanf()", but that's actually way
1429 * too permissive for what we want to check. So do an anal
1430 * object header parse by hand.
1432 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1434 char type[10];
1435 int i;
1436 unsigned long size;
1439 * The type can be at most ten bytes (including the
1440 * terminating '\0' that we add), and is followed by
1441 * a space.
1443 i = 0;
1444 for (;;) {
1445 char c = *hdr++;
1446 if (c == ' ')
1447 break;
1448 type[i++] = c;
1449 if (i >= sizeof(type))
1450 return -1;
1452 type[i] = 0;
1455 * The length must follow immediately, and be in canonical
1456 * decimal format (ie "010" is not valid).
1458 size = *hdr++ - '0';
1459 if (size > 9)
1460 return -1;
1461 if (size) {
1462 for (;;) {
1463 unsigned long c = *hdr - '0';
1464 if (c > 9)
1465 break;
1466 hdr++;
1467 size = size * 10 + c;
1470 *sizep = size;
1473 * The length must be followed by a zero byte
1475 return *hdr ? -1 : type_from_string(type);
1478 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1480 int ret;
1481 git_zstream stream;
1482 char hdr[8192];
1484 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1485 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1486 return NULL;
1488 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1491 unsigned long get_size_from_delta(struct packed_git *p,
1492 struct pack_window **w_curs,
1493 off_t curpos)
1495 const unsigned char *data;
1496 unsigned char delta_head[20], *in;
1497 git_zstream stream;
1498 int st;
1500 memset(&stream, 0, sizeof(stream));
1501 stream.next_out = delta_head;
1502 stream.avail_out = sizeof(delta_head);
1504 git_inflate_init(&stream);
1505 do {
1506 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1507 stream.next_in = in;
1508 st = git_inflate(&stream, Z_FINISH);
1509 curpos += stream.next_in - in;
1510 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1511 stream.total_out < sizeof(delta_head));
1512 git_inflate_end(&stream);
1513 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1514 error("delta data unpack-initial failed");
1515 return 0;
1518 /* Examine the initial part of the delta to figure out
1519 * the result size.
1521 data = delta_head;
1523 /* ignore base size */
1524 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1526 /* Read the result size */
1527 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1530 static off_t get_delta_base(struct packed_git *p,
1531 struct pack_window **w_curs,
1532 off_t *curpos,
1533 enum object_type type,
1534 off_t delta_obj_offset)
1536 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1537 off_t base_offset;
1539 /* use_pack() assured us we have [base_info, base_info + 20)
1540 * as a range that we can look at without walking off the
1541 * end of the mapped window. Its actually the hash size
1542 * that is assured. An OFS_DELTA longer than the hash size
1543 * is stupid, as then a REF_DELTA would be smaller to store.
1545 if (type == OBJ_OFS_DELTA) {
1546 unsigned used = 0;
1547 unsigned char c = base_info[used++];
1548 base_offset = c & 127;
1549 while (c & 128) {
1550 base_offset += 1;
1551 if (!base_offset || MSB(base_offset, 7))
1552 return 0; /* overflow */
1553 c = base_info[used++];
1554 base_offset = (base_offset << 7) + (c & 127);
1556 base_offset = delta_obj_offset - base_offset;
1557 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1558 return 0; /* out of bound */
1559 *curpos += used;
1560 } else if (type == OBJ_REF_DELTA) {
1561 /* The base entry _must_ be in the same pack */
1562 base_offset = find_pack_entry_one(base_info, p);
1563 *curpos += 20;
1564 } else
1565 die("I am totally screwed");
1566 return base_offset;
1569 /* forward declaration for a mutually recursive function */
1570 static int packed_object_info(struct packed_git *p, off_t offset,
1571 unsigned long *sizep, int *rtype);
1573 static int packed_delta_info(struct packed_git *p,
1574 struct pack_window **w_curs,
1575 off_t curpos,
1576 enum object_type type,
1577 off_t obj_offset,
1578 unsigned long *sizep)
1580 off_t base_offset;
1582 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1583 if (!base_offset)
1584 return OBJ_BAD;
1585 type = packed_object_info(p, base_offset, NULL, NULL);
1586 if (type <= OBJ_NONE) {
1587 struct revindex_entry *revidx;
1588 const unsigned char *base_sha1;
1589 revidx = find_pack_revindex(p, base_offset);
1590 if (!revidx)
1591 return OBJ_BAD;
1592 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1593 mark_bad_packed_object(p, base_sha1);
1594 type = sha1_object_info(base_sha1, NULL);
1595 if (type <= OBJ_NONE)
1596 return OBJ_BAD;
1599 /* We choose to only get the type of the base object and
1600 * ignore potentially corrupt pack file that expects the delta
1601 * based on a base with a wrong size. This saves tons of
1602 * inflate() calls.
