i18n: commit: mark parseopt strings for translation
[git/jnareb-git.git] / sha1_file.c
blobaf5cfbde63aed41d95ac7318d064d3012832d2c5
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;
735 * Do not call this directly as this leaks p->pack_fd on error return;
736 * call open_packed_git() instead.
738 static int open_packed_git_1(struct packed_git *p)
740 struct stat st;
741 struct pack_header hdr;
742 unsigned char sha1[20];
743 unsigned char *idx_sha1;
744 long fd_flag;
746 if (!p->index_data && open_pack_index(p))
747 return error("packfile %s index unavailable", p->pack_name);
749 if (!pack_max_fds) {
750 struct rlimit lim;
751 unsigned int max_fds;
753 if (getrlimit(RLIMIT_NOFILE, &lim))
754 die_errno("cannot get RLIMIT_NOFILE");
756 max_fds = lim.rlim_cur;
758 /* Save 3 for stdin/stdout/stderr, 22 for work */
759 if (25 < max_fds)
760 pack_max_fds = max_fds - 25;
761 else
762 pack_max_fds = 1;
765 while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
766 ; /* nothing */
768 p->pack_fd = git_open_noatime(p->pack_name);
769 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
770 return -1;
771 pack_open_fds++;
773 /* If we created the struct before we had the pack we lack size. */
774 if (!p->pack_size) {
775 if (!S_ISREG(st.st_mode))
776 return error("packfile %s not a regular file", p->pack_name);
777 p->pack_size = st.st_size;
778 } else if (p->pack_size != st.st_size)
779 return error("packfile %s size changed", p->pack_name);
781 /* We leave these file descriptors open with sliding mmap;
782 * there is no point keeping them open across exec(), though.
784 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
785 if (fd_flag < 0)
786 return error("cannot determine file descriptor flags");
787 fd_flag |= FD_CLOEXEC;
788 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
789 return error("cannot set FD_CLOEXEC");
791 /* Verify we recognize this pack file format. */
792 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
793 return error("file %s is far too short to be a packfile", p->pack_name);
794 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
795 return error("file %s is not a GIT packfile", p->pack_name);
796 if (!pack_version_ok(hdr.hdr_version))
797 return error("packfile %s is version %"PRIu32" and not"
798 " supported (try upgrading GIT to a newer version)",
799 p->pack_name, ntohl(hdr.hdr_version));
801 /* Verify the pack matches its index. */
802 if (p->num_objects != ntohl(hdr.hdr_entries))
803 return error("packfile %s claims to have %"PRIu32" objects"
804 " while index indicates %"PRIu32" objects",
805 p->pack_name, ntohl(hdr.hdr_entries),
806 p->num_objects);
807 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
808 return error("end of packfile %s is unavailable", p->pack_name);
809 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
810 return error("packfile %s signature is unavailable", p->pack_name);
811 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
812 if (hashcmp(sha1, idx_sha1))
813 return error("packfile %s does not match index", p->pack_name);
814 return 0;
817 static int open_packed_git(struct packed_git *p)
819 if (!open_packed_git_1(p))
820 return 0;
821 if (p->pack_fd != -1) {
822 close(p->pack_fd);
823 pack_open_fds--;
824 p->pack_fd = -1;
826 return -1;
829 static int in_window(struct pack_window *win, off_t offset)
831 /* We must promise at least 20 bytes (one hash) after the
832 * offset is available from this window, otherwise the offset
833 * is not actually in this window and a different window (which
834 * has that one hash excess) must be used. This is to support
835 * the object header and delta base parsing routines below.
837 off_t win_off = win->offset;
838 return win_off <= offset
839 && (offset + 20) <= (win_off + win->len);
842 unsigned char *use_pack(struct packed_git *p,
843 struct pack_window **w_cursor,
844 off_t offset,
845 unsigned long *left)
847 struct pack_window *win = *w_cursor;
849 /* Since packfiles end in a hash of their content and it's
850 * pointless to ask for an offset into the middle of that
851 * hash, and the in_window function above wouldn't match
852 * don't allow an offset too close to the end of the file.
854 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
855 die("packfile %s cannot be accessed", p->pack_name);
856 if (offset > (p->pack_size - 20))
857 die("offset beyond end of packfile (truncated pack?)");
859 if (!win || !in_window(win, offset)) {
860 if (win)
861 win->inuse_cnt--;
862 for (win = p->windows; win; win = win->next) {
863 if (in_window(win, offset))
864 break;
866 if (!win) {
867 size_t window_align = packed_git_window_size / 2;
868 off_t len;
870 if (p->pack_fd == -1 && open_packed_git(p))
871 die("packfile %s cannot be accessed", p->pack_name);
873 win = xcalloc(1, sizeof(*win));
874 win->offset = (offset / window_align) * window_align;
875 len = p->pack_size - win->offset;
876 if (len > packed_git_window_size)
877 len = packed_git_window_size;
878 win->len = (size_t)len;
879 pack_mapped += win->len;
880 while (packed_git_limit < pack_mapped
881 && unuse_one_window(p, p->pack_fd))
882 ; /* nothing */
883 win->base = xmmap(NULL, win->len,
884 PROT_READ, MAP_PRIVATE,
885 p->pack_fd, win->offset);
886 if (win->base == MAP_FAILED)
887 die("packfile %s cannot be mapped: %s",
888 p->pack_name,
889 strerror(errno));
890 if (!win->offset && win->len == p->pack_size
891 && !p->do_not_close) {
892 close(p->pack_fd);
893 pack_open_fds--;
894 p->pack_fd = -1;
896 pack_mmap_calls++;
897 pack_open_windows++;
898 if (pack_mapped > peak_pack_mapped)
899 peak_pack_mapped = pack_mapped;
900 if (pack_open_windows > peak_pack_open_windows)
901 peak_pack_open_windows = pack_open_windows;
902 win->next = p->windows;
903 p->windows = win;
906 if (win != *w_cursor) {
907 win->last_used = pack_used_ctr++;
908 win->inuse_cnt++;
909 *w_cursor = win;
911 offset -= win->offset;
912 if (left)
913 *left = win->len - xsize_t(offset);
914 return win->base + offset;
917 static struct packed_git *alloc_packed_git(int extra)
919 struct packed_git *p = xmalloc(sizeof(*p) + extra);
920 memset(p, 0, sizeof(*p));
921 p->pack_fd = -1;
922 return p;
925 static void try_to_free_pack_memory(size_t size)
927 release_pack_memory(size, -1);
930 struct packed_git *add_packed_git(const char *path, int path_len, int local)
932 static int have_set_try_to_free_routine;
933 struct stat st;
934 struct packed_git *p = alloc_packed_git(path_len + 2);
936 if (!have_set_try_to_free_routine) {
937 have_set_try_to_free_routine = 1;
938 set_try_to_free_routine(try_to_free_pack_memory);
942 * Make sure a corresponding .pack file exists and that
943 * the index looks sane.
945 path_len -= strlen(".idx");
946 if (path_len < 1) {
947 free(p);
948 return NULL;
950 memcpy(p->pack_name, path, path_len);
952 strcpy(p->pack_name + path_len, ".keep");
953 if (!access(p->pack_name, F_OK))
954 p->pack_keep = 1;
956 strcpy(p->pack_name + path_len, ".pack");
957 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
958 free(p);
959 return NULL;
962 /* ok, it looks sane as far as we can check without
963 * actually mapping the pack file.