1604 if (sizep) {
1605 *sizep = get_size_from_delta(p, w_curs, curpos);
1606 if (*sizep == 0)
1607 type = OBJ_BAD;
1610 return type;
1613 int unpack_object_header(struct packed_git *p,
1614 struct pack_window **w_curs,
1615 off_t *curpos,
1616 unsigned long *sizep)
1618 unsigned char *base;
1619 unsigned long left;
1620 unsigned long used;
1621 enum object_type type;
1623 /* use_pack() assures us we have [base, base + 20) available
1624 * as a range that we can look at. (Its actually the hash
1625 * size that is assured.) With our object header encoding
1626 * the maximum deflated object size is 2^137, which is just
1627 * insane, so we know won't exceed what we have been given.
1629 base = use_pack(p, w_curs, *curpos, &left);
1630 used = unpack_object_header_buffer(base, left, &type, sizep);
1631 if (!used) {
1632 type = OBJ_BAD;
1633 } else
1634 *curpos += used;
1636 return type;
1639 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1640 unsigned long *sizep, int *rtype)
1642 struct pack_window *w_curs = NULL;
1643 unsigned long size;
1644 off_t curpos = obj_offset;
1645 enum object_type type;
1647 type = unpack_object_header(p, &w_curs, &curpos, &size);
1648 if (rtype)
1649 *rtype = type; /* representation type */
1651 switch (type) {
1652 case OBJ_OFS_DELTA:
1653 case OBJ_REF_DELTA:
1654 type = packed_delta_info(p, &w_curs, curpos,
1655 type, obj_offset, sizep);
1656 break;
1657 case OBJ_COMMIT:
1658 case OBJ_TREE:
1659 case OBJ_BLOB:
1660 case OBJ_TAG:
1661 if (sizep)
1662 *sizep = size;
1663 break;
1664 default:
1665 error("unknown object type %i at offset %"PRIuMAX" in %s",
1666 type, (uintmax_t)obj_offset, p->pack_name);
1667 type = OBJ_BAD;
1669 unuse_pack(&w_curs);
1670 return type;
1673 static void *unpack_compressed_entry(struct packed_git *p,
1674 struct pack_window **w_curs,
1675 off_t curpos,
1676 unsigned long size)
1678 int st;
1679 git_zstream stream;
1680 unsigned char *buffer, *in;
1682 buffer = xmallocz(size);
1683 memset(&stream, 0, sizeof(stream));
1684 stream.next_out = buffer;
1685 stream.avail_out = size + 1;
1687 git_inflate_init(&stream);
1688 do {
1689 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1690 stream.next_in = in;
1691 st = git_inflate(&stream, Z_FINISH);
1692 if (!stream.avail_out)
1693 break; /* the payload is larger than it should be */
1694 curpos += stream.next_in - in;
1695 } while (st == Z_OK || st == Z_BUF_ERROR);
1696 git_inflate_end(&stream);
1697 if ((st != Z_STREAM_END) || stream.total_out != size) {
1698 free(buffer);
1699 return NULL;
1702 return buffer;
1705 #define MAX_DELTA_CACHE (256)
1707 static size_t delta_base_cached;
1709 static struct delta_base_cache_lru_list {
1710 struct delta_base_cache_lru_list *prev;
1711 struct delta_base_cache_lru_list *next;
1712 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1714 static struct delta_base_cache_entry {
1715 struct delta_base_cache_lru_list lru;
1716 void *data;
1717 struct packed_git *p;
1718 off_t base_offset;
1719 unsigned long size;
1720 enum object_type type;
1721 } delta_base_cache[MAX_DELTA_CACHE];
1723 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1725 unsigned long hash;
1727 hash = (unsigned long)p + (unsigned long)base_offset;
1728 hash += (hash >> 8) + (hash >> 16);
1729 return hash % MAX_DELTA_CACHE;
1732 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1734 unsigned long hash = pack_entry_hash(p, base_offset);
1735 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1736 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1739 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1740 unsigned long *base_size, enum object_type *type, int keep_cache)
1742 void *ret;
1743 unsigned long hash = pack_entry_hash(p, base_offset);
1744 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1746 ret = ent->data;
1747 if (!ret || ent->p != p || ent->base_offset != base_offset)
1748 return unpack_entry(p, base_offset, type, base_size);
1750 if (!keep_cache) {
1751 ent->data = NULL;
1752 ent->lru.next->prev = ent->lru.prev;
1753 ent->lru.prev->next = ent->lru.next;
1754 delta_base_cached -= ent->size;
1755 } else {
1756 ret = xmemdupz(ent->data, ent->size);
1758 *type = ent->type;
1759 *base_size = ent->size;
1760 return ret;
1763 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1765 if (ent->data) {
1766 free(ent->data);
1767 ent->data = NULL;
1768 ent->lru.next->prev = ent->lru.prev;
1769 ent->lru.prev->next = ent->lru.next;
1770 delta_base_cached -= ent->size;
1774 void clear_delta_base_cache(void)
1776 unsigned long p;
1777 for (p = 0; p < MAX_DELTA_CACHE; p++)
1778 release_delta_base_cache(&delta_base_cache[p]);
1781 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1782 void *base, unsigned long base_size, enum object_type type)
1784 unsigned long hash = pack_entry_hash(p, base_offset);
1785 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1786 struct delta_base_cache_lru_list *lru;
1788 release_delta_base_cache(ent);
1789 delta_base_cached += base_size;
1791 for (lru = delta_base_cache_lru.next;
1792 delta_base_cached > delta_base_cache_limit
1793 && lru != &delta_base_cache_lru;
1794 lru = lru->next) {
1795 struct delta_base_cache_entry *f = (void *)lru;
1796 if (f->type == OBJ_BLOB)
1797 release_delta_base_cache(f);
1799 for (lru = delta_base_cache_lru.next;
1800 delta_base_cached > delta_base_cache_limit
1801 && lru != &delta_base_cache_lru;
1802 lru = lru->next) {
1803 struct delta_base_cache_entry *f = (void *)lru;
1804 release_delta_base_cache(f);
1807 ent->p = p;
1808 ent->base_offset = base_offset;
1809 ent->type = type;
1810 ent->data = base;
1811 ent->size = base_size;
1812 ent->lru.next = &delta_base_cache_lru;
1813 ent->lru.prev = delta_base_cache_lru.prev;
1814 delta_base_cache_lru.prev->next = &ent->lru;
1815 delta_base_cache_lru.prev = &ent->lru;
1818 static void *read_object(const unsigned char *sha1, enum object_type *type,
1819 unsigned long *size);
1821 static void *unpack_delta_entry(struct packed_git *p,
1822 struct pack_window **w_curs,
1823 off_t curpos,
1824 unsigned long delta_size,
1825 off_t obj_offset,
1826 enum object_type *type,
1827 unsigned long *sizep)
1829 void *delta_data, *result, *base;
1830 unsigned long base_size;
1831 off_t base_offset;
1833 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1834 if (!base_offset) {
1835 error("failed to validate delta base reference "
1836 "at offset %"PRIuMAX" from %s",
1837 (uintmax_t)curpos, p->pack_name);
1838 return NULL;
1840 unuse_pack(w_curs);
1841 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1842 if (!base) {
1844 * We're probably in deep shit, but let's try to fetch
1845 * the required base anyway from another pack or loose.
1846 * This is costly but should happen only in the presence
1847 * of a corrupted pack, and is better than failing outright.
1849 struct revindex_entry *revidx;
1850 const unsigned char *base_sha1;
1851 revidx = find_pack_revindex(p, base_offset);
1852 if (!revidx)
1853 return NULL;
1854 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1855 error("failed to read delta base object %s"
1856 " at offset %"PRIuMAX" from %s",
1857 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1858 p->pack_name);
1859 mark_bad_packed_object(p, base_sha1);
1860 base = read_object(base_sha1, type, &base_size);
1861 if (!base)
1862 return NULL;
1865 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1866 if (!delta_data) {
1867 error("failed to unpack compressed delta "
1868 "at offset %"PRIuMAX" from %s",
1869 (uintmax_t)curpos, p->pack_name);
1870 free(base);
1871 return NULL;
1873 result = patch_delta(base, base_size,
1874 delta_data, delta_size,
1875 sizep);
1876 if (!result)
1877 die("failed to apply delta");
1878 free(delta_data);
1879 add_delta_base_cache(p, base_offset, base, base_size, *type);
1880 return result;
1883 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1885 static FILE *log_file;
1887 if (!log_file) {
1888 log_file = fopen(log_pack_access, "w");
1889 if (!log_file) {
1890 error("cannot open pack access log '%s' for writing: %s",
1891 log_pack_access, strerror(errno));
1892 log_pack_access = NULL;
1893 return;
1896 fprintf(log_file, "%s %"PRIuMAX"\n",
1897 p->pack_name, (uintmax_t)obj_offset);
1898 fflush(log_file);
1901 int do_check_packed_object_crc;
1903 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1904 enum object_type *type, unsigned long *sizep)
1906 struct pack_window *w_curs = NULL;
1907 off_t curpos = obj_offset;
1908 void *data;
1910 if (log_pack_access)
1911 write_pack_access_log(p, obj_offset);
1913 if (do_check_packed_object_crc && p->index_version > 1) {
1914 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1915 unsigned long len = revidx[1].offset - obj_offset;
1916 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1917 const unsigned char *sha1 =