965 p->pack_size = st.st_size;
966 p->pack_local = local;
967 p->mtime = st.st_mtime;
968 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
969 hashclr(p->sha1);
970 return p;
973 struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
975 const char *path = sha1_pack_name(sha1);
976 struct packed_git *p = alloc_packed_git(strlen(path) + 1);
978 strcpy(p->pack_name, path);
979 hashcpy(p->sha1, sha1);
980 if (check_packed_git_idx(idx_path, p)) {
981 free(p);
982 return NULL;
985 return p;
988 void install_packed_git(struct packed_git *pack)
990 if (pack->pack_fd != -1)
991 pack_open_fds++;
993 pack->next = packed_git;
994 packed_git = pack;
997 static void prepare_packed_git_one(char *objdir, int local)
999 /* Ensure that this buffer is large enough so that we can
1000 append "/pack/" without clobbering the stack even if
1001 strlen(objdir) were PATH_MAX. */
1002 char path[PATH_MAX + 1 + 4 + 1 + 1];
1003 int len;
1004 DIR *dir;
1005 struct dirent *de;
1007 sprintf(path, "%s/pack", objdir);
1008 len = strlen(path);
1009 dir = opendir(path);
1010 if (!dir) {
1011 if (errno != ENOENT)
1012 error("unable to open object pack directory: %s: %s",
1013 path, strerror(errno));
1014 return;
1016 path[len++] = '/';
1017 while ((de = readdir(dir)) != NULL) {
1018 int namelen = strlen(de->d_name);
1019 struct packed_git *p;
1021 if (!has_extension(de->d_name, ".idx"))
1022 continue;
1024 if (len + namelen + 1 > sizeof(path))
1025 continue;
1027 /* Don't reopen a pack we already have. */
1028 strcpy(path + len, de->d_name);
1029 for (p = packed_git; p; p = p->next) {
1030 if (!memcmp(path, p->pack_name, len + namelen - 4))
1031 break;
1033 if (p)
1034 continue;
1035 /* See if it really is a valid .idx file with corresponding
1036 * .pack file that we can map.
1038 p = add_packed_git(path, len + namelen, local);
1039 if (!p)
1040 continue;
1041 install_packed_git(p);
1043 closedir(dir);
1046 static int sort_pack(const void *a_, const void *b_)
1048 struct packed_git *a = *((struct packed_git **)a_);
1049 struct packed_git *b = *((struct packed_git **)b_);
1050 int st;
1053 * Local packs tend to contain objects specific to our
1054 * variant of the project than remote ones. In addition,
1055 * remote ones could be on a network mounted filesystem.
1056 * Favor local ones for these reasons.
1058 st = a->pack_local - b->pack_local;
1059 if (st)
1060 return -st;
1063 * Younger packs tend to contain more recent objects,
1064 * and more recent objects tend to get accessed more
1065 * often.
1067 if (a->mtime < b->mtime)
1068 return 1;
1069 else if (a->mtime == b->mtime)
1070 return 0;
1071 return -1;
1074 static void rearrange_packed_git(void)
1076 struct packed_git **ary, *p;
1077 int i, n;
1079 for (n = 0, p = packed_git; p; p = p->next)
1080 n++;
1081 if (n < 2)
1082 return;
1084 /* prepare an array of packed_git for easier sorting */
1085 ary = xcalloc(n, sizeof(struct packed_git *));
1086 for (n = 0, p = packed_git; p; p = p->next)
1087 ary[n++] = p;
1089 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1091 /* link them back again */
1092 for (i = 0; i < n - 1; i++)
1093 ary[i]->next = ary[i + 1];
1094 ary[n - 1]->next = NULL;
1095 packed_git = ary[0];
1097 free(ary);
1100 static int prepare_packed_git_run_once = 0;
1101 void prepare_packed_git(void)
1103 struct alternate_object_database *alt;
1105 if (prepare_packed_git_run_once)
1106 return;
1107 prepare_packed_git_one(get_object_directory(), 1);
1108 prepare_alt_odb();
1109 for (alt = alt_odb_list; alt; alt = alt->next) {
1110 alt->name[-1] = 0;
1111 prepare_packed_git_one(alt->base, 0);
1112 alt->name[-1] = '/';
1114 rearrange_packed_git();
1115 prepare_packed_git_run_once = 1;
1118 void reprepare_packed_git(void)
1120 discard_revindex();
1121 prepare_packed_git_run_once = 0;
1122 prepare_packed_git();
1125 static void mark_bad_packed_object(struct packed_git *p,
1126 const unsigned char *sha1)
1128 unsigned i;
1129 for (i = 0; i < p->num_bad_objects; i++)
1130 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1131 return;
1132 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1133 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1134 p->num_bad_objects++;
1137 static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
1139 struct packed_git *p;
1140 unsigned i;
1142 for (p = packed_git; p; p = p->next)
1143 for (i = 0; i < p->num_bad_objects; i++)
1144 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1145 return p;
1146 return NULL;
1150 * With an in-core object data in "map", rehash it to make sure the
1151 * object name actually matches "sha1" to detect object corruption.
1152 * With "map" == NULL, try reading the object named with "sha1" using
1153 * the streaming interface and rehash it to do the same.
1155 int check_sha1_signature(const unsigned char *sha1, void *map,
1156 unsigned long size, const char *type)
1158 unsigned char real_sha1[20];
1159 enum object_type obj_type;
1160 struct git_istream *st;
1161 git_SHA_CTX c;
1162 char hdr[32];
1163 int hdrlen;
1165 if (map) {
1166 hash_sha1_file(map, size, type, real_sha1);
1167 return hashcmp(sha1, real_sha1) ? -1 : 0;
1170 st = open_istream(sha1, &obj_type, &size, NULL);
1171 if (!st)
1172 return -1;
1174 /* Generate the header */
1175 hdrlen = sprintf(hdr, "%s %lu", typename(obj_type), size) + 1;
1177 /* Sha1.. */
1178 git_SHA1_Init(&c);
1179 git_SHA1_Update(&c, hdr, hdrlen);
1180 for (;;) {
1181 char buf[1024 * 16];
1182 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1184 if (!readlen)
1185 break;
1186 git_SHA1_Update(&c, buf, readlen);
1188 git_SHA1_Final(real_sha1, &c);
1189 close_istream(st);
1190 return hashcmp(sha1, real_sha1) ? -1 : 0;
1193 static int git_open_noatime(const char *name)
1195 static int sha1_file_open_flag = O_NOATIME;
1197 for (;;) {
1198 int fd = open(name, O_RDONLY | sha1_file_open_flag);
1199 if (fd >= 0)
1200 return fd;
1202 /* Might the failure be due to O_NOATIME? */
1203 if (errno != ENOENT && sha1_file_open_flag) {
1204 sha1_file_open_flag = 0;
1205 continue;
1208 return -1;
1212 static int open_sha1_file(const unsigned char *sha1)
1214 int fd;
1215 char *name = sha1_file_name(sha1);
1216 struct alternate_object_database *alt;
1218 fd = git_open_noatime(name);
1219 if (fd >= 0)
1220 return fd;
1222 prepare_alt_odb();
1223 errno = ENOENT;
1224 for (alt = alt_odb_list; alt; alt = alt->next) {
1225 name = alt->name;
1226 fill_sha1_path(name, sha1);
1227 fd = git_open_noatime(alt->base);
1228 if (fd >= 0)
1229 return fd;
1231 return -1;
1234 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
1236 void *map;
1237 int fd;
1239 fd = open_sha1_file(sha1);
1240 map = NULL;
1241 if (fd >= 0) {
1242 struct stat st;
1244 if (!fstat(fd, &st)) {
1245 *size = xsize_t(st.st_size);
1246 if (!*size) {
1247 /* mmap() is forbidden on empty files */
1248 error("object file %s is empty", sha1_file_name(sha1));
1249 return NULL;
1251 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
1253 close(fd);
1255 return map;
1259 * There used to be a second loose object header format which
1260 * was meant to mimic the in-pack format, allowing for direct
1261 * copy of the object data. This format turned up not to be
1262 * really worth it and we no longer write loose objects in that
1263 * format.