1918 nth_packed_object_sha1(p, revidx->nr);
1919 error("bad packed object CRC for %s",
1920 sha1_to_hex(sha1));
1921 mark_bad_packed_object(p, sha1);
1922 unuse_pack(&w_curs);
1923 return NULL;
1927 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1928 switch (*type) {
1929 case OBJ_OFS_DELTA:
1930 case OBJ_REF_DELTA:
1931 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1932 obj_offset, type, sizep);
1933 break;
1934 case OBJ_COMMIT:
1935 case OBJ_TREE:
1936 case OBJ_BLOB:
1937 case OBJ_TAG:
1938 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1939 break;
1940 default:
1941 data = NULL;
1942 error("unknown object type %i at offset %"PRIuMAX" in %s",
1943 *type, (uintmax_t)obj_offset, p->pack_name);
1945 unuse_pack(&w_curs);
1946 return data;
1949 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1950 uint32_t n)
1952 const unsigned char *index = p->index_data;
1953 if (!index) {
1954 if (open_pack_index(p))
1955 return NULL;
1956 index = p->index_data;
1958 if (n >= p->num_objects)
1959 return NULL;
1960 index += 4 * 256;
1961 if (p->index_version == 1) {
1962 return index + 24 * n + 4;
1963 } else {
1964 index += 8;
1965 return index + 20 * n;
1969 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1971 const unsigned char *index = p->index_data;
1972 index += 4 * 256;
1973 if (p->index_version == 1) {
1974 return ntohl(*((uint32_t *)(index + 24 * n)));
1975 } else {
1976 uint32_t off;
1977 index += 8 + p->num_objects * (20 + 4);
1978 off = ntohl(*((uint32_t *)(index + 4 * n)));
1979 if (!(off & 0x80000000))
1980 return off;
1981 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1982 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1983 ntohl(*((uint32_t *)(index + 4)));
1987 off_t find_pack_entry_one(const unsigned char *sha1,
1988 struct packed_git *p)
1990 const uint32_t *level1_ofs = p->index_data;
1991 const unsigned char *index = p->index_data;
1992 unsigned hi, lo, stride;
1993 static int use_lookup = -1;
1994 static int debug_lookup = -1;
1996 if (debug_lookup < 0)
1997 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1999 if (!index) {
2000 if (open_pack_index(p))
2001 return 0;
2002 level1_ofs = p->index_data;
2003 index = p->index_data;
2005 if (p->index_version > 1) {
2006 level1_ofs += 2;
2007 index += 8;
2009 index += 4 * 256;
2010 hi = ntohl(level1_ofs[*sha1]);
2011 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2012 if (p->index_version > 1) {
2013 stride = 20;
2014 } else {
2015 stride = 24;
2016 index += 4;
2019 if (debug_lookup)
2020 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2021 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2023 if (use_lookup < 0)
2024 use_lookup = !!getenv("GIT_USE_LOOKUP");
2025 if (use_lookup) {
2026 int pos = sha1_entry_pos(index, stride, 0,
2027 lo, hi, p->num_objects, sha1);
2028 if (pos < 0)
2029 return 0;
2030 return nth_packed_object_offset(p, pos);
2033 do {
2034 unsigned mi = (lo + hi) / 2;
2035 int cmp = hashcmp(index + mi * stride, sha1);
2037 if (debug_lookup)
2038 printf("lo %u hi %u rg %u mi %u\n",
2039 lo, hi, hi - lo, mi);
2040 if (!cmp)
2041 return nth_packed_object_offset(p, mi);
2042 if (cmp > 0)
2043 hi = mi;
2044 else
2045 lo = mi+1;
2046 } while (lo < hi);
2047 return 0;
2050 int is_pack_valid(struct packed_git *p)
2052 /* An already open pack is known to be valid. */
2053 if (p->pack_fd != -1)
2054 return 1;
2056 /* If the pack has one window completely covering the
2057 * file size, the pack is known to be valid even if
2058 * the descriptor is not currently open.
2060 if (p->windows) {
2061 struct pack_window *w = p->windows;
2063 if (!w->offset && w->len == p->pack_size)
2064 return 1;
2067 /* Force the pack to open to prove its valid. */
2068 return !open_packed_git(p);
2071 static int fill_pack_entry(const unsigned char *sha1,
2072 struct pack_entry *e,
2073 struct packed_git *p)
2075 off_t offset;
2077 if (p->num_bad_objects) {
2078 unsigned i;
2079 for (i = 0; i < p->num_bad_objects; i++)
2080 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2081 return 0;
2084 offset = find_pack_entry_one(sha1, p);
2085 if (!offset)
2086 return 0;
2089 * We are about to tell the caller where they can locate the
2090 * requested object. We better make sure the packfile is
2091 * still here and can be accessed before supplying that
2092 * answer, as it may have been deleted since the index was
2093 * loaded!
2095 if (!is_pack_valid(p)) {
2096 warning("packfile %s cannot be accessed", p->pack_name);
2097 return 0;
2099 e->offset = offset;
2100 e->p = p;
2101 hashcpy(e->sha1, sha1);
2102 return 1;
2105 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2107 struct packed_git *p;
2109 prepare_packed_git();
2110 if (!packed_git)
2111 return 0;
2113 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2114 return 1;
2116 for (p = packed_git; p; p = p->next) {
2117 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2118 continue;
2120 last_found_pack = p;
2121 return 1;
2123 return 0;
2126 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2127 struct packed_git *packs)
2129 struct packed_git *p;
2131 for (p = packs; p; p = p->next) {
2132 if (find_pack_entry_one(sha1, p))
2133 return p;
2135 return NULL;
2139 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2141 int status;
2142 unsigned long mapsize, size;
2143 void *map;
2144 git_zstream stream;
2145 char hdr[32];
2147 map = map_sha1_file(sha1, &mapsize);
2148 if (!map)
2149 return error("unable to find %s", sha1_to_hex(sha1));
2150 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2151 status = error("unable to unpack %s header",
2152 sha1_to_hex(sha1));
2153 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2154 status = error("unable to parse %s header", sha1_to_hex(sha1));
2155 else if (sizep)
2156 *sizep = size;
2157 git_inflate_end(&stream);
2158 munmap(map, mapsize);
2159 return status;
2162 /* returns enum object_type or negative */
2163 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2165 struct cached_object *co;
2166 struct pack_entry e;
2167 int status, rtype;
2169 co = find_cached_object(sha1);
2170 if (co) {
2171 if (oi->sizep)
2172 *(oi->sizep) = co->size;
2173 oi->whence = OI_CACHED;
2174 return co->type;
2177 if (!find_pack_entry(sha1, &e)) {
2178 /* Most likely it's a loose object. */
2179 status = sha1_loose_object_info(sha1, oi->sizep);
2180 if (status >= 0) {
2181 oi->whence = OI_LOOSE;
2182 return status;
2185 /* Not a loose object; someone else may have just packed it. */
2186 reprepare_packed_git();
2187 if (!find_pack_entry(sha1, &e))
2188 return status;
2191 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2192 if (status < 0) {
2193 mark_bad_packed_object(e.p, sha1);
2194 status = sha1_object_info_extended(sha1, oi);
2195 } else if (in_delta_base_cache(e.p, e.offset)) {
2196 oi->whence = OI_DBCACHED;
2197 } else {
2198 oi->whence = OI_PACKED;
2199 oi->u.packed.offset = e.offset;
2200 oi->u.packed.pack = e.p;
2201 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2202 rtype == OBJ_OFS_DELTA);
2205 return status;
2208 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2210 struct object_info oi;
2212 oi.sizep = sizep;
2213 return sha1_object_info_extended(sha1, &oi);
2216 static void *read_packed_sha1(const unsigned char *sha1,
2217 enum object_type *type, unsigned long *size)
2219 struct pack_entry e;
2220 void *data;
2222 if (!find_pack_entry(sha1, &e))
2223 return NULL;
2224 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2225 if (!data) {
2227 * We're probably in deep shit, but let's try to fetch
2228 * the required object anyway from another pack or loose.
2229 * This should happen only in the presence of a corrupted
2230 * pack, and is better than failing outright.
2232 error("failed to read object %s at offset %"PRIuMAX" from %s",
2233 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2234 mark_bad_packed_object(e.p, sha1);
2235 data = read_object(sha1, type, size);
2237 return data;
2240 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2241 unsigned char *sha1)
2243 struct cached_object *co;
2245 hash_sha1_file(buf, len, typename(type), sha1);
2246 if (has_sha1_file(sha1) || find_cached_object(sha1))
2247 return 0;
2248 if (cached_object_alloc <= cached_object_nr) {
2249 cached_object_alloc = alloc_nr(cached_object_alloc);
2250 cached_objects = xrealloc(cached_objects,
2251 sizeof(*cached_objects) *
2252 cached_object_alloc);
2254 co = &cached_objects[cached_object_nr++];
2255 co->size = len;
2256 co->type = type;
2257 co->buf = xmalloc(len);
2258 memcpy(co->buf, buf, len);
2259 hashcpy(co->sha1, sha1);
2260 return 0;
2263 static void *read_object(const unsigned char *sha1, enum object_type *type,
2264 unsigned long *size)
2266 unsigned long mapsize;
2267 void *map, *buf;
2268 struct cached_object *co;
2270 co = find_cached_object(sha1);
2271 if (co) {
2272 *type = co->type;
2273 *size = co->size;
2274 return xmemdupz(co->buf, co->size);
2277 buf = read_packed_sha1(sha1, type, size);
2278 if (buf)
2279 return buf;
2280 map = map_sha1_file(sha1, &mapsize);
2281 if (map) {
2282 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2283 munmap(map, mapsize);
2284 return buf;
2286 reprepare_packed_git();
2287 return read_packed_sha1(sha1, type, size);
2291 * This function dies on corrupt objects; the callers who want to
2292 * deal with them should arrange to call read_object() and give error
2293 * messages themselves.