1265 static int experimental_loose_object(unsigned char *map)
1267 unsigned int word;
1270 * We must determine if the buffer contains the standard
1271 * zlib-deflated stream or the experimental format based
1272 * on the in-pack object format. Compare the header byte
1273 * for each format:
1275 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1276 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1278 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1279 * in standard loose-object format, UNLESS it is a Git-pack
1280 * format object *exactly* 8 bytes in size when inflated.
1282 * However, RFC1950 also specifies that the 1st 16-bit word
1283 * must be divisible by 31 - this checksum tells us our buffer
1284 * is in the standard format, giving a false positive only if
1285 * the 1st word of the Git-pack format object happens to be
1286 * divisible by 31, ie:
1287 * ((byte0 * 256) + byte1) % 31 = 0
1288 * => 0ttt10000www1000 % 31 = 0
1290 * As it happens, this case can only arise for www=3 & ttt=1
1291 * - ie, a Commit object, which would have to be 8 bytes in
1292 * size. As no Commit can be that small, we find that the
1293 * combination of these two criteria (bitmask & checksum)
1294 * can always correctly determine the buffer format.
1296 word = (map[0] << 8) + map[1];
1297 if ((map[0] & 0x8F) == 0x08 && !(word % 31))
1298 return 0;
1299 else
1300 return 1;
1303 unsigned long unpack_object_header_buffer(const unsigned char *buf,
1304 unsigned long len, enum object_type *type, unsigned long *sizep)
1306 unsigned shift;
1307 unsigned long size, c;
1308 unsigned long used = 0;
1310 c = buf[used++];
1311 *type = (c >> 4) & 7;
1312 size = c & 15;
1313 shift = 4;
1314 while (c & 0x80) {
1315 if (len <= used || bitsizeof(long) <= shift) {
1316 error("bad object header");
1317 size = used = 0;
1318 break;
1320 c = buf[used++];
1321 size += (c & 0x7f) << shift;
1322 shift += 7;
1324 *sizep = size;
1325 return used;
1328 int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
1330 unsigned long size, used;
1331 static const char valid_loose_object_type[8] = {
1332 0, /* OBJ_EXT */
1333 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1334 0, /* "delta" and others are invalid in a loose object */
1336 enum object_type type;
1338 /* Get the data stream */
1339 memset(stream, 0, sizeof(*stream));
1340 stream->next_in = map;
1341 stream->avail_in = mapsize;
1342 stream->next_out = buffer;
1343 stream->avail_out = bufsiz;
1345 if (experimental_loose_object(map)) {
1347 * The old experimental format we no longer produce;
1348 * we can still read it.
1350 used = unpack_object_header_buffer(map, mapsize, &type, &size);
1351 if (!used || !valid_loose_object_type[type])
1352 return -1;
1353 map += used;
1354 mapsize -= used;
1356 /* Set up the stream for the rest.. */
1357 stream->next_in = map;
1358 stream->avail_in = mapsize;
1359 git_inflate_init(stream);
1361 /* And generate the fake traditional header */
1362 stream->total_out = 1 + snprintf(buffer, bufsiz, "%s %lu",
1363 typename(type), size);
1364 return 0;
1366 git_inflate_init(stream);
1367 return git_inflate(stream, 0);
1370 static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
1372 int bytes = strlen(buffer) + 1;
1373 unsigned char *buf = xmallocz(size);
1374 unsigned long n;
1375 int status = Z_OK;
1377 n = stream->total_out - bytes;
1378 if (n > size)
1379 n = size;
1380 memcpy(buf, (char *) buffer + bytes, n);
1381 bytes = n;
1382 if (bytes <= size) {
1384 * The above condition must be (bytes <= size), not
1385 * (bytes < size). In other words, even though we
1386 * expect no more output and set avail_out to zero,
1387 * the input zlib stream may have bytes that express
1388 * "this concludes the stream", and we *do* want to
1389 * eat that input.
1391 * Otherwise we would not be able to test that we
1392 * consumed all the input to reach the expected size;
1393 * we also want to check that zlib tells us that all
1394 * went well with status == Z_STREAM_END at the end.
1396 stream->next_out = buf + bytes;
1397 stream->avail_out = size - bytes;
1398 while (status == Z_OK)
1399 status = git_inflate(stream, Z_FINISH);
1401 if (status == Z_STREAM_END && !stream->avail_in) {
1402 git_inflate_end(stream);
1403 return buf;
1406 if (status < 0)
1407 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1408 else if (stream->avail_in)
1409 error("garbage at end of loose object '%s'",
1410 sha1_to_hex(sha1));
1411 free(buf);
1412 return NULL;
1416 * We used to just use "sscanf()", but that's actually way
1417 * too permissive for what we want to check. So do an anal
1418 * object header parse by hand.
1420 int parse_sha1_header(const char *hdr, unsigned long *sizep)
1422 char type[10];
1423 int i;
1424 unsigned long size;
1427 * The type can be at most ten bytes (including the
1428 * terminating '\0' that we add), and is followed by
1429 * a space.
1431 i = 0;
1432 for (;;) {
1433 char c = *hdr++;
1434 if (c == ' ')
1435 break;
1436 type[i++] = c;
1437 if (i >= sizeof(type))
1438 return -1;
1440 type[i] = 0;
1443 * The length must follow immediately, and be in canonical
1444 * decimal format (ie "010" is not valid).
1446 size = *hdr++ - '0';
1447 if (size > 9)
1448 return -1;
1449 if (size) {
1450 for (;;) {
1451 unsigned long c = *hdr - '0';
1452 if (c > 9)
1453 break;
1454 hdr++;
1455 size = size * 10 + c;
1458 *sizep = size;
1461 * The length must be followed by a zero byte
1463 return *hdr ? -1 : type_from_string(type);
1466 static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
1468 int ret;
1469 git_zstream stream;
1470 char hdr[8192];
1472 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
1473 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
1474 return NULL;
1476 return unpack_sha1_rest(&stream, hdr, *size, sha1);
1479 unsigned long get_size_from_delta(struct packed_git *p,
1480 struct pack_window **w_curs,
1481 off_t curpos)
1483 const unsigned char *data;
1484 unsigned char delta_head[20], *in;
1485 git_zstream stream;
1486 int st;
1488 memset(&stream, 0, sizeof(stream));
1489 stream.next_out = delta_head;
1490 stream.avail_out = sizeof(delta_head);
1492 git_inflate_init(&stream);
1493 do {
1494 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1495 stream.next_in = in;
1496 st = git_inflate(&stream, Z_FINISH);
1497 curpos += stream.next_in - in;
1498 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1499 stream.total_out < sizeof(delta_head));
1500 git_inflate_end(&stream);
1501 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1502 error("delta data unpack-initial failed");
1503 return 0;
1506 /* Examine the initial part of the delta to figure out
1507 * the result size.
1509 data = delta_head;
1511 /* ignore base size */
1512 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1514 /* Read the result size */
1515 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1518 static off_t get_delta_base(struct packed_git *p,
1519 struct pack_window **w_curs,
1520 off_t *curpos,
1521 enum object_type type,
1522 off_t delta_obj_offset)
1524 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
1525 off_t base_offset;
1527 /* use_pack() assured us we have [base_info, base_info + 20)
1528 * as a range that we can look at without walking off the
1529 * end of the mapped window. Its actually the hash size
1530 * that is assured. An OFS_DELTA longer than the hash size
1531 * is stupid, as then a REF_DELTA would be smaller to store.
1533 if (type == OBJ_OFS_DELTA) {
1534 unsigned used = 0;
1535 unsigned char c = base_info[used++];
1536 base_offset = c & 127;
1537 while (c & 128) {
1538 base_offset += 1;
1539 if (!base_offset || MSB(base_offset, 7))
1540 return 0; /* overflow */
1541 c = base_info[used++];
1542 base_offset = (base_offset << 7) + (c & 127);
1544 base_offset = delta_obj_offset - base_offset;
1545 if (base_offset <= 0 || base_offset >= delta_obj_offset)
1546 return 0; /* out of bound */
1547 *curpos += used;
1548 } else if (type == OBJ_REF_DELTA) {
1549 /* The base entry _must_ be in the same pack */
1550 base_offset = find_pack_entry_one(base_info, p);
1551 *curpos += 20;
1552 } else
1553 die("I am totally screwed");
1554 return base_offset;
1557 /* forward declaration for a mutually recursive function */
1558 static int packed_object_info(struct packed_git *p, off_t offset,
1559 unsigned long *sizep, int *rtype);
1561 static int packed_delta_info(struct packed_git *p,
1562 struct pack_window **w_curs,
1563 off_t curpos,
1564 enum object_type type,
1565 off_t obj_offset,
1566 unsigned long *sizep)
1568 off_t base_offset;
1570 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
1571 if (!base_offset)
1572 return OBJ_BAD;
1573 type = packed_object_info(p, base_offset, NULL, NULL);
1574 if (type <= OBJ_NONE) {
1575 struct revindex_entry *revidx;
1576 const unsigned char *base_sha1;
1577 revidx = find_pack_revindex(p, base_offset);
1578 if (!revidx)
1579 return OBJ_BAD;
1580 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1581 mark_bad_packed_object(p, base_sha1);
1582 type = sha1_object_info(base_sha1, NULL);
1583 if (type <= OBJ_NONE)
1584 return OBJ_BAD;
1587 /* We choose to only get the type of the base object and
1588 * ignore potentially corrupt pack file that expects the delta
1589 * based on a base with a wrong size. This saves tons of
1590 * inflate() calls.
1592 if (sizep) {
1593 *sizep = get_size_from_delta(p, w_curs, curpos);
1594 if (*sizep == 0)
1595 type = OBJ_BAD;
1598 return type;
1601 int unpack_object_header(struct packed_git *p,
1602 struct pack_window **w_curs,
1603 off_t *curpos,
1604 unsigned long *sizep)
1606 unsigned char *base;
1607 unsigned long left;
1608 unsigned long used;
1609 enum object_type type;
1611 /* use_pack() assures us we have [base, base + 20) available
1612 * as a range that we can look at. (Its actually the hash
1613 * size that is assured.) With our object header encoding
1614 * the maximum deflated object size is 2^137, which is just
1615 * insane, so we know won't exceed what we have been given.
1617 base = use_pack(p, w_curs, *curpos, &left);
1618 used = unpack_object_header_buffer(base, left, &type, sizep);
1619 if (!used) {
1620 type = OBJ_BAD;
1621 } else
1622 *curpos += used;
1624 return type;
1627 static int packed_object_info(struct packed_git *p, off_t obj_offset,
1628 unsigned long *sizep, int *rtype)
1630 struct pack_window *w_curs = NULL;
1631 unsigned long size;
1632 off_t curpos = obj_offset;
1633 enum object_type type;
1635 type = unpack_object_header(p, &w_curs, &curpos, &size);
1636 if (rtype)
1637 *rtype = type; /* representation type */
1639 switch (type) {
1640 case OBJ_OFS_DELTA:
1641 case OBJ_REF_DELTA:
1642 type = packed_delta_info(p, &w_curs, curpos,
1643 type, obj_offset, sizep);
1644 break;
1645 case OBJ_COMMIT:
1646 case OBJ_TREE:
1647 case OBJ_BLOB:
1648 case OBJ_TAG:
1649 if (sizep)
1650 *sizep = size;
1651 break;
1652 default:
1653 error("unknown object type %i at offset %"PRIuMAX" in %s",
1654 type, (uintmax_t)obj_offset, p->pack_name);
1655 type = OBJ_BAD;
1657 unuse_pack(&w_curs);
1658 return type;
1661 static void *unpack_compressed_entry(struct packed_git *p,
1662 struct pack_window **w_curs,
1663 off_t curpos,
1664 unsigned long size)
1666 int st;
1667 git_zstream stream;
1668 unsigned char *buffer, *in;
1670 buffer = xmallocz(size);
1671 memset(&stream, 0, sizeof(stream));
1672 stream.next_out = buffer;
1673 stream.avail_out = size + 1;
1675 git_inflate_init(&stream);
1676 do {
1677 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1678 stream.next_in = in;
1679 st = git_inflate(&stream, Z_FINISH);
1680 if (!stream.avail_out)
1681 break; /* the payload is larger than it should be */
1682 curpos += stream.next_in - in;
1683 } while (st == Z_OK || st == Z_BUF_ERROR);
1684 git_inflate_end(&stream);
1685 if ((st != Z_STREAM_END) || stream.total_out != size) {
1686 free(buffer);
1687 return NULL;
1690 return buffer;
1693 #define MAX_DELTA_CACHE (256)
1695 static size_t delta_base_cached;
1697 static struct delta_base_cache_lru_list {
1698 struct delta_base_cache_lru_list *prev;
1699 struct delta_base_cache_lru_list *next;
1700 } delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
1702 static struct delta_base_cache_entry {
1703 struct delta_base_cache_lru_list lru;
1704 void *data;
1705 struct packed_git *p;
1706 off_t base_offset;
1707 unsigned long size;
1708 enum object_type type;
1709 } delta_base_cache[MAX_DELTA_CACHE];
1711 static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
1713 unsigned long hash;
1715 hash = (unsigned long)p + (unsigned long)base_offset;
1716 hash += (hash >> 8) + (hash >> 16);
1717 return hash % MAX_DELTA_CACHE;
1720 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
1722 unsigned long hash = pack_entry_hash(p, base_offset);
1723 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1724 return (ent->data && ent->p == p && ent->base_offset == base_offset);
1727 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
1728 unsigned long *base_size, enum object_type *type, int keep_cache)
1730 void *ret;
1731 unsigned long hash = pack_entry_hash(p, base_offset);
1732 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1734 ret = ent->data;
1735 if (!ret || ent->p != p || ent->base_offset != base_offset)
1736 return unpack_entry(p, base_offset, type, base_size);
1738 if (!keep_cache) {
1739 ent->data = NULL;
1740 ent->lru.next->prev = ent->lru.prev;
1741 ent->lru.prev->next = ent->lru.next;
1742 delta_base_cached -= ent->size;
1743 } else {
1744 ret = xmemdupz(ent->data, ent->size);
1746 *type = ent->type;
1747 *base_size = ent->size;
1748 return ret;
1751 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
1753 if (ent->data) {
1754 free(ent->data);
1755 ent->data = NULL;
1756 ent->lru.next->prev = ent->lru.prev;
1757 ent->lru.prev->next = ent->lru.next;
1758 delta_base_cached -= ent->size;
1762 void clear_delta_base_cache(void)
1764 unsigned long p;
1765 for (p = 0; p < MAX_DELTA_CACHE; p++)
1766 release_delta_base_cache(&delta_base_cache[p]);
1769 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
1770 void *base, unsigned long base_size, enum object_type type)
1772 unsigned long hash = pack_entry_hash(p, base_offset);
1773 struct delta_base_cache_entry *ent = delta_base_cache + hash;
1774 struct delta_base_cache_lru_list *lru;
1776 release_delta_base_cache(ent);
1777 delta_base_cached += base_size;
1779 for (lru = delta_base_cache_lru.next;
1780 delta_base_cached > delta_base_cache_limit
1781 && lru != &delta_base_cache_lru;
1782 lru = lru->next) {
1783 struct delta_base_cache_entry *f = (void *)lru;
1784 if (f->type == OBJ_BLOB)
1785 release_delta_base_cache(f);
1787 for (lru = delta_base_cache_lru.next;
1788 delta_base_cached > delta_base_cache_limit
1789 && lru != &delta_base_cache_lru;
1790 lru = lru->next) {
1791 struct delta_base_cache_entry *f = (void *)lru;
1792 release_delta_base_cache(f);
1795 ent->p = p;
1796 ent->base_offset = base_offset;
1797 ent->type = type;
1798 ent->data = base;
1799 ent->size = base_size;
1800 ent->lru.next = &delta_base_cache_lru;
1801 ent->lru.prev = delta_base_cache_lru.prev;
1802 delta_base_cache_lru.prev->next = &ent->lru;
1803 delta_base_cache_lru.prev = &ent->lru;
1806 static void *read_object(const unsigned char *sha1, enum object_type *type,
1807 unsigned long *size);
1809 static void *unpack_delta_entry(struct packed_git *p,
1810 struct pack_window **w_curs,
1811 off_t curpos,
1812 unsigned long delta_size,
1813 off_t obj_offset,
1814 enum object_type *type,
1815 unsigned long *sizep)
1817 void *delta_data, *result, *base;
1818 unsigned long base_size;
1819 off_t base_offset;
1821 base_offset = get_delta_base(p, w_curs, &curpos, *type, obj_offset);
1822 if (!base_offset) {
1823 error("failed to validate delta base reference "
1824 "at offset %"PRIuMAX" from %s",
1825 (uintmax_t)curpos, p->pack_name);
1826 return NULL;
1828 unuse_pack(w_curs);
1829 base = cache_or_unpack_entry(p, base_offset, &base_size, type, 0);
1830 if (!base) {
1832 * We're probably in deep shit, but let's try to fetch
1833 * the required base anyway from another pack or loose.