2295 void *read_sha1_file_extended(const unsigned char *sha1,
2296 enum object_type *type,
2297 unsigned long *size,
2298 unsigned flag)
2300 void *data;
2301 char *path;
2302 const struct packed_git *p;
2303 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2304 ? lookup_replace_object(sha1) : sha1;
2306 errno = 0;
2307 data = read_object(repl, type, size);
2308 if (data)
2309 return data;
2311 if (errno && errno != ENOENT)
2312 die_errno("failed to read object %s", sha1_to_hex(sha1));
2314 /* die if we replaced an object with one that does not exist */
2315 if (repl != sha1)
2316 die("replacement %s not found for %s",
2317 sha1_to_hex(repl), sha1_to_hex(sha1));
2319 if (has_loose_object(repl)) {
2320 path = sha1_file_name(sha1);
2321 die("loose object %s (stored in %s) is corrupt",
2322 sha1_to_hex(repl), path);
2325 if ((p = has_packed_and_bad(repl)) != NULL)
2326 die("packed object %s (stored in %s) is corrupt",
2327 sha1_to_hex(repl), p->pack_name);
2329 return NULL;
2332 void *read_object_with_reference(const unsigned char *sha1,
2333 const char *required_type_name,
2334 unsigned long *size,
2335 unsigned char *actual_sha1_return)
2337 enum object_type type, required_type;
2338 void *buffer;
2339 unsigned long isize;
2340 unsigned char actual_sha1[20];
2342 required_type = type_from_string(required_type_name);
2343 hashcpy(actual_sha1, sha1);
2344 while (1) {
2345 int ref_length = -1;
2346 const char *ref_type = NULL;
2348 buffer = read_sha1_file(actual_sha1, &type, &isize);
2349 if (!buffer)
2350 return NULL;
2351 if (type == required_type) {
2352 *size = isize;
2353 if (actual_sha1_return)
2354 hashcpy(actual_sha1_return, actual_sha1);
2355 return buffer;
2357 /* Handle references */
2358 else if (type == OBJ_COMMIT)
2359 ref_type = "tree ";
2360 else if (type == OBJ_TAG)
2361 ref_type = "object ";
2362 else {
2363 free(buffer);
2364 return NULL;
2366 ref_length = strlen(ref_type);
2368 if (ref_length + 40 > isize ||
2369 memcmp(buffer, ref_type, ref_length) ||
2370 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2371 free(buffer);
2372 return NULL;
2374 free(buffer);
2375 /* Now we have the ID of the referred-to object in
2376 * actual_sha1. Check again. */
2380 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2381 const char *type, unsigned char *sha1,
2382 char *hdr, int *hdrlen)
2384 git_SHA_CTX c;
2386 /* Generate the header */
2387 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2389 /* Sha1.. */
2390 git_SHA1_Init(&c);
2391 git_SHA1_Update(&c, hdr, *hdrlen);
2392 git_SHA1_Update(&c, buf, len);
2393 git_SHA1_Final(sha1, &c);
2397 * Move the just written object into its final resting place.
2398 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2399 * "moving" is only a part of what it does, when no patch between
2400 * master to pu changes the call sites of this function.
2402 int move_temp_to_file(const char *tmpfile, const char *filename)
2404 int ret = 0;
2406 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2407 goto try_rename;
2408 else if (link(tmpfile, filename))
2409 ret = errno;
2412 * Coda hack - coda doesn't like cross-directory links,
2413 * so we fall back to a rename, which will mean that it
2414 * won't be able to check collisions, but that's not a
2415 * big deal.
2417 * The same holds for FAT formatted media.
2419 * When this succeeds, we just return. We have nothing
2420 * left to unlink.
2422 if (ret && ret != EEXIST) {
2423 try_rename:
2424 if (!rename(tmpfile, filename))
2425 goto out;
2426 ret = errno;
2428 unlink_or_warn(tmpfile);
2429 if (ret) {
2430 if (ret != EEXIST) {
2431 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2433 /* FIXME!!! Collision check here ? */
2436 out:
2437 if (adjust_shared_perm(filename))
2438 return error("unable to set permission to '%s'", filename);
2439 return 0;
2442 static int write_buffer(int fd, const void *buf, size_t len)
2444 if (write_in_full(fd, buf, len) < 0)
2445 return error("file write error (%s)", strerror(errno));
2446 return 0;
2449 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2450 unsigned char *sha1)
2452 char hdr[32];
2453 int hdrlen;
2454 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2455 return 0;
2458 /* Finalize a file on disk, and close it. */
2459 static void close_sha1_file(int fd)
2461 if (fsync_object_files)
2462 fsync_or_die(fd, "sha1 file");
2463 if (close(fd) != 0)
2464 die_errno("error when closing sha1 file");
2467 /* Size of directory component, including the ending '/' */
2468 static inline int directory_size(const char *filename)
2470 const char *s = strrchr(filename, '/');
2471 if (!s)
2472 return 0;
2473 return s - filename + 1;
2477 * This creates a temporary file in the same directory as the final
2478 * 'filename'
2480 * We want to avoid cross-directory filename renames, because those
2481 * can have problems on various filesystems (FAT, NFS, Coda).