1834 * This is costly but should happen only in the presence
1835 * of a corrupted pack, and is better than failing outright.
1837 struct revindex_entry *revidx;
1838 const unsigned char *base_sha1;
1839 revidx = find_pack_revindex(p, base_offset);
1840 if (!revidx)
1841 return NULL;
1842 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
1843 error("failed to read delta base object %s"
1844 " at offset %"PRIuMAX" from %s",
1845 sha1_to_hex(base_sha1), (uintmax_t)base_offset,
1846 p->pack_name);
1847 mark_bad_packed_object(p, base_sha1);
1848 base = read_object(base_sha1, type, &base_size);
1849 if (!base)
1850 return NULL;
1853 delta_data = unpack_compressed_entry(p, w_curs, curpos, delta_size);
1854 if (!delta_data) {
1855 error("failed to unpack compressed delta "
1856 "at offset %"PRIuMAX" from %s",
1857 (uintmax_t)curpos, p->pack_name);
1858 free(base);
1859 return NULL;
1861 result = patch_delta(base, base_size,
1862 delta_data, delta_size,
1863 sizep);
1864 if (!result)
1865 die("failed to apply delta");
1866 free(delta_data);
1867 add_delta_base_cache(p, base_offset, base, base_size, *type);
1868 return result;
1871 static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
1873 static FILE *log_file;
1875 if (!log_file) {
1876 log_file = fopen(log_pack_access, "w");
1877 if (!log_file) {
1878 error("cannot open pack access log '%s' for writing: %s",
1879 log_pack_access, strerror(errno));
1880 log_pack_access = NULL;
1881 return;
1884 fprintf(log_file, "%s %"PRIuMAX"\n",
1885 p->pack_name, (uintmax_t)obj_offset);
1886 fflush(log_file);
1889 int do_check_packed_object_crc;
1891 void *unpack_entry(struct packed_git *p, off_t obj_offset,
1892 enum object_type *type, unsigned long *sizep)
1894 struct pack_window *w_curs = NULL;
1895 off_t curpos = obj_offset;
1896 void *data;
1898 if (log_pack_access)
1899 write_pack_access_log(p, obj_offset);
1901 if (do_check_packed_object_crc && p->index_version > 1) {
1902 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
1903 unsigned long len = revidx[1].offset - obj_offset;
1904 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
1905 const unsigned char *sha1 =
1906 nth_packed_object_sha1(p, revidx->nr);
1907 error("bad packed object CRC for %s",
1908 sha1_to_hex(sha1));
1909 mark_bad_packed_object(p, sha1);
1910 unuse_pack(&w_curs);
1911 return NULL;
1915 *type = unpack_object_header(p, &w_curs, &curpos, sizep);
1916 switch (*type) {
1917 case OBJ_OFS_DELTA:
1918 case OBJ_REF_DELTA:
1919 data = unpack_delta_entry(p, &w_curs, curpos, *sizep,
1920 obj_offset, type, sizep);
1921 break;
1922 case OBJ_COMMIT:
1923 case OBJ_TREE:
1924 case OBJ_BLOB:
1925 case OBJ_TAG:
1926 data = unpack_compressed_entry(p, &w_curs, curpos, *sizep);
1927 break;
1928 default:
1929 data = NULL;
1930 error("unknown object type %i at offset %"PRIuMAX" in %s",
1931 *type, (uintmax_t)obj_offset, p->pack_name);
1933 unuse_pack(&w_curs);
1934 return data;
1937 const unsigned char *nth_packed_object_sha1(struct packed_git *p,
1938 uint32_t n)
1940 const unsigned char *index = p->index_data;
1941 if (!index) {
1942 if (open_pack_index(p))
1943 return NULL;
1944 index = p->index_data;
1946 if (n >= p->num_objects)
1947 return NULL;
1948 index += 4 * 256;
1949 if (p->index_version == 1) {
1950 return index + 24 * n + 4;
1951 } else {
1952 index += 8;
1953 return index + 20 * n;
1957 off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
1959 const unsigned char *index = p->index_data;
1960 index += 4 * 256;
1961 if (p->index_version == 1) {
1962 return ntohl(*((uint32_t *)(index + 24 * n)));
1963 } else {
1964 uint32_t off;
1965 index += 8 + p->num_objects * (20 + 4);
1966 off = ntohl(*((uint32_t *)(index + 4 * n)));
1967 if (!(off & 0x80000000))
1968 return off;
1969 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
1970 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
1971 ntohl(*((uint32_t *)(index + 4)));
1975 off_t find_pack_entry_one(const unsigned char *sha1,
1976 struct packed_git *p)
1978 const uint32_t *level1_ofs = p->index_data;
1979 const unsigned char *index = p->index_data;
1980 unsigned hi, lo, stride;
1981 static int use_lookup = -1;
1982 static int debug_lookup = -1;
1984 if (debug_lookup < 0)
1985 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
1987 if (!index) {
1988 if (open_pack_index(p))
1989 return 0;
1990 level1_ofs = p->index_data;
1991 index = p->index_data;
1993 if (p->index_version > 1) {
1994 level1_ofs += 2;
1995 index += 8;
1997 index += 4 * 256;
1998 hi = ntohl(level1_ofs[*sha1]);
1999 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
2000 if (p->index_version > 1) {
2001 stride = 20;
2002 } else {
2003 stride = 24;
2004 index += 4;
2007 if (debug_lookup)
2008 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
2009 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2011 if (use_lookup < 0)
2012 use_lookup = !!getenv("GIT_USE_LOOKUP");
2013 if (use_lookup) {
2014 int pos = sha1_entry_pos(index, stride, 0,
2015 lo, hi, p->num_objects, sha1);
2016 if (pos < 0)
2017 return 0;
2018 return nth_packed_object_offset(p, pos);
2021 do {
2022 unsigned mi = (lo + hi) / 2;
2023 int cmp = hashcmp(index + mi * stride, sha1);
2025 if (debug_lookup)
2026 printf("lo %u hi %u rg %u mi %u\n",
2027 lo, hi, hi - lo, mi);
2028 if (!cmp)
2029 return nth_packed_object_offset(p, mi);
2030 if (cmp > 0)
2031 hi = mi;
2032 else
2033 lo = mi+1;
2034 } while (lo < hi);
2035 return 0;
2038 int is_pack_valid(struct packed_git *p)
2040 /* An already open pack is known to be valid. */
2041 if (p->pack_fd != -1)
2042 return 1;
2044 /* If the pack has one window completely covering the
2045 * file size, the pack is known to be valid even if
2046 * the descriptor is not currently open.
2048 if (p->windows) {
2049 struct pack_window *w = p->windows;
2051 if (!w->offset && w->len == p->pack_size)
2052 return 1;
2055 /* Force the pack to open to prove its valid. */
2056 return !open_packed_git(p);
2059 static int fill_pack_entry(const unsigned char *sha1,
2060 struct pack_entry *e,
2061 struct packed_git *p)
2063 off_t offset;
2065 if (p->num_bad_objects) {
2066 unsigned i;
2067 for (i = 0; i < p->num_bad_objects; i++)
2068 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2069 return 0;
2072 offset = find_pack_entry_one(sha1, p);
2073 if (!offset)
2074 return 0;
2077 * We are about to tell the caller where they can locate the
2078 * requested object. We better make sure the packfile is
2079 * still here and can be accessed before supplying that
2080 * answer, as it may have been deleted since the index was
2081 * loaded!
2083 if (!is_pack_valid(p)) {
2084 warning("packfile %s cannot be accessed", p->pack_name);
2085 return 0;
2087 e->offset = offset;
2088 e->p = p;
2089 hashcpy(e->sha1, sha1);
2090 return 1;
2093 static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
2095 struct packed_git *p;
2097 prepare_packed_git();
2098 if (!packed_git)
2099 return 0;
2101 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2102 return 1;
2104 for (p = packed_git; p; p = p->next) {
2105 if (p == last_found_pack || !fill_pack_entry(sha1, e, p))
2106 continue;
2108 last_found_pack = p;
2109 return 1;
2111 return 0;
2114 struct packed_git *find_sha1_pack(const unsigned char *sha1,
2115 struct packed_git *packs)
2117 struct packed_git *p;
2119 for (p = packs; p; p = p->next) {
2120 if (find_pack_entry_one(sha1, p))
2121 return p;
2123 return NULL;
2127 static int sha1_loose_object_info(const unsigned char *sha1, unsigned long *sizep)
2129 int status;
2130 unsigned long mapsize, size;
2131 void *map;
2132 git_zstream stream;
2133 char hdr[32];
2135 map = map_sha1_file(sha1, &mapsize);
2136 if (!map)
2137 return error("unable to find %s", sha1_to_hex(sha1));
2138 if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
2139 status = error("unable to unpack %s header",
2140 sha1_to_hex(sha1));
2141 else if ((status = parse_sha1_header(hdr, &size)) < 0)
2142 status = error("unable to parse %s header", sha1_to_hex(sha1));
2143 else if (sizep)
2144 *sizep = size;
2145 git_inflate_end(&stream);
2146 munmap(map, mapsize);
2147 return status;
2150 /* returns enum object_type or negative */
2151 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
2153 struct cached_object *co;
2154 struct pack_entry e;
2155 int status, rtype;
2157 co = find_cached_object(sha1);
2158 if (co) {
2159 if (oi->sizep)
2160 *(oi->sizep) = co->size;
2161 oi->whence = OI_CACHED;
2162 return co->type;
2165 if (!find_pack_entry(sha1, &e)) {
2166 /* Most likely it's a loose object. */
2167 status = sha1_loose_object_info(sha1, oi->sizep);
2168 if (status >= 0) {
2169 oi->whence = OI_LOOSE;
2170 return status;
2173 /* Not a loose object; someone else may have just packed it. */
2174 reprepare_packed_git();
2175 if (!find_pack_entry(sha1, &e))
2176 return status;
2179 status = packed_object_info(e.p, e.offset, oi->sizep, &rtype);
2180 if (status < 0) {
2181 mark_bad_packed_object(e.p, sha1);
2182 status = sha1_object_info_extended(sha1, oi);
2183 } else if (in_delta_base_cache(e.p, e.offset)) {
2184 oi->whence = OI_DBCACHED;
2185 } else {
2186 oi->whence = OI_PACKED;
2187 oi->u.packed.offset = e.offset;
2188 oi->u.packed.pack = e.p;
2189 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2190 rtype == OBJ_OFS_DELTA);
2193 return status;
2196 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2198 struct object_info oi;
2200 oi.sizep = sizep;
2201 return sha1_object_info_extended(sha1, &oi);
2204 static void *read_packed_sha1(const unsigned char *sha1,
2205 enum object_type *type, unsigned long *size)
2207 struct pack_entry e;
2208 void *data;
2210 if (!find_pack_entry(sha1, &e))
2211 return NULL;
2212 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2213 if (!data) {
2215 * We're probably in deep shit, but let's try to fetch
2216 * the required object anyway from another pack or loose.
2217 * This should happen only in the presence of a corrupted
2218 * pack, and is better than failing outright.
2220 error("failed to read object %s at offset %"PRIuMAX" from %s",
2221 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2222 mark_bad_packed_object(e.p, sha1);
2223 data = read_object(sha1, type, size);
2225 return data;
2228 int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2229 unsigned char *sha1)
2231 struct cached_object *co;
2233 hash_sha1_file(buf, len, typename(type), sha1);
2234 if (has_sha1_file(sha1) || find_cached_object(sha1))
2235 return 0;
2236 if (cached_object_alloc <= cached_object_nr) {
2237 cached_object_alloc = alloc_nr(cached_object_alloc);
2238 cached_objects = xrealloc(cached_objects,
2239 sizeof(*cached_objects) *
2240 cached_object_alloc);
2242 co = &cached_objects[cached_object_nr++];
2243 co->size = len;
2244 co->type = type;
2245 co->buf = xmalloc(len);
2246 memcpy(co->buf, buf, len);
2247 hashcpy(co->sha1, sha1);
2248 return 0;
2251 static void *read_object(const unsigned char *sha1, enum object_type *type,
2252 unsigned long *size)
2254 unsigned long mapsize;
2255 void *map, *buf;
2256 struct cached_object *co;
2258 co = find_cached_object(sha1);
2259 if (co) {
2260 *type = co->type;
2261 *size = co->size;
2262 return xmemdupz(co->buf, co->size);
2265 buf = read_packed_sha1(sha1, type, size);
2266 if (buf)
2267 return buf;
2268 map = map_sha1_file(sha1, &mapsize);
2269 if (map) {
2270 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
2271 munmap(map, mapsize);
2272 return buf;
2274 reprepare_packed_git();
2275 return read_packed_sha1(sha1, type, size);
2279 * This function dies on corrupt objects; the callers who want to
2280 * deal with them should arrange to call read_object() and give error
2281 * messages themselves.