2483 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2485 int fd, dirlen = directory_size(filename);
2487 if (dirlen + 20 > bufsiz) {
2488 errno = ENAMETOOLONG;
2489 return -1;
2491 memcpy(buffer, filename, dirlen);
2492 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2493 fd = git_mkstemp_mode(buffer, 0444);
2494 if (fd < 0 && dirlen && errno == ENOENT) {
2495 /* Make sure the directory exists */
2496 memcpy(buffer, filename, dirlen);
2497 buffer[dirlen-1] = 0;
2498 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2499 return -1;
2501 /* Try again */
2502 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2503 fd = git_mkstemp_mode(buffer, 0444);
2505 return fd;
2508 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2509 const void *buf, unsigned long len, time_t mtime)
2511 int fd, ret;
2512 unsigned char compressed[4096];
2513 git_zstream stream;
2514 git_SHA_CTX c;
2515 unsigned char parano_sha1[20];
2516 char *filename;
2517 static char tmp_file[PATH_MAX];
2519 filename = sha1_file_name(sha1);
2520 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2521 if (fd < 0) {
2522 if (errno == EACCES)
2523 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2524 else
2525 return error("unable to create temporary file: %s", strerror(errno));
2528 /* Set it up */
2529 memset(&stream, 0, sizeof(stream));
2530 git_deflate_init(&stream, zlib_compression_level);
2531 stream.next_out = compressed;
2532 stream.avail_out = sizeof(compressed);
2533 git_SHA1_Init(&c);
2535 /* First header.. */
2536 stream.next_in = (unsigned char *)hdr;
2537 stream.avail_in = hdrlen;
2538 while (git_deflate(&stream, 0) == Z_OK)
2539 ; /* nothing */
2540 git_SHA1_Update(&c, hdr, hdrlen);
2542 /* Then the data itself.. */
2543 stream.next_in = (void *)buf;
2544 stream.avail_in = len;
2545 do {
2546 unsigned char *in0 = stream.next_in;
2547 ret = git_deflate(&stream, Z_FINISH);
2548 git_SHA1_Update(&c, in0, stream.next_in - in0);
2549 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2550 die("unable to write sha1 file");
2551 stream.next_out = compressed;
2552 stream.avail_out = sizeof(compressed);
2553 } while (ret == Z_OK);
2555 if (ret != Z_STREAM_END)
2556 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2557 ret = git_deflate_end_gently(&stream);
2558 if (ret != Z_OK)
2559 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2560 git_SHA1_Final(parano_sha1, &c);
2561 if (hashcmp(sha1, parano_sha1) != 0)
2562 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2564 close_sha1_file(fd);
2566 if (mtime) {
2567 struct utimbuf utb;
2568 utb.actime = mtime;
2569 utb.modtime = mtime;
2570 if (utime(tmp_file, &utb) < 0)
2571 warning("failed utime() on %s: %s",
2572 tmp_file, strerror(errno));
2575 return move_temp_to_file(tmp_file, filename);
2578 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2580 unsigned char sha1[20];
2581 char hdr[32];
2582 int hdrlen;
2584 /* Normally if we have it in the pack then we do not bother writing
2585 * it out into .git/objects/??/?{38} file.
2587 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2588 if (returnsha1)
2589 hashcpy(returnsha1, sha1);
2590 if (has_sha1_file(sha1))
2591 return 0;
2592 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2595 int force_object_loose(const unsigned char *sha1, time_t mtime)
2597 void *buf;
2598 unsigned long len;
2599 enum object_type type;
2600 char hdr[32];
2601 int hdrlen;
2602 int ret;
2604 if (has_loose_object(sha1))
2605 return 0;
2606 buf = read_packed_sha1(sha1, &type, &len);
2607 if (!buf)
2608 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2609 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2610 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2611 free(buf);
2613 return ret;
2616 int has_pack_index(const unsigned char *sha1)
2618 struct stat st;
2619 if (stat(sha1_pack_index_name(sha1), &st))
2620 return 0;
2621 return 1;
2624 int has_sha1_pack(const unsigned char *sha1)
2626 struct pack_entry e;
2627 return find_pack_entry(sha1, &e);
2630 int has_sha1_file(const unsigned char *sha1)
2632 struct pack_entry e;
2634 if (find_pack_entry(sha1, &e))
2635 return 1;
2636 return has_loose_object(sha1);
2639 static void check_tree(const void *buf, size_t size)
2641 struct tree_desc desc;
2642 struct name_entry entry;
2644 init_tree_desc(&desc, buf, size);
2645 while (tree_entry(&desc, &entry))
2646 /* do nothing
2647 * tree_entry() will die() on malformed entries */
2651 static void check_commit(const void *buf, size_t size)
2653 struct commit c;
2654 memset(&c, 0, sizeof(c));
2655 if (parse_commit_buffer(&c, buf, size))
2656 die("corrupt commit");
2659 static void check_tag(const void *buf, size_t size)
2661 struct tag t;
2662 memset(&t, 0, sizeof(t));
2663 if (parse_tag_buffer(&t, buf, size))
2664 die("corrupt tag");
2667 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2668 enum object_type type,
2669 const char *path, unsigned flags)
2671 int ret, re_allocated = 0;
2672 int write_object = flags & HASH_WRITE_OBJECT;