2283 void *read_sha1_file_extended(const unsigned char *sha1,
2284 enum object_type *type,
2285 unsigned long *size,
2286 unsigned flag)
2288 void *data;
2289 char *path;
2290 const struct packed_git *p;
2291 const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
2292 ? lookup_replace_object(sha1) : sha1;
2294 errno = 0;
2295 data = read_object(repl, type, size);
2296 if (data)
2297 return data;
2299 if (errno && errno != ENOENT)
2300 die_errno("failed to read object %s", sha1_to_hex(sha1));
2302 /* die if we replaced an object with one that does not exist */
2303 if (repl != sha1)
2304 die("replacement %s not found for %s",
2305 sha1_to_hex(repl), sha1_to_hex(sha1));
2307 if (has_loose_object(repl)) {
2308 path = sha1_file_name(sha1);
2309 die("loose object %s (stored in %s) is corrupt",
2310 sha1_to_hex(repl), path);
2313 if ((p = has_packed_and_bad(repl)) != NULL)
2314 die("packed object %s (stored in %s) is corrupt",
2315 sha1_to_hex(repl), p->pack_name);
2317 return NULL;
2320 void *read_object_with_reference(const unsigned char *sha1,
2321 const char *required_type_name,
2322 unsigned long *size,
2323 unsigned char *actual_sha1_return)
2325 enum object_type type, required_type;
2326 void *buffer;
2327 unsigned long isize;
2328 unsigned char actual_sha1[20];
2330 required_type = type_from_string(required_type_name);
2331 hashcpy(actual_sha1, sha1);
2332 while (1) {
2333 int ref_length = -1;
2334 const char *ref_type = NULL;
2336 buffer = read_sha1_file(actual_sha1, &type, &isize);
2337 if (!buffer)
2338 return NULL;
2339 if (type == required_type) {
2340 *size = isize;
2341 if (actual_sha1_return)
2342 hashcpy(actual_sha1_return, actual_sha1);
2343 return buffer;
2345 /* Handle references */
2346 else if (type == OBJ_COMMIT)
2347 ref_type = "tree ";
2348 else if (type == OBJ_TAG)
2349 ref_type = "object ";
2350 else {
2351 free(buffer);
2352 return NULL;
2354 ref_length = strlen(ref_type);
2356 if (ref_length + 40 > isize ||
2357 memcmp(buffer, ref_type, ref_length) ||
2358 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
2359 free(buffer);
2360 return NULL;
2362 free(buffer);
2363 /* Now we have the ID of the referred-to object in
2364 * actual_sha1. Check again. */
2368 static void write_sha1_file_prepare(const void *buf, unsigned long len,
2369 const char *type, unsigned char *sha1,
2370 char *hdr, int *hdrlen)
2372 git_SHA_CTX c;
2374 /* Generate the header */
2375 *hdrlen = sprintf(hdr, "%s %lu", type, len)+1;
2377 /* Sha1.. */
2378 git_SHA1_Init(&c);
2379 git_SHA1_Update(&c, hdr, *hdrlen);
2380 git_SHA1_Update(&c, buf, len);
2381 git_SHA1_Final(sha1, &c);
2385 * Move the just written object into its final resting place.
2386 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2387 * "moving" is only a part of what it does, when no patch between
2388 * master to pu changes the call sites of this function.
2390 int move_temp_to_file(const char *tmpfile, const char *filename)
2392 int ret = 0;
2394 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
2395 goto try_rename;
2396 else if (link(tmpfile, filename))
2397 ret = errno;
2400 * Coda hack - coda doesn't like cross-directory links,
2401 * so we fall back to a rename, which will mean that it
2402 * won't be able to check collisions, but that's not a
2403 * big deal.
2405 * The same holds for FAT formatted media.
2407 * When this succeeds, we just return. We have nothing
2408 * left to unlink.
2410 if (ret && ret != EEXIST) {
2411 try_rename:
2412 if (!rename(tmpfile, filename))
2413 goto out;
2414 ret = errno;
2416 unlink_or_warn(tmpfile);
2417 if (ret) {
2418 if (ret != EEXIST) {
2419 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
2421 /* FIXME!!! Collision check here ? */
2424 out:
2425 if (adjust_shared_perm(filename))
2426 return error("unable to set permission to '%s'", filename);
2427 return 0;
2430 static int write_buffer(int fd, const void *buf, size_t len)
2432 if (write_in_full(fd, buf, len) < 0)
2433 return error("file write error (%s)", strerror(errno));
2434 return 0;
2437 int hash_sha1_file(const void *buf, unsigned long len, const char *type,
2438 unsigned char *sha1)
2440 char hdr[32];
2441 int hdrlen;
2442 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2443 return 0;
2446 /* Finalize a file on disk, and close it. */
2447 static void close_sha1_file(int fd)
2449 if (fsync_object_files)
2450 fsync_or_die(fd, "sha1 file");
2451 if (close(fd) != 0)
2452 die_errno("error when closing sha1 file");
2455 /* Size of directory component, including the ending '/' */
2456 static inline int directory_size(const char *filename)
2458 const char *s = strrchr(filename, '/');
2459 if (!s)
2460 return 0;
2461 return s - filename + 1;
2465 * This creates a temporary file in the same directory as the final
2466 * 'filename'
2468 * We want to avoid cross-directory filename renames, because those
2469 * can have problems on various filesystems (FAT, NFS, Coda).
2471 static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
2473 int fd, dirlen = directory_size(filename);
2475 if (dirlen + 20 > bufsiz) {
2476 errno = ENAMETOOLONG;
2477 return -1;
2479 memcpy(buffer, filename, dirlen);
2480 strcpy(buffer + dirlen, "tmp_obj_XXXXXX");
2481 fd = git_mkstemp_mode(buffer, 0444);
2482 if (fd < 0 && dirlen && errno == ENOENT) {
2483 /* Make sure the directory exists */
2484 memcpy(buffer, filename, dirlen);
2485 buffer[dirlen-1] = 0;
2486 if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
2487 return -1;
2489 /* Try again */
2490 strcpy(buffer + dirlen - 1, "/tmp_obj_XXXXXX");
2491 fd = git_mkstemp_mode(buffer, 0444);
2493 return fd;
2496 static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
2497 const void *buf, unsigned long len, time_t mtime)
2499 int fd, ret;
2500 unsigned char compressed[4096];
2501 git_zstream stream;
2502 git_SHA_CTX c;
2503 unsigned char parano_sha1[20];
2504 char *filename;
2505 static char tmp_file[PATH_MAX];
2507 filename = sha1_file_name(sha1);
2508 fd = create_tmpfile(tmp_file, sizeof(tmp_file), filename);
2509 if (fd < 0) {
2510 if (errno == EACCES)
2511 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2512 else
2513 return error("unable to create temporary file: %s", strerror(errno));
2516 /* Set it up */
2517 memset(&stream, 0, sizeof(stream));
2518 git_deflate_init(&stream, zlib_compression_level);
2519 stream.next_out = compressed;
2520 stream.avail_out = sizeof(compressed);
2521 git_SHA1_Init(&c);
2523 /* First header.. */
2524 stream.next_in = (unsigned char *)hdr;
2525 stream.avail_in = hdrlen;
2526 while (git_deflate(&stream, 0) == Z_OK)
2527 ; /* nothing */
2528 git_SHA1_Update(&c, hdr, hdrlen);
2530 /* Then the data itself.. */
2531 stream.next_in = (void *)buf;
2532 stream.avail_in = len;
2533 do {
2534 unsigned char *in0 = stream.next_in;
2535 ret = git_deflate(&stream, Z_FINISH);
2536 git_SHA1_Update(&c, in0, stream.next_in - in0);
2537 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
2538 die("unable to write sha1 file");
2539 stream.next_out = compressed;
2540 stream.avail_out = sizeof(compressed);
2541 } while (ret == Z_OK);
2543 if (ret != Z_STREAM_END)
2544 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
2545 ret = git_deflate_end_gently(&stream);
2546 if (ret != Z_OK)
2547 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
2548 git_SHA1_Final(parano_sha1, &c);
2549 if (hashcmp(sha1, parano_sha1) != 0)
2550 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
2552 close_sha1_file(fd);
2554 if (mtime) {
2555 struct utimbuf utb;
2556 utb.actime = mtime;
2557 utb.modtime = mtime;
2558 if (utime(tmp_file, &utb) < 0)
2559 warning("failed utime() on %s: %s",
2560 tmp_file, strerror(errno));
2563 return move_temp_to_file(tmp_file, filename);
2566 int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
2568 unsigned char sha1[20];
2569 char hdr[32];
2570 int hdrlen;
2572 /* Normally if we have it in the pack then we do not bother writing
2573 * it out into .git/objects/??/?{38} file.
2575 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2576 if (returnsha1)
2577 hashcpy(returnsha1, sha1);
2578 if (has_sha1_file(sha1))
2579 return 0;
2580 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
2583 int force_object_loose(const unsigned char *sha1, time_t mtime)
2585 void *buf;
2586 unsigned long len;
2587 enum object_type type;
2588 char hdr[32];
2589 int hdrlen;
2590 int ret;
2592 if (has_loose_object(sha1))
2593 return 0;
2594 buf = read_packed_sha1(sha1, &type, &len);
2595 if (!buf)
2596 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
2597 hdrlen = sprintf(hdr, "%s %lu", typename(type), len) + 1;
2598 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
2599 free(buf);
2601 return ret;
2604 int has_pack_index(const unsigned char *sha1)
2606 struct stat st;
2607 if (stat(sha1_pack_index_name(sha1), &st))
2608 return 0;
2609 return 1;
2612 int has_sha1_pack(const unsigned char *sha1)
2614 struct pack_entry e;
2615 return find_pack_entry(sha1, &e);
2618 int has_sha1_file(const unsigned char *sha1)
2620 struct pack_entry e;
2622 if (find_pack_entry(sha1, &e))
2623 return 1;
2624 return has_loose_object(sha1);
2627 static void check_tree(const void *buf, size_t size)
2629 struct tree_desc desc;
2630 struct name_entry entry;
2632 init_tree_desc(&desc, buf, size);
2633 while (tree_entry(&desc, &entry))
2634 /* do nothing
2635 * tree_entry() will die() on malformed entries */
2639 static void check_commit(const void *buf, size_t size)
2641 struct commit c;
2642 memset(&c, 0, sizeof(c));
2643 if (parse_commit_buffer(&c, buf, size))
2644 die("corrupt commit");
2647 static void check_tag(const void *buf, size_t size)
2649 struct tag t;
2650 memset(&t, 0, sizeof(t));
2651 if (parse_tag_buffer(&t, buf, size))
2652 die("corrupt tag");
2655 static int index_mem(unsigned char *sha1, void *buf, size_t size,
2656 enum object_type type,
2657 const char *path, unsigned flags)
2659 int ret, re_allocated = 0;
2660 int write_object = flags & HASH_WRITE_OBJECT;
2662 if (!type)
2663 type = OBJ_BLOB;
2666 * Convert blobs to git internal format
2668 if ((type == OBJ_BLOB) && path) {
2669 struct strbuf nbuf = STRBUF_INIT;
2670 if (convert_to_git(path, buf, size, &nbuf,
2671 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
2672 buf = strbuf_detach(&nbuf, &size);
2673 re_allocated = 1;
2676 if (flags & HASH_FORMAT_CHECK) {
2677 if (type == OBJ_TREE)
2678 check_tree(buf, size);
2679 if (type == OBJ_COMMIT)
2680 check_commit(buf, size);
2681 if (type == OBJ_TAG)
2682 check_tag(buf, size);
2685 if (write_object)
2686 ret = write_sha1_file(buf, size, typename(type), sha1);
2687 else
2688 ret = hash_sha1_file(buf, size, typename(type), sha1);
2689 if (re_allocated)
2690 free(buf);
2691 return ret;
2694 static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
2695 const char *path, unsigned flags)
2697 struct strbuf sbuf = STRBUF_INIT;
2698 int ret;
2700 if (strbuf_read(&sbuf, fd, 4096) >= 0)
2701 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
2702 else
2703 ret = -1;
2704 strbuf_release(&sbuf);
2705 return ret;
2708 #define SMALL_FILE_SIZE (32*1024)
2710 static int index_core(unsigned char *sha1, int fd, size_t size,
2711 enum object_type type, const char *path,
2712 unsigned flags)
2714 int ret;
2716 if (!size) {
2717 ret = index_mem(sha1, NULL, size, type, path, flags);
2718 } else if (size <= SMALL_FILE_SIZE) {
2719 char *buf = xmalloc(size);
2720 if (size == read_in_full(fd, buf, size))
2721 ret = index_mem(sha1, buf, size, type, path, flags);
2722 else
2723 ret = error("short read %s", strerror(errno));
2724 free(buf);
2725 } else {
2726 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
2727 ret = index_mem(sha1, buf, size, type, path, flags);
2728 munmap(buf, size);
2730 return ret;
2734 * This creates one packfile per large blob unless bulk-checkin
2735 * machinery is "plugged".
2737 * This also bypasses the usual "convert-to-git" dance, and that is on
2738 * purpose. We could write a streaming version of the converting
2739 * functions and insert that before feeding the data to fast-import
2740 * (or equivalent in-core API described above). However, that is
2741 * somewhat complicated, as we do not know the size of the filter
2742 * result, which we need to know beforehand when writing a git object.
2743 * Since the primary motivation for trying to stream from the working
2744 * tree file and to avoid mmaping it in core is to deal with large
2745 * binary blobs, they generally do not want to get any conversion, and
2746 * callers should avoid this code path when filters are requested.
2748 static int index_stream(unsigned char *sha1, int fd, size_t size,
2749 enum object_type type, const char *path,
2750 unsigned flags)
2752 return index_bulk_checkin(sha1, fd, size, type, path, flags);
2755 int index_fd(unsigned char *sha1, int fd, struct stat *st,
2756 enum object_type type, const char *path, unsigned flags)
2758 int ret;
2759 size_t size = xsize_t(st->st_size);
2761 if (!S_ISREG(st->st_mode))
2762 ret = index_pipe(sha1, fd, type, path, flags);
2763 else if (size <= big_file_threshold || type != OBJ_BLOB ||
2764 (path && would_convert_to_git(path, NULL, 0, 0)))
2765 ret = index_core(sha1, fd, size, type, path, flags);
2766 else
2767 ret = index_stream(sha1, fd, size, type, path, flags);
2768 close(fd);
2769 return ret;
2772 int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
2774 int fd;
2775 struct strbuf sb = STRBUF_INIT;
2777 switch (st->st_mode & S_IFMT) {
2778 case S_IFREG:
2779 fd = open(path, O_RDONLY);
2780 if (fd < 0)
2781 return error("open(\"%s\"): %s", path,
2782 strerror(errno));
2783 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
2784 return error("%s: failed to insert into database",
2785 path);
2786 break;
2787 case S_IFLNK:
2788 if (strbuf_readlink(&sb, path, st->st_size)) {
2789 char *errstr = strerror(errno);
2790 return error("readlink(\"%s\"): %s", path,
2791 errstr);
2793 if (!(flags & HASH_WRITE_OBJECT))
2794 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
2795 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
2796 return error("%s: failed to insert into database",
2797 path);
2798 strbuf_release(&sb);
2799 break;
2800 case S_IFDIR:
2801 return resolve_gitlink_ref(path, "HEAD", sha1);
2802 default:
2803 return error("%s: unsupported file type", path);
2805 return 0;
2808 int read_pack_header(int fd, struct pack_header *header)
2810 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
2811 /* "eof before pack header was fully read" */
2812 return PH_ERROR_EOF;
2814 if (header->hdr_signature != htonl(PACK_SIGNATURE))
2815 /* "protocol error (pack signature mismatch detected)" */
2816 return PH_ERROR_PACK_SIGNATURE;
2817 if (!pack_version_ok(header->hdr_version))
2818 /* "protocol error (pack version unsupported)" */
2819 return PH_ERROR_PROTOCOL;
2820 return 0;
2823 void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
2825 enum object_type type = sha1_object_info(sha1, NULL);
2826 if (type < 0)
2827 die("%s is not a valid object", sha1_to_hex(sha1));
2828 if (type != expect)
2829 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
2830 typename(expect));