2674 if (!type)
2675 type = OBJ_BLOB;
2678 * Convert blobs to git internal format
2680 if ((type == OBJ_BLOB) && path) {
2681 struct strbuf nbuf = STRBUF_INIT;
2682 if (convert_to_git(path, buf, size, &nbuf,
2683 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2684 buf = strbuf_detach(&nbuf, &size);
2685 re_allocated = 1;
2688 if (flags & HASH_FORMAT_CHECK) {
2689 if (type == OBJ_TREE)
2690 check_tree(buf, size);
2691 if (type == OBJ_COMMIT)
2692 check_commit(buf, size);
2693 if (type == OBJ_TAG)
2694 check_tag(buf, size);
2697 if (write_object)
2698 ret = write_sha1_file(buf, size, typename(type), sha1);
2699 else
2700 ret = hash_sha1_file(buf, size, typename(type), sha1);
2701 if (re_allocated)
2702 free(buf);
2703 return ret;
2706 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2707 const char *path, unsigned flags)
2709 struct strbuf sbuf = STRBUF_INIT;
2710 int ret;
2712 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2713 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2714 else
2715 ret = -1;
2716 strbuf_release(&sbuf);
2717 return ret;
2720 #define SMALL_FILE_SIZE (32*1024)
2722 static int index_core(unsigned char *sha1, int fd, size_t size,
2723 enum object_type type, const char *path,
2724 unsigned flags)
2726 int ret;
2728 if (!size) {
2729 ret = index_mem(sha1, NULL, size, type, path, flags);
2730 } else if (size <= SMALL_FILE_SIZE) {
2731 char *buf = xmalloc(size);
2732 if (size == read_in_full(fd, buf, size))
2733 ret = index_mem(sha1, buf, size, type, path, flags);
2734 else
2735 ret = error("short read %s", strerror(errno));
2736 free(buf);
2737 } else {
2738 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2739 ret = index_mem(sha1, buf, size, type, path, flags);
2740 munmap(buf, size);
2742 return ret;
2746 * This creates one packfile per large blob unless bulk-checkin
2747 * machinery is "plugged".
2749 * This also bypasses the usual "convert-to-git" dance, and that is on
2750 * purpose. We could write a streaming version of the converting
2751 * functions and insert that before feeding the data to fast-import
2752 * (or equivalent in-core API described above). However, that is
2753 * somewhat complicated, as we do not know the size of the filter
2754 * result, which we need to know beforehand when writing a git object.
2755 * Since the primary motivation for trying to stream from the working
2756 * tree file and to avoid mmaping it in core is to deal with large
2757 * binary blobs, they generally do not want to get any conversion, and
2758 * callers should avoid this code path when filters are requested.
2760 static int index_stream(unsigned char *sha1, int fd, size_t size,
2761 enum object_type type, const char *path,
2762 unsigned flags)
2764 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2767 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2768 enum object_type type, const char *path, unsigned flags)
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 (path && would_convert_to_git(path, NULL, 0, 0)))
2777 ret = index_core(sha1, fd, size, type, path, flags);
2778 else
2779 ret = index_stream(sha1, fd, size, type, path, flags);
2780 close(fd);
2781 return ret;
2784 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2786 int fd;
2787 struct strbuf sb = STRBUF_INIT;
2789 switch (st->st_mode & S_IFMT) {
2790 case S_IFREG:
2791 fd = open(path, O_RDONLY);
2792 if (fd < 0)
2793 return error("open(\"%s\"): %s", path,
2794 strerror(errno));
2795 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2796 return error("%s: failed to insert into database",
2797 path);
2798 break;
2799 case S_IFLNK:
2800 if (strbuf_readlink(&sb, path, st->st_size)) {
2801 char *errstr = strerror(errno);
2802 return error("readlink(\"%s\"): %s", path,
2803 errstr);
2805 if (!(flags & HASH_WRITE_OBJECT))
2806 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2807 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2808 return error("%s: failed to insert into database",
2809 path);
2810 strbuf_release(&sb);
2811 break;
2812 case S_IFDIR:
2813 return resolve_gitlink_ref(path, "HEAD", sha1);
2814 default:
2815 return error("%s: unsupported file type", path);
2817 return 0;
2820 int read_pack_header(int fd, struct pack_header *header)
2822 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2823 /* "eof before pack header was fully read" */
2824 return PH_ERROR_EOF;
2826 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2827 /* "protocol error (pack signature mismatch detected)" */
2828 return PH_ERROR_PACK_SIGNATURE;
2829 if (!pack_version_ok(header->hdr_version))
2830 /* "protocol error (pack version unsupported)" */
2831 return PH_ERROR_PROTOCOL;
2832 return 0;
2835 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2837 enum object_type type = sha1_object_info(sha1, NULL);
2838 if (type < 0)
2839 die("%s is not a valid object", sha1_to_hex(sha1));
2840 if (type != expect)
2841 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2842 typename(expect));